!!!!!!!!!!!!2009-Wiley-Handbook of Measuring System Design
Uploaded by
zoranmiskovic!!!!!!!!!!!!2009-Wiley-Handbook of Measuring System Design
Uploaded by
zoranmiskovicVOLUME 1
Contributors.
Preface.
Abbreviations and Acronyms.
Introduction.
PART 1. FOUNDATIONS OF MEASURING.
SECTION 1. THE PROCESS OF MEASURING.
1. Sophistication of Measurement and its Body of Knowledge (Peter H. Sydenham).
2. Organization of Instrument Science (Joseph McGhee).
3. Measures and Metrics; Their Application (Peter H. Sydenham)
4. Economic Consideration of Measurement (Peter H. Sydenham).
5. Humans in the Real World (Joseph McGhee).
6. Substructure of Human-Machine Systems (Joseph McGhee).
SECTION 2. MEASURING THEORY AND PHILOSOPHY.
7. Introduction to Measurement Theory and Philosophy (Ludwik Finkelstein).
8. Formal Theory of Measurement (Ludwik Finkelstein).
9. Nature and Properties of Measurement (Ludwik Finkelstein).
10. Extensions of the Representational Theory of Measurement (Ludwik Finkelstein).
11. Measurement Theory in Physical, Social, and Psychological Science (Ludwik Finkelstein).
12. Fuzzy Approaches for Measurement (Eric Benoit, Laurent Foulloy and Gilles Mauris).
13. Signals, Information and Knowledge, and Meaning (Qing Ping Yang).
14. Hierarchical Aspects of Measurement Systems (Joseph McGhee).
15. Typical Measurement Systems Architecture (Joseph McGhee).
SECTION 3. ENVIRONMENTAL FACTORS.
16. Reduction of Influence Factors (Paul P.L. Regtien)
17. EMC and EMI (Kim R. Fowler).
SECTION 4. FEEDBACK IN MEASURING SYSTEMS.
18. Nature and Scope of Closed-loop Systems (Peter H. Sydenham).
19. Dynamic Behavior of Closed-loop Systems (Peter H. Sydenham).
20. Closed-loop Sampled Data Systems (Peter H. Sydenham).
21. Nonlinear Closed-loop Systems (Peter H. Sydenham).
SECTION 5. MESSAGING THEORY.
22. Characteristics and Theory of Knowledge (Luca P. Mari).
23. Principles of Semiotics as Related to Measurement (Luca P. Mari).
24. Principles of Epistemology as Related to Measurement (Timothy Lindsay John Ferris).
SECTION 6. SIGNAL THEORY.
25. Introduction to Signals in Physical Systems (Eugen Georg Woschni).
26. Signal Classification (Eugen Georg Woschni).
27. Signals in the Frequency Domain (Eugen Georg Woschni)).
28. Signals in the Time Domain (Eugen Georg Woschni).
29. Relationship Between Signals in the Time and Frequency Domain (Eugen Georg Woschni).
30. Statistical Signal Representations (Eugen Georg Woschni).
31. Discrete Signal Theory (Eugen Georg Woschni).
32. Geometrical Signal Representation (Eugen Georg Woschni).
33. Coding Theory and its Application to Measurement (Eugen Georg Woschni).
34. Modulation Theory (Eugen Georg Woschni).
SECTION 7. SYSTEMS THEORY.
35. Systems in the Time-Domain (Eugen Georg Woschni).
36. Systems in the Frequency Domain (Eugen Georg Woschni)
37. Relationship Between Systems in the Time and Frequency Domain (Eugen Georg Woschni).
38. Stability Issues (Eugen Georg Woschni).
SECTION 8. SOURCES OF INFORMATION ON MEASUREMENT.
39. Characteristics of Data, Information, Knowledge, and Wisdom (Timothy Lindsay John
Ferris).
40. Sources of Information (Peter H. Sydenham).
41. Terminology and Classification of Measurement Systems (Peter H. Sydenham).
42. Information Databases of Relevance to Measurement (Peter H. Sydenham).
PART 2. UNITS, STANDARDS AND CALIBRATION.
SECTION 1. STANDARDS SUPPORTING MEASUREMENT.
43. Units (Brian W. Petley).
44. Types of Paper Standards and their Purpose (Halit Eren).
SECTION 2. CALIBRATION.
45. Calibration Process (Halit Eren).
46. Calibration Interval (Peter H. Sydenham).
47. Internet Calibration (Richard A. Dudley).
PART 3. ERROR AND UNCERTAINTY.
SECTION 1. ERROR AND UNCERTAINTY.
48. Common Sources of Errors in Measurement Systems (Dietrich Hofmann).
49. General Characterization of Systematic and Stochastic Errors (Martin Halaj).
50. Errors in Signal Systems (Eugen Georg Woschni).
51. Errors in Digital Signal Systems (Luca P. Mari).
52. Error Models, Error Budgets and their Calculation (Rudolf Palencár).
53. Calculation and Treatment of Errors (Joseph McGhee).
54. Explanation of Key Error and Uncertainty Concepts and Terms (Luca P. Mari).
55. Uncertainty Determination (Joseph McGhee).
PART 4. MEASURING SYSTEM BEHAVIOR.
SECTION 1. MEASURING SYSTEM SPECIFICATION.
56. Transfer Characteristics of Instrument Stages (Peter H. Sydenham).
57. Static Considerations of General Instrumentation (Peter H. Sydenham).
58. Description of Accuracy: Linearity, and Drift (Peter H. Sydenham).
59. Introduction to the Dynamic Regime of Measurement Systems (Peter H. Sydenham).
60. Zero-order System Dynamics (Peter H. Sydenham).
61. First-order System Dynamics (Peter H. Sydenham).
62. Second-order System Dynamics (Peter H. Sydenham).
VOLUME 2
PART 5. MEASURING SYSTEM DESIGN.
SECTION 1. ENGINEERING A MEASURING SYSTEM.
63. Outline of Systems Thinking (Peter H. Sydenham).
64. Executing A Measuring System Design (Peter H. Sydenham).
65. Life Cycle Concept (Floyd Guyton Patterson Jr.).
66. Phases of System Life Cycle (Kim R. Fowler).
67. Principle of Concept of Operations (ConOps) (Jack Ring).
68. Setting the System Boundaries (Joseph McGhee).
69. Requirements Allocation (Andrew Kusiak and Fang Qin).
SECTION 2. DESIGN METHODOLOGIES.
70. Measuring System Design Methodologies (Ludwik Finkelstein).
71. Modeling Methodology (Peter H. Sydenham).
72. Mathematical Methods of Optimization (Halit Eren).
SECTION 3. ELECTRONIC AND ELECTRICAL REGIME.
73. Overview of Electrical and Electronic Technique (Peter H. Sydenham).
74. Basic Electronic Components (Peter H. Sydenham).
75. Electronic System Building Blocks (Peter H. Sydenham).
76. Electronic Systems Design (Peter H. Sydenham).
77. Limits of Detection in Electronic Systems (Peter H. Sydenham).
78. Embedded Systems (Timothy Wilmshurst).
79. Testing Electronic Systems (Patrick D.T. O’Connor).
SECTION 4. FINE MECHANICAL REGIME.
80. Principles of Fine Mechanics – Kinematic and Elastic Designs (Peter H. Sydenham).
81. Principles of Fine Mechanics – Systems Considerations (Peter H. Sydenham).
82. Kinematical Regime – Members and Linkages (Peter H. Sydenham).
83. Kinematical Regime - Fasteners, Bearings (Peter H. Sydenham
84. 83. Kinematical Regime – Rotary Motion (Peter H. Sydenham
85. Elastic Regime of Design – Design Principles (Peter H. Sydenham).
86. Elastic Regime of Design – Spring Systems (Peter H. Sydenham).
87. Elastic Regime of Design – Plates and Bimorphs (Peter H. Sydenham).
88. Error Sources in Fine Mechanics (Peter H. Sydenham).
SECTION 5. VISIBLE RADIATION REGIME.
89. Optical Materials (Pak L. Chu).
90. Optical Elements (Pak L. Chu).
91. Light Sources and Detectors (Miroslaw Jonasz).
92. Optical Measuring Instruments (Peter H. Sydenham).
93. Testing Optical and Other Radiation Systems (Alan J. Cormier).
SECTION 6. HUMAN FACTORS ENGINEERIN.
94. Human Factors Engineering (Nicholas I. Beagley).
95. Human-Machine Interface (Nicholas I. Beagley).
96. The Domains of Human Factors Integration (Nicholas I. Beagley).
97. Design Methodology (Nicholas I. Beagley).
SECTION 7. QUALITY IN MEASURING SYSTEMS.
98. Reliability and Maintainability (Patrick D.T. O’Connor).
99. Safety Organization (Peter H. Sydenham).
100. Safety Analysis Methods (Peter H. Sydenham).
101. Assessing and Demonstrating Safety (Peter H. Sydenham).
102. Introduction to the Legal Process (Christopher Sweet).
103. Legal Liability Issues for Designers – A Case Study (Christopher Sweet).
PART 6. MODELING MEASURING SYSTEMS.
SECTION 1. MODELING MEASURING SYSTEMS.
104. Models of the Measurement Process (Luca P. Mari).
105. Modeling with LabVIEW™ (Wieslaw Ttaczala).
106. Virtual Instrumentation in Physics (Wieslaw Ttaczala).
PART 7. ELEMENTS: A – SENSORS.
SECTION 1. SENSOR FUNDAMENTALS.
107. Principles of Sensor Science (Joseph McGhee).
108. Transducer Fundamentals (Paul P.L. Regtien).
109. Structure and Energy in Sensor Systems (Joseph McGhee).
110. Signal/Energy Matrix Modeling (Joseph McGhee).
111. Classification of Sensors (Joseph McGhee).
112. Systematic Description of Sensors (Paul P.L. Regtien).
113. Force-feedback Sensors (Barry E. Jones).
SECTION 2. THE SENSING INTERFACE.
114. Models of the Sensor Interface (Qing Ping Yang).
115. Designing the Sensor Interface (Qing Ping Yang).
116. Selection of Sensors (Paul P.L. Regtien).
117. Materials in Measuring Systems (Peter H. Sydenham).
118. Ultrasonic Sensors (Peter J. Lesniewski).
119. Ultrasonic Instrumentation Principles (Lawrence C. Lynnworth).
120. Ultrasonic Instrumentation Design (Lawrence C. Lynnworth).
PART 8. ELEMENTS: B – SIGNAL CONDITIONING.
SECTION 1. ANALOG SIGNAL CONDITIONING.
121. Signals in the Presence of Noise (Richard Burdett).
122. Operational Amplifiers (Joseph McGhee).
123. Instrumentation Amplifiers (Joseph McGhee).
124. Specialized Amplifiers for Measurement Systems (Joseph McGhee).
125. Outline of Purpose of Analog Data Filters (Joseph McGhee).
SECTION 2. ELECTRICAL BRIDGES.
126. Electrical Bridge Circuits – Basic Information (Zygmunt L. Warsza).
127. Unbalanced DC Bridges (Zygmunt L. Warsza).
SECTION 3. AI SIGNAL PROCESSING TECHNIQUES>
128. Name and Scope of AI Techniques (Ajith Abraham).
129. Artificial Neural Networks (Ajith Abraham).
130. Rule-based Expert Systems (Ajith Abraham).
131. Evolution Computation (Ajith Abraham).
VOLUME 3
PART 9. ELEMENTS: C – DATA ACQUISITION AND PROCESSING SYSTEMS.
SECTION 1. DAS COMPONENTS.
132. Data Acquisition Systems (DAS) in General (Gerd Wöstenkühler).
133. Amplifiers and Filters for DAS (Gerd Wöstenkühler).
134. Analog Multiplexers (Gerd Wöstenkühler).
135. Sample-hold Circuits (Gerd Wöstenkühler).
136. Quantizing Theory Relevant to DAS (Gerd Wöstenkühler).
137. Coding for Data Converters (Gerd Wöstenkühler).
138. Sampling Theory Relevant to DAS (Gerd Wöstenkühler).
139. Analog-to-Digital (A/D) Converters (Gerd Wöstenkühler).
140. Integrating Type (A/D) Converters (Gerd Wöstenkühler).
141. Digital-to-Analog (D/A) Converters (Gerd Wöstenkühler).
SECTION 2. DIGITAL SIGNAL PROCESSING (DSP).
142. Z-transforms (Armar Bousbaine).
143. DFT and FFTs (Gerd Wöstenkühler).
144. DSP Chip Sets (Iain Paterson-Stephens).
145. DSP Tools (Iain Paterson-Stephens).
146. Principles of DSP Hardware Design (Iain Paterson-Stephens).
147. Ideal Digital Filters Approximation (Joseph McGhee).
148. General Performance of the Digital Filter (Joseph McGhee).
149. Low-, High-, and Band-pass Digital Filters (Joseph McGhee).
150. Finite Impulse Response (IIR) Digital Filters (Joseph McGhee).
151. Finite Impulse Response (FIR) Digital Filters (Joseph McGhee).
SECTION 3. COMPUTERS IN MEASURING SYSTEMS.
152. Fundamentals of the Stored Program Digital Computer (Joseph McGhee).
153. Single Address Instruction Microcomputer (Joseph McGhee).
154. Internal Operation of the Microprocessor (Joseph McGhee).
155. External Operation of the Microprocessor (Joseph McGhee).
156. Memory Management in the Microprocessor (Joseph McGhee).
157. Data Acceleration in Computers (Joseph McGhee).
158. Microcontroller Systems (Joseph McGhee).
159. Designing and Building Software for Measuring Systems (Joseph E. Kasser).
SECTION 4. INTELLIGENT MEASURING SYSTEMS.
160. Smart Sensor System Features (Peter H. Sydenham).
161. Knowledge-based Systems (Dietrich Hofmann).
PART 10. ELEMENTS: D – MEMS.
SECTION 1. MICRO ELECTRO MECHANICAL SYSTEMS (MEMS).
162. Principles of MEMS (Janusz Bryzek).
163. Uses and Benefits of MEMS (Janusz Bryzek).
164. Principles of MEMS Actuators (Janusz Bryzek).
PART 11. ELEMENTS: E – COMMUNICATION IN MEASURING SYSTEMS.
SECTION 1. DISTRIBUTED AND NETWORKED MEASURING SYSTEMS.
165. Introduction to Networked Instrumentation (Joseph McGhee).
166. Instrument Interconnection (Joseph McGhee).
167. Asynchronous and Synchronous Interface Protocols (Joseph McGhee).
168. RS 232 and EIA/TIA 232 Serial Interface (Joseph McGhee).
169. Voltage and Current Loop Transmission (Joseph McGhee).
170. IEEE-488 Instrumentation Bus (Joseph McGhee).
171. Local Area (LANs) and Wide Area Networks (WANs) (Joseph McGhee).
172. Fieldbus Systems (Halit Eren).
173. Scheduling Systems (Emil Michta).
PART 12. ELEMENTS: F – SIGNALS AND NOISE.
SECTION 1. NOISE AND INTERFERENCE.
174. Typical Signals Arising in Measurement (Eugen Georg Woschni).
175. Comparison of Analog and Digital Signal Handling (Joseph McGhee).
176. Signals and Signal-t0-noise Ratio (Richard Burdett).
177. Grounding and Shielding (Kim R. Fowler).
178. Noise Matching and Preamplifier Selection (Richard Burdett).
179. Input Connections; Grounding and Shielding (Richard Burdett).
SECTION 2. SIGNAL RECOVERY IN THE PRESENCE OF NOISE.
180. Bandwidth Reduction of Baseband DC Signals (Richard Burdett).
181. Amplitude Modulated Signals: The Lock-in Amplifier (Richard Burdett).
182. Boxcar and Signal Averages (Richard Burdett).
183. Correlators in Signal Extraction (Richard Burdett).
184. Photon Counting (Richard Burdett).
185. Pulse Height Discrimination, Ratemeters and Pileup (Richard Burdett).
186. The Family of Signal Recovery Methods (Richard Burdett).
PART 13. COMMON MEASURANDS.
SECTION 1. FLOW MEASUREMENT.
187. Flowmeter Selection and Application (Michael Reader-Harris).
188. Differential Pressure (DP) Flowmeters (Michael Reader-Harris).
189. Basic Principles of Flow Measurement (Richard Thorn).
190. Calibration and Standards in Flow Measurement (Richard Paton).
SECTION 2. DISPLACEMENT AND ANGLE MEASUREMENT.
191. Displacement and Angle Sensors Performance and Selection (Halit Eren).
192. Strain Sensors (Peter H. Sydenham).
193. Specialty Displacement and Angle Sensors (Halit Eren).
194. Large-scale Metrology (Stephen Kyle).
SECTION 3. TEMPERATURE MEASUREMENT.
195. Characteristics of Temperature Measurement (Joseph McGhee).
196. Thermocouple Temperature Sensors (Jacek Kucharski).
197. Metalic Resistance Temperature Detectors (RTDs) (Dietrich Hofmann).
198. Calibration and Standards in Temperature Measurement (D.R. White).
SECTION 4. TIME AND FREQUENCY.
199. Characteristics of Time and Frequency Measurement (Michael A. Lombardi).
200. Calibrations and Standards in Time Measurement (Michael A. Lombardi).
SECTION 5. ELECTRICAL QUANTITIES.
201. Voltage Measurement (Halit Eren).
202. Current Measurement (Halit Eren).
203. Resistance Measurement (Halit Eren).
204. Capacitance and Inductance Measurement (Consolatina Liguori).
SECTION 6. VELOCITY AND ACCELERATION.
205. Theory of Vibration Measurement (Peter H. Sydenham).
206. Practice of Vibration Measurement (Peter H. Sydenham).
207. Acceleration Measurement (Peter H. Sydenham).
208. Amplitude and Velocity Measurement (Peter H. Sydenham).
SECTION 7. CHEMICAL PROPERTIES.
209. Characteristics of Chemical Measurements (Peter H. Sydenham).
210. Optical Transducers for Chemical Measurements (Ashutosh Sharma).
211. Mass Spectrometry (Peter H. Sydenham).
212. Chromatography (Brett Paull).
213. Electrochemical Measurements (David Davey).
PART 14. TEST AND EVALUATION.
SECTION 1. MEASUREMENT TESTING SYSTEMS.
214. Accelerated Testing (Patrick D.T. O’Connor).
215. Automatic Test Systems (Patrick D.T. O’Connor).
216. Test Facilities (Patrick D.T. O’Connor).
217. Instrument Evaluation (Steve Cork).
Subject Index.
Sophistication of Measurement and its Body of
1:
Knowledge
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
involved provide data input to the brain, coding it with
1 Sophistication of Measurement as the Degree meaning to suit the required task. Two people viewing a
of Science 5 plant leaf, for example, see the same object with simi-
2 Measurements and the Body of Knowledge 6 lar senses, yet both could ‘see’ quite different attributes.
Related Articles 9 Latent information available has begun to be filtered at
this stage.
References 9
The assembled data is then sorted and classified accord-
ing to various kinds of similarities to detect differences.
Each group forms a crude measurement standard of compar-
1 SOPHISTICATION OF MEASUREMENT ison for the others. This process can often be continued until
AS THE DEGREE OF SCIENCE advanced knowledge is established without using measur-
ing instruments. Linnaeus (1707–1778) was able to make
The decision of whether to use existing, or to create new, a major contribution to botany by introducing his binomial
measuring instruments in the study of a subject comes classification system (see Figure 2). Darwin’s On Origin of
after measurable variables have been identified. The process the Species by Means of Natural Selection of 1859 has been
is, in the physical sciences, usually considerably easier to recognized as probably the greatest generalization yet –
realize than in many areas of the empirical sciences. Many although gene mapping is taking that over. It was made
stages of prior reasoning precede such a decision: this is not from vast quantities of data that were all assembled with
always recognized, especially in engineering. The process little use of measuring instruments to enhance man’s natu-
can be depicted by the chart given in Figure 1. ral senses.
Knowledge seeking begins presumably because of certain At some stage, this qualitative form of science can
inquisitive features of man’s makeup that stirs up interests be subjected to increasingly more quantitative methods.
in directions that seem to have more relevance than others. Attributes of the various classes became apparent in a
The processes involved are complex, and, as yet, not way that allows instruments to be applied that give nat-
adequately known. Paradoxically, it seems that a great deal ural sensing, greater sensitivity, and greater power to move
of knowledge is used in a very general way from the onset from a qualitative mode into the quantitative measure-
to choose candidate paths of action to follow to gain the ment mode. More detailed knowledge becomes available
knowledge sought. as measurements produce data that is referred against more
This process, which involves the cognitive elements adequate, precise, and accurate standards. Thus, it is that
of sensation, perception, apperception, advises the knowl- physical measuring instruments applied as the degree of
edge seeker that certain information is more relevant for science, which is reflected by the degree of quantification
study than other data. It appears that the biological senses used, is improved. This line of reasoning also makes it
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
6 Foundations of Measuring
Total machine sensing suitable for autocontrol of measuring instruments are used varies widely. In some
studies, they are needed at the very beginning.
Finkelstein (1975) sums up the situation in this way:
Increasing application of measurement hardware indicates
Use of physical apparatus to enhance senses
Measurement presupposes something to be measured,
and measures have to be conceived and sought before they
Concentrated filtering of data with human senses can be found in experience. Both in the historical devel-
opment and logical structure of scientific knowledge the
formulation of a theoretical concept, or construct, which
degree of science used
defines a class of entities and the relations among its mem-
Continual feedback exists to all stages
I Groups ‘n ’
bers, providing a conceptual interpretation of the sensed
world, precedes the development of measurement proce-
dures and scales. It is necessary for instance to have some
concept of ‘degree of hotness’ as a theoretical construct,
Classification into groups interpreting the multitude of phenomena involving warmth,
before one can conceive and construct a thermometer.
As measurement procedures are developed, and knowl-
Collection of possibly useful data edge resulting from measurement accumulates, the concept
of the measured class becomes clearer and may to a sub-
stantial extent become identified with the operational pro-
Feeling of need to know cedures underlying the measurement process.
In some cases the concept of an entity arises from the
discovery of mathematical invariances in laws arrived at
by measurement, and the entity is best thought of in such
mathematical terms, but in general one attempts to arrive at
System under study some qualitative conceptual framework for it, if possible.
As the subject matter becomes better known and enables
Figure 1. Simplified hierarchy of application of measuring ins- unattended sensing by an observer, as needed for control or
truments in the study of a problem. monitoring purposes, the use of measuring instruments to
enhance human senses may not be appropriate. Hardware
sensors then totally replace man’s senses.
It is logical, therefore, to expect all of man’s endeav-
ors that require measurements to be made (most of them!)
to trend steadily toward greater use of measuring instru-
ments. Certainly, time has proven this to be so. But this is
also a consequence of man’s method of survival on earth.
Unlike lower animals, man has the ability to modify his
environment to suit his biological structure. He does this
usually by the use of technological developments, which
rarely operate in the same way as natural equivalents or
are made with the same materials. A comparison of natu-
ral and man-made vision sensors is given in Figure 3. The
knowledge man possesses is being built up of a component
about the natural world plus a component about the struc-
Figure 2. Linneaus resting after a botanical ramble. He devised tures that man has created. Measuring instruments are the
the binomial classification system now used, reporting it in ‘Sys- means by which man’s creations operate and these too are
tema naturae’, 1758. (Copyright Uppsala University.) creations of man. The relationship between measurement
and knowledge has been explored (Sydenham, 2003).
vital that appreciation of the qualitative stages preceding
proper measurement and the instruments that evolve are
understood. 2 MEASUREMENTS AND THE BODY OF
This sentiment is not new as the famous Lord Kelvin KNOWLEDGE
quote tells us. The words of Westaway (1937) are in sym-
pathy with this. The concept expressed in Figure 1 is The sum total of knowledge is termed the body of knowl-
simplified: in practice, the stages at which hardware forms edge. As knowledge is a characteristic of man, not of
Sophistication of Measurement and its Body of Knowledge 7
S R
RS
D
X B
z E
Y K G A
K F
L
Q
E F G D
RS
R
z W
(a)
Mosaic
Signal
plate
Collector
To video
amplifier
Electron
gun Signal-plate
load resistor
Yoke
(b)
Figure 3. Man’s creations generally use different materials and techniques as do natural systems. Here, imaging sensors are contrasted
(a) Longitudinal section of eye (Reproduced from Cyclopaedic Science, Pepper J.B., (1874), Copyright Frederick Warne) and (b) RCA
iconoscope – early form of television camera tube from Kloeffler (1949) (Courtesy RCA Ltd, USA).
existence, it began at zero magnitude and grew with time. recoverable in an easier manner than that by which it was
No method has yet been devised to measure its magnitude first generated!
in objective ways but it clearly is enlarged continuously To retrieve knowledge, it is grouped into convenient clas-
with the passage of time. sifications. Convenience is a term in which time of action
It is formed of two groups: that about the natural world is most important. The memory span of man, especially
and that about the unnatural systems created by man. Man’s short term, is very limited, so it has been suggested (Har-
creation grows, the natural world changes; the extent of the man, 1973) that major groupings usually total around seven.
latent information available for conversion into knowledge These in turn are subdivided, giving the various epistemo-
therefore grows continuously. logical groups.
As the body of knowledge grew, various workers of the Measurements assist in gaining knowledge and knowl-
past tried to summarize all that was known. Today, that edge, in turn, assists new forms of measurements to be
must be recognized as an almost hopeless task. Collectively, conducted. A closed-loop mechanism can be observed in
all knowledge must be stored in a manner whereby it is the development of measurements; Figure 4 depicts this.
retrievable. The danger of converting latent information Over the past few decades, the trend toward recogni-
back into another form of latent storage via the knowledge tion of the interdisciplinary studies that replaced the spe-
conversion state is real; what lies in the literature is not all cialisms that came to us previously has highlighted the
8 Foundations of Measuring
Measurement techniques flow back
Cross-discipline for reuse and modification
use of measurement
Techniques applied
Discipline 1
Information
converted to Discipline 2
Information
Untapped knowledge via
latent flow measurement plus
information other skills
Discipline ‘n ’
The system Academic endeavour Disciplines systematizing Application of
under study (research and teaching) knowledge to suit the times knowledge
Figure 4. Relationship of measurement principles in ordering the body of knowledge.
Knowledge of Knowledge of
Natural systems Man-made systems
Growth
Latent information
(as rise) Genesis of man
yet to be converted
into knowledge of coded (no man made
knowledge systems existed)
Breadth of knowledge
Hydraulics
Breadth increases
mechanics
with passage Ancient times
+ Others (man began to generate
of time
unique systems)
(7 liberal arts)
Mathematics
Philosophy Mechs. Optics
Hyd.
Natural
philosophy + Others 1600s
Biological
sciences
Humanities
Physical
sciences
Numerous similar
measurement subsets
in applications Present
Social sciences
Figure 5. Epistemological mountains in the two plains of human knowledge. Measurement techniques are now duplicated on most
contemporary mountains.
fact that not only does such a feedback process exist but The Dewey cataloging system gives librarians a set of
it is also often duplicated (a needless waste of effort, numerical codes, each having a linguistic description of
therefore) and is often cross-fertilized between epistemo- what subject matter each number represents. Of over 40 000
logical groups. numerical assignments, some 600 clearly relate to the
Sophistication of Measurement and its Body of Knowledge 9
measuring process. These are distributed widely over the cation, Volume 1; Article 4, Economic Considerations
whole body of knowledge, as classified by that system. of Measurement, Volume 1; Article 5, Humans in
Pictorially, this means that most clusters of knowledge the Real World, Volume 1; Article 6, Substructure of
possess subclusters concerned with measurement method Human–Machine Systems, Volume 1.
as depicted in Figure 5.
At present, information scientists – those people that
work on the storage, coding, and retrieval of knowledge –
REFERENCES
consider that the major clusters are changing to reflect the
interdisciplinary attitudes. New clusterings are emerging, Finkelstein, L. (1975) Fundamental Concepts of Measure-
one which may well be that of the relatively new discipline ment: Definition and Scales. Measurement and Control, 8,
of measurement science, the pursuit of means to convert 105–111.
latent information into meaningful knowledge by rigorous Harman, G. (1973) Human Memory and Knowledge, Greenwood
and objective procedures of philosophy and practice. Press, London.
Sydenham, P.H. (2003) Relationship between Measurement,
Knowledge and Advancement. Measurement, 34(1), 3–16,
RELATED ARTICLES Special Issue on Measurement foundations.
Westaway, F.W. (1937) Scientific Method: Its Philosophical
Article 2, Organization of Instrument Science, Vol- Basis and its Modes of Application, Hillman-Curl, New
ume 1; Article 3, Measures and Metrics; Their Appli- York.
2: Organization of Instrument Science
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
‘ENERGY transfer from’ a ‘process’ using ‘SYSTEMS,’
1 Definition of Instrument Science 10 which are made up of ‘components or TRANSDUCERS ’
2 The Need and Starting Point for Ordering in forming a ‘STRUCTURE or network’.
Instrument Science 11 A definition that encompasses all of these ideas is given
3 How Instrument Science is Organized 12 in Figure 1. The systemic nature of measuring instruments
demands a holistic approach in design and analysis. It is
4 Orders of Classification 12
apparent that the ordering of information machines depends
Related Articles 14 upon the holistic relations among specific sensor struc-
References 14 tures performing diverse functions within different energy
domains for the acquisition, capture, communication, or dis-
tribution of information in a variety of signal forms.
The diagrammatic summary definition given in Figure 1
1 DEFINITION OF INSTRUMENT
is based upon the functions performed by measurement
SCIENCE systems, the structures that allow them to perform the
function, and the energy form from which the informa-
A science is an organized body of knowledge (Finkelstein, tion is acquired. It may be regarded as the study of the
1994). What then is Instrument Science? To answer this methods and techniques of extending the human abilities
question, we must define what an instrument is. When to handle information using information machines. Since
posing the question, ‘what is an instrument?’ (McGhee, information is predominantly carried by signals, measure-
Henderson and Sankowski, 1986), most people have a vis- ment is concerned with the acquisition, handling, analysis,
ceral feeling for the answer. According to the McGraw-Hill and synthesis of signals in measuring instruments. It may
Encyclopaedia of Science and Technology, an instrument also be considered as the measurement analogy of data
is a SYSTEM, which refines, extends, or supplements the communications.
HUMAN faculties of sensing, observing, communicating, To assist with the generalizations that make measurement
calculating, controlling, and perceiving. In other words, scientific, it is essential to develop a unified metrological
instruments are human-made elements embedded within description of every constituent component making up a
human-machine systems, which help humans to acquire measurement system. A unified approach allows the eval-
information, by the process of sensing, and to handle data, uation of the metrological characteristics of each element.
by performing information handling operations. Using this Thus, the formation and analysis of all contributory fac-
definition as the key, an implicit use of taxonomy led to tors, and in particular, the measurement errors can be per-
the proposal that ordering in instrumentation should involve formed (Solopchenko, 1994). Signals, which are acquired
functional and structural reticulation (McGhee, Henderson using various forms of sensors, are handled using diverse
and Sankowski, 1986). This statement is similar to another forms of metrological components. These may be con-
definition by Peter Stein (1969) who asserted that Mea- ditioners, amplifiers, and filters used in conjunction with
surement combines ‘INFORMATION transfer about’ and suitable multiplexing methods.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Organization of Instrument Science 11
Measurement systems
Refine, extend and supplement the human senses
Perform FUNCTIONS ‘What they do ’
for capturing INFORMATION
The entity characterized by ENERGY flow for
Acquiring INFORMATION ‘Why they do ’
capturing INFORMATION
The physical variable to be handled by generation
Carried by SIGNALS ‘When they do ’
and processing operations
The physical domain of the captured
Extracting ENERGY ‘Way they do ’
INFORMATION classified by COMETMAN
From PROCESSES ‘Where they do ’ The principal source of the INFORMATION
Possess STRUCTURE ‘How they do ’ The physical means of FUNCTIONING
Using SENSORS ‘While they do ’ The principle element of their STRUCTURE
Figure 1. A substantive definition of measurement.
When instruments, which have the primary structure of Thomson, 1926) using objective methods to ensure
systems, are viewed from this position, the field of Sys- that the ordering is justifiable. Such schemes of
tems Science and Engineering (M’Pherson, 1980, 1981; classification have been compared to nominal scales of
Sandquist, 1985), with its related disciplines associated with measurement using an algebraic formulation (Watanabe,
large-scale systems, must play an important part in their 1996). The following quotation (Knight, 1986) indicates
exposition. This systems approach, which possesses holis- the fundamental importance of classification in all of the
tic or totality features, offers a number of advantages. A applied sciences:
principal benefit places instruments within a hierarchy of
both systems and machines by structure, function, energy We are apt to think of classification as a sort of ‘natural
form, and information. McGhee, Henderson and Sankowski history stage’ through which all sciences pass in their
youth before they grow into something handsomer, more
(1986) have stated that these aspects are revealed by the
mathematical and explanatory. . . classification is a highly
methods of reticulation or subdivision. As it happens, retic- theory-laden activity. . .. What one thinks one is classifying
ulation also reveals the places occupied by other types may make a big difference to the system of classificatory
of subsystems within this hierarchy. Thus, advantages are categories one uses.
accrued by using this approach in the study of instru-
mentation. Commencing from this standpoint, the systems It is apparent that classification is of basic importance for
approach is essential for the study of instrumentation. Some all activities in the applied sciences.
broad principles of Systems Engineering for instrumenta- It has been noted that a taxonomy of Instrument Science
tion are adapted for the boundary view of human–machine will be erroneous if it is based upon its ends (McGhee
systems in Article 68, Setting the System Boundaries, and Henderson, 1993) as this will only lead to a cat-
Volume 2. aloging of instruments. Indeed, only by organizing the
constitution of the topic on the basis of contributory
disciplines can Instrument Science be arranged accord-
2 THE NEED AND STARTING POINT FOR ing to its basic nature and inherent characteristics. Con-
ORDERING IN INSTRUMENT SCIENCE sidering the nature and scope of the disciplines con-
stituting the taxonomy, analysis, design, and utilization
Every field of scientific activity requires organization of instruments and instrument systems provides a clear
or ordering. An essential starting point in the ordering view of the contributory disciplines of Instrument Sci-
of Instrument Science is the application of a relevant ence (Finkelstein, 1994; Finkelstein and Grattan, 1993,
taxonomy (Flint, 1904; Durand, 1899; Broadfield, 1946; 1994; Measurement, 1994; Sydenham, 1982, 1983; Syden-
Körner, 1970; Knight, 1986; McGhee and Henderson, ham and Thorn, 1992) within Instrumentation and Measure-
1991; McGhee et al., 1996; McGhee and Henderson, 1993; ment Technology (I&MT).
12 Foundations of Measuring
3 HOW INSTRUMENT SCIENCE IS studied deeply until the nineteenth-century French philoso-
ORGANIZED pher, Durand (De Gros), examined its constitution. Thus, a
clear distinction is drawn between the ordered organization
Instrument science must be holistic by always using the of the theory of Taxinomy (its original spelling) itself and its
‘whole-life-whole-system’ approach characterizing the SYS- principal applications in a specific field. It has been claimed
TEMS ENGINEERING method (M’Pherson, 1980; Mc- that the word Taxonomy (from the ancient Greek taxis
Ghee, Henderson and Sankowski, 1986; Sandquist, 1985). meaning order) was first used by the seventeenth-century
Thus, it is seen that instruments and instrument systems per- Swiss botanist Augustin Pyrame de Candolle (1778–1841).
form a diversity of information handling functions allowing What is the nature and scope of taxonomy or classifica-
the acquisition, capture, communication, processing, and tion science? In the view of Durand, the most elementary
distribution of information about the states of equilibrium form of all classification is the series that depends upon
and motion of solids, liquids, gases, and their constituent the increase or decrease of some variable of the scheme of
systems using a variety of physical sensing structures in dif- ordering. Hence, any legitimate scheme of instrument clas-
ferent energy forms. McGhee and Henderson (1991) have sification must ensure that all of its divisions are always
suggested that this is the starting point, not only for ordering determined by one common principle. Instrument classifi-
in Instrument Science but also as the fundamental context cation will thus be erroneous if it is based upon its ends, as
for ordering in all of the applied sciences. this merely leads to a catalog of different kinds of instru-
The question then arises as to how the science of mea- ments. Rather, instrumentation should always be arranged
surement should be organized into identifiable bodies of according to its basic nature, its inherent characteristics, and
knowledge. A method for the organization of knowledge not upon anything lying outside itself. In other words, the
in the biosciences called Taxonomy or Classification Sci- science of classification in instrumentation is not about the
ence provides the answer to this question. This method can sum of the ends of instrumentation but rather about coordi-
be adapted for the organization of measurement. Obser- nating the science of instruments in such a way as to give
vation and recording are the embodiment of the scientific it an organized or systematized structure.
method, which is of profound importance in the under-
standing and utilization of the physical universe and its
resources. This aim is achieved through the measurement 4 ORDERS OF CLASSIFICATION
of the states of equilibrium and motion of solids, liquids,
gases, and the systems they constitute (McGhee, Hender- The significant contribution Durand made to the science
son and Sankowski, 1986). Instruments are the means of taxonomy was the proposal that there are four princi-
by which these human faculties may be improved and ple orders or problems of classification. These orders are
supplemented (Finkelstein, 1994). However, the acquisition summarized in Table 1. In the First Order, described as
of information, or, more generally knowledge, requires Generality or Resemblance, is embodied what many other
some process of ordering or organization. In the case theorists of classification have called the ‘likeness’ of one
of instrumentation, this ordering of information machines thing with another thing. The thing concept is fundamental
depends upon the holistic relations between various instru- to the whole of categorical ordering, not just in bioscience.
ments. The basic theoretical mechanism, which allows It is also important in the earth sciences (Von Engelhardt
the organization, is the field of taxonomy or classifica- and Zimmermann, 1988) for the classification of miner-
tion science. Although this science is well known in the als, in technology transfer (Zhao and Reisman, 1992), and
biosciences, it is not so well known, or for that matter in KNOWLEDGE ENGINEERING (KE) (Chandrasekaran
understood or applied, in the engineering sciences. This and Goel, 1988; Gomez and Segami, 1991; Mill and Rada,
opinion has been expressed on a number of occasions in 1990; Yasdi, 1991). Hence, this concept also has central
the references quoted in McGhee, Henderson and Syden- importance in instrumentation. Likeness, of course, is that
ham (1999). It is well worthwhile to provide some basic relation between several concrete things that unites them.
information on the nature and scope of taxonomy for use Thus, the application of classification by zoologists and
in measurement. The systemic nature of instruments implies botanists in the discrimination between genera and species
a holistic approach in their ordering. is a good example of the way in which the problem of
Since the time of Plato and Aristotle, many attempts generality and resemblance is approached.
have been made to organize the sciences into hierarchical In taxonomy, there is an important tendency to group
groupings. A scientific approach for the ordering of sci- things on the basis of their Composition or Collectivity.
ence is provided by TAXONOMY . Although this science has Durand distinguished this as the Second Order of taxon-
been used implicitly by bioscientists for centuries (Daly and omy. While this order is concerned with the relationship of
Linsley, 1970), its intrinsic rules and principles were not the part to the whole and vice versa, the Third Order of
Organization of Instrument Science 13
Table 1. A summary of the four orders or problems of taxonomy.
Taxonomy, the science of classification
(Putting things in a scientific order )
Problem or Order Definition and Aspects Comment
Generality or resemblance 1. Concerned with the likeness of separate Also called the Metaphysical Order because
things terms are concerned with theoretical or
fictitious things
2. Likeness is that relation between things
that unites them
3. The thing concept is fundamental to all
Categorical Ordering (i.e. Taxonomy)
Composition or collectivity 1. Concerned with the relationship of a part All other orders are concerned with the actual
of a thing to the whole thing things to be classified
Hierarchy 1. Concerned with the relation between heads Related to the order of composition/collectivity,
or central members of groups of things especially in the places occupied in each order
relative to other things of the same order
Genealogy or evolution 1. Concerned with the kinship of one thing Hinges upon notions of kinship by
with some other thing relationships of
• ascent
• descent
• collaterality.
taxonomy, called Hierarchy, takes account of the relation instrumentation. Phenetic discrimination uses similarity and
of rank between the heads or central members of groups of difference in form or physical feature, while phyletic tech-
things. In their turn, these are related in the order of com- niques are based upon evolutionary criteria.
position, but address each concrete thing in the assessment A summary of taxonomy for instrumentation (McGhee
of the place it occupies in each order relative to the other and Henderson, 1989) points out that it has three objectives
constituents of the same order. Perhaps the most important and three functions that emphasize its importance. Thus, the
Fourth Order in Durand’s theory of taxonomy, especially three objectives of classification are:
in bioscience, is that known as Genealogy or Evolution.
1. the concrete discrimination between different things;
This order hinges upon the notions of kinship through the
2. the consensus regarding standards for the principles of
relations involved in the characteristics of ascent, descent,
description;
and collaterality. As with the orders of Composition and
3. the bringing of order or systematization.
Hierarchy, Genealogy and Evolution are also concerned
with the actual objects or events that are to be classified. Similarly, the three functions of classification should
Although there have been minor developments of this the- allow
oretical constitution of taxonomy, it is still fair to say that
the basis laid by Durand has not been significantly altered. 1. the organization of the means of communication and
As this theory of taxonomy was formulated in the context retrieval of the descriptions used;
of bioscience, it requires modification before being applied 2. the acquisition of new information in the extension of
to instrumentation. descriptions;
Another important aspect of taxonomy is the develop- 3. the highlighting of unifying factors between entities
ment of a system of nomenclature, which is unambigu- without diminishing the importance of any existing
ous. In bioscience, the binomial nomenclature is due to differences.
the eighteenth-century Swedish botanist, Carolus Linnaeus The materials of taxonomy in Instrument Science are the
(1707–1778). For example, in plant kingdom classification, diverse types of instruments and their operating principles.
the first category of the ordering is called a division. This Assembling the various instrument types is the main activ-
is followed by subdivision followed by class, order, fam- ity of classification in Instrument Science because it allows
ily, genus, species, and subspecies. It seems logical and the possibility for further study. The grouping of instru-
convenient to use the same ordering for machine king- ments from the lowest levels of sensors into progressively
dom grouping, although it may cause some controversy. larger groups so that a hierarchical ordering by function,
Adapting the basic phenetic and phyletic methods used by structure, and energy form, constitute the final ingredients
bioscientists allows functional and structural grouping in of discrimination and ordering in Instrument Science.
14 Foundations of Measuring
RELATED ARTICLES McGhee, J. and Henderson, I.A. (1993) Current Trends in the
Theory and Application of Classification to Instrumentation and
Article 1, Sophistication of Measurement and its Body Measurement Science, in State and Advances of Measurement
of Knowledge, Volume 1; Article 6, Substructure of and Instrumentation Science, Proc IMEKO TC1/TC7 Collo-
quium (eds L. Finkelstein and K.T.V. Grattan), City University,
Human–Machine Systems, Volume 1; Article 7, Intro- London (p. 32).
duction to Measurement Theory and Philosophy, Vol-
McGhee, J., Henderson, I.A. and Sankowski, D. (1986) Functions
ume 1; Article 14, Hierarchical Aspects of Measurement and Structures in Measurement Systems: A Systems Engineer-
Systems, Volume 1; Article 22, Characteristics and The- ing Context for Instrumentation. Measurement, 4(3), 11–119.
ory of Knowledge, Volume 1; Article 63, Outline of Sys- McGhee, J., Henderson, I.A., Kulesza, W. and Korczynski, M.J.
tems Thinking, Volume 2; Article 104, Models of the (1996) Scientific Metrology, ISBN 83-904299-9-3, printed by
Measurement Process, Volume 2; Article 107, Principles A.C.G.M. LODART, Lodz.
of Sensor Science, Volume 2. McGhee, J., Henderson, I.A. and Sydenham, P.H. (1999) Sensor
Science–Essentials for Instrumentation and Measurement Tech-
nology. Measurement, 25(2), 89–113.
REFERENCES
Measurement, 14(1), (1994) special issue on Measurement and
Instrumentation Science.
Broadfield, A. (1946) The Philosophy of Classification, Grafton
and Co., London. Mill, H. and Rada, R. (1990) Regularity: Generalising Inheritance
Chandrasekaran, B. and Goel, A. (1988) From Numbers to Sym- to Arbitrary Hierarchies, in Proceedings of 2nd International
bols to Knowledge Structures: Artificial Intelligence Perspec- Conference on Tools Artificial Intelligence Washington D.C.,
tive on the Classification Task. IEEE Transactions on Systems, (p. 635).
Man and Cybernetics, 18(3), 415. M’Pherson, P.K. (1980) Systems Engineering: An Approach to
Daly, H.V. and Linsley, E.G. (1970) Taxonomy, in Encyclopaedia Whole-System Design. Radio and Electronic Engineering, 50,
of the Biological Sciences, 2nd edn (ed. P. Gray), Van Nostrand 545–558.
Reinhold, New York (p. 920). M’Pherson, P.K. (1981) A Framework for Systems Engineering
Durand (De Gros), J.P. (1899) in Aperçus de Taxinomie Générale Design. Radio and Electronic Engineering, 51, 59–93.
(ed. F. Alcan), Paris. Sandquist, G.M. (1985) Introduction to System Science, Prentice
Finkelstein, L. (1994) Measurement and Instrumentation Sci- Hall, Englewood Cliffs, NJ.
ence – An Analytical Review. Measurement, 14(1), 3–14. Solopchenko, G.N. (1994) Formal Metrological Components of
Finkelstein, L. and Grattan, K.T.V. (eds) (1993) State and Ad- Measuring Systems. Measurement, 13, 1–12.
vances of Measurement and Instrumentation Science, Proc of Stein, P.K. (1969) The Engineering of Measurement Systems.
IMEKO TC1/TC 7 Colloquium, City University, London. Journal of Metals, 21, 40.
Finkelstein, L. and Grattan, K.T.V. (1994) Concise Encyclopae- Sydenham, P.H. (ed.) (1982) Handbook of Measurement Sci-
dia of Measurement and Instrumentation, Pergamon, Oxford. ence, Vol. 1 Theoretical Fundamentals, John Wiley & Sons,
Flint, R. (1904) Philosophy as Scientia Scientiarum and A History Chichester.
of Classification of the Sciences, William Blackwood & Sons, Sydenham, P.H. (ed.) (1983) Handbook of Measurement Science,
Edinburgh. Vol. 2 Practice Fundamentals, John Wiley & Sons, Chichester.
Gomez, F. and Segami, C. (1991) Classification Based Reasoning.
Sydenham, P.H. and Thorn, R. (eds) (1992) Handbook of Mea-
IEEE Transactions on Systems, Man and Cybernetics, 21(3),
surement Science, Vol. 3 Elements of Change, John Wiley &
644.
Sons, Chichester.
Henderson, I.A. and McGhee, J. (1993) Classical Taxonomy: An
Thomson, A.J. (1926) Introduction to Science, Williams & Nor-
Holistic Perspective of Temperature Measuring Systems and
gate Ltd, London.
Instruments. Proceedings of IEE-A, 140(4), 263.
Knight, D. (1986) Physics and Chemistry in the Modern Era, in Von Engelhardt, W. and Zimmermann, J. (1988) Theory of
The Physical Sciences Since Antiquity (ed. R. Harre), Croom Earth Science, Cambridge University Press, Cambridge, MA
Helm, Beckenham. (p. 102).
Körner, S. (1970) Categorical Frameworks, Basil Blackwell, Watanabe, H. (1996) Theory of Classification of Objects by
Oxford. Predicates. Measurement, 18(1), 59–69.
McGhee, J. and Henderson, I.A. (1989) Holistic Perception in Yasdi, R. (1991) Learning Classification Rules from Database in
Measurement and Control: Applying Keys Adapted from Clas- Context of Acquisition and Representation. IEEE Transaction
sical Taxonomy. IFAC Proceedings of Series, (5), 31. on Knowledge and Data Engineering, 3(3), 293.
McGhee, J. and Henderson, I.A. (1991) The Nature and Scope of Zhao, L. and Reisman, A. (1992) Towards meta research on
Taxonomy in Measurement Education. ACTA IMEKO XII, 2, technology transfer. IEEE Transaction on Engineering Manage-
209. ment, 39(1), 13–21, 103.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
3: Measures and Metrics; Their Application
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
• Is it being done appropriately?
1 Measures Overview 15 • Is the result expressed appropriately?
2 The Measurement Situation 15
3 Measures and Metrics 16 2 THE MEASUREMENT SITUATION
4 Terms 17
5 Some Metrics 18 Throughout the recorded history of man, there has existed
6 Forms of Cognitive Entity 19 recognition of the connectivity between measurement and
7 The Scientific Process 20 the acquisition of knowledge that, in turn, can be related
to the advancement of man in general – Sydenham (1979);
8 How to Apply Measures 21
Bud and Warner (1998).
9 The Measures Triangle and its Parameters 21 Measurement can be used to support two kinds of knowl-
10 Case Study of the Generation of Measures 22 edge gathering situations:
Related Articles 23
• Controlling a known situation
References 23
A temperature controller in a food storage container
uses the measurement value to switch the cooling on
and off as needed. Here, the physical process is well
1 MEASURES OVERVIEW
understood; the need is to control the flow of cooling as
Measurement is found everywhere; it seems to be a neces- the temperature varies.
sary part of human living (Klein 1975; Ellis 1973). It is the • Investigating a subject under research
process by which we seek to qualify and quantify an issue.
It is a key part in the generation of knowledge for that issue. The need here is to glean new knowledge. For example,
For example, in order to decide if the greenhouse watering a theory has been proposed that suggests a relationship
system needs to be turned on, the moisture content of the between two variables in an illness exists that would suggest
soil is measured, resulting in a number that is compared to a cure. A series of experiments is designed in which
a standard value to decide if it is needed. measurements are made under controlled conditions to
Measurement is not always well set up. The well- reveal if the relationship holds.
experienced measurement scientist or engineer will easily A key statement about the relationship between mea-
be able to point to the inefficient way in which much of surement and knowledge is that of Lord Kelvin. In 1883,
measurement activity is practiced. in a lecture at the Institution of Civil Engineers, he
We need to be clear about such questions as: stated:
• What is the purpose of the measurement? ‘In physical science a first essential step in the direction
• How does measurement advance the issue in question? of learning any subject is to find principles of numerical
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
16 Foundations of Measuring
reckoning and methods for practicably measuring some Hofmann (1999) makes the link between measurement
quality connected to it. I often say that when you can and practical needs in society.
measure what you are speaking about, and express it in
Yang and Butler (1997) approached the problem of
numbers, you know something about it; when you cannot
measure it, when you cannot express it in numbers, your creating a universal framework from the epistemological
knowledge is of a meager and unsatisfactory kind: it may perspective, suggesting it be modeled as a knowledge-
be the beginning of knowledge, but you have scarcely, in oriented system. They propose that an object-oriented
your thoughts, advanced to the stage of science, whatever model (Yang and Butler 1998) be used for representing
the matter may be.’ measurement systems.
This statement is expressing elements of the thinking
paradigm known as reductionism, the main method by 3 MEASURES AND METRICS
which we gather knowledge in the so-called hard sciences.
Reductionism appears to have come down to us from The nature of inquiry used to gather knowledge can be
the contribution of Descartes. In 1693, he stated in his very different across the various disciplines – see Brown,
Discourse on Method – see Hutchins (1952): Fauvel and Finnegan (1981). Mathematicians, scientists,
engineers, social science, management, and so on, do not
‘. . . . .divide each of the difficulties that I was examining
into as many parts as might be possible and necessary, in have the same belief systems and often use different ways of
order to best solve it.’ thinking to solve their problems. They, however, all make
measurements of some kind to support their knowledge-
He suggested that the human mind sorts out its problems development processes.
and finds solutions by breaking them down into succes- It often comes as a surprise to the ‘hard science’ trained
sively smaller elements, until the stage is reached where scientist or engineer that not all situations can make use of
they are adequately understood. reductionist techniques.
Descartes suggested four rules for ‘properly conducting Stumbling blocks for reductionists in accepting the softer
one’s reason’: sciences and humanities approaches are the:
• Avoid precipitancy and prejudice • apparent lack of sufficient rigor of understanding and
• Accept only clear and distinct ideas expression;
• Conduct orderly progression from the simple to the • use of many less familiar terms and words like
complex ‘paradigm’, ‘metaphor’, ‘holistic’, and so on;
• Complete analysis with nothing omitted. • inability to be as precise about ideas as are the laws of
physics;
This is the basis of the measurement methodology out- • inability of humanities practitioners to clearly identify
lined in Section 8. the parameters and relationships of their areas of work;
In addition to being used for the ‘hard science’ physical • lack of applicability of the reductionist approach – that
situation measurements, it is also used in the ‘soft science’ surely should be used; after all, it has and still is serving
situations to obtain qualitative knowledge where measure- much of science and engineering very well.
ment is vital to such situations – for example, audits of the
performance of people and processes. The humanities paradigm is known as the phenomenolog-
Studies on the general nature of measurement are avail- ical approach. Here, the observer does not metaphorically
able. A few are now selected to show the range of dismantle, by reductionism, the system of interest to sepa-
approaches taken. rate its subsystems and then build it up again after changes
Finkelstein has covered a large range of fundamental top- have been made. Instead, the humanities viewpoint is one of
ics. His paper, Finkelstein (1999), is a good summary of metaphorically getting inside the system of interest, insert-
how far the ideas have been taken in formal mathemati- ing intervention actions to see if current understanding is
cal terms. correct, and the ability to change the system as required. A
Sydenham (1979) is a review of the role of measurement, relevant branch of this is called the soft system methodology
which attempted to delve into the reasons and processes. (SSM), Checkland (1981).
The place of measurement in science is covered by In sharp contrast, reductionism requires all of the system
Kariya (1999) and IMEKO (1999). It gives a balanced of interest to be first bounded to form a closed system
overview of the hard science involved along with the that is then dismantled to be built up again in its new
necessary early stages of idea formulation and expression form. The sort of problem that does not lend itself to this
of what it is about as a process of learning. paradigm is one in which the boundaries of influence are
Measures and Metrics; Their Application 17
unclear, preventing the creation of an adequately closed 4 TERMS
system model.
There is also another reason why reductionism often fails Whatever process of measuring is being implemented, a
in the complex systems arena. Success in understanding confusing range of terms are used to describe the mea-
and problem solving is predicated by the assumption that sures used.
the solutions for the subsystems resulting from reticulation The ‘thing which is to be known’ within a measuring
can all be integrated back into the needed whole. One diffi- situation is today, in the engineering world, often called a
culty is that even slight variations in interface specification measurand .
of those subsystems parts can have a significant impact A commonly found general term for measures, used
on the performance of the whole – to the point where the extensively in the process performance arena, is the metric
performance of the new whole differs markedly from expec- (Blanchard and Fabrycky (1998); Sage and Rouse (1999)).
tations. This term is found, where a set of measures (metrics) are
The reductionist concept for problem solving is not established to collectively gain insight into how well the
totally accepted – it does have a severe philosophical pro- whole process is working. This term is not as frequently
blem. used in the physical sciences, for the word ‘metric’ there is
A fundamental difficulty is what philosophers call the associated with the metric system of units.
‘dual body’ problem. Behavior of the physical aspect of Another measures term often used in systems manage-
the human system is well explained by the laws and ment is the technical performance parameter (TPM) – this
rules of physics. The human mind, however, seems to is explained later.
behave quite differently. Its behavior defies reduction to Many other terms will be encountered that mean much
formal description and use of the same method of scientific the same thing – tracking variable or parameter, indicator,
investigation. index, score board value, and so on.
Methods of inquiry, and even the scientific process of Measure terms that have specific and different applica-
knowledge discovery, are not taught in most engineering tion include:
and science courses. A result has been the widening divide
• measure of effectiveness (MOE)
between the thinking styles of the Arts/Humanities and the
• measure of performance (MOP)
Sciences, existing on the modern university campus.
• system performance parameter (SPP)
Many myths about measuring exist – see Sage and Rouse
• technical performance parameter (TPP).
(1999) pg. 584–586 – some are:
Where these fit into a hierarchy of measurement is
• Measurement made with hard quantified measures will explained in Section 8.
lead to the soft issues also being understood – not so; The development and application of truly effective sets
soft systems are different from hard systems and need of metrics is a skilled task based much on experience in the
different approaches in their measurements. application area. It is easy to generate the measures for the
• Measurement is for bean counters and the data cannot be clearly evident physical measurements such as temperature,
translated into useful improvements – not so, provided speed, and load-carrying capacity. It is often not so easy to
it is done well, see Section 8. decide an effective parameter for more elusive, many to
• Measurement is about the past and is not relevant to one mapping situations, such as in setting up a measuring
the different future – not so; applications can mature system for the quality of a social reform program.
as projects change, by the application of sound and At the single-measurand level, seek to choose the mea-
relevant measurement. sure with best overall effectiveness. It will not always be
• Measurement encourages a box ticking culture – not obvious; the process involving the measure needs to be
necessarily so, provided it is done well and not using understood. For example, in jam making, a rapid change
simple-to-measure, yet nonuseful, data. in the pH is a far better indicator of when it is optimally
• Measurement stifles creativity – not so, as measurement cooked than is the viscosity of the mixture.
is about knowing about things in an objective manner. Setting up a truly effective metric is not always easy; sim-
• Measurement thwarts productive human activity – not ple ones are often chosen that, while providing a seemingly
so, if done appropriately. comfortable quantitative number, add little to the overall
• The more the measurement the better the productivity picture being sought. For example, the rate of progress of a
will become – not so, for again it is a matter of devis- software task could be measured as ‘lines of code completed
ing a good measuring system that truly addresses the in a unit time’ compared against the envisaged number of
requirements. lines used as the norm. This is, however, far too simplistic
18 Foundations of Measuring
as the quality of the code and the number of errors to be • Force
subsequently corrected can completely overwhelm the time • Temperature
used to prepare the code for the usable standard. • Viscosity
As the choice and use of metrics is based in considerable • Tensile strength
experience, a company will often be protective of its metrics • Strain
database and not release it to the general public for, over
and so on.
time, it develops to have intellectual property value.
In reality, it is the high-level measures that are of real
interest, physical variables being but a part of ‘many to
one’ mappings of measures. 5.2 General systems use
• Time to market
5 SOME METRICS • Time to completion
• Number of items produced
Thousands of metrics exist. A well-organized systems • Sales made
design operation will have a progressively updated database • Sales returns
of metrics that has been developed to suit its own kind • Defects rate
of industry. • Repair time
Unfortunately, these tend to not be developed in reusable • Mean time between failure (MTBF)
ways that would permit follow-on projects to extract them
and so on.
from a well-setup library. Also, they are often held in
confidence and tied into a project.
They mature as the staff uses them, and for this reason
alone the best way to develop effective ones is to ensure 5.3 Customer responsiveness
they are reused over projects in a controlled manner.
A measure stored in a metric database needs to have the The following are from Sage and Rouse (1999), pg 569.
following information recorded: These require many-to-one measurement mappings to arrive
at a measured quantity – see Section 8.
• Metric/measure name
• Symbol used to represent it • Product features added
• Acronym used, where applicable • Product quality
• Synonym usage explanation • Customer satisfaction
• Definition of its purpose • Speed of response to customers
• Brief description of its uses • Market expansion
• Use in multimeasures mapping sets • Product uniqueness
• Previous projects in which it has been used • Listening to customers
• Person who authored the entry • Customer visits
• Level of confidentiality assigned • Sales improvements
• Authorizing person • Innovation
• Persons who accessed it in past use. • Organizational acceptance to customer evolution
With so much to set up to ensure traceability, sound-
ness, and uniqueness, it is not surprising that good metric 5.4 Innovation measurement
databases are not readily available.
The following short collection of metrics is a motiva- Some of the lists provided in Sage and Rouse (1999), pg
tional starting point. 570 are now given:
• Number of innovative small parts
5.1 Physical measurands • Service innovations
• Number of pilots and prototypes
• Velocity • Number of benchmarked ideas adopted
• Time lapsed • Measures of word-of-mouth marketing
• Mass • Number of innovation awards.
Measures and Metrics; Their Application 19
5.5 Software development • Payload
• Subsystem x weight
• Lines of code • Number of crew needed
• Rate of completion of lines of code • Firing rate
• Efficacy of coder
• Error rate per 1000 lines of code and so on.
• Recursion time
• CPU needed
• Speed of execution of standard benchmark operation
6 FORMS OF COGNITIVE ENTITY
• Latency time
When measuring, it is important to differentiate between
• Number of branches
the terms data, information, knowledge, and wisdom when
• Compilation time
used in relation to knowledge gathering via measurement.
• Reset time
This issue is addressed in more depth in Article 13, Sig-
• Cyclometric complexity
nals, Information and Knowledge, and Meaning, Vol-
• Level of cohesion
ume 1; Article 23, Principles of Semiotics as Related to
• Level of coupling
Measurement, Volume 1; Article 24, Principles of Epis-
and so on. temology as Related to Measurement, Volume 1; and
Article 39, Characteristics of Data, Information, Knowl-
edge, and Wisdom, Volume 1.
5.6 Defence systems It is useful at this stage to classify the four levels of the
development of a cognitive entity. Using the Oxford Uni-
While extracted from defence material, Hoivik (1999), these versal Dictionary, 1968, we get the following definitions:
may also be relevant to civil projects and situations. • Data:
• Quantity of x ‘a thing given or granted: something known or assumed
• Quality of x as fact and made the basis of reasoning or calculation.’
• Coverage of x • Information:
• Survivability ‘the action of informing’ stemming from
• Lethality ‘to put into form or shape.’
• Sea, air, and land worthiness • Knowledge:
• Warhead size ‘the fact of knowing a thing, state, etc.’ stemming from
• Speed ‘to recognize or identify.’ Also, ‘an organized body of
• Range information’.
• Altitude of operation • Wisdom:
• Evaluability ‘the quality or character of being wise’ stemming from
• Weight ‘having or exercising sound judgment or discernment.’
• Power These give some useful clues about the terms, but we
• Computer throughput really need an explanation that is better related to mea-
• Memory size surement systems. Here, follow the author’s definitions
• Cooling capacity Sydenham (1986).
• Target location accuracy
• Reaction time Data: Raw symbols that are obtained from a measurement
• Receiver sensitivity system and that have no assignment of meaning associ-
• Ranging accuracy ated with them. They are just simply numbers, letters,
• Range ikons, cuneiform stabs in clay, and so on. An example is
• Hardness to damage the symbol set of ‘10’.
• Damage tolerance Information: This is data that has associated, either tagged
• Drift rate of guidance unit with it or held elsewhere, a small amount of cognitive
• Radiation hardness material that gives it a certain meaning. Reduction of
• Engine power ‘raw data’ into ‘engineering units’ is an example. An
• Rate of turn example is ‘10 m’. The assignment of a tag that has a
• Climb rate cognitive meaning; here, the distance unit ‘m’ for the
20 Foundations of Measuring
unit of length, the meter, creates useful information from now. The fact is the various cognitive entities have yet to
the number symbol. be consistently used.
Knowledge: This is sets of information put into a con-
text of a particular use. Representational information
is organized into a coherent model structure. As with 7 THE SCIENTIFIC PROCESS
‘beauty’, what constitutes knowledge is in the mind of
the beholder. An understanding is needed of how quantifying measure-
It possesses specificity of application. For example, the ment can contribute to increasing the available knowledge
raw data from a strain gauge on a wing of an aircraft on a topic. This is explained by the scientific process used
for a given location and time, and with known units, as the basis of reductionist thinking.
constitutes a segment of knowledge. This stems from as early as the sixteenth century and
Wisdom: This is a higher level of cognition than knowledge. has gradually become the norm. In 1931, Bertrand Russell
It is a set of knowledge components having associa- published his understanding of the basic process steps of
tions between entities. For example, the pattern of strain the scientific method.
gauge readings across the aircraft wing may have pecu- ‘In arriving at a scientific law there are three stages:
liarities that suggest, to the expert mind, that it is in an
unsafe state. • The first consists in observing the significant facts
• The second in arriving at an hypothesis, which if it is
A level is reserved for the highest level of, as yet true would account for these facts
unfathomable, intelligence. • The third deducing from this hypothesis consequences
These entities form the intelligence tree shown in which can be tested by observation.’
Figure 1. It is clear how measures lead to an increase
The scientific method relies on:
in wisdom.
Fashions in the use of terms change. Overall, what used • reducing the complexity of the variety of the real world
to be called information often tends to be called knowledge to a manageable state;
System requirements Process of application
used to develop CIs of measures tree
Obtain current
performance
Start state of
measuring measured
system design system
Increasing
uncertainty of Critical issues
measures
CIs
Measures of Reticulation
effectiveness to generate Calculation with data
MOEs tree from TPPs and SPPs
to generate
performance of CIs
Measures of
performance
MOPs
System performance
SPPs parameters
Obtain numbers
Technical performance from TPPs
parameters and SPPs
TPPs
Number of measures used
Figure 1. Intelligence tree shows relativity of the various cognitive variables and relationship to measures triangle.
Measures and Metrics; Their Application 21
Table 1. Stages of the measurement process and the role of The scientifically executed process is the only way by
measurement in its execution. which measures are obtained that are as objective as pos-
Generalized Role of measurement sible. The physical experiment performed in measuring is
scientific method theory and practice the only way to obtain verified data on the physical world.
Develop hypothesis A single measurement entity is being measured as part
1. Identify question/problem 1. Develop test objectives of a large array of measurements needed for evaluation
2. Formulate hypothesis 2. Estimate performance purposes of a system of some kind. Examples might well
Experiment be to assess the airworthiness of a new aircraft or to see
3. Plan the experiment 3. Develop test method if a medical intensive monitor unit is operating within all
4. Conduct the experiment 4. Collect test data critical performance parameters.
5. Analyze the results 5. Calculate the measures
The above sample lists of metrics show that for a
Verify hypothesis project numerous things can be measured. The question
6. Check the hypothesis 6. Compare results
7. Refine the hypothesis 7. Rerun tests or extrapolate needing an efficient solution is how can one set up an
optimal measuring system when time, access, and cost,
usually, severely limit the number of measurements that
• performing analysis or experimentation on simple can be made.
models of the world to examine a hypothesis;
• validating a hypothesis by looking repeatedly to see if
it can be disproved – the ‘null hypothesis’ basis. It is 9 THE MEASURES TRIANGLE AND ITS
actually not achieved by showing it to be always true, as PARAMETERS
is commonly understood (infinite testing needed there!);
• building knowledge, therefore, by eventually refuting What is needed is a plan to set up and use many scientif-
the hypotheses and forming an improved one. ically executed physical measurements that are integrated,
in a traceable manner, to form decisions that map into a
The scientific process of inquiry and its stages are
few high-level measures about the overall system.
summarized as Table 1. Alongside are given the various
This leads to the concept of the measures triangle.
functions of measurement in that process.
Figure 2 shows the various levels and types of measures
Areas of measurement are needed to undertake all
stages of this knowledge-gathering activity. Measurement that form this measures treelike diagram.
is, therefore, a key part in its application. Poorly undertaken To set up a system’s measurement plan, the first thing
measurement can lead to incorrect knowledge, or more to do is to identify the critical issues (CI) from the system
usually the case, to less precise knowledge, possibly giving requirements documentation. CIs are those high-level issues
rise to misinformation or negative knowledge. that, will make the development fail if not achieved.
The process acquires new data from measurements made,
and the observer uses that data to draw conclusions about
Highest intelligence
the hypothesis being developed by evaluating the data in
the context of the hypothesis.
So far, we have discussed the role of measurement in
the scientific process. It is an easy step to see that this
Wisdom
process is applicable to any measurement situation itself
for a measurement activity is an experiment to see what
you have. This is the time to review how that data flows Increasing wisdom
into evaluation of the hypothesis. with usually Knowledge Knowledge
reducing provable
objectivity
8 HOW TO APPLY MEASURES
Information Information
We need to ask a fundamental question. What is the holistic
purpose of making a measurement?
In the closest inward looking boundary, it is to satisfy the
need of the person requesting the test. This is, however, far Data Data
too restricted a horizon to take because that test is being
done to integrate into a much larger problem situation. Figure 2. Measures triangle and its levels of measures.
22 Foundations of Measuring
Each CI is then broken down to obtain its measures ideal of perfect measurement appears to be a goal but
of effectiveness (MOE). These are expressed in terms of it, usually, would be too costly, would take too long to
what is to be achieved, for example, the requirement states organize, and more often than not, would not be feasible
that ‘the customers must be satisfied’ so that MOE needs to build.
measures of customer satisfaction to be set up. That it At any point in the tree a nodal value can be extracted,
may not be immediately obvious how to measure it is not and its measure calculated from data below it. The widely
an issue at this stage of the reticulation. One should not used technical performance measure TPM is, in effect, such
start from what can be measured, but from what should be an extraction, but it has little recognition of where it fits into
measured – a commonly ignored requirement! the traceable hierarchy from data measurement to proof of
The MOEs, in turn, reticulate down to give measures requirement satisfaction, for it is usually generated from
of performance (MOP). These break down the MOE into experience without use of a measures tree.
the MOP that, when combined, lead to the MOE value. A more detailed explanation of test and evaluation
Customer satisfaction could be measured in terms of the of large systems with many variables is provided in
return rate of customers, from a direct survey of them using Sydenham (2004).
a written survey instrument, or from use of a video camera
that records their demeanor as they pay for the goods.
MOEs and MOPs cannot be measured directly.
10 CASE STUDY OF THE GENERATION
This in turn gives the number of returns per customer; OF MEASURES
consolidated survey results grade; and an image interpreta-
tion as the system performance parameter SPPs to measure. A simple example of how a measures tree is developed
The final level is the technical performance parameter from the CI is the generation of a suitable measuring system
(TPP) wherein physical measurements are possible. SPP for automatically monitoring the quality of loaves of bread
measures will often go directly to the TPP level as is the coming from a bread baking line. This results in a fully
case here for return rate: a counting exercise. The paper physical measurement situation.
survey will, however, not necessarily need TPP measures Figure 3 shows a set of reticulated measures that allow
as it will be done by discussion and box ticking in a form. the appropriate TPPs (the actual measurements of physical
The paper survey is already able to yield some SPP values. variables) to be measured and their values integrated into
The customer demeanor grade has to be determined from the result that each loaf is acceptable, having met all
image processing set up to measure various indices. This requirements that are extracted as CI. If the integrated
needs a physical measuring instrument to obtain a set of measure for a loaf is not acceptable, then the out-of-
TPPs that map into the SPP of customer demeanor. limit measurements will show where action is needed
for correction.
In this way, from each CI, a tree of measures is generated.
A measures tree can be set up for a measuring system
A measuring system design has been generated. Of these
engineering development project. As the development pro-
layers, only the TPP is hard data obtained by use of
gresses, the respective measures are monitored and used,
the defendable scientific process. The measures on the
with the best estimates of the time, to yield the overall
other layers invariably contain some degree of subjective
value of merit for progress of the project.
decision making.
This methodology can become very complex in large
Once all of the TPP and SPP data has been obtained, the
systems developments for the following reasons:
values are fed upward inside the tree to give the various
other measures their value, by direct calculation up the tree. • CIs can exceed the 100 to 1000s;
In this way, traceability results for all measurements and • reticulation needs time to decide good breakdown sets
only essential measurements are performed. of variables;
When the first draft of the system is available and • measurements cannot always be made in a timely
representative figures can be inserted for each measure, manner;
judgment can be exercised to see which measures are not • uncertainty of measurements is variable;
significant; these can be omitted to reduce the measurement • nature of measure, such as statistical distribution, are
effort needed. different, making combination problematic;
As the calculations rise up the trees, the integrated • measures are often taken from similar prior tests and
knowledge becomes less accurate in its nominal value and need careful application;
will have increasing uncertainty. This situation cannot be • required measurements cannot be made because of a
avoided unless all decisions are made entirely with physical reduced budget, too little time, lack of test object, or
measurements and with nonfuzzy logical combination. This lack of a test facility;
Measures and Metrics; Their Application 23
Critical issues are identified Level of measures
CIs
Legal trade limits are met Customer satisfaction is met Safe to consume
MOEs
Amount of bread provided Look, taste, and feel Foreign metal objects MOPs
Mass and shape Appearance, taste, texture Magnetic profile SPPs
Mass Length Width Height pH Odor Total magnetic field TPPs
Color Shading Stiffness Position
Figure 3. Development of a measuring system for loaves of bread from an automatic bakery.
• prior history, inherited data, and poorly defined pro- Finkelstein, L. (1999) Foundational Problems of Measurement
cesses are forced into the program. Science, in Proceedings of. International Workshop on Advances
of Measurement Science, IMEKO TC-7, Kyoto, (pp. 23–31).
For these reasons, many large programs do not get well- Hofmann, D. (1999) The Role of Measurement for Innovation and
organized measures, with subsequent lack of confidence and Society, in Proceedings of International Workshop on Advances
of Measurement Science, IMEKO TC-7, Kyoto, (pp. 101–109).
timeliness in delivery.
Hoivik, T. (1999) Advances in Test and Evaluation, Short Course,
Small programs need to tailor the measures tree concept
Defence Engineering Group UCL, London.
to suit their budget. An example is Pratt (2004).
Hutchins, R.W. (ed.) (1952) Great Books of the Western World –
31 Descartes and Spinoza, University of Chicago, Chicago, IL.
IMEKO. (1999) All papers, in Proceedings of International
RELATED ARTICLES Workshop on Advances of Measurement Science, IMEKO TC-7,
Kyoto, (pp. 1–235).
Article 4, Economic Considerations of Measurement, Kariya, K. (1999) Construction of Measurement Science as Basis
of Empirical Sciences, in Proceedings of International Work-
Volume 1; Article 10, Extensions of the Representa- shop on Advances of Measurement Science, IMEKO TC-7,
tional Theory of Measurement, Volume 1; Article 11, Kyoto, (pp. 67–98).
Measurement Theory in Physical, Social, and Psycholog- Klein, H.A. (1975) The World of Measurements, Allen and Unwin,
ical Science, Volume 1; Article 14, Hierarchical Aspects London.
of Measurement Systems, Volume 1; Article 15, Typical Pratt R.W. (2004) Framework for Issue Management, MEng
Measurement Systems Architectures, Volume 1; Arti- thesis, University of South Australia.
cle 39, Characteristics of Data, Information, Knowledge, Sage, A.P. and Rouse, W.B. (1999) Handbook of Systems Engi-
and Wisdom, Volume 1. neering and Management, Wiley, New York.
Sydenham, P.H. (1979) Measuring Instruments: Tools of Knowl-
edge and Control, Peter Peregrinus and Science Museum,
REFERENCES Stevenage.
Sydenham, P.H. (1986) Structured Understanding of the Mea-
Blanchard, B.S. and Fabrycky, W.J. (1998) Systems Engineering surement Process Pt.2: Development and implementation of a
and Analysis, Prentice Hall, New Jersey. Measurement Process Algorithm. Measurement, 3, 161–168.
Brown, S., Fauvel, J. and Finnegan, R. (1981) Conceptions of Sydenham, P.H. (2004) Systems Approach To Engineering, Artech
Inquiry, Methuen and Open University Press, London. House, Boston, MA.
Yang, Q. and Butler, C. (1997) On the Framework of Measure-
Bud R. and Warner, D.J. (eds) (1998) Instruments of Science: An
ment Science, in Proceedings of XIV IMEKO World Congress,
Historical Encyclopedia, Garland, London.
Vol. 5, Tampere, (pp. 101–106).
Checkland, P. (1981) Systems Thinking, Systems Practice, Wiley, Yang, Q. and Butler, C. (1998) An Object-oriented Model of
Chichester. Measurement Systems. IEEE Transactions on Instrumentation
Ellis, K. (1973) Man and Measurement, Priory Press, London. and Measurement, 47(1), 104–107.
4: Economic Considerations of Measurement
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
The cost of measurements and their worth has not been
1 Cost Issues 24 a subject that many writers have felt they wanted to openly
2 The Cost of an Instrument 24 address. In general, measurement is seen as a low value
3 Analyzing the Cost of Making a item (around 2–8%) in a much larger overall budget of a
Measurement 26 system and thus needs little attention. It is considered by
many that making a measurement is straightforward and
4 Calculating the Cost of Measurement 28
simple to set up using available systems off the shop shelf.
5 Balancing the Cost-benefit Factors 28 Costs to produce and sell a measuring system are, how-
6 The National Measurement System Matrix ever, very much the concern of the manufacturer who needs
Approach 29 to set a realistic selling price, and to a nation, for it is a
7 Guidance to Management on the Cost-benefit significant factor in a national economy due to the high
of Measurement 29 gearing effect of good measurement practice.
Related Articles 32 This account begins first by looking at the cost of the
References 32 apparatus, broadening outward through the cost of making
a measurement, and then to the cost of measurement to
a nation.
1 COST ISSUES
Measurements are made using physically existing equip- 2 THE COST OF AN INSTRUMENT
ment. The instruments must be created from the basic
materials and applied using the labor of people. Material The cost of finally making a measurement is formed from
and labor resources have never been unlimited or freely several subcomponent costs:
available to a measurement user. Priorities must be decided
in situations appearing to need measurement equipment. • Ci – cost to investigate the measurement need by the
A measuring instrument does not stand alone as far as customer
its costs are concerned. The cost of measurement should • Cm – cost to research need, design, manufacture and
not be assessed by looking only at the purchase or man- market the measurement system hardware and software
ufacturing cost of the apparatus used. Here, attention is • Cs – cost to set up the measurement in its application
drawn to the real cost of the individual measurement, for ready for use
too little attention is often given to this aspect of measuring • Cm – cost of maintenance and spares holdings
instruments. • Ca – cost to apply the measurement.
To provide a balanced economic statement, it is necessary
to consider the worth of a measurement in terms of the Costs to decide what is needed and to locate and pur-
cost-to-benefit ratio for making the measurement. chase the instrument system Ci are often absorbed into other
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Economic Considerations of Measurement 25
project budget lines and thus can be transparent. They, how- result of devoting most of his life to the task. Originally,
ever, can be considerable. Paradoxically, the less expensive instrument making was more akin to the production of fine
the system needed, the higher the relative costs can be for arts products of today than to the fierce open competition
locating a suitable unit; sellers will often assist with this of an industrial market.
cost where the price tag is large. The introduction of production manufacturing methods
The cost Cm of designing and making the instrument changed this situation greatly. Markets widened, production
system is built up from many parts, such as researching run numbers rose, and cheaper means of producing the
the need, design development, embellishment, paper work desired performance were realized.
about specifications, use and service, packing, delivery, The measuring instrument manufacturing market game
and many more contingencies. Figure 1 depicts the factors has the feature that its players seem to be doing their best
involved in costing the making of an instrument. Manufac- to work themselves out of a job! As soon as a significant
turers well understand the issues and costs involved. new principle or technique is discovered that could cause
In classical antiquity, many of these cost factors did major sales, the manufacturers move to produce it for sale
not arise. How to use the instrument was usually self- first at a relatively high price. Demand grows with reduced
evident. Rarely were more than one of exactly the same price, so means are sought to stimulate demand, leading to
model made; often the maker was the user, so he or she larger production runs sizes. This tactic, if successful, drops
did not need to be concerned with records and sales. As the price, which saturates the market; the manufacturer must
the instruments were made by craft artisans, they usually again look for another product.
had more decorative embellishment than was needed to be Some makers will make adequate profits in this cycle and
adequately functional. will return again to play another round. Many, however, do
The time taken to make a piece of measuring equipment not get their timing and marketing methods right. They end
has changed over time. It was not uncommon for instrument up with an abundance of out-of-date products that cannot be
makers up to the seventeenth century to spend not months sold at even the cost-to-produce price. Companies can go
but years on manufacture. Harrison produced his four prize- to the wall if this happens too often. Many an instrument
winning chronometers of the eighteenth century as the firm has poured funds into developing a certain class of
Market research studies
Market surveys
Labor
supply
and Design studies of alternatives Designers, etc.
control Shareholders and
Product selected Meetings other financial
interests
Prototype designed and made Detail designers, etc.
Production Sales and promo Components supply
Adapt prototype for production
Stores Service and
test
Manufacture pilot run
Assemble and test External supply
Modifications
Manufacture for sale
Testing
Stores Packaging Customer training Servicing stations
Delivery Customer use
Figure 1. Factors involved in costing the manufacture of an instrument.
26 Foundations of Measuring
instrument only to have other makers systems sell better 3 ANALYZING THE COST OF MAKING A
than theirs. MEASUREMENT
The instrument market has the following specific
features:
Assessment of the real cost of a measuring instrument
• Production volume is small to medium, rarely large. should be on the basis of considering what costs would
• Cycle time to develop is only a matter of months have been avoided, or incurred through lost benefit, if
or so. it had never been considered – the economists call this
• Costs to produce have fallen dramatically as develop- the ‘opportunity cost’. Figure 2 shows the total costs that
ment by the whole group of this industry increased. comprise the cost of making a measurement.
• Useful life of the physical aspect of a product is com- To begin the costing process, it is necessary to first
paratively long. include the cost of the time spent in deciding what has to be
• Information Technology (IT) content changes rapidly measured (Ci ). Too often this stage is seen as barely existing
but the basic principle remains reasonably static. because the measurand, the parameter to be measured,
• Purchasers expect ultimate reliability for minimal cost. appears obvious. It is the writer’s experience that many
• Applications are not well worked out. users give too little emphasis to this very basic step, ending
• Users expect ‘fit and forget’ to apply despite the low up measuring the wrong entities.
Having decided the variable(s) concerned, it is then
price paid.
necessary for the designer to devote time to generating
• Users expect spares to be available for decades.
sound specifications for the method selected. For example,
• Companies making instruments and measuring systems
assume the task has been defined as one of measuring the
change hands a great deal.
temperature of a product because that seems to indicate
when it is ready to be transferred to another process – as
Furthermore, as the per-unit production cost is lowered in jam making. Is temperature really the best parameter
to make the system more affordable, the reliability of to monitor? If so, what must be the precision, accuracy,
the components, and thus that of the system, are often absolute value, shape of sensor, and so on? To ascertain
sacrificed. when jam is cooked sufficiently, it is better to measure
Another feature that the instrument maker has to contend the pH, and not the temperature! Making these requirement
with is that systems designers consider it reasonable and decisions incurs time and monetary costs.
normal to call up specials when specifying the measuring Consider next that the specific instrument principle and
apparatus required. One-offs are product lines that the firms the specifications have been chosen. The measurement task
must produce to maintain customer faith yet they often cost has still barely begun, for it is then necessary to locate trade
far more than the customer thinks should be paid. People information of available products so that a specific purchase
have become used to the idea that their electric toaster can be organized. Tenders are called – or firms contacted –
works as the designers suggest it should, yet they would to obtain quotations, data sheets, delivery information,
not always agree to make the same degree of compromise installation details, and so forth. At this juncture in costing
in using the instrument makers suggestions of what are of the real cost of a measurement, one must not forget to
reasonable basic units to work with. also include the time spent in preparing budget submissions
Just as the knowledge about measuring systems has to and filing the records of this investigative work.
be restructured for future, more efficient use, so must the Eventually, the stage is reached where a purchasing
range and number of marketed instrument variations be decision can be made. That, too, takes up more time by
restructured. skilled staff. It is unlikely that these above mentioned stages
It is generally acknowledged and understood that the cost can be passed through in less than a minimum of a few
of the instrument itself can be assessed from a long list hours of effort for even the simplest of measurements.
of the production component costs, as shown in Figure 1. Extensive measurement systems take months, sometimes
Management must budget for a manufacturing price to many-years, of work to reach this stage.
decide sales potential and profit margins. The same, how- The instrument system has then been selected and the
ever, cannot be said of users when they assess the cost order placed for supply; the cost of the ordering procedure
of a measurement to their organization or enterprise. The should be included. Delivery is then awaited. If all proceeds
truth is rather startling: the real cost of a measuring instru- perfectly, the instrument arrives on time without the need
ment is usually many times the purchase price of the to make enquiries about late delivery. More often than not
apparatus. it is necessary to chase delivery, adding more cost.
Economic Considerations of Measurement 27
Measurement task arises
Deciding what to measure
Choosing class of sensor to use Experiments needed?
Deciding which instrument to purchase and from which supplier
Placing order Accounting procedures begin
Accreditation
Chasing delivery
calibration, evaluation
Delivery check Inventory records, possible
calibration in laboratory
Familiarization with instrument by user
Installation Maintenance program begins
Instrument in use Preventative and corrective
maintenance, calibrations
Disposal
Cycle begins again
Figure 2. The cost of a measurement far exceeds the purchase price of the instrument.
Upon receipt of the system, the wise purchaser will carry Installation is then needed. This requires the organization
out an immediate overview delivery check to ascertain that of varied skills. It may necessitate inventory records to be
it is undamaged, is generally within specifications, contains created along with an instrument dossier and a spare parts
the necessary documentation (which might be yet another reserve to be set up. Plant drawings usually need to be
cost to consider), and that it will suit the purpose intended. updated with instructions written and issued to operators.
Soon after that a more extensive test is needed; if performed Costs of the measurement continue to mount. To ensure
properly, this can take many hours and will need the use of that the instrument is performing properly, it will be neces-
expensive test equipment, therein also placing a cost into sary for it to have a service program, a calibration program,
the overall account. and a maintenance program in operation, as well as a pos-
If things are found to be wrong at delivery time, addi- sible method of monitoring the measuring system if its role
tional effort will have to be devoted to getting it corrected is in a key area. To remove an instrument, take it to a cal-
through the purchase agreement or perhaps, which is often ibration laboratory, carry out a calibration, and return it to
quicker and perhaps cheaper, having it repaired in-house. duty will take a minimum of another two or three hours of
In parallel with this acceptance stage is the need for work, but it can even take many days!
the user to become familiar with the instrument opera- Too often the decision is made to purchase the instrument
tion. It may need attendance by staff at a training course, offered at the lowest price. This is oversimplifying the pur-
or be simply a case of reading the operating instructions. chase decision; the fact remains that most instruments will
Whichever it is, still more time must go onto the cost- need some unpredictable service during their useful life.
ing sheet. The resources expended on this should not be The cost of this kind of service is high, for it usually occurs
underestimated, for a complex piece of measuring equip- when labor repair costs are at their highest penalty rate.
ment may require the user to spend much unproductive Furthermore, the emergency technician may not be familiar
time learning to apply it. Errors in its use in the early with the system and has to learn at the owner’s expense.
stages will also give rise to costs that have to be met Eventually, the measuring system has provided a useful
somewhere. and satisfactory life; its costs to the user go on. At some
28 Foundations of Measuring
stage, effort has to be expended to establish a replacement This can be done by conducting a risk analysis, for
schedule and the choice of its replacement, and these can example along the lines of developing a safety case – see
be complex issues. The model of instrument used originally Article 99, Safety Organization, Volume 2; Article 100,
will often have been superseded by something that may not Safety Analysis Methods, Volume 2; and Article 101,
be quite the same in some aspects that are critical to the Assessing and Demonstrating Safety, Volume 2. Some
application; it may not be possible to get a replacement simple examples should assist one to understand what can
at all. be at stake.
Finally, the last of the costs associated with the mea- A large blower was used to aerate sludge in a mineral
suring system parts is its disposal. It may be possible to separation process. The 100-kW blower had a temperature
recoup some of the aforementioned costs through sale of sensor installed to shut it down when its internal lobes over-
the instrument, but the opportunity cost of a well used, heated, for that could cause it to seize and burn out the
now superseded instrument is more likely to be a debit, motor. The cost of the blower unit was around $200 000
not a credit. This cost often does not arise until years installed, and the inexpensive and somewhat crude temper-
after installation, for most large complexes tend to store ature sensor used was a mere $200. As might reasonably
old instruments for many years before they are finally dis- be expected, the blower was eventually seriously damaged
carded, finding their way into a museum collection or being when it overheated at a time when the sensor had not oper-
dismantled for use as pieces of some other plant. ated properly. Lost production and repairs cost as much as
a new blower unit. In this case, the outlay of a mere 0.1%
of the purchase value on protection seems to be too low.
4 CALCULATING THE COST OF
Lost benefit can be expected to be one of the higher costs.
MEASUREMENT A second example is found in the factory of a permanent
magnet manufacturer who had taken over an old plant.
Any genuine attempt to estimate the cost of a measurement
These magnets are cast from a molten NiCo alloy that melts
must include all of the above factors. The overall cost issue
at a temperature higher than that of steel. The actual pouring
to be studied is one of establishing what resources were tied
temperature and subsequent annealing process are crucial to
up by the existence of the measuring system that might
the final field strength of the magnet.
have been used elsewhere if it had not been considered.
The whole manufacturing process, with its casting, sev-
A detailed costing on the above-given basis can lead to a
eral annealing and curing steps, takes some 10 days before
satisfactorily quantitative assessment of which measuring
it is learned from tests how well the magnets will perform.
system to purchase and in which way to use it to make the
Along with the process, several measurements are made of
measurement needed.
critical parameters. At the end of the process there existed
All of the above costs are concerned only with the choice,
a final inspection station used to test and certify the quality
purchase, installation, and maintenance of the correct mea-
of each magnet.
suring system. The cost of the whole measurement can
First, the process relied on a very old uncalibrated
continue to rise if the right decisions have not been made.
infrared radiometer and a set of record cards that had
Much of measurement conducted is, unfortunately, not
entirely productive. A wrong measurement has often been been inherited, giving the different temperatures needed (as
the cause of disastrous losses to the enterprise. A serious indicated by that radiometer) for the various shapes of mag-
malfunction in the commissioning stage of a new ship can net cast. As no independent calibration of the radiometer
cost the builder thousands of dollars per hour. Incorrect existed, if it should fail, it would be virtually impossible to
operation of a weapon of war has a cost that goes without decide what temperatures to use in future.
saying. Next, the final inspection area that was set up inside the
dirty foundry area was a dark mess. The force measuring
system for the magnet test was red with rust and its
5 BALANCING THE COST-BENEFIT measuring instrument dials were hardly readable. This was
FACTORS the station audited by the customer! It was rejected, as it
was not a suitable testing process for them to accept; they
The total of all of the above costs has to be weighed then did their own tests with a higher rejection ratio.
against the benefits of making the measurement when it The manufacturer decided he was unable to make the
is expressed in monetary terms. Whereas it is reasonably investment in improvements and eventually was forced to
clear and straightforward to obtain the cost of making the close down.
measurement, it is not so clear how to assess the value of A third example situation was that of a temperature con-
the benefits of making it. troller installed in a greenhouse used for plant research in a
Economic Considerations of Measurement 29
university. After 5 years of apparently stable temperatures of the National Measurement System, 1972–1975’ and
in the greenhouse, the thermocouple controller system was NBSIR 75–949 ‘Structure and function of the National
overhauled. measurement system’.
It was found that the thermocouple circuitry was not in The report included a major impact matrix (Figure 3),
the control regime expected, owing to an incorrect change showing the measurement relationships existing for 25
to the circuit card to hopefully reduce the indication span. groups of societal needs. It graphically portrays the dynam-
This had reset the recorder’s chart zero point to where it was ics and the degree of such interactions and, in doing so,
expected to be, but in doing so the changes had increased well illustrates the great complexity and widespread use of
the chart span to 400 ◦ C, full scale, instead of 40 ◦ C. It measurements in our modern lifestyle.
seemed to behave as expected but it was then seen to be This NBS study was catalytic; many countries now
recording the temperature of the control room, not that of maintain NMS activities. Owing to the importance of the
the greenhouse. The cost of this incorrect measurement NMS to national economic welfare, they are now usually
would be hard to assess but it is clear that the research cared for by departments of trade or industry.
findings for the previous 5 years were subject to much Information in the UK system is found on DTI (2003).
uncertainty. The Australian one is explained in NSC (2003). Taiwan’s
The benefits of a measurement need to be carefully is discussed in NML Taiwan (2003). Others are to be found
investigated using an objective standpoint and methodology from a web search.
that leads to monetary values. These can then be weighed Interestingly, the organization that began the NMS move-
against the monetary cost to make the measurements. ment, NIST, does not list any NMS activity, as such, on its
current information service, NIST (2003).
6 THE NATIONAL MEASUREMENT
7 GUIDANCE TO MANAGEMENT ON
SYSTEM MATRIX APPROACH
THE COST-BENEFIT OF
The cost of measurement also can be of importance at the MEASUREMENT
whole country level; measurement has been identified as
an important factor in a national economy. This discussion Very little has been done to find and develop means by
now considers the cost to a nation of making measurements. which management can be guided in budget decisions
Studies of the cost of national measurements were given concerned with the use of new, or existing, measuring
in-depth consideration in the 1970s. Prior to that, costing equipment. Easy to reveal are the simple benefits of money
was certainly performed on the annual budgets of such saved by implementing a better-controlled process, or by
institutions as the national measurement laboratories and labor costs saved by the use of a more advanced measuring
similar identifiable groups, but little attempt was made to go technique. More difficult to quantify in an adequate manner,
beyond this small section of the total national cost devoted but often equally as important, are such factors as
to measurement. • general improvement of the standard of a product result-
The term National Measurement System (NMS) came ing in increased sales;
to the fore when the then National Bureau of Standards • increased reliability of a product;
(NBS), USA, (now the National Institute of Science and • less wastage in manufacture;
Technology (NIST)) conducted a major nation-wide study • potential to produce a product that was otherwise impos-
of the value of the United States’ NMS. sible to make;
Development of their study can be followed in the • improved performance of the product;
National Conference of Calibration Laboratories (NCSL) • improved working relationships between those groups
Newsletters from 1972, through to 1978. Minck (1977) of a nation that impact together in an area of measure-
defined the NMS as ‘consisting of all national activities ment; and
and mechanisms that provide physical measurement data to • increased confidence by others in a nation’s products,
allow the creation of the objective, quantitative knowledge with subsequently improved exports.
required by our society’.
Previously, Huntoon (1967) had developed the concept At present, measurers are not yet able to measure the
of the NMS in a paper. The NBS study reached the worth of measurements to the same degree as they can
final report stage in 1975 after 151 separate studies had assess the cost of measurement. Perhaps this is the rea-
been made. The report contained 27 documents. Two key son many measurement budget decisions generally do not
readings are NBSIR 75–925 ‘Final summary report study allocate finance to measuring equipment in proportion to
30
DIRECT
MEASUREMENTS
TRANSACTIONS U
MATRIX FOR S
NATIONAL SYSTEM E
OF PHYSICAL R
MEASUREMENTS S
(March 1976) N
B
KNOWLEDGE COMMUNITY
(Science, Education,
Prof. Soc. & Publ.)
INTERNATIONAL
METROLOGICAL
ORGANIZATIONS
DOCUMENTARY
STANDARDS
ORGANIZATIONS
INSTRUMENTATION
INDUSTRY
(SIC Major Gp 38)
OTHER U.S. NATIONAL
STANDARDS
AUTHORITIES
STATE & LOCAL
OFFICES OF WEIGHTS
& MEASURES (OWM's)
STANDARDS & TESTING
LABORATORIES
AND SERVICES
REGULATORY
AGENCIES
(excl. OWM's)
DEPARTMENT OF
DEFENSE
(excl. Stds. Labs)
CIVILIAN FEDERAL
GOVT AGENCIES (exc.
Stds Labs & Reg. Ag.)
STATE & LOCAL
GOV'T AGENCIES (exc.
OWM's Reg. Ag.)
INDUSTRIAL
TRADE
ASSOCIATIONS
AGRICULTURE, FORESTRY
FISHING; MINING
(SIC Div. A & B)
CONSTRUCTION
(SIC Div. C)
FOOD/TEXTILE/LBR/
PAPER/LEATHER/ETC.
(SIC 20–26, 31)
CHEM/PETROL/RUBBER/
STONE/CLAY/GLASS...
(SIC 28–30, 32)
PRIMARY & FAB.
METAL PRODUCTS
(SIC 33–34, 391)
MACHINERY
EXCEPT ELECTRICAL
(SIC Major Gp 35)
ELECTRIC AND
ELECTRONIC EQPMT
(SIC Major Gp 36)
TRANSPORTATION
EQUIPMENT
(SIC Major Gp 37)
TRANSPORTATION &
PUBLIC UTILITIES
(SIC Div. E)
TRADE/INS/FIN/REAL
EST/PERS SVCS/PRINT
(SIC F-H, bal l, 27)
HEALTH SERVICES
(SIC Major Gp 80)
GENERAL PUBLIC
S
SUPPLIERS
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
1 KNOWLEDGE COMMUNITY 4 1 3 1 3 1 2 1 4 1 2 1 3 1 3 1 3 2 2 2 2 2 2 3 1 3 1 2 1 3 2 3 1 2 1 3 1 2 1 2 2
(Science, Education, 6 2 3 3 4 1 1 2 2 3 3 1 1 1 1 1 2 2 2 3 3 2 1 1 1
Prof. Soc. & Publ. 2 1 1 1 2 1 2 1 2 1 2 1 2 1 1 1 1 2 1 1 1
2 INTERNATIONAL 2 1 3 1 3 1 2 2 3 1
Foundations of Measuring
METROLOGICAL 1 3 2 1 3 1 1 1 1 1 1 1
ORGANIZATIONS 1 1 2 1 2
3 DOCUMENTARY 3 2 3 1 3 2 3 1 3 1 3 1 3 1 3 2 3 1 2 1 3 1 2 1 2 1 2 1 2 1 3 1
STANDARDIZATION 1 2 3 3 2 1 2 2 2 2 1 2 1 2 2 2 2 2 2 2 2 1 2 1
ORGANIZATIONS 2 2 1 1 2 1 2 1 1 2 1 1 1 1 2
4 INSTRUMENTATION 2 1 2 1 2 1 4 1 3 1 3 1 3 1 3 1 3 1 2 1 3 1 3 1 3 1 3 1 3 1 3 1 3 1 3 1 3 1 3 2 3 1
INDUSTRY 2 1 2 5 3 1 1 3 2 5 4 3 2 3 4 3 3 4 5 5 5 5 5 4 4
(SIC Major Gp 38) 1 1 2 2 2 2 2 2 1 1 2 2 1 1 1 2 1 2 1 2 2
3 1 3 1 3 1 3 1 3 1 3 1 4 1 3 1 3 2 3 3 1 2 1 2 1 4 1 2 2 2 2 4 2 2 1 3 1 4 1 4 2 4 3 4 3 2 1
5
3 3 3 3 4 1 2 3 2 3 3 1 2 2 1 1 2 2 2 2 2 2 1 2 1
NBS 1 2 2 2 1 2 1 1 2 1 2 1 1 2 2 1 1 1 2 2 2 1
6 OTHER U.S. NATIONAL 3 1 3 1
STANDARDS 1 1 1 1 1 1 1 2 1 1 1 1 1 1 1
AUTHORITIES 1
7 STATE & LOCAL 3 1 4 1 4 3 1 3 1 3 1 4 4 1
OFFICES OF WEIGHTS 1 1 1 2 2 1 1 1 1 1 2 1 2 1 2 3 1
& MEASURES (OWM's) 1 R 1 R 1 2 1 1 1
8 STANDARDS & TESTING 2 1 2 1 2 1 2 1 3 1 2 1 3 1 3 1 3 2 1 3 1 3 3 1 3 1 3 2 3 1 3 2
LABORATORIES 1 2 3 2 3 2 3 2 1 2 1 1 1 2 2 1 2 3 2 1 2
AND SERVICES 1 1 1 R 2 R 1 2 1 1 1 1 1 1 1 2 1 2
9 REGULATORY 3 1 3 1 3 1 3 2 3 1 3 2 2 1 3 1 2 1 4 2 4 1 3 2 3 1 3 1 4 1 3 2 3 1 3 1 3 1 4 2 3 1
AGENCIES 1 2 2 2 2 3 1 2 1 2 2 1 2 2 2 2 2 2 2 2 2 2
(excl. OWM's) 2 2 2 R 2 R 2 R 1 1 R 2 R R 2 R 1 R 2 R 1 R 1 R 1 R 1 R 1 R 2 R 2 R 2 R 3 R 2 R
10 DEPARTMENT OF 3 2 3 1 3 1 2 1 3 1 3 1 3 1 2 1 2 1 3 1 3 1 3 1 3 1 2 1
DEFENSE 2 2 3 3 1 3 1 7 2 1 1 2 2 4 4 2 2 1
(excl. Stds. Labs) 2 1 2 R 1 R 1 R 1 1 1 1 R 1 R R 2 R 2 R 2 R R R
11 CIVILIAN FEDERAL 3 1 2 1 2 1 3 1 2 2 2 1 2 1 3 1 2 1 2 1 2 2 3 3 1 3 1 3 1 4 1 2
GOV'T AGENCIES (excl. 2 2 2 3 1 2 2 1 7 2 1 2 1 1 2 2 1 2 2 4 1 2 3
Stds. Labs & Reg. Ag.) 1 2 2 1 1 2 1 2 1 2 2 1 1 2 1 1
12 STATE & LOCAL 2 1 3 2
GOVERNMENT AGENCIES 1 1 1 1 1 1 1 1 1 1 6 1 1 2 1 1 1 1 1 1 1 1 1
(exc. OWM's & Req. Ag.) R 1 R R R 2
13 INDUSTRIAL 2 2 1 2 1 2 1 3 1 2 1 2 1 2 1 2 2 2 1 2 1 2 1 2 1 2 1
TRADE 1 1 3 2 2 1 1 2 1 1 1 2 1 1 1 1 1 2 2 1 1 1 1 1
ASSOCIATIONS 1 2 2 2 1 2 1 2 2 1 1 1 1 1
14 AGRICULTURE, FORESTRY 2 1 2 2 2
FISHING; MINING 1 1 1 1 1 1 1 1 2 1 1 5 1 2 2 2 1 1 1 2 1 1
(SIC Div. A & B) R R R R
15 CONSTRUCTION
1 1 1 1 1 1 1 1 1 1 1 1 6 1 1 1 1 1 1 1
(SIC Div. C) R R R R
R
16 FOOD/TOB/TEXTILE/ 3 1 3 2 3 2 2 3 1 3 1 2 1
APPAREL/LBR/FURN/PAPER/ 1 1 1 1 2 1 1 1 1 1 2 1 1 6 1 1 1 1 1 2
LEATHER (SIC 20–26, 31) 1 2 R 2 R R R 2 2 1 R R R R
17 CHEM/PETROL/RUBBER/ 2 1 2 1 2 1 3 2 2 2 2 2 1 2 2 2 1 2 1
PLASTICS/STONE/CLAY/ 2 1 2 2 2 1 1 2 1 2 1 1 2 1 1 6 1 1 1 1 1 1 1 1
GLASS (SIC 28–30, 32) 1 2 1 2 1 2 1 2 1 1
Figure 3. Direct measurements matrix from the USA foundational NMS study.
18 PRIMARY & FAB. 2 1 3 3 1 3 2 3 3 1 3 1 2 1 3
METAL PRODUCTS 1 2 2 1 1 1 2 2 2 1 2 2 1 1 1 6 1 1 2 2 1 1 1
(SIC 33-34, 391) 1 R 2 R R 1
19 MACHINERY, 2 1 3 1 3 1 3 2 3 1 3 1 2 3 1 2 3
EXCEPT ELECTRICAL 1 1 2 2 2 1 1 1 1 1 1 2 2 1 1 1 2 6 1 2 2 2 1 2
(SIC Major Gp 35) 1 R R 1
20 ELECTRIC AND 3 1 3 1 3 1 3 1 3 1 3 1 3 1 3 1 3 1 3 1 3 2 2 2
ELECTRONIC EQPMT 2 2 3 2 2 1 3 2 1 2 1 1 1 1 1 2 6 2 2 1 2 2
(SIC Major Gp 36) 2 2 2 2 2 2 1 2 1 3 3 2
21 TRANSPORTATION 2 1 2 1 3 1 2 1 2 1 2 1 2 1 2 1 2 1 3 2 1 2 1 2 1 3 2 3 1 3
EQUIPMENT 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 6 2 1 1
(SIC Major Gp 37) 1 1 1 R 2 R 1 R 2 2 1 1 1 1 2 1 1
22 TRANSPORTATION & 2 1 2 1 2 1 2 1 3 1 2 1 3 2 2 1 2 2 1 3 1 3 1
PUBLIC UTILITIES 2 2 2 2 1 1 2 2 2 2 1 2 2 1 1 1 1 1 1 2 6 2 1 4
(SIC Div. E) 1 2 1 R 2 R 2 1 R 2 1 1 1
23 TRADE/INS/FIN/REAL 3 1
EST/PERS SVCS/PRINT- 1 1 2 1 2 1 2 1 1 1 1 1 1 2 2 2 1 1 1 2 7 1 6
PUB (SIC F-H, Bal. I, 27) R 1 R R R
4 2
24 HEALTH SERVICES 1
1 1 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 5 2
(SIC Major Gp 80) R
R 3 R
2 1 2 1 2 1
25 GENERAL PUBLIC 1 2 1 1 1 1 2 1 1 1 1 1 1 1 1 1 1 2 2 1 7
R R R R 2 R R R R R R R R R R R R R R R 1
KEY TO MATRIX ENTRIES
C−Importance of transactions D−(in) adequacy of services
1 = purely convenience 0 = no improvements needed
2 = strongly desirable 1 = could be improved
3 = no real alternatives 2 = marginal
4 = essential USERS 3 = serious deficiencies
C D 4 = out of control
A
B−rate of change SUPPLIERS A−magnitude of transactions
B R
N = declining 0 = trivial
0 = stable 1 = minor
2 = growing 3 = moderate
4 = growing explosively 4 = important
4 −7 major
R = Flow of requirements info dominates
Figure 3. (Continued ).
Economic Considerations of Measurement 31
32 Foundations of Measuring
the money expended on the whole system. Comparatively benefits, however, are often seen in their historical perspec-
little money goes into calibration, education and research tive as being more significant than the originally argued
of the basic fundamentals of measurement using scientific case; these are the benefits that management would like to
principles and its technology. ‘Fit and forget’ measure- be able to cost in money terms.
ment technology might be the desire of management but,
as with all uses of technology, it is not a reality. Measur-
ing instruments are no exception and they are certainly not RELATED ARTICLES
perfect.
Throughout history, there have been occasions when the Article 3, Measures and Metrics; Their Application,
governing power of nations have considered that certain Volume 1; Article 5, Humans in the Real World,
measurements or measuring instruments would be of great Volume 1; Article 64, Executing A Measuring System
benefit to the nation. Design, Volume 2; Article 65, Life Cycle Concept,
In 1598, the Spanish monarch, Philip III, offered a Volume 2.
large sum of money to any person who could produce
a timekeeper that would reduce navigational inaccuracies
across the Atlantic Ocean regions. In England, Harri- REFERENCES
son was awarded £20 000 from the British Government
who had offered this huge reward, in 1714, for a similar DTI (2003) UK National Measurement System.
reason. URL: http://www.dti.gov.uk/nms/.
On another occasion, James Cook took the voyage to Huntoon, R.D. (1967) Concept of a National Measurement Sys-
southern regions of the globe to measure the transit of the tem. Science, 158, 67–71.
planet Venus in June 1769. This too was concerned with Minck, J. (1977) The National Measurement System – The Hid-
improving navigational accuracy. Here, it was the Royal den Giant. NCSL Newsletter, 17, 3–5.
Society of London that induced the King and the Admiralty NIST (2003) National Institute of Science and Technology, USA.
to expend a considerable sum of money conducting this one URL: http://www.nist.gov.
measurement. NMLTaiwan (2003) National Measurement Laboratory, Taiwan.
In summary, it is relatively easy to make a case for URL: http://www.nml.org.tw/en/NMS/nms− 1.html.
spending money on measurement for clear-cut scientific, NSC (2003) National Standards Commission, Sydney, Australia.
knowledge-seeking reasons. The long-term less quantifiable URL: http://www.nsc.gov.au/PAGES/Nms/nms.html.
5: Humans in the Real World
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
machines may also be referred to as information machines
1 General Remarks on Humans, Machines, and in short or also as technical instrumentation.
Interfaces 33
2 The Overall Human, Machine, and 2 THE OVERALL HUMAN, MACHINE,
Environment Model 33
3 Humans and Technical Instruments – an AND ENVIRONMENT MODEL
Evaluation and Comparison 34
The important interfaces between humans, machines, and
4 Human and Information Machine Model –
their environment are illustrated in Figure 1 (McGhee
The Crucial and Central Measurement Link 36
et al., 1996, 1998). The human biosystem is a complex
5 The Two Problems of Measurement 36 combination of muscles and senses. Muscular activity is
6 Direct and Indirect Measurement 36 associated with effort and effort with work. The practical
7 Inferential Measurement 37 limitations on the ability of human muscle power to work
Related Articles 37 have led to the extension of almost all humans working
References 37 in complex industrial societies using machines that amplify
the human ability to gain mechanical advantage. Simple
examples of this are the spades, pickaxes, and spanners used
in such diverse activities as cultivation, road repair, and
1 GENERAL REMARKS ON HUMANS, motor mechanics. More sophisticated examples used in the
MACHINES, AND INTERFACES same activities, which allow orders of magnitude increase
in the mechanical advantage, are tractors, mechanical
From the earliest times, humankind has attempted to control diggers, and pneumatic spanners. The impact that these
its environment. This compulsion has required the extension energy handling machines have had upon human activity
of the human abilities to sense the environment and so is mirrored in those machines allowing extension of the
to perform useful work upon its materials and substances. human abilities of sensing and perception. For example,
Human-made implements and artifacts are the means by although humans can sense hotness relations through the
which these faculties are extended. A generic name, which tactile and visual senses, they require extension when the
includes all of these items, is embodied in the concept of objects under observation are either very hot or very cold.
machines. There are two main kinds of machines. The first, Thus, the need arose for inanimate implements to extend
which extends the human ability to do useful work, may this human sensing ability. So it is with the other human
be referred to as energy handling or energy transforming senses as well.
machines. A second group of machines to extend human Because of the inextricable link between humans and
sensing, given much emphasis by the present revolution their environment, and between humans and machines, the
is information handling, constituting what may be called ordering of these interfaces helps to explain the links given
information transforming machines. Information handling in Figure 1. Firstly, there is always a direct link between
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
34 Foundations of Measuring
Direct working H
W
E
I
Ability to work Energy
handling Working
M M
Human Machine Universal
M E
biosystem system environment
I I
Information Sensing
Ability to sense handling
H
S
E
Direct sensing
I
Figure 1. Interfaces between humans, machines, and the universal environment.
humans and the universal environment. This interface has Sydenham, 1999), which are freely adapted from Hofmann
two component parts. One of these, which may be called the (1982). The five human senses of seeing, hearing, tast-
human-working environment interface, or (HWEI), explains ing, smelling, and feeling, which involve some 122 million
the activities undertaken by humans in performing use- sensing receptors, can be classified by the energy form of
ful work as part of the process of providing sufficient the observed variable using the COMETMAN acronym.
resources for survival. Related to this, to a greater or lesser The origin of this acronym and the classification of energy
extent, is an interface called the human-sensing environ- are given in Article 109, Structure and Energy in Sen-
ment interface, or HSEI. This embodies the interacting link sor Systems, Volume 2. From the information processing
between the human senses, which interface with the univer- point of view, humans can process information at a rate of
sal environment for the purposes of gathering information. around 100 bps with an unconscious information transfer
From the beginning of the machine era to the present rate of around 1 Gbps from some 4 million nerve strings
day and beyond, an interface has existed between humans linked through the human nervous system to the cortex or
and machines. Known as the human–machine interface, or brain stem.
MMI, it provides the basis for humans to use machines to Figure 2 shows that a biological sensing system is
extend their abilities to perform useful work by acquiring characterized by
some form of mechanical advantage, as well as supplement-
ing the human abilities of sensing and perception. The third • an abundance of sensors also called receptors,
interface, allowing machines to interact with the universal • a very large number of message channels,
environment, is called the machine environment interface, • a parallel processing capacity,
or MEI. • a complex interacting structure with fusion of sensors,
These interfaces between humans, machines, and their • an ability to learn and adapt.
environment give a clear picture of the primary links In the case of technical sensor systems, it is their clear
between them. It also indicates that instruments should function to extend and enhance the human senses. Although
be studied, analyzed, and designed on the basis of how they execute this intention in most cases, Figure 3 shows
humans interact with them and how they interact with the that they are nevertheless characterized by
environment within which they extend the human abilities
to handle information. • a limited range of basic sensors,
• a small number of message channels,
• a series processing of information,
3 HUMANS AND TECHNICAL • a transparent physical structure, which is neither com-
INSTRUMENTS – AN EVALUATION plex nor interacting, nor is there extensive fusion
of sensors,
AND COMPARISON • an as yet underdeveloped ability to learn and adapt.
It is instructive to compare the human sensing system It is fair to say that the advent of large-scale integration
with human-made technical sensors. This comparison is of both analog and digital processing circuits, accompanied
summarized in Figures 2 and 3 (McGhee, Henderson and by the use of microprocessors with large memory capacity,
Humans in the Real World 35
Energy forms Human senses Receptors Nerve strings
107 103
Conscious information
Gustaceptors
C processing = 102 bps
107 106
Olfaceptors
108 106
O
Videceptors
M
106 106
E
Tacticeptors
T
M Unconscious information
104 104 processing = 109 bps
A
Auraceptors
106 106
Nociceptors
N
Figure 2. The human biological sensing system including sight, hearing, tasting, smelling, touching, and pain.
F
Energy forms
Analog (~)
~ ~
PC
~
Analog (~)/Digital (#) converter
T
Analog (~)/Analog (~) modifier
F −Force; T−Temperature
Digital (#)/Digital (#) Modifiers
Intelligence
PC
Digital (#)
Figure 3. Technical sensor systems.
has led to tremendous changes in technical instrumentation. perhaps be renamed as data measurement or DM. In con-
Even so, there is broad scope for further development. temporary measurement, because of the influence of com-
Figure 3 shows that almost all of the measuring system puters, there is an inexorable trend toward the use of a
is actually handling data, even in the strictly analog parts minimum number of analog components. For this reason, it
of the instrument such as the sensor, where an information- seems appropriate to view measurement technology as data
bearing signal is converted into another signal in a different measurement. Data measurement is to measurement as data
energy domain. Hence, measurement technology should communication is to communication.
36 Foundations of Measuring
Human Presentation; Information Data measurement; Universal C
keyboard; signal engineering; O
channel coding; Resources M
Sensing mouse; voice; etc. Channel
Observing Measurement testing; etc. E
Information
Perceiving Contamination T
Creating M
Data capture; sensing;
Displaying; logging; signal processing; A
Biosystem Machine Environment N
VDU; recorder; etc. pattern identification; etc.
Human/machine Machine/environment
interface interface
Human sensing/environment interface
Figure 4. Block diagram of human and measurement information machine in a universal environment.
4 HUMAN AND INFORMATION equilibrium properties of real objects or events. The second
MACHINE MODEL – THE CRUCIAL is to acquire data about the dynamic behavior of these real
objects or events. These may be referred to as direct and
AND CENTRAL MEASUREMENT LINK inferential measurement respectively (Finkelstein, 1994;
McGhee, Henderson and Sydenham, 1999). They may be
Information technology, or IT, has produced radical
referred to as the two problems in measurement. Both
changes through the pervasive influence of the personal
require the use of sensors to measure the states of equilib-
computer, or PC. Each increase in the system clock fre-
rium and motion of solids, liquids, gases, and the systems
quency and corresponding improvement in the information
they constitute. A primary goal in both these measurement
machine input/output, or I/O cards opens up endless mea-
processes is to take account of errors, their sources, and
surement applications.
their quantification (Solopchenko, 1994; McGhee, Hender-
From the functional point of view, the relationship
son and Sydenham, 1999). The first measurement problem
between the information machine aspects of Figure 1 are
is to acquire data regarding the static or equilibrium proper-
given in more detail in Figure 4 (McGhee, Henderson
ties of some real object or event. Acquiring data about the
and Sydenham, 1999). The human faculties of sensing,
dynamic behavior of these objects or events is the second
observing, perceiving, and creating are visualized as being
problem. Thus, real systems, objects, or components may
extended using a measuring information machine. Informa-
be measured from two different points of view known as
tion flows from the human through the MMI, which consists
direct measurement and inferential measurement.
of the PC keyboard, the human voice, the mouse, and so on.
In the reverse direction, information is communicated to the
human using various types of displays, Visual Display Units
(VDUs) as well as data loggers and recorders. The mea- 6 DIRECT AND INDIRECT
suring information machine enhances the human faculties MEASUREMENT
by assisting in the generation of suitable coded interrogat-
ing signals for testing purposes. This process is one of the
Direct measurement per se involves the process of directly
aspects of data measurement. Technical sensors, of the same
sensing those active observable quantities characterizing
general diversity as in Figure 3, enhance the human abilities
the flow of energy and information (Finkelstein, 1994).
to capture information associated with the flow of energy
Figure 3 (McGhee, Henderson and Sydenham, 1999) shows
in the universal environment under observation.
a typical structure for direct measurement using only two
analog sensors and two digital sensors for simplicity.
5 THE TWO PROBLEMS OF A load cell is used to measure an applied force
MEASUREMENT F , while an electrical temperature sensor is used to
measure a temperature, T . Optical methods for digitally
It is important to notice the distinction between the two measuring linear and angular positions are also shown. The
main types of measurement in the above model of measure- instrumentation also includes a diagrammatic representation
ment. Measurement in technical science has two main pur- of multiplexing, amplification, and conversion under the
poses. The first purpose is to acquire data about the static or control of a PC.
Humans in the Real World 37
Indirect measurement uses directly measured quantities Volume 2. Modeling and model building as an essential
to compute the value of another, ostensibly nonmeasur- ingredient of sensor science is justified in Article 107,
able variable. The measuring instrument captures a signal Principles of Sensor Science, Volume 2. Two-space, three-
carrying information about the flow of energy or matter space, and tetrahedron modeling of sensors is described
associated with the system or event. Signal processing is in Article 109, Structure and Energy in Sensor Systems,
used in both cases to extract the information. Volume 2. Signal/Energy matrix modeling of sensors is
considered in Article 110, Signal/Energy Matrix Model-
ing, Volume 2.
7 INFERENTIAL MEASUREMENT Humans have a compelling urge to sense and test. Clas-
sifying and synthesizing signals is a very important activ-
Figure 5 gives a block diagram of the inferential mea- ity in inferential measurement. In its engineering context,
surement problem. All systems may possess any one, or measurement can lead to improved design, since physical
a combination of the characteristics associated with stor- processes become better known. This knowledge also leads
age, transformation, transmission, or dissipation of energy. to more efficient energy utilization and product quality by
These system characteristics determine the dynamic behav- allowing effective control.
ior of the process whose dynamic properties can only
be verified by measurement with the help of interroga-
tive testing, also referred to as inferential measurement
or identification. The dynamic characteristics of a sys- RELATED ARTICLES
tem can be described by models. A model of a physical
process is a representation of its style, form, or pattern. Article 2, Organization of Instrument Science, Vol-
Hence, inferential measurement or identification consists ume 1; Article 6, Substructure of Human–Machine Sys-
of constructing a model of a dynamic system. Modeling tems, Volume 1; Article 95, Human–Machine Interface,
is based upon input and output observations or measure- Volume 2.
ments, as well as on prior knowledge. Dynamic models
have significance in many engineering and nonengineer-
ing systems. Frequently, they play a vital role in the
REFERENCES
proper analysis and design necessary for the efficient
operation of the modeled system. Models are used in Finkelstein, L. (1994) Measurement and Instrumentation Sci-
physics, biology, astronomy, engineering, economics, soci- ence – An Analytical Review. Measurement, 14(1), 3–14.
ology, psychology, medicine, ecology, agriculture, as well Hofmann, D. (1982) Automatic Testing with Intelligent Sensor
as others. A boundary view of the identification problem Systems – Measurement or Classification, in ACTA IMEKO IX
is given in Article 68, Setting the System Boundaries, (p. 545).
McGhee, J., Henderson, I.A. and Sydenham, P.H. (1999) Sensor
Science – Essentials for Instrumentation and Measurement
Ability to store, transform, transmit Technology. Measurement, 25, 89–113.
or dissipate energy or information
McGhee, J., Henderson, I.A., Korczynski, M.J. and Kulesza, W.
(1998) The Sensor Effect Tetrahedron: An Extended Transducer
Space. Measurement, 24, 217–236.
Signal
McGhee, J., Henderson, I.A., Kulesza, W. and Korczynski, M.J.
Test System to be tested
signal (Identified) analysis (1996) Scientific Metrology, ISBN 83-904299-9-3, printed by
A.C.G.M. LODART, Lodz.
Solopchenko, G.N. (1994) Formal Metrological Components of
Figure 5. A block diagram illustrating inferential measurement. Measuring Systems. Measurement, 13, 1–12.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
6: Substructure of Human–Machine Systems
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
The primary functions of instruments inextricably link
1 Boundary Perspective of Human–Machine them with humans (Article 5, Humans in the Real World,
Systems 38 Volume 1). A human–machine system is the context within
2 Resources 38 which men and machines are united in the execution or
pursuit of aims. Hence, human–machine systems provide
3 Information 38
an appropriate starting point for the introduction of instru-
4 Contaminants or Influence Effects 39 mentation. These systems do not exist in isolation. The
5 The Information/Energy Tetrahedron 40 environment, within which they function, regards them as
6 Substructure of Human–Machine Systems 41 subsystems of itself, the wider system. Hence, the bilateral
Related Articles 41 impacts of this linking must be considered. A boundary per-
References 41 spective of these systems, shown in Figure 1 (M’Pherson,
1980; McGhee, Henderson and Sankowski, 1986), allows a
distinction between the three groups of inputs and outputs.
Inputs and outputs in human–machine systems may be
1 BOUNDARY PERSPECTIVE grouped as Resources, Information, or Contamination.
OF HUMAN–MACHINE SYSTEMS
2 RESOURCES
As defined in Article 5, Humans in the Real World, Vol-
ume 1, instruments are human-made systems. Between men Manpower, Money, Machines, and Materials are often
and machines there is an ever-present relation, whose man- called the 4M’s of systems. Manpower is remunerated by a
ifestation alone does not constitute a system. Systems are Money resource, which is also used to purchase Machines
understood to have an opaqueness and obscurity of struc- and Materials. Actuators and tools (forms of energy han-
ture, characteristic of complexity. Although complexity is dlers or transformers) are machines that supplement the
a necessary property, its possession is insufficient for an human capacity to do work. Those that extend the human
object to be classified as a system. The object must also faculties of sensing and perception, belong to the general
have a purposeful function. Hence, a system is a purpose- class of information machines. Raw materials support pro-
fully directed labyrinth of complex operations. It will be duction at all stages and levels. The transformation of input
seen later that a system also possesses recursive proper- resources into consumable output resources is an aim of
ties. Measurement occurs within the context of systems human–machine systems. Electrical power is the most com-
(Article 5, Humans in the Real World, Volume 1). An mon form of resource for modern instrumentation.
exemplary aspect of this systems approach is the totality
of the considerations, which are taken into account. Thus, 3 INFORMATION
systems are examined on the basis of the bilateral influ-
ences and effects, which occur between them and the wider This could be considered as the most important in the
system within which they operate. measurement context. There are three main classes of
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Substructure of Human–Machine Systems 39
Human–Machine
Management
system
Output manifold
Input manifold
Resources Resources
Support Process
Information system system Information
Contamination Contamination
Information
system
System Direct output
contamination
Figure 1. Boundary perspective of a human-machine system.
information. Subjective information is the most nebulous 4 CONTAMINANTS OR INFLUENCE
since it is embodied in those human perceptions resulting EFFECTS
from intuitions, opinions, and feelings. As has been stated
above, technical instruments extend the human faculties of Contamination is properly regarded as implying impurity
sensing and perceiving and so enable humans to acquire and interference. Three types of contaminants (or influ-
Objective information about the physical universe. ence effects) are (i) disruptions, which could have drastic
The theories of Shannon and Weaver (1972), which per- or catastrophic effects due to natural disasters, for example,
mit the treatment of information handling processes with climate, (ii) social factors such as wars, strikes, or prohi-
a higher level of abstraction, provide the means of quan- bitions and (iii) failure of resources due to impurities and
tifying objective information. In this theory, there are two effluents from various sources.
coding theorems, which, although developed for commu- Bilateral Power Flows occur along the interconnections
that link a human–machine system with its environment.
nications instrumentation systems, are also important for
Apart from feedstock-power flows, which are essential for
the other instrumentation systems. One theorem states the
correct operation of the system, other types of power flow
important principles for an economical, coded representa-
can cause interference within the system resulting in poorer
tion of source data, that is, each source alphabet should con-
system performance. This type of contamination may be
tain the minimum number of alphabet symbols. Although
referred to as a disturbance power flow. In most situations,
this alphabet should be used as economically as possi-
as mentioned above, a limitation, called a fundamental limit,
ble in the formation of code words, the code words must
exists. These are of interest in all systems. As they are of
still be uniquely and instantaneously decodable, thus ensur-
special interest in information theory, they are of interest
ing no ambiguity. The other theorem refers to the coding in all information handling processes. If the information
of source coded signals for transmission through inter- handling operation is measurement, then fundamental mea-
ference generating channels. A pretransmission coding of surement limits exist due to the equipartition of energy,
these signals should be such that their statistics optimally which is the most common source of fundamental fluctu-
match the statistics of the communications instrumentation ations. McGhee, Henderson and Sankowski (1986) give a
channel, so allowing high fidelity information transmis- list of references.
sion. Hence, two levels of coding are evident. The limits, Components that are used to build systems, are not
beyond which it is impossible to advance, are usually immortal. The aging of the physical elements gives rise
referred to as Fundamental Limits in all physical processes to a change in their material properties, which can affect
(IEEE, 1981). performance. If burnout occurs, then this type of failure
Fuzzy information, which was first formulated by Lotfi may cause a disruption. Parameters of the system, which
Zadeh (1984) in the 1960s, provides a means of quantify- are instrumental in determining the performance of the
ing the possible meanings for the same group of symbols. system, give the system ‘parameter-sensitivity’. Sensitiv-
It may be concluded that fuzzy events should be treated as ity analysis (Frank, 1978; Lojek, 1982) provides a means
independent, ‘fuzzy Shannon sources’. The average infor- of assessing the effects of all impacts and parameters upon
mation of these ‘fuzzy Shannon sources’ has one compo- the performance of systems. In general, all of these sources
nent associated with the randomness of the two events in of contamination must be considered in instrument systems.
a binary alphabet. The other specifies the average informa- The boundary perspective of human–machine systems pro-
tion resulting from the fuzziness of the fuzzy set relative to vides a basis for highlighting all effects impacting upon
its ordinary or binary set. systems. Hence, it indicates the importance of a detailed
40 Foundations of Measuring
consideration of all causes and effects at the boundaries of of these axes is taken as representing one of the three input
any system (Sandquist, 1985). groups with their COMETMAN classification. This organi-
zation of inputs on the base plane may be regarded as the
input triangle. Such a name is appropriate, as illustrated
5 THE INFORMATION/ENERGY
by the input triangle given by the dotted lines in Figure 2.
TETRAHEDRON Here the support energy form is electrical, the contami-
nation energy/information is chemical, and the measurand
It is also possible to visualize the range of possible infor-
energy form is mechanical using a sensor like a resistance
mation and energy forms for the groups of inputs dis-
strain gauge. The chemical contamination could be respon-
cussed above. A way of providing this view is shown in
sible for parasitic electrochemical potentials at the contacts
Figure 2 (McGhee et al., 1996, 1998) using the COMET-
in the electrical circuit. Each point on the input triangle
MAN acronym. This acronym is one of the ways of
classifying energy types described in much fuller detail axes may be joined to an energy/information form on the
in Article 109, Structure and Energy in Sensor Systems, output axis. When the output is in the form of an informa-
Volume 2. tion bearing electrical signal, the dash-dot line combinations
A spatial representation for the boundary inputs and out- shown in Figure 2 allow the link between all three input
puts of a human–machine system in its most general form forms and the output form to be clearly seen. The geomet-
requires a multidimensional space. As there is a total of rical shape, which is outlined by the dotted and the dash-dot
three groups of inputs and three groups of outputs, the lines, has one base plane surface and three inclined surfaces.
representation becomes too complex for graphical illustra- This represents a tetrahedron or four-faced object. This way
tion. Each of the input and output groups may be specified of representing the boundary impacts on systems may be
as information bearing signals or energy flowing forms. called the information/energy tetrahedron when used in the
Figure 1 greatly simplifies the possibilities by visualizing context of systems. When used in the context of sensors it
the output energy/signal form for an information machine, may be called the sensor effect tetrahedron.
or more specifically a measurement system. In such a sys- Illustration of the manner in which the contamination
tem, the main output is an information output, which also inputs influence the output may be extended to cover con-
corresponds to the resource output. Although contamina- tamination from other energy forms. The dot-dot-dash lines
tion is also produced at the output, it is not necessary to in Figure 2 show the case when the contaminating energy
consider it for the present purposes. form is due to the influence of temperature on the wire
Along the base plane of the spatial representation intro- of the strain gauge. The deep perception allowed by the
duce a topological grouping consisting of three axes. Each information/energy tetrahedron given in Figure 2, which is
Contamination Output energy/
energy/information form information form
Nuclear
Acoustic
N Magnetic
A
Thermal
M
T Electrical
E Mechanical
Optical
M Input energy/
O Chemical
information form
C C O M E T M A N
C
O
Contamination
M
(e.g. Chemical, temperature)
E
T
M Information Information
DM system
A Input Output
N (e.g. Mechanical strain) (e.g Electrical signal)
Support energy/ Support energy
signal form (e.g. Electrical)
Figure 2. The information and energy tetrahedron.
Substructure of Human–Machine Systems 41
Resource input
Maintenance
cycles
Management
system
Contamination
Output
Controls Resource
Support Process
system system Output
Information input Information Information
system Output
Contamination input
Figure 3. Interconnecting substructures in a human-machine system.
fairly straightforward, is still realistic and comprehensive. RELATED ARTICLES
The use of the information/energy tetrahedron in sensor sys-
tems is given a fuller treatment in Article 109, Structure Article 2, Organization of Instrument Science, Vol-
and Energy in Sensor Systems, Volume 2. ume 1; Article 5, Humans in the Real World, Volume 1;
Article 16, Reduction of Influence Factors, Volume 1;
6 SUBSTRUCTURE OF Article 95, Human–Machine Interface, Volume 2.
HUMAN–MACHINE SYSTEMS
REFERENCES
Human–machine systems may be divided into the four
main subsystems that are indicated in Figure 2 and des-
Frank, P.M. (1978) An Introduction to System Sensitivity Theory,
cribed by McGhee, Henderson and Sankowski (1986), Academic Press, London.
M’Pherson (1980) and McGhee et al. (1996). The intercon-
IEEE (1981) Fundamental Limits in Electrical Engineering. Spe-
nection between these subsystems is shown in Figure 3. cial edition of Proceedings of IEEE, 69(2).
A Process system, which can produce a useful resource
Lojek, B. (1982) Sensitivity Analysis of Non-Linear Circuits. IEE
output, provides the means by which the aim of the Proceedings, 129G, 85–88.
human–machine system is realized. Controls and instruc-
McGhee, J., Henderson, I.A. and Sankowski, D. (1986) Functions
tions, generated by a Management system, with the aid of
and Structures in Measurement Systems: A Systems Engineer-
an Information system, are required for the proper coordi- ing Context for Instrumentation. Measurement, 4(3), 111–119.
nation and execution of all system functions at all levels.
McGhee, J., Henderson, I.A., Kulesza, W. and Korczynski, M.J.
The Information system is responsible for acquiring and (1996) Scientific Metrology, ISBN 83-904299-9-3, printed by
processing all of the information required to permit effec- A.C.G.M. LODART, Lodz.
tive operation of the complete system, through the man- McGhee, J., Henderson, I.A., Kulesza, W. and Korczynski, M.J.
agement system. Supplying feedstock and energy is the (1998) The Sensor Effect Tetrahedron: An Extended Transducer
responsibility of a Support system, which also provides Space. Measurement, 24, 217–236.
the human–machine system with its functional means of M’Pherson, P.K. (1980) Systems Engineering: An Approach to
repair and survival. Survivability means that the system Whole-System Design. Radio and Electronic Engineering, 50,
will continue to operate in a hostile environment and will 545–558.
complete its mission or operation. Survivability in the con- Sandquist, G.M. (1985) Introduction to System Science, Prentice-
text of microprocessor based real-time operating systems Hall, Englewood Cliffs, NJ.
is important. An ability for structural modification may be Shannon, C.E. and Weaver, W. (1972) A Mathematical Theory of
required. Reliability analysis provides a method of assess- Communication, University of Illinois Press, Illinois.
ing the probability that a system will complete its mission Zadeh, L.A. (1984) Making Computers Think Like People. IEEE
successfully. Spectrum, 12, 26–32.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
7:Introduction to Measurement Theory and
Philosophy
Ludwik Finkelstein
City University, London, UK
Further, the concept of the property measured must be
1 The Basic Concepts 45 based on empirically determinable relations and not, say,
2 Outline of the Historical Development of on convention.
Measurement Theory 46 This wide definition of measurement is often disputed
3 Nature and Properties of Measurement 46 by those who consider the paradigm of measurement in the
physical sciences as normative, or, at least, they require
References 47
measurement to be a numerical representation in which
Further Reading 47 reflects an order.
For this reason, it is convenient to distinguish between
strongly and weakly defined measurements. Strongly
1 THE BASIC CONCEPTS defined measurement is defined as a class of widely defined
measurement, which follows the paradigm of the physical
Measurement can be defined in the wide sense as a process sciences. In particular, it has precisely defined empirical
of empirical, objective assignment of symbols to attributes operations, representation by numbers, and well-formed
of objects and events of the real world in such a way as to theories for broad domains of knowledge.
represent them, or to describe them. Measurement that constitutes representation by symbols
Description, or representation, means that when a sym- of properties of entities of the real world, based on an objec-
bol, or measure, is assigned by measurement to the property tive empirical process, but lacks some, or all, of the above
of an object, and other symbols are assigned by the same distinctive characteristics of strong measurement, may be
process to other manifestations of the property, then the termed weakly defined (Finkelstein, 2003).
relations between the symbols or measures imply, and It remains to explain the terms theory of measurement
are implied, by empirical relations between the property and philosophy of measurement as used in this introduc-
manifestations. tion.
Objective process in the definition of measurement means Measurement theory is the study of the logical and episte-
that the symbols assigned to a property manifestation by mological basis of representation of property manifestations
measurement must, within the limits of uncertainty, be by symbols.
independent of the observer. Measurement philosophy is the study of the basic con-
An empirical process in the definition of measurement cepts underlying measurement, embracing measurement, as
presented above means, first, that it must be the result of defined above, and concerning the relation of measurement
observation and not, for example, of a thought experiment. to language, reasoning, knowledge, and the like.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
46 Foundations of Measuring
2 OUTLINE OF THE HISTORICAL and model theory, broadly describable as concerned with
DEVELOPMENT OF MEASUREMENT the representation by symbols of a formal language.
Suppes and Zinnes (1965) provided an early, clear expo-
THEORY sition of the theory, in the development of which Suppes has
been one of the key workers. The representational theory
The practical pursuit of measurement arose in the earliest of measurement was then developed and well documented
stages of the material culture of man. However, it was based in a group of important books published between 1971 and
on an intuitive understanding of the process. 1990. The principal works are those of Pfanzagl (1968), a
The ancient Greeks were the first to investigate the very detailed and thorough account of the representational
philosophical foundations of measurement. The schools foundations of measurement in three volumes by Krantz
of Pythagoras, Plato, and Aristotle were concerned with et al. (1971, 1990) and the important treatises of Roberts
the nature of number, quantity, and their relation to the (1979) and of Narens (1985)
real world. The representational theory has now been accepted
The Middle Ages saw much scholarly study of the widely in the social and behavioral sciences. However,
theory of measurement, though scholars were not concerned while it embraces physical measurement, the theory has
with the application of measurement to observation of the not found a recognized place in the theory and practice of
physical world. physical sciences.
The rise of modern science, from Galileo, through New- Modern measurement theory has been debated in the
ton to Lagrange and Maxwell, put mathematics at the measurement and instrumentation technology community
heart of physics, applied measurement to physical observa- since the 1970s.
tion, and developed comprehensive mathematical theories Representational measurement theory has not progressed
of domains of physics. These developments were, however, significantly in recent years, as it seems to have proved
not accompanied by analysis of the fundamental concepts adequate for the purposes for which it has been
of measurement. developed.
Helmholtz (1887) laid the true foundations of the modern However, the rapid development of symbolic compu-
theory of measurement in a thorough logical analysis of tation and soft computing has offered prospects for the
the epistemology of counting and measuring. Hölder (1901) further development and extension of general symbolic
further developed this. representation.
Campbell (1920) provided a lucid and thorough analysis
of the fundamental basis of the measurement of physical
quantities and his theory became generally accepted. 3 NATURE AND PROPERTIES
The classical theory of measurement from Helmholtz to OF MEASUREMENT
Campbell was concerned with physical measurements. The
theory was inapplicable to the social and psychological The properties of measurement arising from its definition
sciences. A report of a committee of the British Association as a process of empirical, objective, assignment of symbols
for the Advancement of Science, published in 1940, which to properties of objects or events of the real world in
considered quantitative methods, rejected the possibility of such a way as to describe them will now be analyzed
psychological measurements. and discussed. They are an explanation of the primacy of
These rigid positions of the classical theory of mea- measurement in science.
surement were broken down by work in the social and Measurement provides an objective description of the
psychological sciences. The concern of the social sciences measurand. It is thus invariant in rational discourse. The
with the concept of utility led to an axiomatic theory of description is not merely a matter of opinion or feeling.
utility that supported development of a new measurement Measurement is based on a well-defined empirical pro-
theory. In psychology, Stevens (1951) carried out much cess of observation. It is thus a basis of justified, true belief;
fundamental work on developing an appropriate analysis in other words it is the basis of true knowledge.
of the nature of measurement. Measurement is not naming. It provides descriptions of
The proceedings of a conference in the United States in relations of the property manifestation measures to other
1959 (Churchman and Ratoosh (1959)) presented a review manifestations of the same property. The value of a mea-
of the classical approaches to measurement as extended to surement process depends upon the richness of the relations
needs of the social and behavioral sciences. it can represent.
Modern formal measurement theory may be said to be Measures are descriptions of great conciseness. A single
based on the work by Tarski (1954) on relational systems number tells us what it would take many words to express.
Introduction to Measurement Theory and Philosophy 47
Measurement gives, further, a description that is precise, Helmholtz, H.V. (1887) Zählen and Messen Erkenntis – The-
pinpointing by a single number a particular entity, where oretisch Betrachet, Philosophische Aufsaetze Eduard Zeller
a verbal description indicates a range of similar but differ- Gewidmet, Leipzig; translated by C.L. Bryan (1930) Counting
and Measuring, Van Nostrand, New York.
ing things.
Measurement is description by a well-defined symbolism. Hölder, O. (1901) Die Axiome der Quantität und die Lehre
vom Mass. Berichte ueber die Verhandlungen der koeniglich
A measure of a property gives us an ability to express facts Saechsichen Geselschaft der Wissenschaften zu Leipzig, Math-
and conventions about it in a formal symbolic language. Phys Klasse, 53, 1–64.
Without the convenient notation of such a language, the Krantz, D.R., Luce, R.D., Suppes, P. and Tversky, A. (1971)
complex chains of induction and deduction by which we Foundations of Measurement, Vol. 1: 1971; Vols. 2, 3: 1990,
describe and explain the universe would be too cumbersome Academic Press, New York.
to express. Narens, L. (1985) Abstract Measurement Theory, MIT Press,
It follows from what has been said that description by Cambridge, MA, London.
symbols is not good in itself. The only value of measure- Pfanzagl, J. (1968) Theory of Measurement, Physica Verlag,
ment lies in the use to which the information is put. Science Würzburg, Vienna.
is not just the amassing of numerical data; it depends Roberts, F.S. (1979) Measurement Theory with Applications to
upon the way in which the data are interpreted, analyzed, Decision Making, Utility and the Social Sciences, Addison-
and organized. Wesley, Reading, MA.
Finally, measurement describes measurands by symbols, Stevens, S.S. (1951) Handbook of Experimental Psychology,
which can be realized as signals. They can be acquired, Wiley, Chichester.
processed, and effectuated by information machines. Suppes, P. and Zinnes, J.L. (1965) Basic Measurement Theory,
in Handbook of Mathematical Psychology (eds R.D. Luce,
R.R. Bush and E. Galanter), Wiley, New York.
REFERENCES Tarski, A. (1954) Contributions to Theory of Models. Indaga-
tiones Mathematicae, 16, 572–588.
Campbell, N.R. (1920) Physics: The Elements, Cambridge Uni-
versity Press, Cambridge; reprinted 1957, Foundations of Sci-
ence, Dover Press, New York. FURTHER READING
Churchman, C.W. and Ratoosh, P (eds) (1959) Basic Concepts of
Measurements, Cambridge University Press, Cambridge. Finkelstein, L. (2000) Foundational Problems of Measurement,
Finkelstein, L. (2003) Widely, Strongly and Weakly Defined in Measurement Science – A Discussion (eds K. Kariya and
Measurement. Measurement, 34, 39–48. L. Finkelstein), Ohmsha Press, Amsterdam (pp. 13–21).
8: Formal Theory of Measurement
Ludwik Finkelstein
City University, London, UK
terms of precisely defined empirical operations. While those
1 Philosophical and Mathematical Basis of the views have run their philosophical course by the sixties of
Formal Theory of Measurement 48 the twentieth century, they remain influential in scientific
2 The Concept of Quality 48 thinking.
3 Representation 49 The other movement on which the formal theory of mea-
surement is founded is the development of modern logic.
4 Some Direct Scales of Measurement 50
From the work of Boole in the mid-nineteenth century, it
5 Uniqueness: Scale Classification and
developed mathematical formalisms, such as the notions of
Meaningfulness 50
symbolic representation, sets, relations, mappings, and the
6 Uncertainty 51 like to represent valid reasoning. It culminated, with respect
References 51 to measurement, in the work of Tarski (1954), on model
theory, concerned with the representation of extralinguistic
systems by symbolic systems.
1 PHILOSOPHICAL AND The theory of measurement defines the concept of mea-
surand quality as an empirical relational system based on
MATHEMATICAL BASIS OF THE
empirically observed data and relations. It maps it into a
FORMAL THEORY OF MEASUREMENT symbol system based on the notions of sets and relations.
It considers the representation of the quality system by the
The formal theory of measurement has its origins in the symbol system and its uniqueness. The foundation of the
needs of psychology and decision sciences. However, its underlying concepts of measurement on empirical observa-
form has been influenced by developments in philosophy. tion gives this theory of measurement a strong validity. The
The approach of the theory is in its essence based on expression of the concepts and reasoning in the formalism
two philosophical movements: logical positivism and the of mathematical logic gives it its rigor.
philosophy of logic.
The formal theory of measurement is presented here
Logical positivism was an important philosophical move-
concisely. Detailed presentations are given in Krantz et al.
ment initiated in the so-called Vienna Circle of philosophers
(1971, 1990), Roberts (1979). A wider discussion is given
in the early twenties of the twentieth century. It sought to
in Finkelstein (2000).
base all knowledge on empirical observation and logical
deductions from them. Central to it was the verification
principle: the notion that the meaningfulness of statements
lies in the specification of the empirical steps required to 2 THE CONCEPT OF QUALITY
verify them. In that view, data obtained by empirical obser-
vation are foundational and knowledge is built bottom-up Measurement presupposes something to be measured. Both
from them. Closely allied to it is the view of operational- in the historical development and logical structure of sci-
ism in the philosophy of science, which defines concepts in entific knowledge, the formulation of a theoretical concept
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Formal Theory of Measurement 49
or construct, which defines a quality, precedes the develop- with the scale of measurement. The danger is that the
ment of measurement procedures and scales. adoption in science of a well defined and restricted meaning
The basic notion is that of a manifestation of a quality, for a quality like intelligence may deprive us of useful
an abstract, a single-sensed aspect of an object or event, insight that the common natural language use of the word
such as, for example, the smell of a substance. Observation gives.
of the real world leads to the identification of empirical
relations among these single manifestations. Examples of
such relations are similarity and difference. As a result, 3 REPRESENTATION
the concept of a quality is formed as an objective rule for
the classification of a collection of empirically observable Consider some quality represented by a relational system
aspects of objects into a single set, together with the family
of objective empirical relations on that set. The resulting Q = Q, R
relational system is a quality, and each single member of
the set is termed a manifestation of the quality. Consider a numerical relational system defined as follows.
We can thus see that there is a difficulty in the mea- Let N represent a class of numbers
surement of such qualities as beauty. The existence and
meaningful use of the word beauty indicates the useful-
N = {n1 . . .} (4)
ness of the concept. However, there is not an objective
rule for classifying some aspect of observable objects as
Let there be on N a family P of relations
manifestations of beauty. Similarly, there are no objective
empirical relations such as indistinguishability or prece-
dence, in respect to beauty. The basis for measurement of P = {P1 , . . . , Pn } (5)
beauty is thus absent from the outset.
When there exists a clearly defined quality, as a set into Then,
which its manifestations can be objectively and empirically N = N, P (6)
classified, together with a set of empirical relations, then we
can always find some symbolical relational set by which it
represents a numerical relational system.
can be represented.
Commonly, N is just the real number line.
Consider some quality and let qi represent an individual
The representation condition requires that measurement
manifestation of the quality, so that we can define a set of
be the establishment of a correspondence between quality
all possible manifestations as
manifestations and numbers in such a way that the relations
between the referent property manifestations imply and are
Q = {q1 . . .} (1)
implied by the relations between their images in the num-
ber set. Formally, measurement is defined as an objective
Let there be on Q a family R of empirical relations Ri
empirical operation M
R = {R1 , . . . , Rn } (2)
M: Q −−−→ N (7)
Then, the quality is represented by an empirical relational
system so that
Q = Q, R (3) ni = M(qi ) (8)
Finally, let us explain the concept of an empirical quantity. such that Q = Q, R is mapped homomorphically into
If there is an order relation in the quality relational system, (onto) N = N, P .
enabling us to order quality manifestations in a way that The above homomorphism is the representation
has formal similarity to the relations equal, greater, and condition.
less, then the quality is termed a quantity. Firstly, it implies that if qi is related to qj by an empirical
In some cases, the concept of a quality arises from relation Rk , that is Rk (qi , qj ), and Pk is the numerical
invariances in numerical laws arrived at by measurement. It relation corresponding to Rk , then Rk (qi , qj ) implies and
will be considered in connection with indirect measurement. is implied by Pk (ni , nj ).
It is usual that once a scale of measurement is established Measurement is a homomorphism rather than an iso-
for a quality, the concept of the quality is altered to coincide morphism because M is not one-to-one, it maps separate
50 Foundations of Measuring
but indistinguishable property manifestations to the same With an empirical ordering operation and an ‘additive
number. combination’ thus established, one proceeds to the setting
up of a scale. A single object with s1 ∈ Q is chosen as stan-
S = Q, N, M (9) dard, assigned the number 1, and chosen as the unit of the
scale. Another object with the equivalent property is then
constitutes a scale of measurement for Q. found and combined by additive combination to produce
nj = M(qj ), the image of qj in N under M is called the the standard 2. This is repeated to produce an extended
measure of qj on scale S. series of integral standards. Fractional standards are gener-
ated by combining two equivalent standards, which when
4 SOME DIRECT SCALES OF combined are equivalent to s1 , and assigning them the num-
ber 1/2, and so on. A property empirically observed to be
MEASUREMENT equivalent to sI is assigned the measure i.
Consider now a matching scale. Such a scale is based on
An attempt will now be made to analyze some qualities as
the establishment on the set of quality manifestations Q of
empirical relational systems and to explain the logical basis
an empirical indifference relation ∼.
of deriving a scale of measurement for them. Extensive
Given Q, ∼, a set of differing elements si ∈ Q are
measurement is the basis of measurement in the strict sense
selected to form a standard set S = {s1 , s2 , . . . , sn }. Num-
and will be considered in detail.
bers (or other symbols) ni ∈ N are then assigned to each
The extensive scales of physical measurement are based
si ∈ S, the same number ni not being assigned to two dif-
on establishing for the quality Q of empirical objects, for
fering elements. The fundamental measurement operation
which a scale is to be determined, an empirical ordering
M of the scale consists of an empirical operation in which
with respect to Q of the class of all objects possessing
measurands qi ∈ Q are compared with members of the stan-
elements of Q, together with an operation o of combining
dard set S. If qi ∼ si , then it is assigned the number ni . An
the objects, elements of, which has, with respect to Q, the
example of this form of scale is a color code in which
formal properties of addition. Such scales are known as
the relation ‘matches’ constitutes the empirical indifference
extensive.
relation.
The above will now be stated more formally.
Consider now ranking scales. In such a scale, an empir-
The basis of a scale of measurement of Q is the definition
ical order system Q, ∼, ≺ is established on Q. A set of
of the set Q.
Secondly, there must be an operational procedure that differing standard objects having si ∈ Q is then selected and
establishes on the set of objects possessing Q an empiri- arranged in an ordered standard series S = s1 , s2 , . . . , sn
cal equivalence relation ∼ and a transitive empirical relation according to Q, ∼, ≺. Numerals are assigned to each si
≺ with respect to Q such that Q, ∼, ≺ is an order system. say i, in such a way that the order of numerals corresponds
Finally, consider objects ω1 , ω2 , ω3 , ω4 ∈ exhibiting to the order in S, of standards to which they are assigned.
property manifestations q1 , q2 , q3 , q4 ∈ Q. Any qi ∈ Q can then be compared with the elements of S
For an extensive measurement scale, there must be an in the same way as in nominal measurement. If qi bears
operation ◦ for combining ω1 and ω2 with respect to q1 the relation ∼ to any si ∈ S, it is assigned the numeral i.
and q2 , which we shall denote by q1 ◦ q2 , with the same If an entity is not equivalent to any si ∈ S, one can deter-
formal properties as addition. mine between which two standard elements it lies in the
For all q ∈ Q, empirical order system.
The best example of a ranking scale of measurement
1. q1 ◦ q2 ∈ Q is the Mohs scale of hardness of minerals in which each
2. q1 ≺ q1 ◦ q2 material successively only scratches those higher than it in
3. q1 ◦ q2 ∼ q2 ◦ q1 the sequence.
4. q1 ◦ (q2 ◦ q3 ) ∼ (q1 ◦ q2 ) ◦ q3
5. if q2 ∼ q3 , then q1 ◦ q2 ∼ q1 ◦ q3
if q3 ∼ q2 then q1 ◦ q3 ∼ q1 ◦ q2 5 UNIQUENESS: SCALE
6. if q1 , q2 , q3 , . . . bear to each other the relation ∼, CLASSIFICATION AND
and q1 ≺ q1 , then there is a number n such that q1 ≺
q1 ◦ q2 ◦ q3 . . . ◦ qn
MEANINGFULNESS
With these definitions, the empirical relation system The requirement that the fundamental measurement proce-
Q, ∼, ≺, ◦ has a structure with the same properties as dure of a scale should map the empirical relational system
the numerical relation system Re, =, <, +. Q homomorphically into the numerical relational system
Formal Theory of Measurement 51
N does not determine the mapping uniquely. There is an Another view of meaningfulness that can be taken is that
element of arbitrary choice in the setting up of scales of only such statements involving measures are meaningful,
measurement. In the case of scales based on additive com- which can be logically traced to the empirical operations
bination, for instance, the choice of the unit standard is on which the measurement is formed.
arbitrary. In the case of a ranking scale of measurement,
the actual numbers assigned to the standards are arbitrary
subject only to the requirement that they should be in the 6 UNCERTAINTY
required order.
The requirement of homomorphism thus defines a class The above discussion of measurement has been given in
of scales that may be called equivalent. The class of terms of deterministic relations and mappings. However,
transformations that transform one member of a class all empirical observations are accompanied by scatter and
of equivalent scales into another is called the class of bias.
admissible transformations. The conditions that admissible The basic empirical observation process is unable to
transformations must satisfy are known as the uniqueness establish exactly the required relations such as indifference
conditions. and order, on which the scale of measurement is based.
We can classify scales by the classes of transformations Thus, for example, the empirical indifference relation is
admissible for them. Let m be numbers representing mea- assumed to be an equivalence relation, which is reflexive,
sures on a scale and let m be the corresponding numbers on symmetric, and transitive. In practice, there is a least
the transformed scale: m = F(m). The generally accepted detectable difference, which means that a1 ∼ a2 does not
classification of scales is as follows: exclude a1 ≺ a2 , or a1 a2 .
Thus, there is uncertainty in the empirical relational
If F() is any 1–1 substitution, the scale is nominal.
system on which the representational measurement is based.
If F() is any monotonic increasing function, the scale is
The problem can be dealt with using the concept of
ordinal.
probabilistic relational systems, or employing fuzzy logic
If F() = am + b, a > 0, the scale is interval.
representation.
If F() = am, a > 0, the scale is ratio.
The problem of the meaningfulness of statements made
about a quality in terms of its measures is important. Such REFERENCES
a statement is meaningful if its truth is unchanged by
Finkelstein, L. (2000) Foundational Problems of Measurement,
admissible transformations of the scales of measurement; in Measurement Science – A Discussion (eds K. Kariya and
in other words, if it reflects the empirical relational system L. Finkelstein), Ohmsha Press, Amsterdam (pp. 13–21).
on which the scale is based and not just the arbitrary Krantz, D.R., Luce, R.D., Suppes, P. and Tversky, A. (Vol. 1:
conventions of the scale. 1971; Vols. 2, 3: 1990) Foundations of Measurement, Academic
Thus, as a very simple example, it is meaningful to speak Press, New York.
of the ratio of two masses, since that ratio is invariant Roberts, F.S. (1979) Measurement Theory with Applications to
with respect to changes of the unit of mass. It is not Decision Making, Utility and the Social Sciences, Addison-
meaningful to speak of the ratio of two hardnesses measured Wesley, Reading, MA.
on the Mohs scale, since that ratio would be changed by a Tarski, A. (1954) Contributions to Theory of Models. Indaga-
monotonic increasing transformation of the scale. tiones Mathematicae, 16, 572–88.
9: Nature and Properties of Measurement
Ludwik Finkelstein
City University, London, UK
on Z to constitute a symbol relational system:
1 Measurement and Other Forms of Symbolic
Representation 52 Z = ZP (2)
2 Special Properties of Measurement as a Form
of Symbolic Representation 53 With F a mapping from R onto P , and M a mapping from
3 Measurement and Information 53 Q into (onto) Z, such that M and F map homomorphically
4 Measurement and Knowledge 54 into (onto) Z, then,
Reference 54
C = Q, Z, M, F (3)
Let zi = M(qi ) be the image of qi in Z under M. Then
1 MEASUREMENT AND OTHER FORMS zi is termed the symbol of (or for) qi under C, and qi is
OF SYMBOLIC REPRESENTATION termed a referent or meaning of zi under C. C is termed
the symbolism or code.
Measurement is only one form of representation of enti- We may extend the concept of general symbolization to
ties by symbols. It is closely related to other forms of the concept of language.
symbolization. A language L may be defined as a subset of all finite
The formal representational theory of measurement based strings, that is, concatenations of elements of an alphabet or
on model theory, outlined here, can be extended to the set of symbols, A. In the context of the symbolization C, the
more general case of representation by symbol systems alphabet consists of the elements of Z and P , supplemented
(Finkelstein, 2000). by function symbols.
A symbol will be defined here as an object or event, A grammar, or syntax, G, is a set of rules, which,
which has a defined relation to some entity, for the pur- operating on the alphabet A, can generate all the strings
pose of eliciting a response appropriate to that entity in of L. An alphabet and grammar constitute a compact
its absence. description of a language.
Let Q be some set of entities and let R be some set of All the considerations of the formal theory of measure-
relations on Q constituting a relational system: ment can be generalized to the symbolization of any rela-
tional systems by any general symbol system and descrip-
Q = Q, ×R (1) tive language.
Formalisms such as symbolic logic and the like enable
Note that Q may now be a set of any objects, events, general symbolic descriptions to be effectively handled.
abstract entities, and so on, and R need not be empirical. The capability of computers to manipulate general sym-
Now, let Z be a set of objects or events to be used as bols has made general symbolic representation techni-
symbols and let P be a set of relations defined or existing cally important.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Nature and Properties of Measurement 53
2 SPECIAL PROPERTIES OF scales, it is only possible to represent the indistinguisha-
MEASUREMENT AS A FORM OF bility, or difference of two entities, and no other relations
between them.
SYMBOLIC REPRESENTATION
Measurement, widely defined, can then be viewed as a 3 MEASUREMENT AND INFORMATION
special form of general symbolic representation. It is, there-
fore, useful to review those properties of measurement If zi is a symbol of or for qi under C, then information
that distinguish it from other forms of symbolic descrip- about qi given zi can be denoted by J (qi |zi ), where
tion. It is useful to note here again the distinction between
strongly defined measurement that is based on representa- J (qi |zi ) = zi , C (4)
tion in the language of mathematics and weakly defined
measurement that may employ representation by more gen- A measure provides information about the measurand.
eral symbolism. In information theory, we consider an information trans-
Measurement is not mere naming, but provides informa- mission channel that transforms symbols xi , elements of a
tion about the relation of a particular manifestation of a symbol set X, acting as inputs into symbols yi , elements of
quality and other manifestations of the same quality. a symbol set Y, constituting outputs. The transformation is
Measurement is an objective description. The symbol in general many to many.
assigned to an entity, given the representation relation, is The quantity of information I about an input xk provided
independent of the observer, within the limits of uncer- by the occurrence of an output yi is given by
tainty of the representation relation. The objectivity of
P (xk /yi )
measurement makes measures invariant and indisputable in I (xk ; yi ) = log (5)
P (xk )
logical discourse.
Measurement is an empirical process. This means first where P ( ) represents probability. The base of the logarithm
that it must be the result of observation and not, for exam- defines the unit of the scale.
ple, of a thought experiment. Further, the concept of the The definition of quantity of information given by infor-
property measured must be based on empirically deter- mation theory and the definition of information based on
minable relations and not, say, on convention. The truth symbolic representation theory are consistent, and indeed
of descriptions based on measurement is thus empirically similar. The information theoretic definition presupposes
justifiable. the concepts of representation by symbols. In both, infor-
The utility of symbolic descriptors depends upon their mation is knowledge about an entity provided by an image
conciseness and the compatibility with further process- of the entity under a mapping. Information theory thus deals
ing. Strongly defined measurement yields measures that are with a restricted class of problems.
numbers, and may be manipulated mathematically. Weakly Measurement is then an information process. Measuring
defined measurement may yield measures that are more instruments are information machines.
general symbols. Their utility depends upon their compati- Information machines are machines or systems of mach-
bility with further processing. While mathematical process- ines that have as their function the acquisition, processing,
ing of data is particularly powerful, there are effective ways outputting, and effectuation of information. They operate by
of processing other symbolic descriptions. transforming input symbols into output symbols by defined
As stated above, measurement is not mere naming. It transformations.
is the representation of a measurand relational system in The signals that carry information in measuring instru-
the real world, by a symbolic relational system. The utility ments may be viewed as sequences of symbols. Regarding
of measurement depends upon the richness of the mea- digital data as a sequence of symbols is intuitively obvious.
surand relational system and the power of the symbolic Analog signals are not so obvious. However, we may view
relational system. In strongly defined measurement, the a continuous analog signal s(t) as a sequence of Dirac unit
measurand relational system is embedded, in general, in a impulses. The elements of the sequence are then symbols.
well formed and complete theory of a domain. The symbol- With regard to symbol processing in information mach-
ization in strongly defined measurement is in the language ines, we may regard the processing as a prescribed mapping
of mathematics, with powerful capabilities of manipulating
the descriptions. In weakly defined measurements, the sym- T: Zi −−−→ Zo
bolization of the domain may not be complete. Further, the
relations representable in the symbolization may be few, from an input symbol set Zi (symbolism Ci ) into or onto
so that the measurement is not powerful. Thus, in nominal an output symbol set Zo (symbolism Co ).
54 Foundations of Measuring
More generally, we may view the transformation as a the theories discussed above. The fact that measurement
prescribed mapping of sentence in an input language Li is a precisely defined empirical and objective process
into sentences in an output language Lo . ensures that it bears a true relation to the fact in question.
These properties also make data that is based on measure-
T: Li −−−→ Lo ment reliable.
We may distinguish between declarative and procedu-
The treatment of measurement as an information process ral knowledge.
and of measuring instruments as information machines is Declarative knowledge is knowledge about ‘that’, and
the basis of systematic approaches to measurement and has been discussed above.
instrumentation science and technology. Procedural knowledge is knowledge about the ‘how’.
It is essentially expressed as transformational rules for
declarative knowledge.
4 MEASUREMENT AND KNOWLEDGE Knowledge based on measurement is essentially declara-
tive knowledge. However, the implementation of the mea-
The result of a measurement constitutes knowledge. surement process is based on procedural knowledge.
The classical view is that knowledge consists of a set of The application of knowledge takes, in general, the form
propositions that are true, that a subject believes to be true, of a problem solving or design process.
and that the subject is justified in believing to be true. The In such a process, the basic operations are an analy-
main problem in this view of knowledge is the nature of sis of requirements, leading to a value model or objective
the justification of the belief. function, possibly multidimensional. Following the gener-
There are theories that provide a basis of justification. ation of candidate solution concepts, which are commonly
Two significant theories in the present context are based knowledge-based symbolic models, the candidates are ana-
on causality and reliability. According to causal theories, lyzed in terms of the value model. On the basis of an
knowledge consists of true belief that bears a true relation analysis of the thus derived values of the solutions, one
to the fact in question. According to reliability theories, or more of the candidates are chosen by a decision process
knowledge requires that it be acquired by a reliable process to be accepted for further development, or implementation.
or method. The value model is based on a measurement-like process
A measurement result is a statement in the form of assigning numbers to attributes. The value model is
however subjective, reflecting the view of the decision-
zk = zl , S (6) maker. It is not measurement.
where zk is the symbolic designation of the measurand,
zl is the symbol assigned by or to the measurand by the REFERENCE
measurement process, and S is the scale on which the
measurement is carried out. Finkelstein, L. (2000) Foundational Problems of Measurement, in
Knowledge obtained by measurement satisfies the con- Measurement Science – A Discussion (K. Kariya and L. Finkel-
stein), Ohmsha Press, Amsterdam, New York (pp. 13–21).
ditions of valid knowledge, meeting the requirements of
Extensions of the Representational Theory of
10:
Measurement
Ludwik Finkelstein
City University, London, UK
here – see Article 7, Introduction to Measurement
1 Development Points of the Representational Theory and Philosophy, Volume 1; Article 8, Formal
Theory of Measurement 55 Theory of Measurement, Volume 1; and Article 9,
2 Indirect Measurement Scales 55 Nature and Properties of Measurement, Volume 1 –
3 Measurement Based on Theory 56 considers measurement scales formed by direct mapping
from a quality relational system to a numerical relational
4 Symbolizations that are not Measurement 57
system. Frequently, however, scales of measurement for
Reference 57 qualities are constructed indirectly through a relation of
the quality to be measured and other qualities, for which
measurement scales have been defined.
1 DEVELOPMENT POINTS OF THE The second problem is that of establishing measure-
REPRESENTATIONAL THEORY OF ment scales for domains of knowledge for which there
are complete, well validated, and grounded theories. The
MEASUREMENT basic representational theory does not rely on any theories
concerning the measurand. However, the establishment of
The formal representational theory of measurement is based
measurement scales in domains for which valid theories
on direct measurement; the representation of particular
exist is generally based on the underlying theory.
attribute manifestations on the basis of an objective empir-
Finally, we have the problem of qualities that have a
ical process that involves no theory either in the formation
number of components, or dimensions. The component
of the concept of the measurand quality, or in its represen-
qualities may be directly measurable in the sense considered
tational process.
above; however, the measurand quality may not. The forma-
The formal representational theory accords with the
tion of measures based on such multidimensional qualities
philosophical perspectives of logical positivism and oper-
is an important practical problem.
ationalism, which base all valid knowledge on empirical
The three problems outlined above will be considered in
observations and logical deductions from them. While both
terms of the representational theory.
these points of view have received much philosophical crit-
A more extended discussion is provided in Finkelstein
icism, they remain an important basis for the approach of
(2000) and the literature cited therein.
the sciences.
Representational theory provides a sound basis for the
foundations of measurement in all its aspects. It neverthe- 2 INDIRECT MEASUREMENT SCALES
less requires extensions to deal satisfactorily with a number
of problems. In direct measurement, scales are formed by direct mapping
The first is the problem of indirect measurement. The from a quality relational system to a numerical relational
exposition of the representational theory that is outlined system. Frequently, however, scales of measurement for
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
56 Foundations of Measuring
qualities are constructed indirectly through a relation of The classical theory of measurement was developed to
the quality to be measured and other qualities for which give an account of measurement in the physical sciences. In
measurement scales have been defined. The reason is, terms of this classical theory, measurement in the physical
generally, the impossibility of setting up a satisfactory sciences is based on the establishment of direct extensive
measurement scale directly. scales of measurement for a number of physical quantities
In its simplest form, consider a case in which every as described here. These quantities are used as the base of a
object that manifests the quality to be measured exhibits system. Scales for other physical quantities are obtained as
a set of other qualities that are measurable. Then, to each derived scales, that is, indirect scales in terms of the base
manifestation of the measurand quality, there corresponds a quantities, in the form of multiplicative monomial functions
set of measures of the associated qualities. These associated of the base quantities.
or component measures can be arranged in an ordered array. This theoretical view does not account for the way in
If manifestations of the measurand quality have identical which measurement units and scales for physical quan-
arrays of component measures, and if and only if they are tities are established. For example, the unit of length in
indistinguishable, then the array of component measures the SI system is now defined in terms of the veloc-
characterizes the measurand. ity of light in vacua defined as standard. The scale of
If we can combine the various component measures temperature is defined in terms of thermodynamic the-
or, in other words, map them into a single number so ory. Thus, the paradigm of measurement that involves
that numbers assigned to the quality manifestations by no theory either in the formation of the concept of the
this process imply and are implied by empirical relations measurand quality, or in its representational process is
between the quality manifestations, then this sets up an inappropriate.
indirect scale of measurement. The situation of physics is that it consists of a number of
Consider as an example the scale of measurement of axiomatized theories such as Euclidean geometry, classical
density of homogeneous bodies. Each such body possesses mechanics, thermodynamics, electromagnetism, and so on.
mass m and volume v (where m and v are assumed to The scales of measurement of classical physics are based
be measures on already defined scales). It is an empirically on the acceptance of these theories as representations of the
established law that objects of the same material, and hence real world and defining the units on that basis rather than
conceptually of the same density, have the same ratio on individual axiomatization and establishment of scales of
(m/v). When different materials are ordered according to particular physical quantities.
our concept of density, they are also ordered according to It is now proposed to outline this formally in terms of the
the respective ratio m/v. Hence, a scale of measurement of processes and notation of formal representational theory.
density is based on the ratio of mass to volume. It is proposed here to define the term theory in the formal
A few observations should be made here. The function terms of model theory. A theory T is then defined as a
m/v that maps measures of mass and volume into measures set of sentences of the language L. It may be possible to
of density is not unique in its order preserving properties present T as an axiomatic system in which all sentences of
with respect to density. For instance, (m/v)2 would be an the theory are clearly derived from a set of axioms that is
equally valid derived measure of density. The form of m/v considered to be true. A fully formalized axiomatic system
is chosen to result in the greatest simplicity of mathematical consists of
relations involving density. The properties of the function
m/v are an idealization of real observations. 1. primitive symbols, (Z), in the notation adopted above;
2. rules for the formation of well-formed expressions (G);
3. axioms;
3 MEASUREMENT BASED ON THEORY 4. rules of inference;
5. a code C establishing the definition of the symbols and
Consider then the establishment of concepts of quantities the interpretation of symbols and sentences.
and of measurement scales in domains of knowledge of the
world, for which there are complete, well validated, and In the strongly defined measurement of physics, L is
grounded theories. the language of mathematics. This establishes the rules for
Physics is one such domain. It is the paradigm for the the formation of well-formed expressions and the rules of
strong definition of measurement. It is proposed to consider inference. The symbols Z are numbers.
measurement in domains for which there are good theories Measurement scales establish a code C constituting the
in terms of the processes of establishing measurement in definition of the symbols and the interpretation of symbols
physics. and sentences.
Extensions of the Representational Theory of Measurement 57
4 SYMBOLIZATIONS THAT ARE NOT meaningfulness of statistics on marks, such as calculations
MEASUREMENT of averages and the like. The conflation of marks, such
as the calculation of weighted sums of marks, contains
The concepts of representational theory can be extended to an element of subjectivity in the conflation scheme, which
assignments of numbers, or other symbols, to properties of probably disqualifies such conflated marks from being con-
objects in such a way as to describe them, but where the sidered measurements.
symbolic descriptions are not measurements. As a final example of descriptive numerical assignments
One example is classification of industries by a standard that have doubtful measurement status, it is important to
industrial classification. Assignments, on such a classifica- consider utility evaluation in formal decision making.
tion, of numbers to enterprises are objective and based on Problems of decision sciences were an important stimulus
an empirical observation. They represent the relation of the to the development of wider concepts of measurement.
enterprise to others symbolized using the same classifica- Formal concepts of utility, with considerable power, have
tion. However, they are not measurement to the extent that been developed. However, it appears, that utility, however
the system of classification is not based on an objective strongly defined, is only a description of the subjective
empirical system of relations of industries. judgment of the decision maker. It is not a measure of any
An important class of the descriptive assignment of objectively defined characteristic of the object evaluated.
numbers, the measurement status of which is problematic,
arises in educational testing. Marks in examinations may
be objective, and are based on an empirical process, but REFERENCE
it is problematic what they measure, other than the per-
Finkelstein, L. (2000) Foundational Problems of Measurement,
formance in a particular test. It is doubtful whether, when in Measurement Science – A Discussion (eds K. Kariya
marks are treated as measures on a ratio scale, they are and L. Finkelstein), Ohmsha Press, Amsterdam, New York
not, in fact, measures on an ordinal scale. This affects the (pp. 13–21).
11:Measurement Theory in Physical, Social, and
Psychological Science
Ludwik Finkelstein
City University, London, UK
describe empirically and objectively the ease with which
1 Measurement Theory in Physical Science 58 a fat can be spread on a surface. In mining practice, there
2 Measurement Theory in Social and arises the need to characterize the hardness of a stratum that
Psychological Science 58 has to be worked upon. In such problems, the measurement
scales formulated are weakly defined. There is, however,
References 59
always the endeavor to investigate the relevant materials
and phenomena in order to model them in terms of estab-
lished theory and to produce strictly defined measurement
1 MEASUREMENT THEORY IN scales.
PHYSICAL SCIENCE Special problems arise in quantum and relativistic
physics, which cannot be handled by the theory of measure-
Measurement in physical science is generally strongly ment presented here. In quantum physics, the interaction
defined. It is based on the mapping of empirically observed between observer and observed system imposes a limit on
relational systems in a physical domain into a symbolical the certainty of the joint measurement of the attributes of
relational system of mathematics (see Article 7, Introduc- the system such as position and momentum. This imposes
tion to Measurement Theory and Philosophy, Volume 1; a fundamental difficulty on the measurement theory. The
Article 8, Formal Theory of Measurement, Volume 1; theory of relativity also has a large impact on the theory
and Article 9, Nature and Properties of Measurement, of measurement in the physical sciences. In terms of the
Volume 1). view presented here it attempts to represent reality using a
In strongly defined measurement theory different from those used in classical measurement
as, for example, in classical mechanics. The rejection of the
1. definition of the quantities is based on the theory; concept of simultaneity and an upper limit to velocity are
2. empirical relational system is also defined by the examples of such differences.
theory; There are thus fundamental developments in the applica-
3. symbolic relational system is rich; tion of measurement theory to the physical sciences to be
4. theory for the domain is complete. undertaken.
However, there are problems in physical measurement A more extended discussion of the problems is given
that present theoretical problems: they arise in measurement in Finkelstein (2000, 2003).
of properties for which there is no adequate theory and also
in quantum and relativistic physics. 2 MEASUREMENT THEORY IN SOCIAL
In technological measurements of physical properties of AND PSYCHOLOGICAL SCIENCE
objects or materials, there often arises the requirement to
characterize properties for which there is no adequate the- It is to be noted that it is the requirements of social and
ory. For example, in testing of food we may need to psychological sciences, which have led to the replacement
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Measurement Theory in Physical, Social, and Psychological Science 59
of the restrictive classical theory of measurement by the There are no wholly axiomatized theories in the social
more comprehensive modern wide definition. and behavioral sciences, which correspond to, say, classical
The social and behavioral sciences are very much con- mechanics or thermodynamics.
cerned with such attributes or qualities as utility, standard Measurement in social and psychological science is thus
of living, alienation, intelligence, and the like. The first generally weakly defined. Such measurement has one or
problem in attempting to measure them is the difficulty in more of the following features:
establishing an adequate objective concept of these qualities
based on empirical relations. The conceptual framework is 1. It is based on a concept of the quality that is often
often absent. fuzzily defined – see Article 12, Fuzzy Approaches
When a scale of measurement for a quality such as ‘stan- for Measurement, Volume 1.
dard of living’ is formed, there remains a fundamental 2. There is significant uncertainty in the empirical rela-
problem of establishing that the measure and concept cor- tional system that it represents.
respond. For example, index figures that are often adopted 3. The symbolic relational system has limited relations
for the purpose of measuring standard of living are dis- defined on it.
puted by those whom they do not suit, as not measur- 4. There is no adequate theory relating the measurement
ing what they consider to be the meaning of standard of to other measurements in the same domain.
living. Finally, it may be stated that in these sciences it is by no
The empirical operations involved in establishing scales means universally agreed that the clear formation of con-
of measurement in social and behavioral judgment, com- cepts in terms of empirical observation, is possible or desir-
monly involve responses by human observers. These are, able. Nor is there agreement that the search for data, through
for example, required to judge whether two stimuli, such measurement, advances knowledge and understanding. The
as light intensities, pitch of sound, and so on, are indistin- opponents of quantification would say that human nature
guishable. As another instance, they are required to give an and behavior are too variable to enable the methodologies
ordinal ranking to a number of alternatives. The data thus of the physical sciences to be applicable to them.
derived may be sufficiently consistent for a population of A more extended discussion of the problems is given
observers to consider their objective; they are nevertheless in Finkelstein (2000, 2003).
subject to considerable random scatter.
The scales of measurement, the social and behavioral
sciences being frequently based on determination of equiv- REFERENCES
alence and order only, are then only nominal or ordinal.
Commonly, some conceptual quality such as the ‘alien- Finkelstein, L. (2000) Foundational Problems of Measurement,
ation’ of a work force in terms of measurable quantities, in Measurement Science – a Discussion (eds K. Kariya
such as worker-days lost through disputes and absenteeism and L. Finkelstein), Ohmsha Press, Amsterdam, New York
(pp. 13–21).
is difficult to establish. The difficulty again is the establish-
ment of agreement between the concept of the quality and Finkelstein, L. (2003) Widely, Strongly and Weakly Defined
Measurement. Measurement, 34, 39–48.
the measures adopted.
12: Fuzzy Approaches for Measurement
Eric Benoit, Laurent Foulloy, Gilles Mauris
Université de Savoie, Annecy, France
from the set of measurements X to a set of fuzzy subsets
1 Definition of Fuzzy Sensors 60 associated with the input, denoted F (Y ).
The inference resulting is an interface that produces a
2 Numerical Fuzzy Sensors 61
new fuzzy subset from the result of the fuzzification, using,
3 Linguistic Fuzzy Sensors 61 for example, a set of rules. The result of the inference
4 Fuzzy Nominal Scales 62 is a fuzzy subset associated with the output. It can be
5 Different Types of Fuzzy Blocks 63 represented by a mapping from the set F (Y ), associated
6 Rule-based Fusion of Information 63 with the input, to a set of fuzzy subsets associated with the
7 Fuzzy Classification Approaches 64 output and denoted F (Z).
Defuzzification is an interface that produces a crisp
8 Interpolation-based Fusion of Information 65
output from the result of the inference, that is, it is a
References 66 mapping from the set F (Z), associated with the output,
to the set U .
The set Y related to the input can either be the set of
numerical values, that is, X, or the set of linguistic values,
1 DEFINITION OF FUZZY SENSORS
that is, LX. In the same way, the set Z related to the output
can either be the set of numerical values, that is, U , or the
Introduced by Zadeh in 1965 the fuzzy subset theory is now set of linguistic values, that is, LU. Therefore, the previous
a common tool in main domains. The basis of this theory definitions can be used to provide an efficient classification,
and its symbology is first defined. A fuzzy subset A of a that is, a typology of fuzzy information handling in four
set X is a set that can contain partially some members of classes as shown in Figure 1.
X. Any member x of X has a membership degree to the From the preceding typology, three types of components
set A. This membership degree is denoted µA (x), and µA can be introduced:
is the membership function of the fuzzy subset A.
The main objective of a sensor is to convert the physical • Fuzzy sensors provide a representation of the measure-
ments by means of fuzzy subsets; they perform the
state of an entity into an information entity (Mari, 2001).
fuzzification.
Fuzzy sensors are defined as sensors producing a fuzzy
• Fuzzy actuators are components that are able to act on
subset as their information entity. The idea of using fuzzy
the physical world depending on the fuzzy subset they
subsets in process control systems arose from Chang and
receive; they perform the defuzzification.
Zadeh’s (1972) study of fuzzy mappings.
• Fuzzy inference components produce new fuzzy subsets
This section provides basic definitions of the three gen-
from fuzzy subsets they have received, thus performing
eral stages involved in fuzzy information systems (Foulloy
fuzzy reasoning.
and Galichet, 1995).
Fuzzification is an interface that produces a fuzzy subset This article focuses on linguistic fuzzy sensors that deal
of information from the measurement; it is a mapping with linguistic fuzzy subsets and on numerical fuzzy sensors
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Fuzzy Approaches for Measurement 61
Fuzzification Inference Defuzzification Thus, the ISO Guide proposed to characterize the mea-
g1 surement result by providing the best estimation of the
Numerical F (X ) F (U )
representation d1 measurand (i.e. in general, the mean value) and the standard
j1 g2
deviation. In fact, it simplifies the probability approach by
X g3 U considering only the first two moments (mean value and
Linguistic j2 variance) of the probability distribution.
g4 d2
representation
F (LX ) F (LU ) Nevertheless, the Guide distinguishes two categories of
uncertainty according to the method used to estimate their
Figure 1. Typology of fuzzy processing in measurement and numerical values: those that are evaluated by statistical
control systems. methods called type A, and those that are evaluated by
other means called type B. Note that for the latter, prob-
that deal with numerical fuzzy subsets. Linguistic fuzzy ability approaches are not more natural than the interval
sensors can be represented directly by means of the ϕ2 approach, when the uncertainty expression comes from
mapping (see Figure 1). They can also be represented experts (Reznik and Dabke, 1999).
indirectly by means of the composition of several mappings In this article, we consider the fuzzy and possibility
as long as the resulting information is a fuzzy subset of theories as an alternative to the conventional probabilistic
linguistic terms. approach (Kendall and Stuart, 1977), and as an extension of
In many situations, information sought is not available the interval calculus (Moore, 1966). Many methods, more
from a single entity, but from many different complemen- or less justified, are proposed in the fuzzy literature to rep-
tary entities. In measurement theory, this case corresponds resent measurement uncertainty (Geer and Klir, 1992; Ree
to the so-called conjoint measurement. Here, it will be and Jager, 1993). The basis of this approach is to represent
called the aggregation of complementary information, and measurement uncertainty by a fuzzy subset (interpreted as
is presented in Sections 4 to 8. a possibility distribution) built to be compatible with the
ISO Guide.
The foundation of this approach is the notion of con-
2 NUMERICAL FUZZY SENSORS fidence interval. Let us first recall this notion: let p be a
probability distribution, and let it be a ‘one point’ estimation
(for example, the mean value of the probability distribution)
The basic idea developed in the design of numerical fuzzy
of the ‘real’ value.
sensors is to take into account and manipulate the measure-
An interval is defined around the ‘one point’ estimation
ment uncertainties. Mauris, Lasserre and Foulloy (2001)
and corresponds to the probability for this interval to
propose a theoretical well-founded approach, compatible
contain the ‘real’ value. Such an interval I1 – α is called a
with the ISO Guide on the expression of uncertainty, for
confidence interval of confidence level (1 – α) (95%, 99%
building a fuzzy subset that is interpreted as a possibility
for example); it is the risk level, that is, the probability for
distribution that represents the uncertainty associated with
the real value to be outside the interval.
a measurement.
Mauris identified each confidence interval of level 1 – α,
According to the ISO Guide prepared jointly by the
with each α-cut of a fuzzy subset F , which thus gathers
International Organization for Metrology; BIPM (Bureau
the whole set of confidence intervals in its membership
International des Poids et Mesures); IEC (International
function. Figure 2 represents this situation.
Electrotechnical Committee); ISO (International Organi-
zation for Standardization); and the OIML (International
Organization of Legal Metrology), the expression of mea-
surement uncertainty must satisfy certain requirements in 3 LINGUISTIC FUZZY SENSORS
order to be widely used at the practical level. The Guide
recommends use of a parametric representation of the mea- Where numerical values are commonly used to repre-
surement uncertainty that sent measurement results some applications would better
manipulate symbolic values or linguistic terms. For exam-
• characterizes the dispersion of the observed values; for ple, a rule-based decision system uses linguistic terms in
example, the standard deviation or a half-width of an its rules. The advantage of using linguistic terms instead
interval at a given level of confidence, of numerical ones is to reduce the number of symbols
• can provide an interval of confidence, which contains involved. Zadeh (1971) subsequently introduced the con-
an important proportion of the observed values, cepts of fuzzy meaning and the descriptor set that are
• can be easily propagated in further processing. extensively used in fuzzy sensors designs. They provide
62 Foundations of Measuring
Possibility degree = 1-confidence level m m(Close) m(Medium) m(Far) md (12)
0 0
1
Pseudo- 1 x (cm) 1
X Close Far LX
triangular 10 12 15
possibility Optimal Medium
distribution possibility
distribution Figure 3. Fuzzy meaning and fuzzy description mechanism.
concepts can be extended to the fuzzy case. It leads to the
0.3 = 1 − 0.7 fuzzy meaning of a term and the fuzzy description of an
e object state, respectively defined by
∀w ∈ W, ∀x ∈ X, µm(w) (x) = µR (x, w) (4)
Nominal value
Interval of confidence at 70% and,
Figure 2. A numerical information entities as a possibility
∀w ∈ W, ∀x ∈ X, µd(x) (w) = µR (x, w) (5)
distribution.
Equivalencies given in (3) become equalities between gra-
des of membership as follows:
a simple means for representing measurement results by
a fuzzy subset of linguistic terms. The conversion from
∀w ∈ W, ∀x ∈ X, µR (x, w) = µd(x) (w) = µm(w) (x) (6)
numeric to linguistic representation presented in the fol-
lowing is called the linguistic fuzzy description or, more Figure 3 shows the link between the fuzzy meaning of each
simply, the fuzzy description. word of the set LX = {Close, Medium, Far} and the fuzzy
Let X be the universe of linguistic discourse associated description of a numerical value.
with the measurement of a particular physical quantity. In
order to linguistically characterize any measurement over
X, let W be a set of linguistic terms, representative of 4 FUZZY NOMINAL SCALES
the physical phenomenon. Denote R the relation defined
on X × W that formalizes the link between items of the Scales have been introduced to model the link between
universe of discourse and the linguistic terms of the lexi- physical states and information entities created by the
cal universe. measurement. Finkelstein (1975) proposed to define any
Two mappings can be defined from this relation. The first scale by the following symbolism:
one is a one-to-one mapping m, called the meaning of a C = X, S, M, RX , RS , F , where
linguistic term. It associates any term w of W with a subset
of X, • X refers to a set of object states and RX is a set of
relations on X,
∀w ∈ W, m(w) = {x ∈ X|xRw} (1) • S refers to a set of information entities, and RS is a set
of relations on S,
The other one is a one-to-one mapping d, called the descrip- • M, called the representation, is a mapping from X to S,
tion of a measurement, which associates any measurement • F is a one-to-one mapping with domain RX and
x of X with a subset of W . range RS .
∀x ∈ X, d(x) = {w ∈ W |xRw} (2) Knowing what kind of scale is used for measurement is
very important. This knowledge defines the set of relations
Since the two mappings are defined from the relation, we and operators allowed for the manipulation of informa-
have obviously the following equivalencies: tion entities.
In the case of nominal scales, RS contains only the
xRw ⇐⇒ x ∈ m(w) ⇐⇒ w ∈ d(x) (3) equality relation on S and RX contains an equivalence
relation on X, denoted ∼. The mapping F is then simply
Consider now that the relation R is a fuzzy relation on X × defined by F = {(∼, =)}. The representation respects the
W . It is characterized by its membership function denoted following equation.
µR . The grade of membership µR (x, w) characterizes the
strength of the link between x and w. Then, the previous ∀xi ∈ X, ∀xj ∈ X, xi ∼ xj ⇔ M(xi ) = M(xj ) (7)
Fuzzy Approaches for Measurement 63
The equivalence relation ∼ defines a partition of the set Comfort sensor
of object states and M associates each item of the parti- Temperature
tion with information entity. The general symbolism for a LF Comfort 1
Humidity LLI
nominal scale is
LF
Comfort 2
C = X, S, M, {∼}, {=}, {(∼, =)} (8) Luminosity LLI
LF
It was shown by Benoit and Foulloy (2003) that the fuzzy LF = Linguistic Fuzzification
description presented in Section 3 can be interpreted as a LLI = Linguistic to Linguistic Inference
nominal scale with the following symbolism. This scale is
named the fuzzy nominal scale. Figure 4. Fuzzy sensor made of fuzzy blocs.
C = X, F (W ), d, {∼}, {≡}, {(∼, ≡)} (9) • Numerical inference: input and output are NFS.
• Numerical to linguistic inference: input is an NFS and
where d is the fuzzy description, and ∼ is a fuzzy equiva- output is an LFS.
lence relation on X, and F (W ) is the set of fuzzy subsets • Linguistic to numerical inference: Input is an LFS and
of W . output is an NFS.
It was also shown that a proximity relation between lexi- • Linguistic inference: input and output are LFS.
cal entities can be deduced from the preceding symbolism, • Numerical defuzzification: it has an NFS as input and
which means that a numerical entity as output.
• Linguistic defuzzification: it has an LFS as input and a
• a fuzzy nominal scale is not only a nominal scale; numerical entity as output.
in addition to equivalence relation, proximity relation
and associated pretopological operators can be used on A measurement system can be created with a graph of
information entities; fuzzy blocks respecting the basic constraint that each end of
• the introduction of fuzziness in measurement can be the connectors must have the same nature and the universe
made only if a proximity relation really exists between of discourse. Linguistic entities must be fuzzy subsets of
physical states. the same lexical set, and numeric entities must be defined
on the same numerical set. Figure 4 shows how this applies
Therefore, fuzzy nominal scales can be useful for tak-
to an example system.
ing into account the fuzziness of a humanlike linguistic
representation, but they must be used carefully.
6 RULE-BASED FUSION OF
5 DIFFERENT TYPES OF FUZZY INFORMATION
BLOCKS
Some inference blocks use linguistic fuzzy subsets for their
Using the preceding approaches, it is now possible to define inputs and their outputs. As the fuzzy nominal scale concept
a set of function blocks that can be merged to create com- is too young to be applied at this level, the linguistic fuzzifi-
plex systems. The classification of the function blocks is cation is usually considered as a simple nominal scale. The
based on the nature of the fuzzy subsets and the nature of only relation that links linguistic entities is an equivalence
the function. Fuzzy subsets can be numerical fuzzy subsets relation. With an equivalence relation, the fusion of infor-
(NFS) or linguistic fuzzy subsets (LFS). Function blocks mation can be made with a set of rules. Generally, most of
can perform the fuzzification, the inference or the defuzzi- the rules are expressed in the following manner:
fication. This classification, issued from the typology pre-
sented in Section 1, distinguishes 8 fuzzy function blocks. If X1 is L1 then Y is a
• Numerical fuzzification: it has a numerical entity as for example, if the size (of a shoe) is large then the price
input, that is, a number or a vector, and an NFS (of a shoe) is high.
as output. In fact, the rule represents a correspondence between the
• Linguistic fuzzification: it has a numerical entity as variable X1 (e.g. the size) and the variable Y (e.g. the
input, and an LFS as output. It performs a fuzzy price). The purpose of this sort of metaimplication is to
description of the numerical input. infer information about Y from information about X1 .
64 Foundations of Measuring
As proposed by Zadeh (1975), the if . . . then . . . metaim- If we have two numeric measurements x1 and x2 , then
plication could be viewed as a fuzzy relation between
X1 and Y represented by the fuzzy graph µ1 (X1 , Y ). µ(E,E ) (L1 , L2 ) = µi(x1 ,x2 ) (L1 , L2 ) = µi(x1 ) (L1 )
Then the image F of a fuzzy subset E of the variable T2 µi(x2 ) (L2 ) (14)
X1 is obtained by the combination/projection principle
(also referred to as generalized modus ponens) accord- Computation of (12) leads to obtaining the description
ing to of (x1 ,x2 ) on LY, that is, µi(x1 ,x2 ) (a) for every a belonging
to LY.
µF (Y ) = µf (E) (Y ) = ⊥X1 (µE (X1 )T µ1 (X1 , Y )) (10) In fact, we have built a new fuzzy sensor that is an
aggregation of the two primary fuzzy sensors.
where T is a norm and ⊥ a conorm. An interesting problem is the determination of the oper-
To compute µF (Y ), we must know µ1 (X1 , Y ) and ators ⊥ and T . This choice is implied by the structure we
µE (X1 ). will obtain for the aggregated information and by the prop-
The problem is first how to build µ1 and secondly how erties of the graph . Such studies have been performed in
to define µE (X1 ). the case in which is a crisp graph (Benoit, Mauris and
The most natural representation of the rules is to consider Foulloy, 1995; Mauris, Benoit and Foulloy, 1996).
that the rules relate the two linguistic terms L1 and a, that
is, to consider that L1 and a are elements of the lexical sets
LX 1 and LY and that 1 is a fuzzy graph on the Cartesian 7 FUZZY CLASSIFICATION
product LX 1 × LY. APPROACHES
In this case, the expression is transformed for every a
belonging to LY into Another approach for fuzzy measurement is to consider
fuzzy sensors as fuzzy classifiers. This approach can be
µF (a) = µf (E) (a) = ⊥L1 ∈LX1 (µE (L1 )T µ1 (L1 , a)) (11) chosen when the initial knowledge about the measurement
is a set of examples instead of an expert set of knowledge.
with µ1 (L1 , a) representing the degree of validity of the The idea of similarity, naturally associated with fuzzy
rule and µE (L1 ) a linguistic description of the variable X1 . sets, has been used mainly in clustering analysis. Indeed, it
Now, if we consider n rules, we have n elementary may be more realistic for describing a data set to look for
graphs i ’s. If we consider that every rule provides a fuzzy clusters when some clusters are not well separated.
relation, a unique graph that represents all the rules could Since Dunn (1973) and Bezdek (1981), the theory of
easily be built by taking = Ui i . If we consider more fuzzy sets has been introduced into several clustering tech-
complex rules of the form niques. In particular, the fuzzy c-means (FCM) and the
fuzzy k-nearest neighbors algorithms (KNN), which are the
if X1 is L1 and X2 is L2 then Y is a, generalization of the crisp ones, have been widely used.
More recently, some authors (Grabisch and Nicolas, 1994),
have used the fuzzy integral in classification problems.
we then consider X1 , X2 , Y as elements of LX 1 , LX 2 and
Introduced by Sugeno (1974), the fuzzy integral has been
LY, and is defined on the Cartesian product LX 1 ×
mostly used as an aggregation operator in problems such
LX 2 × LY .
as decision making. Bezdek and Pal (1992) contains most
The expression becomes
of the seminal works on the use of fuzzy sets theory in the
field of pattern recognition and classification.
µF (a) = ⊥(L1 ,L2 )∈LX1 ×LX2 (µ(E,E ) (L1 , L2 )T1 µ The FCM clustering algorithm is the fuzzy equivalent
(L1 , L2 , a)) (12) of the nearest mean hard clustering algorithm (Duda and
Hart, 1973). Data are supposed to be structured into n
It remains then to define µ(E,E ) (L1 , L2 ). vectors whose dimension is p : Xj ; j = 1,2, . . . , n; each
If we make the assumption that the two variables are vector characterizes an object described with p attributes.
independent µ(E,E ) (L1 , L2 ) is decomposable and we can We assume here that the number of clusters is known.
write For this preselected number c, the FCM algorithm pro-
duces c vectors, which represent the cluster centers, and
µ(E,E ) (L1 , L2 ) = µE (L1 )T2 µE (L2 ) (13) for each data point c membership values, which measure
the similarity of the data points to each of the clus-
(with T2 an operator of intersection, i.e. a norm). ter centers.
Fuzzy Approaches for Measurement 65
Let uij be the membership value of the vector Xj to the This measure gives the importance of each attribute, and
cluster i describing how close Xj is to this cluster’s center every subset of attributes from the point of view of this
Ci . The classification is obtained by minimizing the fol- class in the classification process. This allows us to take
lowing objective function with respect to the memberships into account the redundancy in the information collected
{uij } and the cluster centers {Ci }: or the dependence between two or more features. Natu-
rally, the features, or subsets of features, whose contri-
n
c
bution to the identification of a given class is the great-
(uij )m d 2 (Xj, Ci) est are assigned the greatest measures with respect to
j =1 i=1 this class.
where d is a distance in the space of data. The value of the
fuzzy index m tunes the degree of fuzziness of the clus- 8 INTERPOLATION-BASED FUSION OF
tering. The membership values indicate how well the point INFORMATION
has been classified. When the input is close to a particular
center, the membership value to the corresponding class is An alternate solution to performing a clustering is to create
close to one. the fuzzy meanings with simple interpolation techniques. In
A fuzzy k-nearest neighbors algorithm was proposed this article, we consider initial knowledge about measure-
by Keller, Gray and Givens (1985). The conventional KNN ments. This knowledge is materialized by the meaning of
classification method assigns each input to one of the pos- lexical entities on a small subset V of the measurements
sible classes. First the k-nearest neighbors are found. Then set X.
the input is assigned to the class that includes the majority Then we have to extend the meaning of each symbol
of the neighbors. In the fuzzy KNN classifier, the second on all the measurements set. In order to obtain a fuzzy
step consists in assigning to the input a membership degree partition, the measurement set is cut into n-simplexes. An
to each class. Let x be an input, ui (x) its membership to n-simplex in an n-dimensional space is a polyhedra with
the ith class and wij the one of the j th neighbor zj , then n + 1 vertices. For example, a 2-simplex is a triangle and
2/(m−1) a 3-simplex is a tetrahedron. Then meanings are defined on
k
1 each n-simplex respecting the fuzzy partition constraint.
wij First, the measurements set is partitioned into n-simpl-
j =1
x − zj
ui = exes using Delaunay’s triangulation. The points used to
2/(m−1) (15)
k
1 perform the triangulation are the elements of the set V . As
any triangulation can be used to cut the measurements set,
j =1
x − zj
it must be chosen according to constraints associated with
the measurements. The Delaunay triangulation is preferred
where m is an adjustable parameter that tunes the weighting when no constraint can be found (Benoit, Mauris and
effect of the distance. Foulloy, 1994).
Since the introduction of the Sugeno integral, different The membership function of the meaning of each word
families of fuzzy integrals have been defined. We only is defined on each n-simplex by a multilinear interpolation.
present here the Choquet integral that we use as an aggrega- Suppose the restriction on an n-simplex of the membership
tion operator for multiattribute classification (Grabisch and function of the meaning of a word s is
Sugeno, 1992). Let f be a function from X to [0,1], and
µ a fuzzy measure on X. Assume the elements {xi } are µm(s) (x1 , x2 , . . . , xn ) = a1 x1 + a2 x2 + · · · + an xn + an+1
ordered such that we have f (x1 ) ≤ f (x2 ) ≤ · · · ≤ f (xN ).
The Choquet integral with respect to µ is defined by (17)
The value of this function is known for the n + 1 vertices
N of the n-simplex, therefore the n + 1 factors ai can be
f dµ = (f (xi ) − f (xi − 1))µ(Ai ) (16) calculated by the following equation system:
i=1
x1,1 . . . x1,n 1
where Ai = {xi , . . . , xN } and f (x0 ) = 0 by convention.
A = M −1 B M = . . . ... ... . . .
The {xi } represent the attributes that are taken into x 1
n+1,1 . . . xn+1,n
account for the classification. The {f (xi )} represent the
individual degrees of matching with each of these fea- a µm(s) (v1 )
1
tures. For each class, we define one or more prototypes A = . . . B = ...
(18)
and a fuzzy measure on the set of subsets of attributes. a µ (v )
n+1 m(s) n+1
66 Foundations of Measuring
The ith vertex of the n-simplex is denoted v i , and its j th REFERENCES
component is denoted xi,j .
Once this process is performed on each n-simplex for Benoit, E., Mauris, G. and Foulloy, L. (1994) A Fuzzy Colour
each word, we obtain a fuzzy nominal scale defined on X. Sensor, in Proceedings of XIII IMEKO World Congress, 5–9
This scale is an extension on X of the fuzzy nominal scale September, Torino (pp. 1015–1020).
on V . Benoit, E. and Foulloy, L. (2003) Towards Fuzzy Nominal Scales.
With this method, the knowledge needed for the config- Measurement, 34(1), 49–55.
uration of the sensor is very compact. It can be acquired by Benoit, E., Mauris, G. and Foulloy, L. (1995) Fuzzy Sensor
means of a system called a teacher, which can be a human Aggregation: Application to Comfort Measurement, in Fuzzy
Logic and Soft Computing (eds B. Bouchon-Meunier, R. Yager
or an expert system. During the configuration phase, the and L. Zadeh), World Scientific, London (pp. 219–227).
teacher and the sensor analyze the same phenomenon, and
Bezdek, J.C. (1981) Pattern Recognition with Fuzzy Objective
the teacher gives its description to the sensor. The sensor Function Algorithms, Plenum, New York.
increases its knowledge with its measurement associated
Bezdek, J.C. and Pal, S.K. (eds) (1992) Fuzzy Models for Pattern
with the teacher description. Then it can build the fuzzy Recognition Methods that Search for Structure in Data, IEEE
nominal scale. It should be noted that the description given Press, New York.
by the teacher must be a fuzzy subset of words, which veri-
Chang, S.S. and Zadeh, L.A. (1972) On Fuzzy Mapping and
fies the fuzzy partition condition. If the teacher is a human, Control. IEEE Transactions on Systems, Man and Cybernetics,
then its description is generally a crisp set that contains only SMC-2(1), 30–34.
one symbol. Figure 5 gives the teacher’s knowledge about Duda, R.O. and Hart, P.E. (1973) Pattern Classification and Scene
colors that is expressed in a two-dimensional chrominance Analysis, Wiley, New York.
space and the fuzzy meaning of the color Red. Dunn, J.C. (1973) A Fuzzy Relative of the ISODATA Process and
its Use in Detecting Compact Well-Separated Clusters. Journal
C1 of Cybernetics, 3(3), 32–57.
Finkelstein, L. (1975) Representation by Symbol Systems as an
Red
Extension of the Concept of Measurement. Kybernetes, 4,
215–223.
Yellow Red Magenta Foulloy, L. and Galichet, S. (1995) Typology of Fuzzy Con-
Yellow trollers, in Theoretical Aspects of Fuzzy Control (eds H.T.
C2
Nguyen, M. Sugeno, R. Tang and R. Yager), Wiley, New York.
Neutral
Geer, J.F. and Klir, G.J. (1992) A Mathematical Analysis of
Information-Preserving Transformations Between Probabilistic
and Possibilistic Formulations of Uncertainty. International
Green Blue Blue Journal of General Systems, 20(2), 143–176.
Green
Grabisch, M. and Nicolas, J.M. (1994) Classification by Fuzzy
Integral: Performance and Tests. Fuzzy Sets and Systems, 65,
Cyan 255–271.
Grabisch, M. and Sugeno, M. (1992) Multi-Attribute Classifica-
Teacher,s knowledge tion Using Fuzzy Integral , 1st IEEE International Conference
on Fuzzy Systems, 8–12 March, San Diego, CA (pp. 47–54).
mm (Red )(c1,c2)
Keller, J.M., Gray, M.R. and Givens, J.A. (1985) A Fuzzy
K-Nearest Neighbor Algorithm. IEEE Transactions on Systems,
1 Man and Cybernetics, SMC-15(4), 580–585.
C1
Kendall, M. and Stuart, A. (1977) The Advanced Theory of
0 Statistics, Griffin and Co., London.
Mari L. (2001) The Duality of Sensing and Coding as a Fun-
C2 damental Topic of Information Science, Proceedings of the
World Conference on Systemics, Cybernetics and Informatics
(SCI 2001), Vol. XVI, Orlando, FL (pp. 177–180).
Mauris, G., Benoit, E. and Foulloy, L. (1996) The Aggregation of
Complementary Information via Fuzzy Sensors. International
Membership function of
the meaning of the word Red
Journal of Measurement, 17(4), 235–249.
Mauris, G., Lasserre, V. and Foulloy, L. (2001) A Fuzzy App-
Figure 5. Teacher’s knowledge and fuzzy meaning of the roach for the Expression of Uncertainty in Measurement.
color red. Measurement, 29(3), 165–177.
Fuzzy Approaches for Measurement 67
Moore, R. (1966) Interval Analysis, Prentice Hall, Englewood Sugeno, M. (1974) Theory of Fuzzy Integrals and its Applications,
Cliffs, NJ. Doctoral thesis, Tokyo Institute of Technology.
Ree and Jager (1993) Control-Data Representation at Knowledge Zadeh, L.A. (1965) Fuzzy Sets. Information and Control, 8,
Level, in IEEE International Conference on Systems, Man and 338–353.
Cybernetics, Vol. 4 Le Touquet (pp. 702–706). Zadeh, L.A. (1971) Quantitative Fuzzy Semantics. Information
Reznik, L. and Dabke, D. (1999) Evaluation of Uncer- Sciences, 3, 159–176.
tainty in Measurement; a Proposal for Application of Zadeh, L.A. (1975) The Concept of a Linguistic Variable and its
Intelligent Methods, in XV Imeko World Congress, Osaka Application to Approximate Reasoning. Information Sciences,
(pp. 93–100). Part 1: 8(3), 199–249; Part 2: 8, 301–357; Part 3: 9, 43–80.
Signals, Information and Knowledge, and
13:
Meaning
Qing Ping Yang
Brunel University, Uxbridge, UK
Human cognition is an empirical process, subjective
1 Nature of Cognitive Understanding and to possible a priori propositions and individual experi-
Relationship to Knowledge and Ideas 68 ence (Yang and Butler, 1997). Although the limitation of
2 Definition and Application of Data 68 human senses often results in unsatisfactory knowledge,
3 Definition and Application of Information 69 human perception, and cognition always plays a fundamen-
4 Definition and Application of Knowledge 71 tal role in measurement, in particular, in the definition of a
measurand and its associated measurement scale.
5 Definition and Application of Wisdom 72
Modern sciences have aimed at systematically organizing
6 Definition and Application of a Signal 72 objectively verifiable empirical experience and knowledge
7 Human and Machine Signals 73 as scientific knowledge. Scientific knowledge is often quan-
8 Role of Cognitive Psychology 74 tifiable, precise, concise and thus easy to process and trans-
References 74 form. Measurement plays a central role in the acquisition of
scientific knowledge. Measuring tools, that is, instruments,
are used to extend and improve the power of human senses.
1 NATURE OF COGNITIVE UNDER- The machine perception performed by measuring instru-
ments transforms the identified measurable properties or
STANDING AND RELATIONSHIP TO attributes (i.e. measurands) into some predefined numbers
KNOWLEDGE AND IDEAS or symbols through symbolic representation as in Finkel-
stein (1975). The information carried by these symbols can
The real world exists independently of our perception, and be extracted through data processing, and scientific knowl-
one is aware of the outside real world through seeing, hear- edge about the observed system can then be developed
ing, and other senses. Human perception organizes sensory on the basis of this information. Similar to human cog-
stimulation into percepts of various objects and phenomena. nition shown in Figure 1, this process may be illustrated in
Sufficient experience of external objects and phenomena Figure 3.
enables one to categorize and characterize them by means
of defined properties or attributes. Basic knowledge is thus
formed as ideas, concepts (including conceptual relations)
about these attributes and the associated matters. Figure 1
2 DEFINITION AND APPLICATION OF
shows the basic relation of human experience or cognition. DATA
In fact, human cognition may be split into two processes,
as shown in Figure 2. The first one is essentially a percep- Data is a plural form of datum and generally refers to
tion process, while the second one involves some memory ‘known facts or things used as a basis for inference or reck-
and learning of the empirical relations. oning’ (Pearsell, 1999). In information systems including
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Signals, Information and Knowledge, and Meaning 69
Ideas & concepts measurement, data may be defined as ‘the quantities, char-
acters, or symbols on which operations are performed by
computers and other automatic equipment, and which may
be stored or transmitted in the form of electrical signals,
records on magnetic tape or punched cards, etc.’, accord-
ing to the Oxford English Dictionary (Simpson and Weiner,
2002). Some common examples include numbers, charac-
Human cognition ters, images, and sounds.
Data are one kind of information carrier, and information
is an integral part of data. Data are the mixture or fusion of
carrier and information. Information or meaning can only
be extracted and understood through some kind of data
processing (e.g. brain, computer).
In measurement, data are usually recorded numbers or
Observed system signals stored in various forms. Data have to be collected
and processed before information and knowledge can be
Figure 1. Human cognition. extracted. Typically, data is collected through a sampling
process, both spatial and temporal, as discussed in Yang and
Butler (1997, 1998). They have to be representative, that is,
Concept of measurand Conceptual relations the information they carry should be essentially the same
as or significantly close to that exhibited by the observed
+ system.
It is often necessary to separate data collection or acqui-
sition from data processing, although an integrated process
Perception Learning
would be more desirable and has indeed become increas-
ingly common. The collected data that have only gone
through minimum conditioning and processing are often
called raw data.
+ Since computers can easily process digital data (both
discrete-valued and discrete-time), raw data should ideally
be in digital format. Continuous-time (typically also
Observed attributes Empirical relations
continuous-valued) signals become discrete-time and
Figure 2. Perception and learning in human cognition.
continuous-valued data (or sampled data) through sampling
in time domain. If raw data are already discrete-time, but
continuous-valued, they can be converted to digital data
Measurement information using an (A to D) analog-to-digital converter. The raw
& scientific knowledge data can be simply continuous-time and continuous-valued
signals, for example, recorded on magnetic tape, in which
case both sampling and A to D conversion is required to
generate digital data.
3 DEFINITION AND APPLICATION OF
Machine cognition
INFORMATION
Information may be defined as something or knowledge
communicated concerning some particular fact, subject, or
event. It can be separated from the person informed and can
produce different responses in something, and is capable
of being stored in, transferred by, and communicated to
Observed system inanimate things (Simpson and Weiner 2002).
Information exists objectively and, unlike knowledge, it
Figure 3. Machine cognition. does not require a human subject, although it ultimately
70 Foundations of Measuring
Thorn (1989). Measurement is an information acquisition
process, primarily concerned with the capture (or sampling),
enhancement, extraction, and presentation of information.
Ideally, the information should be preserved as much as
possible in the measuring chain and could be easily sepa-
rated or extracted from the carrier.
Information carrier Information In measurement, information is carried and conveyed
by measurement signals or recorded data. The informa-
Figure 4. Information cell.
tion in a measurement is ultimately knowledge about
the measured object. It is primarily concerned with the
serves human beings. Information becomes knowledge once magnitudes and the relations of the measurand(s), rep-
it is understood and possessed by the informed subject. resented on some predefined scales, as in the repre-
Information cannot exist on its own and it necessarily
sentation theory of measurement discussed in Pfanzagl
requires some kind of information carrier. Information is
(1968), Krantz et al. (1971), Narens (1985), Finkelstein
usually carried and conveyed by signals in various energy
(1982).
forms, for example, mechanical, electrical, optical, and so
The information may be regarded as the meaning of
on. It can also be stored in various media, for example,
the measurement data, and, as such, has to be extracted
printed text, electronic file, and even in the human brain.
using various data transforming techniques. There are two
The relationship between information and information car-
distinct types of data transforming involved in a mea-
rier may be generally presented as an information cell
surement process, as shown in Figure 6. The first trans-
model, as shown in Figure 4.
formation (AB) converts the measurand(s) to signals in
The cell may manifest itself in various forms, for exam-
various forms, which can then be conditioned and enhanced.
ple, electronic signal, data, text, sound, image, brain, and
The central function of this transformation is the gener-
even real-world objects. The cell body denotes the infor-
mation carrier, while the nucleus actually represents the ation of machine percepts or representation of measur-
information carried and conveyed. and(s) in the form of numbers or symbols. The second
transformation (BC) converts the symbols into meaning-
ful information, although the latter also usually takes the
3.1 Information and measurement form of numbers or symbols. The key process of the sec-
ond transformation is calibration. The first transformation
On the basis of the information cell model, machine percep- is data-driven, relating the measurand(s) to the measure-
tion together with human perception may be represented as ment numbers or symbols, whereas the second one is
shown in Figure 5. Measuring instruments therefore essen- information-driven, relating the symbol relations back to
tially function as an information flow channel, or infor- empirical relations.
mation machine as discussed in Draper, McKay and Lees
(1952), Finkelstein (1982) and Sydenham, Hancock and
Numbers/symbols Information/knowledge
Ideas & concepts Measurement information
B C
Knowledge
Information
Information
A
Observed system Observed system
(a) (b) Measurands
Figure 5. (a) Human machine and (b) machine perception. Figure 6. Extraction of measurement information.
Signals, Information and Knowledge, and Meaning 71
4 DEFINITION AND APPLICATION OF Symbols
KNOWLEDGE
Indirect
Knowledge can be generally defined as the familiarity experience
Concepts &
gained by experience of a fact, thing, or person. Knowledge knowledge
is usually in reference to a subject or person, and informa-
tion becomes knowledge as information is interpreted and Machine
possessed by a person (Sydenham, Hancock and Thorn, perception
1989). In fact, human knowledge could also be represented
and stored as information.
From epistemology’s point of view, knowledge can Direct experience
be obtained through direct and indirect experience. Ideas (Human perception)
and concepts are formed from human perception and
from the observer’s direct experience of the real world.
Human knowledge gained also comes from the experi- Observed object or system
ence of the manifestation of the information carried by the Figure 7. Direct and indirect knowledge.
observed system.
Indirect experience can be any experience other than
direct experience. For example, this may be the experience
3rd level of abstraction
of somebody else, or one’s internal experience and mental
activities, that is, reasoning and deduction, which result in Knowledge
rational knowledge. Indirect knowledge should also include
2nd level of abstraction
the knowledge acquired through machine perception. The Data manipulation
role of instruments and measurement is not only extension
of the range and capability of human senses but also the Symbolic data
enhancement of indirect experience and knowledge. Machine perception
1st level of abstraction
Measurement starts with human direct experience, and as
Attributes and propositions
basic concepts of attributes or measurands are formed, some
basic and fuzzy knowledge is developed. Measurement Human perception
serves to quantify and improve this kind of knowledge,
generating, and refining scientific knowledge, which is Real world (objects and phenomena)
usually precise and easy to process and apply.
Figure 8. Knowledge pyramid.
4.1 Direct and indirect knowledge in
measurement
levels of abstraction. In terms of the level of abstraction,
Of course, human knowledge comes from both direct and the relationship between knowledge and human/machine
indirect experiences. Figure 7 shows the combined rela- perceptions can be presented as a knowledge pyramid, as
tions between human perception and machine perception, in Yang and Butler (1997) and as shown in Figure 8.
where the observed object or system is the key link of the The first level of abstraction is the concept of attributes
two perception processes. Since machine perception and and their relations formed after human perception through
measurement may be regarded as a kind of indirect experi- direct experience. The second level of abstraction trans-
ence, Figure 7 also models the interactions between direct forms these attributes to predefined symbols and it repre-
and indirect experiences. It is widely recognized that one’s sents the observed system in symbols. The third level of
direct experience plays an important role generally in his or abstraction further transforms the symbols into information
her cognition process, knowledge acquisition and, in par- and knowledge about the observed system through general
ticular, indirect experience. data manipulation.
4.2 Level of abstraction and knowledge pyramid 4.3 Measurement and other branches of science
Both human and machine perceptions aim at extracting Measurement science requires the underpinning of some
information and knowledge, but the two have different other branches of science, for example, mathematics and
72 Foundations of Measuring
Real Knowledge
Information/
world base
knowledge
Human Knowledge
Measurement
perception refinement
Knowledge
validation
Knowledge
application
Figure 9. Measurement and other branches of science.
other fields of information science including data process- in physics generally involves some kind of measurement,
ing, communication, and control. Measurement also under- including those very advanced ones representing the wis-
pins many other branches of science, for example, physical, dom of mankind, for example, Einstein’s relativity theory
chemical, and biological science. Domain experts rely upon was based upon the measurement of the speed of light.
measurement and instrumentation to study their problems. Having discussed various related concepts like data,
But they work on a larger framework, as shown in Figure 9, information, knowledge, and wisdom, it is common wis-
where the knowledge obtained from measurement about the dom to use simple examples to clarify the relations and
measured attributes help them understand the objects or differences of related concepts!
phenomenon in the system concerned. This understanding Suppose we have received the following parts of texts:
leads to knowledge at a higher level, which may be fur-
ther refined through validation and which constitutes the A: ‘house, prices, UK, rise, fastest, annual, market, last,
science of the subject domain concerned. The objective of peaked, 25.3%, 2002, 1989’.
measurement is knowledge acquisition, whereas those of B: ‘House prices in the UK rose by 25.3% over 2002,
the domain experts are primarily concerned with the direct the fastest annual rise since the market last peaked in
practical experience, knowledge validation, and knowledge 1989.’
refinement (Yang and Butler, 1997). C: ‘Property can be a very good investment in the UK.’
D: ‘What goes up can come down, and usually quite
sharply.’
5 DEFINITION AND APPLICATION OF Part A could best be described as data because it mainly
WISDOM contains individual words and their relations are not explic-
itly given. Part B could best be described as information
Wisdom may be defined as special or advanced knowledge with the relations between words and concepts clearly
that is developed from wider experience and is often not stated. Part C could best be described as knowledge derived
easily or naturally associated or applied. from the interpretation of the information (Part B). Of
As discussed in the previous section, measurement ex- course, additional knowledge can also be developed. Part D
tends our experience and knowledge. It often functions could best be described as wisdom.
as part of an overall system that applies and makes use
of the knowledge generated in measurement. It underpins
many other branches of science. Measurement can thus 6 DEFINITION AND APPLICATION OF A
be a source of intelligence or wisdom, especially in a SIGNAL
complicated system or process in which advanced knowl-
edge is required and/or should be applied. For instance, Signals may be defined as ‘a modulation of an electric cur-
it can be argued that the development of our knowledge rent, electromagnetic wave, or the like by means of which
Signals, Information and Knowledge, and Meaning 73
information is conveyed from one place to another’, accord- humans and may be utilized as human signals. In fact, one
ing to the Oxford English Dictionary (Simpson and Weiner, definition refers to a signal as ‘a sign or notice, perceptible
2002). Signals are closely related to data, and in fact they by sight or hearing, given especially for the purpose of
become data once recorded. But by definition they have conveying warning, direction, or information’ (Simpson
wider existence in natural and man-made systems, and thus and Weiner, 2002).
play a more important role. Like data, they are informa-
tion carriers. Information is carried by the magnitude or
attributes of the variation of signals (Finkelstein, 1994). 7.1 Human signals and machine signals in
A signal is essentially the change or modulation of an measurement
energy state, which can be detected by a human observer
or instrument. Signals can be regarded as factors of energy In instrumentation and measurement, machine signals, and
or power, and energy as the signal carrier. Signals are gener- human signals are closely related in several ways. Firstly,
ated from energy connections and interactions between the human sensation and perception based upon the direct
objects and matter in the system concerned. Various energy detection or sense of human signals are an important and
forms may be involved, for example, electrical, mechanical, necessary part of measurement and knowledge acquisition.
thermal, optical, magnetic, and chemical. Each element of Although they tend to have limited range and power,
the measuring chain can be regarded as a transducer and the human perception can directly lead to the understanding
energy flow may change form in each transducer. Energy of measurand(s) and their empirical relations, which are
flow determines the signal flow in a measuring instrument. the basis of measurement.
Secondly, although machine signals may be quite dif-
ferent from human signals, they are actually often based
upon human signals. This is necessary because they should
6.1 Signal processing
carry and convey the same information. The similarity
may extend to the signal generation mechanisms and
Since the energy associated with a signal is influenced or energy forms.
modulated not only by the measurand but by other effects Thirdly, machine perception extends the range and power
also, noises will always be present in a measuring chain of human senses. Measuring instruments or machines are
together with signals. It is thus necessary to carry out signal able to handle more signals with greater capability. They
conditioning and signal processing to increase the signal- can generate quantified numerical results, which subse-
to-noise ratio. quently lead to refined scientific knowledge.
Signal processing is a technology concerned with the rep- Finally, the output signals of a measuring instrument
resentation, transformation, and manipulation of signals and or machine perception often have to be presented to a
the information they contain, in such a way as to make human observer. Indicating or displaying instruments have
them more useful in a given application. Signal process- been a very important part of a measurement system. As
ing has a long history. The last four decades have seen measurement is often intended for the human observer, the
its extremely rapid growth, partly due to availability of interpretation of the output signals becomes an integral part
the ever-increasing power of devices and computers, both of measurement.
hardware and software, and partly due to its prevalence
in the advances of many branches of science and technol-
ogy (Yang, 2001). 7.2 Signals and semiotics
Signals may be represented and processed in various
domains, for example, time domain, frequency domain, Indeed, as measurement signals may be regarded as sym-
and time-frequency domain. The ultimate goal of signal bols or signs, their interpretation may be explained using
processing is information enhancement and extraction, for a semiotic triadic model or semiotic triangle, in which the
example, noise reduction, classification, detection, estima- meaning of the sign is related to both the specific sign
tion, prediction, feature extraction, and so on. presented and the observer’s empirical experience of the
real-world object.
Similar explanations can be found in measurement. In
7 HUMAN AND MACHINE SIGNALS fact, Figures 6 and 7 presented in the previous sections are
essentially a triadic model and may be referred to as a
Machine signals are predominately used nowadays, and measurement triangle, as shown in Figure 10. Further dis-
they are generated by and also mainly intended for machi- cussions on semiotic issues in measurement are given in
nes. But signals are closely related to the signs intended for Article 114, Models of the Sensor Interface, Volume 2,
74 Foundations of Measuring
Information may even need to be integrated with human senses, for
(or concepts) example, hearing aids.
At a higher level, the understanding of memory, lan-
C guage, reasoning, learning, and decision making can make
a great contribution to information extraction, knowledge
representation, manipulation, storing, and retrieval in mea-
surement. It may also lead to suitable models for intelligent
sensors and instrumentation.
A B
Measured Symbols
object (or numbers)
REFERENCES
Figure 10. Measurement triangle.
Draper, C.S., McKay, W. and Lees, S. (1952) Instrument Engi-
but it is interesting to note here that measurement is pri- neering, Vols. 1–3, McGraw-Hill, New York.
marily concerned with how measurement symbols (signals Finkelstein, L. (1975) Representation by Symbols as an Extension
or data) are related to the observed object(s), whereas semi- of the Concept of Measurement. Kybernetes, 4, 215–223.
otics centers around how the symbols are interpreted. Both Finkelstein, L. (1982) Theory and Philosophy of Measurement, in
of them require the human observer’s direct experience of Handbook of Measurement Science, Volume 1 Theoretical Fun-
the related system or process. damentals (ed P.H. Sydenham), John Wiley & Sons, Chichester
(pp. 1–30).
Finkelstein, L. (1994) Measurement and Instrumentation Sci-
ence – An Analytical Review. Measurement, 14, 3–14.
8 ROLE OF COGNITIVE PSYCHOLOGY
Krantz, D.R., Luce, R.D., Suppes, P. and Tversky, A. (1971)
Foundations of Measurement, Academic Press, New York.
As an empirical science, cognitive psychology is concerned
with mental processes and their effects on human behavior. Narens, L. (1985) Abstract Measurement Theory, MIT Press,
Cambridge, MA.
It studies mental processes such as attention, sensation, per-
ception, memory, language, reasoning, learning, problem Pearsell, J. (1999) The Concise Oxford Dictionary, 10th edn,
Oxford University Press, Oxford.
solving, and decision making.
Human sensation and perception are the basis of mea- Pfanzagl, J. (1968) Theory of Measurement, Physica-Verlag,
Würzburg, Vienna.
surement. As discussed in the previous sections, concepts,
and definitions of measurands as well as their measure- Simpson, J. and Weiner, E. (2002) The Oxford English Dic-
tionary, 2nd edn, Oxford University Press, Oxford. URL:
ment scales are based upon experience and understanding
http://dictionary.oed.com/.
of empirical relations developed from human sensation
Sydenham, P.H., Hancock, N.H. and Thorn, R. (1989) Introduc-
and perception.
tion to Measurement Science and Engineering, John Wiley &
Since human perception is subject to the limitation of Sons, Chichester.
human senses and direct experience, the concept of mea-
Yang, Q. (2001) Introduction to the Special Issue on Signal
surand and its measurement may go through continuous Processing. Measurement and Control, 34(8), 228.
development and evolution. For example, the length stan-
Yang, Q. and Butler, C. (1997) On the Framework of Measure-
dard has been redefined several times to reflect the latest ment Science, in Proceedings XIV IMEKO World Congress,
understanding of the measurand. Vol. 5, Tampere (pp. 101–106).
The study of human sensation and perception could also Yang, Q. and Butler, C. (1998) Object-Oriented Model of Mea-
provide useful examples, references, and knowledge for surement Systems. IEEE Transactions on Instrumentation and
sensing and measurement. In some applications, sensors Measurement, 47(1), 104–107.
14: Hierarchical Aspects of Measurement Systems
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
structural characterization plays a primary role in this sys-
1 The Systems Hierarchy of Measurement tems approach. Thus, the functions and boundary structure
Systems 75 of information machines may be linked to the hierarchical
2 Machines Hierarchy of Measurement Systems 75 structure of the complete system. Technical instrument sys-
tems, also called technical instruments, perform a number
3 Secondary Functions or Orders of
of different functions. Four primary functions, which are
Measurement Systems 77
also shown in Figure 1, distinguish between instruments for
4 Formal Data Measurement Components 77 measuring, calculating, communicating, and controlling.
5 Boundary Perspective of the Four
Information Machines 78
Related Articles 78 2 MACHINES HIERARCHY OF
References 78 MEASUREMENT SYSTEMS
Technical instruments are one of the two types of human-
made machines. It is important to visualize the place
1 THE SYSTEMS HIERARCHY OF occupied by these machines in a hierarchical classification
of entities in the physical universe as illustrated in Figure 2.
MEASUREMENT SYSTEMS This will be recognized as similar to the hierarchy used
in everyday situations by bioscientists and earth scientists
As pointed out in Article 5, Humans in the Real who are interested in the ordering of the variety of living
World, Volume 1; Article 2, Organization of Instru- and nonliving things. The grouping of the physical and
ment Science, Volume 1; and Article 6, Substructure of life sciences allows a clear division between these two
Human–Machine Systems, Volume 1 (McGhee, Hender- important sciences. Two classes or groups of things, which
son and Sankowski, 1986 and McGhee et al., 1996), the may be called superkingdoms, make up the terrestrial
wide view that Systems Engineering permits, highlights hyperkingdom. The superkingdom of living things is made
the placing of information and information systems within up of animals, plants, and other life-forms such as fungi
the structure of human–machine systems. The substructure and microbiological organisms. In the superkingdom of
of human–machine systems (Article 6, Substructure of nonliving things it is possible to identify the separate groups
Human–Machine Systems, Volume 1); (McGhee, Hen- or kingdoms of materials and machines. Materials that are
derson and Sankowski, 1986; McGhee et al., 1996) may important in all engineering activities may occur naturally
be represented by the hierarchical dendrograph or tree dia- as earth and minerals. There is also a diversity of human-
gram given in the classification of Figure 1, where it can made materials.
be seen that technical instruments, also called instrumen- Only the division of the machine kingdom called the
tation, are clearly located as elements within the informa- division of information machines is of immediate inter-
tion subsystem of human–machine systems. Functional and est in this article. These machines may be grouped under
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
76 Foundations of Measuring
Management system four classes of information machines, which allow measur-
ing, calculating, communicating, or controlling operations
to be performed. A significant feature of these diagrams is
Support system Information system Process system
the representation of the calculating function as providing
an important primary support for the other three functions.
Technical instrument Other information
This does not imply that calculation or processing cannot
systems acquisition systems exist independently of the others. Indeed, the emphasis is
meant to indicate the overall importance, which this data
Measuring Computing Communicating Controlling processing function possesses.
instruments instruments instruments instruments
Further analysis indicates that lower levels of classifi-
Figure 1. Instruments in a systems hierarchy. cation can be proposed in the same way as for biologi-
cal systems. Thus, the lower level groups are called the
the subdivisions of deductive or inductive types. Deductive order, family, species, and subspecies of the machine king-
types of machines perform mathematical calculations that dom. The order and family groups will be considered
inexorably lead to only one possible calculated result. On below. Lower orders of this kingdom can be built up by
the other hand, inductive machines use rules for reasoning considering specific kinds of sensors and other elements,
so that logical inference is used to form a conclusion, which that Solopchenko (1994) has called formal data measure-
is not generally numeric. Further distinctions lead to the ment components, or FDMCs.
Hyperkingdoms of the ordering
Levels of the key
The physical sciences The biosciences
Hyperkingdom
Terrestrial things
Superkingdom
Nonliving things Living things
Kingdom
Materials Machines Animals Plants Fungi Other
Division
Natural Man-made
Information Energy
(minerals/earths) Handling Handling
(classes by function/structure/energy)
Class
Measurement Calculation Communication Control
Order
Acquisition Communication Distribution
Information flow
FAMILY
1. Sensing 1. Encoding, encryption 1. Displaying, presentation
2. Conditioning
2. Equalization 2. Logging, recording
3. Screening, guarding
reduction, recovery 3. Data transmission 3. Postprocessing
4. Amplification, filtering, 4. Modulation, 4. Control elements, outputs
preprocessing, conversion demodulation and inverse transducers
5. Identifying, adapting
May be defined in terms of
Formal Data Measurement Components, FDMCs (see Figure 3)
Figure 2. Instruments in a machines hierarchy. From the physical and biosciences to the ORDERS of technical instruments.
Hierarchical Aspects of Measurement Systems 77
3 SECONDARY FUNCTIONS OR ORDERS machines, one of the two types of human-made machines
OF MEASUREMENT SYSTEMS in the physical universe, is given in Figure 2, which has
been built upon the ideas of classification. The relation-
Before any data can be processed, an ACQUISITION SUB- ship between the various key levels of the division of
SYSTEM acquires it. Within the context of measuring sys- information machines as far as the individual Formal Data
tems, further subdivision by function is possible as shown in Measurement Components, or FDMC, is shown in Figure 2.
Figure 2. Hence, a separation of the sensing and transduc- In the present context, they are referred to as Data Mea-
tion functions, from those performed by any conditioning surement components, or DMCs, to emphasize that analog
of the sensor output, is apparent. Conditioning functions and digital signals are used to carry information in the form
may be galvanic isolation, screening, and guarding against of data from the sensor to the output of the instrument.
electromagnetic penetration, circuit completion, and supply Extending Figure 2 allows the classification of FDMCs
of energy to modulating types of sensors. Preamplification, given in Figure 3. It can be seen from this figure that
which is also a subsidiary function appropriate to measure- analog sensors are classified as Formal Analog DMCs,
ment, emphasizes the importance of electronic circuits and or FADMCs, while digital sensors are classified as For-
systems in modern instrumentation (Lang, 1987). mal Analog-to-Digital DMCs, or FA-to-DDMCs. Each of
Information Communication and Distribution are impor- the FDMCs shown in the tree diagram may be repre-
tant from the broader view of measurement in particular and sented by a universal, coherent model (McGhee et al.,
information technology in general. Some subsidiary func- 2001; Solopchenko, 1994), which allows the errors at each
tions appropriate to communication systems are given in stage of a measurement chain to be taken into account.
Figure 2. Aspects of distribution systems, such as data pre- Since such an approach allows a single block diagram
sentation and recording, and human–machine interfaces, to be used to represent widely different DM components,
are relevant for measuring systems. The dendrograph of the primary measurement aim of taking account of errors,
Figure 2 also provides a clear illustration of the place occu- their sources, and their quantification is achieved. The
pied by materials science, which is important for sensors, classification given in Figure 3 is supported by the defini-
in the ordering of the physical sciences. tion of Data Measurement systems in Solopchenko (1994)
and McGhee et al. (2001). It can be seen that measuring
instruments perform a number of functions such as sens-
4 FORMAL DATA MEASUREMENT ing, conditioning, screening, and guarding, amplification,
COMPONENTS and filtering as well as others. These functions may be
described in terms of FDMCs as given in the tree diagram
Measurement systems are formed from the interconnec- of Figure 3.
tion of FDMCs (McGhee et al., 2001; Solopchenko, 1994), The physical world is dominantly analog in nature. Since
which are their subsystems, elements, or building blocks. interfacing with the real world is usually accomplished
The term Data Measurement is used by Solopchenko using analog means, it is clear that there is a group of
(1994). A clear view of the place occupied by information FADMCs. This group of components is appropriately used
(from Figure 2)
Formal Data Measurement Components, FDMC
Formal Analog DM Components formal A-to-D DM Components Formal Digital DM Components
(FADMC) (FA-TO-DDMC) FDDMC
Sensors Linear Correcting Digital Sample-and Operational Operational
conditioners conditioners sensor -hold device procedure function
Modulators Functional Inverse Converters
Self-sustaining converters filters analog-to-digital
cross-converters and digital-to-analog
Analogue
Dummy Galvanic Filter Amplifier switching devices
interfaces isolators (multiplexers)
Figure 3. The classification of formal data measurement components, FDMCs.
78 Foundations of Measuring
when analog information handling operations are to be per- the COMETMAN energy forms described in Article 109,
formed on the information bearing signals. In accordance Structure and Energy in Sensor Systems, Volume 2. In
with the predominance of digital instrumentation, most the case of the measurement system, typical variables to
contemporary instrumentation uses a minimum of analog be measured in different energy forms could be mechanical
circuits. The group of formal components used in this case displacement, fluid flow rate, or thermodynamic temper-
are named FA-to-DDMCs. Thus, all forms of analog-to- ature. As a complete visualization is complex, a simpler
digital and digital-to-analog converters belong within this consideration is appropriate. The information and energy
group. Other components are used for multiplexing pur- tetrahedron, described for systems in Article 6, Substruc-
poses and for sample-and-hold operations. ture of Human–Machine Systems, Volume 1 and more
It has been observed that software is the most important fully for sensors in Article 109, Structure and Energy in
group of components in modern instrumentation systems. Sensor Systems, Volume 2, is an illustration of such a sim-
The two main groups of FDDMCs, may be either opera- plification for the resource, contamination, and energy or
tional procedures or operational functions. signal forms of the input. Again it should be emphasized
that, when this figure represents a measuring system with
electrical information carrying output signals, the impor-
5 BOUNDARY PERSPECTIVE OF THE tance of the disturbance inputs is immediately seen. Con-
FOUR INFORMATION MACHINES tamination of the electrical output signal may be due to any
of the energy forms shown, although only the temperature
A combination of the tree diagrams of Figures 1 and 2 with and chemical form are considered. Thus, the holistic aspect
the boundary view of human–machine systems leads to of the systems engineering approach, when applied to con-
the boundary view of the four information machines given tamination using the information and energy tetrahedron,
in Figure 4. This approach again illustrates the recursive allows the introduction of the concepts relating to error
aspect of the total, or holistic, view, which is a feature sources and error characterization. Solopchenko (1994) has
of the systems engineering method (McGhee et al., 1996; emphasized that taking errors and their sources into account
McGhee, Henderson and Sankowski, 1986). Only one is a primary aim in the analysis and design of measure-
output is shown in these diagrams. However, the totality ment systems.
of influencing impacts is evident.
A distinction is also seen between the control instrumen-
tation system and the other three. Power flow is ultimately RELATED ARTICLES
modulated or regulated by information feedback in con-
trol instrumentation systems, whereas power flow supports Article 2, Organization of Instrument Science, Vol-
information flow in the other three. The information input ume 1; Article 5, Humans in the Real World, Volume 1;
or output at the system boundary may also be in any of Article 6, Substructure of Human–Machine Systems,
Volume 1.
Disturbance Power
Information Information REFERENCES
Measurement
input output
Lang, T.T. (1987) Electronics of Measuring Systems, John Wiley
Disturbance Power
& Sons, Chichester.
Information Information McGhee, J., Henderson, I.A. and Sankowski, D. (1986) Functions
Calculation
input output and Structures in Measurement Systems: A Systems Engineer-
Disturbance Power ing Context for Instrumentation. Measurement, 4(3), 111–119.
Information Information McGhee, J., Henderson, I.A., Kulesza, W. and Korczynski, J.
Communication (2001) Measurement Data Handling: Theoretical Techniques,
input output
ISBN 83 7283 007 X, A.C.G.M. Lodart S.A., Lodz.
Disturbance Power input
McGhee, J., Korczynski, M.J., Kulesza, W. and Henderson, I.A.
Information Power (1996) Scientific Metrology, ISBN 83 904299 9 3, A.C.G.M.
Control Lodart, Lodz.
input output
Solopchenko, G.N. (1994) Formal Metrological Components of
Figure 4. Boundary perspective of the four information machines. Measuring Systems. Measurement, 13, 1–12.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
15: Typical Measurement Systems Architectures
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
before the output of the measuring system is used to pro-
1 Generalized Architecture of a Measurement vide a soft copy in a display, a hard copy in a plotter, or
System 79 printer, or a stored record in a data-logger.
2 Microprocessor-based Instrumentation 79
3 Process Control Instrumentation 80
2 MICROPROCESSOR-BASED
4 Petroleum and Gas Process Instrumentation 82
5 Supervisory Control and Data Acquisition INSTRUMENTATION
System 82
6 Remote Transmitter Units 82 The hierarchical subdivision of measurement systems
reveals the existence of four classes of Information Mac-
Related Articles 83
hines (Article 14, Hierarchical Aspects of Measurement
References 84 Systems, Volume 1; McGhee et al., 1996; McGhee,
Henderson and Sankowski, 1986) that perform the primary
functions of
1 GENERALIZED ARCHITECTURE OF A
1. measurement,
MEASUREMENT SYSTEM 2. calculation,
3. communication,
A measuring system has the general structure shown in 4. control.
Figure 1 (McGhee et al., 1996). There are three main divi-
sions within the system. Input elements are primary sensing Essential support to the remaining three classes of
elements and transducers. This distinction is not trivial as information machines is provided by the calculating
there are many cases in which the sensing element is fash- machine. They may also be used as central elements
ioned for one energy domain, while the transducer as its in microprocessor-based instrumentation (Article 158,
name implies converts the output of the sensing element Microcontroller Systems, Volume 3), as shown in
into a signal in another energy form. The intermediate Figure 2. Note the different structures at the input end
elements are usually electronic circuits and systems that of the system (Lang, 1987; McGhee et al., 1996, 2001;
make extensive use of large scale integration and possi- McGhee, Henderson and Sankowski, 1986). Among the
bly distributed computing, where microprocessors and per- many elements, it is important to note the sample-and-hold,
sonal computers are dispersed throughout the system. Data or S&H, units, the multiplexers and the data converters. The
transmission and telemetry methods would be used if the first p inputs are analog signal inputs, which are analog
measuring point is remote from the observation or display multiplexed before being sampled and held, or S&H. This
point. Signal recovery/detection may be necessary if the S&H operation is sequentially performed on each signal.
information is transmitted using a suitable form of modula- By contrast, each of the analog inputs from (p + 1) to q
tion. Output elements could consist of some postprocessing are synchronously sampled and held before conversion. A
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
80 Foundations of Measuring
Input elements Intermediate elements Output elements
*Observation
*Information acquisition +Data
sensing *Utilization
Information flow display
(Signal transformation) *Processing
*Interpretation +Inverse
+Primary
+Data transducer
sensing +Transducer
element processing
+Data
logging/
*Data acquisition *Transmission and *Data recovery recovery
communication and reduction printing
+Signal +Data
conditioning transmission +Signal
and/or and recovery/
conversion telemetry detection
Principal sub structures; *Primary functions; +Subsidiary functions
Figure 1. Generalized architecture for a measuring system. (Each system must be examined to allow consideration of particular features
such as the absence of certain elements or the existence of more complex interconnection of structures).
Input 1
S A/F M
U S/H
Input p X
M A Motor
A
U D or
Input (p +1) X C
D
S A/F S/H M actuator
U
Input q X
M Plotter
A Micro
Input (q +1) U
processor
S A/F S/H D X
C M
Printer
U
A X S – Sensor
Input r A/F – Amplifier/filter
D
S/H – Sample-and-Hold Display
C
MUX – Multiplex
ADC – Analog-to-digital converter
Input (r +1) DAC – Digital-to-analog converter
S A/F S/H M
U
Input s X
Figure 2. Architecture of microprocessor-based instrumentation.
possibility, which is shown for inputs (q + 1) to r, is for Acceleration in Computers, Volume 3; and Article 158,
individual channels to consist of a sensor, an amplifier/filter Microcontroller Systems, Volume 3, process information
an S&H and a dedicated converter. All of the (r + 1) to using binary operations. These binary operations may
s inputs are assumed to be binary signals, which may be belong to either the arithmetic group of operations or the
generated from switch contacts or other event detectors. Boolean group of operations. Arithmetic operations are
In these lines, the amplifiers/filters may also include some addition, subtraction, multiplication, and division. Boolean
kind of binary signal conditioner such as a Schmitt trigger. operations consist of conjunction, or logical AND, disjunc-
Calculating machines, more commonly referred to as tion, or logical NAND, and exclusive OR, or logical XOR.
computers, minicomputers or microprocessors, which are Microprocessors also allow the generation of diverse signal
considered in Article 152, Fundamentals of the Stored forms for use in system testing.
Program Digital Computer, Volume 3; Article 153, Sin-
gle Address Instruction Microcomputer, Volume 3; 3 PROCESS CONTROL
Article 154, Internal Operation of the Microprocessor, INSTRUMENTATION
Volume 3; Article 155, External Operation of the Micro-
processor, Volume 3; Article 156, Memory Management A Process System is a complex combination of men,
in the Microprocessor, Volume 3; Article 157, Data money, machines, and materials (Article 5, Humans in
Typical Measurement Systems Architectures 81
the Real World, Volume 1, McGhee, Henderson and this comparison, it generates a control signal, which is
Sankowski, 1986) with the specific aim of producing then amplified, if necessary, before being applied to an
some resource output having enhanced or surplus value. actuator and final control element. The actuator and final
Instrumentation is required in process systems (Sydenham, control element cause a manipulation that appropriately
1983) to assist the humans involved in operating the process regulates the flow of energy in the form of say a fluid-
for assessing process behavior and so extending and refining flow rate, a pressure, a level, a temperature, or a chemical
their ability to monitor and control it. composition.
In most cases, humans will be replaced by instruments A loop will also consist of some means of monitoring
for the following: its behavior by displaying the value of the set-point, the
error, and any operational limits. Such facilities may also
• Measurement – to provide more accurate, reliable data.
include a graphical mimic diagram of the process, as
• Calculation – to process large quantities of data more
well as appropriate means for logging and recording of
quickly and efficiently.
• Communication – to transmit information for a wide process data. Status, alarm, and warning indicators are also
range of uses, frequently between remotely separated normally included.
transmitters and receivers. Some severe problems, which must be dealt with using
• Control – to improve the quality of the process resource process instrumentation, are as follows:
output with a minimal involvement of humans
A schematic of a typical process control measure- • Generally, the instruments of whatever class must
ment loop is shown in Figure 3. There are a number remain in operation for 24 h every day with atten-
of identifiable stages of a process control system. The tion only being provided during scheduled mainte-
most commonly occurring structure is called a loop. In nance periods.
a loop, a sensor, also called a gauge, a primary trans- • The reliability of the instruments must be high to ensure
ducer, or pickup, measures a physical variable whose value that they perform their specified function. Failures cause
is to be controlled. As this signal is rarely in a form lost production, which can be compounded if the failure
that is convenient for easy handling, it must be condi- endangers the safety of the plant or more seriously of
tioned by using some appropriate technique. Such con- the personnel or of the environment.
ditioning is most commonly performed using electronic • Instrumentation may operate in a wide range of envi-
circuitry consisting of amplifiers, filters, and data convert- ronments. They may operate in hazardous locations.
ers. All of this circuitry must be designed to be intrinsi- They may also be subjected to extremely high or low
cally safe if the instrument is to be used in a hazardous temperatures, dry or humid. Under failure conditions,
location. the instrumentation should always ‘fail-to-safety’. This
Subsequently, this conditioned signal is transmitted or means that hazardous processes should be rendered safe
relayed to a controller. The controller compares the con- in a controlled manner, and in such a way as to present
ditioned received signal with the process set-point. From no dangerous occurrences or conditions, which could
Hazardous area
Transmission Control
(Sensors, actuators, driving power and signal transmitters)
link across room
All equipment in this area must not store sufficient energy
to cause sparking, which may lead to explosion. the plant (Nonhazardous area)
A/F/P – amplification/filtering/processing
Controller
Conditioner/ Safety Indicators
System Sensor A/F/P Display
transmitter barrier
under Logger
control Printer
Transmitter in hazardous area
Actuator Safety
Power amplifier
barrier
Figure 3. Architecture of a process control measurement loop.
82 Foundations of Measuring
Field storage Crude/products system
Wellhead
Crude storage
Processing Products storage
Gas system Mainframe at
District office Gathering/ control center
Wellhead storage
Pump Compressor Delivery
Products terminal station station points
District office
Location of District office
microcomputers
District office Compressor station
Line valve Metering/measuring station Line sensors
Figure 4. Schematic diagram of the instrumentation in a crude/petroleum products system and a gas process system.
lead to damage of equipment or, more seriously, to 5 SUPERVISORY CONTROL AND DATA
injury or death of personnel. ACQUISITION SYSTEM
To assist in the overall control of the process system and
4 PETROLEUM AND GAS PROCESS its distributing pipeline requires the incorporation of an
INSTRUMENTATION important system known as a Supervisory Control And
Data Acquisition, or SCADA, system. Figure 5 illustrates
the block diagram of a typical SCADA system. The
Computer-based instrumentation is now regarded as indis-
master station includes fully redundant computers that
pensable in these industries. Figure 4 shows the main parts
communicate over suitable links such as microwave or
of the instrumentation for both petroleum and gas pro-
cable links to Remote Transmitter Units (RTUs).
cess systems. Although petroleum and gas process systems
The master station instrumentation also consists of what
handle different materials, they have the following com-
is called a Human–Machine Interface, or MMI. This
mon features:
interface consists of Visual Display Units, or VDUs,
• Both have wellheads, which are used to extract the printers, and data loggers. Other graphic facilities will
raw resources. include a hard mimic hydraulic profile of the process and
• Secure storage facilities are required in both cases. its associated pipeline. This MMI also includes a control
• Distribution is achieved through piping, which may panel for process/pipeline dispatchers. It is also usual to
be extensive. include redundant communication channels between the
• Security against dishonest fiscal accounting and custo- various pumping/compressor stages of the pipeline, and also
dial delivery is essential. between the various important elements of the process.
Master station computers process all-important data at
The figures emphasize the important role played by regular intervals so that hourly or daily reports may
computers as well as the diversity of the systems and be printed automatically. Data to be printed on these
noteworthiness of their distributed nature. There are many summary reports may be inventories of process/pipeline
remote measuring points. These may be in locations, performance, volumes of various materials received and
such as at the wellhead storage, where large numbers of delivered. Volumes and leakage information may also be
temperature, level, flow, and pressure variables need to be indicated. This data may be used for direct invoicing.
measured. It is necessary to integrate these measurements
to provide effective control of such complex systems.
Pipelines, which are an indispensable part of both petroleum 6 REMOTE TRANSMITTER UNITS
and gas process systems, require supervision and control
just as much as the process systems themselves. Indeed, Remote Transmitter Units, which are microprocessor-based
the pipeline system could be considered a complex process units for digitally multiplexing data before transmission to
in its own right. the master station, are connected in a SCADA system as
Typical Measurement Systems Architectures 83
Status Battery backup
Alarms
Prog Direct memory access Prog Local CRT with
Disc Disc Controls IRTU No 1 keyboard
I/O DMA DMA I/O Analog
Standby Standby Set-point Valve
computer computer SCADA system
Peripheral VHF/
switch microwave
Microwave Modem Modem
Mag tape
VHF/
Computer equipment Status
microwave
Man–machine interface Alarms Valve Modem
Control RTU No 1
Alarm, event Analog Battery backup
CRT with CRT with
& report
keyboard keyboard
printer Battery backup
Status
Computer room Alarms
Communications interface Local CRT with
Controls IRTU No 2 keyboard
Analog
Comms Comms Comms Set-point Valve
controller controller controller SCADA system
VHF/
Modem Modem Modem Microwave Modem Modem microwave
Status VHF/
Microwave Microwave microwave
system system Alarms Valve Modem
Control RTU No 2
Analog Battery backup
To corporate data processing
(a) (b)
Figure 5. A Supervisory Control and Data Acquisition, or SCADA, system: (a) computer equipment, human–machine interface and
communications interface, (b) field located Intelligent Remote Transmitter Units (IRTUs) in a SCADA system.
shown in Figure 5. It is also possible that these RTUs are fire and smoke detection, high sump level, and many
independent, in so far as they implement local supervision, more. Of the many control actions, the most common are
data acquisition, and control in their own right. RTUs may starting and stopping pumps and motors as well as opening
also have some intelligence, in which case they are referred and closing valves. Others are set-point control of many
to as IRTUs. system variables such as valve throttling, turbine speed,
Typically, an RTU may acquire data from pumping or vessel fluid level, pressure, temperature, flow rate, and
compressor stations on a pipeline or perhaps from a local discharge pressure regulation, not forgetting station reset
process. Variables such as suction, pump case temperature, or shutdown.
discharge and process temperatures and pressures are typ- Technology convergence has meant that programmable
ical of the wide range, which are monitored. Flow rate is controllers and/or IRTUs now almost exclusively perform
also important, especially if the discharge is an origin or many process operations involving logical sequence con-
destination location for custodial transfer of the product. trol. Continuous analog control uses direct digital control
As volume accumulation is also important, then tank lev- strategies, which provide fast control response while pro-
els and temperatures must be measured, so that temperature cessing or in transit. Design changes are easily implemented
compensated tank volumes can be computed both locally using these programmable controllers, with downloading of
and at the master station. Leaks may also be detected if program revision to the RTUs by the master station.
flow rates are monitored at separate locations in the prod-
uct flow system. Measurement of energy consumption in
kilowatt-hours is also essential. RELATED ARTICLES
Status and alarm signals are required to ensure safe
and reliable operation of the system. It is normal to Article 14, Hierarchical Aspects of Measurement Sys-
indicate the open or closed status of valves, on or off tems, Volume 1; Article 16, Reduction of Influence Fac-
status as well as high levels of vibration, shaft bearing tors, Volume 1; Article 132, Data Acquisition Systems
temperature and so on, of pumps and motors, low suction (DAS) in General, Volume 3; Article 158, Microcon-
pressures, power failures, relief valve open, leak detection, troller Systems, Volume 3; Article 165, Introduction to
84 Foundations of Measuring
Networked Instrumentation, Volume 3; Part 5, Measur- McGhee, J., Korczynski, M.J., Kulesza, W. and Henderson, I.A.
ing System Design, Volume 2. (1996) Scientific Metrology, ISBN 83 904299 9 3, A.C.G.M.
Lodart, Lodz.
McGhee, J., Henderson, I.A., Kulesza, W. and Korczynski, J.
REFERENCES (2001) Measurement Data Handling: Theoretical Technique,
ISBN 83 7283 007 X, A.C.G.M. Lodart S.A., Lodz.
Lang, T.T. (1987) Electronics of Measuring Systems, John Wiley Sydenham, P.H. (1983) Chapter 16.3, in Handbook of measure-
& Sons, Chichester. ment science: Vol 2 (ed. P.H. Sydenham), John Wiley & Sons,
Chichester.
McGhee, J., Henderson, I.A. and Sankowski, D. (1986) Functions
and Structures in Measurement Systems: A Systems Engineer-
ing Context for Instrumentation. Measurement, 4(3), 11–119.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
16: Reduction of Influence Factors
Paul P.L. Regtien
University of Twente, Enschede, The Netherlands
additional input signals: xd and xs represent deterministic
1 Influence Effects 87 and stochastic signals respectively. They model all kinds
2 Isolation of Influence Effects 88 of interference from the environment as well as the equiv-
alent input error signals due to system offset and noise.
3 Compensation for Influence Effects 89
Additive error signals are always expressed as (equivalent)
4 Feedback Methods for Suppressing Influence input signals, which facilitates comparison with the actual
Effects 90 input signal.
5 Correction Methods to Reduce Influence The error inputs εd and εs symbolize deterministic and
Effects 91 stochastic multiplicative errors; they mainly originate from
References 92 environmental influences: temperature changes, electric,
magnetic, or electromagnetic fields, mechanical vibrations,
shocks, and nuclear radiation. For this simplified model, the
output signal of a sensor can be written according to:
1 INFLUENCE EFFECTS
yo = S(1 + εd + εs )(xm + xd + xs ) (1)
Any measurement system has imperfections and any act of
measurement is liable to errors. Measurement errors either where S is the nominal sensitivity (transfer) of the measure-
originate from system deficiencies (for instance system ment system. This model is useful for evaluating system
noise, quantization, and drift), or are due to environmental errors, influence factors, and the various error-reduction
influences such as thermal, electromagnetic, and mechanical methods. When all influence factors are eliminated, the out-
interference. Manufacturers of measuring instruments try to put of the system satisfies:
minimize such intrinsic errors by a proper design of the
instrument structure; the remaining imperfections should yo = S × xm (2)
be given in the data sheets of the system. The user of a
measurement system should minimize additional errors that This situation is hypothetical and will never occur in
could arise from improper use and faulty interfacing. practice. Therefore, it is important to know what is the
Besides influence effects from outside the system, instru- deviation from this ideal situation and to quantify it. It
ments may suffer from many other imperfections. They is equally important to know the type of error: determin-
influence the magnitude of the instrument transfer (mul- istic or stochastic; additive or multiplicative, in order to
tiplicative error) or give rise to unwanted output signals effectively suppress them when necessary. Error suppres-
(additive errors). Both error types may occur as a deter- sion can be obtained by a sophisticated design, or by just
ministic or a stochastic error. Figure 1 shows a simplified some additional signal processing. Some of these methods
model of a measurement system, with an indication of these will reduce mainly additive sensors, others minimize multi-
error sources. In this figure, xm is the measurand and yo plicative errors. We will discuss some basic error reduction
the output signal. Additive error signals are modeled as methods in the next sections.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
88 Foundations of Measuring
ed es and the measurement system, notably the electrical parts at
xd xs
its input (Figure 2a).
+ + Capacitive coupling (in the figure, the distributed capac-
+ +
xm S(edes) yo itances are modeled with only lumped capacitors) results in
an equivalent error signal:
Figure 1. Simplified sensor model including error signals. Cs
Vn,o = V (3)
Cg n,i
2 ISOLATION OF INFLUENCE EFFECTS
with the assumption Cs < Cg . Its effect can be reduced by
Error signals can be reduced or eliminated by isolating a grounded shield around the sensor structure and the input
the measurement instrument from the interference source. leads (Figure 2b). The equivalent error signal is reduced to
The error signals may be kept outside the measurement
Crest
system by several isolation techniques, depending on the Vn,o = V (4)
type of signal. The isolation techniques are also referred to Csh n,i
as shielding, guarding, or (pre)filtering, according to the
This approaches zero when the rest capacitance can be made
implementation. We briefly review some causes and the
very small (so an effective enclosure around the sensitive
associated isolation methods.
parts of the system is provided).
2.1 Capacitively induced errors
2.2 Magnetically induced errors
Capacitively induced errors arise by electric injection of
spurious signals, through a capacitive connection between Magnetically induced signals arise by a time varying mag-
the error voltage source (for instance, wires of the mains) netic flux through the loop made up by the input circuit.
Cs
Sensor system Interface system Vn,o
+
Vn,i
Cg
−
(a)
Crest Cg
Csh
Sensor system Interface system Vn,o
+
Vn,i
−
(b)
Figure 2. Capacitively induced errors: before and after shielding.
Reduction of Influence Factors 89
The induced error signal amounts to 3 COMPENSATION FOR INFLUENCE
EFFECTS
d d dB
Vn,o = = B × dA = A (5)
dt dt A dt Compensation is a simple and effective method of minimiz-
and is proportional to the surface area of the loop, the ing additive errors due to interference signals. The basic
strength of the magnetic field, and the frequency of the field idea is as illustrated in Figure 3, where the technique is
(for instance 50 or 60 Hz from the mains). The interference applied on the sensor level.
In Figure 3(a), the output of the sensor is y1 , which
can be reduced by a shield made of a material with a high
contains unwanted signal components, for instance, due to
magnetic permeability (to reduce the induction B), and by
interference xi or offset. From this output, a compensation
minimizing the area of the loop A (for instance, by placing
voltage yc is subtracted. The condition for full compen-
the input leads close to each other or by twisting).
sation is yc = S × xi , making yo = S × xm , independent
of xi . For correct compensation, the interference signal as
well as the sensor transfer should be known. One way of
2.3 Temperature induced errors accomplishing compensation is by an adjustable compen-
sation signal: at zero input, yc is (manually) adjusted to a
Changes in environmental temperature may introduce both value at which the output is zero. A more elegant way of
additive errors (for instance, offset drift) and multiplicative compensation is by applying a second sensor, as illustrated
errors (for instance, the temperature coefficient of the trans- in Figure 3(b). The measurement signal, xm , is supplied
fer). This interference can be reduced by thermal shielding simultaneously to both sensors, which have equal but oppo-
(an encapsulation with high thermal resistance or a temper- site sensitivity to the measurement input (for example, two
ature controlled housing). strain gauges, one loaded on compressive stress, the other
on tensile stress). A minus sign represents the opposite sen-
sitivity of the sensors. The two output signals are subtracted
2.4 Optically induced errors by a proper electronic circuit. Because of the antisymmet-
ric structure with respect to the measurement signal only,
many interference signals appear as common output sig-
Unwanted optical input can be stopped by optical fil-
nals, and are thus eliminated by taking the difference of
ters; many filter types are available on the market. Mea-
the two outputs.
surand and interferences should have a substantial dif-
The effectiveness of the method depends on the degree
ference in wavelength to make the filtering sufficiently
of symmetry of the double-sensor or differential sensor
effective. Optical interference originating from environ-
structure. From Figure 3 it follows:
mental light (daylight, lamplight) has a low frequency
(DC, 50/60 Hz and harmonics). If the frequency of the y1 = Sm1 xm + Si1 xi (6)
optical measurand is in the same band, filtering is not a
solution. In that case, modulation may help reduce inter- y2 = −Sm2 xm + Si2 xi (7)
ference. If the measurement system contains an auxiliary
light source (as with many optical systems), that signal where Smk and Sik are the sensitivities for the measurement
source can be modulated (using an LED or laser diode). signal and the interference signal. So, the output of the total
If the source itself cannot be modulated, optical mod-
ulation can be performed by, for instance, a chopping + y1 +
xi S yo
wheel, as is applied in many pyroelectric measurement + − y
systems. xm c
(a)
+ y1
2.5 Mechanically induced errors xi S1
+ +
xm yo
− −
Mechanical disturbances (shocks, vibrations) are reduced S2
+ y2
by a proper mechanical construction with elastic mounting,
(b)
performing suitable damping of the vibrations. The method
requires knowledge about the most important resonance Figure 3. General layout of compensation; (a) compensation sig-
frequencies of the structure. nal and (b) balanced configuration.
90 Foundations of Measuring
sensor system equals sensitivities to the measurand, the bridge output voltage
approximately satisfies, the equation:
yo = (Sm1 + Sm2 )xm + (Si1 − Si2 )xi (8)
Vo 1 1
= Sm xm + (Si xi ) (13)
The sensitivities Sm1 and Sm2 are about equal, so the output Vi 2 4
signal becomes
If both sensor parts have equal sensitivity to interference
Si (by a symmetric sensor design), and both sensor parts
yo = 2Sm xm + Si xi = 2Sm xm + x (9) experience the interference equally, the error term is zero,
2Sm i
and the interference is completely eliminated. The equation
Analogous to the definition of the rejection ratio for differ- is useful for making a quick assessment of the error due to
ential amplifiers, we can define a quality measure for the asymmetry, relative to the measurement signal.
imbalance of the differential sensor:
2Sm
H = (10) 4 FEEDBACK METHODS FOR
Si
SUPPRESSING INFLUENCE EFFECTS
a parameter characterizing the system’s ability to distin-
guish between measurand and interfering signals. Feedback is an error reduction method originating from the
The method is illustrated with a two-active element early amplifiers with vacuum tubes. Their instable operation
Wheatstone bridge in Figure 4. was a real problem until the systematic application of
In this half-bridge mode, R3 and R4 are fixed resistances; feedback, which reduces in particular multiplicative errors.
R1 and R2 are resistive sensors: their resistance values Figure 5 shows the general idea.
change with a particular physical quantity as well as an The sensor has a nominal transfer S, but due to mul-
interfering signal, according to tiplicative interference, it has changed to S(1 + εi ). The
feedback is accomplished by an actuator with an inverse
R1 = R(1 + Sm1 xm1 + Si1 xi1 ) (11) transduction effect, and a transfer k. From classical control
R2 = R(1 + Sm2 xm2 + Si2 xi2 ) (12) theory, the error reduction factor can easily be found. The
transfer of the total system, Sf , is given by
Here, Smk is the sensitivity of sensor k (k = 1, 2) to the
measurand (for instance, deformation), and Sik is the sensi- S
Sf = (14)
tivity to the interference signal (for instance, temperature). 1+S×k
Note that in this system, the sensor resistance is just R
A relative change dS in the sensor transfer S causes a
at a particular reference value of the measurand, and at
relative change dSf in Sf according to
zero interference. Assuming both sensor parts experience
the same measurement signal, and have equal but opposite dSf 1 dS
= × (15)
Sf 1 + kS S
+
So, the relative error in the feed-forward part is reduced by
a factor equal to the loop gain 1 + kS of the system. The
R1 R3 penalty for this improvement is a reduction of the overall
sensitivity with the same factor.
Feedback also reduces additive interference signals to a
+ Vo −
Vi degree that depends on the point of injection in the system.
ei xi1 xi2
R2 R4 xm yo xm + + + + yo
S (1 + ei)
+ S
− + −
k
− k
Figure 4. Measurement bridge with two active elements. Figure 5. Interference signals in a feedback system.
Reduction of Influence Factors 91
Two cases are discussed (Figure 5). The output due to two He
Hi Hs
interfering signals xi1 and xi2 equals: Fi Fs
a m Spring Sensor Interface Gain
+ − ∆x ∆C VC Vo
S 1
yo = × xi1 + × xi2 (16)
1 + kS 1 + kS Fa
Actuator
Ha
Obviously, signals entering at the input of the system are
reduced by feedback as much as the measurement signal Figure 7. Model of the differential capacitive accelerometer
(so the SNR is not better); interfering signals injected at with feedback.
the output of the sensing system are reduced by a factor S
more than the measurement signal. and is given in the next equation:
Feedback reduces errors in the forward signal path:
the transfer is mainly determined by the feedback path. Hs He
Vo = Ha (17)
Prerequisites for an effective error reduction are 1 + Ha Hs He i
• high forward path transfer, In this equation, Hi is the transfer from applied acceleration
• stable feedback path transfer. a to inertial force Fi . Hs is the transfer of the mechanical
spring: from force to displacement. The transfer of the
The application of this method to sensors requires a feed- electrical system: capacitive sensor, electronic interface,
back element with a transfer that is the inverse of the sensor and amplifier together) is He . Finally, Ha is the transfer
transfer. The imperfections of the sensor are reduced; how- of the actuator. For Ha Hs He 1 the transfer function of
ever, the demands on the actuator are high. The method the total sensor system becomes:
is illustrated with an example of a capacitive accelerometer
system in which two error reduction methods are combined: Hi
Vo = a (18)
compensation by a balanced sensor construction and feed- Ha
back by an inverse transducer (Figure 6).
A displacement of the seismic mass m results in a Indeed, this is independent of the spring stiffness, the sensor
capacitance difference C; this value is converted to a transfer, the interface, and the gain of the amplifier, accord-
voltage that is compared with a reference value (here this ing to the feedback principle. In equilibrium, the mass is
value is zero). The amplified voltage difference is supplied at its center position. Hence, no particular demands have
to an electromagnetic actuator that drives the mass back to be made on the spring, the sensor, and the interface cir-
toward its neutral position. When properly designed, the cuit; the only requirement is a low zero error. Actually,
system reaches a state of equilibrium where the applied the requirements with respect to the sensor quality have
inertial force is compensated by the electromagnetic force been transferred to those of the actuator. The system trans-
from the actuator. The current required to keep equilibrium fer only depends on the seismic mass m and the actuator
is a measure of the applied force or acceleration. transfer Ha .
A more detailed model of this system, for instance, for There are various sensors on the market that are based
stability analysis, is depicted in Figure 7. All transduction on the feedback principle, which is discussed in more detail
steps are visualized in separate articles. in Article 113, Force-feedback Sensors, Volume 2.
Obviously, the feedback is performed in the mechanical
domain, by counteracting the inertial force Fi with the elec-
tromagnetic force Fa of the actuator. The transfer function 5 CORRECTION METHODS TO REDUCE
of the whole system can easily be derived from this model, INFLUENCE EFFECTS
Seismic mass An erroneous sensor signal can be corrected if knowledge
V(∆C ) about the causes of the errors or the value of the errors is
a Sensor −
interface available. Two different strategies can be distinguished:
Vo
Differential +
capacitor • static correction
∆x
• dynamic correction.
Actuator
In the first class of strategies, correction is performed
Figure 6. Illustration of a feedback system to reduce sensor while leaving the sensor as it is. Figure 8 shows the gen-
errors: capacitive accelerometer. eral configuration for two approaches. In the model-based
92 Foundations of Measuring
+ y1 From these equations, the measurand can be calculated:
xm S Processing yo
+ yo1 − yo2
xi xm = x (22)
yo3 − yo2 ref
Si Error Model
yi data data
Offset and scale errors are completely eliminated if the
Figure 8. Model-based correction and measurement-based errors do not change during the sequence of the three
correction. measurements. The method requires a reference of the
measurand type, as well as the possibility of completely
approach, the sensor signal is corrected on the basis of prior isolating the input from the measurand to find the offset
knowledge about the origin of the error, for instance, non- error contribution. For most electrical quantities, like volt-
linearity or a calibration curve, stored in a look-up table. If age, capacitance, and resistance, this is quite an easy task.
the errors are unknown (interference), the error signal could This is not the case for many other physical quantities:
be measured separately by additional sensors. The output repetitively short circuiting the input of a magnetic field
of these sensors is used to correct the original sensor sig- sensor or a force sensor, for instance, without removing the
nal. The method is straightforward, but requires additional measurand itself requires complicated and therefore imprac-
sensors, at least one for each type of interference. tical shielding techniques.
Dynamic correction involves a particular sensor design. Modulation, as discussed before, is a very effective way
Basically, the method involves a series of measurements of of eliminating offset and the effects of influence factors.
the same quantity in such a way as to eliminate the errors by The basic idea is to shift the (low frequency) informa-
postprocessing. This can be done in various ways, according tion in the input signal to another, higher frequency band.
to the type of quantity and error: After high-pass filtering, the offset and drift are effec-
tively eliminated. However, modulation of the input signal
• multiple input signal measurements, eliminating scale, is not always possible. An alternative way of reducing
and offset errors; offset is flipping the directivity of the sensor, without
• cyclic interchanging of components (dynamic mat- changing the sign and value of the offset (Van Putten,
ching); Van Putten and Van Putten, 1994). Clearly, the method
• cyclic changing in the sensitivity direction of the sen- can only be applied to sensors that are sensitive to the
sor (flipping). direction of the input quantity (which should be a vec-
tor). Another technique, applicable to modulating sensors,
The first method is illustrated in Figure 9. Alternately, the is spinning. An example where this technique is applied is
input of the system is connected to the measurand (yielding the spinning-current Hall sensor (Bellekom, 1998). Here,
an output yo1 ), to ‘ground’ (giving output yo2 ) and to a the auxiliary current is electronically rotated stepwise with
reference (resulting in the output yo3 ). Multiplicative errors respect to the geometry, yielding a number of outputs;
due to the interfering signal xs are represented by the each output contains an offset signal, originating from
relative error εs . layout asymmetries of the chip. However, going around
The three output signals are: the average of the offset is zero, and by averaging the
outputs over a full cycle, offset is almost completely
yo1 = S(1 + εs )(xm + xoff ) (19) eliminated.
yo2 = S(1 + εs )(xoff ) (20)
yo3 = S(1 + εs )(xref + xoff ) (21) REFERENCES
Bellekom, A.A. (1998) Origin of Offset in Conventional and
xoff es
xm Spinning-Current Hall Plates, PhD thesis, Delft University of
+ Technology, Delft.
+
xref S(1 + es) yo
Van Putten, M.J.A.M., Van Putten, M.H.P.M. and Van Putten,
A.F.P. (1994) Full Additive Drift Elimination in Vector Sensors
Using the Alternating Direction method (ADM). Sensors and
Actuators A, 44, 13–17.
Figure 9. Three-point measurement to eliminate scale and off-
set errors.
17: EMC and EMI
Kim R. Fowler
Technical Consultant, Cool Stream, Windsor Mill, MD, USA
• high-voltage equipment (e.g. spark plugs, igniters)
1 Definitions and Examples of EMC and EMI 93 • discharges and sparks (e.g. lightning, static electricity)
2 Basic Principles of Energy Coupling 93 • high-current equipment (e.g. arc welders).
3 Mechanisms of Noise Energy Coupling 94
Examples of susceptibility within receiver circuits
4 Electrostatic Discharge (ESD) 95 include
5 General Guidelines for EMC and EMI 96
6 Standards for EMC 96 • crosstalk on inputs in digital logic causing unintentional
References 97 changes in logic states
• crackle in the reception of radio
• static discharge that destroys components, such as key-
board circuits.
1 DEFINITIONS AND EXAMPLES OF
EMC AND EMI
2 BASIC PRINCIPLES OF ENERGY
Electromagnetic interference (EMI) occurs when one elec-
COUPLING
trical circuit interferes with another. EMI may exist between
components within an electronic circuit, between electronic,
and electrical subsystems, or between separate instruments. EMI has three necessary components: a noise source,
Electromagnetic compatibility (EMC) is the art and science a susceptible circuit, and an energy coupling mecha-
of reducing EMI to acceptable levels. nism. Typically, neither the noise source nor the sus-
General and specific accounts that review EMI and EMC, ceptible circuit may be eliminated; this leaves you to
provide practical solutions to EMI and EMC, and provide reduce or remove the energy coupling mechanism between
key references are Fowler (1996), Johnson and Graham the two.
(1993), and Ott (1988).
Typically, the electromagnetic interference is called
noise, which is any undesired electrical activity coupled 2.1 Reciprocity
from one circuit into another. EMI has three necessary com-
ponents: a noise source, a susceptible circuit, and an energy
coupling mechanism. EMC is concerned with reducing or Anything that reduces emissions usually makes the cir-
eliminating the interference or noise to acceptable levels. cuit less susceptible to disruption. This means that the
Examples of noise sources include same techniques to reduce noise or remove a noise source
will probably work to reduce susceptibility within receiver
• power lines or mains circuits. Reciprocity is a useful guideline but not an abso-
• motors lute rule.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
94 Foundations of Measuring
2.2 Impedance versus resistance Noise A Susceptible
source circuit
One component of understanding energy coupling and inter-
ference in EMI and EMC is to understand how electrical + +
current flows. Current always follows the path of lowest
− −
impedance Z (see (1)), not necessarily the path of lowest
resistance R. Remember that impedance includes reactive
terms for inductance L and capacitance C. Consequently,
electrical charge follows the path that represents the min- B
imum of resistance and inductive reactance and maximum
Figure 1. Conductive coupling. If either connection A or B is
capacitive reactance; generally, this is true for signal fre-
removed, the conductive noise is eliminated. (Reproduced from
quencies above the audio range (>3 kHz). Electronic Instrument Design: Architecting for the Life Cycle,
Fowler, K.R. Copyright (1996) Oxford University Press.)
2
1
Z= R2 + L − (1)
C
3.2 Inductive coupling
Inductive coupling requires a receptive loop circuit and a
2.3 Pseudo-impedance changing magnetic flux through the loop. A current bearing
a loop circuit may be the source of the changing magnetic
A very useful diagnostic tool for signal frequencies above flux. The coupled or induced noise voltage is proportional
the audio range (>3 kHz) is Pseudo-impedance as explained to the time rate of change in current in the source and
by Van Doren (1991). Pseudo-impedance is the diagnostic the loop area of the circuit(s). The Pseudo-impedance or
ratio of the rates of change in voltage to current. diagnostic ratio in (3) gives a value much less than free-
space impedance.
dv
dt dv
Diagnostic ratio = (2)
di dt
Inductive coupling = 377 (3)
dt di
dt
3 MECHANISMS OF NOISE ENERGY Straight wires create small loops that can magnetically cou-
ple. Twisted wire eliminates the effective loop area of the
COUPLING cables and magnetic coupling, thus eliminating inductive
coupling (see Figure 2).
Noise may couple between circuits through four primary Inductive coupling meets the principle of current flow
mechanisms: conductive, inductive, capacitive, and elec- through the path of least impedance in designing circuit
tromagnetic. boards. The return current mirrors the topology of the signal
current; that is, it flows in the return plane immediately
adjacent to the signal line. This means that the return plane
3.1 Conductive coupling of a circuit board should not have any slots in it that cut
across the signal path. A slot, Figure 3, increases the loop
Conductive coupling, see Figure 1, requires a connection area, hence, it increases the inductive reactance and the
between the source and the receiver that completes a generation of a magnetic field.
continuous circuit. Often these connections are inadvertent;
when they are, they are called sneak circuits.
A ground loop is a particular form of conductive cou- 3.3 Capacitive coupling
pling, which has two primary characteristics. It has multiple
ground connections providing multiple return paths that Capacitive coupling requires both proximity between cir-
cause significant current flow in the grounding structure. It cuits and a changing voltage. The coupled or induced noise
unbalances the desired circuit. (See Figure 5 in Article 177, voltage is proportional to the time rate of change in the volt-
Grounding and Shielding, Volume 3.) age in the source and inversely related to separation distance
EMC and EMI 95
+
ZS
−
Noise + ZL
source −
(a)
(a)
+
Susceptible
− ZS circuit
Noise + ZL
source −
(b)
(b)
Figure 2. Effect of twisted wire. (a) Straight wires create small
loops that can couple magnetically and (b) twisted wire eliminates Figure 4. Effect of shielding. (a) Without the shield, stray cur-
the effective loop area of cables and magnetic coupling. (Repro- rents can disrupt susceptible circuits and (b) a properly connected
duced from Electronic Instrument Design: Architecting for the shield can divert capacitively coupled current from susceptible
Life Cycle, Fowler, K.R. Copyright (1996) Oxford University circuits. (Reproduced from Electronic Instrument Design: Archi-
Press.) tecting for the Life Cycle, Fowler, K.R. Copyright (1996)
Oxford University Press.)
Signal
current 3.4 Electromagnetic coupling
+ Radiated electromagnetic energy requires an antenna in
− both the noise and susceptible circuits. The antenna must
be an appreciable portion of the wavelength. Electromag-
ZL netic coupling usually occurs for frequencies greater than
20 MHz and for signal conductors that are longer than a
small fraction of the fundamental wavelength (>λ/20). The
Pseudo-impedance for electromagnetic coupling in (5) is
around the value for free-space impedance or 377 ; prac-
Return current tically, any value between 100 and 500 is a reasonable
in return plane Return value for electromagnetic coupling.
plane
Figure 3. A slot in the return plane of a circuit board increases dv
the current loop area and the self-inductance. (Reproduced from dt
Electronic Instrument Design: Architecting for the Life Cycle, Electromagnetic coupling = ∼ = 377 (5)
di
Fowler, K.R. Copyright (1996) Oxford University Press.)
dt
between the circuits. The Pseudo-impedance or diagnos- EMI through electromagnetic coupling always begins as
tic ratio in (4) gives a value much larger than free-space conductive (i.e. current in wires), becomes radiative, and
impedance. ends as conductive, that is, fields interact with circuitry.
Cables can be primary sources and receivers for frequency
dv of operation above 200 MHz.
dt
Capacitive coupling = 377 (4)
di 4 ELECTROSTATIC DISCHARGE (ESD)
dt
Electrostatic discharge (ESD) is a transfer of electrical
Generally, a properly connected shield will reduce capac- charge at very high voltage, ranging from hundreds to thou-
itive coupling (see Figure 4). sands of volts (Table 1), and very low current. Generally,
96 Foundations of Measuring
Table 1. Some typical values for ESD. ESD. Figure 13 in Article 177, Grounding and Shielding,
Static generation Static voltage (V) Volume 3 illustrates one way to protective sensitive circuits
by shunting ESD to chassis ground.
10% relative 65% relative
humidity humidity
Walking across carpet 35 000 1500 5 GENERAL GUIDELINES FOR EMC
Walking over vinyl floor 12 000 250 AND EMI
Common plastic bag 20 000 1200
picked up from bench
Work chair padded with 18 000 1500
Characterize your instrument or circuit for EMI and EMC
polyurethane foam in the following ways:
Source: Reproduced from Electronic Instrument Design: Architecting for • Establish: grounding options, source and load impe-
the Life Cycle, Fowler, K.R. Copyright (1996) Oxford University Press.
dances, and frequency bandwidth needed.
• Determine the predominant coupling mechanisms.
15 • Diagram the circuit paths and reduce loops, such as
ground loops and inductive loops.
Discharge current (A)
Use good design techniques to reduce EMI and improve
EMC:
8
• reduce frequency bandwidth
• balance currents
4 • route signals for self-shielding: use return planes, short
traces, and decoupling capacitors
• add metal enclosure shielding only when necessary.
30 60 Remember reciprocity, anything that reduces emissions
<1 ns Time (ns) usually makes the circuit less susceptible to disruption,
as well. If you decrease the inductive loop of a circuit
Figure 5. ESD discharge waveform at 4 KV; this is one of sev-
eral different waveforms for testing circuits against ESD. (Repro- to reduce emission, then you will also reduce the mutual
duced from Electronic Instrument Design: Architecting for the coupling that could receive noise.
Life Cycle, Fowler, K.R. Copyright (1996) Oxford University Susceptibility often traces to incorrect return paths or
Press.) long signal lines that are not properly shielded. Remember
that current follows the path of lowest impedance. Under-
ESD has very fast rise times (<1 ns) and short duration stand mechanisms of noise energy transmission:
(∼90 ns), as shown in Figure 5. ESD can easily dam-
• conductive coupling
age sensitive electronic circuits. Any instrument containing
• inductive coupling
integrated circuits is susceptible to ESD if not protected.
• capacitive coupling
ESD transfers charge in three stages: pickup, storage,
• electromagnetic coupling.
and discharge. Usually mechanical rubbing between dry,
insulated materials picks up and transfers the charge from When you do encounter EMI, begin the search for the
a source to storage. Sources include plastic work surfaces, coupling mechanisms using the diagnostic clues in Table 2.
carpeted floors, wool, cotton or synthetic garments, vinyl
or fiberglass chairs, common plastic bags, trays and bubble
wrap, spray cleaners, and electrostatic copiers. Humans 6 STANDARDS FOR EMC
often can be the storage means for the electrical charge
until its discharge occurs. Many different regulations and standards exist around the
Control methods for ESD eliminate activities and mate- world for designing instruments to meet EMC requirements.
rials that create high static charge and provide transient These standards provide levels that tests must pass for
protection to sensitive circuits. Eliminating high static instruments to avoid interference. Commercial regulations
charge includes appropriate grounding, using protective cover the frequency band from 9 kHz to 1 GHz.
handling and materials, and maintaining humidity. A check- Typically, these standards specify conducted and radiated
list in Article 177, Grounding and Shielding, Volume 3 environments for interference and susceptibility. Clearly,
lists some of the activities that will reduce problems with conductive coupling falls into the conducted environment.
EMC and EMI 97
Table 2. Diagnostic clues for noise-coupling mechanisms.
Coupling mechanism Diagnostic clues
Conductive 1. Complete conductive path (such as multiple ground connections)
2. Frequency < 10 MHz
Inductive 1. (dv/dt)/(di/dt) 377
2. Ringing on pulse edges
3. Low source impedance and high load impedance
4. Laying conductive foil over cable decreases noise level (acts as an eddy current shield)
5. Large circuit loops
Capacitive 1. dv/dt)/(di/dt) 377
2. Rounding of pulse edges
3. High source impedance and high load impedance
4. Laying conductive foil over cable increases noise level (acts as a coupling capacitor)
Electromagnetic 1. Distance between noise source and susceptible circuit > λ
2. Frequency > 20 MHz
Source: Reproduced from Electronic Instrument Design: Architecting for the Life Cycle, Fowler, K.R. Copyright (1996) Oxford
University Press.
Table 3. Diagnostic clues for noise-coupling mechanisms. coupling, falls under the domain of the radiated environ-
Type Standard Country ment. Table 3 lists examples of some of these standards.
Commercial CNELEC Europe
IEC 871, 801, 555 Europe
FCC United States REFERENCES
VCCI Japan
VDE Germany Fowler, K.R. (1996) Electronic Instrument Design: Architect-
Military GAM-EG-13 France ing For The Life Cycle, Oxford University Press, New
VG NORM Germany York.
DEF STAN 59-41 United Kingdom Johnson, H.W. and Graham, M. (1993) High-Speed Digital De-
MIL STD-461E United States sign: A Handbook of Black Magic, PTR Prentice Hall, Engle-
Source: Reproduced from Electronic Instrument Design: Architecting for
wood Cliffs, NJ.
the Life Cycle, Fowler, K.R. Copyright (1996) Oxford University Ott, H.W. (1988) Noise Reduction Techniques in Electronic Sys-
Press. tems, 2nd edn, John Wiley & Sons, New York.
Van Doren, T. (1991) Grounding and Shielding Electronic
Generally, the three remaining types of coupling, induc- Systems, Notes from video seminar on the NTU satellite
tive coupling, capacitive coupling, and electromagnetic network.
18: Nature and Scope of Closed-loop Systems
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
was controlled manually using a throttle valve on the
1 History of Closed-loop Control Systems 101 steam inlet.
2 Illustration of an Exemplar Closed-loop In 1788, Boulton, Watt’s coprincipal, in a company
Control System 102 manufacturing the new steam engines, had visited the
3 Setting up the Mathematical Model for an Albion Mill and described to Watt a form of centrifugal
Open-loop System 103 governor that was being used to regulate the grinding
4 Mathematical Model of the Closed-loop speed of the millstones driven by wind. Watt adapted this
System 103 mechanical governor to measure and automatically control
the speed of his steam engines (Atkinson, 1972).
5 Effects of Friction in Real Feedback Systems 104
Since those days, closed-loop control and measurement
6 Nature of Closed-loop Systems used in
have been inexorably linked; indeed to control a physical
Measurement Systems 105
quantity accurately and rapidly in the presence of changing
Related Articles 106 demand, changing internal parameters, and changing load,
References 106 measurement of the control quantity coupled with feedback
control is essential. Such measurement can, of course, only
be accomplished by means of instruments.
Another development down this path is that feedback
1 HISTORY OF CLOSED-LOOP itself is often used extensively within measuring instru-
CONTROL SYSTEMS ments to improve the accuracy of the measurement.
In the nineteenth century, vast numbers of regulated
This account of linear closed-loop control principles and steam engines were in operation in factories throughout
theory is now a mature topic, wherein the control theory Britain. These closed-loop systems were by no means per-
has greatly advanced to allow for sophistication of multi- fect in that the precision of speed control was often rather
variable, stochastic, nonlinear, and digital applications. poorly achieved with large and unpredictable fluctuations
The design of measuring systems does at times demand being the norm.
knowledge of such advanced ideas but the starting point, Engineers of that time had little real insight into the
which often suffices, is the classical theory and its foun- dynamics of systems. They attempted to improve the per-
dation. This set of articles is revised and edited from formance by making smaller, lighter, and better lubricated
material first published in brief by Atkinson in Syden- governors. Today, we can easily see from our understand-
ham (1982). A detailed explanation is available in Atkinson ing of the theory of closed-loop systems that this was not
(1972). the way to go.
Modern closed-loop control systems can be traced back Much to their consternation these expected improve-
to James Watt, who invented the first rotating steam engine ments normally led to an unforeseen associated difficulty
(patented in 1781). Initially, the speed of these engines in that the steam-engine speed then tended to oscillate or
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
102 Foundations of Measuring
hunt continuously about the demanded value. This partic- in control. The following principles apply to all energy
ular form of instability, as it is now called, has plagued domains but the electromechanical one is more intuitively
designers of feedback control systems of all kinds ever obvious than in others.
since. It is the main limitation on performance. Handled Suppose there is a need to control the angular posi-
from an understanding of the physics involved, it can usu- tion of a heavy rotating mass, such as a gun turret or a
ally be limited to allow good dynamic performance – see machine tool turntable. The power resource of mechani-
Article 19, Dynamic Behavior of Closed-loop Systems, cal or electrical power assistance, rather than total reliance
Volume 1 and Article 62, Second-order System Dynam- on muscle power, allows the obvious advantage of rapid
ics, Volume 1. control.
The practical importance of this difficulty stimulated no To simplify the problem, suppose an ideal frictionless
less a person than Maxwell (1868) to investigate the prob- electric motor is available. To achieve maximum accelera-
lem in great depth. He brought mathematical insight to bear tion of the rotating load, the motor is coupled to the load
by relating the existence of instability to the presence of through an ideal frictionless step-down gearbox. Friction
positive real parts in the complex roots of the characteristic is always present in real systems but for the moment it is
equation of the system. His work did not, however, greatly ignored – see Section 5.
influence matters at that time. It is assumed that the motor produces a torque at the
The First World War (1914–1918) caused military engi- load, which is directly proportional to its supply voltage. In
neers to realize that to win wars it is necessary to posi-
order to control the supply voltage to the motor, an ideal
tion heavy masses (e.g. ships and guns) precisely and
power amplifier is connected, which receives at its input a
quickly. Classic work performed by Minorsky (1922) in
control voltage, υi , that is directly proportional to an angular
the United States was on automatic ship steering and it was
positional signal, θi .
Hazen (1934) who defined a servomechanism for the first
This signal is applied manually through a light handwheel
time.
connected to a position-to-voltage transducer (such as a
The concepts of automatic control, as they developed,
rotary potentiometer).
are covered in Bennett (1979). Sydenham (1979) gives an
overview of the historical development of feedback control The notional arrangement is illustrated in Figure 1. This
in its use in early instruments and the chart recorders, system will produce rapid acceleration of the rotating mass
wherein it was first recognized intuitively that velocity in response to small and effortless motion applied manually
feedback could aid loop response times. to the handwheel. When the handwheel is at a nominal
zero position, the mass will cease accelerating; a change
in the handwheel position in one direction will produce
acceleration in one direction, and a change in position
2 ILLUSTRATION OF AN EXEMPLAR in the opposite direction will produce acceleration in the
CLOSED-LOOP CONTROL SYSTEM other direction.
This, the open-loop system will, however, be found to be
The operation of a mechanical position control provides quite unsatisfactory as a controller. To show why, the next
a good example for illustrating the need for feedback section discusses its mathematical model.
Power
Handwheel Gearbox
qi
Electronic uo Electric
Positional ui motor
power
transducer
amplifier
qo
Rotatable mass
Figure 1. Position control system without feedback.
Nature and Scope of Closed-loop Systems 103
3 SETTING UP THE MATHEMATICAL Worse still, in many circumstances (e.g. control of the
MODEL FOR AN OPEN-LOOP SYSTEM position of a gun turret) there will be load disturbances
present, for example, wind gusts which will cause the mass
A simple mathematical analysis for the open-loop system to deviate from the desired position in a random manner.
is as follows. The result is that to get workable control of the position
Let the effective moment of inertia of the moving parts, using open-loop control, only small torques can be applied
referred to the position of the rotating mass, be J . Also, let and the result is a sluggish motion that is prone to system
disturbances and sloppy following action.
υi = kt θi
υa = ka υi (1) 4 MATHEMATICAL MODEL OF THE
and the effective torque acting on the mass T be given by CLOSED-LOOP SYSTEM
T = km υa (2) This position control problem is provided with a much
better solution similar to the one in which Watt solved
where kt , ka , and km are constants. his speed control problem, that is, by adding feedback of
Thus, a measured value of the controlled variable to the power
amplifier so that the system power can be applied with as
T = km ka kt θi = Kθi (3)
much power as is needed and in the right direction at the
appropriate time.
where K is a composite system constant.
The mathematical model for such a feedback or closed-
Expressed in calculus notation, according to Newton’s
loop system is now developed using Figure 3 as the feed-
second law of motion it is given by
back system arrangement.
d2 θo The controlled output variable θo can be measured by
J = T = Kθi (4) means of a transducer identical to that monitoring the input
dt 2
command θi , so that a signal υo (given by υo = kt υo )
The response to a small change in θi from zero to a fixed is available for comparison with υi . The notion is now
positive angle is shown in Figure 2. This can be seen to be that if θi is made equal to the required position of the
a first-order system – see Article 61, First-order System rotating mass, the amplifier can be fed by a difference signal
Dynamics, Volume 1. (υi − υo ), which is given by
In order to change the position of the mass from one fixed
position to another in the shortest time, it is necessary to υi − υo = kt θi − kt θo = kt ε (5)
first accelerate the mass with all the torque available but
then quickly retard it with a reversed torque such that at where ε is defined as the positional error between the
the instant the required position has been reached, both required position θi (termed the command or input) and
the velocity and acceleration are simultaneously reduced the actual output position θo .
to zero. With this simple approach, the manual control This error signal is now amplified as before and applied
problem is extraordinarily difficult to apply to advantage. to the motor. Thus, a driving torque will always be present
as long as θo is different from θi . When they are of the
same value there will be no driving torque and the mass
will hopefully stop moving at the point where it needs to
qo be. An analysis of this system is very revealing.
The effective torque, T , acting on the mass is no longer
Kθi , but is now equal to Kε. Thus, again applying Newton’s
Angle
qi
second law, gives
d2 θo
J = Kε (6)
dt 2
But ε = θi − θo , hence
Time, t
d2 θo
J = K(θi − θo ) (7)
Figure 2. Response of position control system without feedback. dt 2
104 Foundations of Measuring
Electronic
Power
Handwheel differencing device Gearbox
qi
Electronic Electric
Positional +
ui − power motor
transducer
amplifier
qo
Positional
uo transducer
Rotatable mass
Figure 3. Position control system with feedback.
that has some friction present in the bearings, gearing
Output position qo
contacts, and seals.
Friction almost always acts against motion regardless
of direction and this will cause the oscillation to decay
2qi
eventually. There are, however, various forms of fric-
qi tion including:
• Stiction (the torque necessary to just cause motion to
Time t
begin)
• Coulomb friction (a constant torque independent of
Figure 4. Response of position control system with feedback. velocity) and
• Viscous friction (a torque which is directly proportional
and so to velocity).
d2 θo
J + Kθo = Kθi (8) Stiction and Coulomb friction both cause undesirable
dt 2
side effects (stiction producing stick–slip motion when
This will be recognized as the equation for simple harmonic the system is commanded to follow a constant velocity
motion – see Article 62, Second-order System Dynamics, input and Coulomb friction producing a constant offset in
Volume 1. response to a constant input). It is thus essential to minimize
Hence, a sudden displacement in the handwheel posi- stiction and Coulomb friction by good mechanical design
tion will lead to continuous oscillations as illustrated in before any form of electronic compensation is applied.
Figure 4. Mathematicians define this system situation as As will now be shown, it is useful to ensure that the
critically stable because its oscillations will neither increase viscous component dominates or that a similar kind of
in magnitude nor decay; control engineers tend to regard dominating effect is reproduced by other means.
such a system as unstable for this is not the desired state The effect of viscous friction on the form of differential
of control due to the continuous oscillation. This system equation of motion for the system is to add an extra term
is exhibiting exactly the same type of behavior as Watt’s proportional to output angular velocity, thus
regulated steam engines did and it is entirely unsatisfactory
for practical position control systems when compared with d2 θo dθ
J + F o + Kθo = Kθi (9)
what can be achieved. dt 2 dt
where F is the viscous frictional torque per unit of angular
5 EFFECTS OF FRICTION IN REAL velocity.
FEEDBACK SYSTEMS Practically, the required viscous damping can be achieved
by either attaching a physical viscous damper to the rotating
To show how a more stable feedback system can be made, mass or more readily done by feeding back an extra
let us now look at the apparently nonideal practical system signal that is derived from another transducer – this time
Nature and Scope of Closed-loop Systems 105
dqo
ks
dt
Electronic
Power Gearbox
Handwheel differencing
circuit
qi
− Electronic Electric
Positional +
− power motor
transducer
amplifier
qo Tachogenerator
Positional
transducer
Rotatable mass
Figure 5. Position control system stabilized with velocity feedback.
a tachogenerator, which is directly proportional to angular This can be recognized as a second-order harmonic sys-
velocity. tem, but this time with damping – see Article 62, Second-
The second of these arrangements is shown in Figure 5. order System Dynamics, Volume 1.
Electric motor-tachogenerators built as a combined unit
are commercially available for this purpose, or a precision
6 NATURE OF CLOSED-LOOP SYSTEMS
tachogenerator can be added to the gear train, or other
output shaft. USED IN MEASUREMENT SYSTEMS
The electronic differencing circuit will now produce a
signal The analysis of the response of this system to vari-
ous inputs – see Article 60, Zero-order System Dynam-
dθo ics, Volume 1; Article 61, First-order System Dynamics,
k t ε − ks (10)
dt Volume 1; and Article 62, Second-order System Dynam-
ics, Volume 1 – is important for two dominant reasons;
where ks is the tachogenerator constant. firstly, the arrangement forms the basis of most chart
Hence, the drive torque T produced by the motor is given recording instruments, printers, plotters, cruise control in
by automobiles, and numerous other industrial applications
involving measuring systems used in practice; and sec-
dθo
T = k m k a k t ε − ks (11) ondly, it represents the embodiment of the second-order
dt
system which is used as an important reference in the design
of higher-order systems.
Again assuming that all forms of unwanted friction are
Feedback systems are more likely to be associated with
negligible, and using Newton’s second law
the larger power systems but they also find extensive
application in fine mechanisms. For example, the miniature
d2 θo dθo laser reading head in a CD-ROM head has three positional
J = k k
m a k t ε − k s
dt 2 dt servos acting to follow the disk. It is also used in feedback
dθo type sensors – see Article 113, Force-feedback Sensors,
= Kε − km ka ks (12) Volume 2 and Jones (1979).
dt
While the implementation of servos is usually done with
Hence, digital signals today, the underlying thinking is as given
above. The difference in digital systems is that the analog
d2 θo dθ linear theory given above has to be transposed into the
J 2
+ km ka ks o + Kθo = Kθi (13) discrete mathematical domain due to the sampling carried
dt dt
out by the digitizer.
The term km ka ks can be regarded as the equivalent When designing instrument feedback arrangements, it is
viscous frictional constant, F . wise to make them linear systems if at all possible, for then
106 Foundations of Measuring
their behavior can be well modeled and predicted. If not, REFERENCES
then modeling is considerably harder, but still feasible.
Texts on classical linear control theory matured by the Atkinson, P. (1972) Feedback Control Theory for Engineers, 1st
1960’s; today, the topic does not attract much attention by edn Heinemann, London.
publishers in new titles. However, most titles on process Bennett, S. (1979) A History of Control Engineering: 1800–1930,
control engineering will cover the topic. A time-honored Peter Peregrinus, London.
source is Levine (1996). Some titles appearing under con- Hazen, H.L. (1934) Theory of Servo-mechanisms. Journal of the
trol theory are best left for those who need and can handle Franklin Institute, 218(3), 279–331.
the complex mathematics other than simple linear systems. Jones, B.E. (1979) Feedback Instruments. Journal of Physics
Instrument designers will usually make use of modeling E-Scientific Instruments, 12, 145–158.
tools to set up a system – see Article 105, Modeling with Levine, W.S. (1996) The Control Handbook, CRC Press, Boca
LabVIEW , Volume 2. Raton, FL.
Maxwell, J.C. (1868) On Governors. Proceedings of the Royal
Society of London, 16, 270–283.
RELATED ARTICLES Minorsky, N. (1922) Directional Stability of Automatically
Steered Bodies. Journal of the American Society of Naval
Article 37, Relationship Between Systems in the Time Engineers, 42, 280–309.
and Frequency Domain, Volume 1; Article 38, Stability Sydenham, P.H. (1979) Measuring Instruments: Tools of Knowl-
Issues, Volume 1; Article 59, Introduction to the edge and Control, Peter Peregrinus, Stevenage.
Dynamic Regime of Measurement Systems, Volume 1; Sydenham, P.H. (1982) Handbook of Measurement Science,
Article 60, Zero-order System Dynamics, Volume 1; Wiley, Chichester.
Article 61, First-order System Dynamics, Volume 1;
Article 62, Second-order System Dynamics, Volume 1.
This article is based in part on a chapter by P. Atkinson originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
19: Dynamic Behavior of Closed-loop Systems
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
The Laplace transform of a signal θi (t) is formally
1 Use of the Laplace Transform in Control defined as i (s), in which s (p is also often used) is the
Systems Investigations 107 complex variable σ + jω and
2 Analysis of Simple Position Controller with ∞
Damping and Velocity Feedback 108 i (s) = θi (t)e−st dt (1)
0−
3 Concept of Damping Ratio and Undamped
Natural Angular Frequency 109 and
c+jω
1
4 Frequency Response of Closed Loop Systems 110 θi (t) = i (s)est ds (2)
5 Second-order Correlations 111 2πj c−jω
Related Articles 111 Here, c is chosen to be larger than the real parts of all the
References 111 singularities of i (s).
Fortunately, there is rarely any need to evaluate these
integrals in practice because they have been tabulated in
transform pairs to aid the rapid solution of differential
1 USE OF THE LAPLACE TRANSFORM equations – see Table 1 for the commonly seen transforms.
IN CONTROL SYSTEMS Modeling tools will usually have them embedded in their
data libraries.
INVESTIGATIONS In the presence of zero initial conditions, we may trans-
form derivatives by the rule
This article continues the discussion on instrument feedback
systems and is based on original material in Atkinson dn f (t)
L = s n F (s) (3)
(1982). As shown in Article 18, Nature and Scope of dt n
Closed-loop Systems, Volume 1, the differential equation and integrals by the rule
of a simple linear feedback control system can be derived t
by the application of physical laws (Newton’s second law F (s)
L f (t)dt = (4)
of motion in the example given). 0 s
However, in order to determine the behavior of the in which L represents the operation of taking Laplace
system in response to certain inputs, that is, as it responds transforms and F (s) is the Laplace transform of f (t).
to a change in input and has not settled down to the steady In situations in which the initial conditions are nonzero,
state, we need to have available a method of analysis for then
such systems when they are disturbed with a range of
inputs; the method of Laplace transforms now used is also dn
L [f (t)] = s n F (s) − s n−1 f (0−)
covered in Article 36, Systems in the Frequency Domain, dt n
Volume 1 and Article 143, DFT and FFTs, Volume 3. − s n−2 f 1 (0−) · · · − f n−1 (0−) (5)
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
108 Foundations of Measuring
Table 1. Some functions and their Laplace transforms. Volume 1, it is necessary to define the form of the input
f (t) F (s) = L [f (t)] signal and the all-important initial conditions.
Assuming zero initial conditions, both sides of the dif-
Unit impulse function δ(t) 1 ferential equation can be transformed, thus
1
Unit step function u(t)
s J s 2 o (s) + F so (s) + Ko (s) = Ki (s) (7)
1
Ramp function u(t)t
s2 Hence,
1
Exponential delay u(t)e−αt
s+α Ki (s)
−αt α o (s) =
Exponential rise u(t) (1 − e ) + Fs + K
J s2
s(s + α)
1 (K/J )i (s)
u(t)t e−αt = 2 (8)
(s + α)2 s + (F /J )s + K/J
ωrt
u(t)sin(ωrt t) Consider the response of this characteristic equation to
s + ωrt2
2
ωrt various input signals.
u(t)e−αt sin(ωrt t)
(s + α)2 + ωrt2
s+α
u(t)e−αt cos(ωrt t) 2.1 Response to the unit impulse as an input
(s + α) + ωrt2
demand
f (t − τ ) e−sτ F (s)
If θi (t) = δ(t), then i (S) = 1, and so
and
K/J
t
1 f −1 (0−) o (s) = (9)
L f (t)dt = F (s) + (6) + (F /J )s + K/J
s2
s s
0
√
If we take the case in which F < 2 (JK ) and complete
where f (0−), f 1 (0−), . . . , f n−1 (0−) are the values of the the square in the denominator; then
function, and its n − 1 derivatives and f −1 (0−) is the value
of the time integral of f (t) just prior to the application of K/J
o (s) = (10)
the signal at t = 0. (s + α)2 + ωrt2
It should be noted that the limit t = 0− is used here,
whereas in rigorous mathematical texts, in which the deriva- where
1/2
tives of discontinuous functions are not legitimate func- F K F2
tions, the lower limit is quoted as t = 0+. However, in α= and ωrt = − 2
2J J 4J
practical engineering analysis in which the unit impulse
function (δ(t)) is used, a more consistent methodology This does not quite agree with the form of any of the
results by using a lower limit t = 0−. tabulated Laplace transforms, but with slight manipulation
Notice that when control engineers use the single-sided it becomes
Laplace transform, all the driving signals are considered K/J ωrt
to operate after t = 0; they are defined as zero prior to o (s) = (11)
ωrt (s + α)2 + ωrt2
this instant and this may be conveniently represented as
multiplying all the time functions by the unit step u(t). which may be inverse transformed to yield
K/J −αt
θo (t) = u(t) e sin(ωrt t) (12)
2 ANALYSIS OF SIMPLE POSITION ωrt
CONTROLLER WITH DAMPING This is the unit impulse response of the system.
AND VELOCITY FEEDBACK
2.2 Response to the unit step as an input demand
Laplace transforms that have now been introduced allow
analysis of a typical closed loop controller having damping If θi (t) = u(t), then i (s) = 1/s, thus
and velocity feedback.
Returning to the simple position system described in K/J
o (s) = (13)
Article 18, Nature and Scope of Closed-loop Systems, s[s 2 + (F /J )s + K/J ]
Dynamic Behavior of Closed-loop Systems 109
This must be broken into partial fractions (done using q0
special mathematical techniques not covered here) to find
matching transforms, thus
t
1 α ωrt s+α
o (s) = − − (14) (a)
s ωrt (s + α)2 + ωrt2 (s + α)2 + ωrt2 q0
Each term has a recognizable inverse Laplace transform
leading to the solution in the time domain of
α
θo (t) = u(t) − u(t) e−αt sin(ωrt t) − u(t) e−αt cos(ωrt t)
ωrt Mpt
√
= u(t){1 − e−αt [1 + (α/ωrt )2 ]
× sin[ωrt t + tan−1 (ωrt /α)]} (15) (b)
Tp t
This is the unit step response. q0 ess
2.3 Response to the unit ramp as an input
demand
If θi (t) = u(t), then i (s) = 1/s 2 , and so
K/J t
o (s) = (16) (c)
s 2 [s 2 + (F /J )s + K/J ]
e
This must be broken into partial fractions to yield the ep
response
F Ke−αt
θo (t) = u(t) t − + sin(ωrt t + φ) (17) (d)
t
K F ωrt
where Figure 1. Time domain responses of a stabilized position control
system: (a) impulse, (b) step, (c) ramp output, (d) ramp error.
2αωrt
φ = tan−1
2α 2 − K/J
1. F = 0, in which case the system will oscillate continu-
This is the unit ramp response. ously with sinusoidal oscillations of angular frequency
√
These responses are plotted in Figure 1. (K/J ). This quantity is termed the natural undamped
The quantities Mpt , Tp , εss , and εp form some of the basic angular frequency. This response is termed critically
means of performance specification. stable.
√
A similar set of responses are analyzed in Article 62, 2. F < 2 (JK ), in which case the response contains an
Second-order System Dynamics, Volume 1, where sys- exponentially damped sinusoidal mode and will exhibit
tems in general are covered. The different terminologies, overshoot in response to a step input. This is called the
differing parameter, and process of analyses given will underdamped response.
√
perhaps be found useful depending upon the discipline to 3. F = 2 (JK ), for which the response is critically dam-
which they are being applied. ped , that is, it does not quite overshoot in response to
a step input.
√
4. F > 2 (JK ), for which the response is a double expo-
3 CONCEPT OF DAMPING RATIO AND
nential rise in response to a step input. This is called
UNDAMPED NATURAL ANGULAR the overdamped response.
FREQUENCY
The simple example given in article Article 18,
The value of the damping term F relative to the terms J Nature and Scope of Closed-loop Systems, Volume 1
and K governs the dynamic behavior of the system. There concentrates on condition 2 because this is in practice the
are four possibilities: most important case as it allows the system to settle within
110 Foundations of Measuring
The Laplace transform of the output of the system is
z = 0.4
related to the Laplace transform of the input by the equation
z = 0.6
1.2
s 2 o (s) + 2ζ ωn so (s) + ωn2 o (s) = ωn2 i (s) (20)
q0
qi
1.0
Dimensionless output
z = 0.8 For steady state sinusoids, it is permissible to substitute
0.8
jω = s, where ω is the angular frequency of the input,
z = 1.0
and produce the following operational relationship between
0.6
i (jω) and o (jω):
z = 1.5
o (jω) ωn2
0.4 =
i (jω) (jω)2 + 2ζ ωn (jω) + ωn2
0.2 ωn2
=
ωn2 − ω2 + j2ζ ωn ω
2 4 6 8 1
= (21)
Dimensionless time wnt 1 − (ω/ωn )2 + j2ζ ω/ωn
Figure 2. Dimensionless unit step response.
1.4
a given tolerance band around the desired value faster than
any other. 1.2
It is convenient to normalize the effect of the damping
1.0
(ju )
term F by relating it to the value Fc required to achieve
√
critical damping where Fc = 2 (JK ).
q0
qi
0.8
Magnitude
The damping ratio ζ is defined as
0.6 1.5 1.0 0.8 0.6 z = 0.4
F F
ζ = = √ (18)
Fc 2 (JK ) 0.4
so that 0.2
ζ < 1 is an underdamped system
ζ = 1 is a critically damped system 0.4 0.8 1.2 1.6 2.0 2.4 2.8
ζ > 1 is an overdamped system (a) Frequency ratio, u
The differential equation may be rewritten in terms of ωn
and ζ thus 0
d2 θo dθ
+ 2ζ ωn o + ωn2 θo = ωn2 θi (19)
dt 2 dt −40
Phase angle, f (deg)
The response to a unit step input for various values of ζ is
illustrated in Figure 2. −80
4 FREQUENCY RESPONSE OF CLOSED −120
z = 1.5
LOOP SYSTEMS 1.0
0.8
0.6
The steady state behavior of the system in response to a −160 0.4
sinusoidal input is of considerable practical importance.
0 0.4 0.8 1.2 1.6 2.0 2.4 2.8
When we refer to the frequency response of a system, we
(b) Frequency ratio, u
mean the variation of the phase and magnitude of the steady
state output of the system, as the frequency of the input Figure 3. Frequency response characteristics: (a) magnitude,
sinusoid is varied over the range of interest. (b) phase.
Dynamic Behavior of Closed-loop Systems 111
1.0
2.5 0.8
wrf / wrt
0.6
2.0 Mpf
Mp
Mpt
0.4
1.5
0.2
1.0
0 0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8
Damping ratio z Damping ratio z
Figure 4. Time and frequency domain characteristics of the second-order system.
It is simpler to work in terms of a nondimensional equations; these may be combined as graphs, which
frequency ratio u = ω/ωn for which are useful in the approximate design of higher-order
systems (Atkinson, 1968). These graphs are given in
o (ju) 1
= (22) Figure 4.
i (ju) (1 − u ) + j2ζ u
2
Levine (1996) provides a general background to control
From this expression, we can determine the modulus, M, its systems analysis.
peak value, Mpf (if any), and the phase, φ, which is given by
1
M=√ (23) RELATED ARTICLES
[(1 − u)2 + (2ζ u)2 ]
1
Mpf = √ (24) Article 18, Nature and Scope of Closed-loop Systems,
2ζ (1 − ζ 2 )
Volume 1; Article 20, Closed-loop Sampled Data Sys-
and tems, Volume 1; Article 21, Nonlinear Closed-loop Sys-
−1 2ζ u tems, Volume 1; Article 27, Signals in the Frequency
φ = −tan (25)
1 − u2 Domain, Volume 1; Article 28, Signals in the Time
Domain, Volume 1; Article 29, Relationship Between
The magnitude and phase characteristics are illustrated in
Signals in the Time and Frequency Domain, Volume 1.
Figure 3. The angular frequency at which the frequency
response has its peak value is designated by the symbol
ωrf . It may be shown that
√ REFERENCES
ωrf = ωn (1 − 2ζ 2 ) (26)
Atkinson, P. (1968) Feedback Control Theory for Engineers, 1st
edn, Heinemann, London.
5 SECOND-ORDER CORRELATIONS Atkinson, P. (1982) Closed loop systems, in Handbook of Mea-
surement Systems, Wiley, Chichester.
The time domain (step response) and frequency response of Levine, W.S. (1996) The Control Handbook, CRC Press, Boca
the second-order system are connected through correlating Raton, FL.
This article is based in part on a chapter by P. Atkinson originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
20: Closed-loop Sampled Data Systems
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
it is necessary to understand the sampling process. It is
1 Introduction to Closed-loop Sampled Data not usually expected that a measuring systems designer
Systems 112 be familiar with carrying out this level of analysis but its
2 Use of Sample and Hold Circuits 112 principles do need to be generally appreciated.
3 Frequency Response Methods Applied to Sampling is defined in general as the generation of a
Sampled Closed-loop Systems 113 sequence of impulses at the sampling instants, with the area
4 Time Domain Analysis of Sampled Data or magnitude of each impulse being equal to the original
Closed-loop Systems 114 signal value at that time (Figure 1).
The notation x ∗ (t) represents the ideal impulse sampled
5 Stability of Closed-loop Sampled Data
version of the original signal x(t) entering the sampler.
Systems Using the Z-transform 116
Feedback systems can be effectively represented by the
6 Compensation of Sampled Data Closed-loop
block diagram shown in Figure 2, in which the continuous
Systems 116
elements have a transfer function H (s). The precise analy-
Related Articles 117 sis of this type of system can be very involved; however, in
References 117 many practical situations, the sampling interval Ts is short
compared with the time for a transient oscillation of the
whole system. Under these circumstances, it is possible to
1 INTRODUCTION TO CLOSED-LOOP replace the impulse sampler theoretically by a pure gain of
value 1/Ts and analyze the circuit by normal continuous
SAMPLED DATA SYSTEMS theory or simulation.
This article continues the discussion of instrument feed-
back systems and is based on original material in Atkinson
(1982). Today many practical control systems use discrete 2 USE OF SAMPLE AND HOLD CIRCUITS
digital signals. Their operation is an extension of analog
principles usually by addition of sampling of the signal In the arrangement illustrated in Figure 2, the impulse
stream. The prior discussion of closed-loop systems in Arti- sampler is placed in the error channel of the feedback
cle 18, Nature and Scope of Closed-loop Systems, Vol- system. This is a fairly normal situation in control systems;
ume 1 and Article 19, Dynamic Behavior of Closed-loop in a properly designed system, the main elements then act
Systems, Volume 1 have assumed operation using analog as a low-pass filter, which smoothes the sampled error so
continuous signals by an A/D converter – see Article 138, that the output θo (t) follows the input θi (t) over the required
Sampling Theory Relevant to DAS, Volume 3 and Arti- profile of inputs.
cle 139, Analog-to-Digital (A/D) Converters, Volume 3. Even in a system that has been designed correctly and
To understand the complications that can arise and how that has a reasonably high sampling rate, the response will
the digital controller system can be analyzed in depth, not be exactly the same as that of the equivalent linear
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Closed-loop Sampled Data Systems 113
x (t ) Impulse x ∗(t ) x (t ) x ∗ (t )
sampler Ts
Conventional symbol
of impulse sampler
x (t ) i (t ) x ∗(t )
t t t
Analog signal Notional train of Notional impulse-sampled
unit impulses signal
Figure 1. Action of a theoretical impulse sampler.
Impulse sampler,
x i(t ) x ∗i (t ) Zero– x o (t )
sampling interval Ts
order
Impulse hold
qi Continous qo sampler
e
+ elements Ts
− H (s ) x (t )
x i(t )
Figure 2. Closed-loop sampled data system.
x o(t )
system. It is often most economical to use low sampling
rates, in which case the intersample ripple on the output
may be very severe. In order to make the behavior of
the sampled data system more like that of the continuous Ts 2Ts 3Ts 4Ts 5Ts 6Ts 7Ts 8Ts 9Ts 10Ts 11Ts 12Ts t
system and particularly to reduce intersampling ripple on
the output, various forms of filter are used between the Figure 3. Action of the zero-order hold.
sampler and the continuous elements.
The simplest type of filter is the zero-order hold or clamp. past two values to predict (or estimate) the slope of the
The actions of an impulse sampler with a zero-order hold curve to the next sampling instant.
are shown in Figure 3.
The output is held at the previous sampled value until
the next sampling instant. The transfer function of the zero-
order hold takes the form: 3 FREQUENCY RESPONSE METHODS
APPLIED TO SAMPLED CLOSED-LOOP
1 − exp(−sTs )
SYSTEMS
s
The total response of a sample and hold circuit approxi- Although for many purposes frequency domain analysis
mates on average (ignoring harmonics) to a pure time delay may be employed for sampled data systems using the
of Ts /2. This approximation is valid down to quite low sam- approximate continuous equivalents described in Sections 1
pling rates and allows design to be performed by classical and 2 above, at very low sampling rates, or in cases
feedback control methods in the frequency domain. The where more accurate analysis is required, it is necessary
approximation can also be used for approximate transient to use a more accurate technique. Linvill (1951) developed
analysis even at quite low sampling rates. a formula for computing the frequency response of a
Further improvements in signal smoothing can be sampled data system by making a vector addition of all
achieved by using predictive hold circuits, which use the the harmonics produced by sampling.
114 Foundations of Measuring
Given a continuous signal of Laplace transform E(s), The coefficient of z−n corresponds to the value of the time
then the Laplace transform E ∗ (s) of the sampled signal is function at the nth sampling instant.
given by Classic control theory texts, such as Truxall (1955), show
+∞ how to calculate the response of a closed-loop sampled data
1
E ∗ (s) = E(s + jnωs ) (1) control system containing an impulse sampler in the error
Ts n=−∞ channel (Figure 5).
The z-transfer (or pulse transfer function) of the system
where n is an integer and ωs is the angular sampling is given by
frequency. The frequency response is then written in the
usual way by substituting s = jω, that is, o (z) G(z)
= (5)
i (z) 1 + H G(z)
+∞
1
E ∗ (jω) = E(jω + jnωs ) (2) It should be noted that HG(z) = H (z)G(z).
Ts n=−∞
The application of the closed-loop z-transform is illus-
trated by an example (Figure 6) in which the input is a unit
Frequency response loci (i.e. Nyquist diagrams) can be
constructed using this series to investigate stability of
the loop. x (t )
e (t )
4 TIME DOMAIN ANALYSIS OF
SAMPLED DATA CLOSED-LOOP
SYSTEMS k0 k1 k2
k3 k4
The method of z-transforms provides a basis for time-
domain analysis and stability analysis of a closed-loop Ts 2Ts 3Ts 4Ts t
sampled data system. Its main limitation is that it provides
Figure 4. Illustrating z-transform series representation of a sam-
information about signal amplitudes at the sampling instants pled signal.
only. It, therefore, provides no information regarding the
intersample ripple.
An impulse-sampled signal ε∗ (t) has a Laplace transform Impulse sampler
∗
ε (s), which contains s in the form exp(sTs ); the z-
transform is obtained by making the substitution z = Θ i (Z ) Θo(Z )
+ G (s )
exp(sTs ). The z-transform can be represented as a series −
∞
ε(z) = ε(nTs )z−n (3)
n=0
H (s )
where n is an integer.
We can interpret z−1 as a delay operator of Ts seconds, Figure 5. Sampled-data control system.
−2
z as a delay operator of 2Ts seconds, and so on. The
summation will take a general form
Impulse sampler
−1 −2 −3 −n (Ts = 1s)
ε(z) = k0 + k1 z + k2 z + k3 z + · · · + kn z (4)
Each term in the series contains the amplitude kn at the qi e 4 qo
+
− s (s + 1)
sampling instant nTs in the form kn z−n (See Figure 4).
The z-transforms for various functions of time are given
in Table 1 for reference.
Although the z-transform can be inverted in a number of
ways, it is generally best to expand the expression into a
power series in powers of z−1 by algebraic long division. Figure 6. Sampled-data system for worked example.
Closed-loop Sampled Data Systems 115
Table 1. Laplace and z transforms of commonly met time functions.
Time function f (t) Laplace transform Z-transform
1 z
Unit step u(t)
s z−1
1 TS z
Unit ramp u(t)t
s2 (z − 1)2
1 Ts2 z(z + 1)
Acceleration function u(t)t 2 /2
s3 2(z − 1)3
tn 1 (−1)n ∂ n z
u(t) lim
n! s n+1 α→0 n! ∂α n z − exp(−αTs )
1 z
u(t)e−αt
s+α z − exp(−αTs )
1 Ts z exp(−αTs )
u(t)t e−αt
(s + a)2 [z − exp(−αTs )]2
α [1 − exp(−αTs )]z
u(t) (1 − e−αt )
s(s + α) (z − 1)[z − exp(−αTs )]
ωrt zsin(ωrt Ts )
u(t)sin(ωrt t)
s 2 + ωrt2 z2 − 2zcos(ωrt Ts ) + 1
ωrt z exp(αTs )sin(ωrt Ts )
u(t)e−αt sin(ωrt t)
(s + α)2 + ωrt2 z2 exp(2αTs ) − 2z exp(αTs )cos(ωrt Ts ) + 1
s z[z − cos(ωrt Ts )]
u(t)cos(ωrt t)
s 2 + ωrt2 z2 − 2zcos(ωrt Ts ) + 1
s+α z2 − z exp(−αTs )cos(ωrt Ts )
u(t)e−αt cos(ωrt t)
(s + α)2 + ωrt2 z2 − 2z exp(−αTs )cos(ωrt Ts ) + exp(−2αTs )
step function and it is required to calculate the output. Now θi (t) = u(t). Hence, again using the table of trans-
forms:
4 4 4
H (s) = = − (6)
s(s + 1) s s+1 z
i (z) =
z−1
From the table of z-transforms (Table 1) 2.53z z
o (z) = (10)
z2 + 1.16z + 0.368 z − 1
4z 4z
H (z) = − (7) 2.53z2
z − 1 z − exp(−αTs ) =
z3 + 0.16z2 − 0.793z − 0.368
where αTs = 1. Thus,
By algebraic long division this expression yields:
2.53z
H (z) = (8)
z2 − 1.37z + 0.368 o (z) = 2.53z−1 − 0.406z−2 + 2.03z−3 + 0.339z−4
+1.37z−5 + 0.827z−6 + 1.06z−7 + 1.00z−8 + · · ·
and so (11)
As can be seen from this response, although the system is
o (z) H (z) 2.53z highly undamped, its output eventually converges towards
= = 2 (9)
i (z) 1 + H (z) z + 1.16z + 0.368 the input.
116 Foundations of Measuring
5 STABILITY OF CLOSED-LOOP systems. Series and parallel or series/parallel compensa-
SAMPLED DATA SYSTEMS USING tion using continuous transfer functions can be designed
in a similar way using a Nyquist or inverse Nyquist
THE Z-TRANSFORM diagram based on the Linvill approximation (Section 3
above).
For a sampled data control system, the z-transfer function Alternative strategies involve series digital compensators
is given by alone (Figure 7a) or in combination with a parallel contin-
o (z) G(z) uous compensator (Figure 7b).
= (12)
i (z) 1 + H G(z) Digital compensators can be arranged to produce a phase-
lead for which Hc (z) = (z − a)/(z + b), with the zero lying
The stability of the system depends on the positions of to the right of the pole in the z-plane, or a phase lag for
the zeros of [1 + H G(z)] in the s-plane. However, the which Hc (z) = (z + a)/(z − b), with the zero lying to the
transformation of exp sTs = z has been made and the left of the pole.
positions of the zeros can be mapped in the z-plane. The Many other forms of a digital compensator could be real-
mapping, z = exp sTs , maps the imaginary axis of the s- ized by construction from hard-wired logical circuitry or,
plane into the unit circle about the origin of the z-plane more usually today, by means of software implemented on
and the left half of the s-plane into the interior of the a digital microprocessor almost always associated with a
unit circle. closed-loop controller; software implementation is clearly
A sampled data control system will only be absolutely very advantageous.
stable if all the zeros of 1 + H G(z) lie inside the unit The arrangement illustrating control and digital com-
circle. pensation of a continuous process is shown in Figure 8;
Direct application of this criterion can be tedious; use of
the error signal is generated by the computer and is
the bilinear transform z = (1 + w)/(1 − w) maps the inside
of the circle in the z-plane onto the left-hand side of the
w-plane. It is then possible to use the Routh–Hurwitz or
the Nyquist stability criterion directly. Plant
Operator Visual output
input Display
display DAC Plant
computer
unit
6 COMPENSATION OF SAMPLED DATA
CLOSED-LOOP SYSTEMS
ADC Transducer
Sampled data (i.e. digital) systems offer a wider range of
possibilities for compensation than do continuous signal Figure 8. Control and compensation using a digital computer.
Digital Hold or no
Impulse
compensator hold Plant
sampler
qi qo
+ H c(z ) H h(s ) H p (s )
−
(a)
Digital Hold or no
Impulse
compensator hold Plant
sampler
qi qo
+ H c1(z ) H h(s ) + H p (s )
− −
Continuous
parallel compenstor
H c2(s )
(b)
Figure 7. Forms of digital compensation: (a) series digital compensation; (b) series digital with parallel continuous compensation.
Closed-loop Sampled Data Systems 117
then operated on by a suitable algorithm before it is REFERENCES
outputted to the plant via a digital-to-analog converter
(DAC). Atkinson, P. (1982) Closed-loop systems, in Handbook of Mea-
This article has only introduced the foundational surement Science, Vol. 1 (ed. P.H. Sydenham), Wiley, Chich-
approach to understanding the operation and tuning of ester (pp. 591–640).
digital closed loops. More detail can be found in Franklin Bishop, R.H. (1997) Modern Control Systems Analysis and Design
et al. (1997), Nekoogar and Moriarty (1998), Ogata Using Matlab and Simulink, Addison-Wesley.
(2001), Stubberud, Williams and DiStefano (1994), and Franklin, G.F., Powell, D.J., Workman, M.L. and Powell, D.
Wittenmark (1990). Application is now assisted using the (1997) Digital Control of Dynamic Systems, Addison-Wesley.
current, widely available circuit system simulation tools, Linvill, W.K. (1951) Sampled-data Control Systems Studies
Bishop (1997). through Comparison of Sampling with Amplitude Modulation.
Transactions AIEE, 70, Part II, 1779–88.
Nekoogar, F. and Moriarty, G. (1998) Digital Control Using
Digital Signal Processing, Prentice Hall.
RELATED ARTICLES Ogata, K. (2001) Modern Control Engineering, Prentice Hall.
Stubberud, A.A., Williams, A.J. and DiStefano, J.J. (1994)
Schaum’s Outline of Feedback and Control Systems, McGraw-
Article 18, Nature and Scope of Closed-loop Systems, Hill Trade.
Volume 1; Article 19, Dynamic Behavior of Closed-loop Truxall, J.G. (1955) Introductory System Engineering, McGraw-
Systems, Volume 1; Article 37, Relationship Between Hill, Tokyo.
Systems in the Time and Frequency Domain, Volume 1; Wittenmark, B. (1990) Computer Controlled Systems, Prentice
Article 38, Stability Issues, Volume 1. Hall.
This article is based in part on a chapter by P. Atkinson originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
21: Nonlinear Closed-loop Systems
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
under control will be limited in amplitude. The result of
1 Introduction to Nonlinear Systems 118 this will normally be that the response time of the system
2 Methods for Studying Nonlinear Systems 119 will be longer than predicted by means of linear theory,
3 Phase-plane Analysis 120 which is based on the assumption that saturation is not
present.
4 Describing Function Method for Investigating
There are many types of nonlinearity. They were first
Nonlinear Closed-loop Systems 120
distinguished in Gelb and Vander Velde (1968) as explicit
5 How to Approach the Design of Closed
or implicit nonlinearities.
Loops in Measurement Systems 122
Explicit implies that the output of the nonlinearity is
Related Articles 123 explicitly determined in terms of the required input vari-
References 123 ables, whereas implicit implies a more complicated rela-
tionship between input and output through, for example, an
algebraic or differential equation.
1 INTRODUCTION TO NONLINEAR With explicit nonlinearities, we have to distinguish bet-
ween static and dynamic forms, in which dynamic implies
SYSTEMS
that the output is related not only through the input but
This article continues the discussion of instrument feed- also through the derivatives of the input. Among the
back systems and is based on original material in Atkin- explicit, static nonlinearities, we must again divide between
son (1982). single-value (memoryless) nonlinearities such as saturation
Although certain nonlinearities (Coulomb friction and and dead-space, and multiple-value (memory) nonlinearities
stiction) were mentioned in connection with the posi- such as hysteresis.
tional servomechanism in Article 18, Nature and Scope The analysis and design of nonlinear systems is vastly
of Closed-loop Systems, Volume 1, the theory covered more complex than that of linear systems, every nonlinear
there and in Article 19, Dynamic Behavior of Closed-loop system being a miniature universe. Fortunately, simulation
Systems, Volume 1 and Article 20, Closed-loop Sampled tools are available for such work that make design easier
Data Systems, Volume 1 describes the class of closed loop than by long hand formal methods.
that are called linear systems. Although some nonlinearities are intentionally inserted
Linear systems are those which obey the principle of by the designer to obtain improved system performance,
superposition. Although linear theory is an indispensable for the most part, they are a nuisance causing undesirable
design aid, it must be realized that all practical systems do side effects when the error signal is large (e.g. the effect of
inevitably contain nonlinearities to some degree. saturation) or when it is small (such as, backlash in gears
For example, an amplifier can exhibit saturation if that may cause tick to occur). System errors of this kind are
overdriven; when the error signal in a control system discussed in Article 57, Static Considerations of General
becomes large the control signal driving the equipment Instrumentation, Volume 1.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Nonlinear Closed-loop Systems 119
High-gain 2 METHODS FOR STUDYING
linear Nonlinear
element element NONLINEAR SYSTEMS
qi qo
+ G K
− There are no completely satisfactory methods of study-
ing nonlinear systems. Computer simulations are of great
importance; to perform a simulation it is, however, neces-
H
sary to choose a model to represent the important charac-
Linear feedback teristics of a system.
element A computer tool study of any nonlinear system is an
essential final step before the real system is built.
Figure 1. Nonlinear element in high-gain feedback loop. Trial and error design based entirely on computer sim-
ulation is not advisable as it may not give a complete
In some circumstances (as with hysteresis), it may be pos- insight into the system behavior and the strategies which
sible to effectively eliminate the effects of the nonlinearity are available.
by means of a high-gain stable local feedback loop. This In instrumentation systems design, it is often possible
procedure is strongly recommended where it is possible. to avoid nonlinearities by choice of components and their
The procedure is based on the idea that the nonlinear ele- working regimes. For efficient lowest cost design it is,
ment has a variable static gain, K. It is driven from a high- however, often better to make use of nonlinear effects
gain linear amplifier of gain, G, and feedback is applied meaning that more investigation into the behavior will
through an element of gain, H , as shown in Figure 1. be needed.
Analysis shows that the transfer function in such cases The first step in designing a nonlinear system is usually
is given by to attempt to linearize the essential nonlinearities for small
signals about a working point.
θo GK
= (1) For example, if we have a nonlinear device with an input
θi 1 + H GK signal x(t) and an output y(t) such that y(t) = kx 2 (t) then
∂y/∂x = 2kx
The method relies on making HGK > 1 for all possible If we are working at some particular working point
values of K, so that xp , then for small departures from the working point, the
θo 1 effective gain of the device is 2kxp .
> (2)
θi H For working points at values of x greater than xp , the
gain will be greater than 2kxp and for working points at
The effects of the nonlinearity are thus made quite negli- values of x less than xp , the gain will be less.
gible so long as none of the elements saturate, this point Linear systems design application, combined with a sen-
often being overlooked. sitivity analysis and a computer simulation should again
This arrangement is only of use so long as the loop is prove adequate.
relatively stable, so careful design of the loop is essential. Phase-plane analysis is a useful tool in the analysis of
This is probably best performed in the frequency domain nonlinear second-order systems – see Article 62, Second-
using the describing function method given in Section 4. order System Dynamics, Volume 1 – subjected to step
Stiction can cause stick–slip motion in feedback mech- (and ramp) inputs.
anisms following constant velocity (ramp) inputs. This The technique involves the determination of the response
problem can again be overcome by means of a high- in terms of its derivative of output (or error) plotted as a
gain local feedback loop. In this case, velocity feedback function of its output (or error). This plot is called a phase-
is used around a motor–amplifier series combination; the plane diagram. In formulating the relevant equations, time
gain of the amplifier is made very high so that the arrange- is removed explicitly. For many practical nonlinearities,
ment behaves like a perfect integrator of transfer function where the nonlinearities can be represented by linear seg-
1/ks s where, ks , is the velocity feedback constant. The mented characteristics, the phase plane can be divided into
arrangement is again identical to that shown in Figure 1; various regions each of which corresponds to motion on a
K represents the nonlinear motor characteristic H = ks s, particular linear segment of the nonlinearity (see Section 3).
and G is the high gain of the amplifier. There is usually For higher-order systems, on occasions where it is nec-
very little problem in designing this loop to have adequate essary only to determine whether the designed system will
relative stability. remain absolutely stable for the entire envelope of input
120 Foundations of Measuring
signals, the methods of Lyapunov or Popov will provide e•
exact answers without solving the differential equations.
The method of describing functions based on the concept
of quasi-linearization for a given class of input signals
provides a sound basis for the analysis and design of
N =0
nonlinear feedback-control systems (see Section 4).
C S
K K A
B −C D e
3 PHASE-PLANE ANALYSIS K Isocline
Phase-plane analysis considers any second-order differen- N = 1.8
tial equation of form:
Trajectory
ẍ + Aẋ + Bx = F (3) N =0
where A, B, and F are not necessarily constant. A phase Figure 2. Phase trajectory for simple positional servomechanism
portrait consists of a number of phase trajectories in the ẋ- with Coulomb friction.
versus x-plane.
If we define y = ẋ, then in which N is the slope of the phase trajectory where it
crosses the isocline.
dy dy/dt F − Bx − Ay
= = (4) The first term defines the family of isoclines for the linear
dx dx/dt y system, whereas the second term introduces the effect of the
This equation represents the slope of the phase trajectory nonlinearity.
in terms of functions of x and y. In general B, A, and F The focal point is changed from ε = +C/K to ε =
may be functions of x and y. To find the phase trajectory −C/K as ε̇ changes from a negative value to a posi-
itself, this equation must be integrated; sometimes this can tive value.
be performed analytically, but usually it is better done If a trajectory begins at ε = A (where A is equivalent
numerically using a digital computer or, the original long to the value of the input step), it transverses through the
hand way, graphically using the method of isoclines. phase plane as shown in Figure 2. The slope of the phase
By way of example, consider the simple viscously trajectory is given by the value of N as each isocline is
damped second-order servomechanism described in Arti- crossed. The determination of the passage of the trajectory
cle 18, Nature and Scope of Closed-loop Systems, Vol- through the isoclines to point B when ε̇ becomes negative,
ume 1 but now with the additional Coulomb friction. thence to point D when it again becomes positive is a simple
Consider the response of the system with a step input and matter. Motion ceases at D because the generated torque is
let C be the magnitude of the Coulomb frictional torque, less than C.
which always opposes motion. The phase-plane technique can be used to analyze the
The instantaneous accelerating torque in Kε and the behavior of the second-order system for a variety of com-
retarding torque is F θ̇o + C sin θ̇o , where sin θ̇o is positive monly encountered nonlinearities.
for θ̇o > 0 and negative for θ̇o < 0. Thus, applying New- The main disadvantage of the phase-plane method is
ton’s second law, we have that it cannot be extended to higher-order systems in a
satisfactory manner.
J θ̈o = Kε − F θ̇o − C sin θ̇o (5)
4 DESCRIBING FUNCTION METHOD
Now ε = θi − θo , thus θo = θi − ε; also for a step input
dθi /dt = 0 and d2 θi /dt 2 = 0 under steady state conditions. FOR INVESTIGATING NONLINEAR
It is possible to translate the above equation into the error CLOSED-LOOP SYSTEMS
form
In its most simple form, the describing function method is
J ε̈ + F ε̇ + Kε + C sin ε̇ = 0 (6)
an extension of ordinary transfer function analysis to take
into account the effect of single nonlinearities in systems
From this equation we can deduce
excited by sinusoidal input.
Kε C sin ε̇ It is particularly useful as a method for predicting the
ε̇ = − − (7) amplitude and frequency of limit cycles, but it can also be
NJ + F NJ + F
Nonlinear Closed-loop Systems 121
used to assess relative stability by conventional frequency For memoryless nonlinearities the describing function is
response methods. purely real, while memory-type nonlinearities introduce a
In essence, the describing function of a nonlinear element phase shift, causing the describing function to be complex.
is the gain of the element in terms of the ratio of its In general, if the fundamental output is expressed as a
fundamental component output to a sinusoidal input of complex operator
given magnitude and frequency.
Whereas the gain of a linear element is only ever fre- Vo (jω) = V̂o cos θ + jV̂o sin θ (8)
quency dependent the gain of a nonlinear element is always
amplitude dependent and, additionally, can also be fre- where V̂o is the peak fundamental output and θ is the phase
quency dependent. shift, then the describing function GD is given by
Here we will confine our study to amplitude depen-
dent describing functions. In order to illustrate this point, V̂o cos θ V̂o sin θ
GD = +j (9)
consider a very common form of nonlinearity, namely, V̂i V̂i
saturation.
The output is linearly related to the input for small where V̂i is the peak input.
positive or negative excursions of the input, but the output To derive a describing function from this, we must use
reaches a limiting value for large excursions (see Figure 3). the Fourier series representation for the output
The response of this nonlinearity to a sinusoid will be
sinusoid for small signals, a clipped sinusoid for medium vo (t) = 12 Ao + A1 cos(ωt) + A2 cos(2ωt)
size signals, and tending to a square wave of magnitude
KE s for very large signals. + A3 cos(3 ωt) + · · · + B1 sin(ωt) + B2 sin(2 ωt)
The gain of the element, based on the ratio of fundamen- + B3 sin(3 ωt) + · · · (10)
tal output to input thus is constant at a value K for small where
inputs; beginning to decrease as the input goes beyond Es π
2
and eventually trailing off towards zero as the fundamental AN = vo (t) cos(N ωt) d(ωt)
π 0
output tends towards its limiting value of 4KE s /π as the
input tends towards infinity (see Figure 4). and
π
2
BN = vo (t) sin(N ωt) d(ωt)
Output
π 0
vo
The definition of the describing function requires Ao = 0
KE s
^ and AN and BN for N > 1 to be negligible.
v i = V i sin(wt ) −E s vo The fact that most control systems contain low-pass
Input E s Input Output filtering elements, which filter out the harmonics to a
vi
KE s
substantial degree usually justifies these assumptions.
The describing function is then given by
B1 A1
+j in coordinate form
Figure 3. Illustrating a saturation nonlinearity. V̂i V̂i
or
Describing Function G D
K B21 + A21 A1
tan−1 in polar form
^
Vi B1
Evaluation of the Fourier integrals may be quite simple, as
in the case of the saturation nonlinearity but, sometimes,
extremely difficult.
Es Fortunately, the describing functions of some commonly
^ encountered nonlinearities have long been tabulated in
Vi
the past, Thaler and Pastel (1962) and Gelb and Vander
Figure 4. Describing function GD for a saturation nonlinearity. Velde (1968).
122 Foundations of Measuring
Nonlinear
Imaginary
Linear elements elements Linear elements
Θi(jw) w = wc
e Θo(jw)
+ G1(jw) GD G 2(jw)
− w=∞
Real
−1
GD
Figure 5. Nonlinear feedback system schematic.
The describing function for the saturation element is
given by G1(jw) G2(jw)
w 0
2K √ ◦
GD = [(sin−1 R) + R (1 − R 2 )] 0 (11)
π Figure 6. Nyquist diagram for a typical nonlinear system show-
ing −1/GD (unstable situation).
The Nyquist stability criterion – see Article 38, Stability
Issues, Volume 1 – may be involved for the determination
of the stability of systems containing a nonlinear element from which ωc may be calculated (generally by iterative
as shown in Figure 5. trial and error).
The closed-loop transfer function of this system is given The amplitude of the limit cycle can now be determined
by from
o G1 (jω)GD G2 (jω) |G1 (jω)||GD ||G2 (jω)| = 1 (16)
= (12)
i 1 + G1 (jω)GD G2 (jω)
which is the Nyquist amplitude condition for continuous
The Nyquist criterion is based on the characteristic equa- oscillation.
tion The value of GD can now be determined and the value
1 + G1 (jω)GD G2 (jω) = 0 (13) of V̂i , the signal input to the nonlinear element, can be
estimated by trial and error.
To avoid the need to plot numerous Nyquist diagrams The error signal entering the first linear element can
for every value of GD , we may reform the characteristic hence be computed from V̂i /|G1 (jωc )|.
equation as In the absence of any input to the system itself, the
error and the output are identical in magnitude so that
1
G1 (jω)G2 (jω) = − (14) θo = V̂i /|G1 (jωc )|. This is the magnitude of the resultant
GD limit cycle.
The encirclement of the locus −1/GD by the function In general, it is necessary to determine the conditions
G1 (jω)G2 (jω) plotted as a polar diagram now indicates graphically rather than analytically or by iterative trial
absolute instability. and error.
Relative stability can be assessed by treating the locus Atherton (1975) has shown how the method of describing
−1/GD as the equivalent of the critical point (−1, j0) used functions can be used to design compensating elements for
in linear systems design. nonlinear systems and be extended to treat multiple non-
The amplitude and frequency of a limit cycle can readily linearities, to handle transient oscillations, dual sinusoidal
be assessed by the point at which the frequency locus inputs, and random inputs. Systems modeling tools often
include means for investigating nonlinear systems.
G1 (jω)G2 (jω)
intersects the locus of −1/GD . This is illustrated in 5 HOW TO APPROACH THE DESIGN OF
Figure 6, which shows the Nyquist diagram of an unsta- CLOSED LOOPS IN MEASUREMENT
ble system that must limit cycle at angular frequency ωc , SYSTEMS
where
Many lengthy textbooks have been written on closed-loop
systems so that it is hardly surprising that the information
G1(jwc)G2(jwc) = −180° (15)
presented here merely touches the surface of a vast subject.
Nonlinear Closed-loop Systems 123
Material described here is based on the transfer function will incorporate feedback. The designers of instruments of
approach because this certainly offers the designer the most all kinds will, therefore, have to become more familiar with
comprehensive set of techniques. the analysis and design of feedback systems.
It should be understood, however, that many authors In essence, the design of measuring systems loops is
prefer to integrate the classical transfer function approach best kept manageable by avoiding nonlinear effects in any
with the more recent state-variable approach. elements under the designer’s control.
In the state-variable approach the system model is Those nonlinear effects that cannot be eliminated or
described in terms of n first-order differential equations, lessened at source should be reduced by the use of local
each equation being a separate description of the behavior feedback solutions.
of a particular state and its connection with the other states Manufacturing them in hardware form, even with a
and the driving inputs. control prototyping kit, is not recommended until they have
This form of description allows the equations for the been investigated thoroughly with an appropriate modeling
system states to be condensed into the form of a single tool, of which many are available including Matlab and
vector, x(t), and related to the driving input vector, u(t), Simulink – see Bishop (1997).
by the equation
ẋ(t) = Ax(t) + Bu(t) (17) RELATED ARTICLES
where A and B are coefficient matrices. Article 18, Nature and Scope of Closed-loop Systems,
The system outputs can then usually be related to the Volume 1; Article 19, Dynamic Behavior of Closed-loop
internal states by the vector equation Systems, Volume 1; Article 26, Signal Classification,
Volume 1; Article 88, Error Sources in Fine Mechanics,
y(t) = Cx(t) (18)
Volume 2.
where, C, is another coefficient matrix and, y(t), is the
output vector. REFERENCES
The solution to the first of these matrix equations can
be determined by various powerful computerized matrix Atherton, D.P. (1975) Non-linear Control Engineering, Van Nos-
methods and hence the response y(t) for a given u(t) can trand Reinhold, London.
be determined quickly and accurately. Atkinson, P. (1982) Closed Loop Systems, in Handbook of
Apart from their application in time-domain analysis, Measurement Science, Wiley, Chichester, 591–640.
state-variable techniques form the basis of the Lyapunov Bishop, R.H. (1997) Modern Control Systems Analysis and Design
stability analysis and of optimal control theory. It can also Using Matlab and Simulink, Addison-Wesley.
be used in other design techniques, such as pole assignment, Gelb, A. and Vander Velde, W.E. (1968) Multiple-input Describ-
and pole assignment and decoupling methods, using state ing Functions and Non-linear Systems, McGraw-Hill, New
vector feedback. York.
In order to obtain greater sensitivity and better accuracy Thaler, G.J. and Pastel, M.P. (1962) Analysis and Design of Non-
of measurement, it is likely that more and more instruments linear Control Systems, McGraw-Hill, New York.
This article is based in part on a chapter by P. Atkinson originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
22: Characteristics and Theory of Knowledge
Luca P. Mari
Università Cattaneo, Castellanza (VA), Italy
of such positions, the complexity of the topic derives from
1 The Problem of Knowledge 127 its inherent reflexivity, due to the fact that the object of
2 The Status of Realism 127 the analysis coincides with the tool by means of which the
analysis is performed: to know knowledge, only knowledge
3 Semiotics of Knowledge 129
can be employed). A basic dichotomy can be identified,
4 Pragmatic Classification of Models 129 whose elements play the role of competing attractors for
5 The Evaluation of Quality of Knowledge 130 an ideal continuum of positions: objectivism assumes that
6 Data and Inference in Knowledge 131 a world external to the subject exists independently of
7 Inexactness of Knowledge and Measurement 132 him and has predefined properties, existing as such before
8 (Non)foundations of Knowledge 132 they are acquired by the perceptive-cognitive system of the
subject, whose aim is to reconstruct them; on the other hand,
Related Articles 133
solipsism asserts that the cognitive system of the subject
Further Reading 133 projects his own world out of him, and the reality of such
a world is just an image of the laws internal to the system.
The position currently supported by the majority of
1 THE PROBLEM OF KNOWLEDGE scientists and engineers can be plausibly characterized as
a kind of ‘pragmatic realism’, close to but not coincident
with objectivism, according to which the conjoint efforts
Human beings know but do not definitely know what
of science and technology are aimed at reaching, and
knowledge is: traditions, prejudices, expectations, and pro-
actually guarantee, better and better, that is, more and more
jections are more or less always part of knowledge and
objective knowledge of the world whose properties are
make it a combination of objectivity and subjectivity. Ratio-
therefore progressively discovered.
nality allows some critical control on knowledge, and the
Measurement plays a crucial role in supporting this
recognition of the limitations to which human knowledge
realism.
is subject is precisely a result of rationality.
The interest in theorizing about knowledge arises from
the observation that different persons have different beliefs,
and ultimately that beliefs and facts are distinct: ‘theory 2 THE STATUS OF REALISM
of knowledge is a product of doubt’, as Bertrand Russell
wrote. In the history of both western and eastern culture, In acquiring and processing information from the world,
such a doubt has stimulated an impressive amount of human beings constantly produce models (and sometimes
research, ideas, and philosophical systems, and nevertheless theories: we will not emphasize here the distinction between
very different positions have been maintained and still models and theories, grounded on formal logic) of the
remain on the nature of knowledge and its object (it is world they observe, thus generating knowledge on it (see
reasonable to hypothesize that, more than from the plethora Figure 1).
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
128 Foundations of Measuring
Figure 3. A justification for realism: mutual understanding.
• Pragmatics: were knowledge only subjective, our abil-
ity to effectively operate on the world would be an
exception more than a rule (Figure 4).
Figure 1. Human beings produce models of the world they Realism can be then interpreted as a weak form of
observe.
objectivism: the world exists independently of us and both
intersubjective and pragmatic experiences lead us to assume
that it cannot be too different from our models of it.
3 On the other hand, to generate knowledge that can
be shared, subjective models must be expressed in some
socially understandable and usable form, such as statements
in a natural language or mathematical laws. This points
out a further basic issue on knowledge: ‘how can it be
1 2
Figure 2. The relations among subjects, world, and models.
Such knowledge results from the relations among the
three interacting entities, the subject, the world, and the
model, so that the relation between the world and its model
is not direct, but always mediated by the subject who
produced the models themselves as shown in Figure 2.
Whenever it remains individual, knowledge is just tacit
and usually implicit and as such it reduces to personal
experience that can be communicated only by person-to-
person means, as imitation.
Critical is therefore the objectivity, that is, the indepen-
dence from the subject, of the relation between the world
and its models. Realism assumes two operative reasons for
justifying the possibility of some objective knowledge:
• Intersubjectivity: were knowledge only subjective,
mutual understanding would be an exception more than
a rule (Figure 3). Figure 4. A justification for realism: effectiveness.
Characteristics and Theory of Knowledge 129
that mathematics, a product of human thought independent itself. . .) is a rather advanced entity, appearing initially
of experience, is so admirably adapted to the objects of within World 2, in the form of subjective experiences, and
reality’? in the words of Albert Einstein. then fully evolving in the context of World 3. The transition
Philosophers and scientists have formulated different from World 2 to World 3 corresponds to the social ability
opinions in this regard, more or less explicitly in refer- to communicate, and therefore to share, experience: that is
ence to a basic dichotomy: either ‘scientific laws faithfully why the availability of a (textual or nontextual) language
describe how the world is’ or ‘scientific laws are just syn- is considered the first step within World 3. Furthermore,
thetic means to express information about events in an the usage of a language gives knowledge a syntax and
aggregate way’. The former position implies a metaphysi- makes it a semiotic entity (see also Article 23, Principles
cal hypothesis on the nature of the world, classically stated of Semiotics as Related to Measurement, Volume 1).
as ‘numbers are in the world’ (Kepler) or by assuming that Given the complexity of the concept of knowledge and
‘the great book of nature’ cannot be understood ‘but by its fuzzy characterization, rather than trying a definition of
learning its language and knowing the characters in which it we suggest that the (possible) presence and the relative
it is written: it is written in mathematical terms’ (Galileo); importance of the semiotic components, syntax, semantics,
in contrast, the latter position suggests the economic nature and pragmatics, can be adopted as a criterion to distinguish
of science: since ‘in Nature the law of refraction does not among the different entities that are commonly deemed to
exist at all, but only different cases of refraction’, by means be (related to) knowledge. In particular,
of such a law ‘we do not have to keep in mind the count-
less phenomena of refraction in the various compositions • the exclusive availability of pragmatic information (‘to
of matter and under the various incidence angles, but only know how to do’), such as the competence shown
the rule that we call ‘law of refraction’, a much easier by many craftsmen, appears to be a limited kind of
thing’ (Mach). knowledge, if knowledge at all;
Measurement has been often adopted to justify the for- • the exclusive availability of syntactical information and
mer position. the ability of purely symbolic (i.e. only syntactical)
processing, as performed by most automatic devices,
appears to be a limited kind of knowledge, if knowledge
3 SEMIOTICS OF KNOWLEDGE at all.
The designation of ‘knowledge-based’ for the systems
Knowledge can be about the physical world but it is not
operating on the basis of an explicit semantics is a further
part of it. Given the realistic assumption of the inde-
argument to support the hypothesis that meanings are criti-
pendence of the physical world from the subject, both
cal for the emergence of ‘proper’ knowledge, and therefore
subjective and objective knowledge can be interpreted in an
that socially communicable knowledge (‘World 3 knowl-
evolutionary context as the results of mankind to adapt to
edge’) is an entity spanning all the semiotic components.
his (firstly only physical and then also social) environment.
In this regard, Karl Popper has suggestively proposed to
identify ‘some stages of the cosmic evolution’ as organized 4 PRAGMATIC CLASSIFICATION
in three ‘worlds’, as follows:
OF MODELS
World 1
0. hydrogen and helium If the pragmatic component is taken into account, dif-
1. heavier elements; liquids and crystals ferent purposes for knowledge can be recognized: mod-
2. living organisms els can be adopted for description, explanation, prevision,
prescription.
World 2 It is usual that the first stages of the development of a
new field of knowledge are devoted to the production of
3. sensitivity (animal conscience)
models aimed at the description of the system under analy-
4. conscience of self and death
sis. Typical outcomes of this work are the identification of
World 3 properties relevant to describe the system and their evalu-
5. human language; theories of self and death ation to classify the system itself into more or less rough
6. products of art, technology, and science. categories.
To overcome the conventionality of taxonomies and
In this framework, knowledge (whose object can belong whenever the available knowledge allows it, some relations
to either Worlds, and finally could even become knowledge among properties are identified, so that each property is
130 Foundations of Measuring
embedded in a network of dependencies. In such cases, knowledge loop, the quality of knowledge itself and the
the relational information that is (explicitly or implicitly) effectiveness of system control can always be enhanced.
conveyed by properties can be referred to in order to obtain On the other hand, in many situations prescriptions are
an explanation of the system state/behavior: the value of the required even when predictive, explanatory, and sometimes
property x1 is v1 because x1 is connected to the properties even socially agreed descriptive models are not available
x2 , . . . , xn by the relation R, and the properties x2 , . . . , xn (let us quote the crucial examples of medicine and business
have values v2 , . . . , vn respectively, and R(x1 , . . . , xn ). administration). In these cases, experiences and expecta-
Sometimes, models can be further enhanced to include tions (i.e. World 2 knowledge) still play a critical role.
relations connecting properties with an explicit functional
time dependence, ∀i = 1, . . . , n, xi = xi (t), for example, in
the form (known as canonic representation, or local state 5 THE EVALUATION OF QUALITY
transition in System Theory): OF KNOWLEDGE
dxi (t)
= fi (x1 (t), . . . , xn (t)) (1) Given the combination of subjectivity and objectivity so
dt
usually present in knowledge, it is not amazing that the
for time-continuous models, and evaluation of the quality of knowledge represents a basic
issue in the process of acquisition of candidate items for
xi (t + t) = xi (t) + fi (x1 (t), . . . , xn (t))t (2) their integration in an existing body of knowledge.
The quality of a model can only be evaluated in reference
for time-discrete models. Models can then be used also for to the goals for which the model itself has been produced:
prevision, in particular if the integral/time-global versions the general criterion for this quality evaluation is therefore
of the canonic representations is taken into account: the adequacy to goals. Truth, traditionally thought of as
t ‘correspondence to facts’, is regarded as a specific case
xi (t) = xi (t0 ) + fi (x1 (τ ), . . . , xn (τ )) dτ (3) of adequacy, applicable whenever correspondence to facts
t0 is indeed considered an important issue (note how this
n−1 position radically differs from the pragmatist definition of
xi (tn ) = xi (t0 ) + fi (x1 (tj ), . . . , xn (tj ))t (4) truth, according to which ‘a sentence may be taken as
j =0 a law of behavior in any environment containing certain
characteristics; it will be “true” if the behavior leads to
allowing to compute the system state x1 (t), . . . , xn (t) at a results that are satisfactory to the person concerned, and
generic (future or past) time t from a reference, initial state otherwise it will be “false”’ (Russell)).
x1 (t0 ), . . . , xn (t0 ) and by means of the state transitions While adequacy is hardly the object of general treatment,
f1 , . . . , fn . the possibility of evaluating the truth of a model has been
Finally, if an external intervention is possible on the widely debated and is surely one of the most critical topics
system, its spontaneous dynamics can be controlled to let of the Philosophy of Science. Following Karl Popper, it can
the system evolve toward a required target. In this case,
be suggested that the controversy is specifically related to
models are then aimed at prescription: given a generalized
two basic Problems:
version of the local state transition function including in
its domain both the current state and the user input, models 1. how to compare (the statements of) competing models?
specify how to provide such an input, and therefore become 2. how to evaluate (the truth of) a model?
decision-making tools (see Figure 5).
According to the traditional paradigm of science and in reference to which three standpoints can be identified:
its relations with technology, by repeatedly following this
• Verificationism: typical of classical science and brought
to its extreme consequences by the Neo-Positivistic
Description: what is. . . school: Problem 2 admits a solution (and therefore the
truth of a model can be determined), from which a
Prescription: what should. . . Explanation: why is. . . solution to Problem 1 is derived: the reference to truth
is the foundation allowing the advancement of science.
• Falsificationism: also called ‘critical rationalism’, as
Prevision: what will. . . advocated by Popper himself: Problem 1 admits a solu-
tion (in the presence of competing models, the one that
Figure 5. The knowledge loop among the four kinds of models. is chosen is the one that is not falsified and has the
Characteristics and Theory of Knowledge 131
greater empirical content), but a solution to Problem 2 This scheme highlights the complementary role of data
cannot be derived from it: by means of conjectures and and inference in knowledge (see also Article 10, Exten-
confutations, truth is approximated; the preference of sions of the Representational Theory of Measurement,
a model over a competing one can be rationally moti- Volume 1).
vated, but a model cannot be justified in itself. Data, that is, evaluated properties, are aimed at being
• Epistemic relativism, also called ‘irrationalism’, sup- a faithful representative of the observed state, as obtained
ported by philosophers such as Thomas Kuhn and, by either subjective or intersubjective and objective proce-
in its extreme consequences, Paul Feyerabend: Prob- dures, and can be expressed by means of either an informal
lem 2 does not admit a solution (‘the only principle or a formal language. The fundamental operation to empir-
that does not inhibit progress is anything goes. For ically get formal data by means of an intersubjective and
example, we may use hypotheses that contradict well- objective procedure is measurement: according to the repre-
confirmed theories and/or well-established experimen- sentational point of view to measurement theory (see also
tal results. We may advance science by proceeding Article 8, Formal Theory of Measurement, Volume 1),
counter-inductively’ (Feyerabend)), and therefore also such a faithfulness is formalized by requiring that the map-
Problem 1 cannot be solved: no criterion/method that is ping from empirical states to symbols be a homomorphism
absolutely valid holds in scientific research. for the scale type in which states are measured; the exis-
tence of monomorphisms for the measurement scale type
(i.e. admissible scale transformations) manifests the resid-
ual presence of conventionality in the selection of symbols.
6 DATA AND INFERENCE
It is in this regard that one could wonder about the truth of
IN KNOWLEDGE symbols and the related sentences.
Inference is an operation aimed at obtaining new data
We get an insight into knowledge by considering its oper- ‘conclusions’ from the processing of the given inputs
ational side of being a faculty to solve problems, and, in ‘premises’. To understand the structure of inferential pro-
particular, to modify the state of systems according to given cesses, the fundamental distinction between singular and
goals. As human beings, we constitutively have the abil- universal assertions (sometimes called facts and laws
ity to operate state transitions on the systems with which respectively) must be taken into account. In set-theoretical
we interact by means of a ‘World 2 strategy’: we acquire terms, a ∈ P (the element a belongs to the set P ; the prop-
data on the current state through our sensorial apparatus; by erty P holds the element a) is singular, whereas P ⊂ Q (P
means of the brain, we perform inference on such data, and is a subset of Q; for all elements x, if the property P holds
finally we use the data resulting from this process to drive for x then also the property Q holds for it) is universal (it
our motor apparatus whose activity actually carries out should be clear therefore that data obtained by means of
the required state transition (these three steps correspond measurement are singular).
to the tripartite structure of the neural system: sensorium, Two kinds of inference are then traditionally considered,
brain, and motorium; note that more than 99% of the about that in their simplest forms are as follows:
1010 neurons of human beings are part of the brain). In
many cases, this strategy is manifestly both more efficient • From the singular a ∈ P and the universal P ⊂ Q by
and more effective than a blind ‘try-and-error’ approach, deduction, the singular a ∈ Q is obtained; deduction is
although far more complex than it. a truth-preserving inference that, strictly speaking, does
The same conditions, the availability of data and the not lead to new knowledge.
ability to deal with them by means of inference, are also • From a collection of singular a ∈ P and a ∈ Q by
characteristic of the ‘World 3 strategy’ to problem solving, induction, the universal P ⊂ Q can be obtained; induc-
as shown in Figure 6. tion is a hypothetical inference that does not lead to
conclusive knowledge.
Initial state Final state
The problem of foundation of empirical knowledge is
Data acquisition traditionally ascribed to this circularity: deduction leads
and evaluation Actuation
to true conclusions, but only if the truth of its (both
Symbolic representation Result of inference: singular and universal) premises can be assumed; induc-
of the initial state: processed values
evaluated properties Data processing: inference tion is the only means to obtain new universal knowl-
edge, but the truth of such knowledge cannot be definitely
Figure 6. The ‘World 3 strategy’ to problem solving. assumed.
132 Foundations of Measuring
7 INEXACTNESS OF KNOWLEDGE AND not generally so specific to univocally denote (properties
MEASUREMENT of) things (2 + 2 = 4 holds for both apples and aircraft
carriers), at the same time, things are too complex to be
fully described by means of symbols.
We have already noted the relevance of language for
World 3 knowledge: truth is a property of sentences (actu-
ally: of declarative ones), and ‘science, though it seeks
traits of reality independent of language, can neither get 8 (NON)FOUNDATIONS OF KNOWLEDGE
on without language nor aspire to linguistic neutrality. To
some degree, nevertheless, the scientist can enhance objec- Philosophy of Knowledge (and Philosophy of Science, in
tivity and diminish the interference of language, by the very particular) has always quested for a foundation of knowl-
choice of language’ (Quine). That is why formalization (i.e. edge, that is, the elements on which the ‘building of knowl-
the expression of knowledge in a form such that inferential edge’ can be firmly erected. In the course of history, such
processes can be entirely performed on the basis of the syn- a foundation has been found in natural elements (for exam-
tactical component of data) is often regarded as a critical ple, Thales of Miletus affirmed that the principle that causes
requirement for scientific knowledge. On the other hand, all the things is water, while Heraclitus of Ephesus found it
formalized languages can be (and in many cases actually in fire), in physical or conceptual structures (atoms accord-
are) far too precise for expressing empirical knowledge: ing to Democritus of Abdera, numbers in the conception
‘there are certain human activities which apparently have of Pythagoric school), in metaphysical principles (such as
perfect sharpness. The realm of mathematics and of logic is the hypothesis that Nature is simple), in methodological
such a realm, par excellence. Here we have yes-no sharp- assumptions (in particular, the postulation that any empir-
ness. But (. . .) this yes-no sharpness is found only in the ical knowledge cannot be derived but from sense data).
realm of things we say, as distinguished from the realm of The usage of the metaphor of foundations is not conceptu-
things we do. (. . .) Nothing that happens in the laboratory ally neutral: the architectural image of ‘foundations’ reveals
corresponds to the statement that a given point is either on the hypothesis that scientific research can make knowledge
a given line or it is not’ (Bridgman). incrementally grow from its bases, where measurement has
Hence, the same empirical knowledge can be expressed been traditionally recognized as the operation that is able to
in sentences by balancing two basic components: certainty produce the objective data playing the role of such bases.
(a term for some aspects more general than truth) and preci- In the last decades, this confidence in the progressive
sion (also called specificity or, at the opposite, vagueness). development of scientific knowledge has been questioned
Therefore, ‘all knowledge is more or less uncertain and by concentric objections, all emphasizing that definitive
more or less vague. These are, in a sense, opposing char- foundations are beyond the reach of the means human
acters: vague knowledge has more likelihood of truth than beings adopt to know. Complementary to the above-
precise knowledge, but is less useful. One of the aims of sci- mentioned philosophical positions of epistemic relativism
ence is to increase precision without diminishing certainty’ (according to which raw sense data do not exist because
(Russell). data are always theory laden), an important area of scientific
The fact that the length of the diagonal of a physical research is currently devoted to the systems that exhibit rel-
√
1 m side square cannot be 2 m is an important conse- evant structural complexity, a characteristic that makes such
quence of metrological thinking: the information conveyed systems irreducible from the classical paradigm of reduction
by real numbers (and the related concepts of continu- to simplicity through the hypotheses of linearity, principle
ity/differentiability) is too specific to be applicable, as is, of superposition of effects, and so on. Knowledge is rec-
to physical systems. By progressively enhancing the resolu- ognized to be an always evolving process, where ‘there is
tion of the measuring systems, and therefore by increasing never an absolute beginning. We can never get back to the
the specificity of the measurement results, their uncertainty point where we can say, ‘Here is the very beginning of log-
consequently grows, until the object of measurement itself ical structures’ (Piaget). More than the actual availability of
becomes uncertain (in the previous example, at the atomic data, knowledge is recognized to be a potentiality (what is
scale the concept of ‘physical square’ is meaningless), and ‘stored’ in our brain is how to compute multiplications, not
an ‘intrinsic uncertainty’ (also called ‘model uncertainty’) the results of operations such as 1234 × 5678), information
is reached. is always under reconfiguration.
This reflects a basic feature of the relation that by means The role assigned to measurement is paradigmatic of the
of knowledge it is established between World 1 (to which shift toward what could be called reticular (and therefore
the object of knowledge belongs) and World 3 (to which the without foundations) knowledge. Indeed, according to the
sentence that expresses knowledge belongs): if symbols are current standpoints of philosophy of measurement:
Characteristics and Theory of Knowledge 133
• Since measurement results depend on standards through RELATED ARTICLES
a traceability chain, standards themselves could be
thought of as ‘realizations of true values,’ than playing Article 13, Signals, Information and Knowledge, and
the role of actual foundations for measurement; on the Meaning, Volume 1; Article 23, Principles of Semiotics
other hand, standards must be indeed ‘realized’ by pri- as Related to Measurement, Volume 1; Article 24, Prin-
mary laboratories, who maintain their quality by means ciples of Epistemology as Related to Measurement, Vol-
of interlaboratory comparisons: therefore, this claimed ume 1; Article 39, Characteristics of Data, Information,
‘path towards foundations’ cannot but include a com- Knowledge, and Wisdom, Volume 1; Article 63, Outline
ponent of conventionality. of Systems Thinking, Volume 2.
• Any measurement result depends for its evaluation on
the previous measurement of a set of influence quanti-
ties, in their turn being new measurands so that in prin- FURTHER READING
ciple such a dependence should be recursively applied,
with the consequence that a ‘well-founded measure- The scientific literature devoted to this topic is so wide that
ment’ would be impossible to be completed; the usual any selection of items is partial. What follows is a (short and
subjectively selected) list of general references to the issues
operative choice to assume that the quantities influenc- covered here.
ing the initial measurand are not influenced by other
quantities, and therefore, that they can be ‘directly mea- Hacking, I. (1983) Representing and Intervening, Cambridge Uni-
sured’, highlights the conceptual approximation inher- versity Press, Cambridge.
ent in any measurement. Morin, E. (1986) La connaissance de la connaissance, Vol. III of
• While the adequacy of empirical models is controlled La méthode, Seuil, Paris.
by means of measurement, the quality of measurement Piaget, J. (1970) Genetic Epistemology, Columbia University
results depends on the quality of mathematical models Press, New York.
used for designing measuring systems. Popper, K.R. (1994) Knowledge and the Body-Mind Problem – In
Defence of Interaction, Routledge, London.
This complexity makes knowledge the most versatile
tool available to human beings and a fascinating object for
knowledge itself.
Principles of Semiotics as Related to
23:
Measurement
Luca P. Mari
Università Cattaneo, Castellanza (VA), Italy
Instead of investigating here the general properties of the
1 Signs as Entities that ‘Stand for’ Something 134 relation S (but at least the observation should be made that
2 Coding and Decoding 135 in many cases functional substitutability is not a matter of
3 Conventionality of Signs 136 a yes–no alternative, and therefore that S could be usually
4 The Opposition Analog/Digital in a Semiotic formalized as a fuzzy relation: see also Article 12, Fuzzy
Perspective 136 Approaches for Measurement, Volume 1), let us devote
our attention to the specific relation of identification. The
5 The Systemic Nature of Signs 137
functional substitution implied in the identification is such
6 Syntax, Semantics, and Pragmatics 137 that x identifies a given y if and only if x operates as the
7 Semiotics and Communication 138 selector of y in a set of candidate things y1 , y2 , . . . the
8 Applying Semiotic Principles to Measurement exhibition of x being considered functionally equivalent
Systems 139 to the selection of y and the nonselection of any other
Related Articles 139 yi = y in the candidate set. In such a relation Sid (x, y)
Further Reading 139 let us call x and y the identifier and the identified entity
respectively: x stands for y. For example, the sound c-h-
a-i-r (a physical thing as a space-time event produced by
1 SIGNS AS ENTITIES THAT ‘STAND the utterance of a speaker), could be the identifier chosen to
select a chair instead of any other nonchair object (note that
FOR’ SOMETHING
no restrictions have been imposed on the set of the entities
Semiotics is commonly defined as the doctrine of signs, that are the object of identification: y can be a physical
a sign being ‘something which stands to somebody for object but also an informational entity).
something in some respect or capacity’ according to Charles It is a common observation that different x1 , x2 , . . . can
S. Peirce, one of the seminal thinkers about semiotics itself. be adopted as identifiers for the same entity y, Sid (x1 , y),
The emphasis is here on the relation of ‘standing for’, Sid (x2 , y), . . . (a chair can be identified by different sounds,
which, in its simplest form, can be modeled as follows. possibly pronounced by different persons in different lan-
When a purpose is assigned to, or recognized proper guages in different times, but also by writings, drawings,
of, things, they can be evaluated in their ability to satisfy gestures, . . .). In this case, the x1 , x2 , . . . are functionally
it. Any given purpose induces a relation of functional substitutable with each other in their role of identifiers
substitutability S on the set of considered things T such for y, and therefore a derived relation Sid−y holds among
that ∀x, y ∈ T , S(x, y) if and only if x is a substitute of them. Formally (we will continue to forget the fact that
y with respect to the purpose, that is, x is as able as y to also Sid−y could be fuzzy) Sid−y (x1 , x2 ) if and only if
satisfy the purpose itself. Sid (x1 , y) ∧ Sid (x2 , y).
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Principles of Semiotics as Related to Measurement 135
A fundamental step is taken when the class x̂ of all the Mediator
identifiers x for which the relation Sid−y holds is abstractly
considered as the identifying entity for y, thus recognizing
that signs, although instantiated in physical things, are
information entities (see also Article 22, Characteristics Identifier Identified entity
and Theory of Knowledge, Volume 1). Figure 3. The ‘semiotic triangle’.
relation is indirect and a ‘mediator’ is usually required to
2 CODING AND DECODING connect identifiers and identified entities.
In the history of Semiotics, diagrams of this kind have
Signs generally result from the relation between two ele- been widely adopted to present and generalize the relation
ments: identifiers and identified entities. Such a relation is that we have introduced as between identifiers and identified
operatively realized and performed in two phases: entities. For example, de Saussure defined it in terms of
• for a given entity y to be identified, an identifier ‘signifiers’ and ‘signifieds’ (and called ‘signification’ the
is obtained by means of an operation of coding: an relation itself), while Hjelmslev used the terms ‘expression’
information entity x̂ is at first associated with y, and and ‘content’ respectively.
then an identifier x is selected such that x ∈ x̂; for These diversities witness the different interpretations and
example, see Figure 1, emphasis put on the elements of the relation. For example,
• for a given identifier x, the identified entity is obtained the mediator has been thought of as either the sense of the
by means of an operation of decoding: an information identifier (chairs are identified by means of the term ‘chair’
entity x̂ is at first identified, by means of a pattern because of the meaning associated with such a term, that is,
recognition, as the class to which x belongs, and then the set of features that are shared by everything to which the
an entity y is selected as associated with x̂; for example, term applies, the so-called intension of the term), or the set
see Figure 2, (see also Article 33, Coding Theory and of the entities, the entities to what the identifier stands for
its Application to Measurement, Volume 1). (the so-called extension of the term), or the subject with the
competence to maintain the relation, or the social context of
Figures 1 and 2 are instances of a more general ‘semiotic the individuals who agreed to associate the identifier with
triangle’, in which the ‘standing for’ relation is represented the identified entity.
as in Figure 3: where the dotted line expresses that the In the case in which the entity to be identified belongs
to the physical world (let us mention again that it could
be a purely informational entity, as in the case when one
The word
‘chair’ is talking about words or numbers), a common, although
surely not necessary, situation is such that the relation
between the physical thing chosen as identifier and the
identified entity is mediated by two informational entities,
for example, (see Figure 4) so that each arrow in Figure 5
The sound c-h-a-i-r A chair represents a partial realization of the ‘standing for’ relation.
pronounced
by an individual Semiotics has been in particular working on the infor-
mational component of the ‘standing for’ relation, thereby
Figure 1. An exemplification of the structure of a coding often minimizing, or even neglecting, the analysis of the
operation.
relation between the symbols and the physical things
adopted as support for them. From now on we will accept
The word
‘chair’
The word The concept
‘chair’ of chair
The sound c-h-a-i-r A chair
pronounced The sound c-h-a-i-r A chair
by an individual pronounced
by an individual
Figure 2. An exemplification of the structure of a decoding
operation. Figure 4. The structure of the semiotic relations.
136 Foundations of Measuring
A symbol A concept easily, while often implicitly, shared among cultural
communities.
• Symbols (in specific sense), such as those constituting
many elements of textual languages, for which the rela-
An identifier An identified tion symbol-referent is purely conventional (obtained
object object sometimes by an explicit ruling convention, sometimes
by usage), so that it must be learned according to
Figure 5. The ‘standing for’ relations.
an extensive strategy, that is, by explicitly listing the
symbol-referent pairs.
this general standpoint, and follow the terminology pro-
posed by Ogden and Richards who describe the ‘standing
for’ relation in terms of symbols that stand for referents 4 THE OPPOSITION ANALOG/DIGITAL
(see also Article 8, Formal Theory of Measurement, Vol- IN A SEMIOTIC PERSPECTIVE
ume 1).
The latter two categories, icons and symbols, can be mean-
ingfully expressed in terms of the opposition between
3 CONVENTIONALITY OF SIGNS analog and digital, as traced back to the concept of struc-
ture modeled and formalized in Measurement Theory (see
The ‘standing for’ relation is a complex one: the same also Article 8, Formal Theory of Measurement, Vol-
symbol could stand for different referents, and different ume 1). The opposition A/D concerns the strategy adopted
symbols could stand for the same referent (e.g. in the case for coding and decoding the metainformation that comple-
of the linguistic phenomena of polysemy and synonymy ments the information that symbols convey on referents.
respectively). This suggests that such a relation is not Indeed, together with the information enabling the selec-
inherent to the entities involved in it: an entity becomes tion of referents, in many cases, some structural information
a symbol only as the result of a choice. must be maintained in coding and recognized in decoding.
The issue of the (degree of) arbitrariness of signs For example, if the referent is a grade in {A, . . . , E} the
has been thoroughly inquired by many philosophers, who observation of a physical support coding the symbol ‘B’
noticed its fundamental implications in terms of autonomy should bring both the information ‘is B’ (and ‘is not A,
of symbols in relation to referents and therefore, generally and not C, . . .’) and the (ordinal) metainformation ‘is less
speaking, of language in relation to reality. For example, than A but more than C, . . .’. Therefore:
in Plato’s Cratylus, the problem of ‘right names’ for things
• analog is the strategy by which the metainformation
is discussed, and it is concluded that ‘whatever name you
is coded in the support configuration, so that both
give to a thing is its right name; and if you give up that
coding and decoding correspond to the application of
name and change it for another, the later name is no less
a homomorphic (i.e. structure preserving) mapping;
correct than the earlier, just as we change the name of our
• digital is the strategy by which the metainformation
servants; for I think no name belongs to a particular thing
is maintained in the coding rule, so that the physical
by nature’.
support is only required to be able to assume at least
Following Peirce, signs are usually distinguished in three
two distinguishable configurations (those usually sym-
broad categories, characterized by their increasing degree
bolized as ‘0’ and ‘1’), as specified in Shannon’s Theory
in conventionality:
of Information.
• Indexes, such as ‘natural signs’ (smoke standing for a This characterization accounts for the nature of opposition
yet unseen fire) and ‘signals’ (a phone ringing standing of the strategies of analog and digital coding:
for a waiting call), for which the symbol is causally
connected to the referent, so that every subject informed • The definition of the code rule can be intensive in
on the connection is able to infer the existence of analog cases, whereas it must be extensive in digital
the relation. cases: while analog codes can be analytically defined,
• Icons, such as images or onomatopoeic words, for the lack of structure forces digital codes to be defined
which the relation symbol-referent is based on some by explicitly and completely listing the occurrences
mutual resemblance or imitation of the related entities; symbol-referent.
in this case, the relation can be intensively learned (e.g. • The set of information entities to code can be incom-
the higher the sound the angrier the speaker) and is pletely preidentified in analog cases, whereas it must be
Principles of Semiotics as Related to Measurement 137
preidentified in digital cases: for example, to increase Dialectical is also the process by which the decoding of
the cardinality of the set of the possible symbols to code composite structures of symbols, for example, sentences,
on a physical support, an analog code can be adopted is performed: paradigmatically, whenever coding rules are
as is, whereas a digital code must be redefined. context-sensitive (a typical characteristic of natural lan-
guages), not only can the meaning of a sentence be derived
On the other hand; from the meaning of its constituting parts, but the meaning
• analog coding can be adopted only if some metainfor- of such parts could also depend on their role in the sentence,
mation is available, whereas digital coding is always so that it can be determined only after some hypothesis
available for finite sets of symbols; of the meaning of the whole sentence itself. This gener-
• supports adopted in coding must be able to maintain ally makes the recognition of the ‘standing for’ relations
the metainformation in physically distinguishable con- a complex, recursive process, and again this explains why
figurations in analog cases, whereas it can be very the constructs of artificial languages are defined as context-
simple since only two distinguishable configurations are free whenever recognized as adequate (as an example of
in principle required in digital cases. the role of context in rule evaluation, consider two possible
definitions of the disjunction operator OR: if v(x) is the
It should be noted that mixed (partly analog, partly truth value of the sentence x, in classical logic, such an
digital) coding rules are common, as in the case of the usual operator is context-free, since v(x ∨ y) = f (v(x), v(y)) =
numerical notation, in which the single digits are digitally max(v(x), v(y)); on the other hand, in the case of proba-
coded while the positional rule is analog. bilistic logic v(x ∨ y) = v(x) + v(y) − v(x ∧ y) and there-
fore the operator is context-sensitive, because v(x ∨ y) =
f (v(x), v(y)), with the term v(x ∧ y) playing the role
5 THE SYSTEMIC NATURE OF SIGNS of context).
The ‘standing for’ relations are seldom defined as single
and independent coding rules. Rather, their conventionality 6 SYNTAX, SEMANTICS, AND
is considerably restricted by the effects derived by their sys- PRAGMATICS
tematic definition (as an example, consider the possibility
of reconstructing the meaning of a linguistic term, that is, A fundamental classification to isolate the different contri-
‘decoding’ it, by means of its etymological analysis). butions to the complexity of the ‘standing for’ relation was
Natural languages, such as English or Italian, are far more proposed by Charles W. Morris, who suggested three basic
complex than artificial languages, such as the formalism of subdisciplines as the constituting components of Semi-
mathematical logic or computer programming languages, otics: syntactics (also, and more commonly, called syntax ),
also because they include a huge amount of exceptions, that semantics, and pragmatics. Despite their large reciprocal
is, irregularities, in their coding and decoding rules. Never- autonomy, such disciplines can be presented in terms of
theless, the fact that some systematic effects progressively progressive extension of scope:
emerge from historical usage instead of explicit decision,
as indeed in the case of natural languages, does not reduce • Syntactic is the information dealt with as data, taking
their relevance but only the uniformity of the system of into account the collection of available signs and its
rules. structure; a basic issue of syntax is parsing, that is, the
This systemic component was called language (langue, check that a sentence is well formed according to a
in French) by Saussure, who contrasted it with speech given set of syntactical rules.
(parole, in French), regarded as the individual act of selec- • Semantic is the information dealt with as data provided
tion and actualization of symbols that stand for intended with meaning, taking into account (also) the entities to
referents by means of some coding rules of the lan- what the signs stand for; a basic issue of semantics is
guage. Any specific film would be therefore the ‘speech’ truth evaluation, that is, the check of the correspondence
of the ‘language’ of cinema, an example highlighting that between the content of a sentence and the actual state
langue/parole is actually the dichotomy code/instance or of the reality (it should be noted that the evaluation
schema/usage, as Hjelmslev termed it. ‘Each of these two of the truth of a sentence does not always require
terms achieves its full definition only in the dialectical pro- the interpretation, that is, the ‘semantization’ of the
cess which unites one to the other: there is no language sentence itself; for example, tautologies in propositional
without speech, and no speech outside language: it is in logic (e.g. A ∨ ¬A) are true for any interpretation of
this exchange that the real linguistic praxis is situated’ A; this establishes a distinction between linguistic and
(Barthes). empirical truth).
138 Foundations of Measuring
• Pragmatic is the information dealt with as data provided message is the same as the one generated by the sender
with meaning and value, taking into account (also) the even in the presence of a noisy channel – specifically relates
relation of signs to interpreters and therefore issues to syntax. The general semiotic nature of the problem has
related to behaviors, subjective interests, utilities, . . .; a been shown by Roman Jakobson, who suggested that each
basic issue of pragmatics is relevance assessment, that of the six components of a communication system shown in
is, the check that a sentence is actually useful for its Figure 7: is associated with a specific function, more or less
deemed receivers. present and emphasized in each communication act. Indeed,
if the prominence is usually given to the referent (i.e. the
The distinction among these disciplines and their goals can
‘content’ of the message, corresponding to the so-called
be exemplified by means of that particular system of signs
referential, or cognitive, function), in many messages, the
that is mathematics:
role of the other functions must also be taken into account
√
• The formula ‘2) (+ ≤)x’ is not well formed, so that a to fully understand the communication act itself. According
fortiori neither its meaning and truth nor its utility for to Jakobson, a communication can differently stress:
a given subject can be evaluated.
• The formula ‘2 + 3 = 4’ is well formed, its meaning • the position of the sender (emotive function) on the
can be evaluated, and it is actually false in the usual communication subject (e.g. in terms of rage or irony);
interpretation of its constituting signs. • the orientation toward the receiver (conative function),
• The formula ‘1 = 1’ is well formed and it is true, but as typically in the case of imperative sentences, whose
plausibly useless for most subjects. aim is indeed to convey commands to receivers;
• the role of the channel (phatic function), whenever a
Given the centrality of the ‘standing for’ relation, seman- support to the management of the communication itself
tics can be considered the core component of semiotics. is required, for example, to check whether the channel
Nevertheless, the threshold between syntax and semantics is still operative between the sender and the receiver
is not always well defined, and often actually a matter (e.g. ‘are you still there’?);
of ‘point of view’. For example, in the case of Morse • the formal structure of the message itself (poetic func-
code, the physical support is an electric current, whose pat- tion), for example, when homophonies, rhymes, . . .
terns are interpreted as sequences of ‘dots’ and ‘dashes’,
are adopted;
whose patterns are in their turn interpreted as sequences of
• the requirements on the code (metalinguistic function),
alphanumeric characters; the sequence ‘dot-dash’ is then a
typically whenever the sender and the receiver want to
semantic entity with respect to an electric signal but it is a
check whether they are using the same coding rules (e.g.
syntactical entity with respect to its deemed interpretation,
‘what do you mean’?).
that is, the character ‘a’.
The semantic component is critical in communications
with a prevailing referential function, and the problem of
7 SEMIOTICS AND COMMUNICATION
the truth evaluation of their contents can be generally posed.
The other functions are instead oriented to the pragmatics of
An important area of application of Semiotics is commu-
the communication: messages such as ‘fantastic’, or ‘excuse
nication, that is, the transfer of messages conveying some
me’, or ‘repeat please’, or . . . are conveyed to obtain some
sort of information from a sender to a receiver through a
communicational aim more than to state a meaning.
channel, (see Figure 6). The standard model for a basic
communication system has been defined by Shannon: the
message generated by the sender is coded into a pattern of Referent
signs, here called a signal, that is transmitted by the chan- Message
nel and finally decoded again into a message for making it Sender Receiver
Channel
accessible to the receiver. Code
As formalized by Shannon, the communication prob-
lem – how to maximize the probability that the received Figure 7. The components of a communication system.
Message Signal Message
Sender Coder Decoder Receiver
Channel
Figure 6. The structure of a communication system.
Principles of Semiotics as Related to Measurement 139
8 APPLYING SEMIOTIC PRINCIPLES TO set. The degree of semantic information conveyed by
MEASUREMENT SYSTEMS a measurement result depends thus on the degree of
richness in algebraic structure of its scale type (for-
As an operation aimed at expressing in symbols the infor- mally, the semantic information increases as the class of
mation empirically obtained on a system about a quantity, admissible transformations for the scale type becomes
measurement can be meaningfully analyzed in its semi- more specific).
otic character. In comparison with other forms of judg- • From the pragmatic point of view : because of the exis-
ment, there are two general grounds of peculiarity for tence of functional relations connecting them (the typi-
measurement: cal case of physical laws), measurands are embedded in
a network of pragmatic information allowing new mea-
• the mediator between the referent (i.e. the measurand) surement results to be obtained by computation, that is,
and the symbol (i.e. the measurement result) is an by derived measurement. While a completely discon-
empirical entity, external to both the measured thing nected measurand can be defined in a totally arbitrary
and the measurer subject: the measurement system; way, and therefore its evaluation is pragmatically use-
• the symbols adopted as identifiers for the measurands less, the more the measurand is connected (e.g. the
are chosen in a formal language, whose composition greater the number of functional relations in which it
and structure are explicitly known. is present), the higher is the degree of pragmatic infor-
mation conveyed by its values.
While for a general semiotic system only the syntac-
tical component can be usually formalized (the attempt Even this summary presentation shows how the semi-
was made by logics and philosophers of science such as otic perspective can be useful to understand some funda-
Bar-Hillel, Carnap, and Hintikka to quantify the seman- mental characteristics of measurement (uncertainty, scale
tic information conveyed by propositions as their ‘quantity types, and derived measurement) in a general concep-
of content’: with such a broad connotation, the problem tual framework.
remained largely unsolved and was substantially left aside),
these characteristics of measurement allow to consider it in
some more specific terms:
RELATED ARTICLES
• From the syntactical point of view : measurement can
be thought of as an operation of selection of a sym- Article 22, Characteristics and Theory of Knowledge,
bol from a set, the actual granularity of such a set (as Volume 1; Article 24, Principles of Epistemology as
usually formalized in terms of either number of signif- Related to Measurement, Volume 1; Article 25, Intro-
icant digits or expanded uncertainty) depending on the duction to Signals in Physical Systems, Volume 1; Arti-
resolution of the sensing device; the usual Shannon’s cle 63, Outline of Systems Thinking, Volume 2.
concept of quantity of information can be adopted in
this case, such that the quantity of information con-
veyed by a measurement result increases as its uncer- FURTHER READING
tainty decreases.
• From the semantic point of view : measurands are always Chandler, D. (2001) Semiotics: The Basics, Routledge, London.
evaluated relative to a reference, that is, explicitly Devlin, K. (1991) Logic and Information, Cambridge University
reported in measurement results in terms of a measure- Press, Cambridge.
ment scale (and specifically measurement unit whenever Eco, U. (1976) A Theory of Semiotics, Indiana University Press,
applicable) and that expresses the actual meaning for the Bloomington, IL.
(usually) numerical symbols by which the measurand is Kyburg, H.E. (1984) Theory and Measurement, Cambridge Uni-
quantified. Each measurement scale is characterized by versity Press, Cambridge.
a type, the most common scale types (nominal, ordinal, Mari, L. (1999) Notes Towards a Qualitative Analysis of Infor-
interval, ratio, absolute) being linearly ordered accord- mation in Measurement Results. Measurement, 25(3),
ing to the algebraic structure they imply on the symbol 183–192.
Principles of Epistemology as Related to
24:
Measurement
Timothy Lindsay John Ferris
University of South Australia, Mawson Lakes, South Australia, Australia
person and experience of the person who has the idea,
1 Definition of Epistemology 140 whereas knowledge of the idea has the self-consciousness
2 Explanation of Epistemology 140 of enabling the knower to be aware of what the idea is,
and of the fact that they know the idea. Thus, knowledge
3 Relevance of Epistemology to Measurement 141
is always a representation of what is known and is never
4 Applying Epistemological Principles to what is known in and of itself.
Measurement Systems 142 The relation of things and their representation in knowl-
5 Examples of the Use of Epistemology in edge leads us to the central problem of epistemology,
Measurement Systems 142 expressed by the question: ‘How do I know that reality
References 143 is what I conceive and assert it to be in cognitive expe-
rience?’ (Russell, 1898). Epistemology aims at identifying
the relation between matters in reality and the subject’s con-
sciousness of these matters (Perry, 1931), and so it explores
1 DEFINITION OF EPISTEMOLOGY the problem of how judgments joining various pieces of
information may be made. Epistemology, in relation to mea-
Epistemology is the branch of philosophy that is concerned surement, is particularly concerned with the issue of how
with the study of how knowledge is possible. The basic empirical information can be used to form knowledge, that
problem of epistemology is that, in common language, is, how it is possible to make synthetic, as distinct from
people make claims to know certain things, and some basis analytic, judgments (Rescher, 1980). Kant distinguished
for determination of whether the claims are reasonable or analytic judgments, judgments relying only on pure reason
appropriate is necessary in order to ensure that any action and a priori information with synthetic judgments, which
taken consequent upon the belief that one knows the matter rely on a posteriori, empirical information.
is appropriately founded. Epistemology develops theories of truth, evidence (Hall,
The analysis of knowledge requires distinction between 1943), meaning, and verification (Carnap, 1936) in order to
that which is known and the knowledge of it. That which clarify each of these concepts and to make clear when it is
is known is either a tangible thing or phenomenon that is reasonable to trust information that one has received, and
observable in nature, or an idea that belongs as part of when not to trust information.
an intellectual construction. The knowledge of the matter
is distinct from the matter in that it is a representation
of that which is known without actually being that which 2 EXPLANATION OF EPISTEMOLOGY
is known. In the case of the matter being an idea, there
is still a distinction between the idea and knowledge of The classical account of knowledge states that S knows that
the idea in that the idea is an existential part of the p if and only if
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Principles of Epistemology as Related to Measurement 141
1. p is true; 1. ‘Measurement is the process of assigning numbers to
2. S believes that p; and represent qualities: the object of measurement is to
3. S is justified in believing that p (Gettier, 1963). enable the powerful weapon of mathematical analysis
to be applied to the subject matter of science’ (Camp-
The classical account of knowledge has three funda-
bell, 1920).
mental elements: truth, belief, and justification, and the
2. ‘Measuring in a narrow sense (measurement) is the
only serious challenge to this understanding has been that
experimental comparison of a measured quantity with
which is raised in Gettier (1963). The classical account of
a metrological standard. Measuring in a wide sense
knowledge is always satisfied in the case of matters known
(classification) is the experimental comparison of a
analytically. The classical account was developed to deter-
measured object with a particular standard (pattern)’
mine when matters known synthetically could be asserted
(Hofmann, 1983).
as known.
3. Measurement is ‘the quantitative determination of a
A theory of truth develops an understanding of what it
physical magnitude by comparison with a fixed mag-
means to assert ‘p is true’. The assertion that p is true must
nitude adopted as the standard, or by means of a
be explained so that it is clear. In brief, ‘true’ means that
calibrated instrument’ (Sydenham, 1976).
something is as it is represented to be in the statement that
4. ‘Measurement is the assignment of numbers or other
is described as ‘true’. The assertion that p is true is thus an
symbols by an objective, empirical process to attributes
assertion of correspondence between what is the case of the
of objects or events of the real world in such a way as
matter and how the matter is represented in the information
to describe them’ (Finkelstein, 1990).
that is the subject of the assertion of truth.
Knowledge ‘that p’ requires that the subject, S, believes Each of these definitions of measurement emphasizes
that the matter corresponds to the representation embodied that measurement is a process of observation of the real
in p. ‘Believe’ is a verb with several distinct uses, including world that results in representation of the observed in the
‘believe that p’, which pertains to the mental assent that structured framework of the scale used for the investigation.
p is true. Mental assent does not require commitment to The intention of measurement is to convert observation
the truth of that which is believed, but it does require of the world into a form amenable to the application of
that the believer intellectually accepts that the assertion the analytical processes of mathematics. The intention of
believed is in fact true and that there is no confusion or conversion from synthetic knowledge to information that
error introduced by any factor such as may introduce any can be operated upon in an analytic manner is a major leap
distortion of perception of the observed. that demands the analysis of epistemology. The reason is
The requirement for justification in the classical account that the results obtained from the analytic processes obtain
of knowledge demands a theory of evidence as the basis for a status, in ordinary thought, linked to their derivation by
determining those signs that are something other than the an analytic process, rather than the uncertainty associated
signs themselves and that are necessary to determine that with immediate recognition of the empirical nature of the
which is signified as true. Evidence may be directly related observation process.
to the matter and explicit evidence of something, or it may Epistemology emphasizes the limitations that are placed
only indicate what is claimed, or it may be circumstantial, on claims to knowledge by the method and circumstances
resulting in the matter being concluded as probable. of the gathering of the information concerning the mat-
The physical sciences assume that there is a physi- ter. Thus, epistemology encourages skepticism concerning
cal world, which is fundamentally and substantially as it whether what appears to be the case is actually the case.
appears to be. On some occasions, what is seen is not In the measurement situation, this subject matter of episte-
real, but the cause of this difference is distorting circum- mology concerns the development of understanding of the
stances that result in seeing something different than what is possible sources and kinds of erroneous measures that may
present. The physical sciences assume that such distorting be obtained. This interest is at the level of verification of
observational circumstances can be explained through the the data in the measurement system.
physical phenomena of the world, and that such illusions Epistemology is also of relevance in the theory of mea-
can be overcome. surement in developing awareness that the ultimate purpose
of measurement is to obtain understanding of the things that
3 RELEVANCE OF EPISTEMOLOGY TO are observed. This awareness prompts the understanding
MEASUREMENT that measurement concerns the development of knowledge
about the observed and forms the basis for wise action
There are many definitions of measurement in the lite- based, in part, on knowledge of the observed, and on
rature: other knowledge.
142 Foundations of Measuring
4 APPLYING EPISTEMOLOGICAL manifestation is performed with the instrument, and results
PRINCIPLES TO MEASUREMENT are recorded, and used to provide a conversion between
instrument output and true value. Calibration can remove
SYSTEMS inaccuracy related to the instrument characteristics, but
cannot remove inaccuracy arising from noise or influence
Measurement systems apply instruments to systems that factors or loading effects different than those at the time of
are considered interesting. This situation produces an inter- calibration.
face between the observed and the instrument, which is Awareness of epistemology has led to investigation of
an information boundary across which the measurement
the issue of accuracy, and the development of the concept
information must flow. The development of measurement
of calibration, to address the epistemological problems of
processes has respected the skepticism concerning the qual-
illusion and hallucination. Nicholas (1994) has a discussion
ity of information as a basis for knowledge presented
of several specific examples of these effects.
in epistemology. Traditional concerns in the development
of measurement process theory have included the issues
of accuracy and calibration. The ideas of both accu-
racy and calibration address different issues associated 5 EXAMPLES OF THE USE OF
with the perspective on measurement provided through EPISTEMOLOGY IN MEASUREMENT
epistemology. SYSTEMS
Accuracy: The definitions of measurement of Syden-
ham (1976) and Hofmann (1983), Section 3, above, involve Section 4 discussed the application of epistemology to
comparison of the observed manifestation with some kind measurement systems at the data level. In this section, the
of scale or standard. If a measurement were entirely accu- application of epistemology to measurement systems at the
rate, there would be an exact correlation between the real levels of knowledge and wisdom is discussed.
manifestation and the measure of it obtained. The discus- Sacerdoti, Paggi and Sacerdoti (1991) describe measure-
sion of accuracy is premised on the idea that there is a ment processes as ‘the means by which man dialogues with
‘correct value’ of the manifestation in the observed, and the external world in order to enrich his knowledge. The
that that correct value is present to be obtained, should one processes pass through dialectic counterreacted phases, that
have sufficiently good means to measure it. It is not pos- is, ‘historical phase’, during which the ‘receiver’ learns to
sible to prove that there is such a thing as one true value recognize forms and situations; organization of the experi-
of things. mental situations on the basis of results; representation of
the results’. This description of measurement expresses the
Inaccuracy: The assumption that there is a true value idea that measurement concerns the gaining of knowledge
arises from a variety of effects including influence effects, about the observed.
noise effects, instrument imperfections such as hysteresis or Sacerdoti, Paggi and Sacerdoti (1991) provides a frame
wear, error in implementation of the physical effect of the leading to the Object Oriented Model of Measurement
sensor system, and loading of the observed by the sensor (OOMM) of Ferris (1997). The OOMM provides a struc-
system. Influence effects result from the sensitivity of the ture for the analysis of measurement involving a class of
sensor to effects other than the target effect. Noise relates entity, which is characterized as having relations into which
to the presence of random events in the observed and sen- it enters, attributes that it manifests, an attribute observation
sor system affecting the output. Instrument imperfections process, and the belief that repeatable and verifiable mea-
result in the instrument systematically producing erroneous surements are possible. The model of the class then controls
results, either in a simple relationship to true value or in the kind of measurements that are applied to the observed
a relationship dependent on the history of the instrument. and the interpretation of the results, leading to the creation
Error in the implementation of the system results in all out- of knowledge. The major problem is that nothing in the
put being misleading because it is a measure of something structure enables one to determine that the observed is not
other than what was intended. Loading of the observed by a member of the class to which it is assumed to belong.
the sensor is a particular class of poor design, in which the An example of the application of epistemology to mea-
combined system of the sensor and the observed results in a surement at the level of knowledge and wisdom is found in
different manifestation being present than would be present the testing and evaluation of complex products and sys-
if the instrument were removed. tems. The design and development of product requires
Calibration is the simplest approach to the accuracy measurement of the prototypes of the product by which it
problem. A set of measurements of known values of the is determined whether the product can perform its function.
Principles of Epistemology as Related to Measurement 143
Measures of effectiveness REFERENCES
Increasing Campbell, N.R. (1920) Physics The Elements, Cambridge Univer-
Increasing definition sity, Cambridge.
uncertainty of physical
of top level variables Carnap, R. (1936) Testability and Meaning. Philosophy of Sci-
result Measures of performance ence, 3(4), 419–471.
Crouch, V. and Sydenham, P. (1994) Relationship between T&E
and M&I, 3rd Australasian Instrumentation and Measurement
System performance parameters
Conference, 26–29 April, Adelaide, Institution of Engineers
Australia, Barton (pp. 1–6).
Technical performance parameters Ferris, T.L.J. (1997) The Concept of Leap in Measurement Inter-
pretation. Measurement Journal of the International Measure-
ment Confederation, 21(4), 137–146.
Figure 1. The pyramid of measures involved in the measurement
of a product or system intended to perform some function. Finkelstein, L. (1990) Measurement and Instrumentation as a Dis-
(Adapted from Crouch, V. and Sydenham, P. (1994) Relationship cipline in the Framework of Information Technology, Knowl-
between T&E and M&I, 3rd Australasian Instrumentation and edge Based Measurement – Application, Research and Educa-
Measurement Conference, 26–29 April, Adelaide, Institution of tion, IMEKO Secretariat, Karlsruhe (pp. 257–263).
Engineers Australia, Barton (pp. 1–6).) Gettier, E.L. (1963) Is Justified True Belief Knowledge? Analysis,
23, 121–123.
Hall, E.W. (1943) Perception as Fact and as Knowledge. The
Philosophical Review, 52(5), 468–489.
The test and evaluation measurement scenario is expressed Hofmann, D. (1983) Automatic Testing with Intelligent Sen-
in Figure 1. sor Systems Measuring or Classification? Technological and
In Figure 1, the technical performance measures refer to Methodological Advances in Measurement Acta IMEKO 1982,
North Holland, Amsterdam (pp. 545–551).
the measures of physical quantities of fundamental kinds of
manifestation such as length, mass, temperature, and time, Nicholas, J.V. (1994) Measurement Illusions, 3rd Australa-
sian Instrumentation and Measurement Conference, 26–29
expressed in terms of standard units. The system perfor- April, Adelaide, Institution of Engineers Australia, Barton
mance parameters describe system performance in terms (pp. 75–80).
of derived units, which combine basis units, and describe Perry, C.M. (1931) Epistemology Re-Examined. The Philosophi-
aspects of the performance of the whole systems, such as cal Review, 40(5), 444–458.
velocity, or power consumption. Measures of performance Rescher, N. (1980) Induction, Basil Blackwell, Oxford.
concern the performance of the system in performing its
Russell, J.E. (1898) Epistemology and Mental States. The Philo-
design tasks, and include measures such as reliability, and sophical Review, 7(4), 394–396.
maintainability. Measures of effectiveness concern whether
Sacerdoti, G., Paggi, R. and Sacerdoti, D. (1991) Measurements
the system satisfies the need for which it was conceived and Representation, IMEKO XII Measurement and Progress,
and developed. Vol 1, IMEKO Secretariat, Beijing (pp. I-159–I-160).
The pyramid of Figure 1 shows a progression from the Sydenham, P.H. (1976) Introductory Address, Proceedings of the
lowest level of data, through information, knowledge to Regional Colloquium on the Nature and Scope of Measurement
wisdom, at each higher level in the pyramid. Science, IMEKO Secretariat, Budapest (pp. 17–29).
25: Introduction to Signals in Physical Systems
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
2 TASK OF MEASUREMENT AND
1 General Remarks on Signals 147 MATHEMATICAL DESCRIPTION
2 Task of Measurement and Mathematical
Description 147 As represented in Figure 1, the task of measurement
3 Two Fundamental Problems: Signal and technology is to establish the characteristics of actually
System Identification 148 occurring input signals, xr , from known (measured) out-
put signals yr . Signals having several components, xr , yr ,
Further Reading 148
for example, can be summarized to form signal vectors x,
y, z, in the same way as the disturbances zr . The output
quantities are a function of the input quantities, and the
relationship between them is given by the behavior of the
1 GENERAL REMARKS ON SIGNALS measuring system.
The aim is that the behavior should be of a certain
type, which is represented by an ideal mathematical oper-
Within measuring systems, communications take place ation Oid :
between the various parts. These communications take
yid = Oid {x} (2a)
place on signal carriers; today, these are predominan-
tly electrical, but other energy formats are sometimes In the individual case, this operation may be a constant with
used. The theory given here can be applied to them only one dimension, to give an example, but it may also be
also. a differentiation or integration, as for instance, in the case
In addition to the useful signal x, interfering signals of measuring devices used for averaging.
z occur (Figure 1). Both types of signals are carriers The practical measuring system with the connection Oreal
of information: wanted information in the case of between its output and input quantities also includes real,
the useful signal, and unwanted in the case of the falsified output quantities that depend on the disturbances z
interfering signal. Both signals can be treated the same
way. yreal = Oreal {x; y} (2b)
In a signal x(ar ), for the purpose of information transmis-
sion as found in measuring systems, it must be possible to Hence, an error ε occurs given by
change these information parameters. Information param-
ε = yid − yreal (2c)
eters are those parameters of the signal upon which the
behavior of the information to be transmitted is mapped. These relationships will be dealt with in more detail in later
Thus: articles (see Article 19, Dynamic Behavior of Closed-
loop Systems, Volume 1; Article 27, Signals in the
Transmitted information I = Fnc. signal x(ar ) (1) Frequency Domain, Volume 1; Article 28, Signals in the
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
148 Foundations of Measuring
Input signal System Output signal z
x Op y
x Measuring y
system
Disturbance (a)
z
z
Figure 1. Task of measurement.
x y
Time Domain, Volume 1; and Article 29, Relationship System
Between Signals in the Time and Frequency Domain,
Volume 1) of this topic, in particular, those with zm
time-varying signals. The discussion has the following
four objectives:
Measuring
1. description of the signals by their characteristic values system
and functions;
2. description of measuring systems by means of charac- ym
teristic values and functions; (b)
3. description of the errors and deduction of quality cri-
teria; Figure 2. (a) Identification of signals and (b) identification of
systems.
4. means for optimizing the system, that is, for minimiz-
ing errors.
System identification. Figure 2(b), on the other hand, dep-
Objective 1. is discussed in Article 27, Signals in the icts measurement of the parameters of a system. In that
Frequency Domain, Volume 1; Article 28, Signals in case, input test signals x and associated output signals y
the Time Domain, Volume 1; Article 29, Relationship are both fed to the measuring system. The parameters that
Between Signals in the Time and Frequency Domain, are characteristic of the system have to be determined from
Volume 1; Article 30, Statistical Signal Representations, the output quantities of this measuring system ym .
Volume 1; Article 31, Discrete Signal Theory, Volume 1;
Article 32, Geometrical Signal Representations, Vol-
ume 1, the other objectives, which are based upon the FURTHER READING
first, are dealt with in Article 31, Discrete Signal The-
ory, Volume 1; Article 32, Geometrical Signal Repre- Many texts have been published on this topic of linear systems
and circuit theory. Searches for relevant books with ‘signal’ as
sentations, Volume 1; Article 35, Systems in the Time a search word tend to be in the minority now, being largely
Domain, Volume 1; Article 36, Systems in the Frequency overtaken by those on digital signals and systems – which are
Domain, Volume 1; Article 37, Relationship Between not the same. Digital systems are, in fact, partially based on the
Systems in the Time and Frequency Domain, Volume 1 application of this linear systems theory. A selection on linear
and Article 38, Stability Issues, Volume 1. Reference is signals and systems is as follows:
made in each case to descriptions in the time and fre-
Buck, J.R., Daniel, M.M. and Singer, A.C. (2001) Computer
quency domains. Explorations in Signals and Systems Using MATLAB, 2nd edn,
Prentice Hall, NJ.
3 TWO FUNDAMENTAL PROBLEMS: Hsu, H.P. and Hse, H.P. (1995) Schaum’s Outline of Signals and
Systems, McGraw–Hill Trade.
SIGNAL AND SYSTEM
Karu, Z.Z. (1995) Signals and Systems Made Ridiculously Simple,
IDENTIFICATION ZiZi Press, Huntsville, AL.
Oppenheim, A.V., Willsky, A.S. and Hamid, S. (1996) Signals
Figure 2 shows the objectives that have been outlined and Systems, Prentice Hall, NJ.
above and that are tailored especially for the following two
Woschni, E.-G. (1988) Informations Technik, Hüthig, Heidelberg.
fundamental problems of measurement technology.
Woschni, E.-G. (1994) Signal Theory in Measurement and Instru-
Signal identification. In Figure 2(a), the aim is to determine mentation, Concise Encyclopedia of Measurement & Instrumen-
the signal parameters of the input signals in terms of known tation, Pergamon Press, Oxford.
(measured) output signals y, where, if possible, disturbances Zadeh, L.A. and Desoer, C.A. (1963) Linear System Theory,
should have no influence. McGraw–Hill, New York.
26: Signal Classification
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
nondetermined signals. If they are described by a prob-
1 Signals with and without Quantization Over ability distribution, they may also be called stochastic
Time 149 signals.
2 Table of Signal Classifications 149
3 Signal-flow Graphs 149
References 151 2 TABLE OF SIGNAL CLASSIFICATIONS
Further Reading 151
Table 1 gives a survey of signal designations as described
in detail in Section 1 (Woschni, 1988; Woschni and Krauss,
1976).
1 SIGNALS WITH AND WITHOUT
QUANTIZATION OVER TIME
3 SIGNAL-FLOW GRAPHS
A distinction is made between analog signals, which are
signals having no quantization of the information parame- To trace and represent the flow of the signal, signal-flow
ter, and discrete signals in which, because of quantization, graphs are used. Figure 1 gives a survey of the graphical
the information can assume a finite number of values. An
important special case of discrete signals is that involving
binary signals in which the information parameter can take x2 x3
only two discrete values: 0 or 1.
x1 x1 x4
In time-dependent signals – most signals are time depen-
dent, or are converted to become time-dependent signals −
x3 x2
by scanning, as is the case in television – the information
parameter can either change at any time (continuous sig-
nals), or changes are possible at given cycle times only, x1
due to time quantization. x1
If the entire behavior of the signal, including future
x3 x3
behavior, is known, the signal is said to be a deter- ∏
mined signal. The transmission or measurement of this
x2
signal, naturally, does not produce any gain in informa-
x2
tion. This type of signal plays a major role as test signals
(e.g. impulse function, step function). Contrary to this, Figure 1. (a) Branching: x2 = x3 = x1 ; (b) adding or subtrac-
a signal to be measured has little a priori information. tion: x4 = x1 + x3 − x2 (c) multiplication: x3 = x1 × x2 ; (d) divi-
Signals with unknown characteristics of flow are called sion: x3 = x1 /x2 .
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
150 Foundations of Measuring
Table 1. Classification of signals.
Characteristic Without quantization of time (continuous) With quantization of time (discontinuous)
Continuous analog signal Discontinuous Analog signal
x Information parameter I: Information parameter I:
signal amplitude x amplitude of the pulse
I sequence
I
the information parameter
Without quantization of
t
t
x Information parameter I:
Analog
I I I width of the pulse
sequence
t
x Information parameter I: x I I I Information parameter I:
phase shift or frequency phase shift of the pulse
sequence
I t Period t
Continuous multipoint signal Discontinuous multipoint signal
x Information parameter I: x Information parameter I:
Multipoint signal
signal amplitude can have signal amplitude can have
With quantization of the information
n discrete quantities n discrete quantities
I
I
t Period t
Discrete
Continuous binary signal Discontinuous binary signal
x Information parameter I: x Information parameter I: is
can have only quantities given as the code word of
parameters
0 and 1 I I the 0, 1 signals of one
Binary
1 1
period
0 0 x
t Period
Mathem. operation Flow graph symbol Realization
Electrical Mechanical Pneumatical or hydraulical
x2 x2 x2 x2
Branching x1 x1 x1 x1
x3 x3 x3 x3
Rods
R
x1 x1
x1
Adding x1 y
R y
Valve y
x2 y
x2 x2 x2
R1
Lever
Figure 2. Examples of technical realization of branching and adding.
Signal Classification 151
representation of the branching, addition, subtraction, mul- FURTHER READING
tiplication, or division of signals. The arrows point to the
direction of signal flow. Buck, J.R., Daniel, M.M. and Singer, A.C. (2001) Computer
Figure 2 shows some examples of technical realization. Explorations in Signals and Systems Using MATLAB, 2nd edn,
Signals may be represented in the time and frequency (spec- Prentice Hall, NJ.
tral) domains. For the formation of statistical characteristic Grover, D. and Deller, J. (1998) Digital Signal Processing and
values, time or statistical means are used. Furthermore, the Microcontroller, Prentice Hall PTR, NJ.
signal representations are applied by making use of geo- Hayes, M.H. (1998) Schaum’s Outline of Digital Signal Process-
metrical relationships. ing, McGraw-Hill Trade, New York.
More information on this material is given in subsequent Hsu, H.P. and Hse, H.P. (1995) Schaum’s Outline of Signals and
articles. Systems, McGraw-Hill Trade.
Karu, Z.Z. (1995) Signals and Systems Made Ridiculously Simple,
ZiZi Press, Huntsville, AL.
Levine, W.S. (1996) The Control Handbook, CRC Press, NJ.
REFERENCES
Lyons, R.G. (1996) Understanding Digital Signal Processing,
Prentice Hall PTR, NJ.
Books on the topic need to cover linear and digital signals
and systems. Digital titles are added to the linear titles given Oppenheim, A.V., Willsky, A.S. and Hamid, S. (1996) Signals
in Article 25, Introduction to Signals in Physical Systems, and Systems, Prentice Hall, NJ.
Volume 1. Williamson, D. (1999) Discrete-Time Signal Processing: An Alge-
braic Approach, Springer-Verlag, Berlin.
Woschni, E.-G. (1988) Informations Technik, Huthig, Heidel- Zadeh, L.A. and Desoer, C.A. (1963) Linear System Theory,
berg. McGraw-Hill, New York.
Woschni, E.-G. and Krauss, M. (1976) Informations Technik, Ziemer, R.E., Tranter, W.H. and Fannin, D.R. (1998) Signals and
Arbeitsbuch, Verlag Technik, Berlin. Systems: Continuous and Discrete, Prentice Hall, NJ.
27: Signals in the Frequency Domain
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
which is equivalent to (1a) and which has the oriented
1 Simplest Case – Harmonic Oscillation 152 complex amplitude
2 Periodic Signals – Fourier Spectrum 152 = Xe
X jϕ (1c)
3 Periodic Signals – Walsh Spectrum 154
4 Nonperiodic Signals – Fourier Transform 154 can be split
According to Figure 1, the directed quantity X
5 Spectral Power Density 155 into real and imaginary parts:
6 Practical Investigations to Study Signals 156
= A + jB|X|
=X
= B
7 Example of Rectangular Pulse Signal 157 X (A2 + B 2 ) ϕ = arctan
A
References 157 (1d)
Further Reading 157 The advantage of the symbolic method consists, above all,
of the possibility of having a simple and easily under-
standable addition of several partial oscillations having
the same frequency (Woschni, 1988). Periodic signals are
1 SIMPLEST CASE – HARMONIC
particularly useful as test signals, because the same sig-
OSCILLATION nal characteristic is repeated after each cycle period T
and thus can be observed on an oscilloscope synchronized
To study the nature of the frequency domain of a signal, with T .
the simplest case may be used as a foundation for the
explanation. A very important, determined fundamental
signal, which is also of major importance as a test signal,
is harmonic oscillation 2 PERIODIC SIGNALS – FOURIER
SPECTRUM
sin(ωt + ϕ)
x=X (1a)
According to Fourier, it is possible to represent this type
where X is the amplitude, ω = 2πf is the angular fre-
of signal having a cyclic time behavior x(t) by a series
quency, T = 1/f is the oscillation period and ϕ is the phase of sinusoidal and cosinusoidal oscillations with frequencies
angle (often zero). Representation in the complex plane, that are multiples of the fundamental frequency ω0 given
as shown in Figure 1(a), yields according to the so-called by
‘symbolic method,’ complex and oriented indicators x and,
consequently, the relationship 2π
ω0 = 2πf0 = (2a)
T
exp[ j(ωt + ϕ)] = X
x=X cos(ωt + ϕ)
This is shown in Figure 2, which depicts two sinusoids (of
sin(ωt + ϕ) = Xe
+ jX jωt (1b) many) that form the cyclic rectangular oscillation (square
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Signals in the Frequency Domain 153
+T /2
Im ∧ 2
+ jX Bn = x(t) sin(nω0 t) dt (3b)
T −T /2
∧
wt X
X sin f
∧ ∧ From (3a, b), one can see that there are only cosine terms
−X f +X for the even time functions x(t) = x(−t), and only sine
∧
∧
X cos f Re terms for the odd functions x(t) = −x(−t).
Transformation with the aid of Euler’s theorem, leads to
the complex Fourier series
∧
− jX
(a) +∞
1
x(t) = X(jnω0 )ejnω0 t (4a)
T n=−∞
∧
+X
f with the complex coefficient
0 p 2p wt +T /2
x(t)e−jnω0 t dt
0 f p 2p 3p wt + f X(jnω0) = (4b)
−T /2
∧ and the amplitude spectrum |Cnk | given by
−X
(b) 1
|Cnk | = X(jnω0 )
(4c)
T
Figure 1. Harmonic oscillation: (a) indicator representation and
(b) time representation.
x (t ) wal0 (t )
Harmonic (3w0) Fundamental oscillation (w0)
+1
a sal 1 (t )
x (t ) −1
cal 1 (t )
a /2
A0 / 2 sal 2 (t )
DC component
cal 2 (t )
0 T /2 T 3T /2 t
sal 3 (t )
Figure 2. Representation of the rectangular oscillation by
sinusoids. cal 3 (t )
sal 4 (t )
wave signal, see also Figure 7) cal 4 (t )
∞ sal 5 (t )
1
x(t) = A0 + [An cos(nω0 t) + Bn sin(nω0 t)] (2b) cal 5 (t )
2 n=1
sal 6 (t )
with the amplitude spectrum cal 6 (t )
sal 7 (t )
Cn = (A2n + Bn2 ) (2c)
cal 7 (t )
The Fourier coefficients can be calculated from the sal 8 (t )
relationships
+T /2
−T /2 0 t +T /2
2
An = x(t) cos(nω0 t) dt (3a)
T −T /2 Figure 3. Walsh functions up to eighth order.
154 Foundations of Measuring
3 PERIODIC SIGNALS – WALSH As one can see, the Walsh spectra, which are also called
SPECTRUM sequency spectra, are superior to the Fourier spectra, in that
the multiplication with sine and cosine functions respec-
tively, is obviated by a simple reversal of signs. Therefore,
In addition to the system of orthogonal functions based these spectra can be determined more easily by experiment
upon sine and cosine functions, other orthogonal sys- than the Fourier spectra (Figure 4).
tems have been introduced, in particular the Walsh func-
tions (Harmuth, 1970), which are shown up to eighth order
in Figure 3. 4 NONPERIODIC SIGNALS – FOURIER
From (2) and (3) respectively, the corresponding rela-
tionships are developed:
TRANSFORM
∞
Nonperiodic functions are of great importance both as deter-
x(t) = W0 + [Wcn caln (t) + Wsn saln (t)] (5a) mined signals, that is, test signals (step function, impulse
n=1 function) and as nondetermined signals (unknown signals
that are to be measured).
with the Walsh coefficients Wcn and Wsn : A discrete Fourier spectrum exists for periodic signals,
whereas a continuous spectrum follows for nonperiodic
+T /2
1 signals. It is obtained from the Fourier series – see (4) –
W cn = x(t) caln (t) dt (5b)
T −T /2 by passing to the limit:
+T /2
1 2π
Wsn = x(t) saln (t) dt (5c) T −−−→ ∞ ω0 = −−−→ dω
T −T /2 T
Multiplication Integration
A0 or W0
x (t )
× A1 or Wc 1
× A2 or Wc 2
× An or Wc n
× B1 or Ws 1
× B2 or Ws 2
× Bn or Ws n
cos(w0t ) cos(2w0t ) cos(n w0t ) sin(w0t ) sin(2w0t ) sin(n w0t )
cal1(t ) cal2(t ) caln(t ) sal1(t ) sal2(t ) saln(t )
Function generator Function generator
cos(n w0t ) or caln (t ) sin(n w0t ) or saln(t )
Figure 4. Experimental determination of the Fourier and Walsh coefficients.
Signals in the Frequency Domain 155
1 dω and for convolution
−−−→ nω0 −−−→ ω (6a)
T 2π +∞
1 (jω)X
x1 (τ )x2 (t − τ ) dτ O– X 2 (jω) (7f)
In this case, it yields −∞
+∞
X(jω) = x(t)e−jωt dt = F {x(t)} (6b)
−∞
5 SPECTRAL POWER DENSITY
+∞
1
x(t) = X(jω)e jωt
dω = F −1 {X(jω)} (6c) For nondeterministic signals, for instance noise, let us
2π −∞ assume in the following text that they are stationary, that is,
that their time averages are not time-dependent quantities.
Here the integrals are to be understood as the Cauchy To identify such signals x(t), the spectral power density
principal value Sxx (ω) is used. It is defined as the part of the power P ,
+c which falls into a differentially small frequency range ω,
lim ··· that is,
c→∞ −c
P dP
Sxx (ω) = lim = (8a)
F and F −1 are the Fourier and inverse-Fourier trans- ω→0 ω dω
forms respectively. Physically, X(jω) represents the com- In contrast to the spectral amplitude density, which cannot
plex amplitude related to dω, and is therefore also called be determined in the case of random signals, the spectral
spectral amplitude density having the dimension of ampli- power density is a real-valued function of the frequency ω.
tude per frequency interval, that is, V s and V Hz−1 respec- It does not contain any phase information. The latter is lost
tively. Depending on whether the frequency scale [Hz] or in the calculation of the average value, which is necessary
the angular frequency scale [s−1 ] is chosen, the values will for the formation of the power. This can also be seen from
differ by a factor 2π. the relationship existing on the basis of Parseval’s equation,
An identical calculation can be made for Walsh functions. by averaging over a time domain T (Zadeh and Desoer,
This leads to the sequential amplitude density (Harmuth, 1963; Woschni, 1988):
1970).
The Fourier transform has the following important prop- 1
|X(jω)| 2
Sxx (ω) = lim (8b)
erties and theorems. The transform is linear, that is, 2π T →∞ 2T
Consequently, the spectral power density is real and always
aν xν (t) O– (jω)
aν X (7a)
ν positive, and it is an even function for which Sxx (ω) =
ν ν
Sxx (−ω). Since the phase angle is missing, Sxx (ω) does not
contain the full information about x(t); a reverse calculation
(where O– is the sign for ‘assignment’). For a change in
is not possible. The power P of the entire signal existing
the timescale, the relationship is
in the whole frequency domain, can be calculated on the
basis of Parseval’s equation, for the energy W
1 jω
–
x(at) O X (7b)
|a| a
+∞ +∞
1 2
Particular importance should also be attributed to the dis- W = x 2 (t) dt = |X(jω)| dω
−∞ 2π −∞
placement theorems, namely, the time displacement
(Zadeh and Desoer, 1963; Woschni, 1988):
–
x(t − t0 ) O X(jω) exp(−jωt0 ) (7c) +T
1
P = x (t) = lim
2 x 2 (t) dt
T →∞ 2π −T
and the frequency shift +∞
1 |X(jω)| 2
= lim dω
x(t) exp(jωt0 ) O– X[j(ω − ω0 )] (7d) −∞ T →∞ 2π 2T
+∞
= Sxx dω (8c)
For differentiation, one obtains −∞
dn x(t) – From (8a, b) it can be concluded that a random phe-
O (jω)n X(jω) (7e)
dt n nomenon contains a periodic component of frequency ω0
156 Foundations of Measuring
w
Sxx
w
∧
x (t ) y |X ( jw)|
w
w
x
−w 0 +w
w
(a)
Figure 5. Spectral power density of a white-noise signal.
∧
with amplitude X. Dirac delta functions will develop in ∆w y |X (jw)|
Low pass Mixer
Sxx (ω) at the frequencies ±ω0 (Woschni and Krauss, 1976): x (t ) wIF w
w
wA
1 2
Sxx (ω)|ω0 = 2
X δ(|ω| − ω0 ) (8d) x
(b) wA
Furthermore, from (8c) it follows that the power density
Sxx (ω) must decrease rapidly from a certain critical fre-
y (t ) y 2 (t ) y 2 (t )
quency and must vanish at higher frequencies because of Quadrature
the requirement of boundedness of the power P . Depending (c)
on the critical frequency ωc , a distinction is made between
narrowband and wideband signals. Figure 6. Spectral analyzer: (a) switched-filter principle of oper-
ation; (b) variable center frequency of single filter method; and
A typical wideband (broadband) signal is the white-noise (c) formation of the power density.
signal with a constant power density, as shown in Figure 5.
There is a relation between time spectral power density
and the auto correlation function, see Article 29, Rela- The low-pass filter at the input eliminates the image fre-
tionship Between Signals in the Time and Frequency quency ω∗, so that one can cover the entire frequency
Domain, Volume 1. domain required by tuning ωA . Synchronizing the sweep
signal with the x-axis sweep will generate the amplitude
spectrum on the oscilloscope display. It is important to
6 PRACTICAL INVESTIGATIONS TO ensure that the filter has enough time to respond to the start-
STUDY SIGNALS ing surge. For the transient time ttr of the filter, according
to Shannon’s sampling theorem, see Article 31, Discrete
The foundation for investigations in practice is the possi- Signal Theory, Volume 1:
bility of determining the characteristic functions and values
by experiment. For the registration of spectral amplitude 1 π
ttr = = (9c)
density and power density respectively, one applies the 2f ω
same principles. In the case of filtering, several filters of
bandwidth ω are used, which are staggered in the fre- The sampling of the spectrum must, therefore, be carried out
quency, and whose outputs will be connected one after the relatively slowly (low sweep frequency of the oscilloscope),
other to a display unit for the voltage that is proportional to for which reason long-persistence cathode ray oscilloscope
|X(jω)|. According to Figure 6(a), it is likewise possible to methods are required. During the scanning run, the spec-
synchronize the switch with a sweep voltage that deflects trum must be practically stable; it must be a steady signal
the beam of an oscilloscope in the x-direction, proportion- in the statistical sense.
ally to ω; consequently, the spectrum |X(jω)| = f (ω) can In addition to filtering and variable frequency methods,
be recorded. increasing techniques are used today, which make use
Figure 6(b) shows another method that is characterized of computers to implement the Fourier transform directly
by the fact that only one filter of bandwidth ω and of according to (6b). This will be discussed in detail
center frequency ωIF (which is also called intermediate in Article 29, Relationship Between Signals in the Time
frequency) is required. Tuning is carried out by mixing with and Frequency Domain, Volume 1, where specially
the continuously tunable auxiliary frequency ωA , where the adapted techniques for the fast Fourier transform will be
following frequencies will be allowed to pass: dealt with. The computing advantages that arise when
the amplitude sequential spectrum defined by the Walsh
ω = ωIF − ωA ; ω∗ = ωIF + ωA (9a, b) functions is used instead of the amplitude frequency
Signals in the Frequency Domain 157
spectrum, have already been pointed out in connection with Figure 7 shows the spectrum as well as the envelope
Figure 4. While the techniques described in Figure 6(a, b) curve for τ = 0.5. This signal was already treated in
supply the amplitude response of the spectrum |X(jω)| only, Section 2 of this article.
it is possible to obtain additional phase information from The relationship with Figure 2 becomes immediately
the relationship ϕ = arg[X(jω)] using computing methods. evident: the smaller the pulse width, that is, ttr = T , the
To display the power spectrum, the formulation more complicated the problem is in general. The pulse
P /ω = x 2 (t)/ω is implemented, as outlined in height is still correctly indicated in the region up to the
Figure 6(c), between the filter output and the input of the first zero of the envelope curve. For τ → 0, a constant
display unit and the oscilloscope. spectrum results because the first zero shifts towards ω =
∞. This case is significant as a test signal. For τ → 0,
7 EXAMPLE OF RECTANGULAR PULSE the unit impulse becomes the Dirac function δ(t) with the
normalization
SIGNAL +∞ +0
For a periodic sequence of rectangular pulses having the δ(t) dt = δ(t) dt = 1
−∞ −0
pulse width T or the repetition ratio τ = T /T (see
Figure 7a), after substitution into (3a, b) and (2c) or (4b, According to (4b), this has a spectral amplitude density
c) and after an elementary calculation, one obtains pure
X(jω) = 1, with only cosine oscillations of constant ampli-
cosinusoidal oscillations having the amplitudes tude occurring.
Some more signals that are used as test signals as well
sin(nπτ )
Cn = An = 2τ = 2τ sin(nπτ ) (10a) as the approximate representation of measurement signals
nπτ are treated in Article 28, Signals in the Time Domain,
Volume 1.
x
REFERENCES
1
Harmuth, H.F. (1970) Transmission of Information by Orthogonal
Functions, Springer, Berlin.
Woschni, E.-G. (1988) Informations Technik, Huthig, Heidelberg.
Woschni, E.-G. and Krauss, M. (1976) Informations Technik,
Arbeitsbuch, Verlag Technik, Berlin.
Zadeh, L.A. and Desoer, C.A. (1963) Linear System Theory,
−3∆T /2 −T/2 −∆T /2 0 ∆T /2 T/2 3∆T /2 t McGraw-Hill, New York.
(a)
Cn
1 FURTHER READING
Buck, J.R., Daniel, M.M. and Singer, A.C. (2001) Computer
Explorations in Signals and Systems Using MATLAB, 2nd edn,
Prentice Hall, NJ.
Hsu, H.P. and Hse, H.P. (1995) Schaum’s Outline of Signals and
0.5 Systems, McGraw-Hill Trade.
Cn (n)
Karu, Z.Z. (1995) Signals and Systems Made Ridiculously Simple,
A0 / 2 ZiZi Press, Huntsville, AL.
C1 Oppenheim, A.V., Willsky, A.S. and Hamid, S. (1996) Signals
C7 and Systems, Prentice Hall, NJ.
C3 C5
Tolimieri, R., An, M. and Lu, C. (1997) Algorithms for Discrete
0 1 2 3 4 5 6 7 n Fourier Transform and Convolution, 2nd edn, Springer, New
(b)
York.
Ziemer, R.E., Tranter, W.H. and Fannin, D.R. (1998) Signals and
Figure 7. (a) Sequency of pulses and (b) spectrum for τ = 0.5. Systems: Continuous and Discrete, Prentice Hall, NJ.
28: Signals in the Time Domain
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
represents a measure of the power. The root of the mean
1 Mean Values 158 square value is the effective value Xeff given by
2 Correlation Function 158
3 Relationships to Spectral Power Density 159 Xeff = [x 2 (t)] (2a)
4 Practical Investigations in the Signal Time
Domain 160 The following relationship exists between the component
5 Examples of Mean Value and Correlation x(t), the AC component x(t), and the mean square value
˜
Functions 161 Xeff :
References 162 Xeff = x 2 (t) = [x(t)]2 + x 2 (t) (2b)
Further Reading 162 ˜
For a harmonic oscillation, one obtains Xeff = X/ √2 for
the effective value, see Section 5. If the Ergodic-theorem
1 MEAN VALUES is fulfilled, normally in practice the effective value Xeff is
equal to the standard deviation δ and the mean square value
2
A time function x(t) can be characterized by time averages Xeff to the variance δ 2 , see Article 30, Statistical Signal
of the nth order, which are also called moments of nth Representations, Volume 1.
order:
+T
1
x (t) =
n x n (t) dt (1a) 2 CORRELATION FUNCTION
2T −T
or, for nonperiodic signals: The correlation function ψ(τ ) represents a generalized
mean square value, where a function is multiplied by the
+T
1 function displaced by time τ , and then the mean value is
x n (t) = lim x n (t) dt (1b)
T →∞ 2T −T formed. If this function is the same function x, we call it
an autocorrelation function ψxx (τ ),
For n = 1, one obtains the (arithmetic) linear average,
+T
which, from the physical point of view, can be interpreted to 1
be the zero-frequency component (DC value) of the signal ψxx (τ ) = lim x(t)x(t + τ ) dt = x(t)x(t + τ )
T →∞ 2T −T
or, according to Article 27, Signals in the Frequency (3a)
Domain, Volume 1, the Fourier coefficient A0 /2. as Figure 1 shows.
Of particular importance is the average value for n = It is suitable for making statistical statements about the
2, which, as a mean square value according to (8c) of internal relationships between function sections, as is now
Article 27, Signals in the Frequency Domain, Volume 1, shown in a survey of its typical properties:
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Signals in the Time Domain 159
t solution of this measurement problem is the foundation for
adaptive systems (Davies, 1970).
x (t +t) x (t ) In contrast to the autocorrelation function, the cross-
correlation function has the following features:
t 1. It has no even functions, but the following relationship
holds:
Figure 1. Definition of the autocorrelation function ψxx (δ).
ψxy (τ ) = ψyx (−τ ) (4b)
1. In averaging, the phase information is lost, as is the
2. It contains relative phase information concerning the
case for the spectral power density (Article 27, Sig-
two events x(t), y(t). In particular, the cross-correlation
nals in the Frequency Domain, Volume 1). There-
function of two harmonic signals with the same fre-
fore, there are also direct relationships between the
quency disappears if the phase shift is ±π/2, as can
autocorrelation function ψxx (τ ) and the spectral power
be seen following substitution into (4a). Likewise, the
density Sxx (ω), which is discussed in detail in Arti-
cross-correlation of two harmonic oscillations is zero
cle 27, Signals in the Frequency Domain, Volume 1.
if the frequencies are unequal.
However, periodic components in the signal x(t) will
3. The limiting cases are
be maintained without giving consideration to the phase
position, because the following expression applies to
ψxy (0) = ψyx (0) = x(t)y(t) (4c)
the autocorrelation function of the harmonic oscillation,
independent of the phase position: lim ψxy (τ ) = x(t) × y(t) (4d)
τ →∞
2 cos(ωτ ) = Xeff
ψxx (τ ) = 12 X 2
cos(ωτ ) (3b)
Experimental registration of the correlation function is
discussed in Section 4 of this article.
2. The value for τ = 0 represents, according to (3a), the
mean square value and is the maximum value of the
autocorrelation function
3 RELATIONSHIPS TO SPECTRAL
ψxx (0) = x 2 (t) = 2
Xeff (3c) POWER DENSITY
The other threshold value for τ → ∞ is the square of The autocorrelation function, like the spectral power den-
the linear mean value: sity, contains no phase information; this is lost in both cases
because of the averaging operation. There is a relationship
lim ψxx (τ ) = [x(t)]2 (3d) between both functions, as is the case between the time
τ →∞
behavior of the signal x(t) and the corresponding spectral
3. Since it is of no significance whether the function x(t) amplitude density X(jω), via the Fourier transform; this
in (3a) is displaced toward positive or negative times, relationship is known as the Wiener–Chinchine theorem
the autocorrelation function is an even function: (Woschni, 1988; Davies, 1970):
+∞
ψxx (τ ) = ψxx (−τ ) = x(t)x(t + τ ) = x(t)x(t − τ ) 1 1
Sxx (ω) = ψxx (τ )e−jωτ dτ = F {ψxx (τ )} (5a)
(3e) 2π −∞ 2π
+∞
If two different signals x(t), y(t) are being compared one ψxx (τ ) = Sxx (ω)ejωτ dω = 2πF −1 {Sxx (ω)} (5b)
with the other, the measure used for the statistical relation- −∞
ship between them is the cross-correlation function ψxy (τ )
according to the definition: Since the autocorrelation function is an even function (see
Section 2 of this article), cosinusoidal oscillations only
+T occur. Consequently, (5a, b) can be rewritten
1
ψxy (τ ) = lim x(t)y(t + τ ) dt = x(t)y(t + τ )
T →∞ 2T
−T 1 ∞
(4a) Sxx (ω) = ψxx (τ ) cos(ωτ ) dτ (5c)
In measurement technology, the cross-correlation function π 0
∞
plays a major role in solving system identification tasks
ψxx (τ ) = 2 Sxx cos(ωτ ) dω (5d)
during normal operation by means of the disturbances. The 0
160 Foundations of Measuring
If the autocorrelation function in (5a, b) is substituted by the
cross-correlation function ψxy (τ ), then the corresponding 1/2T
y = x2
relationships with the cross-power density will be obtained: x (t ) y (t ) x 2(t )
+∞
1 1
Sxy (jω) = ψxy (τ )e−jωτ dτ = F {ψxy (τ )} (6a) Figure 3. Measurement of the mean square value.
2π −∞ 2π
+∞
ψxy (τ ) = Sxy (jω)ejωτ dω = 2πF −1 {Sxy (jω)} (6b) The mean square value can be found by means of
−∞ the principle shown in Figure 3. To obtain the square
of a value, either electronic circuits with a square-law
A survey of these relations is given in Article 29, Rela-
characteristic, for example, diodes or transistors (as in the
tionship Between Signals in the Time and Frequency
case of transistor voltmeters), or a measuring device having
Domain, Volume 1.
a square-law characteristic, such as a soft-iron or hot-wire
instrument (heating ≈ P ≈ I 2 R) is used.
Figure 4 shows the classical basic system for determin-
4 PRACTICAL INVESTIGATIONS IN THE ing the autocorrelation function or the cross-correlation
SIGNAL TIME DOMAIN function. Delay is due to a delay section, multiplication,
and averaging yield, according to (3a), the autocorrelation
The characteristic functions and values that have been function (switch in position A) or, according to (4a), the
introduced can be found by experiment. cross-correlation function (switch in position B). Since the
For the measurement of the linear mean value, x(t), integration time T cannot be chosen to be infinite, only the
either moving-coil instruments or transistor voltmeters hav- short-term correlation function
ing a series-connected integration link is used; compare +T
Figure 2(a), with Figure 2(b) showing a simple analog cir- 1
ψk (τ ) = · · · dt (8)
cuit for finding the average value according to the equation 2T −T
1 1 will be measured in practice; under certain circumstances it
uo = iC dt ≈ ui dt (7) reflects the actual behavior of ψ(τ ) with sufficient accuracy.
C RC
The just treated classical analog methods have been
using the assumption R 1/ωC. increasingly displaced by digital methods using the princi-
ples of digital signal processing. There the signal function
R iR
x(t) is sampled leading to sampled values x(rTs ) with the
sampling time Ts . These values are processed due to the
ui iC uo
1/2T algorithms given by (1a), (3a), or (4a), leading to the mean
x(t ) x (t ) values or the correlation functions.
C
To fulfil the sampling theorem conditions, high sampling
rates are necessary to avoid aliasing errors, see Article 31,
(a) (b)
Discrete Signal Theory, Volume 1. On the other hand,
Figure 2. Measurement of the linear mean value: (a) basic circuit online processing that is often demanded in signal process-
and (b) simple realization. ing, means that the necessary processing time has to be
Delay system
x (t ) x (t- t)
t
Multiplication Mean value
1
T
2
y....(t)
y (t ) A
B
Figure 4. Experimental determination of the correlation function: (a) autocorrelation function ψxx (τ ) and (b) cross-correlation function
ψxy (τ ).
Signals in the Time Domain 161
smaller than the sampling time. Therefore, in general, the 2. for full-wave rectification
numbers of samples n are chosen due to n = 2s so that
the time for the division through n can be small, because T
2a 2 2a
instead of the time-consuming division now only a shifting x(t) = sin ω0 t dt = (11a)
T 0 π
is necessary.
T
Methods based on the application of Fast Fourier Trans- 2a 2 a2
form (FFT) are used to calculate the correlation functions x 2 (t) = sin2 ω0 t dt = (11b)
T 0 2
from power density functions and vice versa, see Arti-
a
cle 27, Signals in the Frequency Domain, Volume 1; Xeff = x 2 (t) = (11c)
Article 29, Relationship Between Signals in the Time 2
and Frequency Domain, Volume 1. See also Article 143,
DFT and FFTs, Volume 3. Let us deal now with an example of correlation func-
tions for a rectangular pulse according to Figure 6(a). The
5 EXAMPLES OF MEAN VALUE AND
CORRELATION FUNCTIONS yxx(t)
As a first example, we find the mean values of a harmonic x^
oscillation x^ 2
+T
1 sin ωt dt = 0
x(t) = X (9a)
2T −T
+T 2
1 sin2 ωt dt = X
x 2 (t) = X (9b)
2T −T 2 t t
−T1 +T1 −2T1 +2T1
This means the effective value is (a) (b)
X
= √
Xeff = X (9c)
Figure 6. (a) Rectangular pulse and (b) corresponding autocor-
2 relation function.
The next example is the rectifier unit as used in mea-
surement instruments. The signal is shown in Figure 5 for S (w)
half-wave (thin) or full-wave (thick) rectification. a
The results are
1. for half-wave rectification
T
a 2 a
x(t) = sin ω0 t dt = (10a) −w −wg O +wg w
T 0 π
(a)
T
a 2 a2
x (t) =
2 sin2 ω0 t dt = (10b)
T 0 4 yxx (t)
a
Xeff = x 2 (t) = (10c)
2
t=0 t twg
3p 2p p p 2p 3p
(b)
Figure 7. (a) Narrowband noise and (b) corresponding autocor-
Figure 5. Half-wave (thin) and full-wave (thick) rectification. relation function.
162 Foundations of Measuring
autocorrelation function is given in (12). REFERENCES
+T1
ψxx (τ ) =
1 ×X
X 2 1 − |τ |
dt = X (12) Davies, W.D.T. (1970) System Identification for Self-Adaptive
2T1 −T1 +τ 2T1 Control, Wiley, Chichester.
Woschni, E.-G. (1988) Informations Technik, Huthig, Heidelberg.
The result shown in Figure 6(b) reveals that if the pulse
width 2T1 decreases in the limiting case T1 → 0, that is, for
the unit impulse δ(t), the autocorrelation function is also a FURTHER READING
delta function. This also follows immediately from the cal-
culation of the autocorrelation function of white noise hav- Buck, J.R., Daniel, M.M. and Singer, A.C (2001) Computer
ing a constant power spectrum Sxx (ω) = constant, as is the Explorations in Signals and Systems Using MATLAB, 2nd edn,
case for unit impulse. For this calculation, (5a, b) are used. Prentice Hall, NJ.
The relationship between the width of the power spec- Grover, D. and Deller, J. (1998) Digital Signal Processing and
trum and the corresponding autocorrelation function can the Microcontroller, Prentice Hall PTR, NJ.
also be seen if the autocorrelation function for narrowband Hayes, M.H. (1998) Schaum’s Outline of Digital Signal Process-
noise is calculated in accordance with Figure 7(a): ing, McGraw-Hill Trade, New York.
Hsu, H.P. and Hse, H.P. (1995) Schaum’s Outline of Signals and
+ωg sin(ωg τ ) Systems, McGraw-Hill Trade.
ψxx (τ ) = a e−jωt dω = 2aωg (13)
−ωg ωg τ Karu, Z.Z (1995) Signals and Systems Made Ridiculously Simple,
ZiZi Press, Huntsville, AL.
As shown in Figure 7(b), the autocorrelation function Oppenheim, A.V., Willsky, A.S. and Hamid, S. (1996) Signals
becomes smaller as the noise bandwidth increases, degen- and Systems, Prentice Hall, NJ.
erating into a delta function for white noise. Williamson, D. (1999) Discrete-Time Signal Processing: An Alge-
This shows clearly that statistical relationships cease to braic Approach, Springer-Verlag, Berlin.
exist if two noise signals are slightly displaced relative to Ziemer, R.E., Tranter, W.H. and Fannin, D.R (1998) Signals and
each other. Systems: Continuous and Discrete, Prentice Hall, NJ.
29:Relationship Between Signals in the Time and
Frequency Domain
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
representation in the frequency domain of Figure 1 in the
time domain.
1 Nature of Relationships and Transforms 163
2 Fourier Transform as the Means to Move
Between the Domains 163
3 Theorems of Fourier Transform 164 2 FOURIER TRANSFORM AS THE
4 Table of Time–Frequency Relationships 164 MEANS TO MOVE BETWEEN THE
5 Example of Time–Frequency Transform 166 DOMAINS
6 Fast Fourier Transform Concept and its
Application 167 As given in detail in Article 27, Signals in the Frequency
References 167 Domain, Volume 1, the mathematical relation between the
Further Reading 167 time and frequency domain and vice versa is given by the
Fourier transform
+∞
X(jω) = x(t)ejωt dt = F {x(t)} (1a)
t=−∞
1 NATURE OF RELATIONSHIPS AND +∞
1
TRANSFORMS x(t) = X(jω)e jωt
dω
2π −∞
=F −1
{X(jω)} (1b)
To show the main ideas of time–frequency transform,
the square wave time signal, given in Figure 1, may With regard to mathematics, this is a transformation
be considered. between the original range – physically the time domain –
Figure 1(a) shows the fundamental harmonic and the and the image range – physically the frequency domain.
harmonic of first order with three times the basic fre- This transformation is bidirectional, which means that it
quency. The addition of these two frequencies as a first is possible to move into the frequency domain if the time
approximation already yields a similar function (Figure 1b). domain is given and vice versa, as the example in Section 5
If one takes into consideration all harmonics, a reason- of this article demonstrates.
ably exact square wave signal appears, which shows that The fact that the convergence of this transformation
the addition of frequencies leads to the appropriate time is limited is known so that, especially in system the-
function. This means that the amplitude spectrum of sinu- ory, another transformation with no restriction due to
soidal functions, as given in Figure 2, is the equivalent the convergence is used, the Laplace transformation –
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
164 Foundations of Measuring
(a) −T /2 t=0 +T /2 +T +3T /2 t
(b) −T /2 t=0 +T /2 +T +3T /2 t
Figure 1. Square wave time signal (a) time signal with harmonics and (b) addition of the fundamental and first harmonic.
An Domain, Volume 1 and Article 37, Relationship Between
Systems in the Time and Frequency Domain, Volume 1,
these results are used to describe the behavior of systems.
The shifting theorem is the key to solve problems of discrete
signal theory and system theory as sampled data systems
2a as well as prediction or delay elements.
π
Hyperbel 1/n
4 TABLE OF TIME–FREQUENCY
RELATIONSHIPS
A summary is now given of the relationships established
so far between the various signal representations in both
the time domain and the frequency domain. Consideration
0 wo 3wo 5wo 7wo 9wo w
is also given to the possibilities of conversion indicated
2wo 4wo 6wo 8wo
above. Table 2 contains a survey of such relationships.
Figure 2. Amplitude spectrum of the time function of Figure 1(a). While the time function x(t) and the spectral amplitude
density X(jω) contain the full information concerning the
see Article 37, Relationship Between Systems in the signal, this is not the case for the functions resulting from
Time and Frequency Domain, Volume 1. averaging (autocorrelation function and power density).
Phase information is lost due to averaging. For this reason,
conversions are possible only in the direction indicated by
3 THEOREMS OF FOURIER the arrow.
TRANSFORM Conversions are possible, via the Fourier transform,
between functions in the time domain and corresponding
A survey of the theorems of the Fourier transform is functions in the frequency domain. Therefore, it does not
given in Table 1 and these correspond to the theorems matter, as far as the significance of the statement is con-
of Laplace transformation as treated in Article 37, Rela- cerned, which of these two functions is measured; rather
tionship Between Systems in the Time and Frequency it is a question of convenience. For instance, it is use-
Domain, Volume 1. ful to carry out measurements in optical communications
Of great importance are the differentiation and the predominantly in the time domain. In vibration measur-
integration theorems. Corresponding to the differentiation ing technology, however, it is preferable to carry out the
of a sinusoidal function d/dt ejωt = jωejωt , the multiplica- measurements in the frequency domain.
tion with jω in the frequency domain means a differentiation The importance of the conversions has already been
in the time domain and appropriate the division with jω discussed. Mention should again be made of the fact that
an integration. In Article 36, Systems in the Frequency x(t) and X(jω) are related in the same way as are ψxx (τ )
Relationship Between Signals in the Time and Frequency Domain 165
Table 1. Theorems of Fourier Transform.
(1) Addition theorem
F{f1 (t) + f2 (t)} = F{f1 (t)} + F{f2 (t)}
(2) Multiplication theorem
F{af (t)} = aF{f (t)}
(3) Shifting theorem
for a > 0 F{f (t − a)} = e−jωa F (jω)
a
or F{f (t + a)} = ejωa F(jω) − e−jωa f (t) dt
0
(4) Likeness theorem
1 p
F{f (at)} = F if a > 0
a a
(5) Attenuation theorem
F{e−at f (t)} = F (jω + a)
(6) Limit theorem
lim f (t) = lim jωF (jω) lim f (t) = lim jωF (jω)
t→∞ jω→0 t→0 jω→∞
(7) Integration theorem
t
1
F f (τ ) dτ = {f (t)}
0 jω
(8) Differentiation theorem
F{f (n) (t)} = jωn F{f (t)} − jωn−1 f (+0) − · · · − f (n−1) (+0)
if the limits
lim f (t) = f (+0); lim f˙(t) = f˙(+0); . . . ;
t→0 t→0
lim f (n−1)
(t) = f (n−1)
(+0), exist
t→0
Note: Woschni, 1988.
166 Foundations of Measuring
Table 2. Relationships between time- and frequency-domain signals.
Time domain Frequency domain
Function of time Amplitude density
Real function Complex
of a real +∞ +∞ function of
1
variable x(t) = X(jω)ejωt
dω ←−−− F −−−→ X(jω) = x(t)e−jωt dt a real
2π −∞ −∞ variable
Mean value Only
operation; unilateral
1
|X(jω)| 2
phase 1 +T Sxx (ω) = lim conversion
ψxx (τ ) = lim x(t)x(t + τ ) dt
2π T →∞ 2T
information τ →∞ 2T −T
possible
is lost
Autocorrelation function
Real function Real function
of a real +∞ +∞ of a real
1
variable ψxx (τ ) = Sxx (ω)ejωτ dω ←−−− F −−−→ Sxx (ω) = ψxx (τ )e−jωτ dt variable
−∞ 2π −∞
Note: Reproduced from Handbook of Measurement Science, Volume One, Sydenham, 1983 John Wiley & Sons Ltd.
and Sxx (ω). For instance, a constant amplitude density interpretation; only, in this case at t = 0 all spectral ampli-
has the delta function, which is a time function, as an tude will be strengthened leading at least to the Dirac pulse
autocorrelation function, just as the constant power density (Figure 4a). The amplitude density of the unit step func-
in white noise has, see also Article 28, Signals in the Time tion is
Domain, Volume 1. +∞
X(jω) = F {ω(t)} = 1.e−jωt dt = 1 (3a)
ω
0
5 EXAMPLE OF TIME–FREQUENCY
This integral does not converge, so we have to solve the
TRANSFORM
following integral
As an example with great relevance to practical application, ∞ ∞
−t −jωt 1
let us deal with two signals used as test signals: The unit lim e e dt = lim e−(+jω)t
→0 0 →0 −( + jω) 0
step ω(t) and the unit pulse or Delta function δ(t) as shown
in Figure 3(a, b). 1
= (3b)
The spectral amplitude density of the unit pulse func- jω
tion is 1
+∞ and we get a hyperbolic spectrum as shown in Figure 4(b),
containing only sine functions: the step function is an
X(jω)∂ = F {δ(t)} = δ(t)e−jωt dt = 1 (2)
−∞ uneven function so that only a sum of sine functions could
approximate the course of the time function.
This means that only cosine functions with constant spec-
tral amplitude appear, as also follows from the physical
^ ^
|1X (jw)| X (jw)
a ∞
1
w
w
(a) 0 (b) 0
(a) 0 t (b) 0 t
Figure 4. (a) Spectrum of a unit pulse function and (b) spectrum
Figure 3. (a) Unit pulse function and (b) unit step function. of a unit step function.
Relationship Between Signals in the Time and Frequency Domain 167
Another typical example is treated in Article 28, Signals necessary. If N = 2S – that means N is a power of 2 –
in the Time Domain, Volume 1, Section 5. it is possible by means of division in subsumes and using
an appropriate algorithm – the so-called Cooley–Tuckey
Algorithm (Cooley and Tuckey, 1965) – to need only N /2
6 FAST FOURIER TRANSFORM lb(N /2) multiplications. For the often used value of N =
CONCEPT AND ITS APPLICATION 1024, the number of multiplications thus can be reduced
from 1048576 to 4608, which means that the calculation
To accomplish the Fourier transform today, digital comput- time is reduced to 1/227 of the original value!
ers are used. The basis for this is the Fast Fourier Transform Before microelectronics and microprocessors were devel-
(FFT). Using a transform specially tailored to fit the way oped, the application of Fourier transform was limited to
in which the digital computer operates, Discrete Fourier special cases because it was extremely expensive to real-
Transform (DFT) programs have been established, which ize the transformation using analog methods. The classical
save computing time, and where at least 1024 graph support application example in the beginning of the 1960s has
points are quite usual. The calculation supplies the Fourier been the testing of gears at the Volkswagen company using
coefficients. As the number of support points is limited, the a spectral analyzer implemented in the analog technique.
short-term correlation function is determined, which how- Today, the field of application is very broad. Signal ana-
ever, is practically identical with the correlation function, lyzer, test equipment, and software tools now routinely
provided the correlation time is τk < Ts . make use of the FFT algorithms. See also Article 143,
To show the main ideas of FFT, we start from the DFT and FFTs, Volume 3.
equation of the Fourier transformation (1a), writing this
equation for a sampled input signal with the sampling
time Ts . REFERENCES
+∞
+∞
Cooley, J.W. and Tuckey, J.W. (1965) An Algorithm for the
X∗ (t) = ∗ (jω) = Ts
x(kTs )X x(kTs )e−jωkTs Machine Calculation of Complex Fourier Series. Mathematical
Computation, 19, 279–301.
k=−∞ k=−∞
(4a) Woschni, E.-G. (1988) Informations Technik, Huthig, Heidelberg.
Because it is not possible to use an infinite number of
sampling values, a time boundary is realized, which means
that instead of kmax = ∞ the maximum value of k is limited FURTHER READING
to N and instead of (4a) we get
Buck, J.R., Daniel, M.M. and Singer, A.C. (2001) Computer
N Explorations in Signals and Systems Using MATLAB, 2nd edn,
+
2 Prentice Hall, NJ.
∗ (jnω ) = T
X x(kTs )e−jknω0 Ts (4b)
0 s Grover, D. and Deller, J. (1998) Digital Signal Processing and
N the Microcontroller, Prentice Hall PTR, NJ.
k=−
2
Hayes, M.H. (1998) Schaum’s Outline of Digital Signal Process-
or with the fundamental frequency ω0 = 2π/NT ing, McGraw-Hill Trade, New York.
Tolimieri, R., An, M. and Lu, C. (1997) Algorithms for Discrete
N
+ Fourier Transform and Convolution, 2nd edn, Springer, New
2 j2πnk
X ∗ j 2πn
∗ (jnω ) = X = Ts x(kTs )e− N (4c)
York.
0
N Ts N
Williamson, D. (1999) Discrete-Time Signal Processing: An Alge-
k=− braic Approach, Springer-Verlag, Berlin.
2
Woschni, E.-G. (1994) Signal Theory in Measurement and Instru-
From (4c) N complex Fourier coefficients arise. To get mentation, Concise Encyclopedia of Measurement & Instrumen-
these coefficients due to (4c), N 2 multiplications are tation, Pergamon Press, Oxford.
30: Statistical Signal Representations
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
as shown in Figure 1 with
1 Probability Distribution 168
tr
2 Probability Density and Relation to W (x) = (1b)
Probability Distribution 168 T
3 Multidimensional Probability Functions 169
The limiting values of the probability distribution
4 Gaussian Distribution 170
5 Relationship to Mean Values; Ergodic lim W (x) = 0 (1c)
Theorem 170 x→−∞
6 Practical Investigations 171 lim W (x) = 1 (1d)
x→+∞
7 Example of Use in Metrology 172
References 172 are also clearly understandable, because they mean the
impossibility of a value smaller than −∞ as well as the
certainty of the occurrence of any signal value ξ(t) < ∞.
For continuous functions ξ (t), the probability distribution
W (x) is a monotonically increasing function, see also
1 PROBABILITY DISTRIBUTION Figure 3.
In the following text, it is assumed that the probability 2 PROBABILITY DENSITY AND
functions are not dependent on time. In this situation, the
process is a stationary stochastic one. In most cases – RELATION TO PROBABILITY
for instance, with noise – this proposition is met. If the DISTRIBUTION
stochastic parameters are changing relatively slowly, the
process can be approximated by a stationary one for a Now we are interested in the probability p(x) of the
determined time (Mendenhall and Sincich, 1992). function ξ(t) to be within a small region x near the
To describe randomly fluctuating events ξ (t), use is made value x
of characteristic functions, which are based on the theory
of probabilities. x ≤ ξ(+) < x + x (2a)
The probability distribution W (x), which is also called
the first-order distribution function, indicates the probability As Figure 2 shows, this probability is
p that the signal ξ (t) remains smaller than a barrier x,
tr∗
p(x) = (2b)
W (x) = p[ξ(t) < x] (1a) T
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Statistical Signal Representations 169
x x(t ) dW(x)/dx = w (x)
W (X )
x
∆t1 ∆t2 ∆t3 T t 1
Figure 1. The definition of probability distribution.
W (x2)
x x(t )
W (x1)
x + ∆x ∆x
x
∆t1 ∆t2 ∆t3 ∆t4 T t (a) x1 x2 0 x
W (X )
Figure 2. The definition of probability density.
This function is also a relation of x. Therefore, one relates
to x and gains the probability density w(x) W (x2) − W (x1) =
x2
1 ∫ W (x)dx
w(x) = p[x ≤ ξ(t) < x + x]; x → dx (3a) x1
x
When comparing the Figures 3(a) and 3(b) it can be seen
that the following relations exist
x
w(u) du = W (x) (3b) (b) x1 x2 0 x
−∞
Figure 3. (a) Probability distribution and (b) corresponding pro-
and bability density.
dW (x)
= w(x) (3c)
dx lead to n probability distribution functions or probability
density functions with the limits x1 . . . xn
Taking into consideration the limit (1d), the normaliza-
tion is
+∞ W1 (x) . . . Wn (x) (4)
w(x) dx = 1 (3d)
−∞ and
As shown in Figure 3, the probability that ξ (t) lies within w1 (x) . . . wn (x) (5)
the interval x2 to x1 is calculated by
x To describe the relations between these functions, the
p[x1 ≤ ξ < x2 ] = W (x2 ) − W (x1 ) = w(x) dx (3e) compound probability distribution densities w(x1 , x2 , . . . ,
x1 xn ) are introduced
W (x1 , x2 , . . . , xn )
3 MULTIDIMENSIONAL PROBABILITY
= p[ξ1 (t) < x1 , ξ2 (t) < x2 , . . . , ξn (t) < xn ] (6a)
FUNCTIONS
w(x1 , x2 , . . . , xn )
If the stochastic process consists of not only one, but n fluc- ∂n
= W (x1 , x2 , . . . , xn ) (6b)
tuating events ξ(t)1 . . . ξ(t)n , then the same considerations ∂x1 ∂x2 , . . . , ∂xn
170 Foundations of Measuring
Furthermore, conditional probability distributions W(x1 / Table 1. Gaussian probability integral (x) evaluated for a
x2 ) and conditional probability distribution densities range of x (Jahnke and Emde, 1966).
w(x1 /x2 ) are defined. They indicate the probability that x 0 2 4 6 8
the value x1 occurs on condition that the value x2 already
0.0 0.0000 0.0226 0.0451 0.0676 0.0901
exists. The following relationships hold for the compound 0.1 0.1125 0.1348 0.1569 0.1790 0.2009
probability density: 0.2 0.2227 0.2443 0.2657 0.2869 0.3079
0.3 0.3286 0.3491 0.3694 0.3893 0.4090
w(x, y) = w(x|y) × w(y) = w(y|x)w(x) (6c) 0.5 0.5205 0.5379 0.5549 0.5716 0.5879
0 0.0000 0.2227 0.4284 0.6039 0.7421
1 0.8427 0.9103 0.9523 0.9763 0.9891
2 0.9953 0.9981 0.9993 0.9998 0.9999
4 GAUSSIAN DISTRIBUTION Source: Reproduced from Handbook of Measurement Science, Volume
One, Sydenham, 1983 John Wiley & Sons Ltd.
Of utmost importance in practice is the Gaussian distribu- The Gaussian distribution describes all stochastic pro-
tion density (Buja, 1991; Jaffe, 2000) cesses represented by a very great number of single pro-
cesses running independent of each other in the same over-
1 −(x − a)2 all event. Typical examples are the arrival of rain droplets
w(x) = √ exp (7a)
2πσ 2σ 2 or, in electronics, the thermal noise.
Very useful in the application of Gaussian distribution is
where a = x(t) is the linear mean value and σ the standard the Gauss error integral in the form
deviation, related to the square mean value, x 2 (t), by
2 x −µ2
(x) = e dµ (7c)
π 0
σ = [x 2 (t) − a 2 ] (7b) with the values given in Table 1.
Some examples of importance in measurement are treated
Figure 4 shows the Gaussian distribution density for in Section 7.
a = 0.
5 RELATIONSHIP TO MEAN VALUES;
ERGODIC THEOREM
1.6
The expectation value E of a function f (x) is defined as
follows:
+∞
W (x )
E{f (x)} = f (x)w(x) dx (8a)
−∞
1.2
For f (x) = x , the moment Mn of nth order is obtained. Of
n
s = 0.25 particular importance is the moment of first order, denoted
as statistical linear mean value x̃:
+∞
0.8 x̃ = M1 = E{x} = xw(x) dx (8b)
−∞
s = 0.5
and the statistical square mean value phase space average:
+∞
x̃ 2 = M2 = E{x 2 } = x 2 w(x) dx (8c)
−∞
0.4
s=1 Accordingly, the expectation value for the simultaneous
occurrence of x1 (t) and x2 (t + τ ) is obtained:
x1 (t)x2 (t + τ )
−4 −3 −2 −1 0 1 2 3 4 = E{x1 (t), x2 (t + τ )}
x +∞ +∞
= x1 (t)x2 (t + τ )w[x1 (t), x(t + τ )]dx1 dx2 (8d)
Figure 4. Gaussian distribution density. −∞ −∞
Statistical Signal Representations 171
If this is an ergodic event, that is, if the ergodic theorem 6 PRACTICAL INVESTIGATIONS
is satisfied, the statistical mean values x̃ n and time mean
values x n (t) are equal In order to display the probability distribution and density,
electronic majority decision elements having an adjustable
+∞ threshold value x are used, as outlined in Figure 5. Expla-
E{x n } = x̃ n = x n w(x) dx nation in terms of analog processing assist appreciation of
−∞ the process used.
+T
1 This can be done in analog systems using either a
= x n (t) = lim x n (t) dt (9a) triggering circuit or a voltage divider having a biased diode.
T →∞ 2T −T
With this, a normalization is to be carried out such that
With (8d), a definition can be obtained of the correlation the corresponding conditions, (1d) and (3d), are observed.
function ψ xy (τ ), which is based on the generalized mean The arrangement has been for long also used for displaying
value: these parameters on an oscilloscope screen, provided the
voltage of the sweep generator for the x-deflection of
+T the oscilloscope is used to control the threshold value x.
1
ψxy (τ ) = lim x(t)y(t + τ ) dt The sweep frequency must be slow enough to ensure that
T →∞ 2T −T adequate averaging occurs.
= x(t)y(t + τ ) By coupling several installations in accordance with
+∞ +∞ Figure 5(b), compound probability distributions W (x, y)
= x(t)y(t + τ )w[x(t), y(t + τ )] dx dy can also be recorded. For this purpose, the trigger outputs
−∞ −∞ of one arrangement for each event x, y will be connected
(9b) with an AND element and further processed as shown in
For the particularly important Gaussian distribution density Figure 5(b) (Woschni, 1988). Today, displays use digital
according to (7a), one calculates processing to perform the task following digitization of the
signal in an analog to digital (A/D) converter.
M1 = x̃ = x(t) In practice it is possible to describe many events, at
+∞ least approximately, by the Gaussian distribution of (7a).
x −(x − a)2
= √ exp dx = a (9c) The probability for the fluctuation process to lie within the
−∞ 2πσ 2σ 2 range – x ≤ ξ(t) < +x, or a – x ≤ ξ(t) < a + x, where a
M2 = x̃ 2 = x 2 (t) is a constant, is
+∞ +x 2
x2 −(x − a)2 1 −ξ
= √ exp dx = a 2 + σ 2 (9d) p[−x ≤ ξ(t) < +x] = √ exp dξ (10a)
−∞ 2πσ 2σ 2 2πσ −x 2σ 2
Threshold Trigger Mean value Normalization
x
1−W (x )
x(t ) x
x
x
(a)
Trigger with
upper and lower
∆x threshold Mean value Normalization
x
W (x )
x∆x
x(t ) x
x
(b)
Figure 5. (a) Registration of the probability distribution and (b) probability density.
172 Foundations of Measuring
Table 2. Values of the example.
φ/rad 0.31 0.32 0.30 0.29 0.29 0.30 0.28 0.31 0.31 0.29
t/s−1 2.39 2.40 2.41 2.43 2.40 2.38 2.39 2.41 2.37 2.42
To evaluate this, the probability integral is used, which is As an example, it may have been found by measurement
tabulated in the following form (Jahnke and Emde, 1966): that the length of workpieces having an average value of
x a = 10 cm satisfies a Gaussian distribution and shows a
2 standard deviation using (10a) of σ = 3 mm. What matters
φ(x) = √ exp(−u2 ) du (10b)
π 0 then might be the number of workpieces that lie within
an admissible tolerance range of 10 ± 4 mm. Evaluation
Extracts are shown in Table 1, see Section 4. according to (10) shows that with the values of Table 1,
Some examples with importance in measurement will be some 82% of the pieces are within the tolerance range, and
treated in Section 7. that the remaining pieces lie outside this range.
7 EXAMPLE OF USE IN METROLOGY REFERENCES
Where a number of measured values xr are given, we are Buja, A. (ed) (1991) Computing and Graphics in Statistics,
first interested in obtaining the standard deviation σ of the Springer, New York, Berlin.
distribution as given by Hart, Lotze and Woschni (1997) Hart, H., Lotze, W. and Woschni, E.-G. (1997) Meßgenauigkeit,
As an example, consider the rotation of the arm of Oldnbourg-Verlag, Munchen.
an industrial robot being tested for preciseness of place- Jaffe, R.C. (2000) Random Signals for Engineers using MATLAB
ment to the same position. The measured repeated values and MATHCARD, AIP Press, New York.
of the angle φ/rad and the time t s−1 are as shown in Jahnke, E. and Emde, F. (1966) Tables of Higher Functions,
Table 2. Teubner, Stuttgart.
Using (10a) the standard deviation of the angle is σ = Mendenhall, W. and Sincich, T. (1992) Statistics for Engineering
1.247 × 10−2 rad and of the time σ = 1.826 × 10−2 s. and the Sciences, Dellen Publ. The Free Press, San Francisco,
The mean value of the angular velocity is ω = ϕ/t = CA.
0.3/2.40 = 0.1250 rad s−1 . Woschni, E.-G. (1988) Informationstechnik, Huthig, Heidelberg.
31: Discrete Signal Theory
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
The time sequence that results is given by
1 Discrete Time Sequences 173
∞
∞
2 Sampling Theorem 173 x ∗ (t) = x(t) δ(t − rTs ) = x(rTs ) (1c)
3 Aliasing Errors 175 r=0 r=0
4 Leakage Distortion 176
5 Frequency Domain Sampling 177 The coding of the time sequences x(rT s ) into a dig-
ital signal is treated in Article 33, Coding Theory and
6 Discrete Fourier, Laplace, and Z-Transform 177
its Application to Measurement, Volume 1. The inverse
7 Spectral Analysis of Signal Samples 178 process, to gain x(t) again if x ∗ (t) is known is shown in
8 Examples of Signal Processing 179 Figure 2. This smoothing function can be realized in electri-
References 180 cal systems by means of a capacitor connected to a voltage
Further Reading 181 holding circuit.
A key design issue is to decide how many samples are
necessary to retain all of the information of the signal x(t).
This consideration leads to the sampling theorem, treated
1 DISCRETE TIME SEQUENCES in the next section.
Another important issue is what happens if the sampling
The principle of gaining a discrete time sequence x ∗ (t)
theorem is violated. In this case, aliasing errors appear, as
by sampling the original analog signal x(t) is shown
treated in the relevant section of this article. Here also the
in Figure 1. There, the signal x(t) is multiplied with
signal processing following the sampling stage has to be
a switching-function s(t) leading to the sampled signal
taken into consideration as shown in this section.
sequence x ∗ (t). This signal is then converted into a digital
form using the methods of coding as treated in Article 33,
Coding Theory and its Application to Measurement,
Volume 1. 2 SAMPLING THEOREM
The basic mathematical formulation is
x ∗ (t) = x(t)s(t) (1a) There exist exact mathematic foundations for the sampling
theorem (Johnson, 1989; Oppenheim and Schafer, 1989).
Thus, by means of pulse-amplitude-modulation, the original Here, however, another approach using physical explana-
analog signal is transformed into a digital signal. The
tions is used because this is very suitable for dealing with
switching function may be written as
the problem of aliasing errors.
∞
If the spectral amplitude density X̂(jω) given in Arti-
s(t) = δ(t − rTs ) (1b) cle 27, Signals in the Frequency Domain, Volume 1 is
r=0 limited to the critical frequency ωc as shown in Figure 3
with the sampling time Ts . we get, applying the Fourier transform, the corresponding
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
174 Foundations of Measuring
x ^
X( jw)
x (t )
−w −3wc −wc +wc 3wc w
Figure 3. Band-limited spectral amplitude density.
(a) t
time function x(t) (Woschni, 1988).
s
+ωc
1
1
x(t) = X̂(jω)e jωt dω (2a)
2π −ωc
s (t ) The spectral amplitude density function may be periodi-
cally continued as marked with dashed lines in Figure 3.
T
Expansion of the series of this periodical frequency func-
tion yields
(b) Ts 2Ts 3Ts vTs t +∞ jrπω
1 rπ −
x
X̂period (jω) = x e ωc (2b)
2 ωc r=−∞ ωc
x (t )
In this series, the only values are
x *(t )
rπ
x = x(rTs ) (2c)
ωc
Applying the sampling time (from the sampling theorem)
Ts t rπ
(c) = rTs (2d)
ωc
Figure 1. Principle of gaining the discrete signal sequence by
means of sampling. where Ts = π/ωc = 1/2fc
In reality, the amplitude density is not a band-limited
entity as required by the sampling theorem. It is, therefore,
x necessary to carry out low-pass filtering before sampling
called antialiasing filtering. Figure 4 shows that this leads
to a cutoff error calculated due to system theory, where
Sxx (ω) is the spectral power density of the signal (Woschni,
1988)
∞
ε =2
2 Sxx (ω) dω (2e)
ωc
Ts t This error will be treated in detail together with the aliasing
error in the next section.
Another approach also demonstrates the main points and
leads to the aliasing errors if the sampling theorem is vio-
Figure 2. Smoothing with holding circuit. lated. We are starting with the switching function s(t) due
Discrete Signal Theory 175
Sxx (w)
(a) −wc 0 +wc w
e2(t )
(a) ws / 2 ws w
(b) −ws 0 ws 2ws w
Sxx (w)
(c) −ws −wc 0 +wc ws 2ws w
Figure 4. Explanation of sampling in the frequency domain.
(b) ws /2 ws w
to (1b). This periodic time function means a spectral func-
tion as shown in Figure 4(b). Figure 5. Calculation of the sampling errors in the frequency
domain.
+∞
τ jrωs t
s(t) = e (2f)
Ts r=−∞
Figure 5(a), this error means the power of the cutoff part
The multiplication with x(t) due to (1a) represents an of the spectral power density of the signal Sxx (ω).
amplitude modulation as treated in Article 34, Modula- Figure 5 clearly shows the overlapping of the lower side
tion Theory, Volume 1. Figure 4(a) shows the spectral band falling into the frequency range 0 . . . ωs /2 and leading
amplitude density of the signal x(t), and the amplitude to aliasing error.
modulation has the spectral amplitude density as given ∞
in Figure 4(c). While in the case of a band-limited sig-
ε2 =2 ωs
Sxx (ω) dω (3a)
nal (thick line in Figure 4c), the spectra are separated from
2
each other, in the case marked with dashed lines the spectra
overlap. Here, the sampling theorem is violated and aliasing Because the frequencies of the lower side band are mir-
errors arise. rored, one can suppose the two error components – the
cutoff and the aliasing errors – are not correlated to each
other unless they originate from the same signal source.
3 ALIASING ERRORS This means that the total sampling error is given by the
sum of both components.
Different error definitions are used in measurement. As In measurement, these errors originate if it is not pos-
shown in Article 32, Geometrical Signal Representa- sible to realize a low-pass filtering before sampling. This
tions, Volume 1 and especially in Article 50, Errors in case is typically for sensors with direct digital output, as, for
Signal Systems, Volume 1 the mean-square error intro- instance, in the case treated in Article 33, Coding Theory
duced by Gauss has the advantage that it leads to and its Application to Measurement, Volume 1, because
closed mathematical solutions because of the validity of there a low-pass filtering is not feasible. These errors are
Euclidean geometry. becoming of decreasing importance because the develop-
If the sampling theorem is fulfilled, the cutoff error due ment of microelectronics leads to ever higher pulse frequen-
to the low-pass filtering is given by (2e). As shown in cies and then oversampling is possible (Woschni, 2001).
176 Foundations of Measuring
The truncated sampled signal function may be written as
(Figure 7b)
Sxx (w)
Sxx (w)
x(kTs ) k = 0, 1, 2, . . . , N − 1 (4a)
Greal (jw)
This is the product of the infinite sequence xs (kT s ) and a
rectangular time domain truncation window so (t) as shown
in Figure 7(a).
wLP ws w xs (kTs ) = x(kTs )so (t) (4b)
Figure 6. Mean-value operation after sampling. With the switching function, as treated in Section 1 of this
article, (1a)
Another example that is important in measurement sci- xs (kTs ) = x(t)so (t)s(t) (4c)
ence is the influence of a signal processing after sam-
pling. A typical case arises in surface measurement. Here, The Fourier transform of this yields
instead of the whole information of the surface profile,
only the average roughness is of interest. As demonstrated XN (jω) = X(jω)∗ So (jω)∗ S(jω) (4d)
in Figure 6, low-pass filtering ωLP after sampling is nec-
essary to gain the mean value. This means only half of
With the modulus of the Fourier transform of a rectangular
the value of the sampling frequency due to the sam-
time window being as treated in detail in Section 7
pling theorem is to be used without the generation of
errors (Woschni, 1996a,b). (sinπα)
|So (jω)| = si(πα) = (4e)
πα
The results obtained mean that in Figure 7(a) the time
4 LEAKAGE DISTORTION window leads to the spectrum given by (4e) and the convo-
lution with the spectrum of xs (kT s ) results in the spectrum
Another problem of analog-to-digital conversion is that of demonstrated in Figure 7(b) where a rippling distortion can
windowing. The signal function x(t) has to be time lim- be seen.
ited, for example, by means of a time window instead Figure 8 shows the case of discrete components within
of the infinite function. A time-limited function is used the signal spectrum Xs (jω). Convolution with the function
called the truncation window . This truncation in time So (jω) would be replaced by a si-function due to (4e),
leads to a variation of the spectrum, called leakage dis- which is explained later.
tortion (Thrane, 1979). In general, the amount of leakage distortion depends on
As shown in Figure 3, in the time domain a periodic time the length of the truncation window in comparison to the
function appears leading to cutoff errors known as leakage sampling interval as the following example shows.
errors. The main statements are demonstrated in Figure 7. A typical example is now given (Hart, Lotze and Wos-
chni, 1997). A sinusoidal signal may be sampled with a
normalized time truncation window Tfc . Using the methods
so(t ) So( jw)
given above the relative mean-square error F is calculated
t
(a)
as a function of k – the sampling time Ts due to the sam-
−Ts To = NTs t w pling theorem related to a shorter real sampling time Ts –
2
x(kTs) XN ( jw)
Xs( jw) So( jw) XN ( jw)
× =
(b)
w w w w
t
Actual spectrum Transform of window Transform
Figure 7. (a) Truncation window and its transform; (b) truncated
sequence and its transform. Figure 8. Leakage distortion of discrete components.
Discrete Signal Theory 177
k−Ts/T s′ x(kTs)
xp (kTs)
1 102 104 105 T0 = 2 T
1
... ...
k k n
(a) x (t )
F
10−2 x (kTs) xp(kTs)
T0
... ...
fεT −1
k k n
10−4 (b)
fcT −10
2
fε T −10 Figure 10. (a) Width of truncation window equal to two periods
fε T −10
3 of x(t); (b) width not equal to two periods.
4
fcT −10
10−6
periodic repetition of the sample function xp (kT s )
Figure 9. Errors as a function of the relative sampling time k
with the relative truncation time as parameter. xp (kTs ) = x(t)∗ so (t)∗ sf (t) (5c)
Figure 9. It is evident that because of the leakage distortion So we get, in principle, the same relation between sampled
shorter sampling times are necessary to avoid errors. spectrum and time function as shown in Figure 7(b), only
now with a periodically repeated time function.
There exists an important case, found especially in mea-
surement, when x(t) is a periodic signal. Here, two dif-
5 FREQUENCY DOMAIN SAMPLING ferent situations may be distinguished, as demonstrated
in Figure 10. If the width of the truncation window con-
As treated in the last section, a truncated time sequence tains an integral number of periods of the signal, then no
originates an infinite spectrum X(jω). Figure 7(b) demon- error arises as Figure 10(a) shows. Otherwise errors appear
strates that a periodic spectrum appears. For computer (Figure 10b).
application, this frequency function has to be sampled and Other applications of frequency domain sampling con-
limited to a finite number of samples. This procedure, there- cern the design of digital filters (Johnson, 1989).
fore, is the same as the one realized in the last section;
only time and frequency are now exchanged. Thus, here
also the same problems of aliasing errors and leakage dis- 6 DISCRETE FOURIER, LAPLACE, AND
tortion arise.
Owing to (4a–d), X(kjωs ) may be the sampled frequency
Z-TRANSFORM
function gained from the original function X(jω) by means
of a sampling function Sf (jω) in the same way as in the As treated in Article 27, Signals in the Frequency
time domain in Section 4 Domain, Volume 1, between the time and the frequency
domain there exists a relationship known as the Fourier
X(kjωs ) = X(jω)Sf (jω) (5a) transform. In the case of analog continuous signals, use of
this transform yields (from (6b, c) of Article 27, Signals
This infinite frequency function has to be frequency trun- in the Frequency Domain, Volume 1)
cated to N numbers of samples by means of a frequency +∞
domain truncation window So (f ) leading to the relation X̂(jω) = x(t)e−jωt dt = F {x(t)} (6a)
equivalent to (4c) −∞
+∞
1
XN (kjωs ) = X(jω)So (jω)Sf (jω) (5b) x(t) = X̂(jω)ejωt dω = F −1 {X̂(jω)} (6b)
2π −∞
This equation may be transformed to the time domain where In technical systems, it is reasonable to assume that the
the multiplication will be a convolution and x(t) will be a signals start at t = 0. Then, under the limit (6a) is zero
178 Foundations of Measuring
(so-called unilateral Fourier or Laplace transform) and Frequency Domain, Volume 1.
∞ ∞
X̂(jω) = x(t)e−jωt dt (6c) X∗ (s) = X(rTs )e−rsTs (6k)
0
r=0
∞
As shown in Section 1 of this article, instead of the time
function x(t) by means of sampling with Ts , the discrete X(rTs ) = X∗ (s)ersTs (6l)
r=0
time sequence
With the substitution
x ∗ (t) = x(t)δ(t − rTs ) = x(rTs ) (6d)
Z = esTs (6m)
originates leading to the infinite sequence due to (6c)
the Z-transformation results that is especially used in
∞
the theory of digital signal processing and filters – see
X̂∗ (jω) = x(rTs )e−jωrTs (6e) Section 8 of this article.
r=0
In Section 4, it is demonstrated that this function has to 7 SPECTRAL ANALYSIS OF SIGNAL
be time truncated. This means a limitation to N samples SAMPLES
applies, that is, the function x(t) is replaced by a periodic
one. Instead of (6b), it follows A periodic rectangular signal may be considered as shown
in Figure 11. It is built by means of rectangular truncation
1 ∗
N−1
windows in the time domain.
X(r) = X (k)e j(2π/N)kr (6f)
N k=0 Using the method of Fourier spectrum as treated
in Section 2 of Article 27, Signals in the Frequency
The Fourier coefficients are Domain, Volume 1 one obtains only cos functions with
the coefficients
N−1
τ
τ
X∗ (k) = X(r)e j(2π/N)kr (6g) 4 2 4a
An = a cos nω0 t dt = sin nω0
k=0 T 0 T nω0 2
2aT τ
With the definition = si nω0 (7a)
T 2
WN = e−j(2π/N) (6h) or the complex Fourier coefficients
+τ/2
The discrete Fourier transformation is (Oppenheim and 2a τ
X̂(jnω0 ) = a e−jnω0 t dt = sin nω0
Schafer, 1989) −τ/2 nω0 2
aτ τ
N−1 = si nω0 (7b)
T 2
X∗ (k) = X(r)WNkr (6i)
r=0 The spectrum consists of harmonics with an envelope
N−1 well known as the si(x) function that is demonstrated in
1
X(r) = X∗ (k)WN−kr (6j)
N k=0
Using this definition, the algorithms of Cooley and Tuckey a
(1965) now well known as Fast Fourier Transform (FFT) x (t )
are introduced as treated in Section 6 of Article 29, Rela- τ
tionship Between Signals in the Time and Frequency
Domain, Volume 1.
−T /2 −τ/2 +τ/2 +T /2 t
Using the same procedure, the Laplace transform may t=0
be applied to discrete signals leading to the relations – see
Article 37, Relationship Between Systems in the Time Figure 11. Periodic rectangular signal.
Discrete Signal Theory 179
Figure 12. The smaller the pulse duty factor τ /T is the and their advantages are treated in (Johnson, 1989) or
more harmonics that fall into the region up to the first zero (Oppenheim and Schafer, 1989).
point. The method of periodograms may be useful here (Oppen-
Because of the existence of the side maximums, the heim and Schafer, 1989).
required spectrum is relatively wide. On the other hand, a
truncation of the spectrum leads to a time function of si(x)
form instead of the rectangular one of the original signal. To
8 EXAMPLES OF SIGNAL PROCESSING
avoid these disadvantages, other time truncation windows
with better performance, for example, smaller spectra are
used (Johnson, 1989) or (Oppenheim and Schafer, 1989). To illustrate the main methods of signal processing using
For instance, the bell-shaped window leading to the bell- the results of the investigations of this article, consider first
shaped pulse is as shown in Figure 13. From this the integration of a signal x(t).
The first approximation is a rectangular one described by
x(t) = ae−β
2 2
t
(7c) the algorithm
y(t) = y(t − TA ) + T x(t) (8a)
has a spectrum – using Fourier transform as treated
in Section 4 of Article 27, Signals in the Frequency
Domain, Volume 1 Using the discrete Fourier, Laplace, or Z-transform, the
frequency response or Z-transform is
+∞
e−β t e−jωt dt
2 2
X̂(jω) = a 1
GR1 (jω) = T (8b)
t=−∞
1 − e−jωT
ω2 +∞
− 1
e−(βt−jω2β) dt
2
= ae 4β 2
GR1 (s) = T (8c)
−∞ 1 − e−sT
ω 2 ∞ √ ω2 z
2a − 4β −x 2 a π − 4β GR1 (z) = T (8d)
= e 2 e dx = e 2 (7d) z−1
β 0 β
The next better approximation is a trapezoidal one with the
This is also a bell-shaped spectrum having no side max-
algorithm
imums and therefore leading to signals with a smaller
required frequency range. Other time truncation windows T
y(t) = y(t − T ) + [x(t) + x(t − T )] (8e)
2
An
leading to the frequency functions
a
T 1 + e−sT
GR2 (s) = (8f)
a 2 1 − e−sT
2
T z+1
GR2 (z) = (8g)
−w0 +w0 w
2 z−1
6w0
2w0 4w0 T 1 + e−jωT
GR2 (jω) = (8h)
2 1 + e−jωT
Figure 12. Spectrum of the periodic rectangular signal.
The frequency response of the ideal integration is
a 1
G(jω) = (8i)
jω
x (t ) As Figure 14 shows, the trapezoidal approximation is better
because here only an amplitude error and no phase error
a = 0.367a
e appears. On the other hand, we have to pay for this
t = −1/b 0 t = 1/b t
advantage by using more processing time!
The mean-square error in using the results of
Figure 13. Bell-shaped pulse. Article 50, Errors in Signal Systems, Volume 1 for the
180 Foundations of Measuring
wT = ∞ +j
+r −r
1
−j
Figure 14. Frequency responses of integration algorithms; wT = x wT = 0
ideal; - - - - rectangular.; −·−·−· trapezoidal.
−G −1 0 +1 +2 +G
rectangular integration and a model signal S(x) = So /(1 + Figure 15. Frequency responses of prediction algorithms;
ω/ω0 ) (Woschni, 1996a,b) is given by ideal; - - - - first approx.; −·−·−· second approx.
x 2
T So 1 T
ε2 =2 − dω (8j) is, the smaller the error – the longer the time needed
0 ω jω 1 − e−jωT
1+ for processing. Owing to the development of ever faster
ω0 microelectronic circuitry having increased processing per-
formance good approximations are possible with smaller
Another example is the prediction problem. The ideal errors.
frequency response reads as follows: Finally, it may be hinted at the limitation of the frequency
range from f = 0 to f = fs /2 due to the Sampling theorem.
GR (s) = esT , GR (z) = z, GR (jω) = e+jωT (9a) Furthermore, the processing error is a function of the word
length s in bits of the code used. Between the signal-to-
The first approximation is the linear continuation of x(t), noise ratio S/N in dB and the word length s in bits there
for example, the algorithm is exist the general relation (Woschni, 1988)
y(t) = x(t) + x(t) − x(t − Ts ) = 2x(t) − x(t − T ) (9b) S/N = 20 log 1/2s = −20 s/2 log 10 = 6 s (9j).
The frequency functions are, using the same method as
before REFERENCES
GR1 (jω) = 2 − e−jωT (9c) Cooley, J.W. and Tuckey, J.W. (1965) An Algorithm for the
−sT
Machine Calculation of Complex Fourier Series. Mathematical
GR1 (s) = 2 − e (9d) Computation, 19, 279–301.
GR1 (z) = 2 − z−1 (9e) Hart, H., Lotze, W. and Woschni, E.-G. (1997) Meßgenauigkeit,
Oldenboug-Verlag, Munchen 5 (p. 112).
The next better approximation includes the two last values Johnson, J.R. (1989) Introduction to Digital Signal Processing,
Prentice Hall, Englewood Cliffs, NJ.
with the algorithm
Oppenheim, A.V. and Schafer, R.W. (1989) Discrete-Time Signal
Processing, Prentice Hall, Englewood Cliffs, NJ.
y(t) = 3x(t) − 3x(t − T ) + x(t − 2T ) (9f)
Thrane, N. (1979) The Discrete Fourier Transform and FFT
Analysers, Bruel and Kjaer Technical Review, 1, 3–25.
leading to the frequency functions
Woschni, E.-G. (1988) Informations Technik, Huthig, Heidelberg.
Woschni, E.-G. (1996a) Signal Processing after Sampling –
GR2 (s) = 3 − 3e−sT + e−2sT (9g) Today’s Problems and Solutions in Future, Proceedings of
GR2 (z) = 3 − 3z−1 + z−2 (9h) ISMTII’96, Hayama Kanagawa Pref., Japan.
Woschni, E.-G. (1996b) Signal Processing after Sampling –
GR2 (jω) = 3 − 3e−jωT s + e−j2ωT (9i) Today’s Problems and Solutions in Future, Proceedings of
ISMTII’96, Hayama Kanagawa Pref., Japan.
Figure 15 shows the results where the general princi- Woschni, E.-G. (2001) Measurement in Robotics – Trends due
ple is evident. In general, the better the algorithm, that to the Development of Microelectronics Proceedings, 10th
Discrete Signal Theory 181
International Workshop on Robotics RAAD 2001, Techn. Uni- Karu, Z.Z. (1995) Signals and Systems Made Ridiculously Simple,
versitat, Wien. ZiZi Press, Huntsville, AL.
Miller, M.J. (1982) Discrete Signals and Frequency Spectra, in
FURTHER READING Handbook of Measurement Science, Vol. 1 (ed. P.H. Syden-
ham), Wiley, Chichester.
Buck, J.R., Daniel, M.M. and Singer, A.C. (2001) Computer Williamson, D. (1999) Discrete-Time Signal Processing: An Alge-
Explorations in Signals and Systems Using MATLAB, 2nd edn, braic Approach, Springer-Verlag, Berlin.
Prentice Hall, NJ. Woschni, E.-G. (1992) The Importance of Estimation and Approx-
Grover, D. and Deller, J. (1998) Digital Signal Processing and imation Methods in System Theory. Cybernetics and Systems,
the Microcontroller, Prentice Hall PTR, NJ. 23, 335–343.
Hayes, M.H. (1998) Schaum’s Outline of Digital Signal Process- Ziemer, R.E., Tranter, W.H. and Fannin, D.R (1998) Signals and
ing, McGraw-Hill Trade, New York. Systems: Continuous and Discrete, Prentice Hall, NJ.
32: Geometrical Signal Representations
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
norm of the difference between two signal vectors x and y
1 Representation of Signals in Euclidean Space 182 as follows:
2 Representation of Signals in Non-Euclidean
1
Space 183
n 2
3 Hamming Distance 183 d(x, y) = x − y = |xv − yv |2 (1c)
v=1
4 Representation with Codegraphs 184
References 184 The scalar product
Further Reading 185
n
x×y= xv yv (2a)
v=1
1 REPRESENTATION OF SIGNALS IN
EUCLIDEAN SPACE can be used to write the angle α between the two vectors
in the following way:
Signals with n components x1 , . . . , xn can be represented
by a signal vector x in the n-dimensional space: xy
cos α = (2b)
xy
x = (x1 , x2 , x3 , . . . , xn ) (1a)
Figure 1 explains the relations.
where the end point of the vector determines the corre- For continuous analog signals x(t), which are defined in
sponding signal. the range a ≤ t < b, one can accordingly indicate a norm:
For analog signals, use is made of Euclidean space,
wherein the Pythagorean theorem applies. Thus, for the 12
b
magnitude of the vector, also called the norm in geometry, x = x 2 (t) dt (3a)
one obtains a
n 1
2
Physically, it represents the square root of the energy of the
x = xr2 (1b)
signal. A special Hilbert space is thus defined (Blumenthal,
r=1
1961). The distance between two signals corresponds to the
The various components are then the projections onto the root mean square error
different axes of the n-dimensional space, which is called
the signal space. b 12
The notion of the distance d between two or more signals d(x, y) = x − y = |x(t) − y(t)| dt 2
(3b)
is of great practical importance. This distance results as a a
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Geometrical Signal Representations 183
2 according to the norm of this space:
n
x = xv (4b)
v=1
The most important discrete signals are binary signals, that
x2 is, signals in which the individual components can assume
d(x→ →
,y
) the values 1 and 0 only. For such signals, the signal space
y2
constitutes an n-dimensional hypercube having the edge
length 1, where the various edges are occupied by possible
signals only.
→
x →
a y Figure 2 shows such signal words with one, two, and
three bits. Obviously, the representation in non-Euclidean
x1 y1 1 space with the norm according to (4b) has the advantage
that the distance d indicates the number of digits by which
Figure 1. Two-dimensional model of the signal representation in
Euclidean space.
two signal words differ from each other. Similarly defined
minimum distance in a signal alphabet is called the Ham-
ming distance and constitutes an important characteristic
It is often used in measurement technology and cyber-
value for the investigation into a system’s sensitivity to
netics as a measure for the error – see Article 50, Errors
noise (Peterson, 1962). To investigate distances between
in Signal Systems, Volume 1.
signals, use is also made of distance matrices as shown in
the next section. In the analog-to-digital conversion of sig-
2 REPRESENTATION OF SIGNALS IN nals – see Article 132, Data Acquisition Systems (DAS)
NON-EUCLIDEAN SPACE in General, Volume 3 and Article 139, Analog-to-Digital
(A/D) Converters, Volume 3, a transform between the cor-
The above representation in the Euclidean space is suit- responding signal spaces takes place.
able for analog signals. Use is made of a representa-
tion in non-Euclidean signal space for discrete signals
whose importance is constantly increasing as digital sys-
3 HAMMING DISTANCE
tems replace evermore the analog ones. In this space, the
distance between two vectors x, y is defined to be the sum of
the differences of the individual components (Blumenthal, The distance between two code words in Figure 1 is the
1961) minimum length that can be taken only along the axis
x1 , x2 , x3. Another possibility to obtain the distances uses
n
d(x, y) = x − y = |xv − yv | (4a) the so-called distance matrix as demonstrated for the signals
v=1 of Figure 2 in Table 1.
x1 x1 x1
1 10 11 100 110
101 111
0 00 01 000 010
x2 x2
(a) (b)
001 011
x3
(c)
Figure 2. Representation of a binary signal in the signal space: (a) one bit; (b) two bits; (c) three bits.
184 Foundations of Measuring
Table 1. Distance matrices for the signals of Figure 2. a) n = 1; 0
n=1 =2 =3
b) n = 2; c) n = 3. 000 =^ 0
a) b) 1
001 =^ 1
1 0 00 10 01 11
010 =^ 2
0 1 00 0 1 1 2 011 = 3
0 1 0 10 1 0 2 1
01 1 2 0 1 100 =^ 4
11 2 1 1 0
101 =^ 5
c) 110 =^ 6
000 001 010 011 100 101 110 111 111 =^ 7
9
000 0 1 1 2 1 2 2 3
(a) (b)
011 1 0 2 1 2 1 3 2
010 1 2 0 1 2 3 1 2
Figure 3. Representation as a codegraph: (a) signal with m = 10
011 2 1 1 0 3 2 2 1
possible states and (b) binary signal with the length n = 3.
100 1 2 2 3 0 1 1 2
101 2 1 3 2 1 0 2 1
110 2 3 1 2 1 2 0 1 Equation (6) – the application of the principle of an addi-
111 3 2 2 1 2 1 1 0 tional redundancy (Woschni, 1988; Shannon, 1948; Gold-
man, 1953) – is the key to designing data transmission lines
with a very small error probability although the original
Because the distance d between two words is equal to transmission line is of low performance.
the number of bits by which the two words differ from each
other, the number of the possible code words between them 4 REPRESENTATION WITH
is d − 1. If in an alphabet not all possible words are used,
the minimal distance in this alphabet is called the Hamming
CODEGRAPHS
distance dmin or dHamming . This means, for instance, for
For an easily understandable explanation of the signal
dmin = 2, that between each word of the alphabet, one
structure, the codegraph may be used. On the basis of
not used word is situated. From this fact it follows that
the results of the graph theory, the m possible states
1-bit error leads to a not used word. So bit errors of first (symbols) of a signal are shown as an assembly according
order fe1 can be noticed. If there are two vacant words to Figure 3(a). Figure 3(b) shows the application of this
between two used words – it means that dmin = 2 – errors type of representation to a binary signal having three digits.
of second order fe2 can be noticed and so on, leading to Where the individual code words are of the same length,
the general relation between the order n of noticed errors they can be separated one from the other by counting. No
and the Hamming distance characters are required for the separation of the individual
words; the code is irreducible. Obviously, this is the case
fen = dmin − 1 (5) if the end points of the codegraph are occupied by code
words only.
For more details, especially concerning code theory and
Similar reflections lead to the relation between the order
its application to measurement, see Article 33, Coding
n of correctable errors fcn and the Hamming distance dmin .
Theory and its Application to Measurement, Volume 1.
For instance, with dmin = 2 in Figure 2, only the words
000 and 111 are used. So errors of first and second order
lead to unused words and can be noticed due to (5). But REFERENCES
the probability that the first-order error word belongs to
the original word is very much higher than for the second- Blumenthal, L. (1961) A Modern View of Geometry, Freeman, San
error word. Therefore, a correction can be done to correct Francisco, CL.
the words with first-order error, the order of correction is Goldman, S. (1953) Information Theory, Prentice Hall, New York.
1. From this reflection there follows the general relation Peterson, W.W. (1962) Error Correcting Codes, MIT Press,
Cambridge.
(dmin − 1) Shannon, C.E. (1948) A Mathematical Theory of Communication.
fcn = (6) Bell System Technical Journal, 27, 379–423.
2
Geometrical Signal Representations 185
Woschni, E.-G. (1988) Informations Technik, Huthig, Heidel- Hsu, H.P. and Hse, H.P. (1995) Schaum’s Outline of Signals and
berg. Systems, McGraw-Hill Trade.
Karu, Z.Z. (1995) Signals and Systems Made Ridiculously Simple,
ZiZi Press, Huntsville, AL.
FURTHER READING Oppenheim, A.V., Willsky, A.S. and Hamid, S. (1996) Signals
and Systems, Prentice Hall, NJ.
Hamming, R.W. (1986) Coding and Information Theory, Prentice Ziemer, R.E., Tranter, W.H. and Fannin, D.R (1998) Signals and
Hall, NJ. Systems: Continuous and Discrete, Prentice Hall, NJ.
Coding Theory and its Application to
33:
Measurement
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
in Article 34, Modulation Theory, Volume 1. The overall
1 Coders and Decoders in Measurement function of this communication system is to transform
Systems 186 the signal where there are several information sources by
2 Coding Theory 186 multiplexing them onto the same transmission channel,
or to make the signal reasonably immune to unwanted
3 Binary Codes Used in Measurement 187
disturbances.
4 Absolute Optical Encoders and Codes Used 188 The modulated or coded signal is transmitted by the chan-
5 Incremental Encoders and LSB 189 nel or processed by the processor. This channel may be a
6 Error Correction by Redundant Codes 189 microwave or uhf relay link, a wire or cable transmission,
7 An Example of Error Correction Code or an optical fiber, as are commonly used in measurement
Application 190 systems. Noise arising in the channel is considered to be
References 190 additive.
Further Reading 190 In measurement, the signal is usually processed by a
microprocessor containing a memory. As in the case of
an analog system – still favored in some measurement
applications – the behavior of this system may be described
1 CODERS AND DECODERS IN by a transfer function provided the computer program is a
linear one (Woschni, 1988).
MEASUREMENT SYSTEMS
The next link in the serial chain is the decoder or
demodulator, whose task is to reconstruct an estimate
The task of measurement as a part of communication is of the original signal as correctly as possible. In the
the transmission of a message from an information source sense of describing the signal as a vector in the signal
to a receiver, as shown in Figure 1. The output of this space (see Article 32, Geometrical Signal Representa-
information source may be a digital or a continuous signal. tions, Volume 1), it means that the distance between the
In measurement, it is the unknown quantity that is to be output and input signals should be kept to a minimum.
measured: the output is generated by a random mechanism Many similarities link this problem to the problems of char-
having a probabilistic nature. Otherwise, the signal would acter recognition (Finkelstein, 1976).
be completely known and there would be no need to
obtain the output by means of measurement. The signal
coming from the information source is the input signal 2 CODING THEORY
of the encoder, modulator, or transmitter. If the signal is
digital, the term encoder is used, while for analog signals As treated in Section 1 of this article, the first subsystem
the term modulator is used. This subsystem is treated of a communication system is the encoder, converting the
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Coding Theory and its Application to Measurement 187
Signal
Encoder
Information Channel Decoder Information
modulator
source processor demodulator receiver
transmitter
Disturbance
noise
Figure 1. Communication system.
input signal into a series of code words. The task of this used in communication, giving an equation for the optimum
coding is the adaptation (interfacing) of the information length of the source code words lk :
source to the channel or processor.
In communication systems, the redundancy-diminishing lbpk lbpk
− ≤ lk < − +1 (1c)
(optimal) source coding, having the purpose of economiz- lbM lbM
ing the time needed for communication, often plays an
where pk is the probability of the appearance of the kth
important role. In measurements, however, security of the
code word. Some important codes, including those applied
message against disturbances is the most needed criterion.
in measurement, are now considered.
Here, therefore, error-detecting or error-correcting codes are
applied (Peterson, 1962).
For the representation of codes, geometrical descriptions 3 BINARY CODES USED IN
or codegraphs are used in the n-dimensional space
MEASUREMENT
(see Article 32, Geometrical Signal Representations,
Volume 1). Many coding and decoding theorems exist. The
There are a lot of binary codes used in several fields of
decoding theorem deals with the problem of identifying a
measurement. The data-transmission error-correcting codes
code word by the receiver. For this purpose, the decoder
are treated in Section 6 of this article.
compares the incoming code words with the words of the
The simplest code is the counting code, mostly seen
code alphabet deciding which code word the transmitter
in decimal counting. Today, this easily learnable code is
has sent. In the case in which the end of a code word is
displaced in machines that use the binary-coded decimal
not marked by a special symbol, only the endpoints of the
notation because of the smaller number of bits. As an
codegraph may be filled with a code word, otherwise a part
example, in Figure 2, the 1-out-of-10 code is presented.
of a code word would be another code word. The equation
For manual data coding (data input), a particular form of
that guarantees this is
the binary code
K
M −lk ≤ 1 (1a) Zbin = An 2n + An−1 2n−1 + · · · + A1 21 + A0 20
k=1 = An An−1 · · · A1 A0 (2a)
where M is the number of symbols and lk is the length of the
kth code word. In (1a), the equality sign represents the most the binary-coded decimal system or BCD code, is used.
advantageous case without code redundancy. Otherwise, the Here, digit-by-digit, the decimal number is converted into
factor c in the equation
9 8 7 6 5 4 3 2 1 0
K 0 0 0 0 0 0 0 0 0 0 1
c M −lk = 1 (1b) 1 0 0 0 0 0 0 0 0 1 0
2 0 0 0 0 0 0 0 1 0 0
k=1 3 0 0 0 0 0 0 1 0 0 0
4 0 0 0 0 0 1 0 0 0 0
is a measure of the code redundancy. 5 0 0 0 0 1 0 0 0 0 0
6 0 0 0 1 0 0 0 0 0 0
This decoding theorem is of great importance in mea- 7 0 0 1 0 0 0 0 0 0 0
8 0 1 0 0 0 0 0 0 0 0
surement because codes with redundancy are often used, 9 1 0 0 0 0 0 0 0 0 0
see Sections 6 and 7 of this article. The theorem of optimal
coding plays a great role in redundancy-diminishing coding Figure 2. 1-out-of-10 code.
188 Foundations of Measuring
Table 1. BCD code in binary and excess-three notation. y3 y2 y1 y0
Decimal Binary-coded Excess-three
code decimal code notation
First Second First Second
x0
tetrad tetrad tetrad tetrad
x3 x2 x1 x0 y3 y2 y1 y0
0 0000 0000 0000 0011 x1
1 0000 0001 0000 0100
2 0000 0010 0000 0101
3 0000 0011 0000 0110
4 0000 0100 0000 0111
5 0000 0101 0000 1000 x2
6 0000 0110 0000 1001
7 0000 0111 0000 1010
8 0000 1000 0000 1011
OR gate
9 0000 1001 0000 1100
10 0001 0000 0100 0011 x3 AND gate
11 0001 0001 0100 0100
INVERT
20 0010 0000 0101 0011
50 0101 0000 1000 0011
Figure 3. Circuit for conversion of the excess-three code to the
51 0101 0001 1000 0100
binary code.
76 0111 0110 1010 1001
99 1001 1001 1100 1100
used especially with optical absolute encoders as described
in the next section.
the binary code. For each digit, four bits, the so-called
In data transmission, the ISO-CCIT code No. 5 or the
tetrad, is necessary (Table 1).
teletype CCITT code No. 3 are used in combination with
In binary notation, the numbers complement necessary
error-correcting codes as treated in Section 6.
for subtraction in a computer, sometimes leads to a nonexis-
tent code word. The complement of 3 (=0011), for instance,
is 1100, which does not exist in the BCD code (Table 1). 4 ABSOLUTE OPTICAL ENCODERS AND
This disadvantage is avoided by use of the notation of the
BCD code, also presented in Table 1. To each decimal num- CODES USED
ber, 3 is added. This code is well known as the excess-three
code or the Stiebitz code. There exist two types of encoders – the absolute and the
As an example, the conversion of the excess-three code incremental encoders (Carstens, 1993; Pallas-Areny and
to the binary code is now considered. With the relationships Webster, 2001). With the absolute encoders, each input has
between x and y shown, Table 1 yields in Boolean algebra a unique output, while, with the incremental codes, only the
form difference between two inputs is given in the output (see
also next section).
x0 = ȳ0 (2b) The principle of an optical encoder is seen in the dig-
ital shaft encoder demonstrated in Figure 4. It is used to
x1 = (y0 × y1 ) + (ȳ0 × y1 ) (2c) transform the rotational angle directly into a computer rec-
x2 = (ȳ1 × ȳ2 ) + (ȳ0 × ȳ2 ) + (y0 × y1 × y2 ) (2d) ognizable signal. The same principle may be used with a
linear encoding scale to transduce length. For them to work
x3 = (y2 × y3 ) + (y0 × y1 × y3 ) (2e) correctly, it must be emphasized that the sampling theorem
has to be fulfilled in operation. Otherwise, aliasing errors
where occur, because normally it is not possible to realize the
× is the logical AND operation necessary low pass before sampling.
+ is the logical OR operation To minimize the errors when counting from one segment
to the other, it is appropriate to have a code that differs
The realization of these logic equations leads to the only in one digit step from one segment to the next.
digital logic circuit of Figure 3. This group of codes is the reflected codes, arising by
Of importance in measuring systems are the Gray code counting at first forward, that is, from 0 to 9, and then
and the improved variation of this code, the Glixon code backward from 19 to 10. One of this group is the Gray code.
Coding Theory and its Application to Measurement 189
Table 2. Formation of the Gray code.
Decimal number 0 1 2 3 4 5 6 7
Binary number 000 001 010 011 100 101 110 111
Shifted binary number 000 001 010 011 100 101 110 111
0000 00011 0110 0101 1100 1111 1010 1001
Gray code 000 001 011 010 110 111 101 100
90°
112.5° 67.5°
135° 45° l
l′
157.5° 22.5°
Figure 5. Incremental encoder for coding the length l.
180° + 0°, 360°
Counter
fi T z
202.5° 337.5° Figure 6. Measuring of frequency f or time T by counting z.
225° 315°
the detector. So counting is made relative to the datum of
242.5° 292.5° the starting position, but the starting datum itself is not
270°
known. If a count is lost or gained due to noise and so
Figure 4. Coding disk (Gray code): code is read radially from on, the output will be in error from that point on. These,
outside inward. however, have been made to be very reliable in application.
The same principle can be used either to code the length
Table 3. Glixon code. as shown in Figure 5 or the angle shown in Figure 4.
Decimal number Glixon code If a pulse frequency fi is switched on at time T as shown
in Figure 6, either the time or the frequency can be coded
0 0000
in the counter z. The relation to get z is
1 0001
2 0011
3 0010 z = fi T (3a)
4 0110
5 0111 Because it is possible that one more line or pulse is
6 0101 counted as Figure 5 demonstrates – the so-called least sig-
7 0100
8 1100
nificant bit (LSB), the relative error ε is (Woschni, 1988)
9 1000
1
ε= (3b)
fi T
It is obtained from the binary code as shown in Table 2. The
advantage of this code is that any two code words following More details are given in Carstens (1993) and in Pallas-
each other always have unit distance between them, that is, Areny and Webster (2001).
the two code words differ in one digit only. For this reason,
this code is often used in measurement for encoding disks
or linear encoding scales (Figure 4). The disadvantage of 6 ERROR CORRECTION BY
a distance greater than 1 occurring between 9 and 10 is REDUNDANT CODES
avoided by using the improved Glixon code (Table 3).
During data input, transmission, or processing procedures,
errors may arise, as has already been mentioned in the
5 INCREMENTAL ENCODERS AND LSB design of the Gray code in the previous section. Error-
detecting and error-correcting codes having additional code
Less expensive to make is the incremental encoder shown redundancy have been designed. The Hamming distance
in Figure 5. Here, identical lines are counted as they pass dmin , that is, the minimum distance between two code words
190 Foundations of Measuring
of an alphabet, has to be maintained (Peterson, 1962), for Hancock and Thorn, 1989). A typical example is given in
error-detecting codes with the degree fd for errors to be the next section.
detected,
dmin = fd + 1 (4a) 7 AN EXAMPLE OF ERROR
CORRECTION CODE APPLICATION
and for error-correcting codes, if fc is the degree of error
to be corrected As an example of error correction by redundant codes, a
digital data transmission with a bit error rate (BER) of 10−2
dmin = 2fc + 1 (4b) is assumed: this corresponds to a normal telephone, hard
wire line. The line should have a capacity of 2400 bps, a
small value by modern speeds of communication.
If correction to the degree fc∗ < fc only is used, it is
We are first interested in the number of bit errors per
possible to detect additional errors up to degree fd∗ :
year that arise in such a link. With 365 days, there will be
756 864 000 bit errors per year!
fd∗ = dmin − 2fc∗ − 1 (4c) Now we ask what is the Hamming distance dH necessary
to get no more than one bit error in one year. If the bit error
Details concerning the Hamming distance as the proce- rate is to be 10−10 , there will still be 756 864 bit errors per
dure of error detection and error correction in the signal year. The necessary bit error rate has to be 10−12 leading
space are treated in Article 32, Geometrical Signal Rep- to 0.07564 bit errors in one year. From (4b) it follows that
resentations, Volume 1. fc = 5 a Hamming distance of dH = 2 × 5 + 1 = 11.
A very simple, but often used method for data input, is If the capacity of the line were to be 48 kbps, the bit
the addition of parity bits or parity check. This additional error rate necessary would be 10−14 because the value
bit is chosen in such a way that the sum of all digits (the 10−12 would lead to 20 × 0.07564 = 1513 bit errors per
so-called weight of the code) is either an even or an odd year. Now fc has to be six, which means that due to (4b),
number. Table 4 presents the BCD code with parity check, the Hamming distance has to be 13. The realization of the
detecting all errors with odd weight (1, 3, . . .). necessary Hamming distance is made by means of checking
Other and more complicated error-detecting or error- digits in addition to the information digits (Peterson, 1962)
correcting codes are the selector code (w-out-of-n code), and (Hamming, 1986).
the Hamming codes produced by feedback shift regis-
ters (Hamming, 1986), the recurrent or cyclic code (Peter-
son, 1962), and codes with block protection (Heise and REFERENCES
Quattrocchi, 1995).
Carstens, J.R. (1993) Electrical Sensors and Transducers, Re-
If the transmission is bidirectional, it is possible to correct gents/Prentice Hall, Englewood Cliffs, NJ.
wrong data blocks by means of retransmission (Sydenham, Finkelstein, L. (1976) Paper CTH 301, Preprints IMEKO Con-
gress, Institute of Measurement and Control, London.
Table 4. BCD code with parity check. Heise, W. and Quattrocchi, P. (1995) Informations- und Co-
dierungs Theorie, Springer, Berlin.
First Second Parity Decimal 1 Hamming, R.W. (1986) Coding and Information Theory, Prentice
tetrad tetrad check number Hall, NJ.
0000 0000 1 0 1 Pallas-Areny, R. and Webster, J.G. (2001) Sensors and Signal
0000 0001 0 1 1 Conditioning, Wiley, New York.
0000 0010 0 2 1 Peterson, W.W. (1962) Error Correcting Codes, MIT Press,
0000 0011 1 3 3 Cambridge.
0000 0100 0 4 1
0000 0101 1 5 3 Woschni, E.-G (1988) Informations Technik, Huthig, Heidelberg.
0000 1001 1 9 3
0001 0000 0 10 1
0001 0001 1 11 3 FURTHER READING
0010 0000 0 20 1
0101 0000 1 50 3 Wilson, S (1996) Digital Modulation and Coding, Prentice Hall,
1001 1001 1 99 5 NJ.
34: Modulation Theory
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
the carrier oscillation having a carrier frequency 0 gives
1 Survey of Modulation Methods 191 sin θ (t)
sin(0 t + φ) = U
u(t) = U (1a)
2 Amplitude Modulation 192
3 Frequency or Phase Modulation 194 Thus, the input signal is caused to vary by the modulation
References 195 is altered by the input signal
signal x(t). If the amplitude U
Further Reading 195 x(t), amplitude modulation (AM) results:
= f (x(t))
U (1b)
1 SURVEY OF MODULATION METHODS Angle modulation is generated by using x(t) to vary the
argument of u(t):
A special form of processing of a source signal to suit a φ(t) = f (x(t)) (1c)
transmission channel is that of modulation. Transmission
of signals together from several sources all over a single As θ = 0 t + φ, two kinds of angle modulation can be
channel can be accomplished using frequency-division or implemented.
time-division multiplexing systems. Frequency modulation (FM) occurs when 0 is varied as
In measurement, the problem of transmitting the output
signals of many sensors or transmitters over one line is (t) = f (x(t)) (1d)
often solved by means of the time-division multiplexing.
This time-sharing method is shown in Figure 1. A switch Phase modulation (PM) occurs when φ is varied as
sequentially connects the several input channels into the
single output line. The length of the connection time gives φ(t) = f (x(t)) (1e)
the pulse interval tp and the time between switching back
to the same channel again is period to . Pulse modulation, Figure 2 shows the modulation methods mentioned above
with the interleaving of the different signals, is applied for for a sinusoidal modulation signal with the modulation
parallel-to-serial conversion. frequency ω:
In general, a modulator may be interpreted as a controlled sin(ωt)
x(t) = X (1f)
system with the carrier signal as one input and the mod-
ulation signal x(t) as the control input (Figure 2). In the Pulse modulation methods use the principle of sampling –
following text, a survey is first given of the several kinds see Article 31, Discrete Signal Theory, Volume 1.
of modulation. As shown in Figure 4, parameters of a regular pulse
In analog modulation methods (see Figure 3), one, or sequence are changed by the modulation signal x(t) in one
several, parameters of the sinusoidal oscillation u(t), termed of several ways.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
192 Foundations of Measuring
x1 y1
tp tp
x2 Channel y2
(a) xn yn
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0
tp to ts
(b)
Figure 1. Time-division multiplexing: (a) system and (b) pulse frame (for 10 transmitters).
Carrier Controlled Modulated x
signal system signal
Modulation signal (a)
Figure 2. Generalized modulation system. u
Variation of the pulse amplitude provides pulse-ampli-
t
tude modulation (PAM), while with the pulse-duration or
pulse-width modulation, the lengths of the pulses corre-
spond to the modulation signal. (b)
A pulse modulation method of great importance, espe-
u
cially in communication because of its good signal-to-noise
ratio, is the pulse-code modulation (PCM) (Figure 5).
In the usual case, a pulse-amplitude modulation signal
t
is first generated, which is then converted to a coded
pulse sequence using one of the codes of Article 33,
Coding Theory and its Application to Measurement, (c)
Volume 1.
u
2 AMPLITUDE MODULATION
t
We now deal with some details of the bandwidth needed
and application in measurement. (d)
A sinusoidal modulation signal
u
sin(ωt + ϕ)
x(t) = X (2a)
0 sin(0 t) yields the amplitude-
operating on a carrier U t
modulated oscillation
0 [1 + m sin(ωt + ϕ)] sin(0 t)
u(t) = U (2b)
(e)
Figure 3. Analog modulation methods: (a) modulation signal
Representation
where the modulation depth is m = k X. x(t); (b) carrier oscillation; (c) amplitude modulation; (d) fre-
of (2b) in the frequency domain is (Woschni, 1988; quency modulation and (e) phase modulation.
Modulation Theory 193
x x
t
t
(a) (a)
u u
tp ts t
0 ts 2t s 3t s 4t s t
(b)
(b)
u
u
30
18
t 4
(c) 10 10
u 0 ts 2t s 3t s 4t s t
(c)
t 01010 =^ 10 10010 =^ 18 11110 =^ 30 00100 =^ 4 01010 =^ 10
(d)
u
tp ts 2t s 3t s 4t s t
0
(d)
t Figure 5. Pulse-code modulation: (a) modulation signal; (b)
(e)
carrier pulse sequence; (c) pulse-amplitude modulation and
Figure 4. Pulse modulation methods (ts = sampling time): (d) pulse-code modulation (binary code).
(a) modulation signal x(t); (b) carrier pulse sequence; (c) pulse-
amplitude modulation; (d) pulse-duration or pulse-width modula-
tion and (e) pulse-phase or pulse-position modulation. the original form of the modulated signal will arise in the
later recovered signal (Woschni, 1988).
Wozencraft and Jacobs, 1965) In measurement, amplitude modulation results at the out-
put of an AC bridge – see Article 126, Electrical Bridge
sin( t) ± 1 m cos[( ∓ ω)t ∓ ϕ]
u(t) = U (2c) Circuits – Basic Information, Volume 2, operating with
0 0 2 0
inductive sensors, as presented in Figure 7. To obtain sat-
showing that the resultant is a signal with the carrier isfactory dynamic behavior, the condition between the lim-
frequency and two side frequencies (Figure 6a). The com- iting frequency of the measured input ωc and the carrier
position of the several spectral frequencies provides the frequency 0 needs to be
time function, as shown in Figure 6b.
0 ≥ 5 ωc (2e)
In the general case of an input signal with bandwidth
ω1 – ωc to be modulated, the bandwidth needed by the
If this condition is not met, it will not be possible to
transmission link is
correctly demodulate the amplitude-modulated oscillation.
b = 2 ωc (2d) The condition given by (2e) is very important, for it is
the key to avoiding errors in dynamic measurements as
around the carrier frequency 0 . An amplifier, therefore, shown in detail in Article 50, Errors in Signal Systems,
has to have at least this bandwidth otherwise distortion of Volume 1.
194 Foundations of Measuring
^ ^ u(t )
U U0
w
C0 ± ∆C
m U^ m U^
2 0 2 0
Figure 8. Circuit for the operation of a capacitive sensor to
w w Ω provide frequency-modulated output.
(a) Ω0
√
the variation of the natural frequency = 1/ (LC ) is
given by (Woschni, 1962)
1 C
m U^ m U^ = 0 + sin(ωt) = 1− sin(ωt)
2 0 2 0 (LC0 ) 2C0
w −w
3 C 2 2
+ sin (ωt) − · · · (3b)
U^0
8 C0
Because of the nonlinear characteristic differential capaci-
Ω0 tance, sensing methods are often used.
(b) The time function yields
Figure 6. (a) Frequency spectrum of sinusoidal amplitude mod-
ulation and (b) phasor representation.
u(t) = U0 sin [0 + sin(ωt)] dt
sin t − cos(ωt)
=U (4a)
0 0
ω
w w The corresponding function for sinusoidal phase modulation
is given by
L + ∆L L − ∆L
AM
sin[ t + φ sin(ωt)]
u(t) = U (4b)
0 0
A comparison between both equations shows that
φ = (4c)
ω
Ω0
Table 1. Relations between frequency and phase modulation.
Figure 7. Bridge circuit, delivering an amplitude modulation.
Frequency Phase
modulation modulation
3 FREQUENCY OR PHASE Frequency deviation = ω
MODULATION Phase deviation =
ω
For the operation of capacitive sensors having high sensi-
Modulation-
tivity, and in certain cases for inductive sensors, frequency Limiter converter AM demodulation
modulation is used (Figure 8). For sinusoidal variation of
the capacitance of the sensor FM FM AM NF
u (t ) udem
C
C = C0 + C = C0 1+ sin(ωt) (3a)
C0 Figure 9. Principle of demodulation of frequency modulation.
Modulation Theory 195
Circuit Characteristic
i(t ) ^
Udem
Oscillating
circuit udem
demodulator
Ω 0 Ω res Ω
^
i(t ) Udem
u D1
uD1 udem =
Push-pull udem uD − uD
1 2
demodulator
Ω
u D2
uD2
i (t ) ^
Udem
uD1
Riegger udem
demodulator Ω
uD2
Figure 10. Circuits for realizing FM demodulation.
represents the equivalent phase deviation called the mod- REFERENCES
ulation index . Therefore, the relationships contained in
Table 1, between frequency and phase modulation, are valid Webster, J.G. (1999) Measurement, Instrumentation and Sensors
(Woschni, 1962). The spectrum of a frequency or phase- Handbook, CRC Press, Boca Raton, FL.
modulated signal is derived by means of a series expansion Wilson, S. (1996) Digital Modulation and Coding, Prentice
of Bessel functions (Woschni, 1962), leading to a bandwidth Hall, NJ.
necessary for distortion-free transmission given by Woschni, E.-G. (1962) Frequenz Modulation, Verlag Technik,
Berlin.
b∼
= 2[ + kω] (4d)
Woschni, E.-G. (1988) Informations Technik, Huthig, Heidel-
with 1 < k < 2. berg.
The demodulation of a frequency-modulated oscillation Wozencraft, J.M. and Jacobs, I.M. (1965) Principles of Commu-
is realized as demonstrated in Figure 9. A limiting stage nication Engineering, Wiley, New York.
is followed by a modulation converter that converts the
frequency modulation into an amplitude modulation, which
is rectified by means of a diode. FURTHER READING
Some example circuits for FM demodulation are shown
in Figure 10 (Woschni, 1962). See also Webster (1999) or Haykin, S. (2001) Communication Systems, 4th edn Wiley, New
Wilson (1996). York.
35: Systems in the Time Domain
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
by a computer of either analog, or today more likely digital
1 Classification of Technical Systems 199 form (an aspect of modeling).
2 Modeling by Means of Analogies 200 Because of the convenience afforded by a computer to
study the behavior of a given system, the role of program-
3 Block Diagrams 201
ming modeling methods plays an important part in the field
4 Linearization 202
of cybernetics. Measurement as a part of cybernetics – the
5 Description with Differential Equations 203 general science of feedback – can be comprehended as a
6 Simple Harmonic Motion 204 mapping of the input signals onto the space of the output
7 State Space Description 204 signals (Finkelstein, 1977).
8 Flowcharting 205 Technical systems are characterized as those with active
9 General Input Response, Convolution 208 or only passive elements (so-called active or passive sys-
tems) or they are described by the number of ports (two-
References 209
port, three-port, etc.) – see also Article 73, Overview of
Further Reading 209
Electrical and Electronic Technique, Volume 2.
Table 1 gives a survey of the classification of technical
systems. With respect to the difficulties in the treatment
1 CLASSIFICATION OF TECHNICAL of a system, it is of great importance to know whether
SYSTEMS the system is linear or not, because in the linear case the
superposition law is valid. In this article, it is assumed the
According to Figure 1, a system can be interpreted as a system is a linear one. Methods of linearization are treated
black box with a family of input variables xr , which can be in Section 4.
regarded as a signal vector x, and a family of corresponding Another typical characteristic of a system is whether
output variables yr , forming a vector y. The interior of the the parameters describing the behavior are functions of
black box may consist of several elements being either of time or not. Furthermore, most of the systems used in
electrical, hydraulic, pneumatic, or other energy forms. cybernetics are so-called unidirectional systems, meaning
The overall behavior of these systems is given by a that the parameters of the system are independent of those
common mathematical equation of the general form: of the following system. We will only deal with systems
that fulfill this assumption. Otherwise, the results of the
y = O{x} (1) four-pole theory must be applied (Feldtkeller, 1962).
As an example of practical importance in the field of
Different systems may have the same overall behavior so measurement, the spring–mass–damper system, as illus-
this means they follow the relationship of (1). This shows trated in Figure 2, will be considered. This system makeup
that it is possible for a system to be substituted by another is used in numerous sensors, for, among others, in the mea-
one having the same mathematical equation, for example, surement of length or force. If force F is the input variable
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
200 Foundations of Measuring
2 MODELING BY MEANS OF
x1 y1
ANALOGIES
x2 System y2
x
y = O(x)
xr yr As mentioned before, systems with different interior ele-
ym ments or forms of energy can have the same mathematical
xn
relationship between output and input variables. From this
fact it follows that a given system can be represented by
Figure 1. Definition of a system. some other system having the same overall behavior. This
modeling opportunity has the advantage that with response
investigations of a given system its model system can have
the parameters and structures changed by simply program-
x=F ming a computer. Furthermore, it is possible to observe the
input and output quantities in a convenient way by means of
oscilloscopes, computer displays, or plotters and to change
m y=s the scale of coordinates or time axes to suit the needs of
the investigation. Important methods of modeling are the
analogies, application of block diagram representation, and
linearization.
k C Of importance is the fact that mechanical systems, in the
same way as pneumatic, hydraulic, and other systems, can
be presented by electrical systems, as shown in Figure 3.
In the cases illustrated in this arrangement, the following
Figure 2. Spring–mass–damper system. equations are valid:
x, and length s is the output variable y, the following spe- F =^ x
w=
^ y
cial equation corresponding to the general relation (1) is
obtained: M =^ x
m v =^ y
mÿ + k ẏ + cy = x (2) (a) (b)
Θ
j =^ x
This second-order differential equation describes the
system’s dynamic behavior. For more depth of explanation
see Article 59, Introduction to the Dynamic Regime C u =^ x x y
of Measurement Systems, Volume 1; Article 60, Zero- C
(c) (d)
order System Dynamics, Volume 1; Article 61, First-
order System Dynamics, Volume 1; and Article 62, Figure 3. Examples of mechanical and electrical systems:
Second-order System Dynamics, Volume 1. (a) translation, (b) rotation, (c) capacity and (d) computer format.
Table 1. Classification of technical systems.
System Linear system Nonlinear system
Mark Parameters are constant, independent of amplitude Parameters are a function of amplitude
Superposition law is valid Superposition law is not valid
Parameters are time Examples: most measurement systems with small Examples: systems with large input levels: output
invariant input levels: amplifiers; filters; transducers amplifiers, driver stages; Often linearization is
possible
Parameters are functions Examples: controlled amplifiers with Examples: controlled amplifiers with
of time multiplicative properties and small input levels: multiplicative properties and large input levels:
modulators, frequency multipliers, parametric modulators, frequency modulators, frequency
amplifiers multipliers
Systems in the Time Domain 201
1. In the mechanical example (mass m, velocity v,
force F )
1
v= F dt (3a) i
m
iL iC iR
2. In the case of rotary motion (moment of inertia ,
angular velocity ω, torque M) L C R u
1
ω= M dt (3b)
3. In the electrical system (voltage u, capacitance C, (a)
current i)
i L C R
1
u= i dt (3c) uL uC uR
C
4. In the general case of an analog computer, or more ⬃u
likely today, the digital computer (constant c)
(b)
y = c x dt (3d)
Figure 4. Electrical model of the spring–mass–damper system
of Figure 2 (a) parallel circuit and (b) series circuit.
Generalizing the dependencies illustrated in Figure 3
realizes the survey of analogies between electrical
and mechanical systems shown in Table 2. For the a mechanical parallel circuit is modeled as an electrical
spring–mass–damper system with differential (2) and the parallel circuit.
representation of Figure 2, the electrical models shown in Similar analogies exist between acoustic, hydraulic, and
Figure 4 are obtained. Direct analogy (see Table 2) yields pneumatic systems and to a limited extent to optical
the parallel circuit Figure 4(a) with the equation systems. Their theory, first ordered in Olson (1943), is
paramount in the design and testing of measuring sys-
du 1 1 tems. For details, refer to Koenig and Blackwell (1961)
C + u+ u dt = i (4a)
dt R L and Shearer, Kulakowski and Gardner (1997). The analo-
gies are also the foundation of modern disciplines as
while use of indirect analogy provides the series circuit mechatronics and micromechanics. Another example is the
given in Figure 4(b): surface-acoustic wave (SAW) filters.
1 di
i dt + Ri + L = u (4b)
C dt
3 BLOCK DIAGRAMS
As can be seen from a comparison of Figures 2 and 4, the
indirect analogy is more convenient in this instance because Measurement systems are built up from particular combi-
nations of subsystems. Therefore, it is suitable to represent
Table 2. Survey of electromechanical analogies. each subsystem by its own block including a symbol indi-
cating the operation the subsystem has to realize. Table 3
Electrical Mechanical system
contains some of these symbols and signs used for the
system
Translation Rotation demonstration of the interconnections between the systems.
(It must be pointed out that many standards for such are in
Direct Indirect Direct Indirect
use.) It is therefore assumed that the systems are unidi-
i F v M ω rectional and thus it is possible to describe the behavior of
u v F ω M such a system by means of an interconnection between sev-
1 1
R k k eral unidirectional systems. Figure 5 illustrates three typical
k k
1 1 methods of interconnection of systems. The output–input
C m θ
m θ relation of a system is given by the equation
1 1
L C C
C C y = Gx (5)
202 Foundations of Measuring
Table 3. Symbols for block diagrams.
Function Branch Summation Subtraction Sign inversion General system
Symbol y1 x2 y x1 + y x + − x y
x −
x1 x2
y2
Equation y1 = y2 = x y = x1 + x2 y = x1 − x2 y = −x y = f (x, t)
Function Constant factor Integrator Nonlinear system Root calculation Multiplier
Symbol x y x y x y x y x y
k dt f n
√ x
x
√
Equation y = kx y= xdt y = f (x) y= n
x y = x1 x2
3. In opposition (Figure 5c)
x = x1 y1 x2 y2 xn yv xn yn = y
G1 G2 Gn Gn
G1
(a) feedforward, oscillator (6c)
1 − G1 G2
Ge =
x1 y1
G1
G1 negative feedback, control (6d)
1 + G1 G2
x2 y2 The product G1 G2 in (6c, d) describes the frequency
G2
response of the open control loop (Levine, 1996; Woschni,
x y
1988).
xn yv
Gn
4 LINEARIZATION
xn yn
Gn
(b) Linear systems are distinguished by the validity of the
superposition law. Nonlinear systems are not easily mod-
x x1 y1 y eled with linear equations and to get round this difficulty a
G1
+ nonlinear system is often linearized in small steps that col-
Case 1:+
lectively model the whole excursion region. This enables
Case 2:− the advantages of linear systems to be used. The following
y2 x2 preliminary conditions must be fulfilled in such a strategy:
G2
(c)
1. Only small deviations of the characteristics from the
Figure 5. Typical system connections: (a) series circuit, (b) par- linear path can be used.
allel circuit and (c) connection in opposition. 2. Only a relatively small drive range of the nonlinear
characteristic can be tolerated.
The frequency response Ge for an equivalent system hav-
For the linearization, the Taylor series expansion of
ing the same overall behavior as that of the interconnection
the nonlinear characteristic y = f (x) at the working point
of the subsystems shown is given by the following:
y0 = f (x0 ) is employed. Writing only the deviations from
1. Series circuit (Figure 5a) the working point x, y yields
n ∂f 1 ∂ 2 f
Ge = y − y0 = y = x + x 2
Gr (6a) ∂x x0 2 ∂x 2 x0
r=1
1 ∂ 3 f
+ x 3 + · · · (7a)
2. Parallel circuit (Figure 5b) 6 ∂x 3 x0
n
Ge = Gr (6b) In practical measurement technique, the input variable is
often the sinusoidal function x = Xsin(ωt).
r=1
Systems in the Time Domain 203
The proportion of the dominant wave with frequency ω ^ ^
, expressed as a ratio with the amplitude of
at the output Y
Y w /X
ω
the input X, gives
3
f/ x3 > 0
3 f
Yω
∂f ∂ |
1 ∂x 3
2
x 0
= 1 + X (7b)
X ∂x x0 8 f
∂ |x 0 f 3
f/ x3 = 0
∂x x x0
This function is the describing function; it expresses the
frequency response of a nonlinear system. Depending on the 3
f/ x3 < 0
sign of the third-order differential coefficient, the describing
function either increases or decreases with the square of
the amplitude of the input as Figure 6 shows. In practice,
the case ∂ 3 f /∂x 3 |x0 > 0 can be troublesome because the X
^
amplification factor is increasing with increasing amplitude
leading to an unstable oscillating regime (Woschni, 1988; Figure 6. Ratio of the amplitude of the dominant wave of the
Levine, 1996). output Yω to the amplitude of the input X (describing function).
Important in the field of measurement is the rectification
effect using a square-wave characteristic created as a recti- where G0 is the static sensitivity:
fier unit that switches between on and off without any state
b0 y
between.
G0 = = (8c)
1 ∂ 2 f
2 a0 x
y0 = X (7c)
4 ∂x 2 x0 G0 can be measured as an amplification factor by means of
a small alteration x as stated in (8c). The coefficients
In this case, distortion appears that is described by the
distortion factors ar
Tr = r (8d)
a0
2 f
Y 1 ∂ |
2 x0
k2 = 2 ω = ∂x
X, (7d) are time constants. For a differential equation of nth order,
Y
4 f
ω ∂ |x 0 n time constants are defined, the greatest of which is
∂x used to estimate the duration of the transient function. The
3 f solution of the differential equation consists of two additive
Y ∂ |
1 ∂x 3 x0
2
k3 = 3 ω = f X (7e) components, a stationary and a dynamic portion:
Y
24
ω ∂ |x 0
∂x y(t) = yst + yd (t) (9a)
To solve the homogeneous differential equation leading to
5 DESCRIPTION WITH DIFFERENTIAL the dynamic portion yd (t), the assumption
EQUATIONS yd = C ept (9b)
The oldest classical method used in solving problems of is used. The zero points of the characteristic equation
system analysis is the method of differential equations. A
linear system is described by a linear differential equation Tnn p n + Tn−1
n−1 p−1
p + · · · + T22 p 2 + T1 p + 1 = 0 (9c)
of the following form:
signify whether the solution is stable or not. In particular,
an y (n) + an−1 y (n−1) + · · · + ar y (r) + · · · + a2 ÿ these so-called eigenvalues pr prove that
+ a1 ẏ + a0 y = b0 x + b1 ẋ + · · · + bm x (m) (8a)
1. if the real part is less than zero, that is, Re(pr ) < 0, a
This equation may be written as stable solution exists;
2. if the real part is greater than zero, that is, Re(pr ) > 0,
Tnn y (n) + Tn−1
n−1 (n−1)
y + · · · + Trr y (r) + · · · + T22 ÿ an unstable solution exists;
b1 b 3. if the eigenvalues are complex, oscillations with dec-
+ T1 ẏ + y = G0 x + ẋ + · · · + m x (m) (8b)
a0 a0 reasing or increasing amplitudes exist.
204 Foundations of Measuring
With the eigenvalues pr , the dynamic solution yields It may be noted that (10c) is the Fourier transform of the
(Coddington and Levinson, 1955) differential (8b) under the supposition of vanishing initial
conditions, which means x(t)t=0 = 0 and so on.
n
yd = Cr exp(pr t) (9d) The same is valid under the same suppositions with
r=1 respect to the Laplace transform, which means jω → p
If a double root p0 arises
Tnn s n Y (s) + · · · + T1 sY (s)
yd = (C1 + C2 t) exp(p0 t) (9e) b1 b
G0 X(s) + sX(s) + · · · + m s m X(s) (10e)
a0 a0
The stationary solution yst is to be found by means of suit-
able terms satisfying the inhomogeneous differential (8b). leading to the transfer function, see Article 36, Systems in
As may be seen from (9d), the eigenvalue pr corre- the Frequency Domain, Volume 1
sponds to a time constant Tr = 1/pr , the greatest value of
which (Tmax ) is responsible for the duration of the transient b1 b
process. Because e−3 = 1/20 = 5% the transient process G0 + s + · · · + m sm
Y (s) a0 a0
approximately continues and G(s) = = (10f)
X(s) 1 + T1 s + · · · + Tnn s n
ttr = 3Tmax (9f)
where ttr is the transient time.
7 STATE SPACE DESCRIPTION
The basis of the state space description is the classical
6 SIMPLE HARMONIC MOTION differential equation discussed above or, in the case of
sampled-data systems, the difference equation. The state
In practice, the special case of a sinusoidal input signal x(t) of a system is described by a set of state variables yr .
is very important The number of these state variables agrees with the degree
jωt
x(t) = Xe (10a) of the differential equation or in the case of sampled-
data systems with the degree of delay. Today, the method
This test signal is used to gain the frequency response, is of significant importance because of its suitability for
see Article 36, Systems in the Frequency Domain, Vol- computer simulation of systems.
ume 1. In the steady state regime, which means after a time The nth-order differential equation for such systems
longer than the transient time since the signal x(t) started,
the output signal y(t) will also be a sinusoidal signal but Tnn y n + Tn−1
n−1 n−1
y + · · · + ẏ + y = x (11a)
with another amplitude Y
and a phase ϕ
can be transformed into a system of n differential equations
e j(ωt+ϕ)
y(t) = Y (10b) of first order:
We use these functions x, y as a statement in the original y = y1
differential (8b) taking into account that every differentia- ẏ = y2 = ẏ1
tion means a multiplication with jω
ÿ = y3 = ẏ2
e jϕ
e jϕ + Y
+ · · · + T1 jωY
[Tnn ( jω)n Y
e jϕ ]e jωt ..
.
b1
bm m
= G0 X + jωX + · · · + ( jω) X e jωt (10c) y (n−1) = yn = ẏn−1
a0 a0
n−1
1 T1 T22 Tn−1
Reducing this equation to the output/input relation, we yn = − y − y − y − · · · − y
get the frequency response that is treated in Article 36, Tnn 1 Tnn 2 Tnn 3 Tnn n
Systems in the Frequency Domain, Volume 1 in detail 1
+ x (11b)
Tnn
b b
e jϕ G0 + jω 1 + · · · + ( jω)m m
Y a0 a0 This system of equations can be written in the form of
G( jω) = = (10d)
X 1 + jωT1 + · · · + ( jω) Tnn
n a vector differential equation (Zadeh and Desoer, 1963;
Systems in the Time Domain 205
Levine, 1996) With the matrices
y1 ẏ1 −a(n−1) 1 0 ...0
y2 ẏ2 −a ···0
(n−2) 0 1
d
y3 ẏ3 A=
.. .. .. .. (12g)
= . . . . 1
dt .. ..
. . −a0 0 0 ...0
yn ẏn
b(k−1) · · · −bn a(n−1)
b −bn a(n−2)
0 1 0 0 ··· 0 (k−2) · · ·
0 0 1 0 ··· 0 B = . .. (12h)
.. ···
0 0 0 1 ··· 0 .
···
= ... .
.. .. .. ..
b0 bn a0
. . ··· . C = [1, 0 . . . 0] (12i)
1 T T2 T3 n−1
Tn−1
− n − 1n − 2n − 3n ··· − n D = bn (12j)
Tn Tn Tn Tn Tn
0 Consider as an example the spring–mass–damper system
y1 0
y2 of Figure 2 with the differential equation
0
y3
× + . x (11c) mÿ + k ẏ + cy = x (13a)
. ..
..
1
yn The vector equation is given by
Tnn
0 1 0
ẏ1 y
or with the variables y, T, b, and x: = c k 1 + 1 x (13b)
ẏ2 − − y2
m m m
ẏ = Ty + bx (11d)
The state variables are the displacement y = y1 of the
For sampled-data systems (Woschni, 1988; Levine, 1996), mass and the velocity ẏ = y2 .
the difference equation is
n
m
8 FLOWCHARTING
aµ y(k + µ) = br x(k + r) (12a)
µ=0 r=0
Let us start with the simple example of the spring–mass–
Defining a delay element as damper system of Figure 2 and (13a, b).
Figures 7 and 8 respectively show the programming of
a computer for modeling this system. The relationships
D{y(k)} = y(k − 1) (12b)
between programming and state space description are easy
We obtain also a system of n difference equations of first
1
order 1 mx
x m 1 −y2 y1 = y
1 1
Zµ (k) = D − a(n−µ) Zµ (k) 1
+ Z(µ−1) (k) + (b(n−µ) − bn a(n−µ) (k) (12c) k y
−m 2
k
Y (k) = Z1 (k) + bn (k) (12d) m
c y
−m
This system of equations can be written in matrix form
1
c − y1
m 1
Z(k + 1) = AZ(h) + B x(k); (12e)
Figure 7. Direct programming of a spring–mass–damper system
Yk = C z(k) + Dx(k) (12f) on a computer.
206 Foundations of Measuring
In Figure 9, the path of the five state variables displace-
Start ment y = z1 and velocity y = z2 are shown both as a
function of time as well as in the state space.
In Figure 10, the programming of a system of nth order
is shown as described with the differential (11a) and the
Input of the following data:
Step length of time ∆t state (11c). Here, in the flowchart shown in Figure 8, the
Final value of time tmax subroutine of Runge–Kutta is used to realize n differential
Initial values y1 (O), y2(O) equations of first order.
c k 1 Programming for a sampled-data system, as given by
Constants m m m (8c, d) or (8e, f), is presented in Figure 11.
Parameters of the input x (t)
For system description and identification, certain res-
ponse output functions for test signals at the input are
generally used. These are presented here as time-dependent
functions but they can be used with any general vari-
able x, spatial distribution being the one used in opti-
Subroutine of
Runge−Kutta for two cal systems.
differential equations of first
order
Z1, Z2
Output of y1 and y2 Z1 = y
Z2 = y o
t1
No
It − t maxI < I∆t /2 /? t
Yes (a)
Z2 = y o
Stop
y (t1)
Figure 8. Flowchart for the programming of a spring–mass–
damper system on a digital computer. Z Z1 = y
t1 y o(t1)
t
to recognize, especially due to the equation (b)
Figure 9. Displacement y = Z1 and velocity y = Z2 for the
c k 1 spring–mass–damper system (a) as a function of time and (b) in
ż2 = z − z + x (13c) the state space.
m 1 m 2 m
1
T nn
x
x 1n −2n
Tn
1 Zn−1 Z3 −Z2 Z1 = Y
1
1
1
T nn-1
-1 T nn-2 T 22 T1 1
-2
T nn T nn T nn T nn T nn
1
1
1
Figure 10. Programming of a system of nth order.
Systems in the Time Domain 207
x(k )
b0 b1 b2 bn −1 bn
∑ V zn(k) ∑ zn −1(k) ∑ z2(k) ∑ V z1(k) ∑
y (k)
a0 a1 a2 an −1
Figure 11. Programming of a sampled-data system.
y, x 2. The response to a pulse function with an integral
value of 1 (Dirac delta function, δ(t)) yields the unit
y (t ) = h(t )
pulse response or weighting function g(t) shown in
Figure 12(b).
Because of the linearity of the system, the response
1 function y(t) is to be divided by the step amplitude or the
x = w (t ) integral value, respectively to get the normalized function.
In the case of system identification, if the input function
generated by a signal generator is not the ideal function but
a function with a rise time τ as signified in Figure 12,
the condition must be fulfilled that the rise time τ , or the
pulse width τ , is very much smaller than the transient
∆τ t time ttr of the system (Woschni, 1988).
(a) 0 As can be seen from the comparison of the parts of
Figure 12, the pulse function is connected with the step
y, x
function by means of a differentiation, that is, in the sense
of distribution (Gelfand and Schilow, 1960)
dw(t)
= δ(t) (14a)
∞
dt
For linear systems, it is immaterial whether the differenti-
∆τ 0 ation is realized at the input or output side of the system,
y (t ) = g(t )
meaning that
h(t) = g(t) dt (14b)
It is only a matter of convenience whether the transient or
∆τ t
(b)
the weighting functions are used for identification.
0
For a system with first-order delay, as is used for the
Figure 12. (a) Unit step w(t) and transient response h(t); approximation of several systems in measurement technol-
(b) delta function δ(t) and weighting function g(t). ogy, we gain the differential equation
T1 ẏ + y = x (14c)
1. The response to a step function with a step ampli-
tude of 1 (unit step function, w(t)) is the unit step with the eigenvalue
response or transient response h(t) as illustrated in −1
p1 = (14d)
Figure 12(a). T1
208 Foundations of Measuring
The stationary solution may be found by means of the 9 GENERAL INPUT RESPONSE,
assumption CONVOLUTION
yst = c2
Transient response and weighting functions are response
for which the total solution yields functions to special input signals. In the general case, the
input function is broken down into a series of weighted
Dirac delta functions, which are time delayed as represented
−t
y = yst + yd = c2 + c1 exp in Figure 14. The pulse at the time τ1 yields the output
T1
Using the boundary conditions x(τ1 )τ × g(t1 − τ1 )
Because of the linearity of the system the superposition law
y|t=−∞ = 0 y|t=∞ = 1
is valid, which means the entire output is the sum (integral)
of all inputs at the time t − τ1 > 0:
we get for the transient response h(t):
t
−t y(t) = x(τ )g(t − τ ) dτ
h(t) = 1 − exp (14e) 0
T1 t
= x(t − τ )g(τ ) dτ
The weighting function g(t) follows from (14b): 0
= x(t)∗ g(t) (15a)
dh(t) 1 −t
g(t) = = exp (14f)
dt T1 T1 This is the convolution integral, denoted by the sign * for
the convolution. The lower limit of the integral may be
The functions h(t) and g(t) are represented in Figure 13.
If these functions are obtained experimentally by means
x
of a function generator at the input of the system exam- ∆t
ined, the time constant T1 is given by the length of the
subtangent (Figure 13). Furthermore, the figure shows the
transient time ttr to be nominally three times the time con- x (t)
stant T , for e−3 = 1/20. For more details referring to testing
of systems and important examples, see Article 37, Rela-
tionship Between Systems in the Time and Frequency
Domain, Volume 1.
x0 t1 t
t1 t
1/T1
y
x (t1)∆t × g(t −t1)
1
h (t )
g (t )
t1 t
0 T1 2T1 t
t1 t
Figure 13. Transient response and weighting function for a
first-order system. Figure 14. Explanation of the convolution integral.
Systems in the Time Domain 209
extended to infinity since the weighting function must be be difficult. Therefore, methods in the frequency domain
zero before the input is applied. Since the input is zero for were developed as shown in Article 36, Systems in the
t < 0, the upper limit may also be extended to infinity, this Frequency Domain, Volume 1. In the special case of
means (15a) may also be written as white noise with an autocorrelation function ψxx (τ − t) =
2πS0 δ(τ − t) the deconvolution degenerates to the equation
+∞
y(t) = x(τ )g(t − τ ) dτ
−∞ ψxy (τ ) = 2πS0 g(τ ) (16c)
+∞
= x(t − τ )g(τ ) dτ (15b) This means that the weighting function corresponds directly
−∞
to the cross-correlation function (Woschni, 1988).
Another form of the convolution integral, the Duhamel
integral, is obtained by taking into consideration (14b):
REFERENCES
t
d
y(t) = x(τ )h(t − τ ) dτ Coddington, E.A. and Levinson, N. (1955) Theory of Ordinary
dt 0
Differential Equations, McGraw-Hill, New York.
d t
= x(t − τ )h(τ ) dτ Davies, W.D.T. (1970) System Identification for Self-Adaptive
dt 0 Control, Wiley, Chichester.
d Feldtkeller, R. (1962) Einfuhrung in die Vierpoltheorie, Hirzel,
= [x(t)∗ h(t)] (15c) Stuttgart.
dt
Finkelstein, L. (1977) Instrument Science. Journal of Physics E:
In this equation, the upper and lower limits may also be Scientific Instruments, 10, 566–572.
extended to infinity. Gelfand, I.M. and Schilow, G.E. (1960) Verallgemeinerte Funk-
In the field of measurement, convolution is of great tionen (Distributionen), Vol.1, Deutscher Verlag d. Wiss.,
importance for system identification (Davies, 1970). The Berlin.
autocorrelation function at the output of a system and the Koenig, H.E. and Blackwell, W.A. (1961) Electromechanical Sys-
input are related by a double convolution (Woschni, 1988) tem Theory, McGraw-Hill, New York.
∞ ∞ Levine, W.S. (1996) The Control Handbook, CRC Press, New
ψyy (τ ) = ψxx (τ + τ1 − τ2 )g(τ1 )g(τ2 ) dτ1 dτ2 York.
0 0 Olson, H.F. (1943) Dynamic Analogies, Van Nostrand, London.
(16a)
If the autocorrelation function of the input ψxx (τ ) and the Shearer, J.L., Kulakowski, B.T. and Gardner, J.F. (1997) Dynamic
cross-correlation function τxy (τ ) are measured, the weight- Modeling and Control of Engineering Systems, Prentice Hall,.
Englewood Cliffs, NJ.
ing function of the system can be calculated, for which the
following relation is valid (Davies, 1970; Woschni, 1988; Woschni, E.-G. (1988) Informations Technik, Huthig, Heidelberg.
Levine, 1996) Zadeh, L.A. and Desoer, C.A. (1963) Linear System Theory,
McGraw-Hill, New York.
∞
ψxy (τ ) = g(t)ψxx (τ − t) dt (16b)
0
FURTHER READING
Deconvoluting (16b) gives the required pulse response
function g(t); the system may be regarded as being iden- Oppenheim, A.V., Willsky, A.S. and Young, I.T. (1983) Signals
tified. Direct deconvolution techniques using (16b) can and Systems, Prentice Hall, Englewood Cliffs, NJ.
36: Systems in the Frequency Domain
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
complex frequency response is
1 Frequency Response 210
2 Logarithmic Characteristics and Bode Plot
Y
e jϕ = G( jω) = |G( jω)|e jϕ(ω) (1c)
Representation 211
X
3 Transfer Function 212
4 Poles and Zeros and Root Locus This function is represented in the complex plane as a locus
Representation 213 diagram. The complex frequency response may be split into
5 Typical Example: Phase-shift Bridge 215 a real and an imaginary part:
References 215
G( jω) = P (ω) + jQ(ω) (1d)
with the relations
1 FREQUENCY RESPONSE
|G( jω)| = [P 2 (ω) + Q2 (ω)] (1e)
(Note of clarification: Before this article is read, it is
Q(ω)
necessary to point out that in the literature s is often used ϕ(ω) = arctan (1f)
instead of p and vice versa. λ is used for Eigen values as P (ω)
well as for p. G and H are often interchanged. There are
differences in strict mathematical use, but engineers have Figure 1 shows the representation of the frequency char-
found it acceptable to use them interchangeably and it is a acteristics. If the differential equation is given, it is very
matter of personal preference in these explanations.) convenient to obtain the frequency response by means of
Using a sinusoidal input and taking both amplitudes of the terms
the output Y normalized to the input X, and phase ϕ as
a function of frequency ω follows the frequency response jωt ; e jϕ e jωt
with x (µ) = ( jω)µ Xe y (r) = ( jω)r Y (2a)
Y
amplitude characteristic is |G( jω)| = (1a) Substituting the differential equation, see Article 35, Sys-
X tems in the Time Domain, Volume 1
ϕ(ω) ≤ y, x (1b)
b1 b
jωt yields
In the complex presentation, the input x = Xe Tnn y (n) + · · · + T1 ẏ + y = G0 x + ẋ + · · · + m x (m)
a0 a0
the output y = Y e j(ωt+ϕ)
= Y e e , which means that the
jϕ jωt
(2b)
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Systems in the Frequency Domain 211
|G (j w)| The frequency response can be measured, by way of (1a,
b), using a sinusoidal test signal. If the input is a stochastic
|G (j w1)| signal with power density Sxx (ω), the output power density
Syy (ω) may be calculated according to Davies (1970) and
G0
Woschni (1988)
Syy (ω) = |G( jω)|2 Sxx (ω) (2d)
The double convolution in the time domain corresponds
to a multiplication with |G(jω)|2 in the frequency domain
(Woschni, 1988; Levine, 1996).
(a) w1 w
w1 w
2 LOGARITHMIC CHARACTERISTICS
AND BODE PLOT REPRESENTATION
f1 Of great practical importance is the plotting of the ampli-
tude characteristic in a double-logarithmic calibration graph
and the phase characteristic with a linear ϕ-axis and a
logarithmic ω-axis, known as logarithmic frequency char-
− 3π acteristics. In order that a linear scale be used for the y-axis
2
(Bode diagram), the amplitude is generally measured in
decibel (dB) units,
(b) f
Y
+j 20 log = 20 log |G( jω)| (3a)
w=∞ P G0
X
−r f1 w=0 +r
The advantage of this method is illustrated in Figure 2. As
|G
(
Q treated in Article 35, Systems in the Time Domain, Vol-
jw
1 )| ume 1, the overall behavior of series-connected systems is
given by the multiplication of the frequency responses of
w1
each of these systems. Because of the logarithmic repre-
sentation, the multiplication is simplified to a summation,
which can easily be realized graphically.
For a system with first-order delay, used for approxi-
mation of more complicated systems, from the differential
(c) −j equation – see Article 35, Systems in the Time Domain,
Volume 1.
Figure 1. Graphical representation of frequency characteristics:
(a) amplitude characteristic; (b) phase characteristic; and T1 ẏ + y = x (3b)
(c) locus diagram.
it follows that
and solving the output–input relation leads to the complex
1
frequency response G( jω) = (3c)
1 + jωT1
b1 m bm 1
e jϕ G0 + jω + · · · + ( jω) |G( jω)| = (3d)
Y a0 a0
G( jω) = = 1 + ω2 T12
X 1 + jωT1 + · · · + ( jω) Tn
n n
(2c) ϕ(ω) = −arctan(ωT1 ) (3e)
This implies that substitution of nth-order differentiation
by (jω)n and nth-order integration by (1/jω)n will yield the These frequency response functions are featured in
output–input relation. Figure 3. From Figure 3(b) it can be seen that a critical
212 Foundations of Measuring
20 +j
|G 1(jw)|
−r f(w) < 0 +r
|G 2(jw)|
|G 3(jw)|
0
|G ( j w)|
|G (j w)| (dB)
wc = 1
−20 (a) −j T1
1 0
|G e(jw)| 5× 10−1
−40
|G ( jw)| (dB)
|G ( jw)|
−60 10−1 −20
1 wc 1 10 102 wc 3 3
10 5 × 10−2
|
(a) wc 2 w
p
2
10−2 −40
0.1 0.5 1 5 10 50
f2(w) wT 1
p
4 wc = 1
(b) T1
f
f3(w)
0 wc = 1
ωc1 T1
ωc2 ωc3 wT 1
fresult(w) 0.1 0.5 1 5 10 50
−p 0
4
f1(w)
f(w)
−p
2
1 10 10 2 103 −p
4
(b) w
Figure 2. Graphical determination, through addition, of the fre-
(c) −p
quency responses of series-connected systems (Reproduced from 2
Handbook of Measurement Science, Volume One, Sydenham,
1983 John Wiley & Sons Ltd). Figure 3. (a) Locus diagram; (b) amplitude characteristic and
(c) phase characteristic for a first-order system.
frequency fc (or ωc ) given by
1 This complex frequency is the variable p (or s in some
2πfc = ωc = (3f) literatures) used with Laplace transformation (see Arti-
T1
cle 37, Relationship Between Systems in the Time and
can be used to quickly generate approximations of the Frequency Domain, Volume 1). In the physical sense, p
response behavior (Woschni, 2000). means a harmonic oscillation with a exponentially increas-
Important examples of measurement system responses ing or decreasing amplitude (Woschni, 1988):
are discussed in Article 37, Relationship Between Sys-
tems in the Time and Frequency Domain, Volume 1. x = Xe δt ejωt
pt = Xe (4b)
3 TRANSFER FUNCTION If s is represented in the p-plane, the left-hand side of this
plane signifies stable solutions (δ < 0) while the right-hand
A generalization of the complex frequency response arises side leads to unstable solutions (δ > 0).
if the frequency jω is extended to the complex frequency This criterion is used for assessing the stability of systems
p = jω + δ (4a) as treated in Article 38, Stability Issues, Volume 1. The
complex frequency p used instead of ω results in the
with the increased constant δ. transfer function G(p) for the system described by (2c)
Systems in the Frequency Domain 213
p plane G ( p ) plane
+j w ∞
+j
wb
d=0 d>0
d<0
wa
w=0 1/(1+T1d) 1 1/(1+T1d)
−r +r w=∞ w=0 +r
wa
d>0
wb
d=0
−j d<0
Figure 4. Representations of the transfer function for a first-order system.
with b0 = b0 /a0 : +j
p2
b0 b b
+ p 1 + · · · + pm m
a a0 a0 p 1*
G(p) = 0
1 + pT1 + · · · + p n Tnn
m
b −r p1 +r
r
pr
r=0
a0 p 2*
= (4c)
n
p3
Tµµ p µ (a) −j
µ=0
+j
The G(p) plane is a conformal mapping of the p-plane
(Zadeh and Desoer, 1963; Woschni, 1988), which means
that the side directions of the curves remain valid and
small squares remain squares, as the next example shows.
The system with first-order delay with the frequency −r p1 +r
responses (3c, d) has the transfer function
1
G(p) = (4d)
1 + pT1 (b) −j
shown in Figure 4 with several values of δ. It can +j
p2
be observed that Figure 3 is a special case included in
Figure 4. p 1*
Im p2
−r +r
Re p2
4 POLES AND ZEROS AND ROOT LOCUS p 2*
REPRESENTATION
p3
(c) −j
By means of searching the zero points of both the numerator
pr∗ and the divisor pµ of the fraction (4c) one gains the Figure 5. (a) Splitting up the pole-zero plane of a system; (b, c)
equivalent product representation, the polynomial equation into series-connected subsystems (x poles; 0 zeros).
214 Foundations of Measuring
+j
R C
y
−1/RC +1/RC
−r +r
R R
(a) (b) −j
Figure 6. Phase-shift bridge: (a) circuit; and (b) pole-zero plane diagram.
alternative of (4c) +j
p1
(p − p1∗ )(p − p2∗ ) . . . . . . (p − pm
∗
)
G(p) = c p 1*
(p − p1 )(p − p2 ) . . . . . . (p − pn )
m
(p − pr∗ ) −r +r
r=1 p 2*
= c n (5a)
(p − pµ ) p2
µ=1 (a) −j
having the poles pµ and the zeros pr∗ of the transfer +j
function. Thus, the properties of a system are described
completely, up to a constant c, by the position of poles
and zeros, represented in the pole–zero plane (Figure 5). p3 p 1*
The poles agree with the Eigen values of the differen-
tial equation. −r +r
The frequency response can be calculated from the
p4 p 2*
pole–zero plane by
m
r=1 ( jω − pr∗ ) (b) −j
G( jω) = c (5b)
n
µ=1 ( jω − pµ )
+j
p1
Figure 5 shows how, for series connections, the pole–zero
representation of a complicated system can be split p 3*
into a sum of simpler systems. Poles and zeros at
the same point cancel each other; this feature called
compensation is used for the correction of systems by −r +r
means of additional series-connected correcting elements – p 4*
see Article 38, Stability Issues, Volume 1). The position
p2
of the poles contains information as to whether the system
(c) −j
is stable or not: existence of poles in the right half
plane signifies instability, see Article 38, Stability Issues, Figure 7. (a) System of Figure 5c; (b) all-pass system; and
Volume 1. (c) minimal-phase system.
Systems in the Frequency Domain 215
Of special interest in connection with stability problems frequency-dependent phase characteristic as follows directly
is the path of the poles as a function of the parame- from (5b). All-pass systems play an important role in
ters of systems. This method is called root locus repre- the correction of the phase characteristic (Woschni, 1988;
sentation and is the basis for system stability investiga- Haykin, 2001).
tions (Levine, 1996). Every system containing zeros in the right half plane can
be split into an all-pass system and a so-called minimal-
phase system without zeros in the right half plane, as shown
5 TYPICAL EXAMPLE: PHASE-SHIFT in Figure 7 for the system represented in Figure 5c.
BRIDGE
In measurement, the phase-shift bridge (Figure 6) is often REFERENCES
used. The transfer function is
Davies, W.D.T. (1970) System Identification for Self-Adaptive
1 1 − pCR Control, Wiley, Chichester.
G(p) = (6a)
2 1 + pCR Haykin, S. (2001) Communication Systems, 4th edn Wiley, New
York.
and the pole and zero at
Levine, W.S. (1996) The Control Handbook, CRC Press, New
1 1 York; Proceedings XVI IMEKO World Congress, Vienna.
p0 = − ; p0∗ = (6b) Woschni, E.-G. (1988) Informations Technik, Huthig, Heidelberg.
RC RC
Woschni, E.-G. (2000) Approximations in Measurement – Today
The pole–zero plane (Figure 6) shows a symmetrical still Necessary? Proceedings XVI IMEKO Congress, Wien,
position of pole and zero referred to as the imaginary 209–211.
axis. This configuration of poles has, typically, for all- Zadeh, L.A. and Desoer, C.A. (1963) Linear Systems Theory,
pass circuits, a constant amplitude characteristic and a McGraw-Hill, New York.
37:Relationship Between Systems in the Time and
Frequency Domain
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
Domain, Volume 1)
1 Laplace Transform in General 216
∞
2 Laplace Transforms of Typical Signals 217
F (p) = f (t)e−pt dt = L{f (t)} (1a)
3 Theorems of Laplace Transform 217 0
4 Important Correspondences of Laplace c+j∞
1
Transform 217 f (t) = F (p)ept dp = L−1 {F (p)} (1b)
2πj c−j∞
5 An Important Example 217
References 221 In system theory, this so-called one-sided Laplace trans-
Further Reading 221 form is made use of; only the region t ≥ 0 is of interest.
For solving optical problems, the two-dimensional Laplace
transform or Fourier transform is applied in the so-called
1 LAPLACE TRANSFORM IN GENERAL optical transfer function (Goodman, 1968).
The convergence abscissa c in (1b) is chosen in such a
In Article 29, Relationship Between Signals in the Time way that the poles remain to the left of this abscissa. Owing
and Frequency Domain, Volume 1, the relations between to the residue theorem (Kaplan, 1962),
signals in the time and frequency domain are treated. In a
similar way, the relations in systems may also be described. c+j∞
The basic idea of the calculation of the time functions, F (p)ept dp = 2πj Re s(pr ) (2a)
c−j∞
that is, g(t) or h(t), is as follows. Both the spectral
function X(jω) of the input and the frequency response
of the system G(jω) or generalized G(p) are given. It is and for a pole of nth order
then possible to derive the output caused by any of the
several sinusoidal input spectral oscillations by multiplying 1 dn−1
Re s(p0 ) = lim [F (p)ept (p − p0 )n ]
the complex spectral density X(jω) with the frequency (n − 1)! p→p0 dp n−1
response and then applying the summation (integration) (2b)
of all frequency components. This is valid because the Physical considerations lead to the following relationships
superposition law applies for linear systems. between output and input (Woschni, 1988):
In system theory, the Laplace transform is preferred to
the Fourier transform because it converges more quickly. It
L{y(t)} = L{x(t)}G(p)
can be derived by substituting jω → p in the Fourier trans-
form equation (see Article 27, Signals in the Frequency y(t) = L−1 {L{x(t)}G(p)} (3a)
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Relationship Between Systems in the Time and Frequency Domain 217
This equation yields for the transfer function G(p) means of Fourier transforms, is often used (Davies 1970,
Woschni 1988).
L{y(t)} (p)
Y
G(p) = = (3b)
L{x(t)}
X(p)
2 LAPLACE TRANSFORMS OF TYPICAL
Using the Laplace transforms of the unit step 1/p and SIGNALS
the Dirac function 1 (Table 1) provides, instead of (3b),
∞ In Table 1, typical signals including test signals as pulse,
L{g(t)}
G(p) = = L{g(t)} = g(t)e−pt dt (3c) step, and ramp function are put together with the time
L{δ(t)} 0 function and the Laplace transform.
or
∞
L{h(t)} 3 THEOREMS OF LAPLACE
G(p) = = pL{h(t)} = p h(t)e−pt dt (3d)
L{w(t)} 0 TRANSFORM
and for the calculation of the time functions
Table 2 shows theorems of Laplace transform. The initial
c+j∞
1 values are often zero because systems are at rest before
g(t) = L−1 {G(p)} = G(p)ept dp (3e) t = 0 and the input signal starts at t = 0. In this starting
2πj c−j∞
c+j∞ state, the differentiation theorem is very much easier as
G(p) 1 G(p) pt
h(t) = L−1 = e dp (3f) Table 2 demonstrates
p 2πj c−j∞ p
L{f (n) (t)} p n L{f (t)} (4)
From (3a) it follows
Using this result one gains easily, from the differential
y(t) = L−1 {L{x(t)}L{g(t)}} equation, the frequency response or transfer function; and
t
vice versa as treated in detail in Article 36, Systems in the
= x(τ )g(t − τ ) dτ =x(t) ∗ g(t) (3g)
0 Frequency Domain, Volume 1.
or
4 IMPORTANT CORRESPONDENCES OF
y(t) = L−1 {L{x(t)}pL{h(t)}}
LAPLACE TRANSFORM
d t
= x(t)h(t − τ ) dτ
dt 0 Table 3 contains important correspondences of the Laplace
d transform to be used to gain the transform of a given
= [x(t) ∗ h(t)] (3h) time function and vice versa. It is advantageous for
dt
it is possible to gain the transforms directly without
giving the convolution theorem of the Laplace trans- solving the appropriate equations as shown in the next
form (Table 2). Figure 1 demonstrates this fact. By means example.
of these relations, the deconvolution problem may be
solved (Davies, 1970; Woschni, 1988).
5 AN IMPORTANT EXAMPLE
−1 1
x(t) = L L {y(t)}
G(p) As an example, let us deal with a temperature sensor as
t
shown in Figure 2. With the temperature of the environment
= y(t)g ∗ (t − τ ) dτ = x(t) ∗ g ∗ (t) (3i)
0
δe and the temperature of the sensor δs , the volume V , the
surface S, the specific density c, the specific heat or thermal
with equivalent χ and the heat-transfer coefficient α one obtains
1 the differential equation
g ∗ (t) = L−1 (3j)
G(p)
dδs
cχV + αS(δs − δe ) = 0
Deconvolution often becomes extremely difficult because dt
it is generally not possible to realize the inverse sys- cχV
tem functions. An alternative approach to the solution, by δ̇ + δs = δe (5a)
αS s
218 Foundations of Measuring
Table 1. Laplace transforms of test signals.
A periodic test signal function Equation Laplace transform
x (t ) ∞ (Dirac) pulse with integral value F⊥
∋ ∋ 0
1F ∞ for t = 0 F⊥ (p) = F⊥
∋ ⊥ x(t) = F⊥ δ(t); δ(t) =
0 for t = 0
t
x (t ) Step function
A A
x(t) = Aw(t); w(t) = 0 for t < 0 F (p) =
1 for t > 0 p
t
x (t ) Ramp function
A = tan f
0 for t ≤ 0 E (p) =
A
f x(t) =
At for t ≥ 0 p2
t
x (t ) Rectangular pulse
A 1 − e−pT
x(t) = 0 for t < 0, t > T E (p) = A
A for 0 < t < T p
T t
x (t ) Triangle pulse
A 2
0 for t ≤ 0, t ≥ T −p
T
2A 2
2A t for 0 ≤ t ≤ T
E (p) = 1 − e
T T t x(t) = T
2
p p
2
t T
2A 1 − for ≤t ≤T
T 2
x (t ) Trapezium pulse
A
0 for t ≤ 0, t ≥ T
T
A A (1 − e−aT p )(1 − e−(1−a)T p )
t for 0 ≤ t ≤ aT E (p) =
aT T(1−a) t x(t) = AaT aT p2
for aT ≤ t ≤ (1 − a)T
A
t
1− for (1 − a)T ≤ t ≤ T
a T
x (t ) Half-sine pulse
A
0 for t ≤ 0, t ≥ T Aπ (1 − e−pT )
π E (p) = 2
x(t) = A sin t for 0 ≤ t ≤ T p π
T t T p +
2
2 p
Reproduced from Handbook of Measurement Science, Volume One, Sydenham, 1983 John Wiley & Sons Ltd.
Relationship Between Systems in the Time and Frequency Domain 219
Table 2. Theorems of Laplace transform.
1. Addition theorem
L{f1 (t) + f2 (t)} = L{f1 (t)} + L{f2 (t)}
2. Multiplication theorem
L{af (t)} = aL{f (t)}
3. Shifting theorem
for a > 0 L{f (t − a)} = e−pa F (p)
a
or L{f (t + a)} = epa F (p) − e−pt f (t) dt
0
4. Likeness theorem
1 p
L{f (at)} = F if a>0
a a
5. Attenuation theorem
L{e−at f (t)} = F (p + a) if Re (p + a) ≥ p0 > 0
6. Limit theorem
lim f (t) = lim pF (p); lim f (t) = lim pF (p)
t→∞ p→0 t→0 p→∞
7. Integration theorem
t
1
L f (τ ) dτ = L{f (t)} if Re (p) > 0
0 p
8. Differentiation theorem
L{f (n) (t)} = p n L{f (t)} − p n−1 f (+0) − · · · − f (n−1) (+0)
if the limits
lim f (t) = f (+0); lim f˙(t) = f˙(+0); ...;
t→0 t→0
lim f (n−1) (t) = f (n−1) (+0), exist
t→0
9.–10. Convolution theorem
If the integrals e−pt f1 (t) dt and e−pt f2 (t) dt both are absolutely
convergent or at most one absolutely and the other conditionally
convergent yields
L{f1 (t)}L{f2 (t)} = L{f1 (t) ∗ f2 (t)}
with
t t
f1 (t) ∗ f2 (t) = f1 (τ )f2 (t − τ ) dτ = f1 (t − τ )f2 (τ ) dτ
0 0
220 Foundations of Measuring
Table 3. Laplace transforms of time-dependent functions.
Time function, f (t) Laplace transform, F (p) Time function, f (t) Laplace transform, F (p)
t ap
0 0 sinh(at)
2 (p 2 − a 2 )2
sin(at) a
δ(t) 1 arc tan
t p
1
w(t)
p
2 sinh(at) p+a a
g(t) G(p) log = 2 arc tanh
t p−a p
1 p cos b − a sin b
h(t) G(p) cos(at + b)
p p2 + a 2
t n−1 1 p + 2a 2
2
cos2 (at)
(n − 1)! pn p(b2 + 4a 2 )
1 p 2 + 2a 2
eat cosh2 (at)
p−a p(p 2 − 4a 2 )
t n−1 at 1
e
(n − 1!) (p − a)n
eat − ebt 1 t 2ap
cos(bt) sin(at) arc tan
a−b (p − b)(p − a) 2 p 2 − a 2 + b2
2
aeat − bebt p cos(bt) − cos(at) p + a2
2 log
a−b (p − b)(p − a) t p 2 + b2
2
a cosh(bt) − cosh(at) p − a2
sin(at) 2 log
p2 + a 2 t p 2 − b2
2ap a sin(bt) − b sin(at) 1
t sin(at)
(p 2 + a 2 )2 ab(a 2 − b2 ) (p 2 + a 2 )(p 2 + b2 )
Reproduced from Handbook of Measurement Science, Volume One, Sydenham, 1983 John Wiley & Sons Ltd.
x (t ) * g (t ) y (t ) Time domain The time functions are obtained using (3e, f)
c+jω
1 ept
g(t) = dp (5c)
2πj 1 + pT
L−1
c−jω
L
c+jω
1 ept
h(t) = pp (5d)
2πj c−jω p(1 + pT )
x (p ) • G (p ) y ( p ) Frequency domain
and using the correspondences found in Table 3
Figure 1. Convolution relationships.
1 −t
g(t) = e T (5e)
T
t
h(t) = 1 − e− T
de
ds (5f)
aS (ds − de )
V
The time constant is then
Figure 2. Temperature sensor.
V
T = cχ (5g)
Using (4) the transfer function is αS
1 cχV So it follows that the relation of V /S = constant and
G(p) = with T = (5b)
1 + pT αS has to be small for good dynamic behavior, which means
Relationship Between Systems in the Time and Frequency Domain 221
the diameter d of the sensor has to be so small as design Woschni, E.-G. (1988) Informations Technik, Huthig, Heidel-
conditions allow. berg.
REFERENCES FURTHER READING
Davies, W.D.T. (1970) System Identification for Self-Adaptive Hsu, H.P. and Hse, H.P. (1995) Schaum’s Outline of Signals and
Control, Wiley, Chichester. Systems, McGraw-Hill, New York.
Goodman, J.W. (1968) Introduction of Fourier Optics, McGraw- Levine, W.S. (1996) The Control Handbook, CRC Press,
Hill, New York. NJ.
Kaplan, W. (1962) Operational Methods for Linear Systems, Oppenheim, A.V., Willsky, A.S. and Hamid, S. (1996) Signals
Addison-Wesley, Reading, MA. and Systems, Prentice Hall, NJ.
38: Stability Issues
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
3. Decreasing eigenfunctions occur; this means that the
1 Concept of Stability 222 weighting function g(t) fulfills the condition
2 Stability Criteria 222 c
3 Root Locus Representation 223 lim |g(t)| dt ≤ M < ∞ (1c)
c→∞ 0
4 Feedback and Feed Forward as an Important
Example 224 From examination of these stability conditions, certain
5 Optimization of Systems 225 stability criteria can be applied to test the situation to
estimate its degree of stability, for instance, using the
6 Deconvolution 226
Hurwitz and Routh test.
7 Testing of Systems 226 If the following differential equation is given
References 227
Further Reading 228 an y (n) + an−1 y (n−1) + · · · + a0 y = b0 x + · · · + bm x (m)
then all coefficients ar , and the determinants
1 CONCEPT OF STABILITY
a1 a0 0 . . . 0
a3 a2 a1 a0 0
Let us deal, as an introduction, with an amplifier having µ = .. .. .. .. .. (1d)
. . . . .
amplification V as shown in Figure 1. If a part of 1/V of the
a a2µ−2 . . . aµ
output y is coupled back to the input x with the right phase 2µ−1
(feedback), the amplifier will produce its own output y
without there being an input. The system becomes unstable.
The system is stable if one of the following conditions have to be positive, that is, ar > 0, µ > 0.
is met:
1. The eigenvalues of the differential equation pr exhibit 2 STABILITY CRITERIA
a real part less than zero, that is
For practical applications, graphical methods based on the
locus diagram representations are of value. The transfer
Re(pr ) = δr < 0 (1a)
function G(p) consists of polynomials in the numerator
N (p) and the divisor D(p):
2. On the right-hand side of the pole-zero plot there are
N (p)
G(p) = (2a)
no poles of the transfer function; (1b) D(p)
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Stability Issues 223
1/V
If the zero point of the D(p)-plane remains left of the
locus curve D( jω), passes from ω = −∞ to ω = +∞,
the system is stable. If not, it is unstable (Figure 3) to
some degree.
y
The procedure is also suitable for developing a statement
V
x on the stability margin (see Woschni, 1988; Zadeh and
Desoer, 1963 and Section 4 of this article).
Figure 1. Amplifier with a feedback.
3 ROOT LOCUS REPRESENTATION
Application of the stability condition (1b) shows that the
polynomial of the divisor D(p) is not permitted to have As described in (1b), the locus of the poles of the transfer
zeros in the right-half plane including those lying upon the function G(p) facilitates the test of the stability of a system.
imaginary axis (2b) (stability limit). The transfer function
For testing the stability of a given system, the divisor
of the frequency response, that is, D( jω), is drawn on the m
bν ν
b0 b1 m bm p
complex plane (locus diagram) forming a closed curve by +p +··· +p a
a0 a0 a0 ν=0 0
inclusion of negative frequencies. Because of the conformal G(p) = = n (3a)
1 + pT1 + · · · p n Tnn
mapping, the unstable field is always the one lying on the Tµµ p µ
right-hand side of the locus curve drawn from ω = −∞ to µ=0
ω = +∞ (see Figure 2). To test for stability, it is necessary
to verify where the zero point is situated. may be written in the form (Woschni, 1988; Levine, 1996),
see also Article 36, Systems in the Frequency Domain,
Volume 1
p -plane D-plane
+j w (p − p1∗ )(p − p2∗ ) . . . (p − pm
∗
)
jw G(p) = c
w1 (p − p1 )(p − p2 ) . . . (p − pn )
D(jw)
m
w (p − pr∗ )
1)
(jw
D r=1
=c (3b)
Unstable
Unstable
w=+0 w=∞ n
−r 0 w=−0 w =− ∞ + r (p − pµ )
d<0 0 d>0 µ=1
This means the roots of the divisor are the poles of G(p)
and thus they are responsible for stability. Stability means
w that there exist no poles in the right-hand side of the pole-
−j
zero plane (Figure 4).
Figure 2. Locus diagram of the divisor D(p). Sampled-data systems with the sampling time TS yields
z = epTS , (Woschni, 1988; Levine, 1996)
+j
m
+j (z − zν∗ )
(z − z1∗ )(z
− z2∗ ) . . . (z
− ∗
zm )
ν=1
D(jw) G(z) = c =c n
(z − z1 )(z − z2 ) . . . (z − zn )
(z − zµ )
D(jw) µ=1
(3c)
Here, instead of the right-hand side of the pole-zero plane,
−r w=∞ +r −r w=∞ +r as with (1b), the region outside the unit circle is the unstable
w one. So poles have now to be inside this circle to achieve
w stability as Figure 4 shows.
−j The main idea of the root locus representation is to
−j
(a) (b)
allow movement of the poles by varying system parameters
when investigating the stability. The method was intro-
Figure 3. (a) Unstable system and (b) stable system. duced by Evans (1954) and is applied to find the critical
224 Foundations of Measuring
p -plane G(p)-plane G(z)-plane
+j w−∞
wb +j +j
d<0 d=0 d>0
wa Stable
w −0 1/(1+T1d) 1/(1+T1d) w= p w=0
TA
−r +r w−∞ w−0 +r −1 +1
Stable Unstable Unstable w = 2p
TA
d>0 wa Unstable
−j d=0 wb −j
d<0 Stable
Figure 4. Mapping of the p-plane to the G(p)- and the G(z)-plane.
amplification V of closed-loop systems (automatic control +j
systems) as dealt with in Section 4.
w=∞ C1C2
−r −1 w=0 +r
4 FEEDBACK AND FEED FORWARD AS
AN IMPORTANT EXAMPLE
As an example of a commonly met system in instrument
systems, consider the feedback system shown in Figure 5.
For negative feedback (control situation), the frequency
response is given by (see Section 3 of Article 35, Systems G1(jw) G2 (jw)
in the Time Domain, Volume 1)
G1 (jω)
Ge (jω) = (4a) −j
1 + G1 (jω)G2 (jω)
Figure 6. Testing stability of closed-loop systems (control).
The D( jω) function is
1 + G1 (jω)G2 (jω) (4b) a stable to an unstable state. This fact may be caused by
nonlinearity and may lead to unstable oscillatory behavior,
Instead of testing this function, it is more convenient to (see Section 3 of this article).
verify that the open-loop frequency response A very important device is the operational amplifier
shown in Figure 7.
G1 (jω)G2 (jω) (4c) Using (4b) the frequency response is with |Z 1 | |Z 2 |
and V → ∞
fulfills the condition related to the point ‘–1’ as shown
in Figure 6 (Nyquist diagram). The diagram illustrates that
Y V 1 Z
an increasing amplification V will take the system from = V∗ = = = 2 (4d)
X V Z 1 Z Z1
1+ 1
+ 1
Z1 + Z2 V Z2
x x1 y1 y
G1
+
Case 1: + Z1 Z2
Case 2: −
y2 x2 V ^
G2 ^ Y
X
Figure 5. Feedback system. Figure 7. Wired-up operational amplifier.
Stability Issues 225
Table 1. (a) Structurally stable and (b) structurally unstable systems..
a)
c c c c
G1,2 (jω) c
1 + jωT1 jω jω 1 + jωT1
c c
G2,1 (jω) c c c
1 + jωT2 1 + jωT2
b)
c c c
G1,2 (jω) c
jω −ω2 jω
c c c
G2,1 (jω) c
jω jω(1 + jωT2 ) 1 − ω2 T22
Source: Reproduced from Handbook of Measurement Science, Volume One, Sydenham, 1983 John Wiley & Sons Ltd.
With Z 2 = R2 and Z 1 = 1/jωC it is a differentiation net- Optimum
work model
Optimum criterion
V x = jωCR2 (4e)
− Computing
x0 instruction
and with Z 2 = (1/jωC) and Z 1 = R1 an integrator (Miller-
integrator, bootstrap-generator)
Following
Real system correction
1 system e.g.
Vx = (4f) computer
jωCR1 ki
Table 1 also compares structurally stable and unstable
Figure 8. Principle of optimization.
systems.
ki becomes optimum, we have a static optimization. In
5 OPTIMIZATION OF SYSTEMS dynamic optimization, on the other hand, a process x(t)
is sought such that a performance function depending on a
Optimization is treated here specifically with respect to sys- changing functional process assumes an extremal value.
tem theory. Both dynamic errors and errors caused by dis- The previous considerations imply that the central need
turbances, for example, noise, are taken into consideration. of optimization is the creation of a suitable optimiza-
The behavior of measuring systems depends upon a tion criterion.
number of parameters ki , for example, time constants Tr , In most cases, the criterion of the mean-square error,
which, at least to a certain extent, may be freely selected. as shown in Figure 9, is used. If Greal ( jω)Gcorr ( jω) is the
If the parameters of the system cannot be changed, an frequency response of the system corrected by a following
optimization can be effected by a following correction (also correction system Gcorr ( jω) as shown in Figure 8, we obtain
called compensation) system or by a correction system the mean-square error (Woschni, 1988)
connected in opposition, the parameters of such a system ∞
being adjustable. In the commonly found microprocessor- ε2 = Sxx (ω)|Gid (jω) − Greal (jω)Gcorr (jω)|2 dω
based applications, the behavior of the system is set up by 0
∞
suitable programming.
+ Szz (ω)|Gcorr (jω)|2 dω
The fundamental idea is shown for a one-dimensional 0
problem in Figure 8. The difference between the output
= ρ 2 + Pz (5)
variables of the ideal optimum system (model) and the real
system, with a following correction system, is formed and
Ideal system
assessed according to the computing instruction defined yid
by the optimization criterion. The parameters ki of the e e2 e2
correction system are set so that the real system resembles − Quadrature Mean value
x
the optimum system as far as possible.
Distinction is made between static and dynamic optimiza-
tion (Bellman, 1961) according to the definition of the per- Real system
yreal
formance criterion. If the system is dimensioned such that a
performance function depending directly on the parameters Figure 9. Mean-square error generation.
226 Foundations of Measuring
with the assumption having been made that the disturbances
+j
occur prior to the correction computer.
xp1
6 DECONVOLUTION op1*
In Article 35, Systems in the Time Domain, Volume 1,
the convolution integral is introduced to obtain the output −r +r
y(t) if the input x(t) and the weighting function of the op 2*
system g(t) are known (Levine, 1996; Woschni, 1988).
xp2
t
y(t) = g(t)x(t − τ ) dτ (6a) (a) −j
0
+j
The inverse problem is of great importance in measuring
systems and is known as deconvolution. There the output
signal y(t) is measured with the input x(t) or the wanted xp3 op1*
signal. The behavior of the measuring device, the weighting
function g(t), or the transfer function G(p) = L{g(t)} is
known. System theory and Laplace transformation rules
−r +r
yields (Woschni, 1988; Levine, 1996)
t xp4 op 2*
x(t) = g ∗ (τ )y(t − τ ) dτ (6b)
0
(b) −j
With the weighting function g ∗ (t) of the inverse system
+j
∗ −1 1
g (t) = L (6c) xp1
G(p)
As introduced in Article 36, Systems in the Frequency
op3*
Domain, Volume 1, the transfer function of the measuring
system G(p) may be described by the pole-zero plane
plot. Figure 10 shows this plane for a general nonminimal- −r +r
op4*
phase system containing a minimal-phase system and an
all-pass system. The inverse system g ∗ (t) now has poles
at the positions where the original system has zeros and xp2
vice versa. −j
(c)
Until there are no zero points of the original system in the
right-half plane of the pole-zero plane (so-called minimal- Figure 10. (a) Nonminimal-phase system (b) an all-pass system;
phase system), it is possible to realize the inverse system. and (c) a minimal-phase system.
If the degree of the divisor is greater than the degree of the
counter, this is possible as a limiting case. If the original
Because of the relationships between performance in time
system is a nonminimal-phase kind, the zero points of the
and frequency domains, it is possible to calculate either of
all-pass in the right half of the pole-zero plane will be
these functions if the other one is measured.
poles of the inverse system. This means that the system
Often, it is the availability of test equipment that decides
will be unstable and a realization of the deconvolution is
which of these characteristic functions is to be found.
not possible. In this case, it is only possible to fulfill the
For instance, to test the frequency response of a temper-
amplitude conditions of (6c) and to use an additional all-
ature sensor, it is easier to rapidly change its environ-
pass unit to gain a constant and increasing delay time.
ment (impulse stimulus giving a time domain response)
than it is to apply cyclically varying temperature (fre-
7 TESTING OF SYSTEMS quency domain).
Testing of systems is performed by means of test signals
Testing of systems is the task of system identification. produced by a test generator, recording the corresponding
Stability Issues 227
Table 2. Survey of test signals.
Characteristic function Input test signal Output function Characteristic values
Differential equation Not specified Not specified Time constants Tr
In the frequency domain Harmonic oscillation Frequency response G( jω); Critical frequency fc ; ωc
jωt
x = Xe Transfer function G(p)
In the time domain Unit step function w(t) Unit Transient response h(t) Transient time tr , dead time td ,
pulse = Dirac delta Weighting function g(t) delay time tl , compensating
function δ(t) time tc , overshoot x0
Stochastic functions White noise Cross-correlation function Transient time ttr
S(ω) = constant ∧ g(τ )
ψxy (τ ) = Correlation time tcorr
x,y
Output
Test generator System w *(t )
representation
1
h (t )
Figure 11. Testing of systems.
h*(t )
output signals as illustrated in Figure 11. Table 2 gives a T t
survey of the test signals used, the output signals, and the
characteristic values used for approximations. Today, the Figure 12. Testing with a nonideal step function w ∗ (t).
process of obtaining the characteristic functions is usually
automated, making use of digitally programmable function
generators that are controlled by microcomputers, to form In practice, it is usually sufficient that the rise time of the
the appropriate input signals. step function is smaller than one magnitude of the transient
In particular, the frequency response G( jω) is obtained time of the system to be tested.
by measuring both phase angle and proportion of the Of practical importance, especially for self-adaptive sys-
amplitudes of output to a given sinusoidal input. tems, are those methods of identification that use the noise
Before recording the true values, it is necessary to wait as the input of a system (Davies, 1970). The weight-
until the steady state solution appears, that is, t > ttr = ing function g(t) may be calculated by deconvoluting the
1/(2fc ). relation
∞
If the measuring device itself has a nonideal frequency
ψxy (τ ) = g(t)ψxx (t − τ ) dt (7c)
response Gx ( jω), Gy ( jω), the real frequency response 0
G( jω) may be calculated from the incorrect value G∗ ( jω)
by means of the relation (Woschni, 1972) with ψxx (t) autocorrelation function of the input and ψxy (t)
cross-correlation function.
Gx (jω) Because the deconvolution technique becomes difficult,
G(jω) = G∗ (jω) (7a)
Gy (jω) methods in the frequency domain or special noise sources
are used (Davies, 1970). Use of white noise yields
Other principles of calibration make use of comparison
systems or reciprocity principles for systems with reversible ψxy (τ ) = constant × g(τ ) (7d)
operation (Woschni, 1972).
If the characteristic functions in the time domain, g(t) which means that the weighting function becomes a con-
of h(t), are to be found, a problem arises in that the input stant multiplied by the cross-correlation function (Woschni,
signals are not the ideal ones, as shown in Figure 12 for a 1988). For more details, including errors arising from use of
nonideal step function w∗ (t). Instead of the real transient a nonideal correlation function, and the reduction of errors
response h(t), h∗ (t) is used: see Davies (1970).
t
d
h∗ (t) = w∗ (τ )h(t − τ ) dτ (7b)
dt 0 REFERENCES
By means of a deconvolution, it is possible to gain the real Bellman, R. (1961) Adaptive Control Processes, Princeton Uni-
transient response where w∗ (t) is known (Woschni, 1972). versity Press, Princeton, NJ.
228 Foundations of Measuring
Davies, W.D.T. (1970) System Identification for Self-Adaptive Zadeh, L.A. and Desoer, C.A. (1963) Linear System Theory,
Control, Wiley, Chichester. McGraw-Hill, New York.
Evans, W.R. (1954) Control System Dynamics, McGraw-Hill,
New York.
FURTHER READING
Levine, W.S. (1996) The Control Handbook, CRC Press, New
York. Hsu, H.P. and Hse, H.P. (1995) Schaum’s Outline of Signals and
Woschni, E.-G. (1972) Meßdynamik, Hirzel, Leipzig. Systems, McGraw-Hill, New York.
Woschni, E.-G. (1988) Informations Technik, Huthig, Heidel- Oppenheim, A.V., Willsky, A.S. and Hamid, S. (1996) Signals
berg. and Systems, Prentice Hall, NJ.
39:Characteristics of Data, Information, Knowledge,
and Wisdom
Timothy Lindsay John Ferris
University of South Australia, Mawson Lakes, South Australia, Australia
which the phenomenon may be either a tangible attribute of
1 Definition of Data and its Characteristics 231 tangible objects, such as mass, or some intangible attribute,
2 Definition of Information and its such as intelligence. In any case, the instrument applied
Characteristics 231 must be of a suitable kind to distinguish manifestations
3 Definition of Knowledge and its of the phenomenon in such a manner as to effect a map-
Characteristics 232 ping to a scale used for description of the phenomenon.
This relationship is illustrated in Figure 1, which shows in
4 Definition of Wisdom and its Characteristics 233
symbolic form the relation of the observed phenomenon,
5 Relationship between Data, Information,
X, which is an attribute of something, usually a tangi-
Knowledge, and Wisdom 233
ble object, which may or may not itself be tangible. The
Related Articles 234 measuring instrument transforms this to a parameter, ,
References 234 which is the data achieved by the measurement process.
The transformation linking the phenomenon to the param-
eters is called , to symbolize that this transformation is a
physical transformation.
1 DEFINITION OF DATA AND ITS
The domain of the mapping, , is the possible value of
CHARACTERISTICS the attribute manifestation in the observed, X; the range
is the mapping of that domain into the data space of the
Data in a measurement system is the raw form of what possible parameter set, . A data set is produced when
comes out of the measuring equipment. Measurement appa- instruments are correctly applied to the observed. The data
ratus is applied to the phenomenon, which is observed may take many forms but it is an image of the observed
and produces whatever representation of the observed phe- reduced to the dimensions detected by the instrument used.
nomenon is consistent with the transformation effected by The image is ‘flatter’ than the observed because it could
the instrument. A measurement system has two essential be investigated with respect to multiple attributes, but the
elements involved in the process of production of data; image is the result of observation with respect to only
being the observed phenomenon and the measuring appa- a subset of the whole. This section is based on (Ferris
ratus. Data is the result of the application of the apparatus 1997a, b, Section 8.3).
to the phenomenon. The observed phenomenon has some
state, which follows from the fact that there is some real-
ity present, and that state is observed by means of the 2 DEFINITION OF INFORMATION AND
application of a standard instrument of a kind, sensitive ITS CHARACTERISTICS
to variations of some kind in the observed phenomenon.
The explanation above is necessarily abstract, because it Information is a step up on data in complexity and
intends to consider the broad range of possible scenarios, in value. Measurement data are the symbols representing the
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
232 Foundations of Measuring
in the message representation. The quantity of data ISh is
Observed
X dependant on such measurement system characteristics as
Ψ
the range and resolution of the instrument, which together
define the number of states of the observed phenomenon
Parameters that the measurement apparatus is capable of distinguishing.
Π However, the representation of the data demands an integer
number of bits, and so the minimum possible value of ISh
Figure 1. Illustration of the relation of the observed phe- is rounded up to the nearest integer above the number of
nomenon, X, which is observed using a measurement instrument resolvable states of the observed.
performing the transformation, , to produce the data, para-
meters, . IDr pertains to the resolution of meaning effected by the
message and so relates to the number of distinct cases that
can be represented by the data in the system. The number of
observations, whereas measurement information is the rep-
resolvable states of the observed is the number of possible
resentation of the observation, which is embodied in the
meanings in the data received. Therefore, certainty of a
data (Crouch and Sydenham, 1994). Data is not informa-
particular state reduces the uncertainty about the state of
tion, but data contains information, which is the significance
the observed by defining which one of Pm possible states
of the data, and is the vehicle for shifting information
is present. The consequence is that IDr is as defined above.
around. Depending on the representation scheme, the sym-
The above discussion makes it clear that ISh ≥ IDr . This
bols may or may not appear to correspond with their
is proved in Ferris (1997b, Th 3.3).
significance. Information must be conveyed symbolically
because significance cannot be directly transferred (Ayer,
1969; Jolley, 1973). 3 DEFINITION OF KNOWLEDGE AND
The distinction between data and information can be use- ITS CHARACTERISTICS
fully drawn by considering the work done in quantification
of ‘information’ in different disciplines. Let the measure of Knowledge is a more complex and valuable entity than
information developed by Shannon in communication the- information, but is dependant on information. Information
ory be ISh (Shannon, 1964) and that developed by Dretske is not knowledge but is the foundation upon which knowl-
in philosophy be IDr (Dretske, 1981). In this analysis, ISh edge is built. Information is transformed into knowledge
is found: when integrated with other information about the observed
ISh = k × ln Pf (1) phenomenon, which enables the observer to increase under-
standing of the instance of the observed phenomenon
where k ∈ is a constant through the observations that have been made.
The concept of knowledge is commonly associated with
Pf is the number of possible message forms communication effected through natural language and this
is a useful point to develop an understanding of knowledge
and IDr is found: in a measurement situation. Natural language communi-
cates knowledge, that it has meaning, through the existence
IDr = k × ln Pm (2) of many definitions, relating to words and the implica-
tions of grammatical constructions; conventions, relating to
where k ∈ is a constant the interpretation of the combination of linguistic expres-
sion and context; and intention, relating to the purpose for
Pm is the number of possible message meanings. which a particular speaker assembles a particular combina-
tion of words to assemble a means of communication of a
Both these measures of information quantity are impor- thought (Burge, 1975). In a natural language communica-
tant in measurement theory since instruments produce a tion situation, knowledge arises because the listener hears
representation of the observed in the form of parameter the discourse and interprets it in the light of the combina-
data, which must be communicated within the system, and tion of the known definitions, conventions, and perceived
information is carried by that data as a means of commu- intentions of the speaker.
nicating something of substance about the observed. To derive knowledge about the observed phenomenon is
ISh relates to the channel capacity required to convey the a common, and worthy, purpose of measurement activities.
measurement data in a digital form. The number of possible But, in the measurement situation, the data and the informa-
message forms is the total number of possible different tion that the data contains arises in a quite different situa-
messages that could be conveyed using the number of digits tion than natural language. Measurement information arises
Characteristics of Data, Information, Knowledge, and Wisdom 233
from nature, not intention, since measurement information
Primary
is a statement of the state of a particular phenomenon at the system
Other
time of observation (Butcharov, 1964). Therefore, interpre- inputs
tation of measurement information to produce knowledge
Controlling Sensor
of the observed is not affected by the factors used in input
interpretation of natural language, these being definitions,
conventions, and intentions.
Figure 2. A generalized automatic feedback control system
To obtain knowledge from measurement information block diagram.
requires that information existing because of the nature
and structure of the world be assigned significance by
required to adjust the behavior of the primary system so
the interpreter of the information. Situations, such as
that the output becomes the desired output. Determination
measurement, in which information exists independent of
intention are fundamentally different than cases where of the control input is nontrivial because the control system
information exists as a result of intention (Lewis, 1946; is a dynamic system, which normally involves complicating
Brodbeck, 1968). Where information arises out of nature, factors such as time delays, with the result that a simplistic
the interpretive approach of perception, in which the adjustment of the control input is liable to introduce
information is recognized as containing evidence that the significant problems such as overshoot, undershoot, or
observed phenomenon is in a state evidencing one of oscillation of the output relative to the desired output value.
some finite number of meaningful categories, is required. In a measurement-based decision system, the contribution
Perception is a process of recognizing that an observed of the measurement system is to provide knowledge that
phenomenon corresponds to a class of phenomenon that classifies the observed phenomenon in a way that enables
the perceiver knows and so the case observation is linked a decision to take appropriate action. The action may be
into a meaningful structure of knowledge of the world, and taken by either a person or a machine. In a measurement-
so the observed state of the observed become the basis of based decision system, the action taken is not a feedback
knowledge of the contextual significance of the state of control action of seeking to adjust the controlled output,
the observed. but rather is an action of an apparently unrelated kind
that is an appropriate action to take on the basis of the
measurement observations. For example, in medicine, a
4 DEFINITION OF WISDOM AND ITS patient may be measured in one way or multiple ways,
CHARACTERISTICS resulting in knowledge of the state of the patient, embodied
in a diagnosis. The action of wisdom, performed by the
Wisdom is the ability to judge the appropriateness of action doctor, is a particular therapeutic maneuver, performed to
and to behave in the correct manner in a situation. Wisdom correct the situation evidenced by the state of the patient,
has been recognized and valued since ancient times, having but not as a mechanistic adjustment of a control input that
formed one of the traditional branches of philosophy. adjusts the observable output to lead to the conclusion that
Some ancient philosophies were essentially embodiments the patient’s state is satisfactory.
of wisdom, with examples of this kind of writing found
in Confucius and parts of the Old Testament, amongst
other sources. 5 RELATIONSHIP BETWEEN DATA,
In measurement systems, wisdom relates to the taking
of appropriate action on the basis of the observations
INFORMATION, KNOWLEDGE, AND
made through the sensors. This occurs in two ways, the WISDOM
automatic feedback control system and systems in which
measurements contribute to decisions. Data is the lowest level of entity that is used in a system for
In an automatic control system, illustrated in Figure 2, processing matters of meaning concerning things. Data can
there is a controlled variable, the output, which is observed be transformed into information when the data is organized
using a sensor. The signal produced by the sensor is used into a form that distinguishes between resolvably different
as an input to the primary system to adjust the output. cases of the observed phenomenon. Knowledge is gained
The intention is to have a system in which the actual when information from one or more sources is integrated
output approximates as closely as possible, at all times, and analyzed to produce a coherent understanding of the
the desired output. matter of observation. Wisdom concerns the performance of
In the case of a feedback control system, ‘wisdom’ is appropriate action based on integration of various sources
defined in the control loop design as the control input of knowledge. Wisdom normally involves analysis of items,
234 Foundations of Measuring
+ Other knowledge advances up the stack because the simplicity of represen-
analysis and prognostication tation results from increased integration of various fac-
tors into the representation of the observed effected at
Wisdom
Feed each level.
In summary, a measurement process is used to col-
+ Other information
integration lect data about the observed phenomenon; that data is
Knowledge organized as information about the observed and leads to
Feed
the knowledge of the state of the observed phenomenon
when integrated with other information about the observed;
Information and the knowledge, when integrated with other knowledge
Organization
about the observed and its context provides a basis for the
exercise of wisdom.
Data
Figure 3. A pictorial representation of the relationship between RELATED ARTICLES
data, information, knowledge, and wisdom.
Article 13, Signals, Information and Knowledge, and
Meaning, Volume 1; Article 14, Hierarchical Aspects of
probably many, of knowledge, and some kind of modeling Measurement Systems, Volume 1; Article 24, Principles
of the situation in order to provide prognostication of the of Epistemology as Related to Measurement, Volume 1.
outcome resulting from various possible actions. The ideas
of this paragraph are expressed in Figure 3.
REFERENCES
Each of the elements, data, information, knowledge, and
wisdom is a member in a hierarchy of both complexity, Ayer, A.J. (1969) Metaphysics and Common Sense, Macmillan,
order, and value. The relative sizes of the elements in London.
Figure 3 is also significant. Following the relation stated Brodbeck, M. (1968) Meaning and Action, in The Philosophy of
at the end of the Section 2 that discusses information, it Science, (ed. P.H. Nidditch), Oxford University, London, (pp.
is, in general, true that the quantity of information is less 97–120).
than or equal to the quantity of data in a measurement Burge, T. (1975) On Knowledge and Convention. The Philosoph-
system. Knowledge, if quantified, would be less than or ical Review, 84(2), 249–255.
equal to the quantity of information because knowledge Butcharov, P. (1964) Knowledge of Meanings and Knowledge of
concerns categorization of the observation into meaningful the World. Philosophy, 39(148), 145–160.
categories of distinct. The quantity of wisdom is the Crouch, V. and Sydenham, P. (1994) Relationship between T&E
smallest because there is always one best solution, although and M&I’, in Australasian Instrumentation and Measurement
Test and Evaluation Conference, Adelaide (pp. 1–6).
in many situations it may be unclear what the best solution
is, and one rational and justifiable solution is selected from Dretske, F.I. (1981) Knowledge and the Flow of Information, MIT,
Cambridge.
a small group of candidate actions instead.
Ferris, T.L.J. (1997a) The Concept of Leap in Measurement Inter-
The entities are situated in a stack of increasing order,
pretation. Measurement Journal of the International Measure-
because each entity is an expression of a more ordered ment Confederation, 21(4), 137–146.
arrangement of representation of the state of the observed Ferris, T.L.J. (1997b) Foundation for Medical Diagnosis and
phenomenon and the appropriate action arising from that Measurement, PhD thesis, University of South Australia.
state. The order is found in the fact that the entities each Jolley, J.L. (1973) The Fabric of Knowledge, Gerald Duckworth,
express the understanding of the observed phenomenon in London.
greater simplicity of representation of the observed. Lewis, C.I. (1946) An Analysis of Knowledge and Valuation, Open
The complexity of representation decreases as one pro- Court, La Salle.
gresses upwards in the stack of elements. However, the Shannon, C.E. (1964) The Mathematical Theory of Communica-
complexity of the elements themselves increases as one tion, University of Illinois, Urbana (pp. 29–125).
40: Sources of Information
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
internal reports within organizations, in the personal records
1 How Information Propagates from Discovery 235 of workers, or in primary published literature. At times,
2 Information Flow from Sources to books, often regarded as secondary sources, will contain
Depositories 235 primary material.
3 Problems of Scale 236 Within a few months of being established (weeks in
some cases), primary information, of the publicly open
4 Enter the Internet 237
kind, will often be published in primary journals where
Related Articles 238 it is available through library services. Material published
Reference 238 in primary journals is usually subject to expert refereeing
that helps maintain, but cannot guarantee, the quality of
reported material.
1 HOW INFORMATION PROPAGATES Primary information is also available in student theses
(dissertations), patent specifications, and in some confer-
FROM DISCOVERY
ence proceedings. A considerable part of it is held in
company and defense secret locations where it is not openly
The world contains an infinite amount of latent informa-
tion – that residing within systems, waiting to be estab- available. It is also to be found in the proceedings of confer-
lished. The creation of available information begins with ences, a source that may not be as well refereed as reputable
people who convert some of the latent information into a primary journals; but it is usually available sooner.
recordable form that suits the storage classification then in There are no universal rules to regulate what is pub-
vogue. Figure 1 depicts how information passes around a lished, or where it is published. Primary information often
circuit in which the information that is useful to measure- also appears in part in the next class, the secondary outlets.
ment finds its way back to be the basis of the design of Academic researchers work with the moral duty to publish
more measuring instruments, which in turn aids the flow of their findings, but the same does not apply to many others.
more information. Because something cannot be located in open depositories
is no guarantee that it has not been externalized from the
latent state.
2 INFORMATION FLOW FROM Primary information is, for a variety of reasons, not
SOURCES TO DEPOSITORIES always presented in the form that is required, so it gets pro-
cessed in many ways. It may need to be collected together
Figure 2 provides an appreciation of the flow of informa- to form a work on a common topic, or be reexpressed in
tion through the many kinds and places where information a more acceptable form for others to make use of it. In
may be found. Scholarship and experimental observation this class are secondary journals, reviews, books, ency-
lead to latent information being made available; the first clopedias, dictionaries, videotapes, DVDs, CDROMs, the
stage is termed primary information. This may occur as Internet, audiotapes, and films.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
236 Foundations of Measuring
Measurement
Cross-discipline techniques flow
use of measurement back for reuse
and modification
Techniques applied
Information Discipline 1
Information converted to
Untapped knowledge via Discipline 2
latent flow
measurement
information plus other skills Discipline n
The system Academic endeavor Disciplines systematizing Application of
under study (research and knowledge to suit the knowledge
teaching) times
Figure 1. Extraction of latent information leads to new measuring instruments.
People (in organizations)
Invisible colleges doing research
making measurements
thinking
Primary sources
Conference Periodical
Reports Patents Theses
papers articles
Condensation
and control
Bibliographies
Catalogs
Abstracting and Review
indexing publications articles
Current awareness
Information services
Secondary sources
Data
Books Handbooks Encyclopedias Dictionaries
compilations
Figure 2. Flow of information through recorded, retrievable systems of documentation. (Reproduced from Melton (1978) with
permission of Institute of Physics.)
Secondary information is usually easier to apply, for it presentation can present a primary viewpoint, often being
distills the essence from the primary reports, leaving out original in nature.
the detail that does not stand the test of time or that is of
little bearing to its use.
A major kind of published secondary source is that com- 3 PROBLEMS OF SCALE
prising the reference material that aids retrieval of primary
material – bibliographies, abstracts, catalogues, informa- The rate at which information is published is almost incom-
tion services. prehensible. By 1900, the Royal Society in London had to
Reviews and bibliographies are formed from secondary abandon its attempts to catalog all of the scientific publi-
information but their construction, classification, and cations from 1800 to 1900 – the Author Index alone was
Sources of Information 237
19 volumes in length! In 1974, an estimated 80 000 regular The Internet has made information available to just about
scientific and technical journals in a total of 150 000 of all every person on the Earth at a comparatively affordable cost
kinds of journals were being published. Abstracting period- and in almost any geographical location.
icals then numbered 1500, in a total of 3500. An estimate of On the surface it would seem that the Internet gives
that time suggested there were 300 000 books published in its millions of subscribers all they need to know for free.
a year with over 3000 million books in print. By the end of Searches will indeed often discover considerable material
the twentieth century, this increased exponentially to vast on a topic of one’s choice. The reality is, however, that
numbers. The rate is not slowing. the freely available information may not have the veracity
The diverse nature of measurement means that infor- required, its time stamp is not always clear, and it rarely
mation needed could rest almost anywhere in the vast provides a sufficient depth of knowledge to allow serious
system of documentation, if it exists at all. Well-devised decisions to be taken from it.
retrieval is essential to glean information in an efficient Publishers of the traditional technical books have placed
manner. considerable emphasis on offering book material online
via their highly developed Web sites. These sources of
knowledge can be relied upon better and do give organized
4 ENTER THE INTERNET material that is refereed and recorded in a highly profes-
sional and concentrated manner. To provide this service,
To provide a more reliable method of communication infor- they must charge fees by subscription or, along the way, as
mation, the United States created what has today become a ‘pay as you use’ service.
known as the Internet or simply the Web. This has grown Benefits of having information stored in the electronic, or
over the past decade into a vast network of millions of e-file form, as compared to the traditional hard copy form,
computers that are all connected together via the servers of are many:
Internet Service Providers (ISP). The Web is now predom-
inantly used by civilians all over the globe. • Access is to any part of it with immensely faster find
These computers offer information on a virtually infinite times for specific knowledge than by using books.
range of topics. To find the required information, the user • Hugely larger sets of information can be made available
makes use of search engines to mine down into the vast at virtually any location.
resource to select those instances that appear relevant. The • Parts can be selected and reused to form new sets of
user then opens them to see if they are what is needed; knowledge (where this is permitted).
many are not, and much time can be wasted in seeking • Upgrading the information is a matter of editing files;
what is not there! no reprinting of all of the information with its changes
Search engines currently only allow users to locate mate- is needed.
rial on topics of their choice by using symbol searches • Foundational knowledge will become clear and settle
for words and strings in text files. Being developed, and down to allow expansion from sound foundations and
more widely available soon, will assuredly be the facility not be rejigged for each new edition of a book.
to search images for contextual graphical constructs; this
uses ontologies for topics, these being large sets of relevant Measurementor, www.measurementor.com, Sydenham
rules about the structure and content of images of the sup- (2004), is an example of this form of information delivery.
ported domain. For example, the textile industry is adopting Some of its statistics illustrate the problem that will be faced
this method as a means to find patterns from large numbers by those needing up-to-date and definitive knowledge.
of different designs now stored as images in computers. Creating the basic knowledge base listing for it showed
Software applications are available, which use this method that the sound set of information needed by a measure-
for searching images. The development of the ontologies ment system designer would involve over 4000 one-page
needed is a costly business. A typical ontology will contain items and that even this set of information could only serve
over 100 000 rules. satisfactorily as an entry gateway to their fuller needs. It
The availability of the Internet can be likened in the time has been estimated by this writer that at least 1000 books
span of man to the epoch when the printing press of the would be needed to provide a deep library of information
1400s made information available to many more people of for a profession. All of the titles could not be economically
that time. Then, recorded information was only available created by one publishing source alone. Internet and gate-
to a privileged few in painstakingly produced single copies way products, like Measurementor, can provide this degree
of books of the monks, and in the art of religions such as of access, for it allows thousands of commercial providers
stained glass windows and paintings. to offer material that sits in their specialty area.
238 Foundations of Measuring
Several Web sites give useful information on measuring e-sources of information will become routinely linked to
systems. They are useful but too often only get the user parts of all people’s lives, especially in support of their
acquainted with the topics. They also are not coordinated professional duties. Already the law and medical profes-
with respect to their information, nor structured to any sions have taken up Internet use to give their practi-
scheme that stands much scrutiny. tioners rapid and full access to their particular bodies of
GlobalSpec, http://www.globalspec.com/, claimed to be knowledge. Others will follow as the cost of time and
the world’s largest database of technical products and ser- time taken to locate information becomes a major cost
vices, provides rapid location and access to a vast number of factor in such areas as teaching, design, and procure-
items – 9000 suppliers catalogues, 850 000 product families ment fields.
and 45 million parts. Many are supported with explanations
given in its tutorials and application notes of its Techni-
cal Library. The aim of the service is to sell products and RELATED ARTICLES
services to users on behalf of its clients.
The Sensors magazine Web site http://www.sensorsmag. Article 1, Sophistication of Measurement and its Body
com/ provides a Literature Showcase for many compa- of Knowledge, Volume 1; Article 2, Organization of
nies and gives access to the past articles of the Sensors Instrument Science, Volume 1; Article 5, Humans in
magazine. Search facilities are provided to assist in find- the Real World, Volume 1; Article 41, Terminology and
ing material. Classification of Measurement Systems, Volume 1; Arti-
Many manufacturers of measuring systems and instru- cle 42, Information Databases of Relevance to Measure-
ments offer information assistance via their own Web site. ment, Volume 1.
An example of the upper end of this type of information
support is that of National Instruments http://www.ni.com/.
In their offerings is a service called ‘Instrupedia’ – a short REFERENCE
encyclopedia of measurement that will be found to be useful
at times. Sydenham, P.H. (2004), Developing a Measuring Systems Body
The impact of the Internet is already clearly evident. It of Knowledge, Proc.10th IMEKO TC7 International Sympo-
seems reasonable to suggest that within the next decade, sium, June 30–July 2, Saint-Petersburg, Russia.
41:Terminology and Classification of Measurement
Systems
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
meanings. It is only in recent times that standardization
1 Terminology Used in Measurement 239 of terms has begun to be accepted; they are not, as yet,
2 List of Key Terms Relevant to Measurement 239 widely used nor broadly developed. Data bases relevant
3 List of Inspec Terms Relevant to to measurement knowledge are discussed in Article 42,
Measurement 240 Information Databases of Relevance to Measurement,
Volume 1.
4 Classes of Publications Relevant to
Measurement 241 The wide and often incorrect use of synonymous terms
makes literature retrieval difficult. Apparently obvious
5 Library Classification Schemes 241
terms often lead nowhere – metrology, measurement, sen-
6 List of Dewey Classification Codes Where sor, and instrumentation in particular; they are too com-
Measurement Material is Located 242 monly used. The name instrument science, for example,
7 List of Journals Relevant to Measurement 245 might be indexed using at least 10 alternatives, such as
Related Articles 245 measurement systems, measurement physics or measure-
ment science, and so on. The fundamental term metrology
also has at least two different meanings and it often gets
1 TERMINOLOGY USED IN confused with ‘meteorology’.
Terms describing the activity of the process of measure-
MEASUREMENT ment exist in great abundance. Figure 1 lists some of these.
It is easy to see how many different terms are used for
Information retrieval by browsing through original material describing the same thing.
makes use of many recognition features – words, illustra-
tions, mathematical formulas – but this method is invariably
too slow for searching the bulk of the literature.
Realistic retrieval procedures are almost exclusively 2 LIST OF KEY TERMS RELEVANT TO
based on the principle of tracing linguistic terms used by MEASUREMENT
authors in their titles, abstracts, and index keys. Conse-
quently, the choice of terms written into abstracts and used
in searching is of paramount importance. Measurement, in many specific disciplines, has its own
The diversity and uncoordinated multiplicity of the name. Figure 2 lists a compilation of these. Thus, the
sources of generation (and regeneration!) of instrument literature of measurement is spread over the whole body
concepts has brought with it a highly varied and inconsis- of knowledge. Use of a large-size dictionary usually gives
tent nomenclature. Indeed, some terms used have multiple little assistance in separating them, for it is their technical
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
240 Foundations of Measuring
Accounting Effect Prospection Approximation theory Non-electric sensing devices
Accredit Evaluation Quality Artificial intelligence Nuclear instrumentation
Adjustment Expectation Rate making
Administer Experiment Recognize Automatic testing Oceanographic techniques
A measure of Exploration Resolution Character recognition Optical systems design
Analyse Features Sensing Characteristics measurement Optical instruments
Appraisal Group Standardize
Computerized instrumentation Parameter estimation
Appreciation Incidence Style
Arbitrate Indication Taxonomy Cybernetics Patient diagnostics
Assay Influence Tests Data acquisition Performance index
Assessment Information Value Data processing Physical instrumentation control
Choice Inspection Weighing up
Data reduction and analysis Physiological models
Classification Interpretation Worth
Classify Judge Digital instrumentation Predictor−corrector methods
Composition of Logistic Display instrumentation Probes
Criticism Management Dynamic testing Production testing
Criticize Material constants
Decide Measurement Education Program testing
Decision Mensuration Electric variables measurement Quality control
Design Metricate Environmental testing Quantization
Detection Monitoring
Error analysis Radiation monitoring
Determining Number
Diagnosis Performance Error detection Recorders
Distribution Precision Fundamental law tests Relay protection
Document Properties Geophysical equipment Research and development
management
Figure 1. Terms describing the activities of the process of Height measurement Reviews (see also published lists
measurement. of bibliographies with author
indexes)
History Robots
Accounting Measurement systems engineering
Anthropometrics Measurement technology Impulse testing Seismology
Archaeometry Mensuration Information science Self-organising systems
Assay Metallography Instrumentation Sensitivity analysis
Automatic testing Metrology Instruments (53 kinds fall into Sensory aids
Biometry Nuclear engineering this)
Cybernetics Pathology
Decision theory Posology Laboratory apparatus and Signal processing
Diagnostics Psychrometrics techniques
Econometrics Quality control Logic testing Social and behavioural sciences
Epidemiology Quantity surveying Machine testing Spatial variables measurement
Ergonomics Radiometry
Magnetic variable measurement Standardization
Faunistics Scientific instrumentation
Geophysics Seismology Man-machine systems Student laboratory apparatus
Instrument physics Sensitometry Materials testing Switchgear testing
Instrument-science Sensory physiology Measurement (29 specific variables Telemetering systems
Instrument technology Stoichiometry are listed: all appear elsewhere).
Management Symptomatology This is also used instead of term
Man–machine-systems Taxonomy metrology
Measurement engineering Teleology
Measurement physics Time and motion study Measurement standards Testing
Measurement-science Value engineering Thesauri
Measurement systems Transducers
Figure 2. Some names of disciplines involved with measurement. Measurement theory Units (measurement)
Mechanical testing Value engineering
Medical diagnostic equipment Water pollution and detection
usage that is of interest, and dictionaries are not good at
Meteorological instruments
defining things in those terms.
Noise measurement
Nomenclature and symbols
3 LIST OF INSPEC TERMS RELEVANT Figure 3. Selected terms related to measurement in the INSPEC
thesaurus.
TO MEASUREMENT
Use of databases, in particular, requires careful choice of given in a standard thesaurus are allowable inputs. Figure 3
terms in the search profile. In some of them, only terms lists a selection of over 650 (of a total 10 000) terms that
Terminology and Classification of Measurement Systems 241
relate to the activity of measurement these being contained A report can be placed into an outlet that emphasizes,
in the thesaurus of the INSPEC database. with an example of each, its
• principle used (Hall effect)
4 CLASSES OF PUBLICATIONS • philosophical thinking (impact on the philosophy of
science and technology)
RELEVANT TO MEASUREMENT • application area (manufacturing)
• device (SQUID)
Consider now the necessary classification of a measurement • technology (MEMS)
classification from the author’s and reader’s viewpoints. It • theory (sensitivity analysis)
is found that numerous possibilities exist for deciding the • proprietary aspect (XYZ new product)
key list of index terms. • fashionable topic (fuzzy systems)
In general, authors contributing to measurement align • industry (user survey), and so on.
more with their field of application or discipline than with
sources publishing fundamentals of measurement science. Many primary journals publish measurement contribu-
Often, they see the instrument content as insignificant and tions. Those more obviously dedicated to measurement are
not worth abstracting in their report. Many facets of a report many. Figure 4 gives a list of measurement-related journals
might not be cited in the few keywords used. that could be chosen for publication of the above report. It
The reality is that the report on a topic including mea- also gives the rationale for the choice of each.
surement, although given several descriptors, should only Other aspects might be that it relates to include trans-
be published in one journal. Ethics suggest this is the correct portation, power generation, environmental issues, civil
policy, but there is a case for multiple releases in different engineering, environmental safety, and many more.
types of publications so that the contribution can be found Authors should pay attention to the need to prepare
by those who need it, as they approach the material from several papers on a major topic, each on its key aspects
different perspectives. and for a different readership.
As an example, consider a hypothetical report contain-
ing, as part only, the calibration of a novel, laser-based,
alignment system used, by the XYZ Company, to test 5 LIBRARY CLASSIFICATION SCHEMES
the military load-carrying capacity of an historic concrete
bridge while the bridge is conveying a mobile nuclear Now consider classification from the viewpoint of location
reactor. Just where the report should be published is an in a library collection. The location of book literature is
open question. identified in a library by the classification code-numbering
Example: Calibration of a novel, laser-based, alignment system used, by the XYZ
Company, to test the military load-carrying capacityof an historic concrete bridge while
the bridge is conveying a mobile nuclear reactor.
Class basis chosen Example of typical journal for publication
Physical principle used (interferometer) Applied Optics
Discipline based on a device used (laser) Laser Focus
Contemporary nature (novel and timely) New Scientist
Military relevance Classified report of a military organization
Discipline of use (surveying) Photogrammetric Engineering
Parameter of measurement (measurand is Journal of Strain Analysis
related to strain)
Measurement principle used (angle Measurement Science and Technology
measuring interferometer)
Constructional material used (concrete) Magazine of Concrete Research
Standardization (calibration) Newsletter of National Conference of
Standards Laboratories
Design implications Journal of Elasticity
Testing International Journal of Non-destructive
Testing
Instrumentation detail Instrumentation and Control Systems
Engineering heritage Transactions of the Newcomen Society
Trade journal XYZ Affairs (hypothetical name)
Figure 4. Classes and example journals in which an instrument might be reported.
242 Foundations of Measuring
scheme used. Placement and retrieval are based on an from the Dewey Classification. Overall, the system contains
allocated code number to each title, not by the use of a over 30 000 named code numbers of which at least some
set of key terms, as is the case in computer retrieval of 700 are measurement related.
journal literature abstract information. For most effective searching, the full classification would
Several schemes of library location classification are in need consulting – a matter of many days of work! Library
use. Predominant systems in use are catalogs will have an alphabetical subject index in their
entries but this will seldom provide the detail sought.
1. Dewey Decimal Classification; The wide choice of measurement topics requires searching
2. Library of Congress (LC); on the basis of a detailed study of the handbook of the
3. Universal Decimal Classification (UDC). classification system – but even then a book might not be
Other schemes used are local subject code allotment and in the class to which a user thinks it should be allocated.
even author order, but only in some cases. Subject retrieval It is appropriate here to explain the development of the
by that alone is virtually impossible in this last case. Dewey Decimal System, for this has a strong bearing on
Electronic catalogs have made it easier to find material, for current day problems with its use in class engineering and
they can search in words in the title, extended keywords, technology titles.
and abstract where used. The Dewey system was devised in the late nineteenth
Each of the established systems has its prime-listing century when technical publishing was in its infancy. Its
directory – such as the Dewey Decimal Classification and structure is a reminder of who did the intellectual thinking
Relative Index. at that time. Its construction did not cater to the subsequent
The UDC system is fundamentally supported by an explosion of technical knowledge. For this reason, subjects
extensive set of British Standards Institution standards doc- that often began as quite a reasonable and small subset of
uments, which enable great detail in subject identification. knowledge have had to be allocated a relatively extended
For instance, one code number is specifically for the cards code length, thus going deep into the numbering system.
used to print weight in automatic weighing machines! In For that reason, already long code numbers had to be
practice, some libraries using the UDC system make use of greatly extended to cope with what are now major sub-
a much abridged version. sets of knowledge. For example, the 620.0–629.0 series
UDC and Dewey are somewhat alike, but not identical. was devoted to engineering and allied operations. This
The LC system is quite different in code allocations. Equiv- required electrical measurements to have a long call number
alence between LC and Dewey codes is published. of 621.381043.
The diverse and ubiquitous nature of measurement often All this contrasts with the nineteenth-century awareness
makes classification of such material a difficult decision. It of time measurement that allocated the code 529.78 to that
is publishers who assign a class call code to a book, that topic. It can be seen that length of the call number is not
being part of their publishing support – but those that assign an indicator of depth of specialist topics.
them are seldom experts on measurement literature. Most Many aspects of modern measurement are not placed in
acquisition librarians will make use of this call number, the Dewey Decimal Classification where they could now be
for they also will not be expert enough to challenge the expected to be in a library. For example, telegraphy (384.1)
one allocated. is placed in the Social Science 300’s group. Holography
It is, therefore, quite possible for a book to be placed (774) appears in the Arts 700’s group along with games
in one of many alternatives – it is not unheard of for theory at 795.01. Computing has all but taken over class
a second edition of a book to be coded differently than 1.0, originally assigned to knowledge.
the first! In some, but not all, groups, there is a specific code for
relevant instrumentation for that application. An example
is that of flow properties and their measurement (532.053).
6 LIST OF DEWEY CLASSIFICATION This is not at all standard practice – many principles or
CODES WHERE MEASUREMENT application areas of measurement do not possess such a
MATERIAL IS LOCATED specific measurement aspect call number.
Occasionally, to add yet more difficulty, a new volume
The lack of structure of the knowledge of measurement of the classification index itself is published. This requires
science is reflected in library holdings: material will be some changes to codes (deletion, placement elsewhere, new
found diffused throughout many classes. An impression of code numbers). This produces only minor changes in the
the situation is obtained by the study of the selected list of location of titles in the established libraries as it generally
subjects and code numbers given in Figure 5 that is taken affects only books cataloged after that event.
Terminology and Classification of Measurement Systems 243
Dewey Decimal code Classification volume entry
001.422 Statistical method (formerly 311.2)
001.51 Communication theory
001.53 Cybernetics
001.533 Self-organizing systems
001.6 Data processing
003 Systems
016 Bibliographies and catalogs of specific disciplines and
subjects. Note: Other codes used such as —.0 of notation
119 Number and quantity
120 Knowledge, cause, purpose, man
121 Epistemology
152 Physiological and experimental psychology
169 Analogy
330 Economics
339.3 Measures of national income
364.12 Criminal investigation (Detection)
368.011 Rates and rate making (Insurance)
371.26 Educational tests and measurements
378.16 Educational measurement and student placement
384.1 Telegraphy
384.9 Visual signaling
389 Metrology and standardization
501 Philosophy and theory
502.8 Techniques, apparatus, equipment materials
511.43 Error theory
512.7 Number theory
515.42 Theory of measure and integration
519.52 Theory of sampling
519.86 Quality control and other statistical adjustments
522.2–522.5 Instruments (astronomy)
526.3 Geodetic surveying
529.75 Time systems and standards
530.16 Measurement theory
530.7 Instrumentation (for measurement, control, recording)
530.8 Physical units, dimensions, constants
531.382 Elastic constants and their measurement
532.053 Flow properties and their measurement
534.4 Measurements, analysis and synthesis of sound
535.33 Optical instruments (see also specific functions class)
Microscopes formerly 578.1
536.5 Temperature (536.51–536.54 Measurement)
539.77 Detection and measurement of particles and radioactivity
542.3 Measuring apparatus (laboratory-chemistry)
543.07 Instrumentation (chemistry)
544 Qualitative analysis
545 Quantitative analysis
548.1 Geometrical crystallography
551.4 Geomorphology (.4607 is Deep sea surveys and exploration)
573.6 Anthropometry
574.028 (no key title given but biological instrumentation here see 610.28)
578.4 Use of microscopes
591.18 Movements senses, control mechanisms (Zoology)
610.28 Medical instrumentation (see also 574.028)
611.8 Nervous system and sense organs
612.8 Neurophysiology and sensory physiology (extensive sub grouping)
614.7 Environmental sanitation and comfort (614.71–614.77 cover
pollution but no specific class for measurements)
615.19018 Assay methods (pharmacy)
616.07 Pathology (includes diagnoses and prognoses)
617.05 Surgery by instrument and technique
617.752 Optical work
Figure 5. Sample list from the likely Dewey Decimal codes that relate to measurement.
244 Foundations of Measuring
Dewey Decimal code Classification volume entry
617.89 Audiology
617.9178 Surgical instruments, apparatus, material
620.0042 Engineering analysis, synthesis, design
620.0044 Testing and measurement (engineering)
620.1127 Non-destructive testing of materials (subgroups on
radiographic, tracer, ultrasonic, magnetic methods)
620.32 Measurements (of mechanical vibration)
620.72 Systems analysis, synthesis, design
621.313 Generating machinery and converters (includes maintenance and testing)
621.37 Electrical testing and measurement
621.372 Units and standards of measurement
621.373 Recording meters
621.374 Measurement of electric quantities (various subgroups exist)
621.379 Measurement of non-electrical quantities
621.381043 Measurements (electronic)
621.38137 Testing measurements standardization (Classifier instructed to
place measurements in group for application)
621.381548 Testing and measuring devices and their use (Electronic)
621.383 Specific instruments and apparatus of wire telegraphy
621.38417 Measurements and standardization (radio)
621.3843 Specific instruments and apparatus of radio telegraphy
621.3887 Measurements and standardization (television)
621.3895 Underwater devices (subgroups include hydrophics and Sonar)
621.4831 Reactor physics (includes testing of physical phenomena
occurring within reactors)
621.4835 Operation, control, safety measures
621.756 Inspection technology (in factory operations engineering)
621.902 Machine tools (including numerical control and other
automation techniques)
621.994 Measuring tools
622.15 Geophysical exploration (several subgroups)
622.8 Mine health and safety
623.71 Intelligence and reconnaissance topography (military engineering)
623.819 Design tests (nautical)
623.863 Nautical instruments
624.1513 Soil mechanics (includes tests)
624.17720–624.1779 Specific structural elements (includes strength tests)
625.794 Traffic control equipment
627.81 General principles (Dams and reservoirs) (includes surveying)
628.177 Measurement of consumption (sanitary engineering)
629.045 Navigation (celestial in 527)
629.1345–629.1346 Aircraft tests, measurements, standards, maintenance, repair
(see subgroups)
629.135 Aircraft instrumentation and systems (many subgroups)
629.273 Panel instrumentation (vehicles)
629.4775 Control of temperature, humidity, air supply and pressure
634.9285 Mensuration (forestry)
637.127 Quality and purity determinations (cows milk)
640.73 Consumer education (includes guides to quality and value of
products and services)
657.48 Analytical (Financial accounting) (Measurement of profitability,
of financial strength, of income, of liquidity, of
flow of funds)
658.28 Equipment for safety and comfort (includes noise control)
658.516 Standardization (equipment, procedures, in general management)
658.562 Standards and specifications (form, size, dimensions, quality,
materials, performance, including standardization formerly
in 658.16)
658.568 Inspection (includes statistical methods of quality control)
658.834 Consumer research (studies of consumer preferences,
attitudes, motivations, behaviour)
Figure 5. (Continued ).
Terminology and Classification of Measurement Systems 245
Dewey Decimal code Classification volume entry
660.283 Process equipment (chemical instruments, apparatus, machinery)
662.622 Properties, tests, analysis (of coal)
664.07 Tests, analysis, quality controls (food—for texture, taste,
odor, colour, contaminants)
665.0288 Tests analyses, quality controls (oils, fats waxes, greases)
666.13 Tests, analyses, quality controls (ceramics)
669.95 Metallography
674.13 Properties and their tests (lumber)
675.29 Properties, tests, quality controls (leather, fur)
676.121 Properties, tests, quality controls (Pulp)
676.27 Properties, tests, quality controls (paper and paper products)
681 Precision instruments and other devices (several groups)
697.9315 Psychrometrics
697.9322 Temperature controls
771.37 Focusing and exposure apparatus
774 Holography
781.91 Musical instruments
788.971 Mechanical and electrical reproducers
913.0310285 Interpretation of remains (includes dating, use of data
processing in ancient studies)
Figure 5. (Continued ).
Australian Journal of Instrumentation and Control 7 LIST OF JOURNALS RELEVANT TO
British Communications and Electronics
Control MEASUREMENT
Control Engineering
Electronic Engineering Figure 6 is a selected list of journals of major relevance
Engineers Digest
Flow Measurement and Instrumentation
to measurement. These cover application areas such as
IEE Proceedings – Science, Measurement and Technology medicine, agriculture, space, and manufacturing, and will
IEEE Control Systems Magazine also invariably contain useful material on measuring sys-
IEEE Instrumentation and Measurement Magazine tems for their domain.
IEEE Sensors Journal
IEEE Transactions on Instrumentation and Measurement
IEEE Transactions on Ultrasonics, Ferroelectronics and
Frequency Control RELATED ARTICLES
ISA Transactions (Instrumentation, Systems and
Automation Society)
Journal of Applied Measurement Article 1, Sophistication of Measurement and its Body
Journal of Optical Society of America A: Optics Image of Knowledge, Volume 1; Article 2, Organization of
Science and Vision Instrument Science, Volume 1; Article 5, Humans in
Journal of Optical Society of America B: Optical Physics
Journal of Optics A: Pure and Applied Optics the Real World, Volume 1; Article 42, Information
Laser Focus World Databases of Relevance to Measurement, Volume 1.
Measurement
Measurement and Control
Measurement Science and Technology
(formerly Journal of Physics E: Scientific Instruments)
Metrologia
Nature
NCSL Newsletter
New Scientist
Physics Today
Review of Scientific Instruments
Science
Sensors and Actuators A: Physical
Sensors and Actuators B: Chemical
Sensor Review
Strain
Figure 6. Some journals relevant to measuring systems.
Information Databases of Relevance to
42:
Measurement
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
and technical information in the field of measurement and
1 Commonly Used Information Sources for instrumentation and the enhancement of international coop-
Measurement 246 eration among scientists and engineers from research and
2 INSPEC Abstracting Service 248 industry. It was founded in 1958. It offers information on
3 NCSLI Services 248 the many meetings of its Technical Committees but does not
provide detailed database information about measurement.
4 National Laboratories 248
NI (National Instruments), http://www.ni.com provides a
5 NIST Services 248 range of support services. One information service offered
6 IMEKO Book Survey 248 is a database of some 2000 instrument solutions; another
7 Personal Lists 249 is its Instrupedia, an Encyclopedia of Instrumentation pro-
Related Articles 249 vided on CD-ROM. The glossary part is provided online.
Reference 249 Ei COMPENDEX http://edina.ac.uk/compendex is a
comprehensive interdisciplinary engineering information
database having over six million summaries of journal
articles and conference proceedings and 220 000 new
1 COMMONLY USED INFORMATION additions every year.
SOURCES FOR MEASUREMENT It uses thesaurus-style indexing. Citations are arranged by
subject and numbered sequentially; abstracts are provided.
There exist some, but not many, Internet-based services Each entry is indexed by article title, author name, and affil-
that provide valuable information on measuring systems as iation, journal or proceeding source, date of publication,
their main line of support. Major organizations that offer and page length. Access to this database is via personal
lists, guides, and relevant databases, INSPEC, National or shared accounts. It covers mechanical, civil, environ-
Conference of Standards Laboratories International (NCSL) mental, electrical, structural, process, materials science,
and National Institute for Science and Technology (NIST), solid-state physics and superconductivity, bioengineering,
are covered in later sections. energy, chemical, optics, air and water pollution, solid-
Global Quality Services www.GlobalQualityServices. waste management, hazardous waste, road transportation,
com offers a search engine in the field of metrology. It transportation safety, manufacturing, quality control, and
indexes a collection of Web sites in the field of metrology. engineering management issues. Thus, while it does not
IMEKO (International Measurement Confederation) http: have a category for measurement, considerable content will
//www.imeko.org is a nongovernmental federation of 34 be found on that topic by careful searching.
national member organizations individually concerned with GIDEP (Government-Industry Data Exchange Program)
the advancement of measurement technology. Its objectives http://www.gidep.corona.navy.mil is a cooperative activ-
are the promotion of international interchange of scientific ity between government and industry participants seeking
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Information Databases of Relevance to Measurement 247
to reduce expenditures of resources by sharing technical presented at major scientific meetings around the world.
information essential during the research, design, develop- Subject emphasis since 1995 has been in the life sciences,
ment, production, and operational phases of the life cycle environmental sciences, and the aquatic sciences, while
of systems, facilities, and equipment. Its content contains older material also covers physics, engineering, and mate-
metrology data, such as material constants. Utilization of rials science. Information is derived from final programs,
GIDEP data can materially improve the total quality and abstracts booklets, and published proceedings, as well as
reliability of systems and components during the acquisition from questionnaire responses.
and logistics phases of the life cycle, and reduce costs in ENVIROLINE http://library.dialog.com covers the world’s
the development and manufacture of complex systems and environment related information. It provides indexing and
equipment. Again, it is not set up to support measurement abstracting coverage of more than 1000 international pri-
as such but can be useful. mary and secondary publications reporting on all aspects of
ISI (Science Citation Index ) http://www.isinet.com/isi/ the environment. These publications highlight such fields as
products/citation/sci provides access to current and retro- management, technology, planning, law, political science,
spective bibliographic information, author abstracts, and economics, geology, biology, and chemistry as they relate
cited references found in 3700 of the world’s leading schol- to environmental issues. It corresponds to the print material
arly science and technical journals covering more than a Environment Abstracts.
hundred disciplines. The Science Citation Index Expanded GEOARCHIVE http://library.dialog.com is a comprehen-
format and the online version, SciSearch , cover more than sive database covering all types of information sources
5800 journals. in geosciences, hydrosciences, and the environmental sci-
BIOSIS http://www.biosis.org has the objective of facil- ences. The criteria for inclusion in GeoArchive are that the
itating understanding of the living world by helping re- source should be publicly available and have relevant infor-
searchers, educators, students, and others in accessing infor- mation content, even if the reference is to a small news item
mation relevant to the life sciences. It is an entire life in a magazine. It provides, from 1969, international cover-
sciences file corresponding to Biological Abstracts (BA) age of over 5000 serials, books from over 2000 publishers,
and Bio Research Index (BioI). It includes, from 1969, geological maps, and doctoral dissertations. It is indexed
biomedical material and is sourced from over 9000 serials. using Geosaurus, the hierarchical thesaurus that classi-
Biology Browser www.biologybrowser.org is an inter- fies references by subject, geography, and stratigraphy. It
active portal designed for the life sciences community. includes an alphabetical index of around 30 000 terms avail-
It creates a dynamic environment that connects users able from the producer. The printed publications Geotitles,
with valuable information through interactive discussion Hydrotitles, Geoscience Documentation, and the Bibliogra-
forums, quality-controlled Web links, science news, and phy of Economic Geology are derived from GeoArchive.
other free resources. ISMEC Mechanical Engineering Abstracts http://www.
CHEMCON http://www.chemcon.net is a global network cas.org/ONLINE/DBSS/ismecss.html is a bibliographic
of chemical conferences in the field of chemical control database (from 1973) containing in-depth coverage
regulations. of worldwide literature on mechanical engineering,
CDI http://www.virtualref.com Comprehensive Disserta- engineering management, and production engineering.
tion Index (not to be confused with another CDI, the Center Theoretical perspectives and applications for current
for Defense Information) contains all dissertations at aca- research results are included. Records contain bibliographic
demic doctoral level granted in the United States and in information, controlled and uncontrolled terms, and
over 125 other-country institutions. It covers from 1861 abstracts. A specific topic area is Measurement and Control.
and grows at over 500 citations a month. Since 1976, cov- NTIS (National Technical Information Service) http://
erage has been expanded to include European institutions, www.ntis.gov is the central source for US government
and later other institutions worldwide – but their coverage scientific, technical, and business information. It leads to
is rather patchy. titles on research and development activities published
IFI CLAIMS http://www.ificlaims.com is a 45-year old, by over 200 US Federal Agencies. Audiovisual materials,
multifaceted, database holding data on US patents. It is computer data-files, and software are included.
offered as being a trusted and comprehensive source of US WPI (World Patent Index ) http://www.derwent.com files
patent information. (partially, since 1963) patent specifications issued by the
CPI Conference Papers http://library.dialog.com has pro- major industrial nations. It corresponds to CPI and WPI
vided, since 1973, a Conference Papers Index (CPI) that and grows at over 250 000 new inventions per year.
contains over 800 000 entries of papers given at over These few examples illustrate the various kinds of ser-
1500 major conferences. It provides citations to sessions vices that are available. Many may need to be searched
248 Foundations of Measuring
to locate measurement information; there seems to be no 4 NATIONAL LABORATORIES
easier method than to patiently work through them.
Technological countries have at least one of these car-
rying the responsibility for developing and maintaining
2 INSPEC ABSTRACTING SERVICE the prime physical standards. They additionally offer staff
expertise through consultancies and by participation in
industry clubs.
INSPEC http://www.iee.org/Publish/INSPEC is produced The National Physical Laboratory (NPL), UK, http://
by the Institution of Electrical Engineers. It is a major www.npl.co.uk is an example. Their bibliography of staff
English-language bibliographic information service provid- publications can be useful at times but this site does
ing access to the world’s scientific and technical literature not appear to offer lists and guides, as does the next of
in physics, electrical engineering, electronics, communica- the NIST.
tions, control engineering, computers and computing, and
information technology. The INSPEC database was formed
in 1969, being based on the Science Abstracts service pro- 5 NIST SERVICES
vided by the Institution of Electrical Engineers since 1898.
It provides Current Awareness, New Product Information, The National Institute for Science and Technology, NIST,
Technological Forecasting, Competitive Intelligence, and US – formerly the National Bureau of Standards (NBS) –
Patent-Related Searching. Over 3400 scientific and techni- http://www.nist.gov offers a vast array of pull-up informa-
cal journals and some 2000 conference proceedings, as well tion services. One of regular value is a guide to the use of
as numerous books, reports, and dissertations are scanned SI units.
each year for relevant articles to abstract and index for
inclusion in the Database. It contains over 7 million biblio-
graphic records and is growing at the rate of 350 000 records 6 IMEKO BOOK SURVEY
each year. It also publishes an INSPEC Thesaurus of terms
that contains over 10 000 terms, of which at least 800 are The practical problems of establishing the existence of book
relevant to a search for an instrument-related subject. As material on measurement science and technology were rec-
they are diffused throughout the whole set, considerable ognized in 1978 by the Higher Education Committee (TC-1)
time is needed to properly choose suitable key terms. For an of the International Measurement Confederation (IMEKO).
explanation of the difficulties of finding measuring systems It was decided that the Committee should establish a
literature see Article 41, Terminology and Classification list of titles for free distribution to member nations of
of Measurement Systems, Volume 1. IMEKO, and any other interested user. The bibliography
was completed in 1980 (Sydenham, 1980).
Initially it was thought that the list would be quite
short and that key articles in the serial literature could
3 NCSLI SERVICES also be included. Entries were to be for works in the
four official IMEKO languages – English, French, German,
NCSLI (National Conference of Standards Laboratories, and Russian.
International) http://www.ncsli.org was established in 1961 It was soon established that computer searching, at that
as a professional association for individuals engaged in time, was not productive for book location. The best
all spheres of international measurement science. It pro- approach was found to be to select entries from a decade
vides real-time professional career support and advance- collection of publishers’ catalogs and promotional mailings,
ment opportunities and sponsors an annual technical Annual from committee members’ personal book collections, and
Workshop & Symposium. Topics covered by the site from visits to the catalogs of libraries that were closely
include implementing national and international standards, related to measurement interests.
achieving laboratory accreditation, new measurement tech- It was found that individual contributor’s book lists rarely
nology, advances in measurement disciplines, laboratory overlapped each other, yet each person was convinced
management procedures and skills, equipment management, theirs was in touch with the bulk of the related book
workforce training, and new instrumentation. In partic- literature. The final list includes over 800 titles up to that
ular http://www.ncsli.org/resources/index.cfm is an inter- date. Books related to, but not directly concerned with,
nal location for some useful resources. These include an measurement and instrumentation (such as control theory,
Acronym and Abbreviation Glossary, Procedures Database, systems design) were not included. It is, unfortunately not
Training Information Directory and NCSLI Archives. yet available in electronic form.
Information Databases of Relevance to Measurement 249
Sections of the listing (by author only), cover Traditional personal methods involving card indexes have
given way to the use of personal computers, especially to
1. Fundamental concepts (theory of measurement for the application of relational databases. Setting up citation
instance); references for general use is not easy for some 100 attributes
2. Units, standards, calibration; are needed to define the paper and book reference in all of
3. Measurement uncertainty, results processing, errors; the ways required. Publishers ask for so many variations
4. Instrumentation practice (in 10 groups); that it is not easy to accommodate them in a simple
5. Instrument system design, construction, and evaluation; database. ENDNOTES is a database set up for processing
6. Education and training; bibliographies. There is no doubt that making satisfactory
7. Miscellaneous (mainly historical). records of citation lists needs diligence and care to detail
An entry is included only once in the listing. as the sources are found.
The problems of deciding what are significant papers in The tediousness of rekeying reference items into the
the journal literature was addressed and found to be too required form could well do with increased use of standard
problematic to produce. The truly vast quantity of papers formats for bibliographic entries.
available was realized when the first 18 instrument-related Users with personal lists are encouraged to mount them
terms (of a possible 650 from the INSPEC thesaurus!) were in an enduring Web site and maybe then advise Measure-
combined with the term ‘review’ and tested on a short file Mentor of that location for inclusion in its link.
length of INSPEC. It produced over 800 abstracts, most
of which appeared to be valid entries for a bibliography.
Clearly, the user must go direct to the databases to generate RELATED ARTICLES
citations, sorting them online to yield those of individ-
ual relevance. Article 1, Sophistication of Measurement and its Body
of Knowledge, Volume 1; Article 2, Organization of
Instrument Science, Volume 1; Article 5, Humans in
7 PERSONAL LISTS the Real World, Volume 1; Article 41, Terminology and
Classification of Measurement Systems, Volume 1.
Browsing, current awareness services, and other sources
such as reprint requests and conference attendance, pro-
REFERENCE
vide the individual with a steady influx of quite specific
information. When the number of entries reaches a few Sydenham, P.H. (ed.) (1980) A Working List of Books Published
hundred, it becomes necessary to use a computer database on Measurement Science and Technology in the Physical
to record them. For this to be useful, it in turn requires Sciences, International Measurement Confederation IMEKO,
ordered storage. Applied Physics Dept, Technische Hogeschool Delft, Delft.
43: Units
Brian W. Petley
National Physical Laboratory, Teddington, UK
apparent that it was impracticable to have a ‘measurement
1 Units 255 vocabulary’ of only seven base units and the derived units
2 The Definitions of the SI Base Units 258 were also agreed, the number of these being kept to a min-
3 Traceability 259 imum (Table 3).
The current International Vocabulary on Metrology
4 Other Unit Systems 259
(VIM) defines a unit as ‘a scalar quantity defined and
Further Reading 260 adopted by convention, with which other quantities of
the same kind are compared in order to express their
magnitudes relative to that quantity’.
1 UNITS The following notes are added:
All measurement is made in terms of comparison with a like 1. Units are designated by conventionally assigned names
quantity. If we live in isolation, the latter quantity may be and symbols.
anything that we choose. However, we might lose the latter 2. Units of quantities of the same dimension may be
and if we wish to communicate the result with ourselves or designated by the same name and symbol even when
with other people possibly in another place at a later date, the quantities are not of the same kind. For example,
then it is obvious that it is useful, even essential, that we the joule per kelvin, J/K, is the name and symbol of
all use the same reference quantity or can relate our more both a unit of heat capacity and a unit of entropy.
transient reference quantity to it. Thus, one might well use 3. Units of dimensionless quantities are numbers. In some
one’s foot or a convenient stick when spacing plants in the cases, these numbers are given special names, for
garden, but would convert the spacing to meter if one were example, radian and steradian.
writing a book on planting. 4. If there is a risk of confusion, the term ‘unit of
The general arrangements by which this agreement is measurement’ should be used.
obtained are illustrated in Figure 1. The SI or Systéme Units that are named after people take the lowercase
International des Unités (Table 1) has been set up to achieve for the name of the SI unit and the uppercase for the
this universality and all dimensioned measurements may be symbol; thus the ampere, symbol A, is the SI base unit
related to the units of this system. of electric current.
The English language has some 80 000 words, but with
measurement we try to have the minimum possible ‘vocab-
ulary’. Initially, it was intended that all SI measurements
1.1 SI conventions
would be expressed in terms of six basic units but it
was decided that a seventh should be added in order to
The value of a quantity Q is given by the product of the
take account of the needs of chemistry and related fields
numerical value of Q, {Q} and the unit or units of Q,
(Table 2).
[Q], thus:
Many of our measurements are expressed in terms of jaw-
breaking combinations of these basic units. It soon became Q = {Q} × [Q] (1)
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
256 Units, Standards and Calibration
The international organization of metrology
Metre Convention
Diplomatic Treaty
1875
General Conference on
Weights and Measures Governments of
(CGPM) delegates from member Member States
states, meets every four years
International Committee on
Weights and Measures International
(CIPM). Eighteen individuals Organizations
selected by the CIPM, meets annually
Eleven Consultative Committees CIPM: Mutual Recognition
(CC’s) each chaired by a Arrangement
member of the CIPM
Bureau International des Poids et Measures National Metrology
(BIPM). International center for metrology, Institutes
located at Sévres, France, about seventy staff
General National Arrangement for Metrology
National Measurement National users of Metrology
Laboratory
Manufacturing
Realization of SI National calibration
and Related Units network Commerce and Trade
Science
Calibration
Laboratories
Provision of Laboratory Communication
National Standards Accreditation Transportation
Energy Generation
Dissemination Regulations and and Distribution
specifications
Defence
Government Applications
Cenre of Expertise Regulations
in National Navigation and
Measurement Surveying Services
Legal Metrology
Environment
Regulatory and
Voluntary Enforcement of
Standards, etc. Regulations
Figure 1. The international organization of metrology.
The convention is that italic font is used for the symbols derived values was used, the numerical value remained the
for quantities, and upright Roman font for units. same. This excluded popular units such as the liter, degree,
minute, hour, day, and so on. However, certain units such
1.2 Coherence as these were in such widespread use that it was agreed
The agreed derived units had to satisfy the criterion that that certain noncoherent units could be used alongside the
whatever correctly dimensioned combination of base and SI units.
Units 257
Table 1. The composition of the International Sys- to that given in Table 4. Any of these may be attached to
tem of Units (SI) and prefixes: 2002. any SI unit but they must not be cascaded. For example,
Type of unit Number 10−6 kg = 1 mg (1 milligram), but not 1 µkg.
SI base units 7
SI named derived units 21
Named SI prefixes 20 1.4 Non-SI units (i) accepted for use, (ii) values
Experimentally determineda 3
Units outside the SI (day, minute, 10
determined experimentally, and (iii) used
etc.) (off system units)a currently (temporarily) with the International
a
System
Units approved for use alongside the SI.
The International Committee on Weights and Measures
Table 2. The SI base units.
(CIPM) (1969) recognized that users would wish to employ
Base quantity SI base unit name Symbol the SI together with units that are not part of it but are
length metre (meter in USA) m important and widely used, listed three categories of non-SI
mass kilogram kg units: units to be maintained; to be tolerated temporarily;
time second s and to be avoided. In 1996 the CIPM agreed on a new
electric current ampere A classification of non-SI units: units accepted for use with
thermodynamic temperature kelvin K
amount of substance mole mol the SI, Table 5; and units accepted for use with the SI whose
luminous substance candela cd values are obtained experimentally, Table 6.
Table 5 lists non-SI units that are accepted for use with
the SI. It includes units that are in continuous everyday
1.3 The SI prefixes use, in particular, the traditional units of time and of
angle, together with a few other units, which have assumed
The 11th General Conference on Weights and Measures increasing technical importance.
(CGPM) authorized certain prefixes to be used in associa- It has proved useful to have some experimentally derived
tion with SI units. The initial list has since been extended units in fairly common use as well, and these are shown
Table 3. The SI derived units with special names.
Quantity Name Symbol Expression in terms of:
Other units SI base units
angle radian r m × m−1
solid angle steradian sr m2 × m−2
frequency hertz Hz s−1
force newton N m × kg × s−2
pressure, stress pascal P N m−2 m−1 × kg × s−2
energy, work, quantity of heat joule J m2 × kg × s−2
power, radiant flux watt W m2 × kg × s−3
electric charge, quantity of electricity coulomb C s×A
electric potential, potential difference, electromotive force volt V W/A m2 × kg × s−3 × A−1
capacitance farad F C/V m−2 × kg−1 × s4 × A2
electric resistance ohm V/A m2 × kg × s−3 × A−2
electric conductance siemens S A×V m−2 × kg−1 × s3 × A2
magnetic flux weber Wb V×s m2 × kg × s−2 × A−1
magnetic flux density tesla T Wb/m2 kg × s−2 × A−1
inductance henry H Wb/A m2 × kg × s−2 × A−2
◦
Celsius temperature degree Celsius C K
luminous flux lumen lm cd × sr cd × m2 × m−2 = cd
illuminance lux lx m−2 × cd × r m−2 × cd
activity (referred to a radionuclide) becquerel Bq s−1
absorbed dose, specific energy (imparted), karma dose gray Gy m2 s−1
equivalent, ambient dose equivalent, directional dose
equivalent, personal dose equivalent, organ equivalent dose
catalytic activity katal kat s−1 × mol
258 Units, Standards and Calibration
Table 4. The range of SI prefixes. 2 THE DEFINITIONS OF THE SI BASE
Factor Name Symbol Factor Name Symbol UNITS
1024 yotta Y 10−1 deci d
1021 zetta Z 10−2 cento c The definitions of the SI units are carefully framed to
1018 exa E 10−3 milli m reflect the increasingly stringent requirements of modern
1015 peta P 10−6 micro µ science and technology. Although the names and symbols
1012 tera T 10−9 nano n for the SI units are intended to be constant, the definitions
109 giga G 10−12 pico p
are modified in the light of progress in order to increase
106 mega M 10−15 femto f
103 kilo k 10−18 atto a the stability and invariance of the SI. Each time that the
102 hecto h 10−21 zepto z definition is changed care is taken to see that the new unit
101 deca da 10−24 yocto y lies within the reproducibility of the old. The definitions of
the SI units were originally derived from concepts related
to properties of the earth and these were soon transferred to
Table 5. Non-SI units accepted for use with the International artifact units such as the prototype meter and kilogram. One
System.
can readily imagine that the reproducibility of the meter was
Name Symbol Value in SI units ultimately limited by the width of the engraved markings
minute min 1 min = 60 s as well as by the stability of the platinum–iridium bar.
hour h 1 h = 60 min = 86 400 s
day d 1 d = 24 h = 86 400 s
◦
degree 1◦ = (π/60) rad 2.1 Unit of length (meter, symbol m)
minute
1 = (1/60)◦ π
second 1 = (1/60) = (π/6400) rad
The meter is the length of the path traveled by light in a
liter l, L 1 L = 1 dm3 = 10−3 m3
metric ton t 1 t = 103 kg time interval of 1/299 792 458 of a second.
neper Np 1 Np = 1
bel B 1 B = (1/2) ln 10 Np
Note: Other units which are currently approved for use alongside the SI in
2.2 Unit of mass (kilogram, symbol kg)
order to satisfy the needs of commercial, legal and specialized scientific
use include: nautical mile (knot), are (a), hectare (ha), bar (bar), angstrom The kilogram is the unit of mass. It is equal to the mass of
(Å), and barn (b).
the international prototype of the kilogram.
Table 6. Non-SI units accepted for use with the International
System, whose values are obtained experimentally. 2.3 Unit of time (second, symbol s)
Name Symbol Value in SI units
The second is the duration of 9 192 631 770 periods of the
electronvolt eV 1 eV = 1.60217653(14) × 10−19 J
radiation corresponding to the transition between the two
unified atomic u 1 u = 1.66053886(28) × 10−27 kg
mass unit hyperfine levels of the ground state of the cesium 133 atom.
astronomical ua 1 ua = 1.49597870691(30) × 1011 m
unit
Note: The latest CODATA values may be obtained from the National
2.4 Unit of electric current (ampere, symbol A)
Institute for Science and Technology (NIST) and International Bureau
of Weights and Measures (BIPM) Websites: http://physlab.nist.gov/cuu/ The ampere is that constant current which, if maintained
Constants/Citations/Search.html, http://www.ibipm..fr/enus/, or http://phy
sics.nist.gov/cuu/Constants/Table/allascii.txt in two straight parallel conductors of infinite length, of
negligible cross section, and placed 1 m apart in vacuum,
would produce between these conductors a force equal to
in Table 6. The electron volt is probably the most fre- 2 × 10−7 newton per metre of length.
quently encountered of these via its extensive use in mea-
suring X-ray energies. The unified mass unit is particu-
larly important in chemistry and serves as the reference 2.5 Unit of thermodynamic temperature (kelvin,
quantity for atomic weights and atomic masses. The astro- symbol K)
nomical unit serves as the basic length reference unit in
astronomy, particularly within the solar system, and larger The Kelvin, unit of thermodynamic temperature, is the
scale units such as the parsec are subsequently related fraction 1/273.16 of the thermodynamic temperature of the
to it. triple point of water.
Units 259
Because of the way that thermodynamic scales used Of course, even the national and international representa-
to be defined, it remains common practice to express a tions of the units will be slightly in error.
thermodynamic temperature T , in terms of the difference It follows from this that all measurements must be
from the reference temperature To = 273.15, the ice point. made with devices that have been calibrated in some
This temperature is called the Celsius temperature, symbol way against nationally or internationally maintained SI
t, and is defined by the equation standard representations of the SI units, or other agreed and
known units to the appropriate degree of accuracy. This
t = T − To (2) has given rise to the concept of traceability. The current
official definition of traceability is given in the International
The numerical value of a Celsius temperature, expressed in Vocabulary on Metrology as follows:
degrees Celsius, is given by
The property of a measurement or the value of a standard
t T whereby it can be related to stated references, usually
◦ = − 273.15 (3) national or international standards, through an unbroken
C K
chain of comparisons having a stated uncertainty.
2.6 Unit of amount of substance (mole,
symbol mol) 4 OTHER UNIT SYSTEMS
The definition of the mole comprises two parts: There are other unit systems in common use. These have
1. The mole is the amount of substance of a system, which a fixed relationship with the SI, such as the cgs and fps
contains as many elementary entities as there are atoms systems, although the size and spelling between the British
in 0.012 kg of carbon 12; its symbol is the ‘mol’. and the US versions of the derived units in the latter
2. When the mole is used, the elementary entities must be (for example, the US gallon and the Imperial gallon) may
specified and they may be atoms, ions, electrons, other well differ.
particles, or specified groups of such particles. There are other systems that are useful for scientific
purposes whose relationship with the SI must be determined
In this definition, it is understood that unbound atoms of
experimentally. Notable examples of these are the System
carbon, at rest and in their ground state are referred to.
of Astronomical units and the atomic system of units, the
system of natural units, the Planck units, and so on. These
2.7 Unit of luminous intensity (candela, systems are based on the fundamental physical constants.
symbol cd) The accuracy of our knowledge of these has been rapidly
The candela is the luminous intensity, in a given direction,
of a source that has emitted monochromatic radiation of 10−2
ohm
frequency 540 1012 Hz and has a radiant intensity in that
direction of 1/683 watt per steradian.
h/e2
Fractional accuracy
10−4
3 TRACEABILITY
10−6
Traceability to the SI units involved is the desired charac-
teristic of any measurement. Thus, if one measures a length
with a cheap plastic ruler, one often finds that the lengths
so measured will differ from those made with a steel ruler 10−8
because the markings on the plastic ruler are not quite at
the correct separations. One should always assume, as a 1850 1900 1950 2000
matter of routine, that all other measuring equipment, ana- Time/year AD
log or digital, electronic or mechanical, and so on, may
Figure 2. Illustrating how the accuracy of the fine structure
similarly be in error at some level of accuracy (including constant and/or h/e2 caught up with that of the realization of
nonlinearities). Thus, one’s local measurement units will the ohm: the ohm may now be maintained in terms of an
almost certainly differ slightly from the correct SI values. internationally agreed value for the Klitzing constant.
260 Units, Standards and Calibration
catching up with the accuracy with which the SI unit(s) Ferro, A., Giacomo, M.P. and Leschiutta, S. (eds) (1976) Metrol-
may be realized. See, for example, Figure 2. ogy and Fundamental Constants: Proceedings of the Interna-
tional School of Physics “Enrico Fermi, Course LXVIII, Aca-
demic Press, Amsterdam.
FURTHER READING Mills, I.M. (1997) The Language of Science. Metrologia, 34(1),
101–109.
The definitive sources for the current SI, including the definitions Mohr, P.J. and Taylor, B.N. (2000) Reviews of Modern Physics
of the SI units, are contained in The International System of 72, 351. A complete listing of the current CODATA recom-
Units (SI ) (7th edn, BIPM, 1998): http://www.bipm.fr/enus/3 SI mended values for the fundamental constants is available at
also published as NIST Special Publication 330 (NIST, 2000): http://physlab.nist.
http://physics.nist.gov/cuu/Units/index.html and The Interna- gov/cuu/Constants/Citations/Search.html.
tional System of Units, Supplement 2000 , BIPM, Sèvres (2000):
http://www.bipm.sfr/pdf/si-supplement2000.pdf. Petley, B.W. (ed.) (1986) The Fundamental Constants and the
Frontier of Measurement, Adam Hilger, London.
Crovini, L. and Quinn, T.J. (eds) (1989) Metrology at the Fron-
tiers of Physics and Technology: Proceedings of the Inter- Quinn, T.J., Leschiutta, S. and Tavella, P. (eds) (2000) Recent
national School of Physics “Enrico Fermi,” Course CX, 27 advances in metrology and fundamental constants, Proceedings
June–7 July, Elsevier Science and Technology Books, Ams- of the International School of Physics “Enrico Fermi” Course
terdam. CXLVI, 25 July–4 August, IOS Press, Amsterdam.
Cutler, P.H. and Lucas, A.A. (eds) (1983) Quantum Metrology Roche, J.J. (1998) The Mathematics of Measurement, Athlone,
and Fundamental Physical Constants, NATO ASI Series. B: London.
Physics, Plenum, New York.
44: Types of Paper Standards and their Purpose
Halit Eren
Curtin University of Technology, Perth, Western Australia, Australia
and/or processes and procedures. Consequently, there are
1 Introduction 261 several kinds of standards. In addition to the fundamen-
2 Published Standards 262 tal physical measurement standards, there are standards of
practice or protocols that can be defined as documents
3 Standards Organizations Relevant to
Measurement 263 describing the operations and processes to achieve uni-
fied results.
4 Links Between Standards Organizations 263
This article largely deals with the practice and pro-
5 Development of Published Standards 264 tocol standards, which can loosely be termed as pub-
6 Obtaining Standards 265 lished standards.
7 Use of Standards in Support of Design and Both physical and published measurement and metrology
Operation 266 standards play a fundamental role in shaping the efficiency
8 Example of Development of a Published of domestic and global economies. Published standards
Standard 267 are important as the documents that provide textual and
9 Summary of Benefits of Published Standards 268 illustrative information on what and how measurements
Further Reading 268 should be done, and how they are done.
They are necessary for worldwide progress in trade
liberalization, research and development activities, inter-
penetration of different technological sectors, and also
1 INTRODUCTION for ensuring consistency in emerging technologies. Con-
sequently, because of the national and international impli-
The term standard has many definitions. The Oxford cations, there are many institutions that are responsible
English dictionary defines standard as ‘noun 1 a level of to investigate, develop, determine, and maintain the rele-
quality or attainment. 2 a required or agreed level of qual- vant standards to support worldwide scientific and indus-
ity or attainment. 3 something used as a measure, norm, trial activities.
or model in comparative evaluations. adjective 1 used or However, national and international authorities are sub-
accepted as normal or average. 2 (of a size, measure, etc.) ject to various internal and external forces, differences of
regularly used or produced’. expertise opinions, and commercial interests during the
The measurement standards, on the other hand, are those development of the standards that may not always end up
devices, artifacts, procedures, instrument, systems, proto- with the ideal situations. This, unfortunately, may result
cols, or processes that are used to define measurement units in a confusing number of standards with different versions
and on which all echelon measurements depend on the fun- and interpretations for a specific artifact or measurement. In
damental physical standards. addition, publications grow over time as new technological
From the above definitions, the word ‘standard’ implies developments take place, and the standards authorities and
comparative evaluations, which can be physical quantities procedures change.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
262 Units, Standards and Calibration
In this article, an introductory explanation on the nature, the affecting concepts, choosing the languages, working
scope, and generation of the published standards is given out the schedules, and so on.
with some examples. • Working procedure that includes collecting and record-
The process of developing the published standards is ing terminological data, establishing the term list, deter-
explained, and guidance is given on how to go about mining the concept fields and concept systems, for-
finding the relevant standards for a specific application. mulating definitions, establishing and structuring the
After having read this article, the reader should be able entries, order of entries, indexes, graphic representation,
to appreciate the importance of the standards in the design bibliography, and so on.
and application of measuring systems and should have some Given that standards on a specific topic exist, determining
informed idea on the selection of most applicable standards the relevant one for a specific application can be tackled by
for his/her needs. adopting a top-down approach. That is, the standards can
be classified under general headings, such as the following
2 PUBLISHED STANDARDS • generalities, terminology, standardization, documenta-
tion
Published standards are documents describing the opera- • mathematics, natural sciences
tions and processes that must be performed to achieve • environmental, health protection, safety
a particular task. The National Standards Policy Advi- • metrology and measurements
sory Committee (http://www.nist.gov) describes a standard • mechanical systems and components
as ‘a prescribed set of rules, conditions, or requirements • health care and related technology
concerning definitions of terms: classification of compo- • electrical, electronics, telecommunications engineering
nents; specification of materials, performance, or opera- • road vehicles, railways, aircraft and space vehicle
tions; delineation of procedures; or measurement of quantity engineering
and quality in describing materials, products, systems, ser- • civil engineering
vices, or practices’. • domestic and commercial equipment, entertainment,
It is important to realize that measurement standards sports
are developed by various national and international bod- • mining, metallurgy, petroleum and related technologies
ies, sometimes operating totally independently; hence there
and others.
may be different versions for the same subject. Exam-
Each general standard contains many discipline-based
ples of national and international standards bodies are
standards. For example, metrology and measurement stan-
the International Organization of Standardization (ISO),
dards may have the following
the International Electrotechnical Commission (IEC), the
American National Standards Institute (ANSI), and so on. • metrology and measurement including measuring instru-
These organizations have multiple internal departments, ments, preferred numbers, standard measures, general
committees, subcommittees, and working groups to support aspects of reference materials, and so on
their activities. • measurement of volume, mass, density, viscosity
Common to all organizations, in the initial stages • measurement of time, velocity, acceleration, angular
of developing published standards, general terminology, velocity
vocabulary, and definitions concerning standardization • acoustics and acoustic measurements
and related activities are determined. This acts as an • optics and optical measurements
agent to develop a mutual understanding amongst the • thermodynamics and temperature measurements
members, groups, and concerned parties at individual, • electricity, magnetism, electrical, and magnetic mea-
governmental and nongovernmental levels. The aim is that surements
these standards are prepared in such a manner that they can • radiation measurements
be understood, used, and referred to easily. To stimulate
this, standards organizations have strict guidelines to assist and others.
technical groups and committees in drafting and producing Discipline-based standards contain specific standards. For
the standards. example, electrical, and magnetic measurements may com-
The typical preparation of a standard includes two prise the following
main stages: • electrical engineering in general including voltages,
• Preliminary work that includes overall analysis, iden- general electrical terminology, electrical documentation,
tification of target groups, determination of subject electrical tables, safety, fire-hazard testing
delimitation, finding the relevant sources, determining • electromagnetic compatibility
Types of Paper Standards and their Purpose 263
• insulating materials and fluids Published standards available to the consumer
• galvanic cells and batteries
• power transmission and distribution networks
• resistors, inductors, capacitors, semiconductor devices Consortium,
International National Industry
• integrated circuits, microelectronics, logic, and analog standards standards
and defacto
standards
int. stds.
microstructures
• radio-communications, mobile services, integrated ser-
ISO BSI
vices digital network (ISDN) DIN
IEC
• fiber optic communications, telecontrol, telemetering ANSI
• television and radio broadcasting others
and others. Figure 1. World standards organizations.
These specific standards can further be divided into
individual standards. Consider electromagnetic compatibil-
American National Standards Institute (ANSI)
ity (EMC) as an example, which may have the follow-
ing standards:
• electromagnetic compatibility in general Instrument Institute of American American
Society of Electrical & Society of Society for
• emission America Electronic Mechanical Testing & Others
• immunity, and so on. (ISA) Engineering Engineering Materials
(IEEE) (ASME) (ASTM)
Application areas of specific standards may differ greatly.
Continuing with the EMC, there are standards for the Figure 2. Branches of American National Standards Institute.
following
Each organization has several specialist groups that
• wheelchairs and motorized scooters
encompass a broad range of scientific disciplines, such as
• robots
• earth moving machinery electromagnetic, temperature, chemical, mechanical, radia-
• space systems tion, optic standards, and so on.
• agricultural and forestry machinery
• lifts, escalators, conveyors, road vehicles, and so on.
4 LINKS BETWEEN STANDARDS
More information on EMC measurement will be provided
in the examples; see Section 8. ORGANIZATIONS
Standards are important for scientific, research and
3 STANDARDS ORGANIZATIONS development activities and for international trade;
RELEVANT TO MEASUREMENT consequently, there are continuous links and cooperation
between national, regional, and international organizations
Standards are formulated by standards development orga- as listed in Table 1.
nizations (SDOs). These organizations can be government For example, in the case of ISO, there are many affiliated
agencies, professional societies, international or regional standards organizations.
organizations, private companies, or trade associations. Affiliated organizations largely represent countries as
World standard organizations can be categorized in four
major groups as national, international, consortium, and 1. full members that are entitled to participate and exercise
industry organizations, as illustrated in Figure 1. full voting rights on any technical committee and policy
Each organization has various branches that are responsi- committee of ISO;
ble for a particular area. As an example, Figure 2 illustrates 2. correspondent members that do not take an active part
some important branches of the American National Stan- in the technical and policy development work, but are
dards Institute. entitled to be kept fully informed about their work of
A list of some of the national and international organiza- interest; and
tions concerned with published measurement standards are 3. subscriber members that represent the countries with
given in Table 1. small economies.
264 Units, Standards and Calibration
Table 1. List of some of the national and international standard organizations.
Acronym Organization name and web address
ANSI American National Standards Institute (http://www.ansi.org/)
AFNOR Association française de normalization (http://www.afnor.fr/)
BSI British Standards Institution (http://www.bsi-global.com/)
CEI Comitato Elettrotecnico Italiano (http://www.ceiuni.it/)
CEN Comité Européen de Normalisation (http://www.cenorm.be/)
CENELEC Comité Européen de Normalisation Electrotechnique (http://www.cenelec.be/)
CSSN China Standards Information Center (http://www.cssn.net.cn/)
DIN Deutsches Institut für Normung (http://www2.din.de/)
IBN The Belgian Institution for Standardization (http://www.ibn.be/)
IEC International Electrotechnical Commission (http://www.iec.ch/)
ISO International Organization for Standardization (http://www.iso.org/)
JISC Japan Industrial Standards Committee (http://www.jisc.go.jp/)
KATS Korean Agency for Technology and Standards (http://www.ats.go.kr/)
NIST American National Standards and Technology (http://www.standards.com.au/)
ON Austrian Standards Institute (http://www.on-norm.at/)
SAA Standards Australia (http://www.standards.com.au/)
SCC Standards Council of Canada (http://www.scc.ca/)
SII The Standards Institution of Israel (http://www.sii.org.il/)
SNZ Standards New Zealand (http://www.standards.co.nz/)
TSE Türk Standardlari Enstitüsü (http://www.tse.org.tr/)
The membership fees of the subscriber members are Once a published standard is developed by an organiza-
reduced but they are allowed to be represented in some tion, depending on its applicability that standard may be
committees. adapted by other organizations after some modifications.
Currently, the ISO has about 96 full member countries In addition, many standards organizations collaborate as
including Australia, Germany, France, UK, and USA; 36 consortiums to develop and implement particular standards,
correspondent members; and 14 subscriber members. More as in the case of field-bus systems – see Article 172, Field-
information on this can be found in the ISO website bus Systems, Volume 3 – for industrial digital information
(http://www.iso.org). transmission applications.
Apart from the international organizations that are made
up with many voting and nonvoting members, numerous
organizations maintain links between standard institutions 5 DEVELOPMENT OF PUBLISHED
and organizations. Examples of some of these organizations
are as follows: STANDARDS
• The Asia-Pacific Legal Metrology Forum (APLMF) is Standards development organizations (SDOs) undertake the
a grouping of legal metrology authorities in the Asia- task of producing relevant standards in their expertise areas.
Pacific Economic Cooperation (APEC) economies and There are many different national and international SDOs,
other economies on the Pacific Rim. The objective of for example, in the electrical engineering field, some of the
this forum is the development of legal metrology and organizations in the USA are as follows:
the promotion of free and open trade in the region
through the harmonization and removal of technical or • Department of Energy(DOE)
administrative barriers. • Electronic Industries Association (EIA)
• Bureau International des Poids et Mesures (BIPM) • International Electrochemical Commission (IEC)
that ensures worldwide uniformity of measurements • Institute of Electrical and Electronic Engineers Stan-
and their traceability to the International System of dards Association (IEEE)
Units (SI). • National Electrical Manufacturers Association (NEMA)
• World Trade Organization (WTO), an international • National Institute of Standards and Technology (NIST).
agency whose purpose is to help trade flow as smoothly
as possible in a system based on rules, to settle trade Similar arrangements apply for other disciplines.
disputes between governments, and to organize trade At the individual organization level, they have different
negotiation. ways of developing published standards depending on their
Types of Paper Standards and their Purpose 265
structures and sizes. Here, the ISO will be given as a typical The draft agreement is then circulated as a Draft Inter-
standards body; more information can be found on the ISO national Standard (DIS) to ISO’s members for comments
website (http://www.iso.org/) and on the websites of many and balloting.
other organizations. The ISO operates with the following Members are responsible for providing public-review
committees and groups: procedures for making draft standards known and avail-
able to interested parties as well as to the general public.
• Policy development committees that are established by Once the feedback is received from members, the draft
the General Assembly to draft policies. The committees standard is formulated in its final form to be voted by
are open to all full member and correspondent members, the members.
and they directly report to the council. Policy develop- If voting is in favor, the document, with eventual mod-
ment committees have several branches such as CASCO ifications, is circulated to the ISO members as a Final
(Committee on conformity assessment), COPOLCO Draft International Standard (FDIS). If that vote is posi-
(Committee on consumer policy), and so on. tive again, the document is then published as an Interna-
• Council that operates the ISO. It consists of 18 elected tional Standard.
representatives of full members and some permanent Similar processes take place in other organizations, for
officers. example, in the case of International Electrotechnical Com-
• Council standing committees that advise the Council on mission (IEC), there are seven stages to develop a full
policy and strategic matters and raise any related issues. standard document (http://www.iec.org/):
• Technical management board that examines proposals,
monitors the work of technical committees, and estab- 1. Proposal Stage – New work item proposal (NP)
lishes and dissolves technical advisory groups. 2. Preparatory Stage – Working draft(s) (WD)
• Technical advisory groups (TAGs) that advise the 3. Committee Stage – Committee drafts (CD)
Board on basic, sectoral and cross-sectoral coordina- 4. Inquiry Stage – Inquiry draft (DIS, CDV)
tion, coherent planning, and the needs for new work. 5. Approval Stage (FDIS – Final draft international stan-
TAG 4 is responsible for metrology matters. dard)
• Technical committees that are responsible for a partic- 6. Publication Stage (IEC)
ular area such as TC 12 for quantities, units, symbols, 7. If an FDIS fails, the Technical Report 2 (TR2) is
conversion factors; TC 108 for mechanical vibration and published and revised after 2 years.
shock; and so on.
• Ad hoc advisory groups that recommend standardiza- TR2 status means publication of work is still under
tions. technical development, or where for any other reason there
is a future but not an immediate possibility of agreement,
The standards to be internationalized can be brought to on an International Standard.
the attention of the ISO from a country or industry or
business through one or more of the member organizations.
Members propose the new work item to the ISO as a
whole. If accepted, the work item is assigned to an existing 6 OBTAINING STANDARDS
technical committee.
Technical committees are composed of experts appointed There are several hundred thousand standards and technical
from industrial, technical, and business sectors. These regulations in the world containing special requirements for
experts may be joined by others with relevant knowledge, a particular country or region. Finding information on the
such as representatives of government agencies, testing specific standards can be a very difficult task.
laboratories, consumer associations, environmentalists, and Nevertheless, once the relevant standard is identified for
so on. Generally, experts are chosen from the national a particular application, the standards can be obtained in
delegations nominated by the member institutes of the coun- paper or electronic forms from various sources. The insti-
tries concerned. tutions of the country where the standards are applied can
The ISO rules state that member institutions are expected supply all the relevant information. Alternatively, one can
to take account of the views of a broad range of interested purchase the standard directly from the issuing body or
parties and other stakeholders so that acceptance of the through private suppliers. Most of the universities, insti-
proposal becomes a consolidated and national consensus. tutions of higher learning, and public libraries keep the
National delegations of experts of a technical commit- current and the most-used standards. If the standards are
tee meet to discuss and debate a new or amended stan- not available in a particular library, it is possible to get
dard until a consensus is reached on a draft agreement. them through the interlibrary loan services.
266 Units, Standards and Calibration
Nowadays, the standards can be purchased online from the necessary instrumentation, procedures, and control pro-
standards institution or other providers using their catalo- cesses can be determined and used appropriately for the
ges. Some of the web sites that provide full-text standards design, construction, operation, maintenance, installations,
are given below, but the list is not exhaustive. safety, and other related areas.
Use of standards in the design and operations offer
• ETSI, European Telecommunications Standards Insti- many advantages:
tute, offers direct access to publicly available ETSI
telecommunications standards (http://www.etsi.org/). • improves the communication between designers and
• IAEA, International Atomic Energy Agency, offers operators
most recent IAEA publications online (http://
• provides practical application of expert knowledge
www.iaea.org/).
• represents accumulated experience, thus avoiding
• !IEEE Xplore!, Institute of Electrical and Elec-
ground-up starts
tronics Engineers, offers standards for members
• enhances operability and interoperability, thus ensuring
(http://ieeexplore.ieee.org/).
compatibility
• NISO, National Information Standards Organization,
• sets a benchmark for global applications
offers free full-text access to NISO information stan-
• sets the recognized level of quality
dards (http://www.niso.org/).
• increases market access and user acceptance
• ANS, American National Standards Institute, offers
• improves communication between peers in industry and
a range of standards for purchasing (http://www.ansi.
other concerned parties
org/).
• provides immediate access to technical resources
• Global Engineering Documents offers national and
• provides an understanding on the future trends
international standards (http://global.ihs.com/)
• provides consistency in products
• HIS, Information Handling Services, offers most engi-
• provides traceability
neering standards (http://www.ihs.com/).
• facilitates interchangeability of products and quality
• WSSN, World Standards Services Networks, offers
assurance
national and international standards (http://www.wssn.
• ensures safety
net/).
• facilitates migration of products for integrated systems.
• NSSN, National Standards Services Networks, offers
national and international standards (http://www.nssn.
org/). Because of the advantages listed above, standards are
used for many purposes, such as the following:
As an example, the WSSN is a network of publicly
accessible Web servers of standards organizations around
1. To identify and plan the measurement activities. Once
the world. This server contains links to national, regional,
the data is collected, it is processed in a prescribed
and international standardization bodies. It also provides
manner so that it can be easily understood by other
access to other national and international organizations that
parties.
develop standards in their specialized subject areas.
2. To select appropriate instruments and experimental set-
If web browsers are selected to find standards, search
ups to yield credible and reliable results.
can be made by keyword entry, such as EMC, temperature,
3. To design new products and improve the existing ones.
noise, and so on; or by standard number, such as TP1, AB4,
4. To comply with the consumer/customer expectations.
and so on; or by complete title or portion of a title, such as
5. To meet the legal requirements and counteract possible
motors and generators, electrical equipment, and so on; or
challenges, and so on.
by date if the publication dates are known.
As mentioned earlier, one problem with using standards
is that there may be too many of them for a particular
7 USE OF STANDARDS IN SUPPORT OF
application; hence selection of the appropriate one may be
DESIGN AND OPERATION difficult. Usually, customers require the use of a particular
standard that suits his/her requirement within that country.
Standards in measurement and metrology largely define In the case of international applications, the recommended
guidelines for testing methodology, accuracy, and report- ones by the standards authorities need to be selected and
ing criteria for consistency and accuracy. From guidelines, complied with.
Types of Paper Standards and their Purpose 267
8 EXAMPLE OF DEVELOPMENT OF A military are some of the bodies setting standards on
PUBLISHED STANDARD the EMI.
In the USA, the FCC and military are two regulating
In this section, electromagnetic compatibility (EMC) will bodies governing the EMC standards.
be given as an example of the types of standards and the In Europe, each country has its own EMC govern-
selection of the appropriate ones. ing body as well as its own standards. Recently, Euro-
Electromagnetic radiation is a form of energy that can pean countries adopted the IEC 1000 as the common
propagate through a medium. It can be generated inten- EMI/EMC standard.
tionally or unintentionally that can interface with electronic The frequency spectrum over which EMI can be
systems, instruments, and other working equipment, thus generated spans from DC to γ -rays (1021 Hz) and
degrading their performance. beyond. The spectrum for use in EMC mainly covers
This electromagnetic interference (EMI) may be gener- from a few Hertz to 40 GHz. Lower frequencies are a
ated by almost all electrically operating devices. Owing main concern for the biological, while higher frequen-
to increase in man-made EMI generated around the globe, cies cause serious concern for communications, instru-
allowable limits as well as measurement techniques on the ments, most electronic devices, and military applica-
radio frequency (RF) noise have been set at national and tions.
international levels. Supposing that we want to do an EMC test and we are
The Federal Communication Commission (FCC), beginners on the subject, we will face typical information
International Electrotechnical Commission (IEC), and the briefly listed in Table 2.
Table 2. A brief list of standards concerning EMC.
EN 50065-1:2001 Signaling on low-voltage electrical installations
EN 55015:2001 Radio disturbance of electrical lighting and similar equipment
EN 55020:2002 Immunity of broadcast receivers and associated equipment
EN 50083-2:2001 Cabled networks for television, sound signals, interactive services
EN 5009022:1997 Home and building electronic systems
EN 60669-2-1:2000 Switches for household and similar fixed electrical installations
EN 501304:1996 Alarm systems
EN 50148:1996 Electronic taximeters
EN 50199:1998 EMC – Arc welding equipment
EN 50227:1999 Control circuit devices and switching elements
EN 50263:2000 EMC – Measuring relays and protection equipment
EN 50270:1999 Electrical apparatus for combustible gases, toxic gases, or oxygen
EN 55011:2000 Industrial, scientific, and medical RF equipment
EN 55014-1:2001 Household appliances, electric tools, and similar apparatus
EN 60204-31:1998 Safety of machinery
EN 604391:1999 Low-voltage switch gear and control gear assemblies
EN 60945:2002 Maritime navigation and radio communication equipment
EN 60947-5-1:1998 Low-voltage switch gear and control gear
EN 61000-6-4:2001 Generic – Emission standard for industrial environments
EN 611312:2000 Programmable controllers
AS/NZS 3652:1998 Arc welding equipment
AS/NZS 4251:1999 EMC – Residential, commercial, and light industry
AS/NZS 4251: 1999 EMC – Industrial environments
AS 62040.2 Uninterruptible power systems
CSA C108:1977 Electromagnetic interference measuring instrument
CSA-C108.6-M91 Industrial, scientific, and medical RF equipment
IEEE 299:1997 Measuring the effectiveness of the shielding enclosure
IEEE 475:2000 Measurement for field-disturbance sensors, 300 MHz to 40 GHz
IEEE C37.100:1992 Definitions for power switchgear
ISO 7637-1:2002 Road vehicles
ISO 13766:1999 Earth-moving machinery
MIL-HDBK-1857 Grounding, bonding, and shielding design practices
MIL-HDBK-241B Design guide for EMI reduction in power supplies
MIL-STD-469B Radar engineering interface requirements
MIL-STD-1541A EMC requirements for space systems
SAE ARP 958D EMI measurement antennas – Standard calibration method
268 Units, Standards and Calibration
This list in Table 2 is by no means exhaustive. In fact, FURTHER READING
we can find over 500 entries on the subject. However, we
can see in the list that some of the standards are duplicated, American National Standards Institute, ANSI, (2004) (http://
probably originating from the same source. We also notice www.ansi.org/).
that standards organizations specialize in some areas, which ASTM dictionary of engineering, science, & technology, 9th edn,
mostly concerns their area of expertise. (2000) TA402.A86 2000 Engineering Library – Dictionaries,
West Conshohocken, PA.
British Standards Institution, BSI, (2004) (http://www.bsi-global.
com/).
9 SUMMARY OF BENEFITS OF Dyer S.A. (ed.) (2001) Instruments, Survey of Instrumentation and
Measurement, Wiley, New York.
PUBLISHED STANDARDS Eren, H. (2004) Electronic Portable Instruments-Design and
Applications, CRC Press, LLC, Boca Raton, FL.
Both physical and published measurement and metrol- IEEE Xplore, (2004) (http://ieeexplore.ieee.org/).
ogy standards are important in shaping the efficiency of International Electrotechnical Commission, IEC, (2004) (http://
domestic and global economies. They are necessary for www.iec.org/).
worldwide progress in trade liberalization, research and International Organization for Standardization, ISO, (2004)
development activities, interpenetration of different tech- (http://www.iso.org/).
nological sectors, and also for ensuring consistency in National Information Standards Organization, NISO, (2004)
emerging technologies. (http://www.niso.org/).
Many national and international standards organiza- National Standards Services Networks, NSSN, (2004) (http://
www.nssn.org/).
tions are maintained by governments and private enter-
prises. Standards Council of Canada, SCC, (2004) (http://www.scc.ca/).
Many standards exist because of the diversity of organi- Standards Library for Measurement and Control: Guidelines for
zations and accumulation of the documented material over Quality, Safety, and Productivity, (2001) Research Triangle
Park, NC, ISBN: 1556175310.
time. However, a user can select the appropriate standards
Techstreet Thomson, standards catalogs, (2004) (http://www.
for a particular application easily by carefully identifying techstreet.com/).
the area of concern.
Webster, J.G. (ed.) (1999) The Measurements, Instrumentation
The use of standards in design and applications offer and Sensors Handbook, CRC and IEEE Press, New York,
many advantages such as providing expert knowledge, ISBN: 0-8493-8347-1.
setting level of quality and consistency, ensuring safety, World Standards Services Networks, WSSN, (2004) (http://www.
and so on. wssn.net/).
45: Calibration Process
Halit Eren
Curtin University of Technology, Perth, Western Australia, Australia
standards of the same quantity. In the second level, the
1 Definition of Calibration 271 secondary reference standards are calibrated by compar-
2 Benefits of Calibration 272 ing with primary standards of the same quantity using a
high precision comparator and making appropriate correc-
3 Calibration Personnel 273
tions. In the third level, working standards are routinely
4 Calibration Laboratory Practice 273 used to calibrate instruments and processes against the sec-
5 Physical Features to be Calibrated 274 ondary reference standards or their representatives. More
6 Labels Used in Calibration 274 is provided about physical standards in Article 43, Units,
7 Measurement Assurance Planning (MAPS) 275 Volume 1.
8 Costs of Calibration 275 The most important element in calibration is the relation-
9 Trends in Calibrations and Case Studies 276 ship between a single measurement and the reference base
for the unit of measurement: the reference base is the prime
10 Literature on Calibration 277
source of authority. The base units of measurement are the
Further Reading 277
Le Systeme International d’Unites (SI) units maintained in
the Bureau International des Poids et Mesures, Paris. These
are kilogram for mass, meter for length, second for time,
1 DEFINITION OF CALIBRATION candela for luminous intensity, kelvin for thermodynam-
ics, ampere for current, and mole for amount of substance.
Calibration of instruments and processes is essential for Other reference bases such as newton for force, hertz for
checking their performances against known standards. This frequency, and so on, are derived from the base units and
provides consistency in readings and reduces errors, thus maintained by national standards.
validating the measurements universally. The calibration Recently, with the wide applications of digital systems,
procedure involves comparison of the instrument against many intelligent instruments can make self-calibrations –
primary or secondary standards. In some cases, it may be Article 160, Smart Sensor System Features, Volume 3.
sufficient to calibrate a device against another one with In these cases, postmeasurement corrections are made and
a known accuracy. After the calibration of a device or a the magnitudes of various errors are stored in the memory to
process, future operation is considered to be error bound for be recalled and used in the laboratory and field applications.
a given period of time under similar operational conditions. A new trend is that certain calibrations can be conducted
The process of calibration is carried out in a hierarchical over the Internet by entering the appropriate Websites of
order. At the highest level, the primary reference standard manufacturers or calibration authorities – see Article 47,
is assigned a value by direct comparison with the refer- Internet Calibration, Volume 1.
ence base of SI units. The primary standards are designated The Oxford English Dictionary defines the term calibrate
and widely acknowledged as having the highest metrolog- as ‘1 mark (a gauge or instrument) with a standard scale of
ical quantities that have values without reference to other readings. 2 compare the readings of (an instrument) with
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
272 Units, Standards and Calibration
those of a standard. 3 adjust (experimental results) to take different laboratories, and so on. The ISO guidelines, such
external factors into account or to allow comparison with as ISO 11095, give guidance on how to assess, correct, and
other data.’ calculate uncertainties.
Calibration can be defined as ‘a set of operations to
establish the relationship between values of quantities
indicated by measuring instruments and systems under 2 BENEFITS OF CALIBRATION
specified conditions.’
Therefore, calibration is a process that assigns values to Calibration is a process of testing and comparing the errors
the response of an instrument relative to reference standards of measurement instruments and processes with accepted
or to a designated measurement process. Calibration aims standards in order to detect and correct variations in
to eliminate or reduce bias in the measurement system performance. Therefore, calibration assures that devices
relative to the reference base in accordance to a specific and processes meet expected performance specifications
algorithm. The bias may be introduced by the instrument within universally acceptable levels and accuracy. Hence,
itself and remain constant independent of the direction of calibration has the following benefits:
measurement. Alternatively, bias may be caused by linear
• It determines whether measurements made before the
drift over the time of measurement.
calibration were valid.
Some of the important definitions related to calibration
• It gives confidence that the future measurements will
are as follows
be accurate.
Calibration procedure: ‘A documented, verified, and vali- • It assures consistency and compatibility with those
dated process that describes a set of operations in accor- made elsewhere.
dance to a given method.’ • It leads to repeatability and reproducibility assessments
Calibration provider: ‘Laboratory or facility including of the instruments and processes.
personnel that performs calibration in an established • It provides confidence that products meet their specifi-
location.’ cations, thus reducing legal liability – see Article 25,
Errors: ‘Result of a measurement minus the true value of Introduction to Signals in Physical Systems, Vol-
the measurand.’ ume 1.
Reference standard : ‘Measurement standard having the • Without calibration, the product quality may be poor,
highest metrological quality available in an organization.’ thus opening up legal challenges and high failure rates
Uncertainty of measurement: ‘Dispersion of values that can of the products, thus increasing costs.
be attributed to the measurand.’ • It increases efficiency by ensuring that measurements
are correct.
In accordance with the above definitions, a calibration • In the process industry, calibration of devices assures
can serve two main purposes: that the processes are well controlled and that the
1. If the instrument reads in the same units as the products meet expected specifications.
reference standards, the purpose of calibration is to • It leads to documentation of performance of instruments
eliminate biases and minimize the instrumental errors. and processes to meet quality standards such as ISO
9000, ISO 1400, and QS-9000.
2. If the instrument reads in different units than the
• Frequent calibrations can provide a graphical view of
reference standards, the calibration serves the purpose
the equipment uncertainty over time, thus leading to
of converting the instrument readings to the units
reliability of performance. This gives in-service life
of interest.
analysis; hence, depreciation and replacements can be
In both cases, most calibrations are likely to have predicted in an informed manner.
uncertainties. • Measurements made within international standards pro-
Since most calibrations are likely to have uncertainties motes global acceptance, thus increasing competitive-
in the calibrated values, they need appropriate evaluations. ness.
Uncertainties can be evaluated as type A evaluations and • It helps convenient implementation of related regula-
type B evaluations. The type A evaluation applies to both tions and legislation that govern the use of equipment
random error and bias and uses statistical techniques. Type in a particular application.
B evaluations can apply to both random error and bias • As the technology changes, the regulations and legisla-
too, but the calculation of the uncertainty component is tion of test and measuring instruments change continu-
not based on a statistical analysis of data but on factors ally, and calibration helps compliance validity of mea-
such as experience, scientific judgment, scant data, use of surements and processes under changing conditions.
Calibration Process 273
• In some cases, calibration can be used as a gain; laboratories (called calibration laboratories) for calibrating
the value multiplied by some input can produce a different instruments and processes.
scaled output. Some organizations cover a broad spectrum of tech-
nologies within a wide variety of industries such as com-
munication, defense, medical, aerospace, automotive, and
3 CALIBRATION PERSONNEL chemical industries.
Laboratories are accredited by authorities in accordance
The calibration is conducted by the use of appropri- with guidelines such as the ISO Guide 58.
ate devices representing the reference standards that are Accreditation is a formal recognition that a particular
deployed by trained personnel. Appropriate devices are kept laboratory is competent to conduct specific tests and/or
as secondary standards and working standards by national calibrations.
authorities and companies themselves. The calibration pro- Calibration of an instrument is intended to eliminate or
cess involves set rules, regulations, and processes. reduce bias in its readings over a continuous range of
During the calibration, the calibrating personnel make the values. For this purpose, a reference standard with known
following assumptions: values is measured by the instrument and the functional
relationship between the values of the standards and the
1. Handling procedures are consistent. corresponding measurements is established.
2. Test item and reference standard respond in the same A successful calibration procedure requires the following
manner in the test environment. basic steps:
3. Random errors associated with the measurement are
independent. • selection of an appropriate reference standard with
4. Multiple measurements form a distribution curve with known values covering the range of interest;
the same standard deviation. • application of the instrument or the reference standard;
5. The test item and reference standards are stable during • conducting calibration curves (i.e. least-squares fit) to
the measurement. establish the relationship between the measured and
6. The bias of the test item and reference standard are known values of the reference standard;
the same, and hence it can be eliminated by taking the • correction of measurements using calibration curves;
difference between the two readings. • preparation of the appropriate documentation of the cal-
7. Once the calibrations are carried out, ideally the ibration procedure, results, analysis, and interpretation
difference is zero or can be explained by appropriate of results for the client.
charts and statistical relationships.
8. Calibration is repeatable, yielding the same results. Once a procedure is adapted, the calibration relies on the
instrument continuing to respond consistently and in the
Skilled calibration personnel are essential to the preci- same way during the calibrations. If the system drifts or
sion of the calibration. Calibration personnel may need to takes unpredictable deviations, the calibrated values may
have appropriate clothing (static free clothing, gloves, face not be easily corrected for bias and errors, thus degrading
masks, etc.) for safety as well as to avoid the possibility of the accuracy of the measurements.
contamination. Although fully trained and experienced, it
During the calibration process, the readings of the test
is possible for different operators to produce measurements
item are compared with the reference standards such that
differing in signs and magnitudes. To overcome this prob-
values can be assigned on the basis of the known values
lem, measurements by different operators can be plotted
of the reference standard. The reference standards can be:
and compared. Another solution may be to maintain sepa-
resistors, length standards, voltage standards, and so on.
rate calibration curves by the same operator. This may not
However, in many situations, it may be impossible to
be a problem in automated calibrations.
achieve a perfect calibration because of instrument and
measurement biases, and uncontrollable random errors.
Mathematically: Ideal value = Measured value + Bias +
4 CALIBRATION LABORATORY Error
PRACTICE Similarly, a reference value may be subject to bias and
error: Ideal reference value = Reference value + Bias +
Calibration is conducted by organizations that perform Error
tests and or calibrations in permanent, temporary, or This leads to a deficiency in the calibration: Deficiency
remote locations. These organizations may have several = Ideal measured value − Ideal reference value
274 Units, Standards and Calibration
Owing to randomness of errors, this deficiency may not In the case of static-gravimetric liquid flow, a calibration
be zero; therefore, the calibration measurements have to be facility may include a reservoir, a pumping system, a
repeated many times. Then, the statistical techniques can be pipeline, flowmeter under test located on the pipeline,
applied to work out the calibration curves, average readings, collection system, computers and interface, supporting
and standard deviations. software, and so on. The calibration of flow of fluid
The process of collecting data for creating the calibration through the meter can be determined by collecting the
curve is critical to the success of the calibration program. prescribed mass of steady fluid flowing over a measured
As a rule of thumb, a minimum of five reference stan- time interval.
dards is required for a linear calibration curve and ten Sensor calibration: Sensors may have to be calibrated after
reference standards are sufficient for more complicated cal- having their data integrated with a signal conditioning
ibration models. system. This process for calibrating the processing part
Nowadays, most calibrations are conducted using com- of the system requires injection of a known input signal
puters to capture and analyze data. Once the results are into the sensor. By observing the output, a correct output
obtained, software packages can assist in analyzing the scale can be configured for that particular application. If
information. Most packages use the method of least squares the sensor is used for time varying inputs, the dynamic
for estimating the coefficients. Some of the packages are calibration becomes necessary. In most cases, transient
capable of performing a weighted fit if the errors of the behavior of sensor top step response may be sufficient to
measurements are not constant over the calibration interval. assess the dynamic response of the sensor.
The software tools provide information such as the coeffi- Calibration of food products: As food contains many chem-
cients of a calibration curve, standard deviations, residual ical substances, the calibration of devices becomes com-
standard deviation of the fit, and goodness of the fit. plex. In the case of honey, for example, the following
Calibrations of instruments and processes can be made parameters need to be identified by calibrated instruments
under static or dynamic conditions. If the calibrated variable and processes: fructose, glucose, turanose, maltose, mois-
has multiple inputs and outputs, the input is varied in ture level, acidity, and so on.
increments, in increasing and decreasing directions, over Calibration of images: Calibration is one of the first steps
a specified range. in image processing. For example, astronomical images
The observed output then becomes a function of that are calibrated to eliminate the effect of cameras, light
single input. For better results, this procedure may be pollution, and distortions. Various methods are used to
repeated by varying the input, thus developing a family ensure calibrated images by eliminating thermal, readout,
of relationships between the inputs and outputs. and other effects. For thermal effects, the cameras are
In multivariable situations, the input/output relation- cooled below certain temperatures and dark frames are
ship usually demonstrates statistical characteristics. From used to compensate the noise generated by camera
these characteristics, appropriate calibration curves can electronics.
be obtained, and statistical techniques can be applied
for analysis.
6 LABELS USED IN CALIBRATION
5 PHYSICAL FEATURES TO BE Calibration labels are important as they supply information
CALIBRATED on the efficiency and applicability of instruments. They
indicate parameters of instrument, tolerances, and special
The physical features to be calibrated depend on the conditions of use. Calibration labels need to conform to the
characteristics of the measuring instrument or process requirements of ISO/IEC 17025 General Requirements for
under test. the Competence of Testing and Calibration Laboratories.
Calibration is applied in a diverse range of measurement ISO 17025 requires the following
instrument and processes. Some examples follow.
• All measurement equipments shall be securely and
Flow calibration: There are many devices and facilities durably labeled.
available for the measurement of liquid, air, or solid • The labels should clearly indicate the name of the
flow – see Article 190, Calibration and Standards in calibration laboratory, date of calibration, due date,
Flow Measurement, Volume 3. Once the method of usage equivalent, and the authorized officer.
measurement is determined by an appropriate flowmeter • Information on the label must be legible and durable
set up, static or dynamic calibrations can be carried out. under reasonable use and storage conditions.
Calibration Process 275
• When it is impractical to affix a label directly on 7 MEASUREMENT ASSURANCE
an item, the label may be affixed to the instrument PLANNING (MAPS)
container.
• Temperature-resistant seals may be used when neces- Measurement Assurance Planning (MAP) is a group of
sary. activities designed for critical evaluation of a group of mea-
• Functional labels should contain reference standards. surements. The term was generated by the National aero-
nautics and Space Administration (NASA) and National
Institute of Standards and Technology (NIST) for describing
As an example, labels on radioactivity-measuring instru-
the calibration of instruments for their Earth Observing Sys-
ments contain information on calibration parameters, appli-
tem (EOS) programs. The concept is applied to calibration
cable dose levels, type of radioactivity detected, source
of EOS satellite, ground-based and airborne instruments
used, and so on, as illustrated in Figure 1.
where it consists of a series of carefully designed activities
Meters with multiple probes have additional labels for
in space explorations and measurements.
each additional probe containing information such as the
However, the idea of MAPs has been adapted by many
following
other organizations as a guideline for good practices in
measurement and calibration activities under various names
• date last calibrated and call date
such as Volt MAP, Mass MAP, Accelerometer MAP, and
• probe or sensor type
so on.
• information on organization and personnel who con-
NASA applies MAPs to achieve accurate, precise, and
ducted the calibration consistent long-term series of radiometric measurement
• range of measurement and correction factors for each data form multiple instruments and multiple platforms to
range advance scientific understanding of the earth as a system.
• efficiency of the instrument, and so on. It is identified that the success of the project depends
on proper calibration of all instruments against a set of
Calibration labels must be durable and legible. They are recognized physical standards and careful characterization
made from various materials such as metal plates, soft of the instrument performance at the system level.
adhesive aluminum, vinyl overlays, or transparent adhesive Other important points would be adhering to good mea-
tapes. In the case of metal plates and adhesive aluminum, surement practices and established protocols, intercom-
information and markings may be permanently indented paring measurements wherever possible, and establish-
for durability. Calibration labels may have the following ing traceability for all instruments to the common scale
information: of physical quantities maintained at the national stan-
dards laboratories.
• calibrate before use
• not calibrated
• obsolete 8 COSTS OF CALIBRATION
• do not use
• does not conform A successful calibration process requires hardware and
• indication only software, special equipment, and manpower, hence the costs
• information only are variable depending intensity of use of these variables.
• out of calibration The cost of calibration depends on what is calibrated and
• reference only who is calibrating by it.
• uncalibrated instrument In simple cases where a one-off instrument is involved,
cost can be lower than one hundred dollars, but complex
• user-calibrated instrument, and so on.
cases can cost thousands of dollars.
Calibration cost depends on whether the calibration is
Cal. Date:__/__/__ Cal. By carried out on the premises of calibrating laboratories or on
Cal. Due:__/__/__ _____________ Probe the factory floor being outsourced to third parties.
type:_____ Certification by ISO 10012-1, ISO 9001, MIL-STD
Range C.F. Isotope Eff. Isotope
45662A, and MIL-HDBK-52B requires calibration for
Eff. measuring equipment. In many situations, such as weighting
X 0. C14 % % systems calibration, it is a statutory requirement.
One of the major factors for cost is the frequency of cal-
Figure 1. A typical calibration label. ibration of an instrument. Most calibration systems issue
276 Units, Standards and Calibration
a validity period during which the instrument can be used made via the adjustment of the parameters of the digital-to-
without concern for major errors and uncertainties. Some analog converters. Corrected calibration constants are kept
organizations use finely worked out methods for deter- within the nonvolatile memory for permanent use.
mining calibration intervals, while others use conservative As an example of this method, the Agilent electronic
calibration intervals barely able to meet the legal demands. calibration modules 8719, 8720, and 8722 microwave net-
The perception exists that calibration cost can be reduced work analyzers provide a broad frequency range of cali-
if the interval can be stretched legitimately. The use brations from 10 MHz to 67 GHz (http://we.home.agilent.
of uncalibrated instruments in an organization can be com/USeng/nav/). Similarly, Fluke offers electronic cal-
costly as it may affect the product quality and quality of ibration facilities for multifunction process, pressure, and
downstream operations. temperature instruments (http://www.fluke.com/calibrators/
Standards such as MIL-STD 45662A suggest good cal.asp).
calibration intervals. As a rule of thumb, 85 to 95% of Remote calibration: A new trend of calibration is emerg-
all instruments returned for calibration meet the calibration ing in the form of Internet calibration. Undoubtedly, the
limits. The calibration limits are determined by probability use of Internet is likely to expand the calibration capabili-
charts of age of instruments and their failure data. ties without the transportation of equipment from one place
Usually, an instrument must be calibrated if the failure to another. Internet calibration technique is supported by
rate increases or functionality deteriorates when compared web-accessible test procedures and appropriate hardware
to other standard instruments. and software.
It is possible that a planned calibration costs, say, $200 As examples of Internet calibration, Fluke , UK, offers
as opposed to an unexpected failure costing thousands of calibration systems for the Fluke 4950 multifunctio-
dollars. In this respect, a number of different mathemati- nal instruments (http://www.npl.co.uk/npl/news/press/pr16-
cal techniques, such as the Weibull statistics and renewal 2000.html).
equations, can be employed to analyze the costs. There Anritsu is active in investigating the possibilities of
are different software tools (e.g. visualSMITH, Calibration Internet-based calibration using portable OTDR MW9076
Manager, etc.) available for cost analysis and for determin- (http://www.electronicstalk.com/news/anr/anr163.html).
ing calibration intervals. The OTDR is controlled via the Internet with the aid
of appropriate software. The software can be controlled
through a PC via modem, mobile telephone access, or PCI
9 TRENDS IN CALIBRATIONS AND CASE cards.
STUDIES Similar efforts are made by NIST to realize Internet-
based calibrations particularly aimed for radiation process-
With the availability of advancing technology, the classical ing industry – see also the work of NPL in UK, as described
calibration process is changing on at least three fronts, these in Article 47, Internet Calibration, Volume 1.
being the following: In some cases, calibration algorithms and production
automation know-how are available online by combining
1. electronic calibration standard low-cost integrated circuits and Internet access.
2. e-calibration using Internet and communication tech- Online calibration is supported by appropriate software
niques that includes the mathematical models of the interface
3. intelligent and self-calibrating instruments and sensors. electronics and optimizes calibration on the basis of this
model.
These techniques are now explained with some examples. Self-calibrating instruments and sensors: Nowadays,
Electronic calibration: Many modern instruments offer many smart sensors are available in the marketplace, includ-
features for closed-case calibrations so that electronic ing digital programmable calibration features, as shown
calibration can be employed. Electronic calibration is a in Figure 2. Particularly, sensors complying with IEEE
single connection and one- or two-port calibration technique 1451.4 standards provide comprehensive transducer elec-
without disturbing the components inside the case. Once tronic datasheets (TEDs) that contain configuration, scaling,
the calibrating equipment, for example, computer, is linked and calibration information necessary to make measurement
with the device under calibration, appropriate software through mixed mode interface.
generates the necessary calibration information. Errors Many smart sensors – see Article 160, Smart Sensor
due to gains and offsets of the instrument are corrected System Features, Volume 3 – are capable of calibrating
mathematically within the instrument processor to obtain themselves, scaling the incoming data, computing statistics,
the correct measured values. Analog corrections can also be and communicating with other digital systems on the
Calibration Process 277
Standards Council of Canada (SCC), National Institute of
Sensor
Standards and Technology (NIST), National Association of
Serial Testing Authorities (NATA) of Australia and the British
interface
Calibration Services (BCS), and Australian Standard (AS).
Calibration Some important information on calibration can be found
DSP
in ISO, IEC, IEEE, and national standards publications.
User code
Some of these publications are briefly explained below.
ISO 17025 is an international standard for calibration
Figure 2. Main features of self-calibrating smart sensors. and testing laboratories. It requires labs to demonstrate
that they operate a quality system covering processes,
network. Most commercially available software permits documentation, and quality management. The laboratories
calibration of smart sensors and upload the new parameters need to generate technically valid results accounting for
directly to the sensor’s circuitry. the equipment procedures and personnel. Information on
In an application of smart sensors (the Atmos SSP14 ISO 1705 can be found in many books or ISO publications
Sensor Signal Processor family) (http://www.sensorsmag. (http://www.fasor.com/iso25/)
com/isensors/dec00/34/main.shtml), each smart sensor has BS EN ISO 9000:2000 is a family of standards for
a memory that is programmed at the factory with a set of international quality management. These standards are
default zero and span curves defining the relationships with globally accepted and applied by many organizations.
the physical phenomenon. These default curves represent (www.iso.ch/iso/en/iso9000-14000/)
average sensor output adjusted for the most accurate The military standards MIL-STD-45662A are standards
response at room temperature. Each time calibration is that describe the requirements for creating and maintaining
done, the appropriate zero or span curve is adjusted in the calibration systems for measurement and testing. They are
vicinity of the calibration. available at http://store.mil-standards.com/.
10 LITERATURE ON CALIBRATION FURTHER READING
Literature on calibration is available from various sources. BS EN ISO 9000:2000, (2000) http://www.iso.ch/iso/en/iso9000-
These are: 14000/iso9000/iso9000index.html.
Eren, H. (2004) Electronic Portable Instruments-Design and
1. Manufacturers: They supply comprehensive informa- Applications, CRC Press, LLC, Boca Raton, FL.
tion about calibration requirements of their products. Fluke (1994) calibration, Calibration: Philosophy in Practice,
2. Regulating authorities and standards institutions: Fluke Corporation, Everet, WA, http://www.fluke.com/
These supply information on calibration requirements calibrators/cal.asp.
of instruments and devices. Calibration can be ISO/IEC/EN 17025, http://www.fasor.com/iso25/.
statutory, particularly where health and safety is
i-sensor Sensor Interface and Calibration Solutions via the Inter-
important. net http://www.sensorsmag.com/isensors/dec00/34/main.shtml,
3. Calibration services: These provide information on 2004.
calibration processes. Microwave calibration kits, Agilent, http://we.home.agilent.com/
4. Organizations: They provide rules and regulations for USeng/nav/, 2004.
their equipments for assurance planning. Noizette, J.L., Robert, M. and Riviera, J.M. (1997) Intelligent
5. Books: These provide information on calibration issues Sensor Calibration. Complexity, Methodology, and Conse-
and processes. Details of some of these books can be quences, IEEE-IMTC Conference Proceedings, Vol. 2, New
found in the bibliography. York (pp. 948–952).
Sydenham, P.H., Hancock, N.H. and Thorn, R. (1989) Introduc-
Many nations and organizations maintain laboratories tion to Measurement Science and Engineering, Wiley, New
with the capability of primary functions of calibrating York.
instruments and field measuring systems that are used in Webster, J.G. (ed.) (1999) The Measurement, Instrumentation, and
everyday operations. Examples of these laboratories are: Sensors Handbook, CRC Press, Boca Raton, FL.
46: Calibration Interval
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
5. effect of external influence parameters (temperature,
1 Physical Features to be Calibrated 278 relative humidity, atmospheric pressure, magnetic
fields, vibration, radio frequency interference, ambi-
2 Calibration Labels 279
ent illumination level, and more);
3 Calibration Interval Assessment 279 6. power supply variations and noise on supply;
4 Types of Calibration Interval 280 7. response to pulsations on measurand;
5 Calibration of Automatic Test Equipment 8. frequency response;
(ATE) 280 9. step response and other inputs;
Related Articles 280 10. maximum safe voltage levels;
References 280 11. leakage resistance to ground;
12. level of operation of a protective device;
13. general inspection of instrument integrity;
14. battery voltages;
15. transmission and reflectance levels in optical systems.
1 PHYSICAL FEATURES TO BE
CALIBRATED
Before a new instrument is commissioned, it is first neces-
sary to establish a suitable calibration plan. The instrument
Calibration is the process of giving a measuring system should be calibrated before use and proper records created.
traceable accuracy. The general topic of calibration is The group that is responsible will be required to maintain
covered in Article 45, Calibration Process, Volume 1. a record system. Often a history card is used, on which is
This article deals with how to determine and set the interval recorded all required technical details about the instrument
between calibrations. including calibration information. A calibration procedure
There exist many physical features of a measuring instru- should be established in writing and be used each time the
ment that may need calibration. Often, the manufacturer’s item is calibrated.
original test certificate will indicate which parameters will Where automatic testing equipment (ATE) is used, the
need periodic calibration and give a recommendation for system will generally produce a calibration printout, which
what interval to use. The following list provides a checklist can act as the certified calibration document. Clearly, the
of some of the many possible error sources that may need data displayed is selected to suit the need.
quantifying: It is often convenient to hold a copy of the calibration
certificate with the instrument but in such cases the original
1. zero setting; should still be filed elsewhere in a secure location. In the
2. full-scale value; event of severe damage to the instrument, or its complete
3. error of nonlinearity; loss, the certificate will then be most useful to help phase-in
4. gain or transduction coefficient; a replacement.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Calibration Interval 279
2 CALIBRATION LABELS Obviously, the interval will depend upon the equipment’s
characteristics and the purpose of the instrument. In nuclear
Many different reasons exist for requiring one or more of power stations, the torque wrenches used to tighten critical
the many physical parameters listed above to be calibrated. nuts, such as those that hold on the reactor vessel top,
A guide of the types of calibration labels that are in use was are recalibrated after a dozen or so pulls: for each nut
published in the NCSL Newsletter of July 1978, this being and its associated pull, there is an individual record of
the summary of a study organized by one of the Regional the circumstance. At the other end of the scale is the
Groups of the NCSL membership. The following list is multimeter of the servicing electrician that may not need
extracted from that report: to be calibrated after the initial build unless the instrument
becomes damaged.
1. standards calibration. Statements on calibration interval are not easily found.
2. calibration: manufacturer’s specifications. Useful sources could be NCSL (1996), Dietrich (1997),
3. calibration: limited specifications. Morris (1998).
4. calibration: facility specification.
5. calibration: limited facility specification.
6. not usable for acceptance testing. 3 CALIBRATION INTERVAL
7. cross-check.
8. operational check. ASSESSMENT
9. calibration not required.
10. preventive maintenance. A priority problem faced by calibration facilities is estab-
11. calibrate before use. lishing suitable calibration intervals to be used. At present,
12. inactive or do not use. it does not appear that there will ever be developed a system
13. reject. that guarantees minimum cost of calibration and zero loss
14. reference or primary standard. through not having recalibrated an instrument in time. It is
15. transfer standard. necessary to build up information about an instrument as its
16. working standard. life passes, making intelligent informed estimates of need
that are gained from this experience and by study of what
The list was compiled from government, military, and others have done beforehand. The essential state to target
private users. It was thought during that meeting that, for is an adequate state of satisfaction that a responsible
in many cases, simplification of labeling systems might decision has been made and that undue risks are not taken.
be possible. To assist those without such experience, the follow-
Associated with every calibration is the cost of making a ing information is extracted from various issues of the
new calibration, and the cost resulting because of not hav- NCSL Newsletter:
ing done it when it needed to be done. The first cost is A suitable paper that describes how a very large organi-
relatively easy to decide. It depends upon the instrument zation handles the problem – the US Army – is provided
complexity, ease of adjustment, degree of automation, cal- in Westmoreland (1980).
ibration procedure, ability of staff, and less obvious factors The bibliography of the June 1977 Newsletter issue is
such as cost of loss of production while it is being cali- specifically concerned with calibration interval topics.
brated, loan or spares holdings needed to reduce such loss During early 1978, the NCSL Calibration Systems Man-
of production, transport costs, and costs of the associated agement Committee conducted a US survey on this subject.
paperwork. Clearly, the cost of making a calibration far Their findings are published in the July 1978 issue. It
exceeds the staff time of those involved. A balance must contains the following outline of how a typical member
be made between keeping costs low by employing less cal- laboratory characterizes its calibration interval system:
ibrations and the risk of waiting too long.
Costing the effect of using an instrument that might be 1. Initial intervals are established on the basis of the
(or might not be!) out of calibration is the intangible fac- recommendation of others, usually the manufacturer of
tor. Overcalibration is often considered to be – see Gebhart the instrument.
(1980) – the safe path, but in the interests of preserving 2. All instruments with the same function (e.g. DMMs,
maximum financial efficiency as well as, perhaps, maxi- scopes, etc.) are calibrated on the same interval.
mum instrument integrity, it is usually necessary to seek 3. Intervals are adjusted (lengthened or shortened) peri-
a suitable calibration interval along with guiding rules, odically (about twice a year) based on analysis of
which will decide when, in the future, the instrument should data for the manufacturer and model number group of
be calibrated. instruments.
280 Units, Standards and Calibration
4. Major test systems are either calibrated as a system, or probably not surprise many purchasers of instrumentation.
individual instruments within the system are removed A similar situation appears to still exist.
and calibrated separately.
5. A maximum interval (between one and two years) is
imposed, but there is no minimum interval. 5 CALIBRATION OF AUTOMATIC TEST
6. The interval system is intended to provide either at EQUIPMENT (ATE)
least 85% or at least 95% in tolerance of instruments
at recalibration. Where the number of instruments and degree of variation
and type arises and the number of calibrations is high,
An update of the NCSL work has been published, but is a case may exist where automation of many of the test
out of print, NCSL (1996). As ISO9000 is so widely used processes becomes economically viable. Remembering that
to assist quality operations, Morris (1998) may be of use. calibration and testing work is an overhead cost, it is usually
necessary to produce a cost-benefit analysis, which can
clearly show that there is an economic case to support
4 TYPES OF CALIBRATION INTERVAL introduction of automated test equipment (ATE). A case
study (Pearson 1980) provides one group’s experience of
Calibration intervals are often initially based upon one of getting started. The topic of calibration in testing is given
two procedures, these being ‘recall after a given calen- some content in O’Connor (2001).
dar time period’ or, the less favored method being based
on ‘instrument usage time’. These are discussed in Greb
(1973). Once established, the period is varied in accordance RELATED ARTICLES
with experience. The impact of the smart instrument gen-
eration containing self-calibration of a kind adds a new Article 45, Calibration Process, Volume 1; Article 47,
dimension to the task. Internet Calibration, Volume 1; Article 48, Common
The recall periods of a large number of commercial Sources of Errors in Measurement Systems, Volume 1;
instrument products were surveyed and reported in the May Article 49, General Characterization of Systematic and
1976 issue, the topic being followed up in Gebhart (1980). Stochastic Errors, Volume 1; Article 55, Uncertainty
It is not easy to provide any simple rule to adopt. Determination, Volume 1.
Recall for calibration periods vary from 1 to 36 months,
varying considerably from user to user for the same product.
Some instruments have a calibration period as little as REFERENCES
only 10 days.
Dietrich, C.F. (1997) Uncertainty, Calibration and Probability:
It might be considered that the better the quality of an The Statistics of Scientific and Industrial, Adam Hilger, Bristol.
instrument the longer can be the calibration interval. Quality
Gebhart, C. (1980) Recall Period Pilot Program. NCSL Newsletter,
in this case is defined to be a factor of recalibration inter- 20(1), 14–18.
val. An extensive study (Greb, 1976a,b) provides detailed
Greb, D.J. (1973) Calibration on Basis of Instrument Usage.
comment on whether such simplistic assumptions are cor- NCSL Newsletter, 13(3), 26–27.
rect. The results of the study are expressed in terms of the Greb, D.J. (1976a) Calibration Intervals and Instrument Quality –
intolerance rates (ITR) that occur. Introduction. NCSL Newsletter, 16(3), 26–32.
On the quality of precision instrumentation, in the more Greb, D.J. (1976b) Calibration Intervals and Instrument Quality –
general sense of the term, the reader is referred to Moss Parts II and III. NCSL Newsletter, 16(4), 29–35.
(1978), in which is summarized the experience of the Morris, A.S. (1998) Measurement and Calibration Requirements
Aerospace and Guidance Metrology Center, US, about the for Quality Assurance to ISO 9000, Wiley, New York.
acceptance of new equipment from industry. It is stated Moss, C. (1978) Quality of Precision Measuring Equipment.
there that a considerable amount of incoming equipment NCSL Newsletter, 18(1), 23–24.
was rejected – 21% over one period involving 62 con- NCSL (1996) Establishment and Adjustment of Calibration Inter-
tracts. Of those failing to meet specifications, 43% were vals: Rp-1 , National Conference of Standards Laboratories.
adjustable to specification but many were either condemned O’Connor, D.T. (2001) Test Engineering, Wiley, Chichester.
or returned for other, less serious, reasons. The author states Pearson, T.A. (1980) Case Study in Automated Test Equipment
that some improvement was evident at the time of study Use. NCSL Newsletter, 20(1), 19–20.
but that 1 in 6 complex facilities did not comply and 1 in Westmoreland, F.G. (1980) Intervals and Calibration System
10 needed action beyond adjustment. These findings would Management. NCSL Newsletter, 20(1), 10–13.
47: Internet Calibration
Richard A. Dudley
National Physical Laboratory, Teddington, UK
of the transportation, environmental, downtime, and cost
1 Use of Internet to Conduct Calibration 281 issues present in current calibration schemes. Additional
2 Type of Measurand Suited to Internet benefits emerge in the dissemination of measurement tech-
Calibration 281 niques and good practice equally to all laboratories using
3 Procedures for Carrying Out an Internet an Internet service.
Calibration 282 The use of the Internet to assist in metrology has taken a
number of forms. In the simplest case, the Internet acts as a
4 Example of an Internet Calibration Service 282
means of exchanging information between two remote sites
References 283 with human operators at each end using video or interactive
links (Filipski, 1999).
The second stage of development was remote monitoring
1 USE OF INTERNET TO CONDUCT of sensors (Lee and Schneeman, 1999 and O’Dowd et al.,
1997).
CALIBRATION
However, services are emerging that are truly remote
calibrations. These allow a laboratory to log on to a
Regular instrument calibration is an essential part of today’s
site, perform an accredited measurement and create a full
quality driven measurement environment, and a traceable
uncertainty evaluation instantly (Dudley and Ridler, 2000,
link to a national or international standard should be
2001).
present. Achieving traceability requires a laboratory to peri-
Services that offer online standard or instrument calibra-
odically send their standards to be calibrated at a National
tion have become known as Internet Calibration Services
Measurement Institute (NMI), acquiring a certificate and
or iCals.
correction values. The standards are measured under care-
fully controlled conditions at the NMI, but there is no
guarantee that these conditions will be reproduced when the
standards are used at the remote laboratory. Furthermore,
2 TYPE OF MEASURAND SUITED TO
in some cases, the value of the standards can be affected INTERNET CALIBRATION
by transport leading to an uncertainty component, which is
difficult to assess. Clearly, Internet calibration is not applicable to all metrol-
The downtime experienced by laboratories fulfilling cal- ogy areas, because physical constraints or human interac-
ibration schedules can be extremely disruptive and costly tion are essential for some measurements.
while the equipment is sent away. On the equipment’s However, measurements that rely on a standard issue
return, system checks, paperwork, and the update of soft instrument, on artifact, or are distinctive in the software
or hard calibration figures compounds the delays. Imple- developed as part of the calibration process, are open to
mentation of remote calibrations using the Internet as a Internet implementation. The implementation of remote
data transmission medium has emerged as a solution to all calibration via the Internet varies considerably with each
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
282 Units, Standards and Calibration
measurement example but some general rules are that the 2000, 2001). It combines the technology of remote moni-
system must have one of the following: toring, remote control, and NMI calibration techniques for
application in a microwave frequency measurement system
• Stable calibration artifact with easily verified perfor- called vector network analysers (VNA).
mance check. VNAs provide a swept frequency measurement of the
• Easily transported measurement artifact from which the transmission and reflection coefficients for an electrical net-
calibration is derived. work. Calibration is performed using instrument firmware
• Unique software for calibration and uncertainty and a set of standard devices, all of which are assumed to
calculation. be ideal and are available as standard items from the VNA
manufacture.
3 PROCEDURES FOR CARRYING OUT Correction of the measurement data to that of the NMI
comes via precision verification artifacts, air-spaced trans-
AN INTERNET CALIBRATION
mission lines, attenuators and terminations, whose proper-
ties change little over time. An external computer, running
A laboratory undertaking an Internet-based calibration
a specially designed code, controls the calibration and eval-
observes a number of advantages during the measure-
uates the uncertainty in measurement. The extension of
ment process compared to a conventional method. Once a
control across the Internet is ideal for this particular system.
computer has been connected to the measurement system,
A calibration procedure under non-Internet conditions
typically through GPIB (General Purpose Interface Bus) or
RS-232 ports, the laboratory only needs to log on to their requires the client to send the calibration laboratory their
calibration service provider. precision verification artifacts for periodic calibration. The
Once connected to the system, the measurement sys- NMI returns the verification device plus certificate and cor-
tem, standards, and instrumentation are known by the iCal rection data, Figure 1(a). The client must then implement
service. It instructs the operator, in the correct order, to their measurements using the verification information and
perform the measurements. introduce their own uncertainty budget. It should be noted
Effectively, the iCal system provides the operator with a that typically a 1 to 18 GHz calibration for a VNA results
soft procedure to perform the measurements assuming, of in several thousand correction values being generated. The
course, the operator is skilled and has been accredited with majority of VNA firmware does not allow inclusion of these
using the system. correction values and a client will usually have to resort to
Once the measurement procedure is completed, the iCal creation of proprietary control software or use a spreadsheet
system generates the required final data with uncertainties to correct measured data using the calibration information.
ascertained from the measured and database information, The Internet version, while not removing the need for
providing the operator with a certificate if applicable. physical movement between sites of the instrument’s stan-
During the measurement process, there is potential for the dard reference, does simplify the process, Figure 1(b).
iCal system to provide measurement assistance screens with When the NMI receives the client standard, all correction
video or procedural details for specific parts of the process. factors are stored in an online database. The standard is
In no way does an iCal system ‘de-skill’ the calibration returned, but now all a client needs to do to measure a
operator; instead, it provides the maximum assistance to device with traceability is to connect their control PC to
ensure an accurate and efficient calibration. the service Web pages.
In addition, the iCal service provides the administra- While online, the client enters the required measurement
tion with any measurement procedural changes through parameters and is offered options based on the knowledge
new international standards and ensures all groups are fol- the NMI has about the client’s equipment. From this point,
lowing common guidelines. Finally, data warehousing and the entire measurement process is controlled by NMI and
historical records are automatically kept on the iCal service the need for clients to assess their own uncertainty budget
Web site and can be accessed and checked with minimal is removed.
demands on the service users. There exists great potential for many new applications of
Internet calibration and metrology. Applications are being
investigated in the medical and optical networks markets.
4 EXAMPLE OF AN INTERNET The optical communications network provides a method of
CALIBRATION SERVICE a real standard transfer for wavelength measurements.
If telecommunications networks were all-optical – they
The first true iCal service was developed by the UK’s are presently a combination of optical and electrical – it
National Physical Laboratory (NPL), (Dudley and Ridler, would be possible for an NMI to distribute a wavelength
Internet Calibration 283
Cal software VNA firmware
Remote
laboratory
measurements
Instrument Electrical Electrical Instrument
Ref. line standards standards
Database Certificate Certificate Analysis software
Standard and cal data transported
(a)
Cal software
Remote
laboratory
measurements
Instrument User Ref. line
database Instrument
Certificate
Ref. line
calibration
(b)
Calibration laboratory Client laboratory
Figure 1. (a) Basic components of Internet calibration system: artifacts, instrument and control firmware, producing a certificate for
the remote laboratories’ artifacts. (b) The Internet system removes the need for traveling verification artifacts by allowing direct access
to the calibration software and client historical database.
standard into the network creating a ‘real’ international Dudley, R. and Ridler, N. (2001) Internet Calibration Direct
standard artifact. to National Measurement Standards for Automatic Network
Other possibilities such as frequency, noise, and pos- Analysers, in IMTC 2001 Proceedings of the 18 th IEEE
Instrumentation and Measurement Technology Conference,
sibly time stamping could be placed on the back of the 21–23 May, Budapest.
wavelength standard for added value. Futher information is
Filipski, P. (1999) SIMnet – A Collaborative Tool for Metrology
found on http://www.internetcalibrations.com.
in the Americas, in Proceedings of 16 th IEEE Instrumentation
and Measurement Conference, Vol. 2. May.
Lee, K. and Schneeman, R. (1999) Distributed Measurement and
REFERENCES Control Based on the IEEE 1451 Smart Transducer Interface
Standards, in Proceedings of 16 th IEEE Instrumentation and
Measurement Conference, Vol. 2. May.
Dudley, R. and Ridler, N. (2000) Internet-based Calibrations of
Electrical Quantities at the UK’s National Physical Laboratory, O’Dowd, R., Maxwell, D., Farrell, T. and Dunne, J. (1997)
in Proceedings of the NCSL Workshop and Symposium 2000 Remote Characterization of Optoelectronic Devices Over the
(National Conference of Standards Laboratories), 16–20 July Internet, in Proceedings of 4 th Optical Fibre Measurement Con-
Westin Harbour Castle, Toronto, Ontario, Canada. ference, October (pp. 155–158).
48:Common Sources of Errors in Measurement
Systems
Dietrich Hofmann
Steinbeis Transfer Centre for Quality Assurance & Quality Measurement, Jena, Germany
A measurement is an experimental process to acquire new
1 Definition of Errors in Measurement Systems 289 knowledge about a product.
2 How Errors Arise in Measurement Systems 290 A measurement is the realization of planned actions for
3 Terms Used to Describe Errors 290 the quantitative comparison of a measurand with a
unit.
4 Types of Errors in Defined Classes 290
A measurand is the physical quantity that is to be measured.
5 List of Error Sources in Measurements 292
Generally, a measurand is the input of a measurement
6 Standards on Error Description 292 system.
7 Uncertainties of Measurements 293 Measuring is a process for ascertaining or determining the
8 Glossary of Terms for Errors and quantity of a product by application of some object of
Uncertainties 294 known size or capacity, or by comparison with some fixed
Reference 294 unit – see www.iso.ch/9000e/2000rev 8e.htm.
A product is the result of a process after ISO 9000:2000,
point 3.4.2 – see http://www.iso.ch.
1 DEFINITION OF ERRORS IN Four generic product categories exist as follows: service,
MEASUREMENT SYSTEMS software, hardware, processed material.
Many products comprise elements belonging to different
A number of crucial definitions are first needed to ensure generic product categories. Whether the product is then
clarity of discussion. These are as follows. called service, software, hardware, or processed mate-
rial depends on the dominant element. A process is a
Measurement error is defined as the difference between the
set of interrelated or interacting activities that transform
distorted information and the undistorted information about
inputs into outputs (ISO 9000:2000, point 3.4.1) – see
a measured product, expressed in its measurands. In short,
http://www.iso.ch. Inputs to a process are generally outputs
an error is defined as real (untrue, wrong, false, no go)
to other processes.
value at the output of a measurement system minus ideal
(true, good, right, go) value at the input of a measurement A measurement process is a set of operations used to
system according to (1): determine the value of a quantity (ISO 9000:2000, point
3.10.2) – see http://www.iso.ch.
x = xr − xi (1) Measuring equipment is the measuring instrument, soft-
ware, measurement standard, reference material or auxil-
where x is the error of measurement, xr is the real untrue iary apparatus, or a combination, necessary to realize a
measurement value, and xi is the ideal true measurement measurement process (ISO 9000:2000, point 3.10.4) – see
value. http://www.iso.ch.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
290 Error and Uncertainty
If the measurement system has the general structure given
Process in Figure 1, the following errors may appear for a general
Nonelectrical measurand Computer output signal measurement task:
• input error
Sensor Transducer Converter Computer Indicator
• sensor error
• signal transmission error 1
Standardized digital signal
• transducer error
Standardized analog electrical signal • signal transmission error 2
• converter error
Electrical measurement signal
• signal transmission error 3
• computer error
Figure 1. Measurement chain.
• signal transmission error 4
• indication error.
A measurement system or measuring system is the totality
of all measuring equipments and auxiliary means to obtain
a measuring result.
3 TERMS USED TO DESCRIBE ERRORS
A measurement signal is the quantity in a measuring
Globalization of the economy is strongly driving inter-
instrument or measuring equipment, which is unequivocally
national standardization and accreditation programs, for
related to the measurand.
example, ISO 9000 and ISO 17025, are the means of assur-
A measuring chain (Figure 1) is the structure of elements
ing quality control of components fabricated throughout the
of a measuring equipment or measuring system, which
world.
channels the transfer of a measurement signal from the input
Accrediting organizations such as American Association
of the measurand to the output of the measured value.
for Laboratory Accreditation (A2LA) experienced double
A measured value is a value that is unequivocally related digit annual growth in the past decade, particularly in the
to the measurand at the output of a measuring chain. dimensional metrology field (classified under mechanical).
Most of these standards and accreditation programs
require traceable measurements and hence measurement
2 HOW ERRORS ARISE IN uncertainty statements – see www.mel.nist.gov/proj/sfnmi.
MEASUREMENT SYSTEMS htm.
Recent developments in metrology use
A measurement under ideal conditions has no errors.
• the International System of Units (SI),
Real measurement results, however, will always contain
• the Guide to the Expression of Uncertainty in Measure-
measurement errors of varying magnitudes. A systematic
ment (GUM), and
(clearly defined process) and systemic (all encompassing)
• the International Vocabulary of Basic and General
approach is needed to identify every source of error that
Terms in Metrology (VIM),
can arise in a given measuring system. It is then neces-
sary to decide their magnitude and impact on the prevailing as a starting point for the application of terms to describe
operational conditions. errors in the frame of uncertainties (see Figure 2).
Measurement system errors can only be defined in rela-
tion to the solution of a real specific measurement task.
If the errors of measurement systems given in technical
4 TYPES OF ERRORS IN DEFINED
documentation are specified, then one has to decide how CLASSES
that information relates to which
Systematic error (bias) is a permanent deflection in
• measurand the same direction from the true value. It can be
• input corrected. Bias and long-term variability are controlled by
• elements of the measurement system monitoring measurements against a check standard over
• auxiliary means time.
• measurement method Random error is a short-term scattering of values around
• output a mean value. It cannot be corrected on an individual mea-
• kind of reading surement basis. Random errors are expressed by statistical
• environmental conditions. methods.
Common Sources of Errors in Measurement Systems 291
data analysis that employs a variety of techniques (mostly
SI VIM GUM
International Guide to Expression graphical) to
International
System of Vocabulary of of Uncertainty
Metrology in Measurement 1. maximize insight into a data set,
Units
2. uncover underlying structure,
3. extract important variables,
Base units Measurement Uncertainty 4. detect outliers and anomalies,
Derived units Measurand Errors analysis
Measured value Evaluation
5. test underlying assumptions,
Etalons
Calibration chains Standards Reliability 6. develop parsimonious models, and
Coherence Calibration Estimation 7. determine optimal factor settings.
Decimal system Traceability Error types
Most EDA techniques are graphical in nature with a few
Figure 2. Main tasks of SI (http://physics.nist.gov/cuu/units), quantitative techniques included. The reason for the heavy
VIM (http://www.abnt.org.br/ISO DGuide 99999(E).pdf), and reliance on graphics is that graphics gives the analysts
GUM (http://www.gum.dk/home.html). unparalleled power to reveal the structural secrets of data
and to be always ready to gain some new, often unsus-
A fundamental source for theory and application of statis- pected, insight into the measured data in combination with
tical methods is the NIST/SEMATECH Engineering Statis- the inherent pattern-recognition capabilities of man – see
tics Handbook (NIST/SEMATECH e-Handbook of Statisti- http://www.itl.nist.gov/div898/handbook/eda/section1/
cal Method – see http://www.itl.nist.gov/div898/handbook/, eda11.htm.
26.03.2003). Most of the graphical techniques outlined in Figure 3
It describes, and uses in detail, the so-called EDA are demonstrated in the case studies given in the above-
method. Exploratory data analysis (EDA) is an approach for mentioned Web site.
Autocorrelation Bihistogram: Block plot: Bootstrap plot:
plot: 1.3.3.1 1.3.3.2 1.3.3.3 1.3.3.4
Box–Cox Box–Cox Box plot: Complex
linearity plot: normality plot: 1.3.3.7 demodulation
1.3.3.5 1.3.3.6 amplitude plot:
1.3.3.8
Complex Contour plot: DEX scatter DEX mean
demodulation 1.3.3.10 plot: 1.3.3.11 plot: 1.3.3.12
phase plot:
1.3.3.9
Figure 3. Graphical techniques in alphabetical order (excerpt) (http://www.itl.nist.gov/div898/handbook/eda/section3/eda33.htm).
292 Error and Uncertainty
5 LIST OF ERROR SOURCES IN drift is usually not a problem for measurements with short
MEASUREMENTS calibration cycles – see http://www.itl.nist.gov/div898/hand
book/mpc/section4/mpc453.htm.
Systematic errors or bias are repeatable errors existing
with the specified source; these can be adjusted out or
compensated for. 5.4 Hysteresis
The terms ‘bias’ and ‘systematic error’ have the same
Hysteresis is a retardation of the effect when the forces
meaning. Bias is defined in the International Vocabulary
acting upon a body are changed (as in viscosity or internal
of Metrology, VIM, as the difference between the mea-
friction); for example, a lagging in the values of result-
surement result and its unknown ‘true value’. It can often
ing magnetization in a magnetic material (as iron) because
be estimated and/or eliminated by calibration to a reference
of a changing magnetizing force. Hysteresis represents the
standard – see http://www.itl.nist.gov/div898/handbook/
history dependence of a physical system under real envi-
mpc/section7/mpc7.htm#VIM.
ronmental conditions – see http://www.lassp.cornell.edu/
To investigate sources of systematic errors, a general
sethna/hysteresis/WhatIsHysteresis.html.
checklist of error sources in measurement should be used,
Specific devices will posses their own set of additional
which has been collected by specialists working in the field
error sources. A checklist needs to be developed and
concerned. The main sources are
matured. The following is an example of such a list.
Hall Effect measurement error checklist – see http://
5.1 Lack of gauge resolution www.eeel.nist.gov/812/errs.htm.
1. Are the probes or wires making good contact to the
Resolution better called (but rarely done so) discrimination sample?
is the ability of the measurement system to detect and faith- 2. Are the contact I -V characteristics linear?
fully indicate small enough changes in the characteristic 3. Is any contact much higher in the resistance than the
of the measurement result – see http://www.itl.nist.gov/div
others?
898/handbook/mpc/section4/mpc451.htm. 4. Do the voltages reach equilibrium quickly after current
reversal?
5. Is there visible damage (cracks, especially around the
5.2 Lack of linearity
contacts)?
6. Is the sample being used in the dark?
A test of linearity starts by establishing a plot of the
7. Is the sample temperature uniform?
measured values versus corresponding values of the ref-
8. Are dissimilar wiring materials used resulting in large
erence standards. This obtains an indication of whether
temperature gradients across the wiring?
or not the points fall on a straight line with slope equal
to 1, which indicates linearity (proportional variation) –
see http://www.itl.nist.gov/div898/handbook/mpc/section4/
mpc452.htm.
6 STANDARDS ON ERROR DESCRIPTION
Nonlinearities of gauges can be caused by the follow-
Standards, when they are part of a single and coherent set of
ing facts:
standards, promote market efficiency and expansion, foster
• gauge is not properly calibrated at the lower and upper international trade, encourage competition and lower barri-
ends of the operating range, ers to market entry, diffuse new technologies, protect con-
• errors in the values at the maximum or minimum range, sumers against unsafe or substandard products, and provide
• worn gauge, trust and reliability – see http://www.iso.ch/iso/en/comm
• internal design problems (in, say the electronic units of centre/events/2002/standardsforservices.html.
the gauge). On the other hand, people talk about raising standards
when they perceive slackness in the ropes of control, when
they see a sloppiness infiltrating the verities of life, when
5.3 Drift they begin to be fearful about life-diminishing certainties.
Talk of standards is to talk about conservation, about
Drift is defined as a slow change in the response of a gauge. protecting the past in its imagined superiority and security,
Short-term drift is frequently caused by heat buildup in and defending the future through strong leadership – see
the instrument during the time of measurement. Long-term http://epaa.asu.edu/epaa/v6n10/c6.htm.
Common Sources of Errors in Measurement Systems 293
Comprehensive and precise terminologies have an extra- • measurement standards that generally are negligibly
ordinary importance to the industry. The following are small in comparison with the uncertainty of the mea-
examples: surement system in total.
ISO/TR 7178:1983 Standard deviation is the generally used parameter for
Liquid flow measurement in open channels – velocity-area expressing the value of uncertainty. It is defined as the
methods – investigation of total error. point of inflection in the Gaussian normal distribution –
ISO 8655-6:2002 see Figure 4 – of randomly scattering values.
Piston-operated volumetric apparatus – Part 6: Gravimetric Evaluation of uncertainty is an ongoing process that can
methods for the determination of measurement error. consume time and resources. It can also require the services
ISO/IEC 10021-7:1997/Amd 1:1998 of someone who is familiar with statistical data analysis
Security error diagnostic codes. techniques – see http://www.itl.nist.gov/div898/handbook/
ISO/IEC 10166-2:1991/Amd 1:1995 mpc/section5/mpc51.htm.
Minor enhancement for additional error. Measurement result is the best estimate of the value of the
ISO 12142:2001 measurand.
Electronic imaging – media error monitoring and reporting Standard uncertainty in the measurement result is expressed
techniques for verification of stored data on optical digital with a standard deviation – see http://physics.nist.gov/cuu/
data disks. Uncertainty/glossary.html.
ISO 7509:2000 Combined standard uncertainty in the measurement result
Plastic piping systems – glass-reinforced thermosetting has to be calculated by the error propagation law for
plastic (GRP) pipes – determination of time for failure statistical characteristics.
under sustained internal pressure. Type A uncertainty in the measurement result is expressed
ISO 10365:1992 as a deviation evaluated by the method of statistical analysis
Adhesives – designation of main failure pattern. from a series of observations.
ISO 11782-1:1998 Type B uncertainty in the measurement result is expressed
Corrosion of metals and alloys – corrosion fatigue testing – as a deviation evaluated by methods other than the statistical
Part 1: Cycles to failure testing. analyses of series of observations.
ISO 13953:2001 Expanded uncertainty in the measurement result is
Polyethylene (PE) pipes and fittings – determination of the expressed as the quantity of an interval that is large enough
tensile strength and failure mode of test pieces from a butt- to include that fraction of the distribution function of the
fused joint. measured values that might be reasonably attributed to the
measurand.
A special kind of expanded uncertainty is the confi-
7 UNCERTAINTIES OF MEASUREMENTS dence interval of a Gaussian normal distribution – see
http://mathworld.wolfram.com/GaussianDistribution.html –
Uncertainty is a measure of the ‘goodness’ of a result. where the factor a > 1.
Without such a measure, it is impossible to judge the Coverage factor is a numerical factor used as a multiplier
fitness of the value as a basis for making decisions relat- or the combined standard uncertainty in order to obtain
ing to health, safety, commerce, or scientific excellence –
see http://www.itl.nist.gov/div898/handbook/mpc/section5/
Gaussian or
mpc5.htm. ‘normal’
Uncertainty of measurement is a parameter associated distribution
with the result of a measurement that characterizes the fg (X)
dispersion of the values that could reasonably be expected –
see http://physics.nist.gov/cuu/Uncertainty/glossary.html.
In this model, it is understood that the result of the mea-
surement is the best estimate of the value of the measurand, 0.0214 0.0214
and that all components of uncertainty, including those aris- 0.00135 0.1359 0.3413 0.3413 0.1359 0.00135
ing from systematic effects, contribute to the dispersion.
−3s −2s −s 0 s 2s 3s
Typical examples for systematic effects are uncertainties of x
• systematic errors that generally can be corrected either Figure 4. The Gaussian normal distribution (http://hyperphysics.
numerically or technically, phy-astr.gsu.edu/hbase/math/gaufcn.html).
294 Error and Uncertainty
Table 1. Selected glossary of terms on error and uncertainty (http://www.rit.edu/∼vwlsps/uncertainties/uncertaintiespart2.
html#glossary).
Term Brief definition
Absolute error An error in a quantity, having the same units as the quantity. For
c = (2.95 ± 0.07) m s−1 , the absolute error is 0.07 m s−1 .
Accuracy How close a measurement is to being correct. g = 9.7 m s−2 on earth is more accurate
than g = 9.532706 m s−2 . See precision.
Average When several measurements of a quantity are made, the sum of the measurements
divided by the number of measurements.
Average deviation The average of the absolute value of the differences between each measurement and the
average. See standard deviation.
Confidence level The fraction of measurements that can be expected to lie within a given range.
Deviation A measure of range of measurements from the average. Also called error or uncertainty.
Error A measure of range of measurements from the average. Also called deviation or
uncertainty.
Estimated uncertainty An uncertainty estimated by the observer based on his or her knowledge of the
experiment and the equipment.
Gaussian distribution The familiar bell-shaped distribution, assumed for random errors. Also called normal
distribution.
Independent variables Changing the value of one variable has no effect on other variables. Error propagation
assumes this independence.
Instrument limit of Error (ILE) Smallest reading that an observer can make from an instrument. This is generally
smaller than the least count.
Least count Size of the smallest division on a scale. Typically the ILE equals the least count or 1/2
or 1/5 of the least count.
Precision The number of significant figures in a measurement. g = 9.532706 m s−2 is more precise
than g = 9.7 m s−2 .
Propagation of errors Method of determining an uncertainty in a function of independent variables each with
their own uncertainty.
Random error Deviations from the ‘true value’ can be equally likely to be higher or lower than the
true value. See systematic error.
Relative error The ratio of absolute error to the average, Dx/x. Also called percentage error or
fractional uncertainty. See absolute error.
Standard deviation The statistical measure of uncertainty. See average deviation.
Systematic error A situation where all measurements fall above or below the ‘true value’. To realize and
correct systematic errors is difficult.
Uncertainty A measure of range of measurements from the average. Also called deviation or error.
an expanded uncertainty of combined measurements – see Table 1 provides a selected glossary of terms on error
http://physics.nist.gov/cuu/Uncertainty/glossary.html. and uncertainty from that source.
Problems of uncertainties and error propagation and Terms used in the description of static system
their solutions are collected in http://www.rit.edu/∼vwlsps/ performance are also covered in Article 57, Static
uncertainties/Uncertaintiespart2.html#problems/. Considerations of General Instrumentation, Volume 1
The issue of error and uncertainty is covered in more and Article 58, Description of Accuracy, Linearity,
depth in Article 54, Explanation of Key Error and and Drift, Volume 1. Terms used to cover dynamic
Uncertainty Concepts and Terms, Volume 1 and other performance are dealt with in Article 59, Introduction
articles of the Subsection ‘Errors and Uncertainty’. to the Dynamic Regime of Measurement Systems,
Volume 1.
8 GLOSSARY OF TERMS FOR ERRORS REFERENCE
AND UNCERTAINTIES
The Web sites cited above are possibly the best sources of
A glossary of terms for errors and uncertainties is avail- information on the terms and concepts for errors and uncertainty.
Generally, texts are not sufficiently standardized – often using
able that can be used as a checklist for error handling – terms that are now defunct – to act as definitive sources; their use
see http://www.rit.edu/∼vwlsps/uncertainties/Uncertainties is, as other expressions of what is not an easy topic to be made
part2.html#glossary. crisp and accurate.
49:General Characterization of Systematic and
Stochastic Errors
Martin Halaj
Slovak University of Technology, Bratislava, Slovak Republic
Following the definition, let us consider n measurements
1 Definition of a Systematic Error 295 of the same measurand X, carried out under the same con-
ditions that are designated x1 , x2 , . . . , xn . Arithmetic mean
2 Nature of a Systematic Error 296
x̄ is the most used estimation of the measurement result
3 Definition of a Random Error 297
1
n
4 Nature of a Random Error 298
x̄ = x (1)
5 Comparison of Systematic and Random n i=1 i
Errors 299
6 Example of a Systematic Error Situation 300 If xctv is the conventional true value of the measurand,
7 Example of Random Error Situation 301 absolute systematic error x is expressed as
8 Errors and Measurement Result 302
x = x̄ − xctv (2)
Further Reading 303
Relative systematic error δx is calculated as
x
δx = (3)
1 DEFINITION OF A SYSTEMATIC xcvt
ERROR
Each particular measurement xi from the series of n mea-
surement is affected by the measurement error. Its system-
General definition states that error (of measurement) is atic and random part is graphically represented in Figure 1.
a result of a measurement minus a true value of the General interest usually excludes systematic errors from
measurand (see VIM, International Vocabulary of Basic and the measurement results. In this situation, the VIM states the
General Terms in Metrology, 3.10). so-called uncorrected result of a measurement as being the
Since a true value cannot be determined, in practice a measurement result before correction for systematic error
conventional true value is used. Measurement error consists (see VIM 3.3).
of two parts, indicated as systematic and random error. On the other hand, the corrected result of measurement is
The same document defines systematic error as a mean the result of a measurement after correction for systematic
that would result from an infinite number of measurements error (see VIM 3.4). Correction is then a negative value of
of the same measurand carried out under repeatability con- the known part of the systematic error that must be added
ditions minus a true value of the measurand (see VIM 3.14). algebraically to the uncorrected result of measurement.
Such definition implies that systematic error is equal to Like true value, systematic errors, and their causes cannot
overall error minus random error. be completely known. Therefore, only an estimation of the
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
296 Error and Uncertainty
Figure 1. Graphical representation of the systematic error.
systematic error can be expressed. Such an estimation can
be done by three ways:
1. analyzing the measurement process;
2. using available information on measuring instruments
(they can be obtained from manufacturer’s information,
calibration certificates, etc.);
3. utilizing experiences from previous measurements.
Not all systematic errors can be estimated. Real
systematic error can differ from that being estimated and
subsequently being corrected. It appears in the measurement
result in the form of the measurement uncertainty, usually
being evaluated by the type B method (see also Article 55,
Figure 2. Micrometer with shifted zero position.
Uncertainty Determination, Volume 1).
Errors in Signal Systems, Volume 1; and Article 51,
2 NATURE OF A SYSTEMATIC ERROR Errors in Digital Signal Systems, Volume 1):
Systematic error is caused by any factors that systemati- 1. Measuring method : improperly selected methods affect
cally affect measurement of the same measurand performed measurement the same way. For indirect measure-
under the same conditions. ments, rounded or uncorrected values of constants are
Unlike random error, systematic errors tend to be con- often used, for example, π = 3.141 or improper grav-
sistently either positive or negative. Because of this, sys- ity’s acceleration rate g, and so on.
tematic error is sometimes considered to be a bias in the 2. Measuring instrument: systematic error causes constant
measurement (see also Article 50, Errors in Signal Sys- difference of the calibration curve from the character-
tems, Volume 1; and Article 51, Errors in Digital Signal istic curve.
Systems, Volume 1). 3. Measuring conditions: measuring instruments being
For instance, if the micrometer has a shifted zero position used in conditions differing from that specified by the
as shown in Figure 2, all measured dimensions are greater manufacturer. A difference in temperature is the most
than the correct value. Subsequently, the mean of measured common cause.
values exceeds the conventional true value, as shown in 4. Operator: For example, wrong reading from the line
Figure 1. scale. Use of digital indicators avoids this error.
As with all errors, systematic errors also arise from four
main sources (see also Article 48, Common Sources of Theoretical division of systematic errors states that two
Errors in Measurement Systems, Volume 1; Article 50, categories of errors exist – additive and multiplicative.
General Characterization of Systematic and Stochastic Errors 297
Additive errors are those values that do not change with and the specified straight line. When expressed simply
the measured value, multiplicative errors change their value as the linearity error, it is assumed to be independent
with change in measured value. The most common additive linearity error.
errors are offset errors and linearity errors; multiplicative
error is presented as a gain error. Systematic error of the
measuring instrument is manifested in most cases by a 3 DEFINITION OF A RANDOM ERROR
combination of those types of errors. Let us describe them
in more detail: The VIM defines random error as a result of a measurement
minus the mean that would result from an infinite number
1. Offset error: (see Figure 3a) the difference, under spec-
of measurements of the same measurand carried out under
ified conditions of use, between the actual output
repeatability conditions (see VIM 3.13). Again, such defi-
value and the specified minimum value of the out-
nition implies that random error is equal to the error minus
put range when the input is at the lower range value.
systematic error as shown in Figure 1.
Usually is expressed as a percentage of the speci-
Only a finite number of measurements can be made
fied span.
in practice; therefore, it is possible to determine only an
2. Gain error: (see Figure 3b) slope of the characteristic
estimate of random error. A finite number of measurements
curve differs from that of the calibration curve. The
can be plotted in the so-called histogram (see Figure 4),
absolute value of the gain error depends on the input
utilizing data as shown in Figure 1.
value; the relative error remains the same.
One should observe the whole range of obtained data
3. Linearity error: (see Figure 3c) the absolute value of
and then divide the x-axis into suitable even-width bins.
the maximum deviation between the calibration curve
The number of measured values that fit within particular
intervals is plotted on the y-axis. This procedure gives the
so-called frequency distribution of the measured values.
Let us imagine that a large number of measurements with
high accuracy will produce a particular histogram with a
smoother shape. For an infinite number of measurements
and infinitesimally small intervals (bins), this histogram
transforms to a smooth curve that can be described by
a theoretical equation and be replaced by a probability
distribution function (see Figure 5).
Measurement theory most often assumes that a normal
Gaussian distribution, also known as a bell curve, can
describe the occurrence of any given error (see Figure 6).
This assumption is based on the central limit theorem
Figure 3. Systematic errors of the measuring instrument. Figure 4. Histogram of the measured data.
298 Error and Uncertainty
in either direction. The equation describing this curve is
1 −(x − µ)2
G(x) = √ exp (4)
σ 2π 2σ 2
The Gaussian distribution has two independent parameters
that completely determine the shape of the curve – the mean
µ and the standard deviation σ . The standard deviation σ
describes the width of the bell; the mean (xm ) lies on the
axis of symmetry of the bell (see Figure 6). Both parameters
are estimated by statistical methods – see also Article 55,
Uncertainty Determination, Volume 1; Article 52, Error
Models, Error Budgets and their Calculation, Volume 1;
and Article 53, Calculation and Treatment of Errors,
Volume 1.
Figure 5. Replacing histogram by the probability distribution 4 NATURE OF A RANDOM ERROR
function.
A random error is caused by any factor that randomly
affects measurement of the measurand. The important thing
about random error is that it does not have any con-
sistent effect on the group of measured data. Instead, it
pushes values obtained by measurement up or down ran-
domly.
This means that if we could see all of the random errors in
a distribution they would have to add up to 0 – there would
be as many negative errors as positive ones. This introduces
the important property of random error – it adds variabil-
ity to the data but does not affect the average performance
of the group – the mean (see Figure 9). Because of this,
random error is sometimes considered as noise – see also
Figure 6. Gaussian distribution. Article 48, Common Sources of Errors in Measurement
Systems, Volume 1; Article 50, Errors in Signal Systems,
Volume 1; and Article 51, Errors in Digital Signal Sys-
that does not necessarily always apply; its use must be tems, Volume 1.
confirmed by statistical tests. The probability that a measured value appears in the
The probability of finding a measurement in the range intervals ±σ , ±2σ , and ±3σ is equal to the area bounded
[x, x + dx] is equal to the area under the curve in that by the Gaussian curve within particular intervals – see
range. The curve is normalized to have a total area of 1, Figure 7.
which is why its amplitude is not also a free parameter. The As stated before, the standard deviation σ describes the
distribution is symmetric; an error is equally likely to occur width of the bell, as shown in Figure 8. A higher standard
Figure 7. Probability of the presence of measured value within the range.
General Characterization of Systematic and Stochastic Errors 299
The standard deviation, σ , obtained as a second root of
the variance, is characteristic of measured data variability.
Standard deviation σ is also designated as σX2 , σ 2 (X).
Estimation of the standard deviation for the set of mea-
sured data is the empirical standard deviation x1 , x2 , . . . , xn
defined as
1 n
s= (x − x̄)2 (9)
n − 1 i=1 i
Variability characteristics are used as a basis for determina-
tion of the measurement uncertainty. The Type A method
for the uncertainty evaluation, utilizing statistical methods,
states that
Figure 8. Influence of sigma on the bell curve.
s2
uA = (10)
deviation means large errors are more likely to exist. The n
mean µ (estimated usually as the arithmetic mean x̄) lies
on the axis of symmetry of the bell. where uA – standard uncertainty evaluated by the Type A
The mean (also called expectation, expected value) is method – (see also Article 55, Uncertainty Determina-
defined as tion, Volume 1).
∞
µ= xf (x) dx (5)
−∞
5 COMPARISON OF SYSTEMATIC
An estimation of the mean µ for the set of measured values
x1 , x2 , . . . , xn used in practice is the arithmetic mean x̄ is AND RANDOM ERRORS
given by
Systematic errors are caused by any factors that systemati-
1
n
x̄ = x (6) cally affect measurement of the same measurand performed
n i=1 i under the same conditions. They tend to be consistently
either positive or negative; therefore, they are sometimes
The variance of the random variable X (measured quantity)
considered to be the bias in measurement. Their value
is defined as
remains, under identical conditions, the same, or its changes
∞
can be predicted and evaluated.
σ =
2
(x − µ)2 f (x) dx (7)
−∞
On the other hand, random errors are caused by any
factors that randomly affect measurement. They do not
As can be seen from (7), the variance of the random have any consistent effect on the group of measured data.
variable is stated as the square of random variable units. Random errors add variability (estimated by the standard
The empirical variance is the estimation of variance for deviation σ ) to the data but do not affect the average perfor-
the same set of measured values x1 , x2 , . . . , xn mance of the group (estimated by the mean µ). Therefore,
random error is sometimes considered to be noise.
1
n
s2 = (x − x̄)2 (8) These facts are documented in Figure 9. Three different
n − 1 i=1 i situations can occur:
Probability
Probability
Probability
(a) (b) (c)
Figure 9. Influence of systematic and random errors.
300 Error and Uncertainty
1. Systematic errors affecting both measurements are the All measurement is affected by some systematic errors.
same; therefore, the mean of both data sets is the same. The tendency is to find their sources and evaluate their
If their standard deviations differ, it points to a strong values before taking final measurements.
effect of random errors – see Figure 9(a). Systematic errors arise again from four main sources –
2. Systematic errors differ; therefore, the mean of both see also Article 48, Common Sources of Errors in Mea-
data sets differs. Their standard deviations are the same surement Systems, Volume 1; Article 50, Errors in Sig-
so it points to a strong effect of systematic errors – see nal Systems, Volume 1; and Article 51, Errors in Digital
Figure 9(b). Signal Systems, Volume 1:
3. Both systematic and random errors differ; therefore, the
mean and standard deviations of both data sets differ. 1. Measuring method : sometimes uncorrected constants
This is the most general case when both systematic and are used, for example, gravity’s acceleration rate, air
random errors affect measurements proportionally, as density, atmospheric pressure, and so on.
shown in Figure 9(c). 2. Measuring instrument: as described before, additive or
multiplicative errors occur in measuring instruments.
Determination of systematic errors is based on analysis They usually result in (a) offset error, (b) gain error,
of their sources – see also Article 48, Common Sources of or (c) linearity error. A combination of all these errors
Errors in Measurement Systems, Volume 1; Article 50, occurs very often.
Errors in Signal Systems, Volume 1; and Article 51, 3. Measuring conditions: they can differ from that spec-
Errors in Digital Signal Systems, Volume 1. ified by the manufacturer, standards, internal instruc-
We tend to first determine systematic errors and subse- tion, and so on. Temperature affects measurement in
quently either exclude them or correct them. Not excluded most cases; besides that, so does humidity, atmospheric
systematic errors have an unknown value so that they are pressure, and so on.
estimated. Such estimation contains its own error some- 4. Operator: neglecting to set the measuring chain to zero
times called nonexcluded systematic error. before measurement, if required, introduces bias in all
For a particular measurement, it has a random character measurements.
with zero mean and certain variability. Determination and
estimation of systematic errors can be a very complicated Let us mention only a few examples of the sources of
and difficult process. In practice, we often try to exclude common systematic errors:
only large systematic errors and keep the rest of them within
certain limits. 1. Uncorrected constants. Gravity’s acceleration rate on
Random errors are manifested in the case of repeated 45◦ northern latitude, used as normal (standard) gra-
measurement and can be determined from measured data. vity’s acceleration rate is g = 9.806 65 m/s2 , 65◦ nor-
They cannot be known so that they cannot be eliminated. thern latitude is g = 9.823 03 m/s2 .
We can, however, describe their stochastic behavior. The All measurement where g occurs can be affected if
mean of random errors is zero; their variability is described using an incorrect value. For example, measurement
by their standard deviation, which serves as a basis for of the pressure using a U-tube filled with mercury,
evaluating their size. where the same difference in levels h = 0.1 m is mea-
sured on both latitudes, differs by about 22.27 Pa (see
Figure 10). This value represents an approximate rela-
tive error of 0.17% of the pressure measurement.
6 EXAMPLE OF A SYSTEMATIC ERROR
2. Temperature effects. Using the same example of a U-
SITUATION tube pressure measurement, let us measure the pressure
at the same place, having the same difference between
Systematic error is caused by any factors that systematically two mercury levels h = 0.1 m. The first measurement
affect measurement under the same conditions. Their value is performed at 0 ◦ C, the second one at 20 ◦ C. The
(positive or negative) is either constant or can be calculated difference between the two measured values at the
according to a known relationship. two temperatures is approximately 48.52 Pa, resulting
As systematic errors can be estimated, their estimations in relative error of 0.36%. Of interest is the fact that
can be subtracted from the measurement result so that the using water instead of mercury gives a relative error of
corrected measurement result can be obtained. While those 0.8% for the same measurement example.
errors are only estimated, and some of them we are not able
to estimate, they occur in a measurement result in the form The value of systematic error can be obtained by calibra-
of measurement uncertainty. tion – for example, the difference of the real weight mass
General Characterization of Systematic and Stochastic Errors 301
examples of random error sources can be given here, those
mainly experienced being as follows:
1. Measuring method : improperly selected method and
measuring instrument for the intended measurement.
For example, a level position of foamed liquid cannot
D
be measured by the ultrasonic ranger; another type of
measuring principle must be selected that avoids the
effect of foam on measurement results.
2. Measuring instrument: wrong or defective parts of
measuring chain. Simple examples are defective con-
tact surfaces of micrometer, transient phenomena in
electronic parts of the measurement chain, and so on.
3. Measuring conditions: the measuring instrument is
used in conditions that vary among repeated measure-
Figure 10. Example of the systematic error situation. ments – temperature, humidity, atmospheric pressure,
and so on. For example, the ultrasonic distance sensor
from the nominal value. Uncertainty of the systematic error is very sensitive to varying measurement conditions.
estimation is also known. 4. Operator: the classic and often seen source of ran-
dom measurement errors. Usually a wrong reading out
from the line scale occurs, followed by a different use
7 EXAMPLE OF RANDOM ERROR
of the measuring instrument in the repeated measure-
SITUATION ment (different positioning of the micrometer on the
shaft, different depth of the thermometer in the liq-
The nature of random error implies that those errors are uid, etc.).
caused by any factors that randomly affect measurement –
see also Article 48, Common Sources of Errors in Mea- Figure 11 can be used as a slightly artificial example of
surement Systems, Volume 1; Article 50, Errors in Sig- the random errors sources when measuring product length
nal Systems, Volume 1; and Article 51, Errors in Digital with a simple caliper. The measuring method generates
Signal Systems, Volume 1. A wide variety of such fac- random errors by incorrect positioning of the measured
tors can occur in everyday life and, therefore, only a few object. Some kind of fixture should be used to assure the
Figure 11. Example of the random error sources.
302 Error and Uncertainty
same position issued for each measured object. Wrong or to standard measurement uncertainty caused by random
defective jaws of the caliper contribute to random errors of error. Normal distribution of the random error is assumed
the measuring instrument. in most cases.
When positioning the measured object and caliper, dif- Correctness of such an assumption can be verified by
ferent parts of the caliper jaws are in contact with the application of appropriate statistical tests. If repeated mea-
measured object. surements under the same conditions are available, the
Measuring conditions changing between the individual measurement result will be given as the arithmetic mean
measurements can affect the measurement results sig- of the measured value and the uncertainty will be given as
nificantly. Temperature changes have a strong influence an experimental standard deviation of the arithmetic mean.
in cases in which thermal expansion coefficients of the If only one measurement is available, standard deviation
measured object and the measuring instrument differ is determined on the basis of other information on proba-
significantly. bility measurement of the random error (e.g. on the basis
The operator usually contributes to random errors. Read- of previous measurements).
ing out the value of scale can depend strongly on the Estimations of systematic errors are based on analysis
operator’s skills and mood. This can be avoided by using a of the measuring process (measuring system, measure-
digital display readout device; however, the digital mea- ment conditions, measuring method, influence of oper-
suring device might introduce random errors caused by ator, etc.). Uncertainties characterizing those estimations
resolution of the displaying device. are determined on the basis of process analysis and dif-
When measurement is performed once, or only a few ferent information obtained from the manufacturer of the
times, random error is estimated by a different way – measuring system, calibration certificates, literature, and
not by the type A method. For example, in the case of so on.
single measurement, all errors and their uncertainties are The following sample situation occurs often in practice
estimated by the type B method. Therefore, uncertainties (see also Article 55, Uncertainty Determination, Vol-
of the random errors must also be evaluated by the type ume 1):
B method.
1. Limits ±zmax around the estimation of the systematic
error can be determined. Values of the systematic
8 ERRORS AND MEASUREMENT errors fit within this interval. The standard uncertainty
u() of the systematic error is then calculated as
RESULT
zmax
u() = (11)
Measurement result is given as a measured value (esti- k
mated) minus measurement error. If the errors could be
known perfectly, a measurement result as defined above where k is the value corresponding to the selected
would be the final result. As the errors cannot be determined approximation of the probability distribution (see
totally, but only their estimations are known, a measure- Figure 12).
ment result must comprise also of the characteristics of 2. When using a digital measuring instrument, resolution
the quality of such estimation. Error estimation is consid- of a displaying device represents one possible error
ered as a random quantity with an estimation of its quality source. Even if the values in the case of repeated
expressed as standard deviation. Metrology uses the term measurements do not change, uncertainty is nonzero.
measurement uncertainty (see also Article 54, Explana- Uncertainty estimation is based on assumed uniform
tion of Key Error and Uncertainty Concepts and Terms, distribution
Volume 1). Standard measurement uncertainty is equal to δ(z)
standard deviation. u(z) = √ = 0.29δ(z) (12)
2 3
Random errors are estimated as zero and standard devia-
tion of the probability distribution of random errors is equal where δ(z) – resolution of the measuring device.
Figure 12. Probability distributions and corresponding k values.
General Characterization of Systematic and Stochastic Errors 303
3. When using an analog measuring instrument, the scale ISO 3494 (1976) Statistical Interpretation of Data – Power of
division δz is used instead of resolution of the display- Tests Relating to Means and Variances.
ing device. The uncertainty is calculated using (12). ISO 3534-1 (1993) Statistics – Vocabulary and Symbols – Part 1:
Probability and General Statistical Terms.
Measurement result is then stated as ISO 3534-2 (1993) Statistics – Vocabulary and Symbols – Part 2:
Statistical Quality Control .
x = x∗ ± U (13) ISO 3534-3 (1999) Statistics – Vocabulary and Symbols – Part 3:
Design of Experiments.
where U – expanded uncertainty is defined as ISO 5479 (1997) Statistical Interpretation of Data – Tests for
Departure from the Normal Distribution.
U = ku (14) ISO 8595 (1989) Interpretation of Statistical Data – Estimation of
a Median.
where k – expansion coefficient – see also Article 55,
ISO/TR 13425 (1995) Guide for the Selection of Statistical
Uncertainty Determination, Volume 1. Methods in Standardization and Specification.
ISO 16269-7 (2001) Statistical Interpretation of Data – Part 7:
Median – Estimation and Confidence Intervals.
FURTHER READING
EA4/02, EA (1999) Expression of the Uncertainty of Measurement
in Calibration.
ISO 2602 (1980) Statistical Interpretation of Test Results –
Estimation of the Mean – Confidence Interval . BIPM/IEC/ISO/OIML (1993) Guide to the Expression of Uncer-
tainty in Measurement (GUM).
ISO 2854 (1976) Statistical Interpretation of Data – Techniques
of Estimation and Tests Relating to Means and Variances. BIPM/IEC/ IFCC/ISO/OIML/IUPAC/IUPAP (1993) International
Vocabulary of Basic and General Terms in Metrology
ISO 3207 (1975) Statistical Interpretation of Data – Determina- (VIM).
tion of a Statistical Tolerance Interval .
ISO/IEC Guide 2 (1991) General Terms and their Definitions
ISO 3301 (1975) Statistical Interpretation of Data – Comparison Concerning Standardization and Related Activities.
of Two Means in the Case of Paired Observations.
50: Errors in Signal Systems
Eugen Georg Woschni
Technical University Chemnitz, Chemnitz, Germany
Because of the mean operation, the definition (1b) does
1 First Section Heading 304 not contain the whole information of (1c, d). It is, thus,
2 Errors in Dynamic Measurements 305 possible to calculate the mean-square value ε2 (t) if the
3 Typical Wave-shape Distortions 305 amplitude and phase errors are known, but the inverse
operation is not possible.
4 Statistical Errors 306
As shown in Article 30, Statistical Signal Representa-
5 Errors in Optimal Filtering 309 tions, Volume 1, noise signals z lead to statistical errors
6 An Example 310 characterized by the spectral power density S(ω), or the
References 313 mean-square value that is z2 , related to the standard devia-
tion σ by the relation
+∞
1 FIRST SECTION HEADING z2 = σ 2 = S(ω) dω (1e)
−∞
As Figure 1 shows, the error ε is given by the difference
between the output of the real system yreal and the ideal Noise is assumed to be generated at the input of the real
system without errors yid system Sχ (ω), as shown in Figure 1. The mean-square
value of the noise at the output is then the statistical
ε = yid – yreal (1a)
error (Woschni, 1988)
Equation (1a) may be interpreted either in the time or the +∞
frequency domain. In the time domain, it leads to a time z2output = S(ω)|Greal (jω)|2 dω (1f)
error ε(t). Of special interest is the mean-square value of −∞
this error
+∞
1 The total error ε2tot is the sum of both error components
ε2 = lim ε2 (t) dt (1b)
t→∞ 2T −∞ when there is no correlation between both error compo-
nents. Otherwise, correlation has to be taken into consider-
or in the frequency domain, leading to an amplitude error ation. This point is often overlooked.
|Gε (jω)|
|Gε (jω)| = |Gid (jω)| – |Greal (jω)| (1c) ε2tot = ε2 (t) + z2output (1g)
with a phase error of Additionally, there may arise aliasing errors if the sampling
theorem is violated. These are investigated in Article 31,
ϕε (jω) = ϕid (jω) – ϕreal (jω) (1d) Discrete Signal Theory, Volume 1.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Errors in Signal Systems 305
Yid (t ) x, y
G id (jw)
x (t ) Σ
e (t )
S (w) G real (jw)
Yreal (t )
Figure 1. Error definition.
∆T t
|G (jw)| Figure 3. Approximation of the output for the case of a
pulse-shaped input – pulse-shaped input ttr = T - - - -, transient
time too long - - - - -, ttr < T - - - - - -.
1 performance, for example, as in the selection of suitable
measuring instruments for a given task.
1√2
Consider the following example. Figure 3 illustrates
the problem of measuring a pulse-shaped input func-
tion (solid line), an approximation for numerous measur-
ing tasks (Woschni, 1972). In this figure, three cases are
presented. The long, dashed, triangular shaped response
represents the example in which the transient time is equal
to the pulse width, that is ttr = T . The pulse height is
wc,1 wc,u w still correctly indicated, whereas the pulse shape is strongly
distorted. In contrast to this particular case, considerable
Figure 2. Definition of the critical frequencies. error is introduced when measuring the pulse height if the
transient time is too long (short dashes). To permit proper
determination of the pulse shape, the transient rise time
2 ERRORS IN DYNAMIC has to be substantially shorter than the pulse width (chain-
MEASUREMENTS dash curve).
The same considerations may also be useful for analyzing
For a description of the dynamic behavior of systems, measuring errors. If one expects, for instance, a pulse-
it is usual and useful to introduce characteristic values shaped behavior of the output and receives, as the response,
that can be gained by means of approximations from the an output variable with a heavily prolonged trailing edge,
functions in the frequency and time domains (see Figure 2). then an error will be present (short dashes in Figure 3).
The amplitude–frequency characteristic, Figure 2, yields It is noteworthy that minor and medium errors in ampli-
the upper and lower critical frequencies fc,u and fc,l or ωc,u tude measurement frequently produce more detrimental
√ ∧
and ωc,l , where |G(jω)| decreases to 2 = 0.7 = 3 dB of effects in practice than do large measuring errors. For errors
the reference value. From the transient response h(t) (see of some 10%, the machine-made circuit element that has
Figure 4) the important transient time ttr , dead time td , delay been sized according to this measurement will function for
time tl , compensation time tc , and the maximum overshoot a certain time due to the safety margins, and will often
x0 are obtained. withstand initial testing. After having been produced in
The transient time is given approximately as (see Arti- series and operated for some time, however, all elements
cle 35, Systems in the Time Domain, Volume 1) fail at the same point according to the fatigue curve for the
number of stress reversals possible up to the failure of the
ttr = 3Tmax (2a)
element.
From the sampling theorem, the transient time and the On the other hand, major measuring errors will mostly
critical frequency are related by (Woschni, 2000) become evident during the testing period.
1
fc,u = t (2b) 3 TYPICAL WAVE-SHAPE DISTORTIONS
2 tr
In measuring systems engineering, these approximate con- Table 1 gives a survey of the more important measuring
siderations are of importance as convenient ways to estimate systems and their characteristic functions in both the time
306 Error and Uncertainty
and frequency domains (Woschni, 1988). The results in this 4 STATISTICAL ERRORS
table, in principle at least, may be used to find optimal
parameters of a system. For example, the best damping of As treated in Article 30, Statistical Signal Representa-
the spring–mass damping system, typical of a great number tions, Volume 1, the sources of noise may be described
of measuring systems, can be seen to be approximately by the Gaussian distribution density w(x) (Jaffe, 2000) as
1 (precise value 0.7). The transfer functions of typical shown in Figure 4.
√
systems, as illustrated in Table 1, allow the user to gain With the standard deviation σ = x 2 (t) it yields
a survey of typical curve distortions and their causes using 2
the methods for approximations given in Section 2 of this −x
exp
2σ 2
article. These considerations lead to the results summarized 1
w(x) = √ exp (3a)
in Table 2. In these cases, the input function is assumed to (2π)σ 2σ 2
be a pulse-shaped curve.
Table 1. Survey of typical systems and their characteristic functions.
Mathematical Transfer locus Amplitude phase curve
formulation
k +j |G (jw)| j (w)
k k
+r w
−j w
k +j |G (jw)| wg = 1/T1
1 + T1 p −p/4 k
+r k −p/4 w
w
−j w /√2 = 1/T
k /√2 −p/2
g 1 wg = 1/T1 w j (w)
k +j k |G (jw)| kT2 /T1 1/T2 = w0
1 + T1p + T 22 p2 −r +r w
kT2 T1/2T2 = D < 1
T1 w k D>1 D<1
1/T2 = w0 w −p
−j D > 1 1/T2 = w0 j (w)
k /T22 +j T2 1/T2 = w0
−kp2 −r +r
|G (jw)| D<1
w
w D>1
1 + T1p + T 22 p2 k /T2
D>1w −p D<1
1/T2 = w0 1/T2 = w0 j (w)
−j
e−tt p w = n p/tt +j p/2tt p/tt
|G (jw)| w
−r w +r k −p/2
−k +k −p
−j w = n p/2tt w j (w)
ke−tt p +j |G (jw)| w
k
1 + T1p −r +r t
w k /√2 j (w)
−j wg = 1/T1 w
k 1 +j
−r +r w
p 1 + T1 p |G (jw)|
−p/2
w p
−j j (w)
w
kp +j w |G (jw)| j (w)
k /T1 π/2
1 + T1p −r +r
kT1
π/4
−j
wg = 1/T1 w wg = 1/T1 w
Errors in Signal Systems 307
Table 1. (continued ).
Poles (×) and zeros (0) Function of time Examples for systems with
such behavior
Transfer function Weighting function
xa /1 = h(t ) aa /1 = g(t )
None xa 0 ∞ Systems with proportional
xa xa +∞
k
behavior (idealized)
xa 0 ∫ x dt = k
∆t 0 −∞ a 0
t=0 t t=0 t
+j
xa xa Temperature-measuring
−1/T1 k /T1
−r +r k arrangements without
protective tube; very heavily
−j damped systems capable of
t = 0 T1 t t = 0 T1 t
vibrating with proportional
behavior (idealized); system
with delay of the first order
and compensation
w0 +j Spring-mass damping
cos−1 D xa xa system, D ≤ 1,
−r +r k temperature-measuring
1/T2 = w0 arrangements with
t=0 t
−j t=0 t protective tube; D < 1,
approximation behavior
for vibration systems
+j Double Spring-mass damping
zero xa xa system without fixed
−r +r point D ≤ 1
1/T2 = w0
t=0 t t=0 t
−j
Cannot be exactly xa xa ∞ System with pure dead
represented. k ∆t 0
time (idealization), caused,
e.g., by pipeline, transport
t = 0 tt t = 0 tt t path, etc.
t
+j Real systems with dead
xa xa
−r +r time, e.g., temperature-
−1/T1 −j k k /T1 measuring arrangement
with heat conductivity
Factor e−tt p cannot be t=0 tT t t=0 t T t feed to transducer
1 1
exactly represented.
+j xa xa Real integrator
−r k k
+r
−1/T1
−j t = 0 T11 t t = 0 T1 t
+j xa xa ∞ Real differentiator
−r +r k /T1 ∆t 0 (xa ~^ xe)
T1
−1/T1
t = 0 T1 t=0 t
−j t
If this stochastic signal is passing through a linear system of S0 , a system of first order G(jω) = 1/(1 + jωRC) is
(Figure 5), the distribution is not changed. The frequency used – a so-called shaping filter – leading to the output
response of this system G(jω) influences the mean-square power Py
value of the noise at the output Py = y 2 (t) where Sx (ω) is
+∞
the spectral power density of the input signal, as shown 1
Py = S0 dω = S0 ω0 π (3c)
in Article 27, Signals in the Frequency Domain, Vol- −∞ 1 + ω2 R 2 C 2
ume 1 (Woschni, 1988; Jaffe, 2000)
+∞ If the output power Py is measured, the unknown spectral
Py = Sx (ω)|G(jω)|2 dω (3b) power density S0 can be calculated
−∞
An important example is the calibration of a white noise Py
S0 = (3d)
source as shown in Figure 6. To find out the unknown value ω0 π
308 Error and Uncertainty
Table 2. Typical wave-shape distortions and their causes.
- - - - - Output variable xa (t) Typical features Causes: amplitude curve, Typical features Remarks
Input variable xe (t) phase curve
Curve shape |G( jω)|; ϕ(ω)
Sharp corners |G(jw)| Ideal system, i.e. Practically not fully
t Steel edges fgu = 0; fgo = ∞ satisfied; only
Straight top ideal phase curve approximately for
w
j(w) slow process (T great)
T
1 w
Blurred corners |G (jw)| Upper limiting Short peaks of the input
t Oblique edges frequency fgo too function are not
Straight top wg w
low; lower reproduced or are
j(w) o limiting reproduced with
2 frequency fgu = 0 wrong height; remedy:
w
ideal phase curve to increase upper
limiting frequency,
e.g. by reducing
masses (idealized
system)
Sharp corners |G(jw)| Lower limiting Short peaks are correctly
Steep edges frequency fgu too indicated; the error is
t Inclined top high; upper increasing the faster
wgu w
j(w) limiting the process runs; pure
frequency static calibration and
3
w
fgo = ∞ measurement not
possible (e.g.
piezoelectric
measuring device)
Blurred corners |G(jw)| Upper and lower Insufficiencies according
t Oblique edges limiting to 2 and 3, i.e. short
Inclined top wgu wgo w frequencies peaks wrongly
j(w)
incorrect reproduced; no
4 w transfer of static
variables
Curve shape |G (jw)| Ideal amplitude With the exception of a
t correctly curve linear phase time shift, no
reproduced, only 1/tt w curve, i.e. pure alteration of the output
tt entire curve dead time variable; time shift is
shifted by Tt −1 often insignificant in
5 j(w) measurement
engineering; is often
detrimental, however,
in control engineering;
idealized case: with
G( jω) = constant
additional distortions
as 2,3,4
Curve consists of |G (jw)| Upper limiting Short peaks are not
pure e-functions frequency too low correctly reproduced;
T1
Large with time wg = 1/T1 w heavily damped
o
T1 constant T1 systems with
T1 Small
t Oblique edges −p/4 proportional behavior
6 Straight top −p/2 (idealized)
j(w)
Errors in Signal Systems 309
Table 2. (continued ).
- - - - - Output variable xa (t) Typical features Causes: amplitude curve, Typical features Remarks
Input variable xe (t) phase curve
Curve shape |G(jω)|; ϕ(ω)
D<1 For D > 1: blurred |G (jw)| Upper limiting Short peaks are not
corners frequency too correctly reproduced;
D≥1 Oblique edges D<1 low; additionally D− ≷ 1: systems with
D≥1
For D < 1: wo w for D < 1: spring–mass damping
oscillation resonance behavior, e.g.
−p/2 D>1 behavior vibration meters
7 D > 1: system for
−p D<1
j(jw) temperature
measurements with
protective tube
Integral curve |G (jw)| Integrator shows With nonelectric
exhibits no nonconstant phase integrators, the
corners at points 1/w rotation by −π/2; influence of the
where xe jumps amplitude curve masses is decisive for
8 integrates w decreases more this effect; upper
correctly at low strongly than with limiting frequency too
frequencies only −p/2 1/ω; has upper low in electric
limiting techniques
j(w) Ideal frequency for
integration
+∞
Differential curve |G (jw)| Differentiator shows Unavoidable with simple
rounded. nonconstant phase CR component
Differentiates rotation by +π/2; T1 = CR
correctly at low amplitude curve
frequencies only not ideal; has an
upper limiting
t w frequency fgo for
j(w) differentiation
p/2
−∞ Ideal
9
Ideal
5 ERRORS IN OPTIMAL FILTERING From 5 the total mean-square error is given by
∞
Gauss introduced the measure of the mean-square error ε2 , ε2 = Sxx (ω)|Gid (jω) − Greal (jω)Gcorr (jω)|2 dω
0
because this definition leads to closed mathematic solu- ∞
tions as for instance with optimal filtering. As shown + Szz (ω)|Gcorr (jω)|2 dω
in Section 5 of Article 38, Stability Issues, Volume 1, 0
the total mean-square error ε2 consists of two compo- = ρ2 + Pz (4)
nents.
As shown in Section 5 of Article 38, Stability Issues, with the assumption having been made that the disturbances
Volume 1, a correction or compensation system is series occur in the system prior to the correction computer.
connected with the real system; error can arise due to it When the error components and total error are plotted
having insufficient dynamic behavior ρ 2 and to the error against the correction degree a, relationships will always
due to the disturbance Pz . If there is no correlation between result as presented in Figure 7. The dynamic error compo-
signal source and disturbance source, as has been assumed nent ρ 2 decreases as a increases and disappears for a → ∞
here, the two error components must be added to arrive at (ideal correction), whereas the error component due to
the total error. the disturbance increases with a. Figure 8 provides the
310 Error and Uncertainty
1.6
e2
w (x )
Pz
r2 Pz
e2
1.2
s = 0.25
0.8 r2
s = 0.5
aopt a
Figure 7. Behavior of the components of the error as a function
of the degree of correction α.
0.4
s=1
|G (jw)|
|G corr (jw)|
−4 −3 −2 −1 0 1 2 3 4
x |G real (jw)|
Figure 4. Gaussian distribution density.
Szz
Sx (w) Py
G (jw) wc,real wc,corr w
Figure 8. Physical explanation of the increase of the noise in
Figure 5. Calculating the noise at the output of a system. case of correction.
into this range to be raised so that the error component
caused by the disturbance becomes the greater the higher
So R the value selected for ωc,corr . Therefore, there will always be
a minimum for the total error ε2 = ρ 2 + Pz , this minimum
Noise C being deeper and more pronounced the lower the inher-
Py
source ent disturbances of the uncorrected system are, the better
the dynamic behavior of this system will be. This mini-
mum corresponds to the case of an optimum filter according
to Schlitt (1960) and Levine (1996).
Figure 6. Calibration of a noise source.
physical demonstrative explanation for this. In this figure, 6 AN EXAMPLE
the amplitude responses |Greal (jω)| of the uncorrected
system as well as of the correction system |Gcorr (jω)| An example considers the first-order system treated in
are plotted. Article 35, Systems in the Time Domain, Volume 1
To compensate for the decrease of the uncorrected system and Article 37, Relationship Between Systems in the
at frequencies above the limiting frequency ωc,real , the cor- Time and Frequency Domain, Volume 1. The frequency
response there is
rection system must raise these frequencies correspondingly
up to its limiting frequency ωc,corr . However, this inevitably 1
G(jω) = (5a)
causes the spectral frequencies of the disturbances that fall 1 + jωT
Errors in Signal Systems 311
Let Sx0 be a white-structure signal applied to the input, ωc,s
1 + ω 2 T2
let the band be limited to ωx and let white noise Sz0 be the + Sz0 dω (5e)
0 1 + ω2 Tk2
type of interference. Furthermore, no correlation is assumed
between the signal and the interference. In order to obtain values that can be compared with the
The correction program of a series-connected system that original system and enable an estimate of the efficiency of
can be realized, in this case reads the correction to be generated, we calculate the error of the
system without any correction while assuming a bandwidth
1 + jωT limited to ωc = 1/T :
Gk (jω) = (5b)
1 + jωTk
2 ωx
ωc
1 − 1 dω +
ε02 = Sx0 dω
1 + jωT
of which the factor
0 ωc
T ωc,k
ωc
= (5c) ωx π Sz0
Tk ωc + Sz0 dω = Sx0 ωc − − (5f)
0 ωc 4 Sx0
indicates the improvement of the bandwidth by use of the
correction. If a microcomputer is used for the realization To find a more favorable solution in each case, and
of (5b), the limit frequency of the digital conversion, with thus obtain suggestions for synthesis, consider the follow-
the sampling time ts , is ing two cases: (i) the limiting frequency ωc,s , that is, the
sampling frequency, be adapted to the limiting frequency
ωc,s = π|ts (5d) of the corrected system ωc,k ; and (ii) the limiting fre-
quency ωc,s be adapted to the limiting frequency of the
Thus, the mean-square error is signal ωx :
ωc,s
ω2 Tk2 1. ωc,s = ωc,k . From (5e), for ωc,s = ωc,k < ωx , one ob-
ε2 = Sx0 dω + ωx −ωc,s
0 1 + ω2 Tk2 tains
102
102 104 108
102
6 104 106
e2 10
10
e20 e2 10 102
e20 108
1
1
10−1
10−1
10−2
10−2
1 10 102 103 104
−3 wc,k /wc
10
1 10 102 103 104
(a) wc,k /wc (b)
103
104 106
2
102 10 108
e2
e20
10
10−1
1 10 102 103 104 105 106 107
(c) wc,k /wc
Figure 9. Dependence of relative mean-square error on the bandwidth improved by correction: Sx0 /Sz0 = 102 ; - - - - -Sx0 /Sz0 = 104 ;
-·-·-·Sx0 /Sz0 = 106 ; Sx0 /Sz0 = 108 ; (a) ωx /ωc = 10; (b) ωx /ωc = 102 ; (c) ωx /ωc = 104 .
312 Error and Uncertainty
2
Sz0 ωc,k π 2. ωc,s = ωx (5i)
ε2 = Sx0 ωc,k 1+ 1−
Sx0 ωc2 4 2
Sz0 ωc,k
ε2 = Sx0 ωc,k 1+
Sz0 π Sx0 ωc2
ω
+ x −1+ (5g)
ωc,k Sx0 4 ωx ωx
× − arctan
ωc,k ωc,k
and for ωc,s = ωc,k > ωx , respectively,
Sz ωx
ω ω + 0 arctan (5j)
ε2 = Sx0 ωc,k x
− arctan x Sx0 ωc,k
ωc,k ωc,k
Any existing amplification or attenuation in the original
2
Sz0 ωc,k π Sz0 π
+ + 1− + (5h) system (static transmission factor) may be considered in
Sx0 ωc2 4 Sx0 4 the usual manner in the signal to noise ratio Sx0 /Sz0 .
103
10 e2
e2 e20 10
e20 10 102
1
102
102 10
10−1
1
10−2 104
10−1
104
10−3
10−2
106
10−4
10−3
106 108
10−5 10−4
108
10−6 10−5
1 10 102 103 104 1 10 102 103 104
(a) wc,k /wc (b) wc,k /wc
103
10 102
104
e2 102
e20
106
10
108
−1
10
1 10 102 103 104 105 106
(c) wc,k /wc
Figure 10. Continuation of Figure 9: -··-··-Sx0 /Sz0 = 10; . . . . Sx0 /Sz0 = 102 ; - - - - -Sx0 /Sz0 = 104 ; -·-·-·Sx0 /Sz0 = 106 ; Sx0 /Sz0 =
108 ; (a) ωx /ωc = 10; (b) ωx /ωc = 102 ; (c) ωx /ωc = 104 .
Errors in Signal Systems 313
In Figure 9(a, b, c), the results obtained from (5g, h) are cases, the increase in the fraction of errors caused by
shown for different values of the signal to noise ratio of 20, interference predominates because of the increase in the
40, 60, and 80 dB, that is for Sx0 /Sz0 = 102 , 104 , 106 , and high spectral frequencies.
108 , in relation to the bandwidth increase by correction of Finally, it should be emphasized that, in practice, further
ωc,k /ωc . The parameters of ωx /ωc were selected such that limitations in efficiency occur due to the sensitivity of
dynamically good systems (Figure 9a) as well as dynam- parameters, which in this investigation, have not been
ically poor systems (Figure 9c) are involved. The values taken into consideration and may arise due to possibly
for the mean-square error ε2 are related to the error of the existing nonlinearities (Woschni, 1967), (Bellmann, 1961)
uncorrected system ε02 according to (5f), so as to indicate and (Levine, 1996).
directly the reduction of the error.
All the diagrams for the adaptation of the bandwidth ωc,k
to the bandwidth of the program for system correction ωc,s REFERENCES
(Figure 9a) reveals a minimum corresponding to the case of
an optimum filter: while the dynamic portion of error (first Bellman, R. (1961) Adaptive Control Processes, Princeton Uni-
term (5e)) decreases with the rising degree of correction, versity Press, Princeton, NJ.
the interference-dependent portion increases with the rising Jaffe, R.C. (2000) Random Signals for Engineers using MATLAB
degree of correction, this being due to the increase in the and MATHCARD, AIP Press, New York.
spectral portions ω > ωc . The efficiency of the correction, Levine, W.S. (1996) The Control Handbook, CRC Press, New
therefore, will be higher, the dynamically better the system York.
and the smaller the interferences. This substantiates the Schlitt, H. (1960) Systemtheorie fur Regellose Vorgange, Springer,
finding obtained by Woschni (1969) on the basis of physical Berlin.
considerations. Woschni, E.-G. (1967) Parameterempfindlichkeit in der Meßte-
By adapting the bandwidth ωc,k to that of the input signal chnik, Dargestellt an Einigen Typischen Beispielen. Zeitschr.
(Figure 9b), the results shown in Figure 10(a, b, c) are Messen-Steuern, 10(4), 124–130.
obtained from (5j). The parameters were selected such that Woschni, E.-G. (1969) Inwieweit Spielt die Qualität eines Meßg-
direct comparison with the results represented in Figure 9 rößenaufnehmers beim Einsatz von On-Line-Rechnern Noch
eine Rolle? Zeitschr. Messen-Steuern, 12(10), 384–385.
is possible. In contrast to Figure 9(a), dependencies are
obtained, which tend asymptotically to a limit value. This Woschni, E.-G. (1972) Meßdynamik, Hirzel, Leipzig.
is because the interference-dependent portion does not rise Woschni, E.-G. (1988) Informations Technik, Huthig, Heidelberg.
any more because of band limitation. In dynamically very Woschni, E.-G. (2000) Approximations in Measurement – Today
poor systems, that is, for the case of large values of ωx /ωc still Necessary? in Proceedings XVI IMEKO World Congress,
(Figure 10b, c), the error increases with correction. In these Wien.
51: Errors in Digital Signal Systems
Luca P. Mari
Università Cattaneo, Castellanza (VA), Italy
carriers for univocally recognizable and mutually distin-
1 Uncertainty and Error Sources in Digital guishable symbols, so that physical transformations of sig-
Signals 314 nals are actually modeled and dealt with as formal data
2 List of Typical Digital Signal Uncertainties processing operations (i.e., mathematical functions) in the
and Errors 315 symbol domain.
3 Digital Signal Uncertainties and Errors in Transition from hardware to software forms for the
Data Acquisition 315 information medium is the paradigm of this transition.
4 Digital Signal Uncertainties and Errors in While the information content in analog systems resides
Data Processing 316 entirely as features and states of their hardware, in the
Related Articles 317 case of digital systems, a progressive virtualization of
the content can be used that varies from hard-wired
Further Reading 317
logic systems, to microprocessor-based programmable sys-
tems, to the so-called virtual instruments, whose oper-
ation could be even interpreted as if their character-
1 UNCERTAINTY AND ERROR SOURCES izing software layer is executed in an ideal hardware
IN DIGITAL SIGNALS subsystem (see also Article 144, DSP Chip Sets, Vol-
ume 3).
Digital systems are adopted today in a broad range of Corresponding to each of these levels of abstraction, arise
measurement applications. While supported by the current different issues in the metrological characterization of the
remarkable innovations in microelectronics and the related systems, and in particular in the identification of the typical
technologies, the reason for the widespread usage of digital sources of uncertainty/error, related to both the hardware
systems in measurement is that they are inherently less and the (multiple) software layers.
prone to error in communication and manipulation. The combined uncertainty uY summarizing the contribu-
Traditional measuring systems behave as transducers of tions of such multiple sources uXi depends additively on
measurands to quantities directly perceivable by human them, as formalized by the law of propagation of uncer-
beings, such as angular deflections of needles on graduated tainty, as recommended by the ISO Guide to the expression
scales. Their operation can be described in terms of physical of uncertainty in measurement (GUM) (the simplified ver-
transformations for which the interpretation of the physical sion of such a law is shown here, applicable in the case
states as information entities is left to the observer. In of statistically uncorrelated sources) (see also Article 54,
this case, any further data processing (leading to the so- Explanation of Key Error and Uncertainty Concepts
called indirect, or derived, measurement) is accomplished and Terms, Volume 1):
by devices external to the measuring system, if not done
manually by the observer.
N
On the other hand, characteristic of the concept of digi- ∂f 2 2
u2Y = uXi (1)
tal coding is the hypothesis that physical signals are simply i=1
∂Xi
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Errors in Digital Signal Systems 315
where f is the function modeling the relationship that links related to digital signals. On the other hand, ADC charac-
the measurand to its influence quantities. Equation (1) is teristics and behavior significantly influence the quality of
obtained as a first-order Taylor series approximation of the generated digital signals (conceptually definable as the
the model function computed in a (N -dimensional) point degree of correspondence with the originating analog sig-
assumed ‘sufficiently closer’ to the average values of the nals and operatively affecting the possibility of reconstruct-
quantities Xi and under the hypothesis that f is ‘sufficiently ing them from the converted digital signals) – see also Arti-
linear’ in the neighborhood of such a point. cle 138, Sampling Theory Relevant to DAS, Volume 3
While usually reasonably correct in the case of instru- The data processing subsystem is aimed at dealing with
mentation dealing with smoothly varying quantities, these digitally coded entities to transform them by means of suit-
assumptions could become quite critical for digital systems, ably implemented algorithms and to transfer them to remote
for then nonlinearities that sometimes are very strong, such devices. Uncertainties/errors can appear in both hardware
as those manifesting as the consequence of bugs in the and software layers because of the presence of physical fac-
software, are common. tors modifying the quantity on which the symbols are coded
Given the fundamental requirement to formalize any and the low quality of algorithms (or their implementations)
measurement result by expressing both a measurand value adopted in the processing of such symbols. The latter issue
and an estimation of its uncertainty, the usage of digital grows in relevance as the software adopted for metrological
signals and systems (particularly if under software control) purposes becomes more and more complex, as is the case
usually implies to trade off flexibility for complexity. with spreadsheets or virtual instruments.
Current developments in this area are particularly impor-
tant, as witnessed by the emerging applications of pattern
2 LIST OF TYPICAL DIGITAL SIGNAL recognition, automatic control, and data fusion based on
UNCERTAINTIES AND ERRORS the so-called soft computing paradigm in which techniques
such as neural networks and fuzzy logic inference are used
While in some specific cases digital systems integrally to exploit uncertainty and partial information.
operate on digitally coded entities (e.g. in some cases of The output subsystem is finally aimed at making the
counting in which the measurand is inherently discrete), processed data available to users and user devices (e.g.,
they are also widely used in the measurement of contin- actuators of control systems) while possibly converting such
uously varying quantities so that a preliminary stage of data to a corresponding analog form. At this stage, raw data
analog-to-digital conversion is implied. produced by the measuring system must be converted to
Once such a transduction has been completed, the opera- information that is meaningful to the intended users and
tions are performed on coded symbols, that is, on a purely useful to them.
algorithmic basis (see also Article 33, Coding Theory and The sources of possible uncertainties/errors in the expres-
its Application to Measurement, Volume 1). The results sion of measurement results from the digital signals repre-
are then fed into a device that acts as an output transducer, senting the instrument readings are multiple, all basically
which is sometimes required to convert the digital symbols related to the mathematical model of the measurement
back to analog signals. system – see Article 104, Models of the Measurement
As a consequence, a metrological characterization of Process, Volume 2.
digital systems involves the analysis of their behavior in While having been traditionally assigned to a human
reference to three general components, each of them being user, the definition and the metrological qualification of this
affected by specific causes of uncertainties/errors. model is now the main task of the knowledge-based intelli-
The input subsystem is set to acquire information on gent instruments – see Article 128, Nature and Scope of
the measurand from the environment and, when needed, AI Techniques, Volume 2.
converting it into an equivalent digital form.
Its general structure includes a sensor, a signal con-
ditioning component, and an analog-to-digital converter
(ADC) that, in PC-based systems, is usually part of a 3 DIGITAL SIGNAL UNCERTAINTIES
data acquisition card – see also Article 132, Data Acqui- AND ERRORS IN DATA ACQUISITION
sition Systems (DAS) in General, Volume 3; Article 136,
Quantizing Theory Relevant to DAS, Volume 3.
Digital signals are obtained as the output of such a sub- Digitalization of analog signals usually implies their time
system; hence, strictly speaking the input subsystem does and amplitude discretization, the two basic parameters qual-
not contribute to the budget of system uncertainties/errors ifying such operations being the sampling rate and the
316 Error and Uncertainty
amplitude resolution (also called the bit depth) of quan- 4 DIGITAL SIGNAL UNCERTAINTIES
tization, measured in samples per second and bits respec- AND ERRORS IN DATA PROCESSING
tively – see also Article 133, Amplifiers and Filters for
DAS, Volume 3.
The simplest kind of data processing is the one performed
Even in the case of an ideal ADC, limitations in the by systems computing the identity function, that is, pro-
size of the data storage devices and bandwidth of the data ducing as their output the same symbols given at their
transmission channels are sources of errors for the generated input, as the ideal behavior of an ideal digital transmission
digital signals: channel.
The sampling theorem shows that the information con- In this case, the presence of errors (generally caused
veyed by an analog signal is integrally maintained whenever by noise sources external to the channel) is modeled in
the signal is sampled at a rate greater than twice its band- statistical terms by recognizing that for each input sym-
width (for most applications the time interval between bol xi the channel does not deterministically produce
samples is kept constant). The usual technique of low- an output symbol yj but a conditional probability distri-
pass (sometimes band-pass) antialiasing filtering (see also bution P (yj |xi ) (for binary channels xi , yj ∈ {0, 1}, and
Article 125, Outline of Purpose of Analog Data Filters, P (0|xi ) + P (1|xi ) = 1).
Volume 2) is in fact a trade-off between two systematic The average value of − log2 (P (xi |yj )), called equivoca-
errors. Its application helps to avoid aliasing effects but tion and computed from P (yj |xi ) by means of the Bayes
removes any information contained in the cut-off portion theorem, represents the average information lost in the
of the signal spectrum. transmission process because of errors.
The number of intervals (sometimes called channels or From the channel equivocation H (X|Y ) and the source
cells) in which the amplitude range is subdivided in quan- entropy H (X), the channel capacity C is computed as
tization specifies the quantizer resolution, that is, the length
of the binary word coding each sample. This establishes the C = maxX (H (X) − H (X|Y )) (2)
amount of the error introduced by quantization.
In the simplest case of uniform quantization, when all a basic informational quantity, measured in bit per symbol
intervals have the same half-width a, each sample of (and more usually in bit per second by multiplying it by
amplitude x is associated with a channel i whose midpoint the rate of symbol transmission over the channel), whose
(dealt with as the reference value to be coded) is ci : physical grounds are clearly identified in the fundamen-
the quantization error is then x − ci corresponding to a tal relation:
maximum quantization error of ±0.5 least significant bits
S
(LSBs) and a null average quantization error. C = W log2 1 + (3)
N
Here again a trade-off is implied: to reduce the quantiza-
tion error, the bit depth of the code word must be increased. where W and S/N are the channel bandwidth and signal-
Thus, to enhance the accuracy of conversion by the ADC, to-noise ratio respectively).
its precision must also be increased. In the case in which the information flowing from the
To characterize the actual behavior of a physical ADC, source has a rate lower than the capacity C of the channel,
some further parameters have to be taken into account, several techniques can be adopted to reduce the probability
such as internal and external noise, settling time, short- of error at the receiver. All are based on the introduc-
term repeatability, and long-term stability, offset, linearity tion of redundancies and aimed at either error recognition
of gain, and, in the case in which two or more signals are or correction.
acquired at the same time, cross talk. Typical applications of digital signal processing in mea-
It is usual that the specifications for such parameters are surement are digital filtering and DFT/FFT computation
directly given by the ADC manufacturer as the interval (see also Article 143, DFT and FFTs, Volume 3).
±a that certainly (i.e. with probability = 1) contains the Higher-level operations are now common, for exam-
corresponding values/errors. ple, in computing statistical parameters as in the case of
This is the typical case in which the ISO GUM rec- DC/RMS measurement.
ommends type B uncertainty evaluations on the basis The fundamental parameters qualifying the arithmetic
of uniform probability distributions (see also Article 54, of a processor are its overflow, underflow, and round-off
Explanation of Key Error and Uncertainty Concepts error thresholds.
and Terms, Volume 1). The corresponding standard uncer- In the common case of the floating-point number repre-
√
tainties are then computed as a/ 3 and combined by means sentation (in which numbers are expressed as (−1)a × b ×
of (1). 10c where a ∈ {0, 1}, the mantissa b ∈ [1, 10) has a fixed
Errors in Digital Signal Systems 317
Table 1. Values of machine parameters in IEEE floating-point should also be taken into account for a complete metrolog-
arithmetic. ical qualification of the system.
Machine Single precision Double precision
parameter (32 bits) (64 bits)
Machine 2−24 ≈ 5.96 × 10−8 2−53 ≈ 1.11 × 10−16 RELATED ARTICLES
precision
Underflow 2−126 ≈ 1.18 × 10−38 2−1022 ≈ 2.23 × 10−308 Article 48, Common Sources of Errors in Measure-
threshold
ment Systems, Volume 1; Article 54, Explanation of Key
Overflow 2128 (1 − ε) ≈ 21024 (1 − ε) ≈
threshold 3.40 × 1038 1.79 × 10308 Error and Uncertainty Concepts and Terms, Volume 1;
Article 55, Uncertainty Determination, Volume 1; Arti-
cle 132, Data Acquisition Systems (DAS) in General,
number of digits, and the exponent c is an integer spanning Volume 3; Article 139, Analog-to-Digital (A/D) Convert-
between two fixed values, the overflow and the underflow ers, Volume 3.
thresholds depend on the maximum positive and negative
values of the exponent respectively.
On the other hand, round-off errors depend on the number FURTHER READING
of digits reserved for the mantissa; they are expressed in
terms of the machine precision, a value generally related ANSI/IEEE (1985) IEEE Standard for Binary Floating Point
to the characteristics of not only the processor arithmetic- Arithmetic, Std 754, New York.
logic unit (ALU) but also the adopted software platform or ANSI/IEEE (1987) IEEE Standard for Radix Independent Floating
compiler. This is an important source of complexity in the Point Arithmetic, Std 854, New York.
metrological qualification of data processing modules. ISO (1993) Guide to the Expression of Uncertainty in Measure-
A commonly implemented reference for the values of ment, International Organization for Standardization, Geneva.
these parameters is the IEEE Standard (see Table 1). Mathworks, manuals related to the MathLab software platform
The data processing subsystem is usually so complex (in particular Data acquisition toolbox – User’s guide, Version
that, instead of identifying all the relevant sources of 2), downloadable at http://www.mathworks.com.
uncertainty that would be required to apply (1), a black National Instruments, manuals related to the LabView soft-
box solution is sometimes adopted for its metrological ware platform (in particular LabView user manual, 7.2000 Edi-
tion; Measurements manual, 7.2000 Edition; Data acquisition
qualification. A reference data set is chosen that contains basics manual, 1.2000 Edition), downloadable at http://www.ni.
a collection of sampled input data with the corresponding com.
expected output; such input data are fed into the subsystem, National Physical Laboratory, Software Support for Metrology
and the results are compared with the references. From Project, Best Practice Guides, downloadable at http://www.npl.
the analysis of the obtained error, an estimation of the co.uk/ssfm.
uncertainty of the data processing results is then inferred. Patterson, D.A. and Hennessy, J.L. (1998) Computer Organiza-
The contribution of all possible hardware faults (and cor- tion & Design – The Hardware/Software Interface, Morgan
respondingly the degree of fault tolerance of the system) Kaufmann, New York.
Error Models, Error Budgets and their
52:
Calculation
Rudolf Palenčár
Slovak University of Technology, Bratislava, Slovak Republic
where xm (p) – the measured value transformed by Laplace
1 Definition of an Error Model 318 transform; xS (p) – transformed true value.
For simplicity, let us assume that xS is the value
2 Error Budget Principles 319
at the input of the measuring system and xm is the
3 Setting Up an Error Budget 320
value at the output of the measuring system. Sometimes,
4 Calculation of an Error Budget 320 value xS , at the input of the measuring system, is not
5 Evaluation of the Final Error Figures 324 identical with the value of the measurand. Its inaccessibility
6 Methods for Evaluation of Uncertainties 325 and common impact on the measurand and measuring
Further Reading 326 system may cause this. Similarly, the obtained value
xm may not need to be identical with the output value
of the measuring system due to such things as signal
processing taking place after the data leaves the measuring
1 DEFINITION OF AN ERROR MODEL system.
Error is treated as a time function when quantity X is time
Uncertainty determination and presentation is covered dependent, or if the transient processes in the measuring
in Article 55, Uncertainty Determination, Volume 1. system are still active, as shown in Figure 1.
This article introduces how errors and uncertainties of Measurement error x consists of the static error xst
measurement are dealt with where many are present when and the dynamic error xdy .
determining a measured value. Dynamic error arises when the measurand is time depen-
The measurement error x is defined as a difference dent and change occurs before the transient processes
between the measured value xm and true value xS of the are completed or when the measurand is time steady
measurand X but determination of the value in the time t ∗ arrives
before the transient processes in the measuring system are
x = xm − xS (1) over.
Measurement error (static and dynamic) consists of
If we want to introduce a factor of time, error is defined as the systematic and stochastic part (see also Article 49,
General Characterization of Systematic and Stochastic
x(t) = xm (t) − xS (t) (2) Errors, Volume 1).
If the transfer function of the measuring system is
The p domain for linear or quasi-linear measuring sys- represented as
tems is
xm (p)
S(p) = (4)
x(p) = xm (p) − xS (p) (3) xS (p)
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Error Models, Error Budgets and their Calculation 319
This transforms into the p domain as
xm (p) − xS (p)
xr (p) = = S(p) − 1 = Sr (p) (7)
xS (p)
The transfer function Sr (p) characterizes the transfer of the
error in the measuring system and does not depend on the
value at the system input.
Figure 2 shows schemes of the measuring system with
the influence of errors. This figure represents an example
for realistic functioning of a measuring system.
2 ERROR BUDGET PRINCIPLES
Absolute measurement error at the output of the measuring
system can consist of two parts:
Figure 1. Relationship between the measured and true value of 1. Inner error of the measuring system is caused by dif-
measurand. xs (t) – path of the input value; xm (t) – path of the ference between the theoretical St (p) and real transfer
output (measured) value; xst (t ∗ ) – static measurement error in function S(p).
time t ∗ ; xdy (t ∗ ) – dynamic measurement error in time t ∗ . 2. Error injected into the measuring system from the
surrounding by disturbance value z.
the error is then
Stated errors can comprise of systematic errors with
1 transfer functions SS (p) and SZS (p), and random errors
x(p) = 1 − x (p) (5)
S(p) m with transfer functions SR (p) and SZR (p). Figure 3 shows
individual cases of the errors acting.
The error value in the time domain is determined by the Error sources are associated with the
use of the reverse Laplace transformation.
Error x is also called an absolute error. • measuring system,
The ratio of the absolute error to the true value is called • measuring method,
the relative error. • environment,
In the time domain • operator.
xm (t) − xS (t) x (t) Details on individual error sources are stated in other
xr (t) = = m −1 (6) articles – see also Article 48, Common Sources of Errors
xS (t) xS (t)
Figure 2. Scheme of the measuring system and the influence
of errors. St (p) – theoretical transfer function of the measuring
system; Sr (p) – transfer function of the inner error; Sz (p) –
transfer function of the injected error. Figure 3. Influence of errors on the measuring system.
320 Error and Uncertainty
in Measurement Systems, Volume 1; Article 51, Errors Values xmi can be obtained as
in Digital Signal Systems, Volume 1; Article 50, Errors
in Signal Systems, Volume 1; and Article 49, General • results of individual measurements,
Characterization of Systematic and Stochastic Errors, • arithmetic means or medians or other statistical charac-
Volume 1. teristics of the set of repeated measurements,
As stated before, during measurement of the quantity y, • values of physical constants,
individual errors xi arise and the overall error y must • values of different parameters,
be determined. It is then necessary to determine for the • results of previous measurements, and so on,
quantity y, being a function of the quantities x1 , x2 , . . . , xn
(indirect measurement of the quantity y), that is, for estimations of errors when their uncertainties and
covariances are known.
y = f (x1 , x2 , . . . , xn ) (8)
Error is determined by the errors propagation law
3 SETTING UP AN ERROR BUDGET
n
y = Ci xi (9)
Basic principles are available that show how individual
i=1
errors combine and propagate (transfer and combination).
where Ci – sensitivity coefficients being obtained as partial Tables 1, 2, and 3 introduce basic schemes for the
derivation of the f function by individual quantities xi . transfer of the measurement errors in measuring systems,
Equation (9) was obtained by expanding function (8) into transfer functions of the measuring system seen from their
the Taylor series, neglecting members of high order. errors point of view, relative, and absolute errors, and
For determination of the uncertainty u(y) of the uncertainties of the measuring system.
error y (only if its estimation is known) using (9), Similar reasoning leads to the relationships given in
the uncertainties propagation law is described by the Table 4, which shows typical cases for different functions
expression (as the standard uncertainty is equal to the occurring in indirect measurement.
standard deviation):
n
n−1
n
u2 (y) = Ci2 u2 (xi ) + 2 Ci Cj u(xi , xj ) 4 CALCULATION OF AN ERROR
i=1 i=1 j >i BUDGET
(10)
where u(xi ) is the standard uncertainty of the error xi ;
For the measuring system shown in Figure 3, we can
and u(xi , xj ) isthe covariance between errors xi and
now determine the value of the measured quantity xm in
xj .
relation to the real quantity xs , as absolute and relative
Uncertainties u(xi ) and covariances u(xi , xj ) are
measurement error.
determined by the type A and type B evaluation – see
For measured value xm∗ , we can determine an estimation
also Article 53, Calculation and Treatment of Errors,
x ∗ of the absolute errors x and estimation (x/x)∗ of
Volume 1; Article 54, Explanation of Key Error and
the relative errors x/x, as well as for the uncertainty of
Uncertainty Concepts and Terms, Volume 1; and Arti-
estimation of such errors as u(x ∗ ), respectively u(x/x ∗ ).
cle 55, Uncertainty Determination, Volume 1.
The measurement result will then be an estimation xs∗
If the individual errors xi are not correlated, then
of the real value xs , obtained as the measured value xm∗
n corrected for systematic errors (their estimation) and the
u2 (y) = Ci2 u2 (xi ) (11) uncertainty of the estimations of those errors including the
i=1 uncertainty of random errors.
For a measuring system according to Figure 3, the
The measurement result y obtained by the measurement of
the Y quantity will be measured value xm in the p domain is given by:
y = ym − y ± U (y) (12) xm = St [1 + SS + SR ]xS + SZS ZS + SZR ZR (13)
where ym is the value obtained from (8) for measured The absolute error will be:
values xmi ; and U (y) is the expanded uncertainty of the
error y. x = [St (SS + SR ) − 1]xS + SZS ZS + SZR ZR (14)
Error Models, Error Budgets and their Calculation 321
Table 1. Transfer of errors and uncertainties through measuring system with serial sequencing of
the measuring instruments.
Serial sequencing of the measuring instruments
Transfer function S1 S2
x1 x2
Relative error +
x1 x2
Absolute error S2 x1 + S1 x2
x1 x2 x1 x2
Relative uncertainty u2 + u2 + 2u ,
x1 x2 x1 x2
Absolute uncertainty S1 u (x1 ) + S2 u (x2 ) ± 2S1 S2 u(x1 , x2 )
2 2 2 2
Used symbols:
S1 , S2 Transfer functions of the measuring instruments
x1 , x2 Absolute errors of the measuring instruments
x1 x2
, Relative errors of the measuring instruments
x1 x
2
x1 x2
u ,u Relative uncertainties of the measuring instruments (uncertainties of
x1 x2 the relative errors of the measuring instruments)
u(x1 ), u(x2 ) Absolute uncertainties of the measuring instruments (uncertainties of
the absolute errors of the measuring instruments)
x1 x2
u , Relative covariance among measuring instruments (covariance among
x1 x2 relative errors of the measuring instruments)
u(x1 , x2 ) Absolute covariance among measuring instruments (covariance
among absolute errors of the measuring instruments)
Table 2. Transfer of errors and uncertainties through measuring system with parallel sequencing of the measuring
instruments.
Parallel sequencing of the measuring instruments
Scheme of the measuring system
Transfer function S1 ± S2
S1 x1 S2 x2
Relative error +
S1 ± S2 x1 S1 ± S2 x2
Absolute error x1 + x2
S12 x1 S22 x2 S1 S2 x1 x2
Relative uncertainty u 2 + u 2 ± 2 u ,
(S1 ± S2 )2 x1 (S1 ± S2 )2 x2 (S1 ± S2 )2 x1 x2
Absolute uncertainty u (x1 ) + u (x2 ) ± 2u(x1 , x2 )
2 2
If absolute error is expressed as a function of xm , then If we know the transfer functions SS and SR exactly, as
well as the spurious influences and their transfer functions
1
x = 1 − x SZS ZS and SZR ZR , we can determine the real value of the
St (1 + SS + SR ) m measured quantity xS .
1 Where we do not know those values exactly, knowing
+ (S Z + SZR ZR ) (15)
St (1 + SS + SR ) ZS S only their estimations as well as the uncertainties of
322 Error and Uncertainty
Table 3. Transfer of errors and uncertainties through measuring system with feedback.
Sequencing of the measuring instruments with feedback
Scheme of the measuring system
S1
Transfer function
1 ± S1 S2
1 x1 S1 S 2 x2
Relative error ±
1 ∓ S1 S2 x1 1 ∓ S1 S2 x2
S1 S12 S 2
Absolute error x 1 ± x2
(1 ∓ S1 S2 )2 (1 ∓ S1 S2 )2
1 x1 S12 S22 x2 S1 S2 x1 x2
Relative uncertainty u 2 + u 2 ± 2 u ,
(1 ∓ S1 S2 )2 x1 (1 ∓ S1 S2 )2 x2 (1 ± S1 S2 )2 x1 x2
S12 S14 S22 S13 S2
Absolute uncertainty u2 (x1 ) + u2 (x2 ) ± 2 u(x1 , x2 )
(1 ∓ S1 S2 )4 (1 ∓ S1 S2 ) 4 (1 ± S1 S2 )2
Table 4. Determination of errors and uncertainties for indirect measurement.
n
Function y= xi
i=1
n
Absolute error y = xi
i=1
y n
xi xi
Relative error =
y x + x2 + · · · + xn xi
i=1 1
n
2
n−1 n
Absolute uncertainty u(y) = u (xi ) + 2 u(xi , xj )
i=1 i=1 j >i
n
u(y)
xi2 u2 (xi )
n−1 n
xi xj u(xi , xj )
Relative uncertainty = + 2
y i=1
(x 1 + x 2 + · · · + x n )2 x 2
i i−1 j >i
(x 1 + x 2 + · · · + x n )2 xi xj
Function y = x1 − x2
Absolute error y = x1 − x2
y x1 x1 x2 x2
Relative error = −
y x1 − x2 x1 x1 − x2 x2
Absolute uncertainty u(y) = u2 (x1 ) + u2 (x2 ) − 2u(x1 , x2 )
u(y) x12 u2 (x1 ) x22 u2 (x2 ) x1 x2 u(x1 , x2 )
Relative uncertainty = + −2
y (x1 − x2 )2 2
x1 (x1 − x2 )2 x22 (x1 − x2 )2 x1 x2
Function y = x1 x2 · · · xn
n
x1 x2 · · · xn
Absolute error y = xi
i=1
xi
Error Models, Error Budgets and their Calculation 323
Table 4. (continued ).
y xi n
Relative error =
y i=1
xi
n
(x1 x2 · · · xn )2 2
n−1 n
(x1 x2 · · · xn )2
Absolute uncertainty u(y) = u (x i ) + 2 u(xi , xj )
i=1
xi2 i=1 j >i
xi xj
n 2
u(y)
u (xi )
n−1 n
u(xi , xj )
Relative uncertainty = 2
+ 2
y i=1
x i i−1 j >i
xi xj
Function y = x1l1 x2l2 · · · xnln
n
l i x l1 x l2 · · · x ln
Absolute error y = 1 2 n
xi
i=1
xi
y xi n
Relative error = li
y i=1
xi
n 2
li (x1 x2 · · · xn )2 2
n−1 n
li lj (x1 x2 · · · xn )2
Absolute uncertainty u(y) = 2
u (x i ) + 2 u(xi , xj )
i=1
xi i=1 j >i
xi xj
n
u(y)
u2 (xi )
n−1 n
u(xi , xj )
Relative uncertainty = li2 2
+ 2 li lj
y i=1
x i i−1 j >i
xi xj
x1 x2 · · · xn
Function y=
z1 z2 · · · zm
n
x1 x2 · · · xn m
x1 x2 · · · xn
Absolute error y = xi − zj
i=1
x i z 1 z 2 · · · z m z z z · · · zm
j =1 j 1 2
y xi zj
n m
Relative error = −
y i=1
xi j =1
zj
n (x1 x2 · · · xn )2 2 m
(x1 x2 · · · xn )2 2
Absolute uncertainty u(y) = u (x i ) + u (zj )
x 2 (z1 z2 · · · zm )2
i=1 i
z2 (z z · · · zm )2
j =1 i 1 2
n−1
n
(x1 x2 · · · xn )2 (x1 x2 · · · xn )2
m−1 m
+2 u(x i , x j ) + 2 u(zi , zj )
i=1 j >i
xi xj (z1 z2 · · · zm )2 z z (z z · · · zm )2
i=1 j >i i j 1 2
n
m
(x1 x2 · · · xn )2
− u(xi , zj )
i=1 j =1
xi zj (z1 z2 · · · zm )2
u2 (xi ) u2 (zj )
n−1
u(y)
n m n m−1 m
u(xi , xj ) u(zi , zj )
Relative uncertainty = 2
+ 2
+ 2 + 2
y i=1
xi j =1
xj i−1 j >i
xi xj i−1 j >i
zi zj
n
m
u(xi , zj )
−
i−1 j =1
xi z j
those estimations, only an estimation of the real measured uncertainties of those estimations can be determined from
quantity xS∗ can be determined. measured data; or that we know them.
Transfer functions of random errors are estimated as For measured value xm∗ being valid (assuming SR∗ = 0,
zero, that is, SR∗ = 0, SZR
∗
= 0 and we presume that the ∗
SZR = 0) then:
324 Error and Uncertainty
1 S ∗ Z∗
x ∗ = 1 − ∗ ∗ xm + ∗ ZS S ∗ (16)
St (1 + SS ) St (1 + SS )
and the uncertainty of the absolute error is:
∗
(xm∗ − SZS ZS∗ )2 2 ∗
u2 (x ∗ ) = [u (SS ) + u2 (SR∗ )]
St (1 + SS∗ )4
2
1 ∗
+ [u2 (SZS ZS∗ ) + u2 (SZR
∗
ZR∗ )]
St2 (1 + SS∗ )2
(17)
for we do not assume dependence among individual
errors.
For error and its uncertainty determined by this proce-
dure in the p domain, we can then extract the error value
and its uncertainty using the reverse Laplace transforma-
tion.
This is based on the assumption that xs is the value at
the input to the measuring system and value xm is at the
output from the measuring system. Figure 4. Indirect measurement of electric current I , via use of
resistor R to generate voltage U from that current.
5 EVALUATION OF THE FINAL ERROR The manufacturer states that the digital voltmeter has the
FIGURES maximum permissible error 0.01% of the measured value
plus 0.005% of the measuring range for the measuring
If we proceed to use a time domain, the measurement result range 100 mV with the surrounding temperature within 15
will be expressed as to 35 ◦ C.
For the current determination, we use the Ohm’s law
x = xm∗ − x ∗ ± U (x ∗ ) (18) relationship:
U
where U (x ∗ ) – expanded uncertainty of the measure- I =
R
ment – is determined from the equation
where I – measured electric current (mA); U – directly
∗
U (x ) = ku(x ) ∗
(19) measured voltage drop (mV); R – resistance of the mea-
suring resistor ().
where k – expansion coefficient depending on the proba- We have obtained 10 values by measurement, as listed
bility distribution of the error x (value k = 2 is often in Table 5. Measurements were performed under the same
selected in practice). conditions.
Following are the estimation results from the mea-
Example: indirect measurement of the current by mea- sured values:
surement of the voltage drop We wish to measure the
1
10
electric current flowing through a circuit. We will mea- U= U = 50.44 mV
sure the voltage drop on the resistor with a nominal 10 i=1 i
value 1 resistor using a digital voltmeter, as shown in
Figure 4. Estimation of the electric current being measured is:
The surrounding temperature is within (22 ± 2) ◦ C. U 50.44
I¯ = = = 50.44 mA
The flowing electric current is expected to be around R 1
50 mA. The calibration certificate of the measuring resis-
Let us consider only the systematic error of the measuring
tor states that its value for the temperature 22 ◦ C and
resistor. This is
the current 50 mA is (1–0.0002) and the corresponding
expanded uncertainty for the expansion coefficient k = 2 is U 50.44
0.0002 . I = − 2
R = 0.0002 = −0.01 mA
R 1
Error Models, Error Budgets and their Calculation 325
Table 5. Values obtained by measurement.
No. 1 2 3 4 5 6 7 8 9 10
Ui (mV) 50.46 50.44 50.45 50.48 50.41 50.49 50.40 50.40 50.45 50.42
Table 6. Balance of errors and uncertainties.
Error Xi Estimation xi Standard Sensitivity Contribution to the
uncertainty coefficient Ci standard
u(xi ) uncertainty ui (y)
Random 0.00 mV 1.011 × 10−2 mV 1 mA/mV 1.011 × 10−2 mA
Voltmeter U 0.00 mV 0.58 × 10−2 mV 1 mA/mV 0.58 × 10−2 mA
Measuring resistor R −0.0002 0.0001 50.46 mA/ 0.55 × 10−2 mA
I −0.01 mA 1.27 × 10−2 mA
Components of the standard uncertainty are as follows: A summary of the procedure for the uncertainty deter-
mination gives the table of the uncertainties balance, see
1. Standard uncertainty of the voltage drop determined by Table 6 and see also Article 54, Explanation of Key
the type A method: Error and Uncertainty Concepts and Terms, Volume 1;
and Article 55, Uncertainty Determination, Volume 1.
1 10 The final measurement result is
uA (U ) = s(U ) = (U − U )2
10(10 − 1) i=1 i
I = (50.440 + 0.010 ± 0.026) mA = (50.450 ± 0.026) mA
−2
= 1.011 × 10 mV
2. Standard uncertainty of the voltage drop determined by 6 METHODS FOR EVALUATION OF
the type B method: UNCERTAINTIES
For the voltmeter used and considering the measured
value 50.45 mV, the maximum permissible error of the Uncertainties of individual errors are determined either
measuring instrument 0.01% of 50.45 mV plus 0.005% by type A evaluation or by type B evaluation (see also
of 100 mV, that is, approximately 0.01 mV. Considering Article 54, Explanation of Key Error and Uncertainty
the uniform distribution, standard uncertainty is uB (U ) = Concepts and Terms, Volume 1; Article 55, Uncertainty
√
0.01/ (3) = 0.0058 mV. Current flowing through the volt- Determination, Volume 1; Article 53, Calculation and
meter is neglected. The influence of the temperature that fits Treatment of Errors, Volume 1).
within the range 22 ◦ C ± 2 is covered in the basic permis- Type A evaluation is based on statistical analysis of
sible error of the voltmeter. measured data.
3. Standard uncertainty of the measuring resistor: Type B evaluation is every evaluation that is not of
The expanded uncertainty of the measuring resistor for type A.
k = 2 is equal to 0.0002 that matches the standard For example, in the case of a single measurement, all
uncertainty u(R) = 0.0002/2 = 0.0001 . The influence errors and their uncertainties are estimated by the type
of the temperature within the range (22 ± 2) ◦ C on the B method.
electrical resistance can be neglected compared to other Therefore, uncertainties of the random errors must also be
considered sources. evaluated by the type B method (e.g. utilizing experiences,
4. Standard uncertainty of the electric current measure- from data of the manufacturer etc.).
ment: As random errors are generated by different influences
We do not consider any covariances in this case as acting on the measuring process, it is more accurate to
we measure directly only the quantity of voltage drop. perform repeated measurement for their evaluation.
Estimations of the permissible error of the voltmeter and For n repeated measurements under the same conditions,
the measuring resistor are considered not to be correlated. ∗ ∗ ∗
when a set of values xm1 , xm2 , . . . , xmn is available,
Then: the measurement result will be given as the arithmetic
mean of the measured values x̄m∗ reduced on systematic
uI = CU2 u2A (U ) + CU2 u2B (U ) + CR2 u2 (R) = 1.27−2 mA error.
326 Error and Uncertainty
Overall uncertainty of the random error is evaluated by
n
n
n
the type A method according to equation u(y1 , y2 ) = C1i C2i u2 (xi ) + C1i C2j u(xi , xj )
i=1 i=1 j =i
(25)
sx
1 n
u= √ =
and for x1 , x2 , . . . , xn being commonly independent it
(x ∗ − x̄m∗ )2 (20)
n n(n − 1) i=1 mi is then
n
u(y1 , y2 ) = C1i C2i u2 (xi ) (26)
i=1
6.1 Covariances
where C1i respectively C1i – sensitivity coefficients of the
function f1 or f2 respectively.
Covariances among the estimations of individual errors Such an approach is used for excluding correlation when
determine how much those estimations are influenced by a proper measurement model is designed that comprises
common sources. only independent quantities.
Commonly bounded errors contribute to the overall 3. If the correlation coefficient cannot be determined
uncertainty more or less than only merging their uncer- and a proper model (excluding correlation) cannot be
tainties. designed, the maximum influence of correlation to the
Covariances can increase or decrease the resulting over- resulting uncertainty should be determined.
all uncertainty. It depends on their character (whether If y = f (x1 , x2 ) then:
increasing one error increases another error) and also on
the shape of the function that binds them to the output u2 (y) ≤ (|C1 u(x1 )| + |C2 u(x2 )|)2 = C12 u2 (x1 ) + C22 u2 (x2 )
quantity.
Type A evaluation of covariances between estimations of + 2|C1 C2 | u(x1 ) u(x2 ) (27)
two errors xi and xj is used when n measured values
of both quantities xi1 , xi2 , . . . , xin and xj 1 , xj 2 , . . . , xj n is If we do not then have enough information for accurate
available. Then evaluation of the covariances, and consequently also the
resulting uncertainty, the upper limit of the uncertainty can
1 n
be stated.
u(xi , xj ) = (x − x̄i )(xj k − x̄j ) (21)
n(n − 1) k=1 ik
Covariances between two estimations evaluated by the type FURTHER READING
B method can be determined by studying certificates of the Books
measuring instruments, from literature and so on, or from
calculation using the following procedure:
Hofmann, D. (1986) Handbuch Meβtechnik und Qualitätssicher-
1. For each pair of estimators, one estimates (on ung, 3rd edn, VEB Verlag Technik Berlin.
the basis of experiences) the correlation coefficient Sydenham, P.H. (1983) Handbook of Measurement Science I, II,
r(xi , xj ) expressing the degree of dependence John Wiley & Sons, New York.
between estimators.
Covariance is then obtained from the expression
Standards and guides
u(xi , xj ) = r(xi , xj ) u(xi ) u(xj ) (22)
ISO 2602:1980 Statistical Interpretation of Test Results – Estima-
2. If two quantities y1 and y2 act in indirect measurement tion of the Mean – Confidence Interval.
and they are defined by the following functions ISO 2854:1976 Statistical Interpretation of Data – Techniques of
Estimation and Tests Relating to Means and Variances.
y1 = f1 (x1 , x2 , . . . , xn ) (23) ISO 3207:1975 Statistical Interpretation of Data – Determination
of a Statistical Tolerance Interval.
y2 = f2 (x1 , x2 , . . . , xn ) (24) ISO 3301:1975 Statistical Interpretation of Data – Comparison of
two Means in the Case of Paired Observations.
the covariance between the quantities y1 and y2 is calcu- ISO 3494:1976 Statistical Interpretation of Data – Power of Tests
lated as Relating to Means and Variances.
Error Models, Error Budgets and their Calculation 327
ISO 3534-1:1993 Statistics – Vocabulary and Symbols – Part 1: ISO 16269-7:2001 Statistical Interpretation of Data – Part 7:
Probability and General Statistical Terms. Median – Estimation and Confidence Intervals.
ISO 3534-2:1993 Statistics – Vocabulary and Symbols – Part 2: DIN 1319:1985 Grundbegriffe der Messtechnik. Teil 4: Behandlug
Statistical Quality Control. von Unsicherheiten bei der Auswertung von Messungen.
ISO 3534-3:1999 Statistics – Vocabulary and Symbols – Part 3: EA-2/04, EA:1997 Expression of the Uncertainty of Measurement
Design of Experiments. in Calibration.
ISO 5479:1997 Statistical Interpretation of Data – Tests for Guide to the Expression of Uncertainty in Measurement (GUM).
Departure From the Normal Distribution. BIPM/IEC/ISO/OIML, 1993.
ISO 8595:1989 Interpretation of Statistical Data – Estimation of International Vocabulary of Basic and General Terms in Metrology
a Median. (VIM). BIPM/IEC/IFCC/ISO/OIML/IUPAC/IUPAP, 1993.
ISO/TR 13425:1995 Guide for the Selection of Statistical Methods ISO/IEC Guide 2:1991 General Terms and their Definitions
in Standardization and Specification. Concerning Standardization and Related Activities.
53: Calculation and Treatment of Errors
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
typical characteristics of large amounts of data (Wein-
1 Errors in Measurement 328 berg, 1975). McGhee et al. (1996) describes the pro-
2 Classification of Errors 329 cess of treating measurement data as consisting of the
following:
Related Articles 330
References 330
• Assessing the consistency of the data.
• Estimating the uncertainties in the result.
• Predicting the result from theory.
1 ERRORS IN MEASUREMENT • Comparing the data with physical theories and previous,
related experimental work.
All measurements contain some variability, which obscures
the true value of the measurand determination (Galyer and It is very well known that no measurement is exact.
Shotbolt, 1990; Doebelin, 2003; Gregory, 1973; Helfrick However, it should be made to an acceptable degree of
and Cooper, 1990; Hofmann, 1982; Rangan, Sarma and accuracy (Morris, 1997). It is, therefore, important to spec-
Mani, 1983; Taylor, 1982). ify the uncertainty of the measurement along with the
The variability in a measurement is called the error of result of the measurement. Representing uncertainty numer-
measurement, which is defined as follows: The error of a ically and graphically, as given by McGhee et al. (1996),
measurement is the discrepancy or difference between the is also described in Article 55, Uncertainty Determina-
result of a measurement and the absolute true value of the tion, Volume 1. There are a number of different sources
quantity measured. for measurement error, such as the measurement method,
An error of measurement is of little practical use when the measuring instrumentation, or the measurement envi-
considered in isolation. A number of measurements of ronment. Therefore, it is important to identify the most
the same true value, each of which has an associated significant source of errors so that less significant ones may
error of measurement, can be used to provide valuable be neglected if sufficiently small.
information regarding the accuracy and repeatability of There are also a large number of terms used to describe
measurements taken. different types of errors, such as systematic errors, random
Thus, measurement is more than just reading the display errors, parasitic error, method error, observer error, parallax
on the instrument and recording the data. It involves the error, interpolation error, error of indication, repeatability
process of treating and manipulating measurement data to error, rounding error, discrimination error, hysteresis error,
gain maximum benefit from them by extracting as much response error, datum error, zero error, influence error,
information as possible from the data. temperature error, supplementary error, total error, and so
Although this process of data reduction causes a loss on. The first step to be taken in the reduction of errors
of information about each specific measurement, an over- is understanding them. Some of these errors are covered
all benefit is achieved by the result of summarizing the here.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Calculation and Treatment of Errors 329
Classes
Error sources
Levels
Random Gross Systematic
Disturbance
Human Structural
power flows
Naturally occuring Man made Software Defects Others
All cosmic
Fundamental
and Electrical Mechanical Manufacture Material
fluctuations
terrestrial radiation
Other
Inductive energy forms
Others Thermal Shot switch-off
Jars Creep
Capacitive
switch-on Vibration Impacts Aging Drift
Figure 1. Classification of measurement errors.
2 CLASSIFICATION OF ERRORS 2.2 Systematic errors
Errors can be classified according to their characteristics.
These errors are due to systematic disturbance power flows,
Thus, Figure 1, which shows a summary of the three main
caused by human activities, or by structural imperfections
groups of error classes, distinguishes between gross errors,
or defects in software, inadequacies in manufacturing meth-
systematic errors, and random errors (McGhee et al., 1996).
ods, inadequacies in materials of instrument construction,
or some other unidentifiable or unspecified causes.
2.1 Gross errors When a measurand is repeatedly measured under the
same conditions, systematic errors always have the same
These are largely human errors. They may be caused by value. In other words although they are constant under
man-made power disturbance flows in any of the COMET- constant conditions, they are caused by intrinsic errors
MAN energy forms described in Article 109, Structure arising from long-term variations.
and Energy in Sensor Systems, Volume 2. They also Systematic errors have the property that the next indi-
include misreading of instruments leading to incorrect vidual measurement error value can be predicted once the
recording of measurement results, which may compound error law has been determined.
any error due to miscalculation. Other possibilities are Systematic errors are divided into instrumental and envi-
incorrect adjustment of an instrument or improper use of an ronmental errors according to their sources. They can also
instrument and any computational mistakes. Although gross be classified according to their behavior into dynamic and
errors may be difficult to eliminate entirely, they should be static errors.
anticipated and corrected. Some gross errors can be easily
detected but others are difficult to detect. • Instrumental errors are defined as imperfections of the
Gross errors cannot be treated mathematically. They can measuring instrument. They are inherent in measuring
be avoided only by taking care in the approach and by using instruments because of their structure. There are many
error detection routines. For example, errors can be mini- types of instrumental errors, which depend upon the
mized by taking more than one set of readings under the instrument. Instrumental error can be avoided by select-
same conditions, or by taking more than one set of readings ing a suitable instrument for the particular measurement
using different instruments or operators. application, calibrating the instrument against a stan-
In high-precision applications, some measurements need dard, adjusting the instrument before using it, and by
to be taken for the same quantity under the same conditions, applying the estimated factors of correction. Instrumen-
by using different operators. In some conditions, the error tal errors in a measuring instrument can be accounted
associated with an operator is included in the estimation of for, approximately, by replacing it with an instrument
the measurement error. of higher accuracy. In this case, the systematic error in
330 Error and Uncertainty
the instrument under test is the difference between the usually occur, but they become important in high precision
two results. measurements.
• Environmental errors are caused by conditions external The only way to compensate for random errors is by
to the measuring instrument. These conditions include taking a large number of readings and by using statistical
changes in temperature, pressure, humidity, interfer- methods to obtain the best approximation of the true value
ence, magnetic field, electrostatic field, and many more. of the measurand.
If it is possible to control the measuring environment In many cases, it is difficult to distinguish between
by providing air-conditioning, then calibration accuracy systematic errors and random errors. This can be seen
may be improved. Also, if certain components can be in the summary of the random and systematic errors
hermetically sealed, then errors due to adverse environ- in measurement and their sources, which is given in
mental effects may be reduced. When measurements Figure 1.
are conducted using appropriate shielding techniques,
then errors due to electromagnetic interference can be
totally removed. RELATED ARTICLES
• Static errors are caused by the limitation of the mea-
suring instrument or the physical law governing its Article 48, Common Sources of Errors in Measurement
behavior. Systems, Volume 1; Article 49, General Characteriza-
• Dynamic errors are caused by the inability of the instru- tion of Systematic and Stochastic Errors, Volume 1;
ment to respond fast enough to follow the changes in the Article 55, Uncertainty Determination, Volume 1.
measurand. However, although systematic errors always
exist in any measurement, they can be minimized to a
degree suitable to the aimed uncertainty of measure- REFERENCES
ment.
• The terminology for static systems is covered in Doebelin, E.O. (2003) Measurement Systems: Application and
Article 58, Description of Accuracy, Linearity, and Design, 5th edn, McGraw-Hill, New York.
Drift, Volume 1 and Article 57, Static Considera- Galyer, J.F.W. and Shotbolt, C.R. (1990) Metrology for Engi-
tions of General Instrumentation, Volume 1; dynamic neers, Cassell Publishers, London.
description is covered in Article 61, First-order Sys- Gregory, B.A. (1973) An Introduction to Electrical Instrumenta-
tem Dynamics, Volume 1; Article 59, Introduction tion, The MacMillan Press, London.
to the Dynamic Regime of Measurement Systems, Helfrick, A.D. and Cooper, W.D. (1990) Modern Electronic
Volume 1; Article 62, Second-order System Dynam- Instrumentation and Measurement Techniques, Prentice Hall,
ics, Volume 1; and Article 60, Zero-order System NJ.
Dynamics, Volume 1. Hofmann, D. (1982) Measurement Error, Probability, and Infor-
mation Theory, in Handbook of Measurement Science: Vol I
Theoretical Fundamentals (ed. P.H. Sydenham), John Wiley &
Sons, Chichester.
2.3 Random errors
McGhee, J., Henderson, I.A. and Sankowski, D. (1986) Functions
and Structures in Measurement Systems: A Systems Engineer-
Even when all systematic errors have been accounted for, a ing Context for Instrumentation. Measurement, 4(3), 11–119.
residual error will invariably occur. This residual error, due
McGhee, J., Henderson, I.A., Kulesza, W. and Korczynski, M.J.
to unspecified or unknown causes, is normally assumed to (1996) Scientific Metrology, ISBN 83-904299-9-3, A.C.G.M.
have a random nature. Random errors are unpredictable in LODART, Lodz.
that the next occurring single data value cannot be predicted Morris, A.S. (1997) Measurement and Calibration Requirements,
with certainty, whereas the behavior of the group can arise John Wiley & Sons, Chichester.
through irregular chances of random causes normally due Rangan, C.S., Sarma, G.R. and Mani, V.S.V. (1983) Instrumenta-
to fundamental fluctuations or extraterrestrial generation. tion Devices and Systems, Tata McGraw-Hill Publishing, New
McGhee, Henderson and Sankowski (1986) provide a Delhi.
comprehensive but incomplete list of starting point refer- Taylor, J.R. (1982) An Introduction to Error Analysis, University
ences. The random sources cause changes in the reading Science Books, CA.
for each repeated measurement under the same condi- Weinberg, G.M. (1975) An Introduction to General Systems Think-
tions. In well-designed experiments, few random errors ing, John Wiley & Sons, New York.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
Explanation of Key Error and Uncertainty
54:
Concepts and Terms
Luca P. Mari
Università Cattaneo, Castellanza (VA), Italy
definitions (in a sense: they are their definitions) so that
1 Measurement Results and Their Quality 331 they are always identical to themselves. Paradigmatically,
2 The Concept of Error 332 noise exists in the empirical realm, not in the symbolic
one; real numbers exist in the symbolic realm, not in the
3 The Concept of Uncertainty 333
empirical one.
4 Characterizing a Measurement with its
Whenever the two realms interact with each other, as
Uncertainty 333
measurement does by means of the mediation of quantities,
5 The Expression of Measurement Results and these diversities (a) require the introduction of a concept of
Their Uncertainty 334 quality of the symbols (in our case, measurement results)
6 The Procedure for Assigning the chosen as representatives for empirical states and (b) are
Measurement Uncertainty: an Example 335 the cause of several issues affecting such a quality.
Related Articles 335 The typical operative context of measurement that is
Further Reading 335 presented in Figure 1 shows that the required empirical
results (‘the output’) can be in principle obtained as the
transformation of the same empirical states (‘the input’)
by a direct manipulation (an ‘empirical procedure’) or
1 MEASUREMENT RESULTS AND THEIR
a transduction to information entities, to be processed
QUALITY and finally transduced back to the empirical realm (an
‘informational procedure’) shown in Figure 2.
Measurement is a peculiar means of acquiring and formally The benefits of informational procedures are commonly
expressing information about empirical systems. It is aimed, recognized (basically due to the fact that it is much easier
therefore, at setting up a bridge between the empirical to deal with symbols than with empirical things), but they
world and the linguistic/symbolic world, the domains of depend on the faithfulness of measurement results as rep-
the systems under measurement, and measurement results resentative entities for the corresponding empirical states.
respectively. Such faithfulness, and therefore the quality of measure-
A fundamental evidence is that these two realms exhibit ment results, is limited as a consequence of causes related
extremely different characteristics. Empirical systems are to the following:
embedded in the space–time universe, and this generates
their space and time dependency: any system is only par- • Model of the system under measurement: incomplete-
tially isolated from its environment and its dynamic forces ness, if not faults, in the definition of the measured
distinguishing between the system itself and its temporal quantity (the measurand), as in the case of an ill-
versions, that is, the system states. On the other hand, sym- characterized system dynamics or only a partial identi-
bolic entities such as numbers are coextensive with their fication of the quantities influencing the measurand.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
332 Error and Uncertainty
Empirical The repeatability of a measurement, and in more gen-
handling Transformed eral terms its quality, is therefore a relative characteristic
Empirical states
empirical states
to be evaluated in reference to the goals for which the
Measurement operation is performed and the available resources (in epis-
(sensing and Actuation
symbolic expression) temological terms, this can be thought of as a confirmation
that a concept of absolute, or complete, precision is simply
Measurement Transformed
results Information symbols meaningless).
processing It is amazing in this perspective to note that the indication
of the estimated quality of the results became customary in
Figure 1. The role of measurement in the relations between physical measurement only in the late nineteenth century,
empirical and symbolic realms.
and, however, several decades after the Theory of Error
provided by Gauss at the beginning of that century. A plau-
Empirical sible reason for this can be recognized in the commonly (in
procedure the past) assumed hypothesis that measurable quantities are
Transformed
Empirical states characterized by a perfectly precise ‘true value’. The choice
empirical states
of adopting the concept of error to model and formalize a
less-than-ideal quality of measurements originates from this
Informational
procedure hypothesis: any discrepancy between the measuring system
Measurement Transformed outputs and the measurand ‘true value’ should be taken into
results symbols account as an error, and correspondingly dealt with (see also
Article 49, General Characterization of Systematic and
Figure 2. The (possible) equivalence of empirical and informa- Stochastic Errors, Volume 1).
tional procedures. However,
• Operative accomplishment of the measurement proce- • an error can be recognized as such only if a correspond-
dure: poor repeatability, or stability, or selectivity of ing ‘right entity’ exists;
the adopted measuring system (see also Article 104, • errors can be corrected only if their corresponding ‘right
Models of the Measurement Process, Volume 2), if entities’ are known;
not faults in its usage. • ‘true values’, that play the role of such ‘right entities’
in the case of measurement, are in principle unknown
The unavoidable presence of such flaws is the reason (otherwise measurement itself would be useless. . .) and
requiring us to state any measurement result by expressing cannot be operatively determined.
in symbols a measurand value together with an estimation
These assertions imply that the Theory of Error is
of its deemed quality.
grounded on metaphysical, empirically inapplicable bases.
Consider the following two statements:
2 THE CONCEPT OF ERROR • ‘at the instant of the measurement, the system is in a
definite state’;
It is a well-known fact that the repeatability of measure- • ‘at the instant of the measurement the measurand has a
ments can be increased by definite value’.
• improving the measuring system in its empirical char- Traditionally they would be considered as synonymous,
acteristics, whereas their conceptual distinction is a fundamental fact
• reporting the results with a reduced number of signifi- of metrology: the former represents a basic assumption for
cant figures, measurement (we are not considering here measurement
in quantum mechanics), while the latter is epistemically
that is, by adjusting the sensing device or modifying the unsustainable and, however, operationally irrelevant. Mea-
symbolic expression respectively (see Figure 3). surement results are symbolic, and not empirical, entities:
Empirical Sensing and Transduced Symbolic Measurement
states presentation states expression results
Figure 3. Abstract schematization of a measurement.
Explanation of Key Error and Uncertainty Concepts and Terms 333
what in the measurement is determined, and therefore con- The recognition that even measurement, an operation tra-
sidered preexisting, is the system state, not the measurand ditionally deemed as the paradigm of objective information
value that is instead assigned on the basis of the instrument acquisition, requires the introduction of some subjective
reading and the calibration information. evaluation is of capital importance. Therefore, the shift from
‘error’ to ‘uncertainty’ is far more than a terminological
issue, and witnesses a conceptual transition from an onto-
3 THE CONCEPT OF UNCERTAINTY logical position to an epistemic one: according to the GUM
standpoint, to establish the quality of measurement results
The search for a more adequate framework reached a cru- is an issue related to the state of knowledge of the mea-
cial point about 30 years ago, when it was understood that a surer, and therefore ‘absolute quality’ cannot be reached
common approach for modeling and formally expressing a simply because some intrinsic uncertainty is always part of
standard parameter describing the quality of measurement the measurement system.
results was a condition for establishing strict cooperation The possible sources of uncertainty that the GUM itself
among the national calibration laboratories. To build up lists are exemplar in this regard: together with the ‘vari-
and maintain mutual confidence between accreditation bod- ations in repeated observations of the measurand under
ies and to build up and maintain compatibility for their apparently identical conditions’, the reason usually recog-
calibration certificates required having the quality of their nized for random variability, and some causes related to
measurement results evaluated and expressed according instrumental issues such as ‘approximations incorporated in
to some harmonized protocol. To this goal, the Interna- the measurement procedure’ and ‘instrument resolution or
tional Committee for Weights and Measures (CIPM), started discrimination threshold’, the GUM identifies several epis-
a project together with several international organizations temic sources, among them the incomplete definition of
involved in standardization (International Organization for the measurand and the imperfect realization of its defini-
Standardization (ISO), International Engineering Consor- tion (see also Article 48, Common Sources of Errors in
tium (IEC), International Organization of Legal Metrology Measurement Systems, Volume 1).
(OIML), . . .): its final result is the Guide to the Expression
of Uncertainty in Measurement (GUM), first published in
1993 and later introduced as a Standard by each of such 4 CHARACTERIZING A MEASUREMENT
organizations. While originally intended for calibration lab-
oratories, the GUM is presently to be considered as the basis WITH ITS UNCERTAINTY
for expressing the results of any measurement performed in
accordance with an international standard. To accomplish a measurement process, three distinct activ-
According to the GUM, the uncertainty of a measure- ities must be sequentially performed:
ment result is ‘a parameter, associated with the result of a
• Acquisition: by means of a sensing device, the measur-
measurement, that characterizes the dispersion of the val-
and is transduced to a quantity suitable for direct access
ues that could reasonably be attributed to the measurand’.
by the measurer (e.g. the angular position of a needle
Apart from this rather classical definition, the most impor-
tant innovation of the GUM stands in its recognition that with respect to a reference scale), possibly through the
the uncertainty of measurement results can be evaluated mediation of an ‘intermediate’ quantity (a typical role
according to two distinct and complementary methods: for electrical quantities) to drive processing and presen-
tation devices (see Figure 4).
• some uncertainties, designated as ‘of type A’, are com- • Evaluation: the access to the transduced quantity (i.e.,
puted as suitable statistics of experimental data, usually the instrument reading) concludes the empirical part of
obtained as repeated instrument readings; the operation; by gathering and processing the available
• some other uncertainties, designated as ‘of type B’, are information (the transduced quantity itself, together
instead estimated on the basis of the observer’s personal with everything that is known on the measurement
experience and the available a priori information, and system: the measurand definition and realization, the
therefore express a degree of belief on the possible instrument calibration diagram, the values of relevant
measurand values. influence quantities, etc.), the measurer evaluates the
Intermediate Directly accessible
Measurand Sensing Presentation
quantity quantity
Figure 4. Abstract schematization of the empirical component of a measurement.
334 Error and Uncertainty
measurand value and its uncertainty; this inferential • The measurand value is estimated as the mean value of
process is based on both objective and subjective the random variable; in the case of type A evaluations,
information. for which an experimental population X of n repeated
• Expression: the obtained information is expressed in reading data xi is available, it is computed as
symbolic form according to an agreed formalization.
1
n
It should be noted that the same information could m(X) = x (1)
be in principle expressed in different forms for different n i=1 i
needs, by adopting, typically, a statistical or a set-theoretical
formalization (or some generalization of the latter, as in the • The uncertainty of the measurand value is estimated as
case of representations based on fuzzy sets: we will not deal the standard deviation of the measurand value, being
with such generalizations here). Consider the traditional itself a random variable; this parameter is termed as the
indication, x ± y, that admits two distinct interpretations: GUM standard uncertainty and denoted u(m(X)); in
the case of type A evaluations it is computed as
• the measurand value is expressed as the scalar x, with
y as its estimated uncertainty;
• as the measurand value, the whole interval (x − y, x + 1 1 n
u(m(X)) = (x − m(X))2 (2)
y) is taken, whose half-width, y, expresses the qual- n n − 1 i=1 i
ity (sometimes called precision) of such a measure-
ment result.
Measurement results can then be reported, for example, as
Neither of them is the ‘right one’: they should be selected mS = 100,021 47(35) g, meaning that the evaluated mass
according to the specific application requirements. m of the system S (whose specification should include
The GUM adopts this approach, and while basing its the indication of the operative condition in which the
procedure on the first interpretation, recognizes that ‘in measurement has been performed) is 100,021 g with a
some commercial, industrial, and regulatory applications, standard uncertainty of 0,35 mg.
and when health and safety are concerned’, it is often The same couple of values (measurand value, standard
necessary to express the measurement results by means of uncertainty) is adopted to express measurement results
intervals of values. as intervals. To this goal, a coverage factor k (typically
Measurement results must be therefore assigned accord- in the range 2 to 3) is introduced, such that U (X) =
ing to the goals for which the measurement is performed; ku(m(X)), termed expanded uncertainty, is adopted as the
they are adequate (and not ‘true’) if they meet such goals. half-width of the interval representing the measurement
By suitably formalizing them, the measurer is able to result: [m(X) − U (X), m(X) + U (X)], commonly written
express the available information of both the measurand as m(X) ± U (X) (if the probability distribution of the
value and its estimated quality. Quoting the GUM again, random variable is known, this interval can be thought of
no method for evaluating the measurement uncertainty can as a confidence interval whose confidence level depends
be a ‘substitute for critical thinking, intellectual honesty, on k).
and professional skill’: indeed ‘the quality and utility of In the case of derived measurement, that is, when the
the uncertainty quoted for the result of a measurement ulti- measurand Y is a quantity depending on N input quantities
mately depends on the understanding, critical analysis, and Xi , i = 1, . . . , N :
integrity of those who contribute to the assignment of its
value’. Y = f (X1 , . . . , XN ) (3)
5 THE EXPRESSION OF MEASUREMENT and for each quantity Xi , the estimated value m(Xi ) and
RESULTS AND THEIR UNCERTAINTY uncertainty u(m(Xi )) are given, the issue arises of how to
obtain the corresponding values m(Y ) and u(m(Y )) for Y .
For both type A and type B evaluation methods, the GUM The measurand value m(Y ) is simply obtained by intro-
assumes that measurands (but the same holds for all the ducing the estimates m(Xi ) in the model function f :
quantities involved in the measurement system: influence
quantities, correction factors, properties of reference mate- m(Y ) = f (m(X1 ), . . . , m(XN )) (4)
rials, manufacturer or reference data, etc.) can be formalized
as random variables, and as such characterized by statisti- The uncertainty u(m(Y )) is instead evaluated by means of
cal parameters: the so-called law of propagation of uncertainty, that for
Explanation of Key Error and Uncertainty Concepts and Terms 335
statistically noncorrelated quantities is According to the GUM viewpoint, any measurand Y
should be actually evaluated by derived measurement, that
N
is, by firstly identifying its dependence on a set of ‘input
u (m(Y )) =
2
ci2 u2 (m(Xi )) (5) quantities’ Xi , such as influence quantities, calibration
i=1
parameters, correction factors, and so on. For each Xi ,
where the ‘sensitivity coefficients’ ci that define the extent the values m(Xi ) should be obtained by statistical or other
to which Y is influenced by variations of the input quantities methods (e.g. as part of instrument specifications), and for
Xi are computed as each of such m(Xi ), the corresponding standard uncertain-
ties u(m(Xi )) and covariances u(m(Xi ), m(Xj )) should be
∂f evaluated, again by either type A or type B procedures.
ci = evaluated at Xi = m(Xi ) (6) In the case in which the functional relation f is known in
∂Xi
its analytical form, the sensitivity coefficients ci can then be
In the general case of correlated input quantities (i.e. their computed; if, on the other hand, the complexity of the mea-
covariance u(m(Xi ), m(Xj )) = 0), (5) becomes surement system prevents the explicit formalization of (3),
each coefficient cj can be experimentally estimated by a
N suitable setup of the system in which Y is repeatedly mea-
u2 (m(Y )) = ci2 u2 (m(Xi )) sured while all the quantities Xi but Xj are kept constant.
i=1 When at least some u(m(Xi )), u(m(Xi ), m(Xj )), and ci
N−1
N are available, (6) can be computed to obtain an estimation
+2 ci cj u(m(Xi ), m(Xj )) (7) of the measurand uncertainty u(m(Y )), to be compared to
i=1 j =i+1 the specified target uncertainty.
in which the combined standard uncertainty of the mea-
surement result m(Y ) is computed on the basis of a RELATED ARTICLES
first-order Taylor series approximation of (3) (see also
Article 55, Uncertainty Determination, Volume 1; Arti- Article 10, Extensions of the Representational Theory of
cle 52, Error Models, Error Budgets and their Calcula- Measurement, Volume 1; Article 48, Common Sources
tion, Volume 1). of Errors in Measurement Systems, Volume 1; Arti-
cle 52, Error Models, Error Budgets and their Calcula-
tion, Volume 1; Article 55, Uncertainty Determination,
6 THE PROCEDURE FOR ASSIGNING Volume 1; Article 104, Models of the Measurement Pro-
THE MEASUREMENT UNCERTAINTY: cess, Volume 2.
AN EXAMPLE
FURTHER READING
Measurement uncertainty is a pragmatic parameter: its
value is not intrinsic to the measurand but is to be estab- CIPM, BIPM (1981) Procès-Verbaux des Séances du Comité
International des Poids et Mesures, 49, 8–9, 26, (in French);
lished in reference to the specific goals according to which Giacomo, P. (1982) News from the BIPM, Metrologia 18,
the measurement is performed. No ‘true uncertainty’ exists, 41–44.
and the preliminary step of a procedure aimed at assigning ISO (1993) Guide to the Expression of Uncertainty in Measure-
a value to the uncertainty of a measurand value is there- ment, International Organization for Standardization, Geneva.
fore to decide a target uncertainty, the maximum value of ISO (1993) International Vocabulary of Basic and General Terms
uncertainty compatible with the given goals. In any step in Metrology, 2nd edn, International Organization for Standard-
of the procedure, if the estimated value is reliably consid- ization, Geneva.
ered less than such a target uncertainty, then the procedure ISO (1999) Geometrical Product Specification – Part. 2: Guide
should be stopped with a positive result: the measurand to the Estimation of Uncertainty in Measurement in Calibration
can be evaluated with a satisfying uncertainty, and no fur- of Measuring Equipment and Product Verification, International
Organization for Standardization, Geneva.
ther resources are required to refine the procedure. On the
other hand, whenever the estimated uncertainty becomes Mari, L. and Zingales, G. (2000) Uncertainty in Measurement
Science, in Measurement Science – A Discussion (eds K. Karija
greater than the target uncertainty, the procedure must be and L. Finkelstein), IOS Press, Ohmsha, (pp. 135–147).
definitely stopped with a negative outcome, conveying the
Taylor, B.N. and Kuyatt, C.E. (1994) Guidelines for Evaluating
information that better measurements are required to meet and Expressing the Uncertainty of NIST Measurement Results,
the specified target uncertainty. NIST Technical Note 1297 (available on the NIST website).
55: Uncertainty Determination
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
Whenever numerical information is presented, it must be
1 General Comment on Uncertainty 336 made clear what unit is being used; this is obvious but
2 Representing Numbers and Uncertainties 336 occasionally forgotten. In most technical writing, the units
3 Example of Uncertainty in Determining employed belong to the SI system.
Temperature 339
4 Rounding Off Measurement Results 339
5 Example of Rounding off of Laboratory 2 REPRESENTING NUMBERS AND
Measurements 340 UNCERTAINTIES
6 Prose Methods of Results Presentation 341
7 Tabular Methods of Results Presentation 341 Before continuing it is necessary to introduce the manner
in which a number is represented. Suppose the number five
8 Graphical Methods of Results Presentation 341
thousand three-hundred and forty-two point 00. . . is to be
9 Use of Equations to Describe Results of
represented. Following the usual procedure using digits, the
Measurement 342
number is 5342.00. . . This is the correct representation.
10 Comparison of Methods for Presenting However, if the number is fifty thousand three-hundred and
Measurement Results 343 forty-two point 00. . . , then writing it as 50 342.00. . . is a
Related Articles 343 contravention of the relevant international standard. The
standard recommends that, for numbers equal to or greater
than ten thousand, the thousands should be separated from
1 GENERAL COMMENT ON the hundreds by a space and not by a comma. The correct
UNCERTAINTY representation is thus 50 342.00. . ..
For numbers greater than this, the digits should be
Numerical information, like most other kinds of infor- grouped in threes, corresponding to increases of 103 .
mation, can always be presented in ordinary prose or If the long-hand version of 2.35 × 106 is required, then
text. However, unless the amount of numerical information it should be written as 2 350 000. In a similar way, the
is small, this method of presentation is clumsy. Conse- fractional decimal part should be written by grouping the
quently, other methods of presenting technical information decimal digits in threes so that 2.354 × 10−1 should be
are often used. written as 0.2354, while 2.3546 × 10−1 should be written
In many cases, the corresponding values of two or in long-hand as 0.23546 (while the standard calls for these
more variables have to be presented. It is, therefore, often formats, common usage still will place a point or comma
more appropriate to use tables, graphs, or equations to as the separator).
show the relationships between variables. Before these When numerical information is quoted, the intention is to
methods are discussed in detail, a few points regarding achieve a higher degree of precision than would otherwise
units, uncertainties, and rounding off need to be settled. be possible. However, perfect precision cannot be achieved.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Uncertainty Determination 337
UA′ UB :- Full-range uncertainties UA
UAL′ UBL :- Lower half-range uncertainties Nominal values
UAL′ UBL :- Upper half-range uncertainties UAL UAU
UB
UBL UBU
0.6 0.7 0.8 0.9 1.0
Remanence of carbon steel
Figure 1. Graph comparing interval estimators and illustrating notation for uncertainties.
Precision is a statement about how exact a particular value Y Y
is. Almost every quoted numerical value has some degree 6 6
of uncertainty associated with it. 4 4
Quite often there is an interest in knowing just how
large the uncertainty is. In particular, it is often impossible 2 2
to make meaningful comparisons between numerical val- 0 0
ues unless their uncertainty is known. The combination of 9 10 11 12 X 9 10 11 12 X
the numerical value and the uncertainty is called an inter- (a) (b)
val estimator. Figure 2. Interval estimators: (a) in the X and Y directions and
Interval estimators may be illustrated as shown in Fig- (b) area of uncertainty of the estimators.
ure 1. The full-range uncertainty defines the band within
which the true value of the measurement is most likely to
be located, while the half-range uncertainties, sometimes in the X direction from 9 to 11, and in the Y direc-
also called tolerances, define the upper bound and lower tion from 2.5 to 3.5. Hence, interval estimators are also
bound of this range. known as uncertainty bars as the example in Figure 2(a)
Suppose, for example, that the measured remanence of illustrates. This style of representation leads to draw-
carbon steel is reported as 0.86 ± 0.05 T. Here, 0.86 T is ing the point (10, 3) with the given uncertainty rectan-
the measured value, and ±0.05 T is the uncertainty in gle. However, it is acceptable to say that the ‘correct’
that value. This style of presentation indicates that the point is anywhere within the shaded rectangle shown in
remanence is believed to lie close to 0.86 T and that any Figure 2(b).
value between 0.81 and 0.91 T is also acceptable as being To appreciate how the interpretation of a graph may be
in agreement with the reported result. aided by the inclusion of interval estimators, consider firstly
If a second report specifies a measured value of 0.80 T ± the points plotted in Figure 3(a).
10% for the same quantity, this means that values in the Is there a ‘kink’ in the graph at about X = 10, or does
range 0.72 to 0.88 T are acceptable. Values in the range 0.81 the suggestion of a ‘kink’ result merely from random errors
to 0.88 T are acceptable to both writers. It can, therefore, in the measured values?
be seen that the two measured values do not differ signif- On the information given in Figure 3(a), it is not possible
icantly. Although such a graphical representation is not of to answer this question. Figure 3(b) and (c) shows the
much value where only two values are being compared, it same plotted points as does Figure 3(a), but with interval
can be useful where three or more values are involved. estimators added. In Figure 3(b), the uncertainties are small,
Obviously, any comparison of values depends on some and the reader accepts that there is a kink in the graph
knowledge of the uncertainty in both, or all, of the values occurring at about the value X = 10. This ‘kink’ probably
being compared. In the above example, if only the two corresponds to a real physical effect.
measured values of 0.86 and 0.80 T were reported, then it In Figure 3(c), where the uncertainties are much larger,
is not possible to judge whether the difference is significant it is clear that the information presented does not clearly
or not. Specification of only these values is widely known establish whether any such ‘kink’ exists.
as giving point estimators. In practice, it is usually not necessary to show uncertainty
Where numerical information is presented in ordinary bars on every point on a curve. To avoid unnecessary
prose, in text, in a table, or by means of an equation, clutter, they can be shown on only two or three points,
the statement of uncertainties is usually straightforward. In which are typically one near each end of the curve, with
graphs, the method of representation shown in Figure 2 perhaps one about the center as well. If the uncertainties are
extended to two dimensions is used. For example, sup- small, they will be invisible on the graph. In such cases,
pose that X = 10 ± 1 and Y = 3.0 ± 0.5. Bars are drawn their values should be stated elsewhere.
338 Error and Uncertainty
Y
8
0
2 4 6 8 10 12 X
(a) Point estimators
Y
8
0
2 4 6 8 10 12 X
(b) Small interval estimators
Y
8
0
2 4 6 8 10 12 X
(c) Large interval estimators
Figure 3. Graphical kink with different types of estimators: (a) point estimators, (b) small interval estimators, and (c) large interval
estimators.
The above approach to the subject of uncertainties has To develop a more realistic approach to the subj-
been oversimplified. According to it, if a value is quoted as ect of uncertainties, it is necessary to use statistical
3.5 ± 0.5, this means that a value of 4.01 will be interpreted concepts.
as being in total disagreement with the result. This is clearly The ability to estimate the uncertainty that exists in a
an unrealistic view. measured value is an important part of the skill required
Uncertainty Determination 339
by experimenters. It often involves a blend of science part of the upper half-range uncertainty of measurement
and intuition. (iv) are also included in the lower half-range uncertainty
The theory of probability and combinations provides the of (ii).
foundation upon which the calculation and treatment of The state of affairs illustrated by this example is far from
errors is based. satisfactory from the measurement point of view. Steps
should be taken to see if paying closer attention to the
measuring method could reduce the differences between
3 EXAMPLE OF UNCERTAINTY IN the measured nominal values and the uncertainties, the
DETERMINING TEMPERATURE instrumentation used, and the treatment of the measuring
data from which the estimators are obtained.
Four different experimenters measure the same temperature,
and report their results as (i) 231.0 ± 2.5 ◦ C (ii) 245 ◦ C ±
5% (iii) 227.2 ± 5.2 ◦ C (iv) 225.0 ◦ C ± 5% 4 ROUNDING OFF MEASUREMENT
Draw a diagram that will allow these results to be com- RESULTS
pared easily. Comment upon the results. Do they agree?
When a numerical value is quoted, a choice as to the
number of digits to be used will need to be made, that is, it
3.1 Solution
is necessary to decide whether to round-off the number by
discarding some of the digits at its right-hand end. Where
Before sketching the interval estimators it is necessary to
the first digit to be discarded is 5 or higher, the last digit
convert the percentage tolerances to actual values of the
retained is normally rounded up. Thus, 4.83500 rounded off
measured temperature. Thus, for (ii), 5% of 245 equals
to three significant figures becomes 4.84.
12.25 and 5% of 225 equals 11.25. The interval estimators
Suppose a numerical value is quoted as being 5.12. Usu-
may now be drawn as shown in Figure 4.
ally, this does not mean 5.12 precisely; it means any number
Examination of Figure 4 shows that the measurements
which, when rounded off to three significant digits, becomes
agree due to the overlapping range of uncertainty indicated
5.12. Thus, it means any number between 5.1150000. . . and
by the shading. This conclusion is justified for the follow-
5.1249999. . ., that is, it practically implies 5.12 ± 0.005.
ing reasons.
Thus, rounding-off to the second figure after the decimal
Measurements (i) and (iii) have nominal values lying
point has introduced an implied uncertainty of ±5 in the
within the tolerance band for that of (iv). Hence, mea-
digit immediately following the last digit quoted.
surements (i), (ii), and (iv) exhibit good agreement. The
A number that is correctly rounded off fulfills two
nominal value of measurement (ii) lies outside the tolerance
requirements. Firstly, the implied uncertainty is much
bands of (i), (iii), and (iv). However, as its lower half-
smaller than the actual uncertainty, so that no meaningful
range uncertainty totally includes the full-range uncertainty
information is lost in the process of rounding-off. Secondly,
of measurement (i), then measurements (i) and (ii) may be
the implied uncertainty is not so small that the figures
taken to be in agreement. Because of this and in the inter-
quoted suggest a higher degree of precision than is justified.
ests of noncontradictory and consistent analysis, it must
In cases in which the uncertainty is not calculated, a
be concluded that all four measurements agree within the
rough idea of how large the uncertainty might be is usually
limits of the tolerances given.
available. In rounding off, it is important to ensure that
Further justification of this opinion can be obtained by
both these requirements are met. Where the uncertainty is
examining the relationship between the interval estimators
calculated, it is possible to be more specific.
for measurements (ii), (iii), and (iv). The nominal value
Suppose, for example, that it is required to quote the
and upper half-range uncertainty of measurement (iii) and
circumference of a circle, having measured its diameter
as 20 mm with uncertainty of 1/30 mm in either direction.
(i) Range of overlapping uncertainty The circumference is then π(20 ± 1/30) mm and may be
(ii)
evaluated as (62.831853 ± 0.10471976) mm.
(iii) Looking at the uncertainty term, it can be seen that all
(iv) Temperature, °C digits before the decimal point are zero, while there is a
1 in the first place after the point. In the nominal value
210 220 230 240 250
(62.83. . .), therefore, the first two digits are not directly
Figure 4. Interval estimators used to compare different measures subject to any uncertainty, while the third digit, which is the
of the same quantity. 8, in the first place after the decimal point, could reasonably
340 Error and Uncertainty
go either up or down by 1. With this uncertainty in the third Table 1. Reported measured nominal values and uncertainties of
digit, if we wish to ensure that the implied uncertainty is six laboratories.
much smaller than the actual uncertainty, all digits beyond A B C D E F
the fourth should certainly be discarded. On this basis, it is
Nominal 18.503 20.5 19.0 21.0 19.5 20.0
reasonable to round off the value of the circumference to Uncertainty 0.374 0.205 0.571 0.525 0.585 1.20
(62.83 ± 0.10) mm.
This example suggests a useful working rule for rounding
off, which can be adopted, as follows: method and comment upon the result of each of the mea-
suring methods relative to that of the standards laboratory.
4.1 Rule
5.1 Solution
In rounding off, include one digit beyond the first uncertain
digit, both in the nominal value and in the uncertainty. Applying the rounding-off rule to the reference measured
The two exceptions to the rounding-off rule are as fol- value gives a rounded-up value of 19.51. In a similar
lows: way, the rounded-off uncertainty is 0.85. The corrected
statement of the reference value is thus 19.51 ± 0.85, which
• Where a whole number ends in a zero. For exam- is illustrated in Figure 5.
ple, 105 000 does not necessarily imply 105 000 ± 0.5. The correctly rounded-off values for the results quoted
To make the implied uncertainty clear, this number in Table 1 are given in Table 2.
could be written as 105 × 103 , 105.00 × 103 . . . as The interval estimators may now be drawn as shown in
appropriate. Figure 6.
• Where the number quoted has no uncertainty such as in Although the measurements conducted by laboratories A,
the statement that 1 inch = 2.54 cm, the number 2.54 is B, C, E, and F agree with the reference value, measure D
absolutely accurate by definition, or when the number
quoted can take on only well-defined specific values,
Table 2. Correctly rounded-off nominal values and uncertainties
such as whole numbers. For example, the valency of
of six laboratories.
oxygen is 2, or the number of cylinders in a certain car
engine is 4. A B C D E F
Nominal 18.50 20.50 19.00 21.00 19.50 20.0
Uncertainty 0.37 0.21 0.57 0.53 0.59 1.2
5 EXAMPLE OF ROUNDING OFF OF
LABORATORY MEASUREMENTS Reference quantity
A B
A standards laboratory measured the nominal reference
value, xr , of a quantity as 19.5081 and the reference value C
D
uncertainty, ur , as ±0.8506. Round these values off and E
sketch the interval estimator representing it. F
When the six laboratories, A, B, C, D, E, and F, use dif- Quantity
ferent measuring techniques to measure the same quantity,
18 19 20 21 22
the nominal value and the uncertainty obtained by each
method are as given in Table 1. Correct any mistakes in Figure 6. Interval estimators for the nominal values of A to F
the rounding off, sketch the uncertainty intervals for each and their uncertainties.
Nominal value Uncertainty
19.5081 ± 0.8506
First uncertain digit in the nominal value First uncertain digit in the uncertainty
Figure 5. Illustration of rounding-off numerical values.
Uncertainty Determination 341
Table 3. Comments upon comparison of rounded-off values for the measurements
of six laboratories.
Values Comment
Reference 19.50 ± 0.85 tolerance band is ±5% of nominal value
A 18.50 ± 0.37 nominal value outside tolerance band of
reference; equipment requires recalibration
B 20.50 ± 0.21 nominal value outside tolerance band of
reference; equipment requires recalibration;
check measuring method
C 19.00 ± 0.57 nominal value inside tolerance band of
reference; acceptable but check equipment
and measuring method
D 21.00 ± 0.53 nominal value outside tolerance band of
reference; equipment requires recalibration;
check measuring method
E 19.50 ± 0.59 good consistency with the reference; measuring
method is acceptable
F 20.0 ± 1.2 nominal value within tolerance band of the
reference; check method to see if errors may
be reduced
does not. The situation is not at all satisfactory. Table 3 7 TABULAR METHODS OF RESULTS
contains appropriate comments. PRESENTATION
6 PROSE METHODS OF RESULTS Tables can be useful where corresponding values of two
or more variables are to be presented as in the example in
PRESENTATION Section 5.
A table is preferable to a graph where:
As has been mentioned above, the prose method of pre-
sentation is unattractive where a large amount of numerical • the amount of information presented is small; there is
information is involved. It can be useful for small amounts, little to be said for a graph that shows only two or
as for example: three points;
• the information is to be presented with a higher degree
The sample tested showed a resistivity of 12.42 m ±2% of precision than can be achieved in reading values from
when tested with an applied voltage of 100.0 V ±1% at a
a graph.
temperature of (20 ± 0.5) ◦ C.
Compared with a graph, a table has the following dis-
This example is not quite in accord with recommendations advantages:
given above regarding uncertainties and rounding off. How-
ever, even partial adherence to these recommendations can • the form of the relationship between the variables is not
make the writing seem rather pedantic. Consequently, there immediately apparent;
is a temptation to be less precise. Alternatively, a rear- • interpolation between the values given is usually not
rangement of the information given in the above example easy.
is possible:
The sample tested gave the following result: 8 GRAPHICAL METHODS OF RESULTS
Applied voltage 100.0 V ±1%
Temperature (20 ± 0.5) ◦ C PRESENTATION
Measured resistivity 12.42 m ±2%.
A brief examination of almost any manufacturer’s data
Obviously, this kind of list can be extended to include sheet or any scientific or technical specialist journal will
quite a large amount of numerical information without show that graphs are more widely used than any other
becoming as objectionable as a long prose statement con- method for presenting a set of numerical information.
taining a lot of numbers. Such a list in effect constitutes a A graph allows a fairly quick and transparent impression
one-dimensional table. of the kind of relationship that exists between two variables.
342 Error and Uncertainty
Two or more graphs, plotted on the same axes, allow rapid Two points in this example are worth emphasizing.
comparisons to be made. Interpolation between the plotted Firstly, the statement indicates the range over which the
values is usually fairly easy. Unfortunately, it is usually not equation is valid. Secondly, it indicates how closely the
possible to read values from a graph with a high degree of measured values lie to the values given by the equation.
precision. Appropriate choice of scale and section of range Without these two pieces of information, the communi-
can make it more precise. cation of the ideas would not be so effective. It is also
Where there are no limitations in the space allowed, necessary to exercise care by also giving information on
the ideal arrangement is often to present the numerical the uncertainties in the measured values of Bm and P .
information in both a table and a graph. There are at least three different situations that can
give rise to a statement of the kind illustrated in the
above example.
9 USE OF EQUATIONS TO DESCRIBE In some cases, the equation will have been derived
RESULTS OF MEASUREMENT from theoretical considerations. The statement will then
constitute a comparison between theory and measurement.
When numerical information is obtained as a result of a In other cases, when theory may indicate the form of the
theoretical analysis, the information naturally makes its first equation, the parameters will have to be evaluated from the
appearance in the form of equations. measured results.
In the case of numerical information obtained by mea- In yet another case, the measured values will be used to
surement, it is also sometimes appropriate to express the indicate the form of the equation as well as the values of
information by an equation. This can lead to a statement the parameters. In this last case, the equation is referred to
like the following: as an empirical equation as it has no theoretical basis.
Graphical methods are useful if it is necessary to check
For the ten cores tested, the measured power loss P (watts) or find the form of an equation that measured results fit, and
lay within ±5% of the value given by P = 120Bm2.7 where in determining the parameters of the equation. A discussion
Bm is the peak operating flux density in Teslas. This of this process is called curve fitting. Where there are no
equation is valid over the range 0.1 T ≤ Bm ≤ 1.7 T limitations in the space allowed, the ideal arrangement is
often to present the numerical information in both a table
This kind of statement can summarize a lot of information and a graph, as also stated above.
in a small space.
Table 4. A summary of methods for information presentation.
Method Advantages Disadvantages
Ordinary prose or text 1. Useful for small amounts of information 1. Unattractive where the amount of
numerical information is large
2. Writing can seem rather pedantic
3. Temptation to be less precise
Tables 1. Can include a large amount of numerical 1. Interpolation between given values is not
information so easy
2. Less objectionable than a long prose
statement
3. Preferable to a graph if the amount of
information is small
4. Higher degree of precision than in reading
from a graph
Graphs 1. Easy interpolation between values 1. Usually not possible to read values from a
2. Kind of relationship usually transparent graph with a high degree of precision
3. Rapid comparisons of two or more graphs
possible
Equations 1. Result of a theoretical analysis 1. Must be supported by experimental
2. Appropriate way to express the measurement
information
3. Information density is high
4. Possibility to compare theory and
experiment
Uncertainty Determination 343
10 COMPARISON OF METHODS FOR RELATED ARTICLES
PRESENTING MEASUREMENT
Article 48, Common Sources of Errors in Measurement
RESULTS Systems, Volume 1; Article 49, General Characteriza-
tion of Systematic and Stochastic Errors, Volume 1.
A summary of the advantages and disadvantages of the
above methods of information presentation is given in
Table 4.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
56: Transfer Characteristics of Instrument Stages
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
converted quantity, such as where a sensor converts relative
1 Representing the Overall Response of a humidity into a voltage and angular motion equivalents for
System 349 a visual indicator.
2 Tachometer Example of System Description 350 The ratio of output to input quantities is termed the trans-
3 DC and Dynamic Characteristics Compared 351 fer characteristic. The magnitude of, and phase between,
4 Use of Simplifications for System the two quantities are expressed as numerical quantities in
Performance Description 352 terms of the units of the two variables involved. For exam-
5 Performance Variation with Signal Parameter ple, the sensitivity of a humidity sensor might be designed
Values 352 to be 0.1 mV per 1% RH.
When the system is investigated further, it is seen that
6 Transfer Function Example for Printed
this transfer expression (the usual one quoted by man-
Armature Motor 353
ufacturers) is a great simplification of the existing full
Related Articles 353
characteristic; it provides no information about the sensor’s
References 353 response to dynamic changes of relative humidity. What is
missing is some form of statement as to how the system
will respond to changing input signals. This can be done by
1 REPRESENTING THE OVERALL adding a frequency-dependent component to the expression
RESPONSE OF A SYSTEM already given to represent the static behavior.
As simplification, to cover the static operational region
To begin explanation of how to specify measuring systems, of use only, suffices for many tasks, it is common practice
it is first necessary to understand the role of the various to quote only the static transfer characteristic. The steady
stages of an instrument system and how they change the state equivalent is given in the case of a carrier system or
energy variables from the input form into the required out- a purposefully designed system made to operate at a given
put. This article deals with the transfer characteristic. With frequency. The steady state is also sometimes called the
that foundation in place, Article 57, Static Considerations quasi-static condition.
of General Instrumentation, Volume 1 and Article 58, In practice, the state-variable form of relationship bet-
Description of Accuracy, Linearity, and Drift, Volume 1 ween the input and the output for a stage often cannot be
describes the static regime with Article 59, Introduction easily arranged in the form of independent input and output
to the Dynamic Regime of Measurement Systems, Vol- signals. For this and other reasons, it is common practice to
ume 1 then introducing the more complex dynamic regime. convert that relationship into its Laplace transform equiva-
The stages used to form a measurement system can be lent, which can then be arranged to produce a transformed
represented as blocks of hardware that effectively trans- transfer characteristic called the transfer function.
fer an input to its output either in the same energy As an example, the relationship between an input variable
form, such as the voltage ratio of an amplifier, or as a x and an output variable y existing through a measurement
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
350 Measuring System Behavior
instrument stage might be Being a driven rather than driving mechanical device,
its mechanical inertia and friction are not of importance
d2 y 3 dy dx in obtaining the transfer function between angular velocity
2
+ + 2y = x + (1)
dt dt dt input θ and the generator voltage output vo . It is designed
such that the voltage generated is proportional to the shaft
from which we require the ratio y/x to be formed to
speed according to vo = Kg dθ /dt; Kg being the generator
obtain the transfer characteristic. It will be noticed that
or transducer constant.
measurement system behavior is generally described in
The mechanical parts of the generator are forced (an
terms of linear differential equations, this being in line with
assumption) to follow input demands and cannot cause
general physical systems description.
lag or lead action because of the generator’s compliance
As the system is linear in the mathematical sense, the
or angular mass. The electrical output circuit, however,
Laplace transform can be taken obtaining
contains energy storage in the form of inductance of the
s 2 Y (s) + 3sY (s) + 2Y (s) = X(s) + sY (s) (2) windings, and this can influence rapidly changing signals.
Energy is dissipated via the two resistances, one of which
This can be rearranged to give is the load resistance of the next stage that it drives; let
R = RL + Ra .
Y (s) s+1 The output voltage vo , that generated across RL , is, in
= 2 = G(s), the transfer function (3)
X(s) s + 3s + 2 Laplace form,
RL Kg sθ
If the process involves a time delay, then terms of e−sτ υo = (5)
form will be included, for example, R + sLa
Y (s) e−sτ which can be rearranged to yield the transfer function:
= (4)
X(s) s+1
υo RL Kg s
= (6)
It is in the Laplace form that transfer characteristics are θ La s + R/La
most usually expressed as they are immediately in a form
that can be operated upon or studied directly. Having developed the expression in this form, for this
It is generally permissible to use the operator D in place example, we can move on to explore the three regions of
of s, and some authors use p instead of s. When the system transfer characteristics that could be involved when describ-
is operating with sinusoidal signals (which can usually be ing the transfer characteristics of a stage in a measuring
the case because complex signals can be broken down into instrument or of the whole instrument system.
sinusoidal components), it is permissible to replace s by jω, It can be seen, from the above example, that when s
enabling the response at any chosen angular frequency ω (that is, jω for sinusoidal excitation) is zero, the system is
to be calculated. operating with zero frequency, that is, at the DC point, in
the electrical circuits. The generator is turning to produce
an output related to the generator constant Kg , modified by
2 TACHOMETER EXAMPLE OF SYSTEM
the loading divider effect of Ra and RL .
DESCRIPTION In this situation, all frequency-dependent components
have no effect on the output. Transients that can arise in the
As an example, the tachogenerator, a device that generates electrical circuit when an input mechanical or an output load
output voltage proportional to angular shaft velocity and, transient occur have, in fact, settled down: the generator
therefore, maps shaft speed into an equivalent voltage is producing its steady state output. Thus, the generator
representation, has the schematic electrical circuit shown constant Kg is, in fact, the response existing after internal
in Figure 1. transients have settled to be insignificant. It represents the
steady state transfer characteristic, and this is generally the
Ra L sensitivity figure quoted.
The generator will, however, not produce the simple
steady state transfer relationship during dynamic changes
Eg RL vo
to its state, that is, during transients. It may be neces-
sary, in application, to make allowances for the transient
characteristics, which are more complex. This certainly is
Figure 1. Electrical schematic circuit of tachogenerator. the case if rapid-response, closed-loop operation with the
Transfer Characteristics of Instrument Stages 351
tachogenerator is sought. Phase-shift and gain changes dur- for very complex systems can often take many years of
ing transients can cause instability. effort. Systems identification procedures attempt to obtain
If the component is to be used in applications where them from practical testing.
transient response is important, then it will sometimes be Further explanation about transfer functions, in general,
found that the manufacturer will quote the necessary addi- is presented in the classic work of DiStefano, Stubberud
tional component values (resistance and inductance in this and Williams (1976) and Doebelin (1975). Atkinson (1972)
case) so that the user can construct the full transfer func- explains the difference between steady state and transient
tion in order to obtain the transient behavior characteristics characteristics in terms of the solution of linear differential
needed to condition a tight, stable system. equations having constant coefficients.
It is, however, often necessary for the user to conduct Having shown how a stage can be modeled theoretically
tests to obtain the additional values, the only parameter in terms of its transfer function, and therefore be uniquely
quoted being the steady state characteristic. For example, described on paper or in a computer model, it is important
the data sheets for the previously mentioned humidity next to realize that linear systems of interconnected transfer
sensor contain no information about transient behavior, it functions, no matter how complicated, can (in theory) be
being unnecessary for the market intended. reduced to a single-stage transfer function. This means
that a complex measurement instrument system, built from
many stages as a serial path including feedback loops
3 DC AND DYNAMIC with other paths, can be reduced mathematically to just
CHARACTERISTICS COMPARED one transfer function that describes its complete overall
static and dynamic behavior. This reduction allows the
There exists another description that can cause confusion transient, steady state, and DC responses of any system
for this too is often quoted as though the component only to be calculated or simulated without need to assemble
possesses a characteristic mode of behavior: this is the all stages.
DC response. As the name suggests, this is the response There is no need here to elaborate on the techniques used
characteristic of the stage when the input, measurement- in block-diagram algebra; they are well covered in DiSte-
bearing signal is at zero frequency – direct current or direct- fano, Stubberud and Williams (1976). It basically stems
coupled would appear to have formed the symbol DC. Not from the prime fact that transfer functions are commuta-
all systems possess a finite value for DC characteristics, nor tive, provided it can be assumed that they do not load each
is it always relevant. other when connected. Interaction of stages destroys the
As an example, consider the DC electronic amplifier. rule, making reduction more complicated. Interaction sig-
This has a certain gain value when operated at zero fre- nificantly changes the response, and so it may be necessary
quency. What is often not realized is that it has a transient to build a slightly more complex system that incorpo-
response during which gain varies and it will exhibit dif- rates buffering between blocks. This being done allows
fering gain with signal frequency. When used as an AC the system performance to be accurately modeled and
amplifier at a given frequency, it then has a certain steady thus optimized.
state gain. Thus, use of the apparently simple amplifier When compounding stages together to build up a mea-
block must make allowances for the various modes of trans- surement system, it is, therefore, necessary to consider if
fer characteristic applications. Furthermore, the input signal the stages interact to a significant degree. In electronic
level can often also alter the transfer characteristic. systems, this problem has generally been eliminated by
It has been shown, by example, that stages used in a designing integrated circuits to connect on the basis of low
measurement system can possess three distinctly different output impedance being connected to high input impedance
transfer characteristic functions depending upon the time in order to preserve voltage levels. In cases where power
duration and form of the signal. Which are important must must be transferred efficiently, it is necessary to use stage
be decided for each application. The easiest to measure impedances that are equal. For current transfer, another
and quote are DC gain and/or the steady state value. The information transfer mode, the opposite to the first given
transient characteristic is much more complex to assess voltage case, applies.
and express and it also varies with signal conditions. Any Electronic systems connections generally present few
sensing system usually simply breaks down, by analysis, connection problems (provided well-established rules are
to yield its DC characteristics, but realization of complete adhered to), but the same is not true of mechanical, thermal,
dynamic transfer functions that would enable the transient fluid, and optical stages, for in those disciplines, component
response to be studied would require expenditure of consid- design does not enable ideal matching arrangements to be
erably more effort. The development of transfer functions performed so easily.
352 Measuring System Behavior
Transfer characteristics of a stage can be stated in terms are used to map a many-variable parameter situation into
of the response in either the time or the frequency domain. a single one. A moving-coil loudspeaker (as an actuator or
Conversion from one domain to the other is achieved via the a sensor) mechanism is generally characterized by stating
Fourier transformation. When the signal is of a steady state, the impedance it presents at a given audio frequency, this
time-domain condition, because it is periodic, it can easily being expressed as a pure resistance of x. In reality, the
be transformed to yield the singular set of frequency compo- terminal impedance is a complex electrical quantity having
nents. Signals of transient nature present greater difficulties in both real and imaginary parts, and it varies greatly with
transformation for the spectrum changes during the transient. frequency. This applies to the input and output impedances
It is, therefore, quite reasonable for the response of a of many stage blocks used. This is, however, not always
stage to be stated in terms of the frequency domain. The the case, a notable exception being the transmission line,
same information could be expressed in the time domain which provides constant purely resistive impedance at any
in terms of the stage’s response to a suitable time variant length, provided it works within certain assumptions of
input. In some applications, it is more convenient to obtain being lossless and properly terminated. Somewhat confus-
or describe a transient in time response form (such as to a ingly, the term impedance is often used synonymously with
step or impulse input) than it is in the frequency domain. resistance: correctly, the former is the resistance to flow
Zero-frequency response is, in the frequency domain, when both real and imaginary components exist as a com-
the value of amplitude, or phase, at zero frequency. As plex variable.
frequency response plots are generally expressed using
logarithmic scales, zero frequency cannot actually be rep-
resented. Amplitudes are generally expressed in decibel 5 PERFORMANCE VARIATION WITH
form, but for some instrumentation, unconverted ratios may SIGNAL PARAMETER VALUES
be preferable.
The above remarks apply to stages having at least one The tacit assumption often exists that components and sub-
each of a clearly identifiable two-terminal input and out- system blocks are what they are defined to be for all regions
put port, refer to Article 108, Transducer Fundamentals, of operation. In practice, a component can behave quite
Volume 2. In use, a two-terminal device, such as a capaci- differently for signal frequencies, levels, noise contents,
tor or a spring, becomes a two-port device, either as a series and other parameters from those for which it was purpose-
(through-variable) or as a parallel (across-variable) arrange- fully made to handle. For example, a capacitor becomes an
ment, by suitable connection with common lines. The above inductor with rising frequency; a mechanical mass provides
remarks then apply, provided it is clearly stated what the properties of spring compliance under certain conditions.
connection arrangements and the terminating conditions Ideally, all components and stages used in measurement
existing are. Alternatively, the two-terminal device can systems should be defined to provide the user with infor-
be described in terms of its terminal impedance response, mation about their whole spectrum of response conditions,
called the driving-point impedance (or admittance), which but this is not practicable. The instrument designer, in par-
can also be fully expressed mathematically in Laplace form. ticular, must be aware that simplistic definitions, such as
Simple common components, such as the electrical those giving no more than steady state or DC information,
inductor or mechanical mass, are so well known that they may not suffice. Lack of observance and understanding of
usually do not need to be specified (for response reasons) in this defect in subsystem description is a prime reason for
greater detail than the numerical constant of the parameter failure of many designs when first assembled and for incor-
they provide, for example, inductance in henry units or mass rect operation when fed with input and influence quantity
in kilogram units. This information is sufficient for users signals that do not conform to those it is designed for.
to construct the response characteristics required. When It is, therefore in principle, possible to model any com-
the component is uncommon, a mathematical driving-point plete or part measuring system as a transfer function
impedance expression is needed to provide adequate infor- expression that enables its behavior to be understood with-
mation about its static and dynamic behavior. out needing to resort to actual hardware. Whether this is
worthwhile, however, depends upon the application and
the extent of the system. A very practical point is that
4 USE OF SIMPLIFICATIONS FOR many of the constants required to provide a boundary to
SYSTEM PERFORMANCE the mathematical model must come from measurement of
DESCRIPTION existing phenomena.
Reduction, in this way, of an arrangement of many sub-
In many cases, it is not convenient to express the true systems’ mathematical models into one, retains all informa-
response behavior of a terminal port, and approximations tion about the overall behavior but does not allow interstage
Transfer Characteristics of Instrument Stages 353
signals to be studied. This realization is not to be con- Printed motor
fused with identification processes and simulations that ω 1
can also provide overall input–output relationships that G(s) = out (s) = C1
υin τm s + 1
apply for some modes of operation and not others and (7)
Field controlled motor
that may be realized by quite different internal processes. 1
If the internal subsystem transfer functions are adequate G(s) = C2
(τm s + 1)(τf s + 1)
models of the stage that they represent, so also will be
the overall reduction obtained by block-diagram algebra where
(provided conditions of connection and range of operation ωout = shaft velocity
are met). υin = drive voltage
Transfer function representation is only as good as the τm = mechanical time constant
assumptions on which it is developed. Many are, and τf = electrical time constant of field winding
need to be, simplifications of the real situation; often, C1 , C2 = constants related to generator and torque
they ignore the higher-frequency terms that can be con- constants and to mechanical friction and
sidered to be of no practical consequence to response, field-coil resistance.
thereby making them only applicable for a given range of
frequencies.
RELATED ARTICLES
Article 25, Introduction to Signals in Physical Systems,
6 TRANSFER FUNCTION EXAMPLE FOR Volume 1; Article 26, Signal Classification, Volume 1;
PRINTED ARMATURE MOTOR Article 27, Signals in the Frequency Domain, Volume 1;
Article 59, Introduction to the Dynamic Regime of Mea-
To illustrate the above ideas, consider the transfer func- surement Systems, Volume 1.
tion of a high-performance instrument printed-armature DC
motor. This needs only to allow for response lag caused
by the mechanical storage aspects of the armature disk, REFERENCES
the electrical storage time constants of the armature cir-
Atkinson, P. (1972) Feedback Control Theory for Engineers,
cuit being much smaller than those of mechanical sources Heineman, London.
(50 µs compared with 30 ms). A conventional style, field-
DiStefano, J.J., III, Stubberud, A.R. and Williams, I.J. (1976)
controlled, DC motor, however, requires a more complex Feedback and Control Systems: Schaum’s Outline Series,
transfer function, one having a second storage term to allow McGraw-Hill, New York.
for the delay caused by significant storage effects in the Doebelin, E.O. (1975) Measurement Systems: Application and
electrical field-coil circuit. The two transfer functions are Design, McGraw-Hill, New York.
57: Static Considerations of General Instrumentation
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
An overall transfer sensitivity (S1 and S2 ) can be devel-
1 Description of the Static or Steady State oped through each path by multiplication of the static
Behavior 354 transfer characteristic for each of the stages connected to
2 Range Factors 354 form a link. This yields
3 Zero Values of Operation 355
Vout L ξ
4 Drift 355 S1 = = a b
%RH %RH L
5 Hysteresis and Dead-band 356
R V V
6 Signal-to-noise Ratio Entities 357 × c d e out
ξ R V
7 Frequency Response Description 358
θ L L θ
8 System Constants 358 S2 = = a f g (1)
%RH %RH L L
Related Articles 358
where a, b, . . . , g are the static, or quasi-static, trans-
ducer constants of the various stages that relate input to
output variables when the stage is in a settled state of
1 DESCRIPTION OF THE STATIC OR response.
STEADY STATE BEHAVIOR Thus, the static sensitivity is found as the product of static
transducer constants for all stages in a path. If a different
As an example to illustrate the terms introduced in this gain is needed, it could be arranged anywhere in the path by
article, consider the relative humidity sensor in which a altering any one of the constants by design alteration. This
humidity-sensitive membrane changes length with moisture expression assumes that the instrument is not operating in
take-up. The length change causes a metal spring strip to a transient state as none of the stages have any frequency
be strained. On the strip are strain gauges that transduce terms associated with their transfer characteristic.
the strain into a voltage using a simple Wheatstone bridge.
The voltage is amplified and adjusted to give a calibration
and sensitivity facility. Static performance parameters are 2 RANGE FACTORS
first considered to construct a block diagram showing the
complete instrument broken down into serially connected Range, the factor that decides the extent of measurement
paths formed as the stage described above. The spring also ability from the smallest to the largest signal amplitude that
drives a mechanical pointer on a scale. can be accommodated without the system being overloaded
Two measuring paths can be identified from the sens- or the signals becoming limited (saturated), is decided by
ing membrane through the instrument to the two out- the capability of all stages. If the signal level of any one
puts that produce an electrical and a mechanical output stage moves into a region of improper operation, the whole
signal. system will fail to operate correctly.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Static Considerations of General Instrumentation 355
The effective measuring or working range is the ‘range the above example, a 200-V input may deliver 0 V to
of values of the measured quantity for which any single the indicator. Any other system of level shifting might
measurement, obtained under specific conditions of use of be chosen. The degree of suppression is expressed as the
a measuring instrument, should not be in error by more than ratio of scale interval omitted, above or below zero, to that
a specified amount’. presented. A similar and wider concept is when the range
The upper and lower range limits form the limiting is expanded at any point. This is often termed band or
boundaries inside which the instrument has been con- scale spreading.
structed, adjusted, or set to measure the specified quantity. Range of operation for static and quasi-static use is often
The effective range may not necessarily be all of the instru- different to that in the instrument’s dynamic transient mode.
ment’s interval between these two limits. Rangeability is the The term dynamic range is used to indicate which regime
term sometimes used to relate, by some form of expres- the quoted range relates to. It is generally stated in the same
sion, the relationship between upper and lower limits of ratio form as is given above for rangeability and is presented
useful range. in decibel form. At times, its common use would, however,
In the humidity sensor example, the manufacturer states appear to be really defining the static and quasi-static
that the instrument has a range between 20 and 100% RH, performance, not the time-dependent dynamic behavior.
but on the instrument itself is a warning that use outside Related to the range of an instrument is the scale fac-
of the interval 20 to 85% RH may lead to temporary loss tor that might need to be applied to the actual numerical
of calibration. Furthermore, careful reading shows that the value obtained in order to arrive at the correct magni-
usability is only within given accuracy limits over 30 to tude. Scaling is often practiced to allow a normalized
85% RH; thus, the range of reliable operation is within 20 instrument to be applied to many ranges of tasks. Except
to 85% RH, rangeability perhaps being given (there is no where the instrument is multiranging, the system should
standard methodology for expressing this) as providing a be marked accordingly. Multiranging instruments usually
ratio of maximum to minimum limits of 4.25:1. Range- indicate automatically the appropriate scale factor as the
ability is more useful with instruments having unbounded range changes.
measurand possibilities, such as occurs in flowmetering
where the two limits of practical flowmeters are often rather
restricted compared with the interval that might be called 3 ZERO VALUES OF OPERATION
for in practice.
A multirange instrument is one in which the range of Many instrument systems require an operation to set the
operation can be selected by manual or automatic means, position of the zero. For example, a gas analyzer and
the latter being referred to as an auto-ranging instrument. its sampling line may need to be purged with a suitable
When describing the effect on a calibration of influence clearing gas before use. Electronic DC systems usually
quantities, it is common practice to state the range over need periodic zero setting to compensate for the drift that
which such quantities can be tolerated. This is defined as the has occurred.
reference range. The humidity meter performance is stated
for a reference range of 0 to 40 ◦ C temperature variation.
As a guide, ambient environmental conditions may usu- 4 DRIFT
ally be taken as having a mean temperature and reference
range of 25 ± 10 ◦ C; relative humidity of 90% or less Drift tends to be regarded as a DC property of a system,
for general use (but 40–60% for storage of paper mate- but this depends upon the time response needed. In many
rials); and barometric pressure lying within 90 to 110 kPa electronic systems, frequencies of operation less than 10 Hz
(900–1100 mbar). are regarded as being at DC. In some disciplines, however,
Many measurements can be performed more easily, or examples being earth tides or temperature variations in a
with improved accuracy, if the instrument is built to provide cool room, responses may be concerned with frequencies
a limited effective range. This can be done by designing the of the order of cycles per year or per day respectively.
lower range limit to be well above zero (the converse holds Drift, if studied closely enough, will be found to be
for measurands having values less than zero). Such systems a slowly moving dynamic phenomenon and may need
are called suppressed-zero systems. A common example is description to allow for its actual behavior. The simplest
the voltmeter used to indicate mains voltages in the range form of expression is statement of the linear component in
200 to 250 V AC. terms of drift occurring per variable of interest – such as
For such systems, sensing signal levels will not have time or supply level. In electronic amplifiers, drift with time
a zero level coinciding with the output zero, that is, in is often quoted as though it were a linear-time function, but
356 Measuring System Behavior
in reality, it is often formulated on a square-root basis, for provide different values for each direction of approach to
drift in this case does not accumulate linearly. Drift is a the point of interest. Where it cannot be removed, it is
complex, often poorly understood, parameter that can defy common practice to approach the final value in the same
generalization on a causal basis. direction each time. Dead-band can become particularly
Related to the problem of drift specification is how to significant as a cause of instability in closed-loop systems;
adequately describe aging effects. These also defy formal some, however, make use of it to allow the system to settle
generalization. Aging is a factor, like drift, that brings about in the dead-band.
lack of reproducibility in an instrument because over time Dead-band produces a kind of hysteresis, but that latter
the values of components, and hence transducer constants, term is more generally used to describe the magnitude of
will change. Aging can also arise because of operational use the hysteresis phenomenon arising internally in a material.
of the components. There are no general rules to follow to Figure 1 shows a typical hysteresis loop with its measured
reduce aging, but the problem usually is concerned with quantities. Magnetic material, for instance, does not show
appropriate knowledge of material properties. the same magnetic field strength for rising and falling induc-
In the humidity sensor example, aging is to be expected tion levels. The magnitude of the strain condition for a
in the membrane, the curved spring strip, in the strain cyclically stressed mechanical member will vary depending
gauges, and in the electronic components. Each of these on the direction of the stress. In each, hysteresis magnitude
changes will gradually give rise to a shift in calibration increases with increasing excursion. Hysteresis magnitude
and in the position of the zero. Internal adjustments are can be quoted as the ratio of maximum difference of the
provided so that these can be reset. Being an inexpensive upscale and downscale variations to the full-scale excur-
design, intended for low-accuracy use, no figures of aging sion. Other expressions are used. In some applications, the
or drift are quoted. Higher-priced instruments usually quote total area is more important than this difference.
such figures. In the humidity sensor example, mechanical hysteresis
The task of deciding when an instrument should be arises in the strained member carrying the strain gauges
recalibrated, to make up for errors of this kind, is very
subjective. The history of an instrument’s performance Output
is an important factor in being able to predict when it +
will have drifted, or aged, out of the calibration tolerance Indicated output for th
return to zero input n pa
band. s io
ur
Drift is often allowed for in a measurement by returning
xc
le
to a known stable and fixed measurement point periodically
tia
Hysteresis loop
through the test run. It is then possible to extract the drift Ini for constant
Maximum hysteretic excursion
and adjust the values. amplitude cycle
cale
Full excursion of output
Downs
5 HYSTERESIS AND DEAD-BAND
− 0 Input +
Measuring instruments do not respond in a completely
Upscale
continuous manner, nor do they always give the same value
for both directions of closure to a point.
The term threshold is used to signify the smallest level of
signal to which the system will respond. All measurements,
no matter how well designed, will ultimately be limited System
vibrating
by some form of discrete phenomenon. Another example to rest at zero
is seen as the threshold of the zero-suppressed voltmeter
mentioned above, which occurs at 200 V. Threshold is
either caused by a feature of the physical principle involved Loop for
(e.g. a turbine flowmeter will not respond correctly until – backlash
form of
there is sufficient fluid in the pipe) or it may be deliberately Alternative origin with zero-shift error hysteresis
introduced (such as to suppress the zero). It provides a
dead-band of operation. Full excursion of input variable
Many instruments exhibit dead-band effects that are
caused by backlash and play in the drives. These will Figure 1. Schematic of a generalized hysteresis loop.
Static Considerations of General Instrumentation 357
and certainly in the membrane. Backlash arises in the pin frequency of the noise signal. Often, the CMRR quoted
joints of the mechanical system (but is kept in control by is for DC input signals.
a biasing tension), in the sector gear driving the pointer
pinion, and because of slack in the bearing system. The art
of good design is to ensure that these are not significant 6 SIGNAL-TO-NOISE RATIO ENTITIES
for the intended purpose and that they do not increase
significantly as wear occurs overtime. A highly damped, In some instances, the need is for definition of how well
resonant system and slowly moving signal systems will system signals exceed noise levels: this can be assessed
exhibit greater hysteretic effects than more rapidly moving from the signal-to-noise ratio (S/N or SNR). This is the ratio
arrangements. of the signal amplitude to the noise amplitude, again usually
Larger excursions also increase hysteresis error; systems expressed in decibels. The amplitude used could be average,
that can resonate cyclically to rest will usually show less root mean square (rms), or peak-to-peak. Steady state AC
error from this source because the hysteresis loop traversed signals are involved, and they may be of sinusoidal or
in each cycle will become progressively smaller, the system complex waveform. Rarely are the waveshape or amplitude
finally coming to rest in the center of the loop. criteria quoted, leaving considerable room for imprecision,
Electronic systems generally do not pose these problems, especially if the signal and the noise do not have the
but some circuits purposefully introduce backlash. Storage identical waveforms that a ratio properly demands.
effects in semiconductor elements can lead to backlash in The introduction of a stage into a measurement system
fast-acting circuits. usually degrades the SNR to some degree. The ratio, in
Backlash, hysteresis, drift, creep, and aging are all sub- decibels, of SNR at the output to SNR at the input is termed
classes of a general noise problem that instruments of the noise figure.
measurement must cope with. Measuring instruments are designed to fulfill a certain
Noise is the general name given to unwanted sources of task at a specified accuracy level. As increasing the accu-
apparent signal that could lead to error in the measurement racy of a measurement implies an increasingly costlier
mapping process. Every energy regime has its own sources design, the general rule is to choose an instrument that just
of noise. The user of an instrument will often need to know copes with the task needed.
how tolerant an instrument is to common noise sources. It is quite pointless to specify an instrument that has accu-
By suitable design, a stage can often be made such that racy exceeding the task; its capability will not be utilized
and the maintenance costs will be greater than necessary.
noise quantities (of certain kinds) influence two features
To assist the specification and selection of instruments
of the system in such a manner that their combined effect
of appropriate accuracy, many disciplines issue standards
on the true signal is differenced, thus canceling. This is
documents that state accuracy classes. There is no general
called common-mode rejection. Thus, deliberate introduc-
rule for the development of such classes, a dominant reason
tion, through a second influence point, of correlated noise
being that each kind of measurement presents different
into the system can be used to dramatically reduce the orig-
ranges of possibility and need.
inal, impossible to eliminate, noise effect.
For example, it is quite feasible to call up time measure-
As no process is perfect, these common-mode rejection ments to errors of parts in 1012 but photometry can only be
techniques need some means of stating their effectiveness. best achieved to parts in 104 .
This is done through the use of the common-mode rejection Where the need has arisen, agreed accuracy classes are
ratio (CMRR), which is defined as the ratio of the common- defined, within which ‘the instrument will have certain
mode noise signal amplitude to the common-mode error declared metrological properties within specified limits of
signal remaining after rejection. The term is used exten- error’. A class is ‘usually identified by one of a set of
sively in the description of electronic systems where it is symbols, adopted by convention, and usually referred to as
usually quoted in decibels. It is, however, quite a general the “class index”’. Definition of accuracy often ties in with
concept and can be applied to any common-mode signal- that of linearity.
to-noise rejecting system. The term conformance is sometimes used to relate an
As with many parameter statements, the need to keep the instrument to a specified accuracy class. Interchangeability
statement simple can lead to the assumption that CMRR is is the term used to describe how well an instrument can
a simple fixed ratio. This is not always so; it will often be be exchanged with another having the same specifications;
a function of many system variables. Average figures are tolerances of manufacture and differences in the various
frequently quoted. It can, at least in electronic systems, vary manufacturers’ designs cause units to be slightly different
with temperature and usually decreases with decreasing to each other yet still be suitable for a given task.
358 Measuring System Behavior
10 factor, and true and apparent power. Each of these terms
Corner is used in instrumentation in the same way as it is in
frequency
electrical power systems. Probably, the greatest error that
Amplitude ratio
1 arises in their use in instrumentation is the lack of care
in ascertaining that the system is operating with sinusoidal
waveforms (which electrical power systems are generally
0.1 assumed to be), for the parameters are often quoted in terms
of sinusoidally based quantities such as r.m.s. and average
amplitudes and power levels.
0.01
0
8 SYSTEM CONSTANTS
Phase (deg)
− 60
− 120 Constants quoted are often far from being so. Many are
− 180 functions of several operational parameters. Noise figure
− 240
for an electronic amplifier, for example, is a function of
0.01 0.1 1 10 100 signal amplitude, frequency of operation, and the actual
Frequency (Hz) device used. Constants may vary with time.
There do exist, however, certain constants, called phys-
Figure 2. Amplitude and phase responses expressed in Bode
diagram form. ical constants, that are believed to be unchanging. It is
therefore often necessary to know more about the formu-
lation of a given constant if more intelligent use is to be
7 FREQUENCY RESPONSE made of the number quoted.
DESCRIPTION The designer or user is usually left to explore such
ramifications through reading the primary literature. In
Although the response of a stage is dependent upon its essence, consensus obtained from practice provides the
dynamic behavior, the frequency response of a stage is published information generally required.
usually that of a steady state that is taken after the stage has The difficulty is who shall decide what the general
settled to the final value after excitation has been applied demand is, and how does the user know whether the
at each frequency. factors given are entirely adequate for the task in hand.
In the simplest form, the frequency response can be Overspecification can be useful but will increase the cost.
stated in a verbal manner describing it in such terms as Terms used in the description of measuring systems are
‘flat to “x” Hz’ (or ‘between’, or ‘above’). In such cases, many. A useful service, but membership is needed, is that
the extremes are denoted, by convention, to be where the provided by the NCLS International http://www.ncsli.org//
response has fallen off (or risen) to the point where the ratio index.cfm. It is a searchable listing for a glossary of terms
of input to output signal magnitudes has changed by 0.707. and for acronyms and abbreviations.
In some cases, the 3-dB points are used, both ratios being
the half-power points. Some times, the response obtained is
a function of signal amplitude. In such cases, it is necessary
to state the amplitude of the test. RELATED ARTICLES
A more adequate, but harder to publish, description is to
provide a magnitude and a phase plot as a Bode diagram. Article 26, Signal Classification, Volume 1; Article 48,
Examples are given in Figure 2. The true Bode diagram Common Sources of Errors in Measurement Sys-
uses only straight-line segments to approximate the actual tems, Volume 1; Article 56, Transfer Characteristics of
response, but this distinction is not always made. The upper Instrument Stages, Volume 1; Article 58, Description of
curve is, by convention, the amplitude response. Accuracy, Linearity, and Drift, Volume 1; Article 59,
Systems involving continuous cyclic signals require Introduction to the Dynamic Regime of Measurement
description of amplitude, phase angle, phase shift, power Systems, Volume 1.
58: Description of Accuracy, Linearity, and Drift
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
as a percent of actual output reading, such as ±1.0% of
1 Simplification of Terms for Accuracy 359 actual reading.
2 Linearity Description 359 Confrontation with this variety of simplifications leads
one to refer to approved standards documents, such as
3 Drift Description 362
BS 5233 or AS 1514, to find that accuracy is defined only
Related Articles 362 in a general sense and that no suggestion for a uniform
References 362 methodology is provided. An ISA document, formulated
by its SP37 subcommittee, on transducer specification,
provided greater depth in the manner given above (issued
1 SIMPLIFICATION OF TERMS FOR as S37.1 or ANSI MC6.1:1975.)
In reality, the concept of accuracy is more properly
ACCURACY discussed in terms of errors and the uncertainty of a
measurement made with the measuring instrument. The
Having introduced the static regime in Article 56, Trans-
term accuracy, one source suggests, should only be used
fer Characteristics of Instrument Stages, Volume 1 and
as a general descriptor of system performance.
Article 57, Static Considerations of General Instrumen-
Implicit in any statement of uncertainty of an instrument
tation, Volume 1, this article concentrates on the descrip-
is the apparent assumption that the instrument will provide
tion of a measuring system’s accuracy, linearity, and drift.
the same absolute or proportionate performance at any
For reasons of simplicity of specification and description,
part of its usable range. This is rarely a totally valid
the system’s performance features of accuracy and related
assumption. Uncertainty is covered in detail in Article 52,
linearity are often quoted in a simplistic manner by the use
Error Models, Error Budgets and their Calculation,
of a single numerical value. This practice might hide, from
Volume 1.
those persons who are not aware of the facts, that these
are often gross simplifications of a much more complex
reality. To illustrate this point, the following paragraph
is taken from the terminology section of a Foxboro- 2 LINEARITY DESCRIPTION
Yoxall statement:
Accuracy includes the combined conformity, hysteresis, The manner in which error varies over the effective range of
and repeatability errors. The units being used are to be an instrument is expressed as the linearity of the instrument
stated explicitly. For example, expressed in output units, a (but which correctly actually relates to the nonlinearity or
typical expression might be ±1 ◦ C; expressed in percent of lack of linearity between desired and actual performance).
output span, the expression could be ±0.5% of output span; A process is said to be linear when the relationship
expressed in percent of the upper range value, it might read between two quantities is such that a change in one quantity
±0.5% of upper range value. For indicators, it might read is exactly proportional to the associated change in the
±0.5% of scale length; or in some cases, it is expressed other quantity.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
360 Measuring System Behavior
Expression of the nonlinearity of an instrument is Actual performance
achieved in several different ways most of which, again, 100
Span of output (%)
Best straight line BSL
result in a simple statement that conveys information about
how the signal deviates as it ranges over the available span.
Fundamentals of measurement indicate that a single Worst deviation error
numerical value can only characterize a single attribute of
a situation. This is, therefore, the case in the statements
of linearity: each method of description conveys different
0 100
information about the linearity. Seldom is an actual curve
(a) Span of input (%)
presented, the exception being when a calibration chart
is provided. 100
Span of output (%)
Linearity is expressed as deviation from a specified
straight line. Four predominant reference lines might be
used.
The first reference line uses a line placed so that it lies in BSL through zero
the position satisfying the least squares criterion calculated
for the data set available. It is often called the best straight
0 100
line (BSL). It might also be referred to as a least squares
(b) Span of input (%)
line. Figure 1(a) is an example of this line.
In many instances, the system provides zero output for End point at 100%
100
zero input, in which case a more appropriate line to use
is a BSL that is computed so that it is constrained to Span of output (%)
make the line pass through the zero point. In suppressed
or elevated-zero systems, the point of interest will be Line through
the pseudozero used. Figure 1(b) illustrates this line. This end points
method describes the zero-based linearity of the system.
One variation of this is to not use a zero-based line but 0 End point at zero 100
some other suitable reference point. This is termed BSL (c) Span of input (%)
with forced reference linearity.
In terminal-based linearity, a reference line is adopted 100
Span of output (%)
that links the upper and lower limits of the span, a
line drawn between its terminal points. It is also called
terminal linearity or end-point linearity. The line so drawn Theoretical line
is sometimes referred to as the end-point line. Figure 1(c)
shows this line.
Added later in AS 1514 was another line that can be 0 100
used (Figure 1d), being one that lies in the theoretical (d) Span of input (%)
linear position of the principle invoked. This is called the
theoretical slope linearity. Figure 1. Four methods of specifying linearity of a response.
Having first decided which line is the most appropriate
to use, the next step is to select a method of description subject to many causes of error in their determination,
that will express the magnitude of deviations from the line. expression, and subsequent interpretation. A process with
In general, this will usually be the greatest departure from significant hysteresis can exhibit quite different linearity
the line of the output variable. Linearity of instruments is with each direction of signal change and with signal
generally determined as the result of a calibration test in amplitude. The magnitude of influence variables at the time
which only a limited number of values would be deter- of the determination also can alter the error magnitudes. The
mined. Some calibration processes provide a considerable method of calculation of the best fit line and the number
number of values and if they are of random nature, the of data samples available will further influence the value
deviation might then be expressed in terms of the standard that will be stated as the singular value of linearity of the
deviation or some other statistical descriptor. instrument.
A little thought reveals that the above methods of Expressions of the linearity of a system are invari-
expressing linearity and accuracy of an instrument are ably determined for a system in the steady or static
Description of Accuracy, Linearity, and Drift 361
state. Rarely will the same relationship apply when the application to sensors used for rapidly changing measur-
stage is in the dynamic state. Storage parameters will ands. Pressure gauges, accelerometers, position sensors,
alter the instantaneous transfer characteristic according thermometers, and the like will give very different instan-
to the signal amplitude and history of previous signal taneous performance figures to those measured in the
excursions. This point is often overlooked when accu- steady state and stated in their specification of calibration
racy and linearity figures are taken from data sheets for chart.
Gravity meter at fixed location
Output (mgal)
1h
(a) Time
Temperature (°C)
24 h Temperature in a room
Time
(b)
Settlement (mm)
Settlement of structure
(c) Time
Identifiable linear
Magnitude
regions
(d) Time
Electronic amplifier
Output drift (mV)
Safety margin
Added butterfly lines
Response
curve
Lower limit Room temperature Upper limit
(e) Temperature
Output drift (mV)
Added butterfly margins
Response Room
curve temperature
(f) Temperature
Figure 2. Typical drift curves, for example, instruments and measurands.
362 Measuring System Behavior
Another factor to be considered may be that the stage must be fitted from which the drift can then be assessed as
transfer characteristic may possess adequate linearity of for the previous cases.
response but that the BSL may slope at the wrong rate When the curve clearly possesses two or more linear
because of an inaccurate DC sensitivity coefficient for regions (Figure 2d), each of a different slope, a more
the stage. extensive statement will be required. No standards of
Where possible, it may be preferable to provide a practice appear to exist for such cases, so the description
mathematical expression that describes a curve fitted to the should carry a statement of the additional features rather
data, an example being the relationship used for the output than simply quote a fixed numerical drift coefficient.
of a thermocouple. If the drift curve returns to the original value (Figure 2e),
It can be seen, therefore, that usually stated expressions quoting the mean drift that has occurred between the
of accuracy and linearity are invariably simplifications. two end points would imply that the system is perfectly
Specifications of these parameters must be generated and stable. Furthermore, nonlinear curves, such as that given in
interpreted with care and understanding if ambiguity is to Figure 2f, would be grossly misrepresented. One approach,
be avoided. outlined in Stata (1969) for example, is to use a butterfly
Extending these ideas further, to cover other nonlinear characteristic curve. Typical boundary curves have been
responses, again introduces likely confusion as the direct superimposed onto the two amplifier drift curves. Drift is
result of seeking simplified statements of performance. calculated as that occurring from the reference point (which
Specification of drift is a related area where such prob- is often room temperature at which the amplifiers were
lems arise. adjusted for the test) to the extremities of workable range,
plus a margin of safety. This coefficient then provides a
conservative figure within which drift can be expected to
3 DRIFT DESCRIPTION lie for a given temperature excursion. Such simplification,
however, is at the expense of possible overdesign on the
Drift is the feature of a system that characterizes how a part of the user.
system variable, that is intended to be constant, changes These examples clearly show that the specification of
with time. The drift of a gravity meter is one example. The drift can be a complicated matter and that simplistic
term is also used to describe effects of influence quantities definition, as with those for nonlinearity, can be misleading.
on the output, an example being the temperature drift of Manufacturers have been known to use the shortcomings
an electronic amplifier. It is also a term applied to explain in the understanding of descriptions to their advantage when
how a measurand varies with time or other variables. An specifying equipment.
example of this use is to characterize settlement of a Further detail on accuracy and related concepts is to be
structure with time. It is sometimes used synonymously found in CORD (1999) and Westgard (1997).
with the term stability. Some kind of qualifying statement
is needed to uniquely define which context is being used. RELATED ARTICLES
Drift conveys very low frequency response information
about the measuring instrument or of the measurand. It can, Article 50, Errors in Signal Systems, Volume 1; Arti-
on occasion, be very difficult to separate the two. cle 51, Errors in Digital Signal Systems, Volume 1; Arti-
Figure 2 shows several often-met drift curves. For each cle 52, Error Models, Error Budgets and their Cal-
curve, the need is to formulate a simple expression that culation, Volume 1; Article 56, Transfer Characteris-
will convey information about the trend of the output away tics of Instrument Stages, Volume 1; Article 57, Static
from a chosen reference value. Considerations of General Instrumentation, Volume 1;
When the drift is linear, it is a simple matter to express Article 59, Introduction to the Dynamic Regime of Mea-
the slope in terms of the two parameters involved. The surement Systems, Volume 1.
gravity meter example of Figure 2(a) might be expressed
as drifting at x milligals h−1 . Being entirely linear, it would REFERENCES
not matter where the slope was determined along the curve.
The situation becomes a little more difficult when it CORD. (1999) Unit 13: Precision, Accuracy & Tolerance Com-
is required to express the drift for the case shown on munications, CORD.
Figure 2(b). If the characteristic possesses a cyclic com- Stata, R. (1969) A Selection Handbook and Catalog Guide to
ponent of constant amplitude, then a mean line is easy to Operational Amplifiers, Analog Devices, Inc., Cambridge, MA.
determine. If, however, the curve does not have obvious Westgard, J.O. (1997) Opspecs Manual: Operating Specifications
linear features, like that of Figure 2(c), then a best fit line for Precision, Accuracy, and Quality Control, AACC Press.
Introduction to the Dynamic Regime of
59:
Measurement Systems
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
of the dynamics of systems. (This is not to be confused
1 Introduction to the Dynamic Regime of with a subject area called Systems Dynamics, SD, that is
Measurement Systems Performance 363 somewhat different to that being described here.) It has pro-
2 Forcing Functions 364 gressively grown from isolated, unconnected explanations
3 Application of Forcing Functions 365 of the analogous situations existing mainly in electrical,
mechanical, and acoustical areas, into one cohesive sys-
Related Articles 366
tematic approach to the study of the dynamics of linear
References 366
physical systems.
The approach has been widened considerably (see Klirs,
1972), to take in philosophical concepts, such as are found
1 INTRODUCTION TO THE DYNAMIC in social systems, this broader assembly of knowledge usu-
REGIME OF MEASUREMENT ally being referred to as general systems theory. General
systems theory certainly has application and potential in
SYSTEMS PERFORMANCE
some instrument systems, but most designers and users of
The dynamic regime of an instrument stage is generally instruments will find that the more confined and mathemat-
a more complex aspect to characterize and test than the ically rigorous assembly of technique and knowledge (sys-
stage’s static performance. To fully understand the behav- tems dynamics) will be that found to be usefully applicable
ior during the dynamic state, it is necessary to allow in practical realization and understanding of measuring sys-
additionally for the transient solution of the transfer char- tems. The use of the word systems is extremely common
acteristic when forced by an input function, a factor that is and carries numerous connotations ranging from a totally
not important in understanding the steady state and static general concept to the quite specific use that is based in
regime behaviors. mathematical explanation.
Dynamic behavior of systems has been well explained As indicators of the part that is useful to measuring
at the theoretical level if the performance remains lin- instrument dynamic studies the reader is referred to some
ear during the dynamic state. It has been shown that classic texts that sorted out this field well. Karnopp and
the energy relationships of electrical, mechanical, ther- Rosenberg (1975), MacFarlane (1964), Paynter (1961),
mal, and fluid regimes can each be characterized by the Shearer, Murphy and Richardson (1967), and Zahed and
same mathematical description and that a system con- Desoer (1963) for general expositions, and to Coughanowr
taining a cascaded chain of a mixture of these, as is and Koppel (1965), Fitzgerald and Kingsley (1961), Koenig
typical of instrument systems, can be studied in a coher- and Blackwell (1961), and Olson (1943) for more spe-
ent manner by the use of such mathematical techniques. cific uses in areas of chemical plant, electrical machines,
This body of knowledge and technique lies in the field electromechanical devices, and the so-called analogies,
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
364 Measuring System Behavior
respectively. Other more recent works of the engineering influence quantities, through numerous other unintentional
systems kind are Franklin (1997), Hsu and Hse (1996), Nise ports. In reality the true two-port, four-terminal, instrument
(2000), Oppenheim et al. (1996). stage can rarely be realized for virtually all designs are
Several authors have extracted, from the total systems influenced to some extent by unwanted noise perturbations
knowledge contained in such works, the much smaller part entering through many mechanisms.
that is needed by measuring systems interests, presenting When considering the dynamic response that might arise
this as chapters in their works. Such accounts are to be for a stage it is, therefore, necessary to first decide the
found in Beckwith and Buck (1969), Doebelin (2003), and ports through which a forcing function signal might enter.
Neubert (1976). This decided, the next step is to assess which kind of forc-
Notable foundation papers have also been published ing function is relevant and apply this to the real physical
that further condense the information for measurement device, or to its correct mathematical model expressed in
systems application (see Bosman, 1978; Finkelstein and the state-variable or transfer function form. Alternatively, it
Watts, 1978). Finally, many control-theory-based works are might be simulated in a digital computer. When the trans-
of relevance. fer function model is used, the product of the Laplace
There exists, therefore, a considerable quantity of well- transform of the forcing function and the transfer function
organized knowledge about the physical behavior that may can be solved to provide the transient dynamic behavior
be encountered in measurement systems studies. When
in a reasonably simple manner, see Article 19, Dynamic
response is linear it is possible to make use of mathemati-
Behavior of Closed-loop Systems, Volume 1. This method
cal models to obtain a very workable understanding of the
is dealt with in detail numerous electrical engineering and
dynamic behavior. If it is nonlinear, however, then the situ-
systems texts. Here are presented some typical forcing
ation is not so well catered for because no such widespread
functions that might be used to test or study an instru-
and generally applicable mathematical foundation has been
ment stage.
forthcoming in truly formal description terms. In practice,
however, designers and users of instruments can find con- Certain types of forcing function lend themselves to ana-
siderable value in assuming linearity, if only for a limited lytical solution being easy to apply to the Laplace method
range of operation. For this reason, it is important to know of response evaluation. They are also simple to procure
how to recognize whether a system functions in a linear as practical test signals that, although not perfect, come
manner and which techniques apply in such cases. Further, close enough to the mathematical ideal. For these rea-
it is often up to the designer to chose methods and, thus, it sons, testing and evaluation of systems tends to attempt
is recommended that consideration be given to the use of first to make use of one or more of several basic forc-
those modules that do respond in a linear manner for their ing functions.
operation can be modeled precisely using long hand math These functions (see Figure 1) are the discontinuous unit
or modeling tools. step and the unit impulse, the ramp plus the continuous sine
The purpose of this article is not to expound the mathe- wave. They are described elsewhere and, therefore they do
matical modeling and design of transducers in a concise and not need further theoretical expansion here.
rigorous manner, but to present the commonly seen charac- These functions are easy to produce and apply in prac-
teristics of linear dynamic systems that will be encountered tice, but it must be recognized that they may not pro-
so that they can be recognized and operated upon by using vide an adequate simulation of the real forcing signals
a fundamental approach that is based on knowledge of existing. This point is not always made clear in the gen-
their characteristics, see Article 19, Dynamic Behavior of eral treatment of the dynamics of common linear sys-
Closed-loop Systems, Volume 1. It also provides the basic tems stages.
descriptive terminology needed in specification of response When such simplifications are not adequate, other more
characteristics. suitable ones must be applied. If they can be transformed
with the Laplace transform expression, and if the product
with the transfer function can be arranged so that a solution
2 FORCING FUNCTIONS into time-variant terms is obtained, then a theoretical study
can be made. In many cases, however, this is not possible
To see how a stage performs to various kinds of changing within bounds of realistic adequacy, and alternative meth-
input signals, it is forced into its transient state by one or ods of study must be employed if no way can be seen to
more external inputs. These are termed forcing input or simplify satisfactorily the forcing transfer function. Simula-
excitation functions. They may be applied to the intentional tion using a computer-based tool often provides the means
input of the stage or cause transient behavior by entering, as to a solution.
Introduction to the Dynamic Regime of Measurement Systems 365
provide a perfect step function to a pressure sensor to obtain
Amplitude
1 Step of unit amplitude after its step response. Faster acting systems, such as found in the
t = 0, zero t < 0 electronic and optical disciplines, can supply rapid changes.
When the response to a slow-to-rise input signal is
0 Time
needed, the use of the step or impulse function may provide
∞ misleading information. A ramp function is more appli-
Impulse (theoretical) of
Amplitude
infinite amplitude and cable in such cases. Many of the published systematized
zero time duration transient behavior descriptions have not included this par-
ticular forcing function, the solutions usually presented
0 Time being for impulse, step, and sine wave inputs. Sinusoidal
1/∞ excitation can, however, sometimes be used to approximate
ramp responses.
A Impulse (Dirac ), practical The unit impulse function represents the input provided
Amplitude
impulse of area A units, by a sudden shock in a mechanical system – it might be
width being much less than
amplitude due to slackness in a drive link taking up. In electronic
systems, it represents such events as a high-voltage pulse
1/A
Time generated by lightning or the switching surges from a
nonzero crossing controlled silicon controlled rectifier. In
practice, a pseudoimpulsive function, called the Dirac or
Amplitude
R Ramp, of slope R :1 delta function, is used instead of the true impulse. As will
beginning at t = 0 be seen, the transient solutions of typical linear systems
1
0 Time to impulse and step functions are somewhat similar in
transient shape.
The third most commonly used input function is the sine
Amplitude
Terminated ramp
wave (or cosine wave, which is the same function, phase-
R (All above are discontinuous,
1
shifted in time). This acts to excite a system in a continuous
singular events. They may be applied
0 Time with time delay after t = 0) cyclic manner forcing it to be excited in both the transient
and steady state when the sine wave is initially applied, the
former dying away to leave the latter as the solution most
+1 Unit sine wave
0 usually discussed.
−1 In practice, systems are often likely to be disturbed by a
continuous complex input waveform. As complex wave-
Figure 1. Typical forcing functions used in testing and study of forms can be broken down, by Fourier techniques, into
the dynamic response of systems. a set of sine waves of different frequencies, amplitudes,
and phases, the use of sine waves of the correct ampli-
tude and frequency enables the system to be studied one
3 APPLICATION OF FORCING component at a time. Sine wave response is also called
FUNCTIONS frequency response, see Article 27, Signals in the Fre-
quency Domain, Volume 1; Article 28, Signals in the
A step or impulse will excite a system into its dynamic Time Domain, Volume 1; and Article 29, Relationship
transient response as might a sudden change of demand in Between Signals in the Time and Frequency Domain,
a control loop, or a sudden change in a measurand as exists Volume 1.
when an indicating voltmeter is first connected to a live cir- The concept that a complex continuous waveform can be
cuit. Many physical systems, however, cannot provide such so resolved into separate components rests on the assump-
rapidly changing signals. Mechanical and thermal systems tion that the system is linear and that superposition applies.
often cannot supply a rate of rise that is great enough to Nonlinear systems can behave quite differently to com-
be regarded as a step because of the presence of signifi- plex signals, creating, for instance, harmonics of lower
cant storage of energy within the components. For example, frequency than those existing in the original signal.
attempting to square-wave modulate the dimensions of a It is often considerably easier, and more reliable to obtain
piezoelectric crystal at relatively high frequency will pro- the transient response of a system by practical testing than
duce quasi-sinusoidal output response, not a square wave – it would be to develop a mathematical model. This is one
the crystal acts as a filter. Similarly, it is not possible to of the reasons why the data sheets of instrument products
366 Measuring System Behavior
often include graphical statements of transient response. Finkelstein, L. and Watts, R.D. (1978) Mathematical Models
Further, a detail about forcing functions can be found in of Instruments – Fundamental Principles. Journal of Physics
E:Scientific Instruments, 11, 841–55.
chapters of classical control texts such as Atkinson (1972),
Coughanowr and Koppel (1965), and Shearer, Murphy and Fitzgerald, A.E. and Kingsley, C. (1961) Electrical Machinery,
McGraw-Hill, New York.
Richardson (1967).
Franklin, G.F. (1997) Digital Control of Dynamic Systems,
Addison-Wesley.
Hsu, A. and Hse, W. (1996) Schaum’s Outline of Signals and
RELATED ARTICLES Systems, McGraw-Hill.
Karnopp, D. and Rosenberg, R.C. (1975) System Dynamics: A
Unified Approach to Physical Systems Dynamics, MIT Press,
Article 18, Nature and Scope of Closed-loop Systems, Cambridge, MA.
Volume 1; Article 19, Dynamic Behavior of Closed-loop
Klirs, G.J. (1972) Trends in General Systems Theory, Wiley, New
Systems, Volume 1; Article 27, Signals in the Frequency York.
Domain, Volume 1; Article 28, Signals in the Time
Koenig, H.E. and Blackwell, W.A. (1961) Electromechanical Sys-
Domain, Volume 1; Article 29, Relationship Between tems Theory, McGraw-Hill, New York.
Signals in the Time and Frequency Domain, Volume 1; MacFarlane, A.G.J. (1964) Engineering Systems Analysis, Harrap,
Article 30, Statistical Signal Representations, Volume 1. London.
Neubert, H.K.P. (1976) Instrument Transducers, 2nd edn, Claren-
don Press, Oxford.
REFERENCES Nise, N.S. (2000) Control Systems Engineering, Wiley, New York.
Olson, H.F. (1943) Dynamical Analogies, Van Nostrand, New
Atkinson, P. (1972) Feedback Control Theory for Engineers, York.
Heineman, London.
Oppenheim, A.V., Willsky, S.H., Nawab, A. and Nawad, H.
Beckwith, T.G. and Buck, N.L. (1969) Mechanical Measure- (1996) Signals and Systems, Prentice Hall, Englewood Cliffs,
ments, Addison-Wesley, Reading, MA. NJ.
Bosman, D. (1978) Systematic Design of Instrumentation Sys- Paynter, H.M. (1961) Analysis and Design of Engineering Sys-
tems. Journal of Physics E:Scientific Instruments, 11, 97–105. tems, MIT Press, Cambridge, MA.
Coughanowr, D.R. and Koppel, L.B. (1965) Process Systems Shearer, J.L., Murphy, A.T. and Richardson, H.H. (1967) Intro-
Analysis and Control, McGraw-Hill, New York. duction to Systems Dynamics, Addison-Wesley, Reading, MA.
Doebelin, E.O. (2003) Measurement Systems: Application and Zahed, L.A. and Desoer, C.A. (1963) Linear System Theory,
Design, 5th edn, McGraw-Hill, New York. McGraw-Hill, New York.
60: Zero-order System Dynamics
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
The coefficients an , . . . , a0 are constants. They could be
1 General Remarks on Linear Systems functions of t but not of y. Powers of derivatives do not
Responses 367 occur in this class of differential equation. Not all orders
2 Summary of Systems Orders in Transfer of derivatives less than the highest are necessarily present.
Function Form 367 The expression on the left-hand side is the character-
3 Definition of the Zero-order System Response 368 istic equation of the differential equation. It contains the
4 Design Aims for Measurement Systems information that characterizes the physics of the system of
Behavior 369 interest and how it will provide output when it is excited.
5 Examples of Zero-order Measurement The expression on the right-hand side characterizes the forc-
Systems 369 ing function, the input signal, applied to that system. Thus,
both sides contain independent information about, on the
6 Specifying Zero-order Systems 369
one hand, the inherent transfer characteristic of the stage
Related Articles 370 and, on the other hand, the forcing function applied to the
Reference 370 stage. As both functions are linear expressions, by the above
definition of requirement, they can both be transformed into
transfer functions by Laplace methods.
1 GENERAL REMARKS ON LINEAR System performance can thereby be obtained theoreti-
SYSTEMS RESPONSES cally if the nonhomogeneous linear differential equation for
the specific system can be solved to yield the output versus
The following discussion applies only if the dynamic behav- time relationship.
ior of the system of interest can be adequately characterized In providing the explanation of dynamic response, au-
by a suitable linear ordinary differential equation, this often thors can make use of the differential equation format given
being found to be the case in instrumentation, but by no here but often using other symbols: they may make use of
means always so. The theoretical foundation given in this the operator D to replace derivatives; the explanation might
article provides the basis for understanding the behavior of be presented directly in transfer function form; differential
linear systems in general. The first-order system is covered equations might be presented in complex number form, plus
in Article 61, First-order System Dynamics, Volume 1 other variations. These tend to confuse the issue somewhat
and the second-order system in Article 62, Second-order but each produces the same result.
System Dynamics, Volume 1.
These equations will have the general form (called
2 SUMMARY OF SYSTEMS ORDERS IN
the nonhomogeneous linear differential equation) for input
variable x and output variable y and unit amplitude input: TRANSFER FUNCTION FORM
dn y dn−1 y The above given general differential equation can be written
an + an−1 + · · · + a0 y = x(t) (1) in the specific sense as an ascending order of equations
dt n dt n−1
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
368 Measuring System Behavior
when additional derivative orders are progressively added. not alter the time features of a time-dependent function
For unit amplitude input, these become introduced to the system by the forcing function. Thus,
the zero-order stage cannot introduce phase shifts between
zero order a0 y = x(t) the input and output, alter frequencies, or provide chang-
dy ing amplitude with frequency. It can, however, provide
first order a1 + a0 y = x(t)
dt constant attenuation or gain and it can transform energy
2
dy dy variables. The numerical coefficient a0 is the static sensi-
second order a2 + a1 + a0 y = x(t)
dt dt tivity of the stage, the constant of transduction. In principle,
dn y dn−1 y the zero-order stage provides perfect dynamic response
nth order an n + an−1 n−1 + · · · + a0 y = x(t)
dt dt for situations where the input variable must not be pro-
(2)
Zero-, first-, and second-order systems are all that need cessed with respect to its time features as it passes through
usually be considered in this discussion of the transient the stage.
response of instrument stages because higher-order differ- Considering the transfer function form, it can be seen
ential equations can be reduced, if need be, to these, through (but this will become clearer when the higher-order systems
suitable manipulation. Also, establishing analytical solu- are studied) that the stage contains no storage mechanism
tions, which requires the roots to be found to higher orders because there is no time constant associated with it.
can be far more difficult, if at all possible, than for these In many design situations, a zero-order response is
three cases. This may seem rather short of real requirements highly desirable. For example, consider a resistance strain
but, in general, a very large part of instrument testing and gauge (Figure 1a) in which extension of the electrical
characterization can be done with the understanding of these resistance wire, attached to a sample, alters the resis-
three types of system response. tance proportional to the strain so induced. In normal
Transformed into typical unitized transfer functions, the use, the gauge, being apparently purely resistive electri-
differential equations become cally, will respond completely in phase with strains of
the sample. No matter how rapid the input signal is,
Y (s) within practicality, the output follows without delay, with-
zero order =1
X(s) out signal amplitude change, and without frequency change
Y (s) 1 (Figure 1b).
first order =
X(s) τs + 1
Y (s) 1
second order =
X(s) (τ1 s + 1)(τ2 s + 1) Gauge factor k = ∆R /R
∆L /L
Y (s) 1
nth order = ∆l
X(s) (τ1 s + 1)(τ2 s + 1) · · · (τn s + 1) l
(3)
where τ1 , . . . , τn are termed time constants and Y (s) and
X(s) the output and inputs respectively. This groundwork Output (electrical resistance
laid now enables the three types of systems to be studied in change of ∆R )
R
order. This article covers the simplest, almost trivial one –
the zero-order system. See Article 61, First-order System
Dynamics, Volume 1 for first order, Article 62, Second-
order System Dynamics, Volume 1 for second order. Input ∋ = ∆l (mechanical strain)
l
(a)
Input, ∋
3 DEFINITION OF THE ZERO-ORDER
SYSTEM RESPONSE
The zero-order system is so trivial mathematically that it is Output, ∆R
R
rarely mentioned in texts on dynamic response. In practice,
however, it is often (but not always) a most desirable Time
response, one that the designer strives to achieve but rarely
(b)
obtains in a perfect sense.
The differential equation form, given above, shows that Figure 1. Example zero-order system – a resistive strain gauge:
there are no derivatives, meaning that the system does (a) gauge and its mathematical model; (b) responses.
Zero-order System Dynamics 369
As the gauge is presumably being used to monitor the in many applications by the long storage time of the mem-
strain behavior of the sample, zero-order response will brane. This helps average (integrate) fluctuations that might
yield the observer true information about what is actu- occur in a room in which a fan circulates pockets of air
ally happening to the sample with time and in ampli- differing in %RH that would otherwise cause a closed-loop
tude. In control-loop design, zero-order sensing components controller to attempt to smooth out to the detriment of the
often help the designer obtain a very tight, responsive real need.
system.
In practice, the designer or user must decide if zero-order
characteristics are what are needed (see below). Assuming 5 EXAMPLES OF ZERO-ORDER
this to be so, it is then necessary to ascertain that the device MEASUREMENT SYSTEMS
does indeed produce them to the dynamic performance
level desired, for, in reality, a perfect zero-order system Of the texts already referenced concerning dynamic res-
is a mathematical abstraction. All systems include some ponse, only Doebelin (2003) makes specific mention of
measure of storage effects; it is these that will degrade the zero-order systems. His description includes another exam-
zero-order response into a higher-order system where they ple – that of a resistance potentiometer used as a position-
become significant. sensitive sensor.
In the strain gauge example, storage might arise through It will probably now be clear that it is not possible to
thermal capacity of the gauge and sample; through after- provide a list of zero-order instrument devices and stages
elastic mechanical effects in the gauge material; through for that attribute is decided by the use to which the stage
parasitic electrical inductance and capacitance when the is applied. As a guide, if the energy storage effects are
frequency of response or excitation increases sufficiently; minimal compared to existing energy levels and if only
through mechanical compliance (and its capacity to enable relatively low-frequency forcing functions are to be applied
the mass of the gauge to move out of phase with the sample) to the stage, then zero-order response probably occurs. The
in the bonding of the gauge to the sample via the thin important factors are the relative values of these parameters
layer of adhesive; plus other reasons. These, in practice, and not their absolute magnitudes. A device operating at
will each introduce time-dependent features situated over picowatt energy levels can provide delays of hours and
the frequency spectrum of response, from virtually DC to conversely, a megawatt energy system might respond in
the radio frequencies. milliseconds.
It can be very dangerous, in practice, to assume the
response order for a given system. Simple testing will easily
4 DESIGN AIMS FOR MEASUREMENT establish the order for the application in question. This is
SYSTEMS BEHAVIOR to be recommended where possible. Many of the problems
that arise in measuring system design and application occur
The art of good instrument design and application as a result of assuming that a sensor or stage provides zero-
is to select a stage that performs with the response order response when, in the specific application, it actually
that just suffices. In a relative humidity sensor, it operates with higher-order responses present. One exam-
can now be seen that the hygroscopic membrane, the ple is the use of a classical flat bed plotting table that
strained metal beam, the strain gauges, and the electrical will filter signals that are too fast for it to respond to.
resistance Wheatstone bridge used can each be regarded It will still record them, but with diminishing amplitude
as zero-order elements for the purpose of studying and as the frequency of the component rises. Phase shift will
designing the sensor for the application it was built to also occur between signals being recorded on multichannel
fulfill. recorders.
If a response less than several minutes is needed, then
the various elements must be viewed as higher-order com-
ponents, and a more complex model must be used to study 6 SPECIFYING ZERO-ORDER SYSTEMS
and improve the overall dynamic behavior.
Design of instrument systems does not, however, always Each kind of response can be described in simple terms
require zero-order performance. In many instances, there is by stating certain simple descriptors. In the case of the
need for storage in a system. Storage helps provide such zero-order system, these are as follows:
often-desirable effects as integration and differentiation,
which, in turn, enable frequency filtering to be devised. • Systems type is of zero order
Applicability of the relative humidity sensor is enhanced • Amplification constant.
370 Measuring System Behavior
There is no need to describe any frequency features Issues, Volume 1; Article 56, Transfer Characteristics
for it is known that zero-order systems do not exhibit of Instrument Stages, Volume 1; Article 57, Static
any frequency varying behaviors. The other orders need Considerations of General Instrumentation, Volume 1;
additional descriptors. Article 59, Introduction to the Dynamic Regime of
Measurement Systems, Volume 1.
RELATED ARTICLES
Article 35, Systems in the Time Domain, Volume 1; REFERENCE
Article 36, Systems in the Frequency Domain, Volume 1;
Article 37, Relationship Between Systems in the Time Doebelin, E.O. (2003) Measurement Systems: Application and
and Frequency Domain, Volume 1; Article 38, Stability Design, 5th edn, McGraw-Hill, New York.
61: First-order System Dynamics
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
Rearranging the coefficients to yield a unit coefficient for
1 Descriptive Equations for First-order System y gives
and an Example 371 a1 dy 1
2 Response of First-order System to Step + y = x(t) (2)
a0 dt a0
Function Input 373
3 Response of First-order System to Ramp Taking Laplace transforms produces
Function Input 374
4 Response of First-order System to Sinusoidal a1 1
sY (s) + Y (s) = X(s) (3)
Function Input (Frequency Response) 375 a0 a0
5 Examples of First-order Measurement
Systems 376 Rearrangement again gives the transfer function as
6 Specifying First-order Systems 377 Y (s) 1 1
Related Articles 377 G(s) = = (4)
X(s) a0 (a1 /a0 )s + 1
References 377
At zero frequency, and with sinusoidal excitation, the oper-
ator ‘s’ validly becomes jω and is equal to zero. Thus:
1 DESCRIPTIVE EQUATIONS FOR 1
G(s) = (5)
FIRST-ORDER SYSTEM AND AN a0
EXAMPLE
The coefficient 1/a0 represents the static regime trans-
duction constant or amplification factor, often denoted as
The dynamic regime of systems is introduced in Article 59,
K. It may involve either attenuation (or gain) of a com-
Introduction to the Dynamic Regime of Measurement
mon input–output variable or relate one form of variable
Systems, Volume 1. For linear signal systems the vari-
to another.
ous responses can be ordered. The zero-order system is
The ratio a1 /a0 is termed the time constant for reasons
covered in Article 60, Zero-order System Dynamics, Vol-
that will become clear below. It has dimensions of time
ume 1. This article deals with the dynamic behavior of
and it is generally denoted as τ . All first-order systems
first-order systems. Second-order responses are discussed in
have the same characteristic kind of time response to any
Article 62, Second-order System Dynamics, Volume 1.
given forcing function and τ specifies the actual numerical
The differential equation for a first-order system is, with
magnitude of that response, this being regardless of the
a unit amplitude forcing function:
energy regime of the system.
dy The mercury-in-glass thermometer is what is chosen by
a1 = a0 y = x(t) (1) many authors to illustrate a first-order system. In use, the
dt
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
372 Measuring System Behavior
thermometer is immersed in a fluid bath and is intended least two more important requirements that do not strictly
to transduce fluid temperature into an equivalent capillary- hold are that heat is only stored in the mercury, not in the
tube mercury movement without introducing delay or being glass, and that no heat is lost to the environment via the
affected in amplitude by the speed of the input temperature heat conduction path of the thermometer stem.
change. In practice, the thermometer can only be regarded Despite the obvious presence of these many imperfec-
as a zero-order system when very slow changes are to be tions, the first-order model that can be developed for the
monitored. For changes that take place in a few seconds, thermometer does provide quite a reasonable estimate of the
the thermal storage effect of the mercury and the heat flow dynamic performance of the thermometer under changing
restriction of the thermal film transfer effect that thermally bath temperature conditions.
couples the fluid to the mercury, provide first-order action. Derivation of the differential equations, on the basis of
Figure 1 is a schematic of the system including all energy balance and knowledge of the physical process by
variables that experience shows to be relevant in order that which heat changes cause a rise in the mercury meniscus,
the system can be adequately treated by a first-order model. leads to the actual development of numerical quantities for
The physical operation of the system must first be studied the time constant and the transduction constant. This exam-
to discover assumptions that can be made in order to obtain ple is worked through in both Coughanowr and Koppel
an adequately descriptive model that is also reasonably (1965) and Doebelin (2003) each providing a slightly dif-
workable, mathematically. ferent form of expression of the resulting numerical factors.
Assumptions in this example relate to such factors as the In terms of the parameters defined in Figure 1, it can be
majority of resistance to heat transfer between the fluid and shown that
the mercury being provided by the heat transfer film, the y V × V
glass being of insignificant thermal resistance. The mercury =K= (6)
x ω=0 a
is presumed to have perfect heat conducting properties
and to respond to heat changes producing instantaneous and
corresponding volume changes. Other factors assumed are ρsV heat stored
that the glass envelope changes volume at the same rate τ = = (7)
hA resistance to heat flow
as the mercury, an assumption that is not exactly true
when high discrimination is needed since the glass takes Note that the time constant parameter concerns one storage
considerable time to return to a smaller volume after being and one energy dissipative parameter.
expanded. Other factors that must be presumed to hold are It is, therefore, possible to determine theoretically not
that heat transfer is achieved over a constant surface area only the transduction constant as a numerical value (as also
and that it is not a function of the actual temperature. At could a practical test) but it also provides knowledge of
(Cross sections)
Area of mercury, a
Capillary
y tube
Differential
volume change, Heat retarding Specific heat of
Area of heat ∆V film coefficient, mercury, s
transfer, A
h
Density of Volume of Bulb
mercury, r mercury, V
Mass of mercury, m Glass transfers heat
without loss
Input temperature, x
Figure 1. Basic physical system of a mercury-in-glass thermometer modeled at a first-order system level.
First-order System Dynamics 373
the factors of the system upon which it depends; simi- are thus satisfied. If energy is already being discharged,
larly so for the time constant. Knowing these parameters that is, the system is still in the transient state of a pre-
enables the system to be tuned to provide whatever static vious forcing function application, then the response will
and dynamic response is needed or to learn that existing be the result of two forcing functions and not the step
materials and other practical constants might not allow a applied here.
specified performance to be realized. They also enable the Being generally applicable, this time function can use-
system to be easily specified in terms of three statements: fully be drawn as a graph in which the input is normalized
the system order, K, and τ . The constants, however, vary to a unity maximum and the time scale is expressed in terms
greatly depending on the system conditions into which the of a time-variable ratio. This is shown in Figure 2.
thermometer is coupled. They cannot be quoted without If the step is applied as a decrement, or fall, rather than
reference to the model conditions that are applicable. the rise shown here, the curve is simply reversed, the shape
The importance of τ can now be considered in its is identical. It will then follow the form
capacity as a unique descriptor of the transient behavior
of a first-order system when excited by certain given y(t) = AKe−t/τ (12)
forcing functions.
Of the forcing functions introduced above, the step (the
It can be seen that τ is an important descriptive constant
impulse, being similar, need not be covered in the same
of the system, deciding the curve’s actual magnitude. It
detail here), the ramp, and the sine wave are the most
also directly describes the initial rate of rise; a tangent
useful inputs to consider. Other input functions will require
to the original line will intercept the final value level at
a special mathematical study of the transfer function along
the 1 τ point. This implies that if the system were to
the lines now outlined.
maintain (which it does not) the initial slope, the output
would reach the final value in the time of 1 τ . In general,
2 RESPONSE OF FIRST-ORDER SYSTEM this slope, for any order of system, is termed its slew-
TO STEP FUNCTION INPUT ing rate.
It can be seen from Figure 2 that the continued response
The Laplace expression for a step function of amplitude reaches what might be considered to be close enough to
A is the final value, at 86.5, 95.0, and 98.2%, levels for 2τ , 3τ ,
A and 4τ , respectively. By 5τ , it is certainly close enough to
X(s) = (8) a settled value for most applications. Although the time
s
constant is the key parameter of definition, it must be
Response of a first-order system, which has G(s) = realized that for final values to be reached, it takes a longer
K/(τ s + 1) (where K is the transduction constant and τ time than that given by the appropriate factors presented
the time constant) is given by above. In 1τ , the response has only reached 63.2% of the
final value. Furthermore, it rises to 63.2% of the remainder
K A in the next 1τ period, and so forth.
G(s)X(s) = (9)
τs + 1 s
Expansion by partial fractions yields Initial rate of rise
AK AK
Y (s) = − (10) Rising step
s s + 1/r 0.8
Using tables of Laplace transformations of common func-
tions to invert this expression gives, for t from zero onward:
Y (t )
0.4
y(t) = AK(1 − et/τ ) (11) AK
Falling step
This expression describes the dynamic behavior of the
first-order system to which a step function of a given ampli- 0
2 4
tude is applied at t = 0. For the above to hold when the
t/t
step is applied, the system must already be in the steady
state, meaning that the storage element is completely dis- Figure 2. Normalized response of a first-order to a step input
charged through the dissipative path; the initial conditions function.
374 Measuring System Behavior
Respective values for a falling unit step are that it falls reverting a, b to physical constants
to 36.8% in 1τ , to 13.5% in 2τ , to 5% in 3τ , and 1.8% by
the time 4τ has passed. RK RKτ RKτ
Y (s) = − + (17)
In order to provide a general descriptive term that applies s 2 s s + 1/τ
to any instrument response, linear or nonlinear, instrument
users have adopted the term settling time. This is usually These are now in the form from which inverse Laplace
quoted as the time needed for the system to settle to within transforms can be recognized from standard tables. Three
given percentage limits of the final value. These limits may time functions result to make up the whole response
be expressed as an error band. For example, it will take a function:
first-order system, excited by a unit step function, a time
of 3τ to come within ±5% limits. Note that a first-order Y (t) = RKt − RKτ u(t) + RKτ e−t/τ (18)
system cannot overshoot the final value but approaches it
from one side only. that is, output = ramp + step + exponential transient.
Detailed analysis of impulse input function response is This is interpreted to be a ramp function response of slope
to be found in Coughanowr and Koppel (1965). rate RK. It will have a different rate to the input when
the stage has either attenuation or gain for the common
input and output variables (that is, volts in, and volts
out, as would occur in an RC stage) and the rate will
3 RESPONSE OF FIRST-ORDER SYSTEM
differ when the rate is not linked to the input because
TO RAMP FUNCTION INPUT the transduction constant is one of conversion (that is,
temperature change yielding a length change output). If the
Step inputs may give an unrealistically severe rate of rise. transduction constant is unity, then input and output slopes
Using a step or impulse in such cases will provide an output will be identical.
response that is much greater in transient deviation than The second term, a step of RK τ amplitude, and negative
what actually occurs. Slower rise rates, as represented by in sign, tells us that the output response begins to occur at
the ramp input function, are often more realistic forcing t = 0 after a time lag. If the rates of the input and output
functions to use. As will be seen, the output response are identical, there will exist a constant lag with time, but
to a ramp follows the input more faithfully but quite an if they differ, the lag magnitude will be proportionately
unexpected delay and possibly a progressively increasing related to the time that has elapsed after initiation of the
deviation error results. input signal. This lag effect has been interpreted by some
Using the same mathematical technique as was applied in authors as meaning that this combination of input to a first-
the previous step response explanation, it is first necessary order system will produce a system in which the output
to obtain the Laplace transform for a ramp function of rate measured value at a given time is that of a fixed time
of rise, R. Forming the product of this with the transfer before. This observation only holds true if the rates of the
function for a first-order system having a transduction input and output ramps are identical. If not, the steady state
constant of K yields error changes with time.
Finally, the third component of the total response is
K R a transient signal of exponential form that occurs at the
Y (s) = G(s)X(s) = (13)
τ s + 1 s2 commencement of the ramping action to die away to virtual
zero within a short time.
Rewritten, this becomes The total response is shown diagrammatically in Fig-
RK 1 a ure 3. The form of the transient part of the response, when
Y (s) = = 2 (14) isolated, follows that given as the falling curve in Figure 2.
τ s (s + 1/τ )
2 s (s + b)
Clearly, if the output is to be a faithful replica of the ramp
where input, then the time constant of the first-order stage must be
RK 1 sufficiently small in magnitude. Furthermore, the transduc-
a= b= (15) tion constant K needs to be unity. When these conditions
τ τ
are met, a ramp will be followed with negligible droop, with
Expanding this into partial fractions gives minimum time delay, with minimal transient component,
and with a small fixed, rather than changing, magnitude
a/b a/b2 a/b2 error. (Under such conditions it virtually becomes a zero-
Y (s) = − + (16) order system.)
s2 s s+b
First-order System Dynamics 375
Input then inverse Laplace transformation:
AKτ ωe−t/τ AK
Y (t) = +√ 2 2 sin(ωt + φ) (22)
τ ω +1
2 2 (τ ω + 1)
Steady state
Output where φ = tan−1 (−ωτ ); that is,
error
output response = transient + steady state components
(23)
R RK (slope may be The transient term is short lived, decaying to leave
greater than input only the steady state component, that is, the contribution
Amplitude
1 1
described as the frequency response of a system. The
steady state term is a sine wave of the same frequency
Steady state time lag as the forcing function but it is phase shifted as a lag,
by angle φ. It will have the so-called DC gain decided
0 by the amplitude of sine wave at that frequency and
Step of Time the transduction constant. The gain factor, however, will
RK t Transient added to progressively decrease (be attenuated) as frequency rises,
steady state component
the magnitude of this attenuation being dependent upon the
Figure 3. Response of a first-order system to a ramp input time constant and the frequency.
function. It is clear, from the presence of the transient term,
that a first-order system will not immediately respond in
the steady state form but will undergo initial exponential
4 RESPONSE OF FIRST-ORDER SYSTEM behavior. Thus, the system will not ring (oscillate) but
does need a certain amount of time to establish its intended
TO SINUSOIDAL FUNCTION INPUT frequency response action.
(FREQUENCY RESPONSE) The steady state normalized frequency responses of a
first-order system with transduction constant K, to input
sine wave frequencies of amplitude A are plotted in
The requirement for testing, or understanding, may relate
Figure 4, being given with respect to the amplitude and
to the behavior under conditions of continuous steady
phase of signals.
state excitation of the system stage. This is fundamentally
approached using the sine wave signal train as the forcing
function since all other continuous complex signals can be
reduced to the sum of such responses. The forcing function 0.8
y
x
in this case in the untransformed time domain is
KA
Amplitude 1
0.4
X(t) = A sin ωt (19)
In Laplace form
0 2 4 6 8 10
Aω wt
X(s) = 2 (20)
s + ω2 0
Phase (deg)
Thus, the response of a first-order network to this will be −30
K Aω
Y (s) = G(s)X(s) = −60
τ s + 1 s + ω2
2
AK ω
= (21) −90
τ (s + ω )(s + 1/τ )
2 2 0 2 4 6 8 10
wt
This can be shown to be as in Coughanowr and Koppel Figure 4. Normalized response of a first-order system to sinu-
(1965), after rearrangement, taking partial fractions and soidal signals of varying frequency.
376 Measuring System Behavior
A first-order system will, therefore, not alter the applied loading mechanisms. For example, a simple electrical low-
frequency but may attenuate the amplitude and introduce pass RC stage will have, in isolation, τ = RC . When the
phase shift rising, in a lagging sense, to a maximum of 90◦ , input source impedance and the output load impedances
the degree of these effects depending upon the time constant are connected, they each modify the effective value of
of the system and the frequency of operation in question. resistance because the load resistance is added in parallel
By way of an example that shows how these dynamic with C and the source resistance is added in series with R.
features can provide incorrect measurement action by a This situation has been described in Section 1, where
first-order sensor, consider a temperature sensor. The exam- the mercury-in-glass thermometer was considered; in that
ple is fully worked through in Coughanowr and Koppel example, the properties of the fluid can greatly vary the
(1965), so only the results need be given here. A ther- effective time constant experienced.
mometer with a time constant of 6 s (that is, for its use in It might be thought that the type of a stage can be
a defined application; it is not a feature of the thermome- recognized uniquely from the shape of the step response.
ter alone) is placed in a bath to detect bath temperature However, this is not necessarily a unique indicator because
variations of ±1 ◦ C amplitude that are sinusoidal at around systems of any order can, if suitably conditioned, exhibit
three cycles per minute. Because of its inherent frequency somewhat similar transient responses. This occurs when the
time constants of the respective order contributions are such
response, these cycles would be indicated to be half their
that the components producing them effectively work in
real magnitude, the sine wave output being of the same
phase. As the order rises, however, the shape of the curve
frequency but now delayed by some 4 s.
changes more noticeably enabling the nonconformity to a
A second example is found when considering the mecha-
first-order response to be more easily recognized.
nical pen recorder. When adjusted for critical damping, the
response of a plotter can be regarded as being essentially of
first-order character. If the recorder response is inadequate,
5 EXAMPLES OF FIRST-ORDER
it will plot signals at reduced amplitude. If the signal of one
channel is of lower frequency than the other but of the same MEASUREMENT SYSTEMS
amplitude, the recorder will not record a true indication of
the amplitudes and phase existing between the two signals The suitably thermally lagged thermometer has already
being recorded. been discussed. From that example, it is possible to rec-
For similar reasons, filters added to smooth out noisy ognize that a first-order system results when there exists a
signals (in any energy regime), may also lead to simi- significant energy storage mechanism and the inability of
that stored energy to discharge through a dissipative ele-
lar inaccuracies when more than one frequency component
ment in a relatively short time. The direct combination
exists in the forcing function, that is, any complex contin-
of these two system properties decides the all-important
uous signal. The response can be found by making use of
time constant.
superposition, which is applicable when the system is lin-
As the dissipating property (often called the loss) decides
ear. This allows each component to be evaluated separately
how well a response is damped (assuming, in such cases
for magnitude and phase values, the set then being added to
that a fixed level of energy storage capacity exists) this
obtain the total response. A worked example for a forcing is often called the damping property. The simplest form,
function with two components is given in Doebelin (2003). mathematically, is that in which the rate of damping is
Complex signals will not, therefore, give exactly the fixed independently of the rate of the signal. In mechanical
same normalized responses as are given in Figure 4, but systems, this is called Coulomb friction. Almost always,
they will always produce attenuation and phase shift as τ however, it will be found that the damping property in
and ω increase. It can be seen that a square wave, being a system is dependent upon influence variables, such as
a set of sine wave components, will produce a response temperature. Friction that is proportional to the velocity
in which each cycle follows a step response profile as the of the signal is termed Newtonian. Add to these two
square wave rises and falls, the starting point of each new the fact that the friction coefficient for static systems is
transition depending upon the level to which it rose, or fell, generally much higher than for those after breakaway has
in the previous cycle. occurred (nonlinear friction), it will be seen that practical
As has already been mentioned, the time constant used use of linear systems theory is a much-simplified version
must be the effective value of the combination present of reality. Although expressed in now deprecated units, the
when the stage is interfaced. It will not be the same value practical information on damping is contained in a classic
as that of the stage in isolation, unless the input source statement. Drysdale and Jolley (1924) is very relevant to
and output load added to the stage do not interact through fine mechanical systems.
First-order System Dynamics 377
In electrical systems, the obvious first-order example is RELATED ARTICLES
the RC combination, such as is met in a low-pass filter
stage. Cascading first-order stages raises the order of the
system producing a different result from that obtained by Article 37, Relationship Between Systems in the Time
the same components arranged to produce the same time and Frequency Domain, Volume 1; Article 38, Stability
constant as a single stage. Issues, Volume 1; Article 59, Introduction to the
In fluid systems, first-order stages occur in tanks of Dynamic Regime of Measurement Systems, Volume 1;
fluid being drained, the resistance to outward flow and the Article 60, Zero-order System Dynamics, Volume 1.
capacity of the tank forming the time constant pertinent
to level of the fluid. Chemical systems provide this order
action when solutes are mixed into solution. In acoustics,
an air enclosure having a baffled (resistance to airflow) REFERENCES
output port, as is used in some speaker enclosure designs,
is another fluid regime example. The same applies to a Atkinson, P. (1972) Feedback Control Theory for Engineers,
Heineman, London.
similar situation met in pneumatic instrumentation; much of
pneumatic equipment, however, is not linear in operation. Coughanowr, D.R. and Koppel, L.B. (1965) Process Systems
Analysis and Control, McGraw-Hill, New York.
First-order systems are also referred to as providing, or
being, an exponential lag stage. Doebelin, E.O. (2003) Measurement Systems: Application and
Design, 5th edn, McGraw-Hill, New York.
Further details of practical examples are to be found in
Atkinson (1972). Coughanowr and Koppel (1965), Olson Drysdale, C.V. and Jolley, A.C. (1924) Electrical Measuring
Instruments, Part 1, Ernest Benn, London.
(1943), and Shearer, Murphy and Richardson (1967). Other
works that may be found to be useful are Oppenheim et al. Hsu, A. and Hse, W. (1996) Schaum’s Outline of Signals and
Systems, McGraw-Hill.
(1996), Hsu and Hse (1996), Nise (2000).
Nise, N.S. (2000) Control Systems Engineering, Wiley, Chi-
chester.
6 SPECIFYING FIRST-ORDER SYSTEMS Olson, H.F. (1943) Dynamical Analogies, Van Nostrand, New
York.
First-order systems require the following features to be Oppenheim, A.V., Willsky, S.H., Nawab, A. and Nawad, H.
given when specifying their response: (1996) Signals and Systems, Prentice Hall, Upper Saddle River,
NJ.
• system type is of the first order Shearer, J.L., Murphy, A.T. and Richardson, H.H. (1967) Intro-
• amplification factor duction to Systems Dynamics, Addison-Wesley, Reading,
• time constant, usually in seconds. MA.
62: Second-order System Dynamics
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
Characteristics of a second-order stage will first be
1 Descriptive Equations for Second-order derived, followed by the study of its response to step, ramp,
System and an Example 378 and continuous sine-wave forcing functions.
2 Response of Second-order System to Step The second-order system is described by the nonhomo-
Function Input 379 geneous ordinary linear differential equation:
3 Response of Second-order System to Ramp
Function Input 380 d2 y dy
a2 + a1 + a0 y = x(t) (1)
4 Response of Second-order System to dt 2 dt
Sinusoidal Function Input (Frequency
Taking Laplace transforms for f (0) = 0 (where the system
Response) 381
is initially in the steady state) gives
5 Selectivity Issues for Second-order Systems 381
6 Specifying Second-order Systems 382 a2 s 2 Y (s) + a1 sY (s) + a0 Y (s) = X(s) (2)
Related Articles 383
References 383 Rearrangement into the transfer function form gives
Y (s) 1 1
G(s) = = (3)
1 DESCRIPTIVE EQUATIONS FOR X(s) a0 (a2 /a0 )s 2 + (a1 /a0 )s + 1
SECOND-ORDER SYSTEM AND AN For this expression, three parameters of practical implica-
EXAMPLE tion can be defined. They are
Second-order systems can arise when two energy-storage • K = 1/a0 transduction constant, the DC gain or con-
mechanisms exist in a system and act with some degree version constant.
of interaction. They must also be able to act with a 180◦ • ωn = (a0 /a2 )1/2 the angular natural frequency (with
phase shift between them, for example, a capacitor and an zero damping present); it has time dimensions.
inductor in an electrical system but not two inductors or two • ζ = a1 /2(a0 a2 )1/2 damping ratio, a dimensionless num-
capacitors. They need not be of the same energy type, for ber relating, in practice, the magnitude of actual damp-
energy conversion mechanisms make them mathematically ing to that at ζ = 1, called critical damping.
and practically compatible. Dissipative qualities are, again,
important as this is the factor that decides the damping These parameters occur as the result of interpretation of the
of responses. solution of the above equations that show that the system is
A most noticeable feature of second-order systems is that oscillatory in nature and that the level of damping present,
they can overshoot final values, even providing oscillation due to losses, controls the degree of oscillation. Being
(when sufficiently small damping exists). properties of the stage and not of the stage coupled to any
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Second-order System Dynamics 379
particular forcing function, they can be used as descriptors the quadratic. The damping ratio ζ is the dominant factor
of a second-order block. here. The roots are complex when ζ < 1, real and equal
The period of the natural angular frequency of the system when ζ = 1, and real when ζ > 1. These three regions
can be expressed instead of the angular frequency as its are termed underdamped or oscillatory, critically damped,
‘reciprocal’ and can be given the symbol τ . This is done and overdamped or nonoscillatory for the three regions
by some authors and is adopted here. It must be made clear respectively.
that τ is not used in the same context as it was for the first- It must be clear that the response range itself passes
order system nor is it exactly the period of the oscillating through a gradual gradation from the under to the over-
waveform (it is only for a zero damping situation). damped response. It is a need of the mathematical method
Using these alternative symbols enables the transfer func- of solution, not a property of the real physical system,
tion to be expressed either as that dictates that there are three different regions. The per-
formance characteristic is a smoothly varying effect, not
K having sudden changes of behavior as the equations might
G(s) = (4)
(1/ωn2 )s 2 + (2ςs/ωn ) + 1 suggest happens.
A family of response curves is presented, in a normalized
or as
manner, as in Figure 1; these show the response as the
K
G(s) = (5) damping ratio is altered from zero to around 5.0. Note
τ 2 s 2 + 2ςτ s + 1 that at large damping ratios, the response curve takes on
both forms of which are commonly used. a similar form to that of a first-order system, the two being
As the denominator of the transfer function is a quadratic slightly different in shape at the origin.
function, it can be factored into a form that shows that the It can be seen that second-order system response can
system could be formed by using two suitable first-order overshoot the final value, and if inadequately damped, it
systems placed in cascade. can continue to oscillate after a step is applied. It can be
Note that the natural frequency and damping ratio are seen that the second-order system, with zero damping, acts
interdependent, each also being functions of the transducer as a sinusoidal signal generator if the losses are truly zero
constant. Thus, gain changes and damping-ratio changes or are actually supplied to the system.
can alter the natural frequency. A step function is the most severe form of input-level
Changes of K matched by appropriate changes to the change that can be applied. In practical systems, this rate
damping ratio can retain a given natural frequency. of rise may be unrealistic, and in such cases, a step
The implication of these constants becomes clear when input would produce a response that is considerably more
the step response of such a system is studied. oscillatory than what reality might yield. The transient is
sometimes called ringing. A slower rate of rise, as would be
simulated by a ramp that reached a terminated level at given
2 RESPONSE OF SECOND-ORDER amplitude, would often be more reasonable to apply. This
SYSTEM TO STEP FUNCTION INPUT
Damping ratio, z = o
2.0
Forming the output expression in Laplace form from the
Laplace transforms of the transfer function given above and
the unit step function yields 1.6
0.25
1 1 1
Output, y (t )
Y (s) = (6) 1.2 0.50
a0 s (a2 /a0 )s 2 + (a1 /a0 )s + 1 Final 0.75
value 1.00
0.8
This reveals that the response will be one involving roots 1.50
of a quadratic. As the actual working is not needed to 2.00
0.4 2.50
understand the output response and because it is fully
presented elsewhere, such as in Coughanowr and Koppel 5.00
(1965), the method of solution only needs to be considered. 0
2 4 6 8 10
Forming the partial fractions and inverting the Laplace wn t
expressions gives the time-domain solution. The solutions
can only be obtained mathematically as three separate Figure 1. Normalized response of a second-order system to a
regions, these depending upon the nature of the roots of unit step function.
380 Measuring System Behavior
situation is worked through in Doebelin (2003) to show that system is a function of the damping ratio. As damping
with a terminated ramp forcing function, the underdamped increases, natural frequency falls according to
(ζ < 1) response can be made to be virtually error free if
ω
the natural frequency of the system is large compared with = 1 − ς2 (9)
the reciprocal rise-time duration of the ramp applied, this ωn
also applies for virtually zero damping. Doebelin makes the
point that this means that stages having little damping may As with first-order systems, although these factors appear
be satisfactory if they have high natural frequencies and are to define the system as a stand-alone unit, they are sub-
not excited in use by rapid rate of rise steps. The example ject to modification by the terminations used. For exam-
cited in such a case is the piezoelectric accelerometer, ple, the damping ratio that would be used when study-
when it is used to measure responses of relatively low-pass ing the response of a high-speed plotter would depend
mechanical functions. upon the damping added by the plotter to the system of
The general relevancy of the step response for many interest.
practical systems has given rise to several descriptive terms Impulse response is not covered here, the reader should
that relate to the underdamped responses given in Figure 1. refer to Coughanowr and Koppel (1965) and Doebelin
Overshoot is defined as the, less than unity, ratio of the (2003) for worked derivations and normalized response
magnitude to which the first overshoot rises over the final- curves that match those given in Figure 1. A some-
value line to the magnitude of the final value. Numerically, what similar oscillatory response, described in identical
this is given by terms, results.
Although the treatment has covered the response in a
−πς general manner, many texts include worked examples of
overshoot = exp (7) specific practical systems of mechanical, electrical, elec-
1 − ς2 tromechanical, acoustical, and pneumatic nature. In the
discipline of physics, it is covered most usually under the
and is, thus, decided only by the damping ratio. Note that
titles of wave motion and simple harmonic motion. It is also
the half magnitude of the overshoot cannot exceed the
dealt with in many texts on the mechanical-design aspect.
final value.
Closed-loop control systems often result in second-order
Decay ratio is defined as the, less than unity, ratio of
open-loop transfer functions and can sometimes be regarded
successive amplitudes of decay cycles. Mathematically,
more simply as a stage with this form of response, instead
it happens to be the square of the overshoot function
of needing to be seen as a closed-loop stage.
given above.
These two expressions enable the damping for a given
system to be calculated from a given record or the overshoot
to be estimated from the tail end of a record, as occurs when 3 RESPONSE OF SECOND-ORDER
phase delay loses the first part of a record in an oscilloscope.
SYSTEM TO RAMP FUNCTION INPUT
Their behavior with varying damping ratio is plotted
in Coughanowr and Koppel (1965).
Rise-time does not follow the same simple relationship The response of the second-order system to a ramp forcing
with the defined τ as it did in the first-order case. It is function is derived by the same procedure as was used for
quoted here as the time for the signal to rise from zero to that of a first-order system. It is worked through in Doebelin
the first crossing of the final value. (2003) and Atkinson (1972).
Settling time, defined in Article 61, First-order System The response is similar to that of a first-order system
Dynamics, Volume 1, applies with the same meaning to fed with a ramp, the exception being that now the system
these waveforms. It is also called response time. might oscillate about the final ramp line during the ini-
The period of oscillation, T , the time duration from one tial transient portion of the response curve. Again, there
peak to the next of the cyclic transient, is not given directly exists a step function component that delays the output
by τ when damping is present, but it is found from ramp by an amount decided by the damping ratio ζ and
natural angular frequency ωn . The resulting steady state
2πτ time lag and error are both reduced by increasing the nat-
T = (8)
1 − ς2 ural frequency and reducing the damping ratio. Doebelin
(2003) provides a normalized chart of how deviation error
the term τ being the period of the undamped second-order varies with varying damping ratio as the transient solution
system. As suggested earlier, the natural frequency of the dies away.
Second-order System Dynamics 381
4 RESPONSE OF SECOND-ORDER Because the system can produce output signals larger
SYSTEM TO SINUSOIDAL FUNCTION than what is provided to it, it has the ability to provide
signal gain or amplification; the output amplitude variable
INPUT (FREQUENCY RESPONSE) is, therefore, termed the magnification ratio.
Whether magnification is to be deliberately adopted or
The forcing function here is that used to establish the not depends upon the application. In some detection sys-
second-order frequency response. The methodology used tems, a second-order sensor is purposefully designed to ring
is that given in Article 61, First-order System Dynamics, (by virtue of absence of damping) at its natural frequency
Volume 1. A solution is needed for the Laplace form of the so that magnification is gained. In other systems, it may be
resultant function: desirable to keep the frequency response flat for as wide a
bandwidth as possible.
Aω K The degree of resonance can be described in terms of
Y (s) = (10)
s2 + ω τ s + 2ςτ s + 1
2 2 2 the Q-factor, a quality factor or figure of merit. It can be
shown (Shearer, Murphy and Richardson, 1967) to be
This involves solution of the roots of two quadratics.
This solution is worked through in several texts (Beckwith 1
Q= (15)
and Buck, 1969; Coughanowr and Koppel, 1965; Shearer, 2ς 1 − ξ 2
Murphy and Richardson, 1967). The resulting time-domain
parameters of phase and amplitude are obtained from the or, for low levels of damping, it approximates to
output expression:
1
AK Q (16)
Y (t) = sin (ωt + φ) 2ς
[1 − (ω/ωn )2 ]2 + (2ςω/ωn )2
(11) Alternatively, it is defined in more direct terms of the
where ‘peakiness’ of the resonance curve as the bandwidth at the
half-power points (where the power level has dropped to
2ς
φ = − tan−1 (12) half that at the peak), divided into the resonant frequency.
ω/ωn − ωn /ω
Both methods give the same result; direct measurements
on the amplitude–frequency response curve can be used to
or in the form where τ = 1/ωn ,
determine the damping ratio.
A word of caution is needed here. In many second-
AK
Y (t) = sin (ωt + φ) (13) order systems, resonances need time to develop when
[1 − (ωτ )2 ]2 + (2ςωτ )2 fed with a swept frequency forcing function. Response
curves swept too rapidly may, therefore, yield resonance
where peaks below their final value if given sufficient time
to build.
−1 2ςωτ
φ = − tan (14)
1 − (ωτ )2
These expressions show that the output is a sine wave 5 SELECTIVITY ISSUES FOR
of the same frequency as the input signal but that the SECOND-ORDER SYSTEMS
magnitude can now be greater, or lesser, as the frequency
rises. When ω = ωn , the amplitude can rise to a theoretical The factor Q is also a measure of the selectivity of the
infinity magnitude, the magnitude being decided by the system. This usage applies when the system is used to
damping ratio. deliberately detect signals near the resonant frequency, this
The phase component of the equations shows that it lags being achieved by attenuating all others with respect to that
from zero to a maximum while approaching, but never quite small bandwidth.
reaching, −180◦ if the damping ratio approaches zero. It Quotation of flat-response bandwidth is usually achieved
is −90◦ at the natural frequency, ωn . The expressions are by specifying the range of frequencies over which the
plotted in normalized form as the two plots of Figure 2. In response remains within the maximum and the half-power
this regard, second-order systems are quite different to first- (3 dB) points. From Figure 2, it can be seen that the widest
order stages. These curves are plotted in a particularly accu- flat response is obtained with a damping factor of around
rate manner in Shearer, Murphy and Richardson (1967). 0.5. By comparison with the frequency response curve
382 Measuring System Behavior
10 0
Damping
ratio, z 0.1
0.2
0.4
1.0 0.6
0.8
1
2
3
5
10
Y (t )
20
0.1
0.01
0.001
0.1 Natural 1 10 80
frequency w /wn (or wt)
(a)
0
0.1 Damping ratio, z
0.3
−30 0.5
10.7
2
−60 53
Phase (deg)
97
10
20
−90
20
910
−120 3
57
2
1
−150 0.7
0.5
0.3
0.1
−180
0.01 0.1 Natural 1 10 100
frequency
w /wn (or wt)
(b)
Figure 2. Normalized response of second-order system to sinusoidal excitation: (a) amplitude; (b) phase.
of the first-order system Article 61, First-order System of the signal is to be preserved, then each frequency must
Dynamics, Volume 1, it can be seen that a second-order be transmitted through the stage without attenuation or
system will provide a flat response of similar bandwidth phase shift.
but the first-order system will not fall off as rapidly.
Additionally, the phase shift of each might be important.
At the 3-dB point of each, a first-order system introduces 6 SPECIFYING SECOND-ORDER
approximately −60◦ shift, while the second-order stage SYSTEMS
introduces some −110◦ of lag.
Complex signals fed to a second-order stage can be Second-order systems require the following features to be
treated as the sum of sinusoids. If the original character given when specifying their response:
Second-order System Dynamics 383
• system type is of the second order REFERENCES
• magnification ratio at DC
• damping ratio Atkinson, P. (1972) Feedback Control Theory for Engineers,
• resonant frequency. Heineman, London.
Beckwith, T.G. and Buck, N.L. (1969) Mechanical Measure-
Other sources of detail on second-order systems are ments, Addison-Wesley, Reading, MA.
Oppenheim et al. (1996), Hsu and Hse (1996), and Coughanowr, D.R. and Koppel, L.B. (1965) Process Systems
Nise (2000). Analysis and Control, McGraw-Hill, New York.
Doebelin, E.O. (2003) Measurement Systems: Application and
Design, 5th edn, McGraw-Hill, New York.
Hsu, A. and Hse, W. (1996) Schaum’s Outline of Signals and
RELATED ARTICLES Systems, McGraw-Hill.
Nise, N.S. (2000) Control Systems Engineering, Wiley, Chi-
Article 37, Relationship Between Systems in the Time chester.
and Frequency Domain, Volume 1; Article 38, Stability Oppenheim, A.V., Willsky, S.H., Nawab, A. and Nawad, H.
Issues, Volume 1; Article 59, Introduction to the (1996) Signals and Systems, Prentice Hall, Upper Saddle River,
Dynamic Regime of Measurement Systems, Volume 1; NJ.
Article 60, Zero-order System Dynamics, Volume 1; Shearer, J.L., Murphy, A.T. and Richardson, H.H. (1967) Intro-
Article 61, First-order System Dynamics, Volume 1. duction to Systems Dynamics, Addison-Wesley, Reading, MA.
63: Outline of Systems Thinking
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
Systems thinking is a mental attitude to problem solving
1 Definition of Systems Thinking 389 that intuitively assists one allow for issues of the ‘whole’ –
2 Reductionism Versus Postmodernist Thinking 392 this is the holistic mode of consideration.
3 Soft Systems Thinking 392 Its implementation is done by application of Systems
Engineering (SE) principles that, in turn, leads to the
4 Hierarchy of Systems 393
detailed engineering design that is well developed in aca-
5 Introduction to the 5-layer Systems demic curriculum; systems thinking and systems engineer-
Engineering Model 395 ing are seldom developed in an overt didactic manner. The
6 Life Cycle Concept 395 material given here is extracted from Sydenham (2004)
7 Emergent Properties 396 where the issues of modern engineering design are covered
8 Systems of Systems 396 in detail.
9 Application of Systems Thinking to A simple definition of Systems Engineering starts devel-
Measurement Systems Design 397 opment of this thinking. QinetiQ, UK, staff has developed
Related Articles 398 this conveniently short description of SE:
References 398 ‘A set of activities which control the overall design, imple-
mentation and integration of a complex set of interacting
components or systems to meet the needs of all users.’
1 DEFINITION OF SYSTEMS THINKING
This clearly recognizes that engineering design tasks
Too often designs for a system are generated from the have to include numerous interacting issues (often not
‘bottom up’; the core need is hastily matched against an properly allowed for in engineering design) in order to
apparently suitable technical solution that appears to fit obtain a sound solution to the customer’s needs.
the need. As the detailed design progresses, it becomes SE makes use of a set of principles and processes
clear that other factors should have been allowed for and that efficiently apply resources to optimize a development
that it is by then getting expensive and time consuming to project’s progress toward a sound solution of a customer’s
change things. product or service need. The definition of SE also high-
On the other hand, the more aware and experienced lights the fact that it deals with more than the physical
designer will carry out the design from the ‘top down’ energy/mass relationships that are covered by detailed engi-
finding appropriate solutions much later when the full neering work.
requirements have been established. This article and Arti- Application of Systems Engineering is not a case of
cle 64, Executing A Measuring System Design, Vol- applying a set of rules, but more a way of thinking that
ume 2 provide the background information to become a extends conventional engineering design practice.
‘top down’ designer, who rarely needs to make drastic Where the memory requirement needed to track the
redesign changes. many issues that arise exceeds one person’s ‘brain-full’
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
390 Measuring System Design
Customer’s need is the input
Knowledge and data
Needs statement constantly flows
between teams
SE brain(s) in
overall control of
all teams activity
Design teams
Leader for each team
Real-time interfaces,
many and vital between
all teams Mission:
Teams work in
unison toward
Output is customer’s need fastest, right,
solution
Figure 1. Teaming in an engineering development.
it becomes necessary to employ commonly understood In the seventeenth century, Descartes suggested that
and agreed recorded processes and activities. This helps problems are handled by successively breaking down a
ensure each person designs the ‘right’ thing as their part of problem until a level is reached where sufficient understand-
the whole. ing exists. This paradigm for problem solving is known
Application of systems engineering is about deciding, for as ‘reductionism’. It is the basic thinking methodology of
the technical aspects of a project, what should be done by science and engineering and has been applied with great
whom, and by what time? effectiveness.
This kind of management task differs from that of office, Taking this reduction idea a step further suggests that a
corporate, or project management; however, the distinctions system rebuilt from a set of subsystem solutions must be
are not always black and white. Figure 1 provides a simple a sound overall solution. This is, however, not necessarily
model of the human teaming aspect of the engineering of a so. Small deviations in subsystem solutions can, all too
systems task; it represents the overall situation taking place easily, propagate upward leading to major errors in the final
in a development. performance.
All teams must be efficient in their duty in order that Difficulty in not meeting requirements is also due to the
the whole set of teams delivers the ‘best practice’ design fact that the traditional limited engineering detail viewpoint
needed. Note that numerous interactions will take place often cannot cope with the complexity of real systems.
between the teams as the project outcomes pass from ‘cus- Reductionism needs to be supplemented with other kinds
tomer need in’ to ‘satisfactory system out’. An individual of thinking.
must work as a team player; a design team leader is needed In the reductionism design approach, a closed model of
to help the process along. the design situation has to be realized to complete and close
Applying SE to the measuring system design task is a the design boundaries. Care is needed in setting up these
matter of being able to recognize what kind of activity is boundaries – see Article 68, Setting the System Bound-
appropriate to be done at any particular time. This ability aries, Volume 2 and Article 64, Executing A Measuring
is developed by reading, taking courses, working with System Design, Volume 2.
experienced practitioners, and using every opportunity to Many engineering design situations contain issues that do
bring fresh and better solutions to design situations. not lend themselves to reductionism thinking. The ability to
Underpinning the professionalism of SE is an apprecia- recognize the nature and scope of these limiting parameters
tion of Systems Thinking. An appreciation of how engineers needs skill in design team operations. Seeking to force
and physical scientists go about their thinking is needed at reductionist ideas to fit where they are not appropriate can
this point. lead to project failure.
Outline of Systems Thinking 391
Systems Thinking includes attention to • Systems consist of hierarchies that relate to each other
through numerous interfaces, each having their own
• human activity systems, not just the inanimate physical kind of requirement.
objects that make up the whole; • All parts of the whole are interconnected (interface is
• operational readiness and suitability, that is, will it do an alternative term) to a varying degree, some being
the job when called upon and will it continue to perform very dominant and thus having greater influence on the
its task for as long as expected? behavior of the whole.
• systems of all levels and types.
The various methods and parts of Systems Thinking are
Key statements have been published on Systems Think- illustrated in Figure 2.
ing by engineers involved in major projects; Hitchins (1992) A model of the layers of system openness starts with the
is worth visiting for its mind-opening views. outer total shell that includes everything thought to be of
Key tenets of Systems Thinking are as follows: relevance to the problem that the diagram represents.
Inside this layer is placed the study of how the different
• It is concerned with the concept of ‘whole’ and its systems viewpoints are expressed. This has two thinking
properties. aspects – philosophical systemic thinking that is often hard
• It needs systemic thinking (i.e. including all of the to apply, and the various pragmatic working areas that
issues) as well as systematic thinking (i.e. being the various kinds of thinking need to advance their prob-
methodical in tackling the problems). lem solving.
Area: Philosophy studies:
Sources of Area: Natural/life sciences: Ontology, epistemology, semiotics,
Area: Social sciences:
knowledge: Biology, cognitive science, cognitive language, nature of enquiry, logic
Humanities, Economics, anthropology,
psychology, neural nets, genetics, and truth, nature of knowledge
Soft Sciences organizational science
evolution and wisdom
Area: Theoretical
development of
systems thinking
Cybernetics, control theory, hierarchy Area: Soft systems
theory, system architecting, information Soft systems methodology
theory, systems of systems
Area: Problem-solving development of
Area: Study of systems thinking applied to real-world
systems ideas as problems
in general
Area: Support for
Area: Hard systems:
decision-making:
SE methodology, Systems analysis, operations
computer systems, Research, management
artificial intelligence Science
Area: The systems movement in general
Area: Applications of systems
thinking in other disciplines
Sources of
knowledge:
Engineering & Area: Engineering and hard sciences:
Sciences
Hard disciplines Mechanical, electrical, chemical, etc
Physics, chemistry, math, etc.
Figure 2. Representation of the relationships of the key disciplines.
392 Measuring System Design
Engineering can be seen there to make use of all of science disciplines. Their phenomenological approach –
the domains shown in Figure 2 with the exception that often called the postmodernism approach – is not as
the application of the soft kind of systems is not well precise as reductionism but it can make progress in soft
developed – see Section 3. science situations.
2 REDUCTIONISM VERSUS 3 SOFT SYSTEMS THINKING
POSTMODERNIST THINKING
Figure 3 shows the flow of activities for finding and imple-
Reductionism has served engineering and science well, but menting a solution to a system need in a ‘soft’ situation.
it cannot provide all of the solutions to systems design – The SSM process begins with the problem being identi-
see Section 4. Design engineers overly seek to use the fied as unclear and lacking an obvious reductionist solution.
reductionist approach because it is the main method in their The need is then expressed in writing as that helps clarify
design toolbox. They are not as well equipped to find sound the situation.
solutions for systems involving human activity. Purposeful activities are identified and conceptual models
In the 1970s, an engineer in charge of major engineering generated. These are compared with the starting point of the
projects in the United Kingdom was dissatisfied with the real situation. Extreme differences are rectified in the model
fact that the reductionist engineering approaches were not and the implementation adjusted to make it acceptable to
working well enough. He developed a methodology for the people involved in the change.
these situations that is now known as the Soft Systems This process is repeated until the best available plan
Methodology (SSM), Checkland (1981), Checkland and results. Considerable iteration may be needed to reach an
Howell (1998). SSM recognizes that not all disciplines acceptable situation.
make use of reductionist thinking. Success cannot be assured with the soft system method-
In the humanities, it has long been recognized that the ology, but it is one of the few ways known to seek design
human machine has two aspects. The physically existing solutions for problematic system designs. Here the solu-
‘body’ part can be explained through the laws of physics. tion solver ‘gets inside the problem’ making a considered
The cognitive ‘mind’ part, however, cannot – yet they form intervention to see what happens.
the same whole. There is currently no meeting of the In sharp contrast, the hard engineering solution path
minds on this dual body problem; it continues to provoke develops a separated model of the whole through metaphor-
philosophical debate. ically dissecting that which identifies its subsystem. This
This has led to the humanities disciplines solving set of component subsystems, and their interrelationships,
problems differently to how they are tackled in the hard are then simulated using a computer to investigate the
Real world
Problem situation Action to
considered improve the
problematic problem Changes:
situation systematically
desirable culturally
feasible
Comparison of
Problem situation the models
expressed and the real
world
Root definitions of Conceptual models of
relevant purposeful the systems named in
activity systems the root definitions
System thinking
about the real world
Figure 3. Flow of activities in finding and implementing a solution in a soft system situation.
Outline of Systems Thinking 393
sensitivities of the various critical issues. Optimization Factors Integration, Volume 2; and Article 97, Design
methods are applied to facilitate beneficial changes to Methodology, Volume 2.
the model.
When the right model and its parameters have been
realized, the real system is rebuilt to form a new system; 4 HIERARCHY OF SYSTEMS
altered parts are put back into place to see if the whole
works as intended. Representing the whole of all systems activities and rela-
Humans cannot be adequately understood as technical tionships would be a massive task; there are too many issues
machines alone. People within systems can be slow and and variations to cover. Instead, a collection of thought pro-
reluctant to respond to the direct process of change. voking models is used that develop insight into aspects of
Changing organizational cultures can take a long time the whole.
to achieve. It has been summed up as trying to ‘herd Three different models are now given, each revealing
cats’. different characteristics of the same generalized whole.
It is often, however, the engineers trained in reductionism The first relates the groups of people involved, the
who are expected to develop solutions for systems with sciences, and their thinking, Figure 4.
which they are not well versed, and for which there is Three key kinds of interrelated activities are shown – the
often low chance of clear-cut success, as measured in hard natural world, the sociopolitical system, and engineers and
science ways. scientists at work. The needs of all three must be met for a
The design team leader needs to recognize the kind of system design to be successful. Engineering, until recently,
system class in which the design team is involved and set largely neglected the other two; today sociopolitical and
up appropriate team membership. natural world aspects must be taken into account. Installing
When designing measuring systems, any human element a measuring/detection system can sometimes seriously
present needs careful treatment to ensure it will operate impact people’s perceptions producing disastrous outcomes,
properly. This arises when generating the needs statement as did the ‘Star Wars’ missile protection system program.
in order to establish what should be measured, and for Each of the three regimes is represented by a triangle.
what purpose. This stage – see Article 64, Executing A At the base of each triangle sits the scientific, formal,
Measuring System Design, Volume 2 – usually involves quantitative thinking workers. Moving up each triangle, the
thinking in the soft regime as customers develop their thinking style used changes from essentially quantitative to
ideas. The operational regime of measuring systems – almost entirely qualitative, for at that level it is taking in
the knobs, dials, panel layouts, handling properties, people’s feelings and emotions.
colors, and emitted sounds – all involve soft thinking Engineering and science disciplines operate best in the
as the human factors – see Article 94, Human Factors lower areas.
Engineering, Volume 2; Article 95, Human–Machine The middle ground is where the use of Systems Engi-
Interface, Volume 2; Article 96, The Domains of Human neering finds effective application. At the top, all manner of
Commerce politics
Natural systems
Management
Uncertainty increasing
Fringe
sciences
Sociopolitical
systems Systems
Soft sciences engineering
Engineering systems Conventional
detail
Hard scientific disciplines engineering
Number of persons involved
Figure 4. ‘People and Science’ model of a systems activity.
394 Measuring System Design
Environment external to company:
It greatly influences how an enterprise operates and its impact on projects and
hence, downward to design team operations
Full enterprise environment in which all projects of enterprise sit:
The established overall SE methods, etc. impact on how project, and hence
design team, operates
Your Centralized
Support for
design enterprise
your
team support for
project:
1 all projects
2
3
Project in which there are many design teams
Figure 5. Environment layers in which a design team works.
often inexplicable decision-making takes place, not because
of lack of skills but for lack of any formalized way to do Metaphysical world of thought and belief systems 7
it better. The natural world under its own control
Increasing disagreement of solution form
The width of the triangle at any given level crudely All systems in which man is involved
indicates the number of people involved. It is interesting in change 6
to note that as little as one person at the top can decide 5
how the many people below use their resources and skills;
4
large groups of designers are involved in taking the ideas of Social
systems
a few to fruition. A designer generally has little influence
over the top-level needs and has to work within given Systems designed
3 in the abstract–
requirements. includes those
The second model, Figure 5, shows how the design 2 Physically concerned with
team works within a multilayered set of quite different made systems human behavior
1
–reality!
environments. For overall success, a project must make
allowances for the nature of the limitations and controlling Increasing uncertainty on solution form
factors that exist for the type of enterprise in which the
design team works. These issues vary greatly. For example, Figure 6. Diagrammatic representation of a hierarchical classifi-
cation of systems.
a private organization does not have to disclose as much
information about its processes to the public as does a
issues that exist versus the degree of uncertainty of their
government institution. To get on top of many problematic
characteristics.
issues, it pays to appreciate the higher layer affairs that are
The types of systems shown in the diagram are
impacting on a design’s situation.
The third model given in Figure 6, after Boulding (1956), 1. straightforward technical design tasks that inherit con-
assists appreciation of the classes of types of system design siderable know-how and have low risk in execution if
that can arise. It helps a designer identify what kind of done by experienced experts;
difficulties the design team might expect to encounter. (Examples: simple road bridge; electronic amplifier
This model is based on mapping the various kinds of board; weighing platform)
systems that exist onto a modeling space represented by 2. technical tasks with a modest degree of design change,
two variables – the degree of disagreement on systems thus including a clear degree of risk;
Outline of Systems Thinking 395
(Examples: original network for 3G mobile phones; It is important to be able to recognize the class of system
major automobile model change with advancing func- in which a design task is working; the surrounding climate
tionality such as moving to all-wheel drive from dual of thinking can make a large impact on progress and on the
wheel drive, measuring instrument using novel princi- type of solutions that will be accepted.
ple of detection)
3. engineering systems involving considerable human
control and intervention in their operation, but not so 5 INTRODUCTION TO THE 5-LAYER
much in the overall organization; SYSTEMS ENGINEERING MODEL
(Examples: production line manufacturing systems
instrumentation; transport systems) The various holistic levels of a project task have been
4. systems where their major subsystems components identified by Hitchins and are reproduced here from his
are associated heavily with human organization; here Web site www.hitchins.net, where it is free to download.
engineering risk issues are low compared with the risks Overall human system endeavors fit into levels as
of understanding the human behavioral aspects; follows:-
(Examples: building detection and evacuation systems;
flight control systems; educational support systems) Level 5 : Socioeconomic Engineering, what constitutes reg-
5. systems where human attitude is dominant and largely ulation and government control at national and interna-
unpredictable; tional operation.
(Examples: change management taking place in a fac- Level 4 : Industrial Systems Engineering, or engineer-
tory; speed control on roads; engendering profession- ing of complete supply chains/circles. Many industries
alism. Improving the quality of the design of measur- make a socioeconomic system. A global wealth cre-
ing systems) ation philosophy.
6. systems that are as complex as man thinks he can build Level 3 : Business Systems Engineering – many businesses
and so often attempts to make; make an industry. At this level, systems engineering
(Examples: fighting wars, peace making, societal pol- seeks to optimize performance somewhat independent of
icy systems. Government’s economic measurement other businesses.
models.) Level 2 : Project or System Level – many projects make
7. systems that can only yet be represented by abstrac- a business. Western engineer managers operate at this
tion in the thinking world of science fiction and the- level, principally making complex artifacts.
ology. (Examples: Utopian worlds; Godlike abilities Level 1 : Product Level – many products make a system.
of design.) The tangible artifact level. Many engineers and their
institutions consider this to be the only ‘real’ systems
As the risks rise and the systems nature becomes engineering
more problematic, it becomes increasingly impossible to
be certain about numerous critical systems issues. Those The design of measuring systems is most likely to fit into
involved are increasingly unable to agree on what kind of Levels 1 and 2. Detail of the factors of each of the layers,
solution to use. Without clear definition of the critical issues provided on the Hitchins Web site, provide appreciation of
it is problematic for a sound engineering solution to be the kind of problems, and thus the boundary limits that a
achieved – see Article 3, Measures and Metrics; Their particular development needs to consider.
Application, Volume 1.
Most commercial projects sit in the lower two classes of
1 and 2 because they tend to exploit proven technologies 6 LIFE CYCLE CONCEPT
and because they need to work in relatively low-risk areas.
The measuring system engineering detail design team The popular way to provide a representational foundation
is usually working, by necessity of delivering a reliable for systems development uses the system life cycle model.
and low-risk outcome, in the high-certainty and low- Development and use of all product or service systems fol-
disagreement area with respect to their design solutions. low the same generic sequence of life cycle activity stages.
However, they can sometimes get involved in the exe- While various specific expressions of this life cycle exist
cution of tasks in many of the classes shown. As the they follow the generic one illustrated in Figure 7. The
magnitude of the position number increases, the detail engi- phases (stages) are the following:
neering design component becomes of lesser importance
compared with execution of the whole as it, in itself, is • conceptual formation of ways in which the task might
unable to provide solutions to the problems. be solved;
396 Measuring System Design
• Systems integrators
• Sub-systems
• Process paradigm • Operators
design/build
• Support tools with little
• Traditional design
• Tightened specs engineering knowledge
• Requirements analysis disciplines
• Complex techno-mgmt
• Requesters • Repeated
• Risk earlier stages
• Politics • Build to Relative
• Requirements design numbers of
• Assessment people
involved
Concept Feasibility Detailed Manufacture Use Upgrade Disposal
formation assessment design
Time (years)
Figure 7. Life cycle stages of a product or service development.
• feasibility assessment of selected, apparently sound, They can exert a great influence on the other systems with
design ideas; which they interface. Unexpected, nonbeneficial, emergent
• detailed design of the component parts of the cho- properties can become very apparent once the system is
sen design; nearing, or after being put into service.
• manufacture of the real system based on detailed For example, it might have been decided to use a
engineering design, using physical or digital model microminiature wireless telemetry system to communicate
prototypes to prove the design; temperature data from inside the flying suit of a pilot,
• use in the application for which the system was only to find after commissioning that it causes the flight
designed; navigation system to be inaccurate. Today, this is an
• upgraded for further use with improved or different obvious design factor to expect, but that was not always
requirements; the case or we would not have to turn off mobile phones
• disposal after use, to follow regulations, and so on. in hospitals and in aircraft.
Good design and effective application of systems think-
ing are key issues within the tasks of all of the stages.
Each is not a stand-alone stage and all need some level of 8 SYSTEMS OF SYSTEMS
consideration from commencement of a project. To appre-
ciate the differences consult Article 66, Phases of System As man has learned how better to organize and design
Life Cycle, Volume 2 and Article 65, Life Cycle Concept, technical systems they have grown in size to an enormous
Volume 2. extent. Many of today’s major man-made systems have
evolved by progressively combining systems to form very
large conglomerates.
7 EMERGENT PROPERTIES For example, the first railway system in Britain created
emergent requirements for engine water and fuel supply
Parts of the whole will have their own important ‘emergent depots, spare parts provision, hotels en route, signaling,
properties’. These are key performance parameters that track rules, special tools for making parts, training schemes,
may, or often may not, have been expected as the result and so on. What might have been overlooked as it was
of implementation. developed was the effect it would have on the existing water
Outline of Systems Thinking 397
canal transport system that it replaced, and the impact it • Measuring systems are unlikely to form into SoS
would have on the nature of the industrial revolution taking units in their own right but their common nature
place at that time. and universal methodologies will be used across those
In recent years, such complex and extensive systems were systems comprising an SoS assembly.
seen to be too large to be considered as suited to the usual
methods of management and design. The name Systems of
Systems (SoS) has been coined for such systems. 9 APPLICATION OF SYSTEMS
A main driver for SoS developments has been in defense
systems. These evolved as the following: THINKING TO MEASUREMENT
SYSTEMS DESIGN
• First there existed the personal weapon system.
• Then came team use of weapons combining the vari- Applying the systems thinking culture is a case of ‘being a
ous forms of firepower with behind the lines support specialist at being a generalist’.
logistics and intelligence inputs. When a design team is part of a large systems provider
• This was followed by many kinds of platforms com- house, a person will usually be appointed who is responsible
bined with the necessary command and control needed for organizing the systems support for all teams. In such sit-
in a campaign structure. uations, teams usually will be provided with the following:
The sophistication and number of cooperating systems
has continuously increased in defense, civilian commercial, • a company specific SE Process Manual;
and government systems, and in the search for solutions to • a computer-based support tool system and support
societal and humanitarian problems. tutors;
It became obvious that the former paradigm of first • special development facilities as are needed according
building general utility platforms (the ship, the airframe, the to the project;
armored vehicle, etc.) on which are then mounted control • sound archiving and communications for design records
and command systems, weapons, and other systems, all and the configuration management system;
as separate entities, was inadequate. It needed the SoS • safety control process;
approach. Similar thinking is essential in such systems as • design controls;
civil aircraft control systems that now span countries, and • sources of advice and mentoring for junior staff;
in integrated power grid operations. • in-house training and more.
So, what are the differences between Systems Thinking
and SoS thinking? This is a matter of degree. SoS is an As a guide, design team leaders need to be familiar
extension of general holistic considerations and has the with the basics of systems thinking and the culture of
following characteristics: systems engineering. They need to have on hand copies of
foundational books on SE practice such as Blanchard and
• High complexity comprising relatively independent sys- Fabrycky (1998), Sage and Rouse (1999), Sydenham (2004)
tems that can each be regarded as a sophisticated system and they need to refer to these for concepts and methods
in its own right to employ as project problems arise. A sound source of SE
• Continuously evolving as the emergent properties of knowledge is found in the general pages, and also in those
each system interact of the technical committees, of the International Council on
• No obvious start or end-point goals for their existence Systems Engineering, INCOSE (2004).
• Parts are often geographically distributed The US DOD Military Handbook on Systems Engi-
• Viewed as a set of interacting, separate, systems neering MIL-STD-499B, USDoD (2004), although being
• Component systems retain much of their indepen- comparatively old as SE standards go, is still an excellent
dence, pursue their own goals, and have indepen- source of ideas for use at the various stages of the life cycle.
dent management Other works giving various views of Systems Engineering
• Systems Engineering activity is dispersed and loosely are Stevens et al. (1998), Buede (2000), Westerman (2001),
controlled Faulconbridge and Ryan (2003).
• Understanding the behavior of constituent systems An issue that will arise in the design of measuring
needs transdisciplinary (each is learning from the other) systems is how much of the SE process should be carried
approaches, not just multidisciplinary approaches (each out when executing the systems engineering for a project.
does its own thing usually with an insufficient number Some design tasks are so small as not to need any ordered
of disciplines) process. Some, like that for flight-testing an aircraft where
398 Measuring System Design
over 100 000 test points might be set up, will need to adopt Setting the System Boundaries, Volume 2; Article 69,
the complete best practices of SE. Requirements Allocation, Volume 2.
Overhead costs of an SE support process are not always
easy to justify because the benefits of its use are abstract in
nature, cover long-term issues, and appear not to produce REFERENCES
easily measured value-adding components to a project. For
this reason, SE activities are often seen as costly luxuries Blanchard, S.B. and Fabrycky, W.J. (1998) Systems Engineering
by those who control the project budget. and Analysis, 3rd edn, Prentice Hall, Upper Saddle River, NJ.
They are, however, as important to project success as Boulding, K.E. (1956) General Systems Theory – The Skeleton
accounting and management activities in that they assist in of Science. Management Science, 2(3).
Buede, D.M. (2000) The Engineering Design of Systems: Models
• controlling the final performance obtained;
and Methods, John Wiley, New York.
• providing early detection and control of design error.
Checkland, P. (1981) Systems Thinking, Systems Practice, Wiley,
Who in their right mind would make a development Chichester.
journey without sound assurances that all design work is Checkland, P. and Howell, S. (1998) Information, Systems and
on track and that the design is moving forward in the Information Systems, Wiley, Chichester.
right direction? Faulconbridge, I. and Ryan, M. (2003) Managing Complex Tech-
The design team leader has to use personal judgment nical Projects: A Systems Engineering Approach, Artech House,
in setting the scale of use when applying the techniques Norwood, MA.
or support mechanisms of SE. A single team comprising Hitchins, D.K. (1992) Putting Systems to Work, Wiley, Chichester,
a dozen or so staff working in a start-up company will Free download version is available from www.hitchins.org/prof,
probably not be able to devote the time to writing a April, 2002.
dedicated SE manual. They might instead simply align with INCOSE. (2004) International Council on Systems Engineering,
an SE standard. The team leader in that case, however, still www.incose.org.
needs to apply SE principles as part of routine technical Sage, A.G. and Rouse, W.B. (eds) (1999) Handbook of Systems
management. Article Article 64, Executing A Measuring Engineering and Management, (ed. A. Sage), Wiley, New York.
System Design, Volume 2 moves this discussion on to Stevens, R., Brook, P., Jackson, K. and Arnold, S. (1998) Systems
show how a project should progress. Engineering: Coping with Complexity, Prentice Hall PTR.
Sydenham, P.H. (2004) Systems Approach to Engineering Design,
Artech House, Boston, MA.
RELATED ARTICLES USDoD MIL documents, (2004) http://astimage.daps.dla.mil/
quicksearch/.
Article 65, Life Cycle Concept, Volume 2; Article 66, Westerman, H.R. (2001) Systems Engineering Principles and
Phases of System Life Cycle, Volume 2; Article 68, Practice, Artech House, Norwood, MA.
64: Executing A Measuring System Design
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
• manufacture of the measuring system
1 Understanding the Whole 399 • installation and commissioning of the system
2 Closed-systems Design 400 • operation by obtaining the desired measurements
3 The Open-system Challenge 401 • removal from service when useful life has expired.
4 Specifying the Need 402 The first step is to obtain a sound briefing of the need
5 Requirements Generation 404 from the customer or their representative. Sounds easy!
6 Conversion into a System Design 406 Customers are not always experienced enough in stating
7 Partitioning 406 a need in terms that clearly lead to design parameters.
8 Interface Design 407 Do not expect a swift resolution of this issue. Usually it
requires a lot of patience as both the designer and the
9 Test and Evaluation 408
customer undergo a joint learning process. Development
10 Applying SE to a Project 409 of the requirement is a journey of discovery; more on
Related Articles 409 that later.
References 410 It is very important to glean as much as possible about
the need and its influencing factors as the starting point
to success.
1 UNDERSTANDING THE WHOLE Issues to be explored are
Having introduced system thinking in Article 63, Outline • the overall statement of customer need, in customer’s
of Systems Thinking, Volume 2, it will be clear that a terms;
sound design of a measuring system begins with a general • purposeful requirement – what capability is to be ful-
appreciation of many system issues before a commitment filled;
is made to firm design parameters. • operational requirements – how well, and in what cir-
Obviously, the main aim of a design exercise is to cumstances it is to be operated;
produce a system that satisfies the need. Getting to that • the extent of the limits of influencing issues in the
state needs systematic activities that drill down from the technical, social, and legal arenas;
top of the situation. • the key critical issues that were introduced in Article 3,
The whole process will pass through several life-cycle Measures and Metrics; Their Application, Volume 1;
stages such as • personnel that will operate and maintain the system;
• decide when it is needed.
• formation of some suitable solution concepts
• assessment of two or three likely candidate concepts It is necessary to appreciate the role of the various
• detailed design of the chosen concept that has become types of players in the development. Figure 1 shows their
more defined by now relationships as the project progresses.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
400 Measuring System Design
System designer Subsystem and
System end user
and manufacturer component vendors
Designers
Functional Contractor Vendors
performance,
technical Purchase System design
specification specifications
Items to
Purchaser vendors
Operational $ specs
requirements, System & hardware
specifications Payments design
Start $ Special vendor
Payments OEM items as
per designer’s
Operational System Manufacture and calls
system delivered factory test Purchased
items
Install, commission, OEM
Users acceptance test designers
End Public
Figure 1. Role of the various players in a system development.
Each player – customers, users, designers, contractors, This does not need to be that complex to start with –
vendors and the public – has different reasons for partici- include just enough detail to portray the overall situation.
pating. Their incentives are different, and the designer needs Figure 2 shows such a model for a measuring system being
to appreciate these differences, Sydenham (2004). set up for monitoring an oil refinery system.
Just how the system will finally look is still to be
developed. For example, the sensors may well be connected
2 CLOSED-SYSTEMS DESIGN on a circular digital bus instead of the star configuration
shown. At this stage, do not rush to start the detailed
Begin with a general assessment of the type of design design; the specifications need to be well developed before
situation, using the tenets of Systems Thinking given in that commences.
Article 63, Outline of Systems Thinking, Volume 2. Ask A good way to develop a concept is to develop a Concept
the following questions and reflect on the findings. of Operations (ConOps) – see Article 67, Principle of
Concept of Operations (ConOps), Volume 2. This is a
• Is it going to easily lead to a closed-system design? case of working through the things that need to be done to
• What is the level of system in the layered Boulding achieve the right outcomes for the system in use. Figure 3
model given in Article 63, Outline of Systems Think- shows one developed by Ring (2000) for a program to train
ing, Volume 2? systems thinking design engineers.
• Does the degree of human activity suggest a reduction-
While this particular ConOps is developed for a system
ist approach that will need the use of a Soft System
in which humans are the subject, the same rules of devel-
Methodology (SSM), as described in Article 63, Out-
opment apply for purely physical or mixed systems; apply
line of Systems Thinking, Volume 2?
the main rule – what is it that happens to the flowing entity
• What is the impact of the characteristics of each life-
as it migrates from an input to an output?
cycle stage? For example, is there a serious disposal
At some stage, the overall system architecture becomes
problem?
• What is the useful lifetime likely to be and will it need clear enough to define the boundary limits of the whole.
upgrading? This can depict the extent to which influencing effects
will impact the system, for good or bad. Figure 4 shows
After obtaining an appreciation of these issues, start such a diagram set up for a temperature measuring system
building an overall contextual systems model of the sit- operating in a high hazard situation. The influence param-
uation. This will help clarify matters and provide material eters that will heavily impact the actual sensor system are
for briefing others who are associated with the project. defined.
Executing A Measuring System Design 401
Communications
and control center
Housing
and offices
Wellhead Wellhead Refinery
Material
flows
Pumps Rail
terminal
Tank farm
Sensor locations Sea
Envisaged sensor communications links terminal
Figure 2. First cut measuring system layout.
Candidate SE education environment SE practitioner Employer Sponsors
in situ experiences
Professional
Funds the
Standards
Academia
Suppliers
systems
society
Designs, Projects in work
leads, creates,
operates, and SE improvement
improves Non-SE work
Candidate
enters
Produces
Systems in operation
Value adder Value carrier Value realizer
Figure 3. ConOps developed to train engineers.
It is useful to form a matrix of the likely influence para- as the requirements are developed; they settle down as
meters as the design proceeds. For each parameter, list the investigation proceeds.
• parameter name
• effect by which it arises
• likely magnitude, frequency, and phase of the influence
3 THE OPEN-SYSTEM CHALLENGE
signal
• likely effect on the output signal in terms of error If humans are involved, the task can be considerably com-
• suggested means for its sufficient reduction. plex to design. After gaining oversight of the whole system,
and its working environment, prepare a list recording where
It will not be possible to fill in all of the details of these human activity is likely to be needed in the construction,
various mechanisms until after the requirements have been installation, operation, and the maintenance of the measur-
well established. They, however, need to be understood ing system.
402 Measuring System Design
Outer operational
environment
System design boundary Signal carrying
value of
temperature to
transponder,
recorder, and
Thermometer enclosure display
Temperature
being
measured
Constant EM field,
temperature, and humidity
Electromagnetic Temperature Humidity
shield shield (moisture) shield
Electromagnetic Temperature Humidity
field variations variations variations
Figure 4. Simple closed-system diagram for a small sensor system.
Each of these human situations will need human factors Phases of System Life Cycle, Volume 2 for a description
(HF) consideration – see Article 94, Human Factors of the alternatives. This has to be decided early as it dictates
Engineering, Volume 2; Article 95, Human–Machine what process is needed to develop the early SE (systems
Interface, Volume 2; Article 96, The Domains of Human engineering) items and specifications.
Factors Integration, Volume 2; and Article 97, Design
Methodology, Volume 2.
Assess whether the situation is suitable for reductionist 4 SPECIFYING THE NEED
methods by testing it for the following:
If not already familiar with design, this is the time to
• Is it only concerned with energy or mass flows and not consider the engineering design process. It has the same
people activities? elements as the SE life-cycle process but it is expressed as
• Can a truly closed system be modeled? a flowchart of tasks with key feedback paths incorporated –
• Will all elements be able to be modeled with formal shown in Figure 5 as links back to (1). Detail designers
models? usually view the design process as shown in Figure 5.
If the system boundaries are not clear, then it is of There may seem to be too much process management
little value to advise the client that the system has no needed, but it is imperative to follow the general thrust of
clear-cut solution. The design has to proceed with the the flowchart or serious design errors can appear late in the
fuzzy boundaries that are modeled as best as one can as process. Design errors are expensive to fix if not picked
things progress. Where they are fuzzy, application of the up early. For example, a resistor in an electronic circuit
soft systems methodology is suggested – see Article 63, that does not have a sufficient power rating will be hard
Outline of Systems Thinking, Volume 2. to replace once the circuit boards have been manufactured,
Explore how the boundary might be limited to a closed for the following reasons:
system by employing risk reduction measures, such as
• too little space is then available for the larger packaging
reducing human involvement, or by circumventing a dif-
needed;
ficult area with bypass type of solution.
• there may be overheating of adjacent components;
Areas of design uncertainty must not be ignored as they
• a board change will be needed that entails removal from
have a habit of showing up at the wrong time. Also, make
service, freight, replacement, and testing cost.
sure they are recorded so that they are not overlooked.
To this point, little detailed engineering design has been To change the design at the early design stage, the
undertaken, but what to do is becoming clearer. cost will be as little as a dollar, but once in service, it
It will also become clear as to what type of life- shoots up to hundreds/thousands of dollars if it gets to the
cycle development model should be used – see Article 66, recall situation.
Executing A Measuring System Design 403
Trade-studies and
Design request
analyses
Identify concept of Optimum solution(s)
operations with modeling
customer
Optimum solution(s)
Identify customer prototyping and
needs/requirements trialing
System specification Prelim/Detailed
design review
Customer review
Design
1 No
approved?
Yes
Specification
No
approved? Preproduction
manufacture
Yes
Function
1 identification and Test and evaluation
modeling
Operation trialing
(OT&E)
Physical allocation
1
and synthesis
Production manufacture
System design
review
Test and evaluation
(PAT&E)
System design
No
approved? In-service support
(FOT&E)
Yes
1 System reticulation System upgrade/
disposal
Figure 5. Flowchart of the design process (Courtesy: Tim Welburn).
Getting a design ‘right’ requires considerable effort in to have more resources than needed – at least that is too
the early phases of the life cycle. As a guide, considerable often the perception of senior managers – the total is less
design resource should be placed into this front-end loading overall.
(FEL). Figure 6 indicates the differences in FEL. There is little doubt that allocating the right level of FEL
The area under the curve is an indication of total cost. expenditure is a wise move; but there are many forces at
It will be seen that while the large FEL at the start seems work against proper allocation.
404 Measuring System Design
Need should be based on real deficiencies, that is,
Dreaded
budget takeoff, • the current system is inadequate, is
and with no
completion • unavailable or unsupportable, or is
Hopeful
• too costly to operate.
budget
High Low FEL recovery, The statement of need should be presented in specific
Cost
FEL but late
qualitative and quantitative terms, as sufficient detail is
provided to allow the systems engineering process to
Happy times! proceed in functional terms.
• On budget
• On time Requirements extraction is just that – extraction of hope-
• To specifications
fully ‘hard’ objective detail from what is initially often quite
Concept and Design and abstract thinking. It is usually necessary to provide assis-
feasibility studies manufacture
tance to the customer to transform the initial expressions of
Figure 6. Effects of different FEL in project outcomes. need into the format and knowledge content needed to later
flow down tasks to the detailed engineering activity stage.
The first difficulty is that most situations are operating Customers are not often experts at this task and would
in a fixed financial supply climate. This means, overall, not have usually approached it by establishing their need
managers will tend to opt for more project starts by cutting on well-structured lines that suit those who follow. To get
the FEL budget on each in order to allocate funds to more a project approved, they would have been working in a
new starts. When the project overruns the forecast time and different direction to the designer’s need, perhaps coming
budget expectations, the lack of the earlier FEL budget is up, instead, with a business plan or budget bid that makes
not seen as the culprit. its case differently to the needs of requirement extraction.
High FEL can be interpreted by some as a luxury Requirements extraction starts with the formulation of
situation wherein designers will waste the funding by doing the key factors – often called the Critical Issues (CIs) –
too many unnecessary things, at too slow a pace. If the via a process of discovery, Sage and Rouse (1999). A
senior managers have little faith in use of high FEL, they representation of the steps of the complete development
will probably also feel that they are not keeping back process is given in Figure 7.
sufficient funds to fix the system errors that always seem to Although the customer will usually provide a written
arise in projects around delivery time – a circular argument needs statement, it is the contractor that prepares the first
that leads to failure all too often. Early planning is needed requirements list. This developing document needs to be
to avoid the risk of late errors. regularly checked with the customer to ensure whether
this is what is needed. In this activity, it is important to
ascertain that the customer is doing a sound job of verifying
5 REQUIREMENTS GENERATION suggestions; they may well be distracted by other more
urgent internal matters.
Determining a sound set of specifications is one early task Always keep in mind that this is a most critical step in
that needs appropriate resources. The requirements extrac- the life-cycle process; this is where directions of effort of
tion activity generates the inputs for System Specifications, many people are decided.
which in turn, establishes design requirements. A client needs a system that
Establishing the set of requirements for a project needs
patience, understanding, and experience. It can take a • provides operational capability, that is, does the right
considerable time to assemble a sound set of issues that task;
characterize the customer’s need. They must be set up • possesses adequate operational suitability, that is, does
carefully and completely. the above tasks when needed, for as long as needed;
This section can only give an introduction; more detail • meets certain cost conditions for ownership and ongoing
is found in several texts and standards on the topic and operation;
in numerous published papers and standards documents • fits within a host of legally required environmental
such as IEEE (1993), Sage and Rouse (1999), Bray (2002), conditions;
Hull, Jackson and Dick (2002), INCOSE (2004), and • meets appropriate health and safety regulations.
Sydenham (2004).
Requirements need to cover much more than the opera- To provide this large list of ‘must haves’, it is necessary
tional task that the measuring system has to meet. to tease out a long set of requirements. A sound way to
Executing A Measuring System Design 405
Rewrite Rewrite
requirements requirements
No No
Ask why
Define Validate the No Remove
Problem Write system Customer each
concurs?
figures of set of Valid? requirements
statement requirements requirement
merit requirements from list
is needed
Yes
No Design and
Yes perform tests
Determine Design and
Verify? Yes verification Test? No perform
method analyses, et al.
Requirements
listing
Use to Create risk
mitigate Yes mitigation Risk mitigation
risk? program Used to set
up technical
designs
Create technical
Use for Yes performance measures TPM tracking
TPM? (TPMs)
Figure 7. Requirements development process. (Reproduced from Handbook of Systems Engineering and Management, Sage and Rouse,
Copyright (1999), Reprinted with permission of John Wiley & Sons, Inc.)
get started is to generate an operational concept or ConOps This activity is looking for directions as well as for
document for the project, as is discussed in Section 2. specific engineering needs. Engineering design must start
After becoming familiar with every statement about the into its detail only when the needs and constraints are
project that is then available, the next step is to explore the adequately appreciated. It should not leap off with a specific
nature of the problem to be serviced. Key issues to address solution at this early stage, but be top-down driven from a
are as follows: sound requirements extraction.
A situation assessment should be developed. General
• Why is this measuring system needed? approaches to apply here are as follows:
• What role does it fulfill?
• Is it a new system with few precedents or is it an • What should it be?
updated copy needed to give the customer a place in • What are the factors that matter over time?
a market or to suit an enhanced operation? • What to do and how to do it?
• How does it provide for that need? An important aim of requirements development is to lead
• Is it that a major activity will be needed to push forward to the recommendation of some candidate solutions. The
the technical edge? most likely approaches need their performance, effective-
• Is it a new innovative use of existing largely proven ness, maintenance, and logistic support to be made clear.
technology, such as a new-generation digital measure- This critical identification step is all too often jumped
ment and control system? into using ‘intuition’ alone to start an investigation of
requirements. This may well start to motivate the mind,
Differences realized here will highlight many require- but its outcomes must be followed up with analytical
ments issues. This is also the time to make a short study of assessment of the findings.
the type of system needed in terms of its hierarchical and It will not be feasible to address every issue that is
organizational natures as discussed in Article 63, Outline discovered. Concentrate on the Critical Issues (CI); large
of Systems Thinking, Volume 2. The kind of development systems will identify hundreds of CIs. Learn from similar
it will be will then emerge, revealing the generic kind of past projects from which many CI definitions and metric
engineering solutions that might be applicable. units may be reused.
406 Measuring System Design
CIs group into four kinds; these have to do with the From the requirements statements, the technical specifi-
following: cations (also called technical requirements) are identified
that say how the design will be approached. Specifications
• Operational task (What is to be done?) of many kinds are prepared from these.
• Suitability (Will it perform when needed?) Many types of specifications are used in a project.
• Political, social, legal, and environmental (What are the Table 1 gives a summary of the essential types that are
key constraints?) usually involved in the larger projects. USDoD MIL
• Programmatic ability (How well will the contractor documents are given as examples as many of them are
perform in the development?). available for free download. There exist numerous others.
The ones to use are those mandated for a project in its SE
Being able to provide adequate operational service when Plan or elsewhere.
needed is as important for a system design as is making Other kinds of specification are for
sure that it does the right job. Ability of a system to
successfully fulfill its mission objective is highly dependent • documentation
on the effectiveness of the support infrastructure provided • installation
by the design. This is variously called the R&M (Reliability • packaging and transport
and Maintenance), ‘ilities’ or ‘special functions’ aspect, • standard items
Sydenham (2004). • modified items
Life-cycle costing could be a major consideration at the • special subcontracts
requirements extraction stage. • incoming inspection
A Maintenance Concept should therefore be developed • use of vendor specifications in projects
that leads to a maintenance plan. • service and maintenance
Requirements are best prepared as descriptions of the • safety
operation of the system under development and do not state • security.
any design solutions – they emerge later. Avoid writing any The specifications that are to be used in a project are
requirement that dictates a technical solution. usually designated in the Systems Engineering manual
Small measuring system projects can maintain a simple for the project. Clearly, small projects will roll these
hard-copy record of the requirements. However, when the together, or often not have them! The list does give an
number of individual requirements exceeds a few hundred, idea of the aspects that a design project has to address.
it becomes essential to use a computer-based requirements Standard specifications are also useful reference works, for
management tool. All these are based on the creation they often embody practical methods and workable ways
of a tree structure by intelligently generated functional of proceeding.
decomposition activities, starting out with the CIs. Caution
is needed. As tools allow rapid generation of statements,
it is easy for poorly stated requirements to be entered. 7 PARTITIONING
What happens later depends on these statements being well
thought through. As the initial ideas develop for a concept design solution,
Detailed design should only start after a careful study of there arises the need to break down the whole into parts,
the requirements has developed a sound set of requirement each of which is then broken down again. This is called
statements that have been fully validated with the customer. system partitioning, box cutting or system reticulation. This
Pointers on the features of a good requirement have been activity concerns the physical association of component
published, Sage and Rouse (1999) or Sydenham (2004). parts and how their interfacing is done in the real world
and in the design meta world.
A single sensor measuring instrument can be simple,
its breakdown being into a serial chain of blocks – see
6 CONVERSION INTO A SYSTEM Article 71, Modeling Methodology, Volume 2. In these
DESIGN cases, the partitioning is reasonably obvious.
As the number of subsystems that interface grow, just
With the requirements in place, it is now the right time to how they are grouped is crucial.
put effort into detailed design. Each requirement statement For example, development of a multiple sensor monitor-
will outline something to be done and what it is to. It ing system for an agriculture project will involve sensors,
will not describe how to achieve it; that is the technical processors, communication channels, central processing,
designer’s task. maintenance, calibration, training, and spares replacement.
Executing A Measuring System Design 407
Table 1. Essential types of specifications, Sydenham (2004).
Specification type Description Example format
Operational A specification describing the requirements of the MIL-STD-498 DI-IPSC-81430 ‘operational
requirements proposed system in terms of user needs, its concept description’.
specification relationship to existing systems or procedures,
and the way in which it will be used and
maintained.
System/subsystem A specification describing the requirements to be MIL-STD-498 DI-IPSC-81431 ‘system/subsystem
specification satisfied by the system and the methods to be specification’.
used to verify that the requirements have been
satisfied.
The SSS is used as the basis for design and
verification of the system.
Interface requirements A specification describing the requirements MIL-STD-498 DI-IPSC-81434 ‘interface
specification associated with a system that defines its requirements specification’.
interface(s) with other systems.
Interface requirements may be defined as part of
the SSS or as a stand-alone IRS.
Subcontract A specification identifying requirements and MIL-STD-498 DI-IPSC-81431 ‘system/subsystem
manufacture/quality standards of workmanship for items of a system specification’, tailored to identify requirements
specification supplied by a subcontract agency. associated with safety, system quality factors,
Such requirements may be identified in a and design and construction as well as other
subcontract Statement of Work (SOW) as part SOW-style clauses associated with the
of a contract or the SOW/Contract may refer to manufacture and quality of items.
the suggested stand-alone specification.
Acceptance test A specification identifying the criteria for N/A
specification acceptance of system requirements described in
(subcontractor, the SSS.
factory, field) The specification may be tailored to address
acceptance criteria for requirements verified by
a subcontractor, at the facilities of the
contractor, prior to delivery to the end customer
or in the field (i.e. when the system is delivered
to the end customer).
Installation A specification identifying requirements associated MIL-STD-498 DI-IPSC-81431 ‘system/subsystem
specification with the installation of the system in its target specification’, tailored to identify requirements
(end-use) environment. associated with installation of the system in its
end-use environment only.
If the system development is divided to have all sensors applies to software modules. It also applies to design
being developed as one kind of subsystem and mainte- activities wherein the design of an item needs the data of
nance as another set of activities, there could well be the another item’s design to be completed and also the data on
need to call all sensors in to carry out the maintenance. where the activities of operation link up.
Alternatively, they could be designed with the sensor and This means a design exercise always progresses with
preprocessing, with maintenance being done for each sensor numerous information interfaces. Each one is a high-risk
unit. An example of this going wrong was seen in a fighter design-error situation.
plane for remote airfield use that needed 20 plus different For example, the design of an aviation electronic ampli-
maintenance kits, each with different training needs. The fier will need to interface with the packaging, weight, power
project folded when the airfield maintenance staff made it supply, cooling, and shielding designs. The design require-
clear that such a division could not be supported in the field. ments of all must be satisfied simultaneously.
Practice has shown that in design activity, there will be
8 INTERFACE DESIGN almost as many interface design links as there are available
interfaces, this can be a huge number.
A physical interface is where signals, components, or If the many interfaces involved are not well thought
subsystems connect together. The same interface concept through, prohibitive mixes of activities can arise. There was
408 Measuring System Design
once a space telescope that was designed in its testing and The need for T&E is summed up by asking three key
calibration functions such that it could only be designed questions of a system development to see if they are
and built when the complete system was assembled and in place:
the unit placed into a flight simulation chamber. Any thing
found to be wrong at that late stage could only be fixed by • What are the system’s development teams trying to
iteration back to the very roots of the design. achieve?
By partitioning the system in different ways, the number • How will those concerned know when the performance
of interfaces can be kept to a minimum. Another way to objectives are reached?
regard this is by seeking a high degree of isolation of design • Who has the responsibility for a satisfactory perfor-
items. There exists a formal way to investigate this using mance outcome?
the n2 matrix method.
For example, in the design of a sensor, it is sound A well-run project will not be using testing as an
practice to combine the preamp, supply voltage regulation, experiment to find out what has been developed but
conversion, and shielding packaging in one unit sending a to verify that the performance of the system is where
high-level signal over the communication system. it is expected to be. A ‘no-surprises’ project situation
A poor design situation was one in which a television- should be the aim – T&E is a key mechanism to achieve
set maker used resistors as a support to wind inductor that condition.
windings with the insulation set up to deliberately incorpo- As such, the first text on T&E seems to be the one by
rate capacitance – very ingenious but hard to troubleshoot Stevens (1989). The case for T&E to be given more status
as components could not be isolated for testing or even in systems development and operation has been well made
be seen. in Crouch (1992) and Reynolds (1996). There is, however,
little other published material on this topic. It is quite
surprising to see project after project being done without
9 TEST AND EVALUATION sound T&E.
Investing more resources in T&E for a project has
The three drivers for all systems developments are cost, the potential to prevent cost overruns and failed systems.
time, and performance – the CTP factors. Improvements Unfortunately, all too often, the T&E budget component is
in all three are always necessary. At first sight, optimizing the first to be pruned when overruns arise.
all three together seems to be impossible. Process improve- T&E is often practiced in an ad hoc informal manner
ments and improved design support systems have, however, as a band-aid activity to find out things when a project
actually improved all of these in many projects. is not going well. In this form, it has the following
As projects are always set up using normal business deleterious features:
practices, it is reasonable to assume that the control of cost
(accounting) and time (project management, PM) will be in • No adequate traceable or recorded control process
place for a project. exists.
Management of performance, however, is often left to • Success relies on the various designers’ abilities to
the design engineers to implement on an ad hoc informal know when and what to test, after which they often have
basis – see Article 3, Measures and Metrics; Their no adequate records addressing the three T&E questions
Application, Volume 1 for an overview of the use of given above.
measures and metrics. • There is a real chance that the system elements that the
This general area has been called Test and Evaluation various design teams are developing will not integrate
(T&E). Too many people think of this only as testing, that without considerable rework. This situation can arise,
is, measuring a Technical Performance Parameter (TPP) not because of lack of competency in performing a good
without appreciating T&E is really about evaluating the detail design but by simply designing the wrong thing.
performance of a system as it develops to ensure it delivers • Omitting the overhead of planned T&E activity can
the expected performance throughout its life. indeed save short-term cost. Doing this, however, can
Implementation of sound through-life T&E practices can increase the risk of not obtaining final success. T&E
provide ongoing data on the maturation of the system’s expenditure can save wasteful rework later by detecting
CIs. These can be used to tell managers, designers, clients, early design errors.
and financiers that the development is moving toward com- • Last-minute decisions are often made on what to test.
pletion on time, within budget, and with the performance This can lead to poor testing as the materials and
required. T&E should be regarded as a whole-of-life pro- equipments, not being planned ahead, are often not
cess, not just as a set of tests made at strategic times. available.
Executing A Measuring System Design 409
• There exists too much flexibility in setting up tests and • special development facilities as are needed according
in how to process the data for evaluation purposes. to the project
This leaves things very open for biased tests to be • sound communications
implemented to obtain an apparently satisfactory result. • effective design records archiving
• configuration management system
An important thing to recognize is the need to plan T&E • safety control process
activity from the commencement of a project, not at the • design controls
end of a development stage. A suggested methodology for • sources of advice and mentoring for junior staff
implementing T&E is given in Sydenham (2004). A T&E • in-house training
Master Plan is essential. This outlines the what, how, and • key documents library.
when of the T&E activities to be done.
Three main planning needs of a successful T&E program As a guide, design team leaders need to be familiar
are as follows: with the basics of systems thinking and the culture of
systems engineering. They need to have at hand copies of
• Ensuring that scarce test resources efficiently address foundational books and standards on SE practice and refer
the project’s CIs. Every possible system parameter to these for concepts and methods to employ as project
cannot be tested, there are far too many of them. problems arise.
• Ensuring that those that need to know can observe how Finally, when executing SE and T&E practices for a
well the CIs are maturing according to well thought project, there is the question of ‘how much is enough’.
through expectations. Overhead costs of a support process are hard to justify
• Providing information on the plan and its maturity when the results of that process are of an abstract nature,
to all who need to know about these issues. For cover long-term issues, and do not appear to produce
example, those who plan the use of test resources easily measured value-adding components to a project. SE
need considerable lead time for organizing availability and T&E activities are often seen like this – as costly
of test objects, observers, test equipment, test suite luxuries. They are, however, as important to success as
programming, and so on. are accounting and project management in that they also
assist early detection and control of design error. Who in
As development moves through its stages, the CI deter-
their right mind would make a major development journey
minations should, by tight design control, gradually trend
without sound assurances that all design work is on track
toward the target value, with their uncertainties reducing to
and moving the design forward is in the right direction at
acceptable values by the required time.
the right speed and within budget?
With such a system available, the project managers and
The design team leader has to use personal judgment
design team leaders can see how the CIs are maturing
in setting the scale of use when applying these techniques
toward their goals. Time trend graphs can show the current
or support mechanisms. A single team comprising a dozen
state of performance maturity compared with the target
or so staff working in a start-up company will probably
value. Severe deviations can then be given more design
not be able to devote the time to writing an SE manual.
attention to bring a particular CI back on track. An openly
They might instead simply align with an SE standard. The
observable system such as this also exposes errors early,
team leader in that case, however, still needs to apply SE
where it can be less expensively corrected with less impact
principles as part of routine technical management. It is
on the careers of the staff.
a matter of judgment. First, you need to know what is at
stake, then, what tools are available to assist, and then to
have the experience to reflect on what to do.
10 APPLYING SE TO A PROJECT
When a measuring system design team is working within
RELATED ARTICLES
a large systems provider house, there will be a person
responsible for organizing systems engineering support for
all teams. In such situations, teams usually will be provided Article 63, Outline of Systems Thinking, Volume 2;
with a Article 66, Phases of System Life Cycle, Volume 2; Arti-
cle 67, Principle of Concept of Operations (ConOps),
• company-specific SE process manual Volume 2; Article 68, Setting the System Boundaries,
• computer-based support tool system and support men- Volume 2; Article 69, Requirements Allocation, Vol-
tors ume 2.
410 Measuring System Design
REFERENCES Institute of Electrical and Electronic Engineers, Guidance for
the development of the set of requirements, IEEE Std 1233, 1998
Ed. (R2002) (Includes IEEE Std 1233–1996 and 1233a-1998),
Bray, I. (2002) An Introduction to Requirements Engineering,
IEEE Standards Department, New York, 1998.
Addison-Wesley, New York.
Reynolds, M. (1996) Test and Evaluation of Complex Systems,
Crouch, V.H. (1992) Test and Evaluation as an Important Emerg-
Wiley, Chichester.
ing Discipline, in Proceedings Australasian Instrumentation
and Measurement Conference, Adelaide, (pp. 7–17). Ring, J. (2000) Development of Educational Methodology, Work
in progress for Education and Research Technical Committee
Hull, E., Jackson, K. and Dick, J. (2002) Requirements Engineer-
of INCOSE, www.incose.com,.
ing, Springer Verlag, Berlin.
Sage, A.G. and Rouse, B.W. (eds) (1999) Handbook of Systems
INCOSE. (2004) International Council on Systems Engineering,
Engineering and Management, Wiley, New York.
SE Management Technical Committee, Requirements Working
Group, www.incose.org/rwg. Stevens, R.T. (1989) Operational Test and Evaluation, Krieger,
FL.
Institute of Electrical and Electronic Engineers. (1993) Guide
for Developing Systems Requirement Specifications, (IEEE Sydenham, P.H. (2004) Systems Approach to Engineering Design,
P12330), New York; IEEE Standards Department., and later, Artech House, Boston, MA.
65: Life Cycle Concept
Floyd Guyton Patterson, Jr
George Mason University, Fairfax, VA, USA
is the study of emergent properties of systems that is the
1 Concept of the Life Cycle of a System 411 principal subject of systems engineering.
Emergent properties of systems, when they are suffi-
2 The Serial Nature of the Life Cycle 411
ciently well understood and properly managed, provide both
3 Life Cycle Versus Team Working Model
solutions and constraints that directly influence designs. On
Comparison 412
the other hand, unanticipated emergent properties can cause
4 Typical Life Cycle for a Measurement products to fail to meet structural, functional, or purposeful
System 414 requirements of the system. It is well known that system
Reference 415 defects, which require rework of the product specifications,
Further Reading 415 are much less costly to correct when the problems are antic-
ipated – detected and corrected early – rather than later in
the design after the product system, and perhaps the pro-
duction and delivery systems, systems developed to enable
1 CONCEPT OF THE LIFE CYCLE OF A the production and delivery of the product as well, have
SYSTEM already been built.
The development of a system has both a positive and a
negative aspect. In the negative sense, it is in the nature of
The life of a system is modeled by its life cycle. In the
engineering a system that every design decision, beginning
case of engineered systems, systems that are built using the
with the development of the initial concept, forecloses
principles of engineering, the word life cycle is synonymous options for the structure, function, and purpose of the
with process model , and it refers to the processes by product. A poorly considered decision early in the design
which the system is conceived, specified, designed, built, could, in theory, eliminate excellent design alternatives
deployed, and retired. We will see that life cycles may be before they are ever considered. Engineering a system
quite generic, or quite detailed. A detailed life cycle is is subtractive – each decision subtracts the options not
shown in Figure 1 for an arbitrary large, complex system. selected from the total option space – when viewed in
Use is also made of other life cycle models, see Article 66, this way, since every decision reduces the size of the
Phases of System Life Cycle, Volume 2. option space. The positive aspect will be discussed in the
Everything that is divisible into parts can be considered next section.
to be a system. It is often said of systems that ‘the whole is
greater than the sum of its parts’. This means that there are
attributes or properties of the whole that did not exist or that 2 THE SERIAL NATURE OF THE LIFE
were not manifest in the collection of parts from which the CYCLE
system was assembled. These attributes may be properties
of the structure, function, or purpose of the system. Such In the previous section, it was noted that the development
attributes are called emergent properties of the system. It of a system has both a positive and a negative aspect;
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
412 Measuring System Design
DEFINITION system is additive when this viewpoint is adopted, since
1. Elicitation of requirements each decision leads to more options, which lead to more
2. Classification of requirements
3. Analysis of requirements
decisions, ad infinitum. Neither point of view is altogether
4. Prototyping of requirements correct or incorrect. However, in both cases, the importance
5. Requirements documentation and specification of early decision in constraining or creating later decisions
6. Preparation of transition documents: request for proposal
and a proposal, containing a statements of work, a is self-evident. For this reason, systems engineering places
systems engineering management plan, a work strong emphasis on the earlier design decisions, those upon
breakdown structure, and negotiated list of other
deliverables
which later design decisions are based.
In all this, there is the concept of growth. A given system
DEVELOPMENT begins as an embryonic entity: a concept, idea, vision, or
1. Creation of a system architecture and an organizational
structure consistent with the system engineering manage-
the like. Through the repeated application of the process
ment plan of deciding among options as described above, the system
2. Establishment of configuration control grows in size and maturity, perhaps changing state several
3. Identification of subsystems
4. Production of subsystem specifications and subsystem times, until the system is realized as a product. The different
test requirements states may include the mental image, a written concept, a
5. Specification of subsystem components
6. Design of subsystem components
drawing, a prototypephase, subsets of the total development
7. Fabrication and testing of hardware components process that are relatively complete, or self-contained, with
8. Coding and unit testing of software and firmware respect to their structure, function, or purpose.
components
9. The development of environments and test equipment for In broad terms, the simplest life cycle of any system can
hardware testing and scaffolding for software unit testing be described in terms of
10. Integration of components into subsystems
11. Integration testing of each subsystem • a definition phase,
12. Validation of the built and tested subsystems
13. Integration of subsystems into a system • a development phase, and
14. Integration testing of the system • a deployment phase.
15. System validation against the system specification
16. Integration test procedures
17. Provide product information to planning & marketing This is referred to as the DDD model, shown in Figure 2.
18. Preparation of transition documents and other aspects of The DDD framework facilitates understanding of the
preparing to deliver the system, such as installation
materials and procedures, training manuals and special
resource requirements needed by the system during the
training equipment, maintenance equipment and manuals, various stages of its existence. The DDD model is recursive,
and user's guides in that any of the Ds may be further broken down into steps
DEPLOYMENT
by application of the DDD model to it (Patterson, 1999).
1. Delivery and installation of the system
The utility of this approach to engineering a system is
2. Acceptance testing of the system found in the ability of the management to treat each phase
3. Operational testing and evaluation of the system or subphase as a self-contained project, with discrete inputs
4. System acceptance by the customer
5. Formal maintenance agreement with a contractor that and outputs, and with its own schedule and cost profile. As
includes a maintenance plan, maintenance test plans, we shall see, the phases are arranged serially, such that the
and maintenance test procedures
6. Trade-off study to compare options for continuing to output from any phase represents the starting point for the
support the business process successor phase; and the input for any phase is provided by
7. Decision to decommission the system its predecessor.
8. Provide current system description to requirements team
(linkage to new cycle)
3 LIFE CYCLE VERSUS TEAM
Figure 1. A thirty-two step life cycle. (Reproduced from Hand-
book of Systems Engineering and Management, Sage and Rouse, WORKING MODEL COMPARISON
Copyright (1999), reprinted with permission of John Wiley &
Sons, Inc.) It has been pointed out that the systems engineering
life cycle provides a simple, useful management tool for
and the negative aspect was explained. In the positive estimating, allocating, and accounting for resources dur-
sense, it is also true that each decision opens up a new ing the life of a system. It is an oversimplification to say
set of viable options, prompting yet another decision. The that phases of the life cycle are self-contained from the
sequence of options and decision from concept to product standpoint of engineering the total system.
can be represented as a decision tree, which can provide a The life cycle concept is a growth model that embodies
useful record to support subsequent developmental efforts, the concept that early design decisions determine options
provided enough information is captured. Engineering a for later design decisions throughout the life of the system.
Life Cycle Concept 413
Begins with the need for or the
concept for the system, ending with
the creation of requirements and
DEFINITION specifications
Begins with the system specification,
includes all software and hardware
DEVELOPMENT manufacturing, integration, and
testing, and ends wiht deployment
Begins with delivery of the system and
DEPLOYMENT ends with system retirement
Figure 2. The DDD life cycle. (Reproduced from Handbook of Systems Engineering and Management, Sage and Rouse, Copyright
(1999), reprinted with permission of John Wiley & Sons, Inc.)
It follows that no subset, including a life cycle phase, can Product size
be entirely self-contained. While it is true that certain very
well-defined and well-understood parts of the work effort
can be considered to be miniprojects, these parts are always
subject not only to verification against a specification, but
also to validation in the context of the system of which it
is a part.
A common error is to define subsets of the system with
nontrivial or poorly defined interfaces to other parts of the
system or with insufficient acceptance criteria to ensure the
successful validation – testing of the system with the goal of
determining whether the customer’s expectations have been
met – of the miniprojects. It is very difficult, for example,
to imagine that one group of engineers can write a set of
requirements for a system and deliver the requirements to
a different group of engineers for development of a design
with the expectation of success. While such an effort may Team size
pass a verification test (for example, a design may meet
each and every one of a set of written requirements), it is
much more difficult to imagine that such a design could be Figure 3. Relationship of project risk to product and team size.
validated against the expected result. (Reproduced from Handbook of Systems Engineering and Man-
agement, Sage and Rouse, Copyright (1999), reprinted with
From the standpoint of the engineer, it is much more permission of John Wiley & Sons, Inc.)
important to create, communicate, and develop a common
vision of the system than it is to attempt to create or to fol-
low a perfect specification. Because documents, drawings, or products experience much more risk – the probability of
and other design representations have a very low informa- loss of time, money, or quality – than small teams on small
tion content relative to the amount of knowledge carried projects. This is depicted in Figure 3, where darker shading
by an engineer who has participated in the design process, indicates more risk.
various strategies have been proposed to avoid the loss of Organizations organize themselves around processes
experience and information through the introduction of new according to some partition of the enterprise. Their strategic
people. Such strategies typically involve the formation of a plan is much enhanced by introducing a life cycle model
system team, charged with the responsibility of communi- with proven success for the domain. An ad hoc organiza-
cating the common vision to the group. The system team tion will occur – this is the message of Stogdill – if not
remains with the project throughout its lifetime. templated – provided with a structural pattern – especially
Because difficulty of communication varies with the size if the organization is involved in an enterprise with which it
of a project team and also with the amount of data that is unfamiliar. Introducing the systems engineering life cycle
one needs to communicate, large teams with large projects template saves much time and helps to assure success.
414 Measuring System Design
4 TYPICAL LIFE CYCLE FOR A The input to the definition phase is a need for a particular
MEASUREMENT SYSTEM measurement system that has a well-defined purpose and,
perhaps, other success criteria in terms of the structure or
function of the system, and any known constraints on the
design. The output of this phase is a well-researched and
The life cycle of a system begins when there is suffi-
analyzed list of attributes or properties to be measured by
cient need to motivate the formation of a concept. The life
some means.
cycle ends when the system is decommissioned, the DDD The list of attributes is the input to the development
model can be applied to a measurement system in a very phase, where metrics and measures, which may be qualita-
straightforward way. Recursive application of DDD yields tive or quantitative, are proposed, validated, and packaged
increasingly detailed descriptions of the activities that com- as a metrics framework for use. Thus, the deployment phase
pose the effort that is to be managed. It is worth noting here, can be concerned with using the metrics framework as a
before discussing the Figure, that the Figure is only a par- tool for actually measuring attribute levels in the data –
tially complete illustration. The effort required to develop see Article 3, Measures and Metrics; Their Application,
a life cycle that is custom built for the development of sys- Volume 1.
tem is repaid in many ways. It is a necessary operational In Figure 4, the DDD model has been reapplied to each
framework that needs to be reinvented for each new system of the three phases to yield subphases as illustrated by the
development project. color coding. To further illustrate, the definition subphase
Figure 4 depicts a partially completed life cycle for a of the definition phase has also been refined. This recursive
generic measurement system, dividing the life cycle into descent into greater levels of detail may be continued until
three phases, definition, development, and deployment. further detail yields no further useful information.
At this point, it is appropriate to articulate the expecta- Further information on systems engineering and the life
tions for each of these three phases, especially the inputs cycle model application is found in the part on ‘Part 5,
and outputs. Measuring System Design, Volume 2’.
DEF Definition (DEF)
DEF Identify objectives
DEF
Identification of goals within existing value system
DEV
Evaluation of present quality measurement system
DEP
Enumeration of obectives
DEV Identify, analyze, and enumerate attributes of each objective
DEF
Identification of known attributes, and postulation of additional attributes
DEV
Qualitative analysis and classification of attributes, organized by objectives
DEP
Enumeration of objective attributes
DEP Candidate measurable factors (measures) and units of measure for each attribute
DEF
Identification of validated measures, and postulation of additional measures
DEV
Qualitative analyis and classification of candidate measures, organized by attributes
DEP
Enumeration of candidate objective attribute measures
DEV Development (DEV)
DEF Classification of measures
DEV Validation of measures
DEP Framework for deployment
DEP Deployment (DEP)
DEF Collecting data
DEV Analyzing data
DEP Reporting results
Figure 4. Measurement example using the DDD model.
Life Cycle Concept 415
REFERENCE FURTHER READING
Bass, B.M. and Stogdill, R.M. (1990) Handbook of Leadership:
Patterson, F.G. Jr. (1999) Chapter 1: Systems Engineering Life
Theory, Research, and Managerial Applications, 3rd edn, The
Cycles: Life Cycles for Research, Development, Test, and Eval-
Free Press Division of MacMillan, New York.
uation; Acquisition; and Planning and Marketing, in Handbook
of Systems Engineering and Management, (eds A.P. Sage and Sage, A.P. and Rouse, B. (1999) Handbook of Systems Engineer-
W.B. Rouse), John Wiley, New York (pp. 59–112). ing and Management, John Wiley, New York.
66: Phases of System Life Cycle
Kim R. Fowler
Technical Consultant, Cool Stream, Windsor Mill, MD, USA
closely defined procedures and methods and, in the United
1 Life Cycle Approach to the Engineering of States, follow the stringent requirements of the Depart-
Systems 416 ment of Defense (DoD) DOD-178B standard; military
2 Concept Phase 418 devices might follow, instead, DoD STD 2167; finally,
commercial companies often adhere to the International
3 Detailed Engineering Design 419
Standards Organisation (ISO) ISO 9000 standard for pro-
4 Manufacture 420 ducing sytems. Each one of these standards has different
5 Operation 420 specific requirements but they all have similarity, if not the
6 Upgrade 420 exact same terminology, in their phases. These phases are
7 Disposal 421 described in detail below.
8 Concurrent Development 421
9 Evolution of the Life Cycle 421 1.2 Development processes
References 421
Development can move through these phases in several dif-
ferent ways. You can use the top-down or waterfall method
to layout and build your system (Figure 1); its premise is
1 LIFE CYCLE APPROACH TO THE
that the requirements completely specify a design and the
ENGINEERING OF SYSTEMS development follows precept upon precept. Alternatively,
you could build from the bottom up using the prototyping
Systems engineering necessarily takes a life cycle approach. method (Figure 2); you take a basic idea, add some fea-
It views the end from the beginning and considers carefully tures to it, and try it out. Finally, you can combine the
everything in between. waterfall and prototyping methods into a spiral develop-
ment (Figure 3) where requirements and development occur
incrementally (Fowler, 1996; Sydenham, 2004).
1.1 Standards and life cycle variations Each stage of the waterfall process, Figure 1, aligns
closely with the phases described below. The waterfall
The life cycle for each product or system varies according to process is the original model for life cycle development. It
its function and market. You tailor the life cycle approach may be the most well understood but it fails to accurately
to fit the product; a medical device could emphasize the portray most development. It does not account for major
US Food and Drug Administration (FDA) Design Control changes in requirements without starting over. It assumes
Guidance, whereas an electric food mixer will focus on that iteration can only happen between adjacent phases of
manufacturability to reduce final cost. development, which is not typically the case. Problems in
Standards for specific markets drive the life cycle pro- one stage, such as manufacturing, may force you to go right
cesses: software for avionics, for instance, requires very back to revisit the requirements and change some of them.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Phases of System Life Cycle 417
Quality assurance Spiral line
Analysis
and feasibility shows progress
of development Design
Concept
Analysis
Requirements
Design Capability
and
completeness
Programming
Testing
Verification
Requirements Evaluation
and testing
Maintenance
Figure 3. Spiral model of development. (Reproduced from
Figure 1. Waterfall method of development. (Reproduced from “Electronic Instrument Design: Architecting for the Life
“Electronic Instrument Design: Architecting for the Life Cycle”, Fowler K.R. Copyright (1996) Oxford University
Cycle”, Fowler K.R. Copyright (1996) Oxford University Press.)
Press.)
Spiral development, Figure 3, evolved from the per-
ceived strengths of the other two methods to supersede
Concept Prototype their shortcomings. It takes a basic set of requirements and
develops a workable system with a reduced set of desired
features. This allows you to understand more fully the
interactions between critical components and subsystems
Evaluation early in development. You then add more requirements that
by customers
lead to additional features and proceed around the spiral to
integrate, test, and prove greater functionality. It is often
favored in software development.
Redesign
Complete
design
1.3 Judgment
Understanding the principles of systems engineering and
applying the appropriate life cycle development are not
Program enough to ensure an optimal solution. These are necessary
but not sufficient conditions for success. Most final prod-
Figure 2. Prototyping model of development. (Reproduced from ucts are a suboptimal set of compromises and trade-offs;
“Electronic Instrument Design: Architecting for the Life a fighter aircraft, for example, is never a perfect solu-
Cycle”, Fowler K.R. Copyright (1996) Oxford University tion – it trades off many concerns, such as threat, mission,
Press.)
speed, maneuverability, range, weapons loads, and reliabil-
ity. Most products and services of a mature and new kind
The prototyping method, Figure 2, often suffices for will be of this nature.
proving a concept, but it does not have the rigor to transfer Ultimately, good judgment, based on knowledge, expe-
sufficient information and knowledge for manufacturing rience, and wisdom – some call it refection – will do the
and deployment. Its utility resides in one-of-a-kind systems most toward guaranteeing a good result. We can only get
and custom development. there by learning and applying, which means that we need
418 Measuring System Design
to risk some elements of failure. Learning should perfect breaks them down into successively more detailed func-
your engineering judgment, and the only way to get there tional components. These functional blocks or components
is through experimentation, iteration, and failure. ‘No one eventually specify who does what, and when and where
wants to learn by mistakes, but we cannot learn enough they do it. This kind of analysis is much like tracing the
from successes to go beyond the state of the art’ (Pet- roots of a tree down to finer and finer rootlets. It can be
roski, 1985). an exhaustive but worthwhile exercise to understand all the
possibilities and interactions.
Once you define the functional components, you then
2 CONCEPT PHASE model the problems and their solutions through mathemat-
ical analysis, simulation, and prototyping. Mathematical
The life cycle begins by defining the problem. You have analysis can be tedious and limited in its efficacy, but it
a number of sources for understanding the problem better; often can give bounds or constraints to the problem and its
they include the following: solution. Simulation provides insight into the interactions
between components and allows ‘what if’ type of studies.
• customer objectives (which are not necessarily the Prototyping produces a subset or subscale model that allows
users), indepth studies of functionality without committing to a
• user needs (users are not necessarily the paying cus- complete production cycle – see also Article 72, Math-
tomer), ematical Methods of Optimization, Volume 2. These
• mission (defines the regions of operation), methods and techniques allow you to characterize the poten-
• constraints from physical processes and environment, tial solutions, which then enter the feasibility analysis.
• government regulations, Typically, a CoDR or Conceptual Design Review com-
• commercial standards. pletes this phase of development – see also Concept of
Operations in Article 67, Principle of Concept of Opera-
The real brilliance in setting requirements is to see what tions (ConOps), Volume 2.
people do, not what they say. Only by doing this can you
really understand the true need (Christensen, 1997). People
do not necessarily understand the capabilities of technology
and innovation, so they do not often express their true 2.3 Feasibility
needs. They tend to express their desires, clouded as they
are with misconceptions and misperceptions. Setting up the Feasibility is the evaluation portion that refines the potential
need is usually a journey of discovery for both the client solutions. Typically, you will iterate several times through
and the designer. definition and refining requirements, functional analysis,
modeling, and feasibility analysis (Figure 4). As a problem
grows more complex, the analyses become more detailed
2.1 Requirements and time consuming.
Requirements generate the ‘what’ to which designers then Modeling
prepare the detailed descriptions (schematics, source list- Functional Analysis
Definition Simulation
ings) that give the ‘how ’. A sizable portion of time is analysis
Prototyping
devoted to generating requirements in the Concept Phase.
Requirements may include some or all of the following: per-
formance, reliability, maintainability, availability, human
Feasibility Characterization
factors, safety, fault tolerance, operation regimes, environ- analysis of solution
ment, and logistics. See also Article 69, Requirements
Allocation, Volume 2.
2.2 Analysis Requirements
Figure 4. The iterative nature of concept development and feasi-
After defining the problem, you enter an iterative cyle of bility analysis. (Reproduced from “Electronic Instrument Design:
analyses to refine the possibilities toward a potential solu- Architecting for the Life Cycle”, Fowler K.R. Copyright (1996)
tion. Functional analysis defines basic operations and then Oxford University Press.)
Phases of System Life Cycle 419
Feasibility has several categories, business, physical, and updates. Version and configuration control has been man-
risk, which can overlap in their outcomes. The business ual until the recent past, software and systems now exist
case looks for the best outcome, most likely to maximize that assist control of versions and archive documents.
revenues or profit. The physical case evaluates whether the
proposed solution violates any physical principles and other
needs such as safety and legal issues. Finally, risk analysis 3.2 Execution
evaluates the potential problems and margins of tolerance;
these can include Fault-Tree Analysis (FTA) and Failure Execution is the actual design work but it is usually a very
Modes Effects Analysis (FMEA) – see also Article 101, small portion of the total effort. This is the creative part
Assessing and Demonstrating Safety, Volume 2; Arti- of development where coding happens or schematics origi-
cle 100, Safety Analysis Methods, Volume 2; and Arti- nate. You should maintain a disciplined habit of keeping a
cle 99, Safety Organization, Volume 2. These two types detailed engineering notebook, or its equivalent in software;
of risk analysis are complementary. FTA takes each iden- it becomes indispensable when problems arise or modifi-
tified potential fault in the system and traces the cause(s) cations are needed in the future. The notebook needs to
down to the possible originating components or interac- provide clues and reasons for your work.
tions. FMEA hypothesizes a failed component or interaction
and then traces the effects of its failure up through the sub-
systems to its ultimate manifestation at the system level. 3.3 Review
Review has a variety of forms. It may be formal, such as a
3 DETAILED ENGINEERING DESIGN design review. It may be informal, where you consult with
a colleague about a problem. Formal review can include
Detailed engineering design takes the ‘what’ from require- some or all of the following specific instances:
ments and molds it into an integrated whole. In other
words, design synthesizes the ‘how’ of the system. Detailed • Peer review – a process of examining a module or sub-
engineering design follows after the Concept Phase and system, particularly software that evaluates the imple-
feasibility. mentation for soundness. Usually four or five peo-
For many complex systems, detailed engineering design ple take part – a moderator (best being a person
splits into two design phases: preliminary design and crit- not too close to the project) to facilitate the review,
ical design. Typically, you complete each of these phases the author/designer, two colleagues to do the review,
after a formal review that satisfies all open action items and sometimes a note taker to record comments and
or concerns. action items.
Design, and indeed all development, should follow a • CoDR – conceptual design review, which typically
simple, iterative cycle: plan, execute, review, and report. completes the Concept Phase. It evaluates the basic
Every activity should follow this process. Even the overall approach and the feasibility.
phases should follow this general pattern. It is a useful habit • PDR – preliminary design review, which completes
to inculcate. the Preliminary Design phase. It evaluates the basic
approach and the initial designs.
• CDR – critical design review, which completes the
Critical Design phase. It evaluates the final design and
3.1 Documentation approves the production go-ahead.
• Manufacturing transfer or production hand off (PHO) –
The primary output from detailed engineering design is this is an additional step that follows a CDR in a vol-
documentation. Documentation includes source listings ume production scheme. It evaluates the manufacturing
from software, schematics of hardware, plans, and issues and gives final approval for production.
procedures, checklists, memos, messages, manuals, reports,
and reviews. Documentation records the plans, the
execution, and the reviews and tests. These components of 3.4 Test
documentation become the directions and instructions for
manufacturing, producing, and using the product or system. Tests usually begin during the detailed design phase and
Documents need version control to record an accurate evaluate the individual components, modules, and subsys-
historical picture and to maintain a contiguous string of tems. Tests can cover one or a variety of concerns, such as
420 Measuring System Design
performance, failure modes, and environmental tolerance. Otherwise, outsourcing the production effort is a worthy
Examples of tests include the following: consideration, particularly if your company does not have
the expertise to manufacture the product. It does, however,
• Software white box testing – where knowledge of pass control out and thus the supplier has to be reliable.
the code leads the tester to exercise all reasonable
paths of operation (white box for where only internals
are known).
• Software black box (for where only externals are avail-
5 OPERATION
able) testing – where the tester exercises the module in
all the anticipated operations without knowledge of the Just designing and building something is not the end of the
code operations. matter. You need to see it into operation and ensure that it
• Hardware thermal testing – cycling the circuitry through functions reliably. This means that you need to consider the
cold and hot temperatures to find potential areas of effort to install the equipment, train the operating personnel,
operational noncompliance. and provide technical support.
• Vibration and shock – testing the circuitry and mechan- One area of this that is continually overlooked is the
ical hardware for potential problems and stress points. user manual. Successful operation requires a good manual,
• Integration – exercising the system as subsystems are training, and access to appropriate information. For more
attached for functionality and correct operation. complicated systems, a troubleshooting and repair manual
• Verification – testing for conformance to the metrics in is essential.
the requirements. These metrics may originate in per- Once operational, you need to consider the logistics
formance, electromagnetic compatibility (EMC), fault and inventory. What are the consumable resources that
tolerance, reliability, and maintainability. your system needs – fuel, lubrication, batteries, or auxil-
• Validation – testing of the system for apparent fulfill- iary supplies, for example, paper for printers? Then there
ment of the customer’s or users’ desires. Usually, you are maintenance and repair, which are two completely sep-
exercise the user interface or graphical user interface arate activities. Maintenance involves periodic care for the
(GUI) for operational responsiveness, utility, and ease equipment – labor, spares, inventory, and equipment down-
that leads to user satisfaction. time are all factors. Repair is the fixing or replacement of a
• Field-testing – is a combination of integration, veri- failed component; here the same concerns as maintenance,
fication, and validation. It can accurately portray the such as diagnostics, prognostics, and test equipment accom-
environment and functionality of a system in actual use. pany additional issues.
These types of tests are extremely valuable but tend to
be the most expensive; they also often require a com-
pleted system, but today the trend is to make use of a 6 UPGRADE
sophisticated model of the system.
If your system operates with some degree of success, then
invariably someone will want to upgrade its capabilities in
4 MANUFACTURE the future. You need to consider this when you do the initial
design. Prepare design hooks in the original design. These
Manufacturing is an entire field of study unto itself. Con- are openings or attachment points for potential upgrades.
cerns for manufacturing should be addressed during the Careful coding of software can make the addition of new
previous phases of concept and detailed engineering design. modules and capabilities easier in the future. Providing a
You should consider design for manufacture, where fewer, margin in memory, performance, size, weight, and power
simpler parts are almost always better than greater com- will ease the addition of future hardware.
plexity. Another area for consideration is design for test; Designing a system for expandability does add com-
this will help you and those who follow you to also diag- plexity to the system. It generates new possibilities for
nose problems more quickly and efficiently as well as unforeseen circumstances and interactions. Consequently,
show compliance. complete and correct documentation is absolutely essential.
One particular manufacturing concern that you should Necessary documentation includes
consider is building the system in-house or outsourcing
the work. If you have the assets and resources in the • software source listings,
company, which are available and which can be efficiently • hardware schematics, and
and wisely used (meaning keep them busy full time for • design documentation that explains the theory of
most situations), then in-house production is reasonable. operation.
Phases of System Life Cycle 421
The engineering notebook can be indispensable for 9 EVOLUTION OF THE LIFE CYCLE
upgrades. It provides the reasons for why and how things
were done. It can also help with explaining the theory The concept of the life cycle has evolved into various
of operation. forms from its origins in the 1950s and 60s. The waterfall
model gave way to several different models including
the prototyping and spiral models that better fit actual
7 DISPOSAL development. Complex systems, for example, a fighter
aircraft might use a combination of approaches – first it
Environmental concerns for disposing off equipment that is is built to a set of requirements that provides a major
no longer in use are rising. You need to consider, during the subset of functions (basically a waterfall development), then
early phases of concept and detailed engineering, how to it goes through extensive flight tests that determine other
dispose of your system. Plan for recycling of the materials, capabilities. These flight tests are iterative and can cause
particularly of the enclosure and packaging aspects. Reuse changes to requirements, now development looks more like
of components may be an avenue for disposal as well. Toxic the spiral method.
compounds, such as heavy metals used in solder, circuit The facility provided by the Internet is forcing further
boards, and components, are becoming an environmental evolution of systems design and life cycle development.
crisis; think about how you might reduce their use in Embedded systems are gaining Internet ports so that soft-
your products. ware upgrades may occur quickly and easily in the field.
These kinds of upgrades significantly reduce downtime and
allow more frequent tuning of operation within embedded
8 CONCURRENT DEVELOPMENT systems. The capability of the Internet can now combine
field-testing with operation; successive versions of the soft-
The life cycle of development necessarily involves paral- ware can be far more numerous. This type of life cycle
lel streams of effort. Someone must evolve the hardware development strongly resembles prototyping but it is much
designs as someone else designs and codes the software. A faster. The iteration time is much shorter. The problem with
different group must prepare the mechanical mechanisms fast field upgrades is that people will be tempted more than
and packaging alongside the circuits and software. This is ever to release ‘buggy’ code that does not work quite right
the business of systems engineering oversight. to make market deadlines and hope to ‘fix’ the problems
One principal concern is the definition of interfaces later through field upgrades.
within the system. These interfaces will control the archi- Engineering knowledge and judgment is the one area
tecture and the eventual operation of the system; they also over which you have some control during your career. You
affect the concurrent development of the system. A general should be continually learning, monitoring, optimizing, and
philosophy of interface coupling, whether loose or tight, improving what you do and how you do it. Your experience
will help define the architecture. can lead to changes and improvements in the life cycle of
Loose coupling means that the interface is straightfor- development. There is much more to engineering than the
ward with the minimum in communications between mod- design itself!
ules. It tolerates changes and tends to be more robust. Loose
coupling can reduce rework and redesign and can increase
reuse. It aids concurrent and independent development, can REFERENCES
adapt to the market, and ease the test burden. The disad-
vantage with loose coupling is that it tends to produce a Christensen, C.M. (1997) The Innovator’s Dilemma: When New
bigger, heavier system. Technologies Cause Great Firms to Fail, Harvard Business
School Press, Boston, MA (p. 188).
Tight coupling optimizes performance or size by spread-
ing function across interfaces. Its advantage is that it can Fowler, K.R. (1996) Electronic Instrument Design: Architect-
ing for The Life Cycle, Oxford University Press, New York
reduce the initial cost of manufacture with more optimized (pp. 407, 415, 416).
structure or smaller form factor. Concurrent development,
Petroski, H. (1985) To Engineer is Human: The Role of Failure in
however, is much more difficult and testing is more exten- Successful Design, St. Martin’s, New York (p. 62).
sive at the system level. Redesign is more difficult because
Sydenham, P.H. (2004) Systems Approach to Engineering Design,
it affects the entire system; changing one component more Artech House: Boston, MA.
readily affects others within the system.
67: Principle of Concept of Operations (ConOps)
Jack Ring
Innovation Management, Scottsdale, AZ, USA
the operations situation. This latter facet sets the stage
1 Definition of a Concept of Operations for the Measures of Effectiveness (MOEs) by which the
(or Operational Concept) 422 eventual measurement system will be judged.
2 The Concept of Operations Applied A ConOps describes why a measurement system is
to System Development 423 needed, how the measurement system will be used, and
3 Allocations of Measures of Effectiveness the consequent value of such a system. The user’s mis-
to the Conops Model 423 sion objectives and goals regarding the phenomenon to be
4 Documentation 424 measured are mapped to measurement system capability
Further Reading 425 scenarios. It tells a story about the expected behavior of the
phenomenon to be measured and about a day, week, month,
or other episode in the life of the users of such measure-
ments and thus of the desired behavior of the contemplated
1 DEFINITION OF A CONCEPT OF measurement system.
OPERATIONS (OR OPERATIONAL A ConOps is not the same as an Operations Concept.
CONCEPT) The ConOps describes the usage of the intended system by
articulating ‘we need a system that will enable us to do
A Concept of Operations (ConOps) describes how an the following: “The authors are standing in problem space
observer intends to use a contemplated measurement sys- looking out”’.
tem in a specific actual or anticipated situation. It is the The other view, the Operations Concept or Concept of
starting point for engineering a measurement system – see Operation, describes the operators view of a system as
Article 63, Outline of Systems Thinking, Volume 2 – and designed or produced. The authors are standing in solution
other articles on that topic. space looking toward the problem space.
In the vein that ‘understanding the problem is 90% of the This subtle distinction is very important because it
solution’, a ConOps purposefully treats the intended mea- emphasizes an understanding of the operational need with-
surement system as a black box. ConOps authors describe out the clutter, confusion, and complexity of implementa-
the operational situation – what the users need to accom- tion ideas.
plish and how a measurement system can help make sure For example, a ConOps says, ‘here’s what we need
that the right issues are reached. Authors must studiously to measure about the weather and how we intend to use
avoid making any assumptions about the internal content those measurements’. In contrast, an Operations Concept
and structure of the eventual measurement system. document says ‘given that we have designed this ocean tem-
The ConOps focuses on both the stimuli to which the perature, salinity, and current flow system, here is the way
measurement system is expected to accommodate and the it should be deployed, operated, and maintained’. Accord-
user-visible responses to such stimuli. Further, the ConOps ingly, the Operations Concept is a response to the ConOps,
describes the effect the responses are intended to have on not an alternate name for the same thing.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Principle of Concept of Operations (ConOps) 423
A good ConOps may reflect ‘out-of-the-box’ thinking decided. These are the basis for the measurement system
and is not concerned with immediate perceptions of imple- acceptance criteria.
mentation feasibility. Because a ConOps describes a mea- The design contract for the measurement system should
surement system yet to be designed, it is necessarily spec- be based on the ConOps and the MOEs. Of course, this
ulative – a vision, though hopefully not a hallucination. is what Customer Oriented, Voice of the Customer, and
To many readers and especially to system designers and similar ministrations in a variety of business publications
developers, a ConOps can be irritatingly vague, presump- have meant for the last 10 years.
tive, or unrealistic. However, a ConOps should be judged Products can fail in the marketplace because the contract
solely on how well it describes the situation in which the document was based on the Specification for the System
measurement system is to be used. rather than on its value to the system users.
A ConOps is not the same as a Use Case. The ConOps is Time and again, marketing and prospective customers
a big picture scenario of the total usage envisioned, whereas collaborate to describe the problem that exists and how
a Use Case presumes to describe what the system will do they would like their world to be, and even the tolerable
internally to serve each user. Use Case Scenarios are similar therapies they will endure to get there only to watch
to Operations Concepts rather than to ConOps. as engineering people proceed to reexpress the intent in
the form of requirements, then ‘manage requirements’. In
response to this, they can fend off corrections with threats
2 THE CONCEPT OF OPERATIONS of cost and delivery time explosion.
APPLIED TO SYSTEM DEVELOPMENT After consensus is reached on the ConOps, MOEs, and
acceptance criteria, the system development process can
A ConOps often serves to converge multiple stakehold- proceed to draft Use Cases about each of the stimuli (and
ers toward a common understanding regarding the usage their sources) on the system. These begin to reveal the
and value of the desired measurement system. This should system requirements.
be accomplished before getting into details about what the
measurement system will look like, what technologies it
will employ and what its cost and development schedule 3 ALLOCATIONS OF MEASURES OF
are expected to be. In other words, it is not about imple- EFFECTIVENESS TO THE CONOPS
mentation detail. MODEL
The ConOps description included all of the stimuli and
responses to be monitored by the measurement system. Although a measurement system may have to handle hun-
These include all interfaces between the intended measure- dreds of stimuli and generate responses for each, it is
ment system and the User, Phenomenon System, and all necessary that the MOE is the minimum set that adequately
collateral systems. Interface types must be specified. Exam- characterizes the effect of the measurement system on the
ples are as follows: user situation. Some call these the system’s critical issues
(CI) that then can be analyzed to reveal its MOEs – see
• Coexist = two or more entities in contact.
also Article 14, Hierarchical Aspects of Measurement
• Compete = two or more entities in tension regarding
Systems, Volume 1.
contact.
The generic MOEs for a measurement system each
• Connect = two or more entities structurally involved.
address and include the following:
• Conduct = two or more entities passively exchange
energy. Adequate: The system must measure and report enough
• Communicate = two or more entities passively ex- attributes of the phenomenon of interest to satisfy the user
change data. needs. For example, if the phenomenon is a single event,
• Cooperate = two or more entities change behavior in then the measurement of that event is adequate. However, if
order to exchange energy or information (usually as the phenomenon is the popularity rating of a politician, then
function of energy or information). dozens of attributes may have to be measured as well as the
• Collaborate = two or more entities change structure or demographics and psychographics of those interviewed.
content in order to exchange energy or information
(usually as a function of result being pursued). Accurate: The inherent precision and uncorrectable dis-
turbance error in each measurement are important as is the
This level of information enables an analysis of the frequency of measurement and duration of observation.
effect of all the responses relative to user objectives and
goals. These effects are the basic Measures of Effectiveness. Timeliness: This factor concerns the time lapse from
Next, the acceptable performance levels for each MOE are (a) occurrence of the phenomenon to (b) the availability
424 Measuring System Design
of the measurement in useful form. A highly accurate 4 DOCUMENTATION
measurement too late is not valuable to the user.
The ConOps is often produced as a file document but the
Interoperability: This factor concerns the behavior of
better ConOps is formed as an executable model (or a model
the measurement system with respect to other systems in
and simulation) such that the dynamic interaction of the
the operational situation. This goes beyond mere existence
ConOps factors can be observed.
of interfaces to consider the interoperation scenario (e.g.
control vs. slave and multiple modes of interoperation). Regardless of form, the following checklist of ConOps
content should be used:
Resilience to operator error(s): Often called the ease
1. Mission of the Measuring System (MS).
of use, which is too ambiguous; the better term is the
2. Scope/context of the MS.
effectiveness factor of the measurement system’s resiliency
3. Anticipated phenomenon system (content, structure,
to disturbances caused by operator error, and especially to
a scenario of errors. behavior, pertinent exogenous attributes, pertinent
endogenous attributes, persistence, variability, glint,
Operational Availability index: This factor describes scintillation, position, etc.).
the likelihood that the measurement system will be opera- 4. Anticipated User objectives and resulting needs for a
tional at all times it is required to be operational. The opera- measurement system, such as
tional availability (OA) index is a combination of reliability,
maintainability, and operations calendar. An important sub- • the existence of something, both the exogenous
factor is the calendar date for the first intended use of the attributes and the endogenous attributes;
measurement system. This, of course, is an attribute of the • the location of something, including the related
measurement system development project rather than of the factors;
measurement system itself, but the OA index goes to zero • the persistence of something, including the related
if the measurement system is not delivered on time when factors;
it is needed. • the variability exhibited by that something (such
as range of values, the rep rate of pertinent
Total cost of ownership: Total cost of ownership (TCO) changes, the least count of observations).
is important even if the cost is not the direct responsibility
of the user. 5. Desired observation and interpretation scenario
The use of Standards of Acceptance is illustrated in (regardless of implementation feasibility). Includes
Table 1. For each MOE (generic ones are used here), the usage of the measurements, such as
acceptable values for each are specified. The ‘X(i)’ values
indicate the capability of the system if it is to be of • surveillance,
maximum value to the users. The ‘Y(i)’ values indicate • metrication,
the effectiveness level that the users will find acceptable. • alarm,
Because the TCO of the system typically increases with the • control.
level of system effectiveness, a measurement system can be
Also describes the roles, accountabilities, authorities
judged acceptable while significantly below the Max Value
interfaces and responsibilities of the operations per-
Threshold.
sonnel.
Table 1. Relationship of MOE factors to quantified level of each 6. The boundary of the MS (and what is not MS such
MOE. as the transport system that positions the MS or the
MOE Max Value Acceptance reference standards that enable calibration of the MS.
Threshold Threshold 7. Any technology constraints or opportunities consis-
tent with operational considerations.
Adequate X1 Y1
Accurate X2 Y2 8. The permissible support and maintenance scenario for
Timely X3 Y3 ensuring OA of the MS.
Interoperability X4 Y4 9. Test capability the MS must provide or endure.
Resilience to operator X5 Y5 10. Operator readiness capability the MS or a supporting
error(s) subsystem must provide.
Operational availability X6 Y6
index 11. The preferred or permissible Cutover scenario.
12. The MS evolution that must be provided for within
Total cost of ownership X7 Y7
the OA timeline.
Principle of Concept of Operations (ConOps) 425
13. The interoperability with related systems (communi- FURTHER READING
cating, cooperating, collaborating, coevolving).
14. Permissible environmental impact considerations. Buede, D.M. (2000) The Engineering Design of Systems, Wiley,
15. Permissible societal impact considerations. NY, (p. 42).
Wray, R. (ed.) (2000) Systems Engineering Handbook, version
Effort expended in developing a sound ConOps can save 2.0, INCOSE, Seattle, WA, URL: http://www.incose.org.
the existence of late errors that are costly to fix. Errors are
far less costly when fixed early.
68: Setting the System Boundaries
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
combination of the associated components justifies treating
1 Boundary Perspective of Measurement them as systems. An exemplary aspect of the systems
Systems 426 approach is the totality of the considerations that are taken
2 Extrinsic and Intrinsic Boundaries 427 into account.
3 Application of Boundary Concept to Systems A basis for highlighting all of the effects impacting
Identification 428 on a measurement system may be adapted from
4 Influences on a Real System 428 the boundary view of human–machine supersystems,
given in McGhee, Henderson and Sankowski (1986),
Related Articles 429
in McGhee et al. (1996), and in Article 6, Substructure
References 429 of Human–Machine Systems, Volume 1 where it has
been shown that this boundary perspective indicates the
importance of a detailed consideration of all impacts in
1 BOUNDARY PERSPECTIVE OF measurement systems. It also allows a distinction to be
MEASUREMENT SYSTEMS drawn between the three groups of inputs and outputs
as shown in Figure 1, where the systems engineering
The wider system, within which a human–machine system approach, described in Article 5, Humans in the Real
is embedded, is sometimes referred to as the supersys- World, Volume 1; Article 2, Organization of Instrument
tem (McGhee, Henderson and Sankowski, 1986; McGhee Science, Volume 1; and Article 6, Substructure of
et al., 1996). Human–Machine Systems, Volume 1, can be applied
Since each of the subsystems, described in Article 6, fruitfully by visualizing a boundary view of an
Substructure of Human–Machine Systems, Volume 1, instrumentation system.
may also be considered as systems in their own right, this In the same way as for human–machine systems, as
means that the human–machine system is the supersystem described in Article 6, Substructure of Human–Machine
in which each is embedded. Hence, it can be seen Systems, Volume 1, the three types of input and output may
that measurement systems, which may be considered as be grouped as Resources, Information, or Contamination.
subsidiary blocks of the information system, cannot exist These inputs and outputs in measurement systems may
in isolation. appear in any of the COMETMAN energy forms described
The human–machine system environment, or supersys- in Article 109, Structure and Energy in Sensor Systems,
tem, within which they function, regards them as subsys- Volume 2. The principal resources in measurement systems
tems of itself. Hence, the bilateral impacts of this linking are the energy forms involved in supporting the flow of
must be considered. information.
Instruments perform a purposeful function. They also In the case of a measurement system, there is no
possess an opaqueness and obscurity of structure, charac- need to emphasize the presence of a resource output
teristic of complexity. This functional complexity of the since the information output is also the resource output.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Setting the System Boundaries 427
COMETMAN
Contamination/interference
energy/information
[Ec/Ic]
C C
O O
M M
Measurand/input Output
E DM system E
energy/information energy/information
T T
[Em/Im] [E0/I0]
M M
A A
N N
[Es/Is]
Resource/support energy
COMETMAN
Figure 1. Boundary impacts in an instrumentation system.
Contamination, which is also known as an influencing principles of recursion and reticulation (or subdivision)
input, or simply as interference, consists of disruptions, of systems. Consistent with this view is the possibility
impurities, and disturbance power flows. Although not of distinguishing between extrinsic and intrinsic system
shown in Figure 1, it is generated at the output of the, say, boundaries. The reticulation, or breaking down, is based
digital measurement (DM) system represented in this figure. upon the methods of taxonomy described in Article 2,
It may also appear as interference in other information Organization of Instrument Science, Volume 1.
carrying channels. Disturbance power flows are the primary Consequently, a boundary may be visualized between the
source of deviations and errors in measurement systems. measurement system and the human–machine supersystem.
This boundary may be referred to as an extrinsic, or exter-
2 EXTRINSIC AND INTRINSIC nal, boundary (McGhee, Henderson and Sankowski, 1986;
BOUNDARIES McGhee et al., 1996) as represented in Figure 2. Impacts
from outside the instrument penetrate the extrinsic boundary
It is also important to distinguish whether the sources and cause effects within the system. These external inter-
of contamination occur from inside or outside the ference sources distinguish the extrinsic boundary from the
measurement system. The systems approach uses the intrinsic, or internal, boundary.
Extrinsic boundary
Controlling Parameter Internal
input fluctuations fundamental
fluctuations
Information/ Information/
energy input Intrinsic boundary energy output
Instrument
system
Support energy form
(Resource)
Direct input Disturbances and Direct output
contamination fluctuations in contamination
power flows
Figure 2. Extrinsic and intrinsic boundaries in instrument systems.
428 Measuring System Design
Linking the extrinsic boundary with another boundary system. It is also valid to look upon this as a representation
called an intrinsic boundary, as shown in Figure 2, provides of an identification channel.
a way of classifying different extrinsic or intrinsic interfer- In the case of identification, the extrinsic boundary of
ence sources. Taking an intrinsic boundary into account Figure 2 may be appropriately called the identification
allows a separation and distinction between the various boundary. Impacting upon this boundary are the demand
impacts within the system from those outside the system. input and the contaminating influences. These interferences
Thus, modeling instruments using the extrinsic and can affect the system in a direct manner at input and
intrinsic boundaries illustrated in Figure 2 is essentially the output. Alternatively, they may be distributed throughout
same as modeling, for example, an operational amplifier. the system. An important class of extrinsically arising
The real operational amplifier is enclosed within an imag- interference is that caused by disturbances and fluctuations
inary extrinsic boundary. An idealized noise and drift-free in the various types of power flow.
operational amplifier is contained within its intrinsic bound-
ary. The internal noise and drift in the real amplifier are
accounted for by including appropriate models for these 4 INFLUENCES ON A REAL SYSTEM
effects within the space between the intrinsic, or internal,
boundary, and the extrinsic, or external, boundary. Propos- As systems operate in a real environment, they are subjected
ing extrinsic and intrinsic boundaries allows a separation to the full range of influences, which this environment can
and distinction between effects in the system due to external exert. The boundary perspective provides an important aid
or internal causes. It should be noted that inputs and outputs to the visualization of these impacts, which may be from
in Figure 2 could be in any of the COMETMAN energy energy or information sources. Energy can be classified in
forms. They have not been included for the sake of clarity. a number of different ways, as described more fully in
The COMETMAN acronym is explained in Article 109, Article 109, Structure and Energy in Sensor Systems,
Structure and Energy in Sensor Systems, Volume 2. Volume 2. The classification of energy using the COMET-
MAN acronym, which may be regarded as complement-
ing other energy classification schemes (see Article 109,
3 APPLICATION OF BOUNDARY
Structure and Energy in Sensor Systems, Volume 2), is
CONCEPT TO SYSTEMS useful because it is a good memory aid. It may also be
IDENTIFICATION conveniently extended to the measurement system shown
in Figure 2.
Systems Identification (Eykhoff, 1984; Ljung, 1985; Söder- The inputs then become the energy and signal form
ström and Stoica, 1988), which has special importance in of the input, the resource energy form supporting the
model building for control engineering purposes, is essen- flow of information, and the energy and signal form of
tially a problem in high precision inferential measurement the contamination. Thus, the resource and contamination
(Finkelstein, 1994). It is shown in Article 5, Humans in classes have a possible total of eight energy form inputs
the Real World, Volume 1 that inferential measurement is each. The signal at the output of the identification channel
one of the two classical problems in measurement. has captured information about the ability of the system
It is concerned with inferring the physical properties to store, transmit, transform, or dissipate energy. This
of a system, which characterize its physical ability to information is the resource output of the channel.
store, dissipate, transmit, or transform energy. The close The intrinsic boundary is subjected to two kinds of
symbiosis between Systems Identification and Physical internally occurring disturbance or interference.
Measurement, which is manifested in their many and The first of these gives rise to inevitable fundamental
diverse similarities, gives the identification problem a limits, which are important in all information handling sys-
predominant relevance in measurement. tems. If the information handling operation is measurement,
This example provides a convenient means of demon- then a fundamental measurement limit exists. Fundamental
strating the importance of a boundary perspective for sys- fluctuations occur as a result of the equipartition of energy
tems. This view is especially appropriate, as the systems’ between the thermodynamic state and the states of the other
context of identification highlights the importance of the energy forms.
holistic aspects of the Systems Engineering approach when In mechanical systems, this is manifested in Brownian
applied to Instrumentation Systems (McGhee, Henderson motion, while Nyquist or Johnson noise occurs in electrical
and Sankowski, 1986; McGhee et al., 1996). From the van- systems, and temperature fluctuations occur in thermal
tage point provided by this view, it is possible to propose systems – the latter being also the result of mechanical
that Figure 2 is a general representative model of a process molecular motion. Although the fundamental fluctuations
Setting the System Boundaries 429
have been shown at the output of the system in Figure 2, RELATED ARTICLES
in practice they are more likely to be distributed throughout
the system. Article 5, Humans in the Real World, Volume 1; Arti-
Components used to build systems are not everlasting. cle 6, Substructure of Human–Machine Systems, Vol-
Aging of the elements gives rise to a change in their prop- ume 1; Article 63, Outline of Systems Thinking, Vol-
erties, which can affect performance. If burnout failure ume 2; Article 95, Human–Machine Interface, Vol-
occurs, then a disruption of the operation of the system ume 2; Article 109, Structure and Energy in Sensor
occurs. Parameters of the sensor system, which are instru- Systems, Volume 2.
mental in determining the performance of the system, give
the system some level of parameter sensitivity. Parameter
sensitivity is caused by the inexorable aging of all of the
REFERENCES
components that make up the system structure.
Eykhoff, P. (1984) Identification Theory: Practical Implications
All of the above sources of interference, in whatever and Limitations. Measurement, 2, 75–84.
energy form they manifest themselves, must be considered
Finkelstein, L. (1994) Measurement and Instrumentation Sci-
in data measurement systems. Any synthesis or analysis of ence – An Analytical Review. Measurement, 14(1), 3–14.
measurement systems that does not take them into account
Ljung, L. (1985) System Identification – Theory for the User,
will almost certainly have very serious shortcomings. Inter- Prentice Hall, Englewood Cliffs, NJ.
fering influences, which arise from inside the measurement
McGhee, J., Henderson, I.A. and Sankowski, D. (1986) Functions
system, are due to either parameter sensitivity or inter- and Structures in Measurement Systems: A Systems Engineer-
nal noise. ing Context for Instrumentation. Measurement, 4(3), 11–119.
External interference is caused by disturbance power McGhee, J., Henderson, I.A., Kulesza, W. and Korczynski, M.J.
flows or by other forms of direct contamination. Highlight- (1996) Scientific Metrology, ISBN 83-904299-9-3, printed by
ing the various sources of contaminating inputs exposes the A.C.G.M. LODART, Lodz.
limits of the ultimate resolution of a measurement system Söderström, T. and Stoica, P. (1988) System Identification, Pren-
in executing its primary function. tice Hall, Englewood Cliffs, NJ.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
69: Requirements Allocation
Andrew Kusiak
The University of Iowa, Iowa City, IA, USA
Fang Qin
The University of Iowa, Iowa City, IA, USA
A methodology for hydraulic circuit synthesis based on
1 Introduction 430 functional and structural design building blocks is pre-
2 Requirements Forming 430 sented in Kota and Lee (1990). Kannapan and Marshek
3 Decomposition 431 (1990) presented procedures for specification and synthesis
using algebraic and logical transformation rules. Rinderle
4 Solution Approach 431
and Finger (1990) used a graph-based language to describe
5 Structural Decomposition 432 the behavioral specifications of a design as well as the
6 Module Design 433 behavior of the available physical components. Morrell
7 System Design 435 (1988) described the concept of quality function deploy-
8 Vending Machine Configuration 435 ment (QFD) applied to improve product quality based on
References 436 customer requirements.
1 INTRODUCTION 2 REQUIREMENTS FORMING
Design is information-processing activity resulting in the Design requirements are ‘demands’ and ‘wishes’ that clar-
creation of an object. An early design stage is referred to as ify the design task in the space of needs (Pahl and Beitz,
conceptual design (Pahl and Beitz, 1988) – see Article 64, 1988). They provide an abstraction of the design task
Executing A Measuring System Design, Volume 2 and from the most general demand (the overall requirement) to
Article 66, Phases of System Life Cycle, Volume 2. In more specific demands (subrequirements). A design func-
conceptual design, design requirements are transformed into tion indicates the behavior that is required for the device
a functional and then physical description. At an early to satisfy a given requirement (Kota and Ward, 1990).
design stage, requirements are formed and then allocated Requirements and functions are domain specific and rep-
to functions. This article covers the methodology. resent part of the knowledge base of the design system. A
To date, relatively little research has been done in the designer, as the needs or his/her design experience changes,
area of design specifications. Kota (1990) developed a can also add them.
function-decomposition hierarchy to identify a set of basic A design task is given overall requirements that are
design building blocks. Qualitative functional specifications domain dependent. They are further decomposed into a set
and synthesis for conceptual design of microelectrome- of subrequirements. The number of levels of requirements
chanical systems are presented in Crary and Kota (1990). depends on the complexity of the design task.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Requirements Allocation 431
R1
R2 R3 R4
Specify or
select a
function
R5 R6 R7 R8
Specify or Specify or
select a select a
function function
R9 R10 R11 R12
Specify or Specify or Specify or Specify or
select a select a select a select a
function function function function
Figure 1. Decomposition of requirements.
Figure 1 shows the overall requirement R1, its decom- ItemA_out
position into subrequirements, and the corresponding func- Nickel_in ItemB_out
tions (Kusiak and Szczerbicki, 1992). Dime_in ItemC_out
Quarter_in
An arc between nodes of the tree in Figure 1 represents a
conjunction. A node without an arc represents a disjunction. Nickel_out
In the example presented in Figure 1, the overall require- ItemA_select Dime_out
ItemB_select Dimes_out
ment R1 is satisfied by each of the following four sets of
ItemC_select Quarter_out
subrequirements:
Return_money
Return_money_out
1. {R11, R8}
2. {R12, R8} Reset Excess_money
3. {R10, R5, R3} Clock
BCD_low [0…3]
4. {R9, R5, R3} BCD_high[4…7]
Each of the above sets may lead to a different conceptual Figure 2. Model of a vending machine controller.
design. Some such designs are discussed in Kusiak (1999).
deposited since the last transaction (Return money out will
be activated). The machine accepts the money only up to
3 DECOMPOSITION
95 cents. If more change is inserted, then the Excess money
output variable activates a bypass chute and extra change
Decomposition is useful in analysis of requirements and
inserted is routed directly to the change return. Two 4-bit
functions (Kusiak 1999). It reduces the complexity of the
binary coded decimal code combinations represent the cur-
problem at hand. The application of decomposition will be
rent amount of money (in cents) the user has deposited in
demonstrated here using the design of a vending machine
the vending machine.
controller. The controller of a typical vending machine has
9 inputs and 11 outputs as shown in Figure 2.
The vending machine needs to be designed for purchas-
ing items, for example, A, B, C, by a providing a change 4 SOLUTION APPROACH
and then selecting an appropriate button to get the item. If
the change exceeds the price of the item selected, the vend- Having formed the design requirements for the vending
ing machine returns the difference. If the Return money machine, the next step is to analyze the various Product
button is selected, the controller will return all the coins Functions. The main functions are now discussed.
432 Measuring System Design
1. Coin counting Select item
The function ‘count coins’ of the vending machine is
illustrated in Figure 3.
Change
2. Money return Change calculation
The vending machine returns change if the button of Total amount
Return money is selected (see Figure 4).
3. Excess money
Figure 7. Change calculation.
The excess of 95 cents is returned (see Figure 5).
4. Item selection
When the customer selects an item and the amount Dime_out
deposited in the vending machine is equal to or greater
than the price of the selected item, the item is delivered Change Dimes_out
Change out
(see Figure 6).
Note: Item select includes itemA select, itemB select,
Quarter_out
itemC select.
Item out includes itemA out, itemB out, itemC out.
5. Change calculation Figure 8. Change out.
If the customer drops in more money than the price
of the item selected, the change needed is calculated.
Total BCD_high
The change equals the total amount deposited in the
BCD indicator BCD_low
Change
Nickel_in
Figure 9. BCD indicator.
Dime_in Coin counting
Total
Quarter_in
vending machine minus the price of the selected item
(see Figure 7).
Figure 3. Coin counting.
6. Change out
After the vending machine has calculated the amount
of change, the proper change amount is returned to the
customer (see Figure 8).
7. BCD indicator
Money return
Return_money Return_money_out The vending machine needs to indicate the current
amount of money (in cents) the user of the vending
machine has deposited with two 4-bit BCD code com-
Figure 4. Return money.
binations (see Figure 9).
5 STRUCTURAL DECOMPOSITION
Excess money
Total amount > 95 Excess_money
On the basis of the functions defined in Figure 2 through
to Figure 9, the function-input/output matrix is formed (see
Figure 5. Excess money. Figure 10).
Each “*” in Figure 10 indicates a relationship between
the corresponding function i, i = 1, . . . , 7 and input/output
Item_select
j , j = 1, . . . , 16.
A list of functions is obtained from the rows in Figure 10:
Item selection Item_out
1. Coin counting
Total amount 2. Return money
3. Excess money
Figure 6. Item selection. 4. Item selection
Requirements Allocation 433
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 5 6 7 8 9 11 12 13 14 15 16 4 10
1 * * * * 1 * * * *
2 * * 2 * * Module 1
3 * * 3 * *
4 * * * 4 * * Module 2 *
5 * * * 5 * Module 3 * *
6 * * * * * 6 * * * * Module 4 *
7 * * * * 7 * * * *
Figure 10. Function-input/output incidence matrix. Figure 11. Function-input/output matrix of vending machine
clustered with the branch-and-bound algorithm.
5. Change calculation
6. Change out that is more suited to design allocation. Four groups of
7. BCD indicator. functions – {1, 2, 3}, {4, 5}, {6} and {7} – and four groups
List of inputs/outputs (columns in Figure 10): of inputs/outputs – {1, 2, 3, 4, 5, 6, 7}, {8, 9}, {11, 12, 13,
14} and {15, 16} – are visible in Figure 11. Inputs/outputs
1. Nickel in 4 and 10 interact with more than one function.
2. Dime in Introducing redundant inputs/outputs for performing the
3. Quarter in functions allows decomposition of Figure 11 into the four
4. Total amount mutually separable submatrices shown in Figure 12 (for
5. Return money details see Kusiak (1999)).
6. Return money out
7. Excess money
8. Item select (includes ItemA select, ItemB select,
ItemC select) 6 MODULE DESIGN
9. Item out (includes ItemA out, ItemB out, ItemC out)
10. Change
11. Nickel out The four function modules are introduced to fulfill the
12. Dime out requirements from Figure 12. VHDL logic is now used to
13. Dimes out describe the behavior of each module (Perry, 1994).
14. Quarter out
Module 1: Coin handler
15. BCD high
A coin handler counts coins, calculates the total amount,
16. BCD low.
returns money, and handles excess money according to the
Clustering the matrix with the branch-and-bound algo- VHDL logic presented next and the function illustrated in
rithm (Kusiak, 2000) results in a new matrix (Figure 11) Figure 13.
1 2 3 4 5 6 7 4’ 8 9 10 11 12 13 14 10’ 4” 10” 15 16
1 * * * *
2 * * Module1’
3 * *
4 * * * Module 2’
5 * * * Module 3’
6 * * * * * Module 4’
7 * * * *
Figure 12. Structured function-input/output matrix of vending machine after introduction of redundant inputs/outputs.
434 Measuring System Design
Nickel_in Total_amount The price of each of the three items A, item B, and item
Dime_in C is listed in Figure 15.
Return_money
Quarter_in Coin handler
1. Select Item A
Excess_money
Return_money_
IF ItemA select = ‘1’ THEN
IF total amount >= 55 THEN
Figure 13. Coin handler. ItemA out = ‘1’;
Change := total amount −55;
END IF;
1. Count coins END IF;
IF nickel in = ‘1’ THEN
total amount := total amount + 05; 2. Select Item B
END IF; IF ItemB select = ‘1’ THEN
---------- IF total amount >= 65 THEN
IF dime in = ‘1’ THEN ItemB out = ‘1’;
total amount := total amount + 10; Change := total amount −65;
END IF; END IF;
----------
END IF;
IF quarter in = ‘1’ THEN
total amount := total amount + 25;
END IF; 3. Select Item C
---------- IF ItemC select = ‘1’ THEN
2. Return money IF total amount >= 75 THEN
ItemC out = ‘1’;
IF return money = ‘1’ THEN Change := total amount −75;
return money out = ‘1’; END IF;
END IF; END IF;
3. Handle excess money Module 3: Change maker
IF total amount > 95 THEN Change maker outputs provide change to the customer
excess money = ‘1’; (see Figure 16).
END IF; The input/output of the change maker is shown in
Figure 17:
Module 2: Item processor
The following VHDL statements illustrate the process of
The Item processor accepts purchase requests and deter-
handling a change of 35 cents.
mines if the money entered is enough to purchase an item
(see Figure 14). IF change = 35 THEN
Nickel out = ‘1’;
Item_out
Dime out = ‘1’;
Item_select Dimes out = ‘1’;
Quarter out = ‘0’;
Total amount := 0;
Item processor Change := 0;
END IF;
Total_amount Change
Nickel_out
Figure 14. Item processor.
Dime_out
Item Price(cents) Change
A 55 Change maker
Dimes_out
B 65
C 75 Quarter_out
Figure 15. Item price list. Figure 16. Change maker.
Requirements Allocation 435
Amount
Change Nickel_out Dime_out Dimes_out Quarter_out BCD_high BCD_low
(cents)
(cents) (5 cents) (10 cents) (20 cents) (25 cents)
0 0000 0000
0 0 0 0 0 5 0000 0101
10 0001 0000
5 1 0 0 0
15 0001 0101
10 0 1 0 0 20 0010 0000
15 1 1 0 0 25 0010 0101
30 0011 0000
20 0 0 1 0
35 0011 0101
25 1 0 1 0 40 0100 0000
45 0100 0101
30 0 1 1 0
50 0101 0000
35 1 1 1 0 55 0101 0101
60 0110 0000
40 1 1 0 1
65 0110 0101
Figure 17. Input/output table of the change maker. 70 0111 0000
75 0111 0101
80 1000 0000
Total_amount BCD_high
85 1000 0101
90 1001 0000
BCD indicator
Change BCD_low 95 1001 0101
Figure 19. Input/output of BCD indicator.
Figure 18. BCD indicator.
Coin Total Item Change Change
Module 4: BCD indicator
handler processor maker
Two 4-bit BCD code combinations represent the current
amount of money (in cents) the user of the vending machine
has deposited. Before purchase, the amount equals the total
amount, after purchase, the amount equals the change (see BCD
indicator
Figure 18).
The input/output of change maker is presented in
Figure 19. Figure 20. Inner connections of modules of vending machine.
Suppose the amount is 35, the following VHDL sequence
is used: 8 VENDING MACHINE CONFIGURATION
IF amount = 35 THEN
BCD LOW = ‘‘0101’’; After the module design is finished and the inner con-
BCD HI = ‘‘0011’’; nections established of modules, the top-level vending
END IF; machine configuration is obtained. Clock and Reset are
added to synchronize the operation of each module (see
Figure 21).
7 SYSTEM DESIGN This process has systematically worked its way through
the many options. By reorganizing the relationship matrix,
7.1 Module interaction several units have been identified that have less interfaces
to other units than was first developed. With less interfaces,
From Figure 11, it was determined that input/output 4 (total the chance of system failure is reduced and design made
amount) and input/out 10 (change) interact with multiple simpler. Measuring systems can similarly be broken down
modules. The relationship of the modules is obtained from in this manner. More information on the nature and scope
Figure 11 and shown below. Thus, the inner connections of of requirements is found in Sage and Rouse (1999) and
modules are established (see Figure 20). Sydenham (2004).
436 Measuring System Design
Nickel in
Return money
Dime in
Excess money
Quarter in Coin
handler
Return money
Total amount
Item out
Item select Item Nickel out
processor Change
Change Dime out
Clock maker Dimes out
Quarter out
Reset
BCD BCD high
indicator
BCD low
Figure 21. Vending machine.
REFERENCES Kusiak, A. (1999) Engineering Design: Products, Processes, and
Systems, Academic Press, San Diego, CA.
Crary, S. and Kota, S. (1990) Conceptual Design of Micro- Kusiak, A. (2000) Computational Intelligence in Design and
Electro-Mechanical Systems, in Proceedings of the Micro Sys- Manufacturing, Wiley, New York.
tems Conference, Berlin.
Kusiak, A. and Szczerbicki, E. (1992) A Formal Approach to
Kannapan, S.M. and Marshek, K.M. (1990) An Algebraic and Specifications in Conceptual Design. ASME Transactions: Jour-
Predicate Logic Approach to Representation and Reasoning nal of Mechanical Design, 114, 659–666.
in Mechanical Design. Mechanism and Machine Theory, 25,
335–353. Morrell, N.E. (1988) Quality Function Deployment; Disciplined
Quality Control. Automotive Engineering, 9, 122–128.
Kota, S. (1990) Qualitative Motion Synthesis: Towards Automat-
ing Mechanical Systems Configuration, Proceedings of the Pahl, G. and Beitz, W. (1988) Engineering Design, Springer-
NSF Design and Manufacturing Systems Conference, Society Verlag, New York.
of Manufacturing Engineers, Dearborn, MI, (pp. 77–91). Perry, D. (1994) VHDL, McGraw-Hill, New York.
Kota, S. and Lee, C.-L. (1990) A Computational Model for Con- Rinderle, J.R. and Finger, S. (1990) A Transformational Approach
ceptual Design: Configuration of Hydraulic Systems, Proceed- to Mechanical Design Synthesis, Proceedings of the NSF
ings of the NSF Design and Manufacturing Systems Confer- Design and Manufacturing Systems Conference, Society of
ence, Society of Manufacturing Engineers, Dearborn, MI, (pp. Manufacturing Engineers, Dearborn, MI, (pp. 67–75).
93–104).
Sage, A.G. and Rouse, W.B. (eds) (1999) in Handbook of Systems
Kota, S. and Ward, A.C. (1990) Functions, Structures, and Con-
Engineering and Management, (ed. A. Sage), Wiley, New York.
straints in Conceptual Design, in Proceedings of The 2nd Inter-
national Conference on Design Theory and Methodology, (ed. Sydenham, P.H. (2004) Systems Approach to Engineering Design,
J. Rinderle), Chicago, IL, (pp. 239–250). Artech House, Boston, MA.
70: Measuring System Design Methodologies
Ludwik Finkelstein
City University, London, UK
signal into a symbol that may be conveniently handled
1 Measuring System Design Task 439 by the following block, which performs any required
functions of information transformation and communi-
2 The Design Process 440
cation.
3 Design Concept Generation 441 This system passes the information to the effector block,
References 442 to further processing, or to the human operator. The
measuring instrument system operates under the control
of a control block. An important part of the system is
the operator or human–machine interface. Through this
1 MEASURING SYSTEM DESIGN TASK interface, the operator effects supervisory control of the
measurement process. The interface embodies also any
The design of a measuring system is the transformation displays.
of knowledge about the measurement task to be per- Measuring system analysis and design uses description
formed into a specification of a system that can perform by mathematical or like models – see Article 71, Modeling
the required task in such a way that the system can be Methodology, Volume 2; and Article 104, Models of the
realized. Measurement Process, Volume 2.
The methodology of measuring system design is deter- Models involve abstraction, that is, the omission of
mined by the fact that measurement is an information irrelevant detail. This is required to reduce complexity,
process, measuring instruments are information machines, ensure tractability, and give access to analytical tools. In
and that measuring instrumentation can be treated for the the case of instruments, we may distinguish a number
purposes of analysis and design as a system. of levels of abstraction in increasing levels of abstrac-
To discuss the principles of design of a measurement tion:
system, it is convenient to do so on the basis of a simple
model of such a system – see Figure 1. 1. Embodiment models in terms of the physical variables
Measuring systems are invariably a subsystem of a large and the shape, dimensions, and material properties
system. The measuring system consists of a number of of the system components and their physical inter-
subsystems. There is firstly the system under measurement connection.
(also called the system under test SUT). This is connected 2. Models of flow of power and power transformation;
to a sensor system and acts on the sensor by a flow of models of signals and their transformation.
matter or energy. 3. Models of flow of symbols and of symbol transforma-
The sensor converts this flow into a signal, maintain- tion.
ing a functional relation between the input flow and the 4. Models of information and knowledge flow, that is,
information carrying characteristics of the signal. There representation and transformation of information and
is usually a signal-conditioning block that converts this knowledge by the system.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
440 Measuring System Design
Observed Information from
system preceding stage
Sensors Definition
actuators
Concept generation
Signal condition Analysis
Operator
Control
interface
Evaluation
Info. process
Decision
Information to
Effectuation next stage
Figure 2. The design process.
Figure 1. Measuring system. The initial step of the process is that of definition of an
initial design concept to be refined. The model and related
We may also distinguish between functional models that knowledge is, in general, provided by the previous stage of
represent only the transformation of input into output and design. Associated with this model is a set of evaluation
embodiment models that represent some details of the criteria, or a value model, which express the degree to
physical realization of the system. which the concept meets the requirement.
In modern measuring systems, the information process- The next step in the process is the generation of a more
ing block, the control block, and to some extent the effec- refined candidate design concept. The generation of candi-
tuation and signal-conditioning blocks are designed and date design concepts is central to the design process. The
implemented by the standard methods and equipments of candidate design concept is represented as a configuration,
information technology. that is, in terms of attributes of its construction.
In general, functional blocks for performing the required It is next analyzed to determine its performance attributes
operations are available. Special problems arise in the and evaluated using the value model previously established.
sensor block and in the interaction between the measuring A number of such design concepts are normally generated
system and the observed system. They need to be designed and stored.
at the embodiment level. The evaluation step passes information to a decision
step in which the candidate design concept judged to be
the most satisfactory is accepted for implementation, or as
2 THE DESIGN PROCESS the initial concept on which the next stage of design will
be based.
The design of measuring systems follows the classical If none of the candidate design concepts is acceptable, it
model of the design process. may be necessary to return to an earlier stage in the process,
Classically, design activity is represented as a process for example, to alter the value criteria, or to generate more
consisting of a sequence of stages, starting at the perception candidate design concepts.
of need and terminating at the communication of the final Viewed overall, design proceeds from a global view
firm knowledge necessary to implement or realize the of the artifact or system to progressively more localized
artifact, or system. Each stage is itself an elementary design considerations, and from abstract and fluid descriptions to
process, which starts with an initial concept and refines that concrete and firm ones.
concept. The stage is a sequence of steps, subprocesses, The design stages typically start with a definition of the
or operations. overall system that arises from the original requirement.
Consider the elementary design process depicted in This establishes the main features of the system as a set of
Figure 2. interconnected subsystems and also fixes the nature and
Measuring System Design Methodologies 441
specifications of the subsystems. In turn, the subsystem One approach is the systematic listing of physical laws
design stage uses this information to establish the main that involve a physical variable of interest, or that relate two
features of the subsystem and the nature and the specifica- physical variables that are to act as the input and output of
tions of its simpler components. a device. For example, if we require to generate concepts
This sequence of stages proceeds to the design of ele- for sensing electric current, we may list the physical laws
mentary components. If at any stage it appears that it is not involving electric current, each of which may suggest a
possible to meet the specification generated by the preced- design concept. One law is that of force between current-
ing stage, it is necessary to return to an appropriate point carrying conductors, which suggests that we can sense
of that latter design stage. current by sensing the force between two coils, as in a
dynamometer instrument. The law of force on a current-
carrying conductor in a magnetic field suggests the principle
3 DESIGN CONCEPT GENERATION of the moving-coil ammeter. Ohm’s law relating voltage
and current suggests sensing current by sensing the voltage
The core of the design process is the generation of design across a standard resistor. Joule’s law of heat generated by
concepts; its basis is decomposition and abstraction. a current in a resistor suggests the principle of the hot-wire
The components of a measuring system that handle ammeter. This list is illustrative but not exhaustive.
knowledge and information are generally realized by the The second approach is the examination of a single law
selection of available functional blocks and the use of essen- involving the variables in question that then suggests a
tially algorithmic methods of design. The design method- design concept. Consider the generation of design concepts
ology of those parts of the system is not specific to instru- for instruments to measure viscosity. The relevant law is
mentation and is basically that of information technology. F = −A (dv/dx), where F is the force acting on an area
In the case of sensors and the interaction between the A normal to the velocity gradient dv/dx of a fluid. This
sensors and the object under observation, it is commonly suggests two viscometer principles:
necessary to find an appropriate working principle and a
• generating a velocity gradient in a fluid and sensing the
physical embodiment of the principle. resultant shearing force;
There are a number of basic methods of determining such • generating a shearing force and sensing the resultant
working principles and finding their embodiments. They are velocity gradient.
useful heuristics rather than algorithmic procedures.
From this we can proceed sequentially to elaborate vari-
ants of the principles. Taking two solid bodies with fluid
3.1 Use of established equipment and design between them, we may have the two bodies fixed with fluid
concepts moving, or one body fixed with the other body moving, or
all may move. The mode of motion may be translation
Established design concepts may be used in the design or rotation, and the motion may be steady, harmonic, or
of a system. They may be retrieved from systematically transient. We may then derive the idea of two coaxial cylin-
organized knowledge bases and tools for searching them. ders with fluid between them. We may drive one cylinder
Formal knowledge bases of sensor principles may be (inner or outer) at constant angular velocity and measure
organized on the basis of their mathematical models or in the torque on either moving cylinder (rotor) or stationary
terms of their function. cylinder (stator). Alternatively, we may drive the rotor at
Catalogs of commercially available sensors and direc- constant torque and measure the resultant velocity of stator
tories of such catalogs are increasingly available on the or rotor. This systematic process then suggests oscillatory
Internet. motion of the rotor. A complete range of variant design
Searching through literature, also increasingly facilitated concepts is thus generated.
by the Internet, is an important method.
3.3 Analogies
3.2 Design concepts deduced from the relevant
physical effects A method of generating design concepts is the consideration
of analogies. When considered in terms of an abstract
Instrument components that maintain a functional relation description, a system may seem to be essentially similar
between physical variables may be deduced from the rele- to another and aspects of the second may suggest design
vant physical effects. solutions for the first.
442 Measuring System Design
Physical analogies such as those between electrical, act as a starting point for generating new concepts by
mechanical, fluid flow, and thermal systems are often transformation. Starting with a particular design, one may
useful. proceed to derive new concepts.
Another important source of analogies for instrumen- The basic steps are the functional description and decom-
tation and information processing are living organisms, position. The instrument is considered as a system of com-
especially the animal system of receptors, nerves, brain, ponents and each component is described in terms of the
and effectors. functions it performs. Variations are then derived by sys-
tematic analysis of the concept and its systematic variation
to generate new options.
3.4 Divergent or lateral generation of concepts Foundational material on engineering design is available
in Finkelstein and Finkelstein (1983, 1994) and Sydenham
Design concepts can be generated by creative thinking (1994, 2004).
in which ideas diverge and move laterally. Design con-
cepts are considered in abstract form, neglecting irrelevant
details until the latest possible stage. Every attempt is made REFERENCES
in the first instance not to consider an established prac-
tice, authority, or apparent obstacles so as not to allow Finkelstein, L. and Finkelstein, A.C.W. (1983) Review of Design
Methodology. Proceedings of the IEE, 130, Part A, 213–221.
promising concepts to be abandoned prematurely. Finally,
for the same reason, idea generation and evaluation are Finkelstein, L. and Finkelstein, A.C.W. (1994) Instruments and
Instrument Systems: Design Principles, in Systems and Con-
separated.
trol Encyclopedia, Vol. 4 (ed. M.G. Singh), Pergamon Press,
Oxford (pp. 2533–2536).
Sydenham, P.H. (1994) Disciplined Design of Sensor Systems.
3.5 Transforming concepts Measurement, Pt 1 Types of Design, 14, 73–80; Pt 2 Knowl-
edge Based Systems, 14, 81–87.
Some existing design concept usually forms the first step Sydenham, P.H. (2004) Systems Approach to Engineering Design,
in the process of generating a design. This may then Artech House, Norwood, MA.
71: Modeling Methodology
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
• laws of physics – formal description
1 Models Relevant to Measurement Systems 443 • process involved
2 Types of Modeling Representation 445 • hardware artifacts – called icons.
3 Modeling Inside Blocks 446
4 Process for Creating Computer-based 1.1 Physics-based formal models
Models 446
5 Setting up the Architecture 447 In these, a real-world activity, existing or not yet existing,
6 Identifying Equations 448 is represented by the formal laws of physics that are
expressed in terms of mathematical equations. The ideally
7 Assignment of Model Coefficients 449
desirable level of representation of the complete enough
8 Exercising the Model 449
real world open system requires extensive mathematical
9 Limitations of Models 450 development that is not always available to the degree
10 Model Validation 450 of completeness needed. Corners are often cut by making
Related Articles 451 unproven assumptions that can be overlooked later to the
References 451 peril of the user.
Much of engineering practice has been modeled by this
physical means, but it has been long recognized that not
all systems aspects can be handled by use of ‘hard’ formal
1 MODELS RELEVANT TO descriptions. Adding in empirical data is sometimes needed.
MEASUREMENT SYSTEMS Examples of this are found in the flight dynamics of aircraft,
temperature distribution on an electronic circuit board, wing
Designers will invariably need to develop some form of lift, and other fluid-flow designs.
model of the system or service being developed or operated
(Neelamkavil, 1987; Holland, 1996). Models enable the
system to be studied in the computer-based virtual domain. 1.2 Process-based models
They allow the designer to study the behavior of a system
before it is physically built. The place of measurement in Here, the modeled domain is represented by a set of heuris-
the life cycle of a system is shown in Figure 1. tic and formal rules embedded into an algorithm or flow
No one modeling tool suits all needs. There exist many diagram. This ‘softer’ thinking approach can sometimes be
factors of the system that will decide which particular tool superior to the hard-science model in terms of applicabil-
has to be used. The first to consider is what basic kind of ity and can often yield a model where the nature of the
model is appropriate. problem totally defeats formal description.
Three types of models are used. They will be based on Examples of these types of models are the queuing
one or more of the following: of traffic in a road system, a logistics supply system in
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
444 Measuring System Design
Engineering system development life cycle → For this reason, there is now a diminishing demand for
physical model-testing facilities. Reduced demand pushes
Concept Assessment Demonstrate Manufacture In-service Disposal
up costs, which can render them unaffordable. Some large
wind tunnels, for example, have been closed down for lack
Modeling and simulation of system of sufficient business. Physical test facilities are now used
more for periodically testing a model under development
Measurement facility creation and testing
and for application in design optimization.
Integration lab testing Setting up a mathematical model to represent the sim-
ple first-level principle of the issue under investigation, as
Hardware-in-the-loop facility testing represented by known laws of physics, is rarely sufficient
for engineering design use. Useful models usually must
Installed system test facility also allow for the numerous second-level effects caused
by the imperfections of real materials, design inadequa-
Open range testing cies, and external influence effects. These greatly extend
the model’s sophistication. Models rapidly expand to be
Distributed testing–new feature today– live, virtual, constructive
complex, needing ever-increasing effort to establish the
particular equation-coefficient values that the model needs
Figure 1. Place of measurements in the systems development life
to run.
cycle.
Over time, these have, however, been developed for some
industries to the point where they can adequately represent
operation such as loading a cargo ship in best time, flows
the need – but at a cost that is not always evident, for much
in a manufacturing facility, and an Expert System control
of the development has come from past projects and before
room support system.
that from university style research. Sophisticated models
This type can also accommodate the softer human
shown as selling demonstrator applications do not reveal
issues – but not that well – using the class of problems
how much effort has been expended in their development!
called Systems Dynamics (SD).
Some regimes have been particularly good for modeling
realistic systems. One example is that of electronic circuitry
1.3 Iconic models where sophisticated models and tools can deliver an almost
complete virtual development environment. Here it is often
An icon is an object representing something. It can be an possible to design the systems in the model domain –
image or a 3-D object. It is used to represent the physically few changes will be needed after it is made. Usually no
existing model of something. prototype build is needed to tune the system.
Examples abound: a scale model of an automobile used Now, computational fluid dynamics (CFD) and mechan-
in a wind tunnel; a full-size aircraft cabin mock-up; and ical and manufacturing design also have tool support for
a scaled-down 3-D model of a large rainwater catch- the design and development that needs minimal final phys-
ment area. ical testing.
Being largely hand-crafted, these Thermal modeling of instrument circuit boards and cases
is also highly developed today.
• are costly to make;
With use of the tools, the new knowledge gained is fed
• take a long time to build;
back to continuously improve the model. The successful
• are easily destroyed or damaged in testing;
development of many quite remarkable tool packages may
• cannot be altered with ease;
give the impression that all engineering design can, and
• need special crafts skills to build;
should be, model based. That is a fine aspiration, but there
• can only represent a small range of conditions;
are many areas where such support is coming slower for
• require costly testing instrumentation installations.
reasons of complexity not allowing adequacy of model-
Once commonplace, they are rapidly becoming a thing ing or, perhaps, because the industry sector is too small to
of the past as the economic means for supporting design. support the costs involved in developing the special tools it
Computer simulation is overtaking the need for these, for would like to use. This is the situation for much of the mea-
that alternative is often superior to an iconic model. suring systems design. For example, it is affordable, indeed
Furthermore, the computer model can also be placed in essential, to develop models of a spacecraft instrument sys-
a virtual world environment to put the system under virtual tem but it is far less so to fully model a custom-designed
test. In this domain, little is physically built. single, weighing platform.
Modeling Methodology 445
In the measuring system field, the problem in tool use is Complex situations will often be formed as hybrid com-
that the particular class of design need often cannot support binations of the above.
the development of the sophisticated tools needed. Derivate Flows of both types in a formal model can be formally
use of tools built for other uses is usually necessary. represented by different descriptions of behavior:
Modeling of the formal part of the information tech-
nology (IT) aspect of a measuring system is, however, • Deterministic
well supported with virtual instrument modeling tools like – Future behavior is formally predicted from knowl-
LabView – see Article 105, Modeling with LabVIEW , edge of the model on a single-point basis.
Volume 2. For example, given the laws of heat flow, the tempera-
Most design regimes – mathematical, electrical, mechan- ture rise of a thermometer allows upcoming values to be
ical, optical, chemical, thermal, and vibrational – have calculated at that point in the system space.
excellent tools available. However, the mixed regime use
needed for many measuring system investigations is often • Stochastic
complicated by the need to integrate the findings of several – Future behavior can also be predicted, but only for
tools that are not directly compatible. It will be necessary the behavior of a group of values having a known
to repeatedly transfer datasets between tools in instrument statistical relationship.
design. The time taken to become truly skilled in the use of
a tool, and its cost, often limits its use to the higher-volume For example, the height of the individual vehicle driver
runs – that are not characteristic of measuring systems. who will next use a truck driver’s seat cannot be predicted,
but the range of heights for the set of likely drivers can
be predicted.
1.4 Aim for simplicity in model building • Chaotic
– Behavior appears to be random but does not follow
Models should be kept as simple as is just necessary to well-known stochastic laws; it appears to be hap-
extract the information needed. hazard, that is, chaotic. Mathematical expressions
Models are NEVER a complete representation of a topic. can handle some forms of chaos using different
They are built to exhibit chosen aspects of behavior and will means from the above two behavior descriptions.
be set to certain limits of dynamic performance.
Static, or steady state, models have their uses, but the For example, work activity in a system engineering
most used are those that allow the dynamic behavior of the process often follows process model descriptions similar
modeled system to be investigated – they are much more to that of a pile of sand grains that is sliding down from
complicated. the top – that has been formalized as sand-pile chaos.
If the external environment conditions are changed from To make a model perform, they need a stimulator (driver,
those for which the model was designed, it is not likely to forcing function, excitation input) to make them represent a
give accurate outcomes. particular situation. The resulting system behavior depends
both on the characteristics of the stimulator and the model.
Stimulators take several forms:
2 TYPES OF MODELING
• Time driven
REPRESENTATION These are the equations given as a function of time,
such as heat transfer as an item is heated. A suitable
Formal models of real-world systems can be formed by
time interval and discrimination is required to use the
virtue of the fluxing entities of a design situation. The entity
model. The time variable can be speeded up or slowed
of interest will either flow as a
down in the formal mathematical models, which is not
• distributed continuous flow (mechanical stresses in the always possible in physical prototypes. This is useful
skin of an aircraft; weather-flow patterns; pollutants for prediction of future behavior and for understand-
entering a water channel; thermals in an electronic ing events that are seemingly stationary or are arising
circuit board) or very fast.
• channeled flows, wherein the ‘substance’ flows in dis- • Event driven
tinct channels and/or with varying discrete arrival times These lead to a process-orientated model, such as oper-
(manufacturing materials flow; public transport move- ators conducting an assembly operation in which their
ments; digital control system; alarm instrumentation task has one kind of statistical variability and the arrival
system). of the parts they need has another.
446 Measuring System Design
3 MODELING INSIDE BLOCKS Such models in engineering often have to use ‘fiddle’
factors to facilitate the modeling when the sufficient under-
Models of a system, or part of it, can be in a black or white standing is not available due to lack of time to develop
box form. more knowledge or where no explanation can be realized.
For example, in fluid-flow modeling, various ‘relationship’
numbers are used, such as Reynolds Numbers.
It is highly useful that the formal descriptive equations
3.1 Black box models
for electrical circuits also are the same as that for the
mechanical and the acoustic regimes. This gives rise to
Here, the internal working of the model is not necessarily what is known as the method of the analogies. Basically,
operating in the same way as the internals of the modeled the DE set for each describes the same simple harmonic
system do. The output/input relationship is, however, a motion (SHM) behavior.
faithful representation of the characteristics of the block’s Real systems often involve a mix of energy regimes. For
ports as it would be in the real thing. example, an audio loud speaker has a mechanical cone sus-
An example of this is the use of a simple low-pass filter pended by an elastic mount being driven in and out of
model to represent the torque-speed characteristics of an the magnet assembly by an electromagnetic moving coil
engine in an overall automobile performance model. Here, actuator. The air column acting on the speaker cone damp-
the output/input does represent the torque behavior of the ens motion. Thus, provided all movements are operating
engine as a whole, but with this model, it is not possible to in the linear mode, the system exhibits SHM. Similarly,
investigate such things as thermal behavior and fuel flow an automobile suspension system is formed with a coil
of the engine. spring and thus has energy storage in the spring and in
the mass of the suspended structure, and there is damping
of the energy effects with the shock absorber. This is also a
3.2 White box models SHM system. Thus, linear SHM systems can all be modeled
with the same set of equations, provided the appropriate
In this form, the internal workings are transparent and coefficients are used. It also allows one regime to be mod-
are modeled as a replica of the internal functions. Each eled with a set up in another. Equivalent electrical Resis-
subsystem, down to an appropriate level, is formed of white tor–Inductor–Capacitor (RLC) models are usually used as
boxes that simulate true internal operation. These can be they suit the regime of operation of the electronic computer.
much harder to build than the black box form and are often Given certain system assumptions (or some trickery), it
not needed. For instance, to model an automobile engine as is possible to write formal equations that relate the behavior
a fully white box model would be the most complex task, of most energy regimes into one set of coherent integrated
for the overall model would have to cover numerous energy differential equations. This is the basis of many modeling
regimes in a fully integrated manner. tools. Details of SHM system behavior are covered in Arti-
Care is needed to appreciate which type of box is being cle 59, Introduction to the Dynamic Regime of Measure-
used in a model. When constructing a model, it is necessary ment Systems, Volume 1; Article 60, Zero-order Sys-
to look well ahead to ascertain whether white box modeling tem Dynamics, Volume 1; Article 61, First-order System
will be needed at a later stage of development. Starting with Dynamics, Volume 1; and Article 62, Second-order Sys-
a black box model realized by, say, system identification tem Dynamics, Volume 1.
methods cannot be delved into to give reticulated behavior
of internal behavior.
Several methods exist for building the formal mathemat- 4 PROCESS FOR CREATING
ical description of models. These are then converted into a COMPUTER-BASED MODELS
suitable form for digital computation. The methods are
There is a clear-cut process for developing a formal, and to
• equation – differential equations (DEs) some extent other forms, model. The following key steps
• finite element – (FE) need to be undertaken with care and quality consideration
• discrete – nodal. (refer to Figure 2).
Engineering system or component models using DEs are 1. Study the situation and all known requirements in
based on there being available an in-depth understanding of order to ascertain the particular features that decide
the dynamic behavior, on the basis of the laws of physics the characteristics of the type of model needed, as are
that can be expressed in DEs. described above.
Modeling Methodology 447
Input I Output O
Development stages →
First box dissected into several black box
functions, forming the measuring chain
….
Each black box is now ‘identified’ internally with
suitable equations– possibly as white boxes
Fnc 1 Fnc 2
To ….. n
O/I = a O/I = Ix + b 2
I O
Output of all stages = Fnc 1 × Fnc 2 × Fnc n ………× Input
(Equations multiplied and simplified)
Figure 2. Simplified development of an identified model of a measuring instrument.
2. Decide the limits of the situation for modeling the The model is ready to use in an investigative manner
subject as an adequately closed system – see Arti- once all of these issues have been decided.
cle 64, Executing A Measuring System Design, Vol- Many general-purpose formal mathematical tools support
ume 2 and Article 68, Setting the System Bound- the above process – examples are MatLab , LabView ,
aries, Volume 2. Ithink , and so on.
3. Set up the architecture of the system in terms of Specialized tools are often available for particular app-
the fluxes of energy or mass, process steps, or dis- lications.
crete activities.
Consideration about which modeling tool to use com-
mences at this stage. The final choice of model and 5 SETTING UP THE ARCHITECTURE
tool may well take place as the result of many trial
starts in which the features of a chosen tool become Generally, measuring system architecture is developed
mainly by the intuition driven by one’s past experience
clearer to the user and are seen to be not suitable
of how things were done before and from the knowledge
after all.
of new technologies and marketed ‘block’ products.
4. Assign blocks and their connectivity to system func-
It is important to also decide where in the hierarchy of
tions that can be described in terms of black or
models does the one being developed sit (Davies, 1996).
white boxes.
The integration of models in a hierarchical architecture
5. Assign a suitable equation for the output/input rela-
needs addressing early in development as this can determine
tionship of each box. key characteristics needed.
6. Assign suitable values to the coefficients to char- Although measuring systems are usually thought of as
acterize the equations of the boxes for the task in small subsystems sitting inside a much larger whole system,
question. some are very large indeed. For example, the model of a
7. Integrating the whole equation set into one total equa- measuring and simulation system for testing aircraft warfare
tion may be useful here. This may be done automat- engagement can demand the use of numerous sensors and
ically for the user in a modeling tool. major computing. Figure 3 shows one such system.
8. Set in the boundary conditions to limit the operation A defined and methodical way to generate the architec-
to that of interest. ture of a measuring system model is to apply the energy
9. Set up the appropriate stimulating functions, their matrix method – see Article 112, Systematic Descrip-
coefficients, and initial conditions. tion of Sensors, Volume 2 and Article 110, Signal/Energy
10. Run calibrating tests to exercise the model, using Matrix Modeling, Volume 2. This is based on the fact that
stimulations that have well-known and proven most measuring systems are formed as a chain of energy
outcomes. conversion stages that progressively convert the measurand
448 Measuring System Design
Virtual target
Virtual jammer
Target
flight
simulation
OAR facility Live missile
target
control
facility Jammer
flight Missile
simulation stimulation
facility facility
Telemetry
facility
Live Chase
shooter
Tanker
Mission
control
facility
Figure 3. Example of a major measurement and simulation system.
of interest into the usually needed electrical entity required and more accurate modeling. If nonlinearity exists, the
for the following communication and processing system. SHM models cannot be used, except in piecewise modeling
Each conversion in the serial chain can usually be treated operations. Modern modeling tools can handle nonlinear
as a black box. White box models of each will be used components, but that is not ‘keeping it simple’.
where appropriate, but for much of measuring system mod- For each box, examine the known performance and struc-
eling in support of design, black box representation suffices. tural and physical characteristics of the thing it represents.
For each box, a suitable dynamic (or static in the simpler From this study, it is usually clear which type of mathemat-
cases) describing equation can be assigned to obtain, by ical description will adequately represent the operation of
overall consolidation, the whole measuring system dynamic the box. An exact equation match is not usually necessary
performance model. and not feasible.
Assign to each box a suitable equation type. Table 1
shows the simplest SHM equations used for modeling
6 IDENTIFYING EQUATIONS measuring systems stages.
For example, consider the design situation wherein force
Where the box behavior is to be selected by design – as is is being converted by use of a load cell on which a strain
often the case in measuring system design – it is wise to gauge is mounted to monitor the change in strain of the
keep them linear in operation because that allows easier calibrated mechanical ring. This system clearly possesses
Table 1. Equation forms to select for a linear model stage.
Equation type Example equation Measuring example Comments
Zero-order linear a R-only circuit; amplifier, or stage, Fixed value in any situation
SHM system gain or attenuation
First-order linear bx + a R–C or R–L circuit: temperature Rise, or decline, without any
SHM system change on heating or cooling oscillation, in exponential manner
Second-order linear cx 2 + bx + a Weighing scale spring balance; Some level of oscillatory motion,
SHM system R–L–C circuit depending on degree of damping
Modeling Methodology 449
mass, springiness, and a low level of damping so that it can the overall dynamic performance. This always needs to be
to be modeled with the second-order SHM equation. At the verified as such, for latency and cable-filtering effects may
early stage of development, the higher second-order level is be significant.
selected, ruling it into a lower order when the coefficients It is best to reduce the forms in the above sequence. Seek-
of the real situation show that simplification is sufficient. ing reduction after all equations have been integrated may
In some cases, it is immediately clear that a constant (e.g. make it less clear as to which blocks are less significant.
an amplifier working only in its constant gain frequency As an example, in the design of the serial set of blocks for
range), or first-order (e.g. a temperature sensor immersed an electromechanical seismic recorder system (an inherently
in a fluid) description will suffice. low-pass system), the electronic amplifier response will
If the describing equation is already known, then use it usually be well below its cutoff frequency, meaning that
after reduction to the simplest form that suffices. An exam- it can be modeled as a constant (zero-order system). In
ple of this availability is the dynamic describing equation contrast, in the case of an acceleration sensor (which is a
often quoted on data sheets for operational amplifiers and high pass system), the amplifier response may be critical
instrument motors. and it must be modeled as a second-order system,
In many modeling tools, a library of linear and nonlinear
functions and statistical distributions is provided for selec-
tion in mathematical, tabular, or graphical form. The need
7.1 Initial conditions for models
to appreciate the mathematics has been lessened by these
tool features, but one still needs to appreciate what is being
used from a foundational viewpoint lest the wrong ones be All models need to have their initial conditions set correctly
selected. A safeguard is that good testing of the eventual when they are started up. This usually means setting con-
model will reveal faulty versions – but at that stage, rework ditions to zero, but it is not always so. System components
can be tedious and expensive. that need setting are storage elements, positions, voltage
levels, fluid levels, and the like.
It is not that well appreciated that some kinds of systems
7 ASSIGNMENT OF MODEL will, given the same initial condition, start up differently
giving different solution paths. For example, there exists
COEFFICIENTS a class of vibration situations where Eigen values are
involved that give differing trajectories.
When all blocks have equation forms assigned, it is time to
characterize the model. This process identifies the value
of several coefficients of the equations. For example, a
spring-mass-damper system of the weighing platform unit 8 EXERCISING THE MODEL
will possess gain, damping factor, and resonant frequency,
each of which will need to be identified in numerical It is possible for a person with good knowledge of the
terms. model and situation to set up tests that seemingly work
When these are assigned, the system should be evaluated to show that a model is sound – when it is not! A well-
by inspection to verify if this higher-order description is still developed test must exercise the system well – not just at
needed. The degree of damping, or the resonant frequency, obvious points.
could well be so high that the system can be modeled The person accepting a model must be familiar with
with a first-order model, or even zero-order model, thus the range of modeling needed. For example, an electronic
simplifying the model. warfare (EW) instrumentation detection system may well
When all blocks have appropriate equation types and be able to respond as needed for an aircraft flying through
coefficients assigned, the serial chain of equations is multi- a zone containing just one EW threat. This is, however,
plied together to arrive at the overall model of the specific not a realistic situation, for most real cases will have many
measuring system being modeled. more than one hazard present.
Again, inspection of relative coefficients may well show Testing should commence from a simple situation for
that some stages are insignificant compared to the oth- which there is a well-established output. For instance, a new
ers and that they can be approximated with a lower- signal spectrum analyzer model can be fed a sine wave. The
order model. output should then be a single spike, as shown in Figure 4.
With modern electronic components and systems, the A square wave should give a series of reducing spikes.
following communication and processing will often be of As confidence grows, the input is expanded to be
zero order (i.e. no phase shift) and, thus, not deteriorate more complex.
450 Measuring System Design
Sine wave model test input incorrectly, to be used synonymously; each of them is quite
different in purpose.
10.1 Verification
time →
In general, verification is the process applied first to deter-
mine that a model implementation accurately represents,
Spectrum analyzer output to a sufficient level, the developer’s conceptual description
and specification. In short, is it what was intended?
Verification is done largely by study of the model needs,
→
Spurious energy as can be found from documentation.
indicates model error
frequency →
10.2 Validation
Figure 4. Use of predictable tests to verify models. Validation is the process of determining the manner and
degree to which a model is an accurate representation of
the real world from the perspective of the intended uses
9 LIMITATIONS OF MODELS of the model, and also the level of confidence that can be
placed on the assessment.
Models are not perfect representations. The following list This is performed largely by testing and evaluating the
of commonly met model performance parameters can be model against the earlier verified requirements.
applied to ascertain limitations:
• Shared data consistency – how well does it facilitate 10.3 Accreditation
data sharing with other models?
Accreditation is the official certification given that a model
• Real-time interaction – does it work well enough in real
is acceptable for use for a specific purpose. Obtaining a
time or is it really working in ‘near real time’ that is
clear understanding of the ‘specific purpose’ is a problem
not ‘near enough’?
at times – models are used in other contexts all too easily.
• Scalability – can the model parameters be scaled up or
This is performed after reviewing the verification and
down and with what loss of precision?
validation results. If all is adequate, accreditation is given
• Extensibility – can its boundaries of operation be ex-
once a clear statement of its purpose and scope of use is
tended as the modeling exercise grows?
recorded to ensure it is not pressed into incorrect use. This
• Bandwidth – how fast can it model behavior? Lack of
should be carried out by an independent authority who has
adequate response can attenuate the results without any
not been too deeply involved in the model’s development.
obvious indication that it is now in error.
These three sequential steps seem to be clear-cut, but
• Reliability – obviously, one seeks a tool that does not
they can rarely be done on a totally crisp and formalized
need constant repair and resetting.
basis in large systems. Conversely, in smaller systems, they
• Latency – this refers to the lag time for data to arrive.
are all too often ignored to save time and money!
This is not the same as lack of dynamic response, for
VV&A can easily be done against the inappropriate
lag produces a different kind of error in a response.
level of documentation and with too narrow and selective
• Heterogeneity – does it work uniformly over its work-
a viewpoint.
ing range or do strange and unexpected things happen
There is no such thing as absolute truth – not even a
in certain situations?
real test will necessarily duplicate all of the important
real parameters – they are just not known until the actual
10 MODEL VALIDATION uses take place. Models must always be used with care,
and regular real-world testing must be used to ensure they
Verification, Validation, and Accreditation or VV&A are are adequate.
the cornerstone activities for ensuring that any model is a More detail on modeling in design and its role in proto-
faithful representation and can be used with the confidence typing is available in Sydenham (2004). Several texts are
that whatever is built from it, or whatever uses it, will have devoted to modeling, but they do not address measuring
a sound basis. Each term is covered in turn in the order in systems in particular – Severance (2002), Zeigler, Kim and
which they are applied in practice. The terms tend, quite Praehofer (2000), Fjeldly, Ytterdal and Shur (1997).
Modeling Methodology 451
RELATED ARTICLES Institute of Aeronautics and Astronautics, London AIAA-96-
3335-CP, (pp. 25–33).
Article 57, Static Considerations of General Instru- Fjeldly, T.A., Ytterdal, T. and Shur, M.S. (1997) Introduction to
mentation, Volume 1; Article 59, Introduction to the Device Modeling and Circuit Simulation, Wiley, New York.
Dynamic Regime of Measurement Systems, Volume 1; Holland, R. (1996) The Virtual Reality Homebrewer’s Handbook,
Article 61, First-order System Dynamics, Volume 1; Wiley, New York.
Article 62, Second-order System Dynamics, Volume 1; Neelamkavil, F. (1987) Computer Simulation and Modelling,
Article 70, Measuring System Design Methodologies, Wiley, New York.
Volume 2; Article 72, Mathematical Methods of Opti- Severance, F.L. (2002) System Modeling and Simulation: An
mization, Volume 2. Introduction, John Wiley, New York.
Sydenham, P.H. (2004) Systems Approach to Engineering Design,
Artech House, Boston, MA.
REFERENCES Zeigler, B., Kim, T. and Praehofer, H. (2000) Theory of Modeling
and Simulation, Academic Press, New York, NY.
Davies, D.R. (1996) Using Structured Modelling Strategy to Build
Validated Computer Model, in Proceeding of Second T&E
Institute of Aeronautics Forum, RAes, 25–27 June, American
72: Mathematical Methods of Optimization
Halit Eren
Curtin University of Technology, Perth, Western Australia, Australia
and complexity of the problem. The system may be lin-
1 Use of Mathematical Analysis in ear or nonlinear, stationary or time-varying, continuous or
Optimization 452 discrete, static or dynamic, constrained or unconstrained.
2 Setting Up the Generalized Mathematical Therefore, appropriate formulation of performance index
Description 453 involves a compromise between a meaningful evaluation of
3 Exercising the Mathematical Description 454 the system performance, availability of feasible mathemat-
ical descriptions, and application of meaningful algorithms
4 Comparison of Mathematical Description and
for their solutions.
Modeling 454
It is important to understand that an optimal system
Further Reading 455 under a selected performance index may not be optimal
for another performance index.
Figure 1 illustrates a typical optimization problem. In
1 USE OF MATHEMATICAL ANALYSIS this figure, the sign of the first derivative dJ /dU of the
cost function changes as the function passes through the
IN OPTIMIZATION maximum point. This method commonly uses an initial
estimated point (1) where the first derivative is considered
Optimization is applied in many disciplines, such as engi- (guessed) to be zero. This point (2) is then used, in the
neering, science, control systems, cybernetics, information direction of steepest ascent (3) or descent as the case
theory, computer systems and networks, economics and may be, for the next point until the absolute maxima or
other social sciences, and so on. minima is found without getting trapped in local extreme
In many engineering, science, and technology applica- points.
tions, optimization is used to maximize or minimize the An optimization problem involves two basic stages; first
value of a function chosen as the performance index or stage is the appropriate mathematical description of the
cost function, J . system and the second stage is the solution of the prob-
The performance is optimized for numerous reasons such lem using suitable techniques and algorithms. For solv-
as improving the quality, increasing production, maximiz- ing an optimization problem, the following steps may be
ing profits, decreasing waste, obtaining greater efficiency, necessary:
maximizing safety, saving time and energy, and so on.
However, in many problems, boundary conditions exist 1. A mathematical description has to be formulated.
because of limited power, storage capacity, capability of 2. The desired form of the objective function has to
machinery, various temperature constraints, limits in speed, be specified.
force, acceleration, and so on. 3. The variables have to be specified.
The cost function (see Figure 1) can be continuous and/or 4. The constraints have to be determined.
discrete. Appropriate selection of the performance index for 5. A suitable optimization method needs to be selected
a given system is important since it determines the nature and implemented.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Mathematical Methods of Optimization 453
Global maxima
J 2 3
1
Local maxima
ST5
Local minima
U1 U2 U3 U
ST4
dJ /dU
ST3
Figure 1. A typical cost function with local and global extremes.
ST2
6. Once the problem is solved, solutions must be related
to the physical system.
ST1
Physical systems are defined (modeled) mathematically Direction of
usually by identifying the related parameters and using the motion ST0
known properties. The application of optimization allows
the selection of those parameters that give the best proper- Figure 2. Illustration of hill-climbing with gradient method.
ties for a chosen preference function.
Once described mathematically, the optimization problem
may be solved using computers backed up by powerful soft- 2 SETTING UP THE GENERALIZED
ware tools, such as linear and nonlinear programming meth- MATHEMATICAL DESCRIPTION
ods, stochastic methods, tabu search algorithms, genetic
algorithms, artificial intelligence methods, and so forth. Optimization has two main objectives. The first is to iden-
For the solution of an optimization problem, depending tify a set of conditions for a function, f , where min or
on the nature of the problem, many different methods may max{f (x, θ )|x ∈ D(θ )} for constraints, D, will exist as a
be used. For example: solution. The second is characterization of the set of optimal
points and the application of algorithms.
1. If the performance index and constraints can be for-
The mathematical problem involves the determination of
mulated as linear algebraic functions of the variables,
a set of values of variables xi such that the cost function
then selection of an appropriate linear programming
ϕ|xi | is as small (or as large) as possible. xi represents the
method, such as simplex programming, is sufficient.
set of all possible variables for i = 1, 2, . . . , n.
2. If the equations describing the system are nonlinear,
However, each variable may be subject to equality or
then the solution may involve nonlinear techniques or
inequality (or both), linear or nonlinear constraints in the
linearization of the problem in some subregions.
form of
3. If the problem involves determination of the largest
value of a function with two or more variables, then the fj (|xi |) = 0, j = 1, 2, . . . , p (1)
steepest ascent (or descent) method, sometimes termed
as hill climbing, or gradient method may be a good or
choice, see Figure 2. In this figure, the steps taken to gk (|xi |) ≥ 0, k = 1, 2, . . . , q (2)
find the optimum point are shown by ST0 , ST1 , . . . , ST5 .
4. If the problems are complex (termed NP hard prob-
lems), traditional optimization methods based on local 2.1 Linear systems
information are not adequate to solve them; global
optimization methods need to be applied. Some of Many practical problems lead to linear systems of equa-
these methods are Lagrange duality in partly convex tions, which can involve hundreds of variables thus requir-
programming, robust minimizers, adaptive optimiza- ing the solutions of numerous simultaneous linear equations.
tion, decomposition techniques, iterative topographical The simplest case is the optimization of a single variable,
methods, and so on. ϕ(x). In simple cases in which there are not many variables,
454 Measuring System Design
linear programming can be employed yielding to solutions
in n-dimensional space.
If a linear programming problem has a unique optimal
solution, then it is an extreme point; if not unique, there is
an optimal solution at an extreme point, thus the solution
obtained is within a finite set of extreme points.
ST7
ST6 ST8
2.2 Nonlinear systems
Many problems lead to nonlinear (transcendental) sys- ST5
ST3
tems of equations, which can only be solved by compu-
tational methods. ST4
ST2
A problem is nonlinear if the objective function f (x)
ST1
in which x is a vector or any of the constraint functions
is nonlinear. Nonlinear functions are classified according
to whether their defining functions have the appropri- ST0 Direction of
ate convexity. motion
Solutions of nonlinear functions may often be dependent
Figure 3. Illustration of random walk method.
on the differentiability of the constituent functions. If all
constituent functions are differentiable, there is an analo-
gous result on the necessary conditions of optimality that Therefore, they find the nearest minimum point as quickly
can be closely related to the Kuhn–Tucker conditions. as possible and search for all potential solutions in an opti-
mum manner.
Recently developed, genetic algorithms are used for opti-
3 EXERCISING THE MATHEMATICAL mizing the parameters of an unknown system whose math-
ematical description is either too complicated or unknown.
DESCRIPTION
The genetic algorithms are usually applied to neural net-
works in two levels. The first level determines the suitable
Standard computational methods for solving nonlinear opti-
weights in the neurons and in the second level, the num-
mization problems may be divided into three main groups:
ber of hidden layers as well as the number of neurons
1. Zero-order methods (coordinate comparison, simplex is determined.
methods, and stochastic methods);
2. First-order methods (gradient and quasi-gradient, meth-
4 COMPARISON OF MATHEMATICAL
ods of associated directions, variable-metric methods);
3. Second-order method (Newton method). DESCRIPTION AND MODELING
Nonlinear optimization aims at mathematical formulation The nature of the mathematical description and computa-
of problems and implementation of algorithms. In the tional method indicate if the type of problem is to be linear
implementation stages, many different algorithms exist, or nonlinear. For example, a linear program is defined as
such as numerical optimization, random walk (Figure 3), the optimization of a linear cost function whose variables
convex analysis, calculus variations, stochastic methods, satisfy a system of linear equalities or inequalities.
game theory, statistical and probability techniques, and The term mathematical programming usually refers
more recently, genetic algorithms, and neural networks. to linear programs. Mathematical programming has a
Stochastic methods consist of the calculation of the val- long history and has been influenced by operations
ues of the cost function at a large number of selected points. research, game theory, mathematical economics, control
The selection is made such that each point has an equal theory, dynamic programming, calculus variations, and
probability of being selected. The best points are deter- combinational theory.
mined by comparing the function values at a large number The standard form of a linear program is expressed as an
of points. Although requiring intensive computational time, instruction to maximize or minimize a linear expression in
stochastic methods are useful for finding the global opti- n nonnegative variables.
mal points, that is, unlike the conventional methods, their Any system of linear inequalities can be reduced by sim-
likelihood of getting trapped in local minimums is small. ple substitutions. Associated with each linear programming
Mathematical Methods of Optimization 455
problem is a duality as a second linear programming prob- new neighborhood is created for a given current solution. A
lem. In this association, the original problem is referred to hill-climbing algorithm modified in this way systematically
as the primal. searches the entire area in which the global minimum of a
However, the general study of nonlinear programs usu- function is to be found.
ally involves such items as integer programs, stochastic The genetic algorithm is used to optimize systems with
programs, and network flow theory. unknown parameters. It is useful when other algorithms
One broad class of nonlinear programming problems is fail in such problems as where computational time is an
that concerning minimizing the cost function subject to no exponential or factorial function of a number of variables.
constraints; another is the linearly constrained nonlinear In the applications, it is sufficient to know the price of
programs that include subclass quadratic programs con- each individual in the population. Therefore, the pricing
cerning minimization of quadratic function subject to linear selection for each individual bears special importance as
constraints. it is directly related to the outcome of the optimized
There are many different ways of solving nonlinear problem. By application of artificial neural networks in two
programs depending on whether they display convexity stages, genetic algorithms are used to determine population,
properties and/or differentiability. pricing, and optimum points. There are improved versions
Some optimization problems are difficult to solve by of genetic algorithms such as scatter search, path relinking,
conventional numeric optimization methods, even if they star-paths, and so on.
converge to an acceptable optimum point in a reasonable
time. To solve these problems, other methods have been FURTHER READING
developed. For example, despite the large size of informa-
tion to be handled, the stochastic methods reach the global Agrawal, K.S. and Fabien, B. (1999) Optimization of Dynamic
optimum point easily. Systems, Kluwer Academic Publishers, Dordrecht.
In the hill-climbing algorithm, it is possible that an Bishop, R.H. (2002) The Mechatronics Handbook, CRC Press,
extreme is reached without finding the global minimum. Boca Raton, FL.
Here, the starting solution is generated at random. Once Floudas, C.A. and Pardas, P.M. (1996) State of the Art in
a local solution is obtained, it is used as the center of a Global Optimization-Computational Methods and Applications,
new neighborhood in which the optimization is repeated to Kluwer Academic Publishers, Dordrecht.
determine the other extremes. Kuo, B.C. (1992) Digital Control Systems, 2nd edn, Harcourt
Even this way, this method might revert to the local Brace Jovanovich, New York.
minimum that has already been passed in a previous step, Lewis, L.L and Syrmos, V.L. (1995) Optimal Control, 2nd edn,
called the problem of looping. The program may need to Wiley, New York.
be repeated several times with different randomly generated Rapcsak, T. (1997) Smooth Nonlinear Optimization in Rn , Kluwer
initial values. Academic Publishers, Dordrecht.
The tabu search algorithm is based on a hill-climbing Rubinov, A. and Glover, B. (2001) Optimization and Related
method that tries to avoid the problem of looping. The hill- Topics, Kluwer Academic Publishers, Dordrecht.
climbing algorithm has a short-time memory in which it Webster, J.G. (1999) Encyclopedia of Electrical and Electronic
remembers the optimal solution transformations to generate Engineering, Wiley, New York.
new centers for iterations. These optimal solution trans- Whittle, P. (1996) Optimal Control-Basics and Beyond, Wiley,
formations are prohibited (hence the name tabu) when the UK.
73: Overview of Electrical and Electronic Technique
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
Here is presented an introduction to the capability and
1 Relevance of Electrical and Electronic methodology of electrical and electronic techniques. It
Technique 459 must be made clear that although electrical and electronic
2 Historical Development of Electrical Regime 460 technique is so extensively used and is very attractive it is
3 Systematic Nature of Electrical and not always the optimum means of solving a measurement
Electronic Regime Engineering 461 systems design need. Indeed, there are many applications
4 Literature of Electrical and Electronic where the use of electrical methods are either not applicable
Regime Engineering 462 or they are not preferred, even if usable, for a variety
5 Signals in the Electrical Regime 465 of reasons.
Despite the fact that electrical methods appear to be able
Related Articles 466
to supply every measurement system data handling needs,
References 466 it must be remembered that the natural world is usually
the system that must be interfaced to man’s inventions;
that world does not possess many signal sources having the
1 RELEVANCE OF ELECTRICAL AND same form of physical manifestation as occurs in man-made
ELECTRONIC TECHNIQUE electrical systems. Thus it is that the other energy regimes
will always be important in measurement system design
Although electricity was the most recent technological dis- to assist formation of the input and output transudation
cipline to develop, it was very rapidly found to be the interfaces. By itself, an electronic system has no practical
most appropriate for many tasks required in a measuring use – it must be interfaced to an application with some kind
system. Certain classes of instrument can be constructed of transformation device – the sensor.
without recourse to electrical principles, examples being Electrical technique includes both the traditional electri-
the microscope, the micrometer, the direct recording water cal and the more recent electronic disciplines. Initially, the
level gauge, and many more. However, if an application two aspects could be reasonably well divided for purposes
needs extensive information processing, signal transmission of exposition and administration. Prior to 1910, electron-
over long distances, uniformity of manufacture at low cost, ics was virtually unknown and electricity was, therefore,
ability to be extended, and use systematic design proce- clearly defined.
dures, then electrical technique will usually (but not always) Electrical methods were partially displaced by electronic
provide a better design solution than other alternatives. methods (that grew out of the application of thermionic
In the context of the thousands of years of measuring devices) since electronics provided the circuit theory
systems development, this situation has only arisen in the considered necessary, and initially much of the practice.
last half century, wherein solid-state integratable electronic As time progressed, electronic methods were devised that
circuitry has provided unprecedented ability to produce low could also handle most power needs, previously only
cost, very powerful processing and data handling ability. possible through the use of electrical devices. Today the
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
460 Measuring System Design
distinction between electrical and electronic methodology significant progress came only after the invention of the
is very blurred, although many institutional mechanisms simple primary cell by Volta in 1800. Prior to that time
and groupings continue that might imply otherwise. experimenters had only very high-voltage sources possess-
Until very recent times, electronic technique was gener- ing high output impedance. It was not exactly the preferred
ally only suitable for stages following the sensor. In the apparatus to use in an investigation of the practical appli-
1970s decade, however, silicon integratable devices were cations of electrical knowledge!
first developed to yield a new generation of electronically In 1800, the electric primary cell became generally
based sensors that are now in prolific use. As time pro- available and, being simple to make, was rapidly adopted
gresses, electronic capability is steadily being extended; the by experimenters. Its low output impedance, reasonably
sensor principles are still as current as when first applied. long duration of operation, low voltage, and ease of
It is certainly true that in any new application requiring manufacture was admirably suited to the needs of the
a measuring instrument the designer must now consider the gentleman scientists of those times. Progress was steadily
appropriateness of electrical methods for the specific task made in gaining understanding of the fundamental nature
in question. of practically useful electrical circuits and devices and in
their everyday use.
2 HISTORICAL DEVELOPMENT During the nineteenth century, the laws of circuits, both
DC and AC, were formulated through careful experimental
OF ELECTRICAL REGIME and theoretical work. As new laws were published, they
were taken up to form technological devices. For exam-
The origins of electricity and electronics can be traced
ple, the relationship between electricity and magnetism
to ancient Greece. The distinction between the two is
was established first as the motor effect in the 1820s.
usually based on the premise that both are concerned with
electron flow, the former being at a macrolevel, the latter This enabled electrical indicators and electromagnetic actu-
at the discrete electron level; this is somewhat artificial ators to be made. Better indicators of electrical quantities
in modern terms of application. Their contribution was, (meters) enabled more laws to be discovered and weaker
however, minimal compared with advances that began in effects to be applied.
the seventeenth century A.D. Furthermore, their knowledge The laws of electromagnetic radiation were also enun-
was rarely put to use and there was certainly no ordered ciated in the middle of the nineteenth century. By 1888,
use of principles of design. Hertz had experimentally proven that the theory was cor-
Interest in, and application of, electricity started in rect and that practical apparatus could be constructed that
earnest with work on electrostatic devices of the form of enabled it to be put to use. Marconi, in the last decade of
charge generators and detectors, this being in the century that century, assembled the first practical radio system using
preceding 1800 A.D. Knowledge of charge, its quantity and simple, purely electrical, components and understanding.
storage, its transfer from place to place by the use of By 1900, many of the basic circuitry laws and physical
insulated wires, and its physical effects gave eighteenth principles, which are today used to devise instruments,
century experimenters a tool for further experimentation. were known, if only as scientific entities. The Victorian
Out of their work, gradually, there grew useful applications. era produced much of the electrical systems groundwork
Electrical telegraphy first began in the late eighteenth needed in twentieth century instrumentation.
century in a static electricity form. Static electricity led to A chance encounter occurred in the latter part of the nine-
knowledge about the quantity and potential of electricity teenth century, this being discovery of the Edison effect. It
and to the understanding that there exist two forms of proved, experimentally, that a thermionic device could rec-
electric charge: positive and negative. What did not develop tify alternating current. Edison did not, however, make use
so well at first was an appreciation of charge flowing as a of this at the time. By 1900, designers of telegraph, tele-
current in wires. phone, and radio communications needed three important
Experiments with static electricity, coupled with the improvements in technological capability. Telecommuni-
improvement in vacuum technique, gradually provided cation needs provided strong commercial reasons to seek
knowledge of discharge effects in gases. These provided them. These needs were: how to amplify a weak elec-
a place for the understanding of the behavior of electrons, trical signal, how to rectify radio frequency, alternating
a maturity that took some 200 years to reach the point where currents, and how to generate radio frequency currents.
the electron was finally proven to exist as a discrete entity Each of these needs was being catered to at the time by
in J. J. Thomson’s experiments of 1899. very unsatisfactory methods.
Although there was much activity in seeking knowledge The first decade of the twentieth century saw these
about the nature of electricity in the eighteenth century, three basic needs being met by the invention of one
Overview of Electrical and Electronic Technique 461
device, the thermionic valve. It first emerged as the development. Integration made devices smaller, cheaper,
purposefully built diode in which the Edison effect was and far less power consuming. They enabled the basic
utilized. The diode was capable of rectification and signal level of systems to be gradually, at a quickening pace,
generation. Soon after, the triode valve was devised; it extended in sophistication. As an example, in the 1960s a
could perform all functions of amplification, rectification, digital counter would normally have been assembled from
and generation. By the 1920s, the thermionic device had discrete components using individually picked transistors.
evolved into many forms satisfying many needs, not only Today, the whole multidecade counter, with its display,
in telecommunications but also in the emerging, derivative, would normally be purchased as a single basic commercial
instrumentation areas. chip unit that is vastly smaller, far more reliable, and much
Electrical and electronic techniques began their partner- less costly.
ship from the onset of the electronic discipline in the 1910s. The discipline of electrical and electronic engineering
Tuned circuits, special purpose devices built with valves as has, for use in the design of instrumentation systems,
their basis, developed; the bistable flip–flop and the DC passed from a stage where originally only relatively small-
amplifier being examples that initially fulfilled some spe- extent systems could be handled by the average person
cial need. to one in which very complex and powerful systems are
The First World War proved that electronic devices were assembled with few technological constraints being placed
practical. Training in radio, given to military servicemen, upon the designer. It is now very much a procedure of being
helped the general public to subsequently take an interest able to more fully express a designer’s innovative powers
in electronics for wireless applications. and ability with extensive systems built from basic, very
Gradually electronic technique spread into industrial powerful, marketed units, doing this to the limit of his or
usage. By the 1930s, industrial electronics was clearly her imaginative ability.
established. It first began to have an impact through the Space only allows a very short history to be given here.
use of the electric eye, a photocell coupled to a suitable A detailed account is to be found in Sydenham (1979a),
electronic amplifier. where reference to other historical studies is given in the
Thermionic components, along with purely electrical general history of instrumentation presented there.
devices such as the magnetic and amplidyne amplifiers,
were able to cope with most of the heavy current industrial
needs that arose; but, in general, electronic method was, by 3 SYSTEMATIC NATURE OF
modern standards, slow to be accepted. Basically valve- ELECTRICAL AND ELECTRONIC
based electronic circuitry was then too expensive, too REGIME ENGINEERING
unreliable, and too sophisticated for the times to find the
widespread use that we have come to accept as the norm Electrical engineering, as has been described above, devel-
of the electronics of today. oped with a well-ordered basis of theoretical understanding.
It is the field of computing that helped generate incentive Indeed, much of what resulted would not have emerged so
to find better technological ways of carrying out the rapidly without the theory being known. Unlike mechani-
functions developed for valve-operated apparatus. Digital cal engineering, where a large part of its output was able
computing, especially, provided impetus for activity, for, by to emerge through the technological path of experimental
the 1940s, it was clear that very powerful computers could experience and intuition, most electrical topics can only be
be built that far exceeded those economically possible with understood and applied if the user has a good theoretical
valves. A superior replacement for the amplifying valve was grounding in the principles involved. The physical nature of
needed. Valves used too much power, they ran at too high the subject material has allowed its constituent knowledge
a temperature, they were too large, they cost too much, and to be ordered systematically.
they were not reliable enough to be used in huge numbers. Electronic technique rests upon well-understood, theo-
In the 1940 decade, the transistor was devised at the Bell retically enunciated principles. When used at the level that
Laboratories. It did not emerge quite as suddenly as it might most users assemble systems (from commercially marketed
seem but came about from gradual evolution of earlier work modules), it also requires a considerable degree of inven-
on the solid-state diode of the 1900 era and after. Valves, tive skill. It can, however, be efficiently applied using
in fact, being able to satisfy many of the designers’ early reasonably straightforward and simplified theoretical aids,
needs and expectations, tended to slow down development such as designing procedures and rule of thumb relation-
of solid-state devices. ships. The nature of the range that electronic parameter
From the time of the invention of the practical transistor, values cover implies that the components, and their assem-
there has been great increase in the pace of electronic bly, do not usually require very tight tolerances of design.
462 Measuring System Design
People with very little training in the basics of electronics may well overwhelm the beholder but the use of the
can build worthwhile systems. Naturally, specially trained systematic realization procedure to form the constituent,
persons can handle the topic with greater ease and better higher-level blocks will reveal how the system operates as
performance. each circuit group is identified as a known and familiar
This fact should not be interpreted as meaning that building block.
electronics does not somewhere require in-depth scientific
understanding and application of sophisticated methodol-
ogy. That has become today more the role of the component 4 LITERATURE OF ELECTRICAL AND
and subsystem designer, those people have provided a ELECTRONIC REGIME ENGINEERING
buffer for the end user and general systems designer.
Marketing forces have assisted simplification of elec- It is of value to provide an overview of the extensive
tronic and electrical instrument design. The quest for sales, printed literature before continuing. There exist numerous
in large numbers, has enticed manufacturers to provide a texts on electronic and electrical engineering. They range,
truly useful and extensive application service. They have in marketing approach, from trivial expositions to esoteric
sought ways by which sophisticated devices can be used academic theoretical works. Each has its own usefulness.
effectively by semitrained persons. Although popular texts are often denigrated by experts,
This account is aimed at the semitrained person. It is not they have, over the years, risen in quality, now providing
intended for the specialist in some aspect of electronic or excellent introductions for persons who wish to use and
electrical engineering. It begins by presenting information understand electrical and electronic technique at the level
about the most basic components used to form electronic of practical use yet not become formally trained in this way.
systems. Whereas many of these may not be used as discrete Academic works may often be found to be too advanced
devices very much in these times, it is, nevertheless, for middle-levels users. The reader is cautioned to study a
important to have an understanding of the concepts and text carefully to establish if it adequately matches the need.
hardware entities that go into a marketed system module. The following selection of published works is presented
These basics presented then allows discussion to rise to a to assist selection. It must, however, be stressed that many
higher level in the systems hierarchy wherein description others are available. Electrical engineering principles are
can be given of the marketed building blocks that are presented in Baitch (1972), Bureau of Navy (1970), and
commonly used to assemble the specific system needed in Smith (1976) as introductory courses. Similar material
a given application. It is then feasible to rise yet further is also available in the many handbooks of electrical
to the final design level in order to discuss how extensive engineering, but where it is not usually presented in a
systems are generated. didactic manner.
Other conceptual aspects of system design that need Electrical engineering overlaps electronic technique in
consideration are sources of noise in these systems and the the area of circuit networks. Linear networks are the easiest
power supplies that supply energy for the building blocks. to design and theorize (see Leach, 1976; Van Valkenburg,
As an example of the systematic nature of electronic 1974). Nonlinear systems are covered in Fox (1957), Van
systems consider the diode-based electronic thermometer der Ziel (1977), and Willson (1975). Older texts are still
shown in Figure 1. It illustrates how a system is progres- useful for they lay down enduring principles.
sively broken down into subsystem blocks, each of which Systems of circuitry originally were developed by ana-
in turn can be further broken down, the process continu- lyzing given arrangements, but a more positive design
ing until adequately basic levels of description are reached. approach was developed through which the system can be
In many systems, there is no need to progress to the indi- synthesized to provide a prestated performance: Fox (1952)
vidual component level, as the necessary modules can often is but one text following this approach.
be assembled from inexpensive high-performance units that One particular class of network is the so-called filter.
only need wiring into the system in the appropriate manner. Modern design procedures for these use previously devel-
The above procedure allows any system to be portrayed oped mathematical expressions to arrive at the desired
on paper at the level needed to adequately specify its performance using feedback placed around an active ampli-
functional architecture. Often the reverse occurs and the fier element. Several texts exist on these forms of filter
user may be initially confronted with the final subsystem (refer to Burr-Brown, 1966; Hilburn and Johnson, 1973;
details, not the primary stage block diagram. At first this Huelsman, 1977; Spence, 1970).
Figure 1. Breakdown of a solid-state, diode-based, temperature meter system showing how an electronic system is formed from basic
components arranged in a systematic order.
Overview of Electrical and Electronic Technique 463
DC supply Output voltage
(battery) proportional to
temperature
and zero set
at selected
Adjustable Vb temperature
constant source −
+
Adjustable
voltage gain
Constant
current source
Vd
Sensing diode
Output voltage
Design requirements:
(1) Supply diode with constant current
(2) Back off standing voltage from diode to obtain zero output
for selected temperature
(3) Provide adjustable gain to give desired millivolts /°C output
Functional descriptive block diagram
Printed circuit board
SW1a
Battery
+ R2
100
e RV 1 RV 2
b Q2 + 9 V− + 9 V−
BC558 R5 C1
c C1 Q2
Q1 e e c 1 B
b
ZD1
IC 1
33 pF
R6
R1
R4
R7
R2
BC558 R5 Q1 b
c e c
RV1 b
C3 C2
R1 10 k 1 R3
2×9V A 10 k
3 8
76
Sensor
Battery IC1
D1 ZD1 R3 LM308
2 +
5.1 V 10 k 4
R6
B 100 k R7 Output
+ RV2
100 k C2
R4 0.047 mF −
100 k C3
0.047 mF Sensor Output Power switch SW 1
SW1b
Detailed wring diagram Constructional layout
Compensation Compensation
Diode
1 8
7 V+
Add piece Epoxy
of insulation R4 R5 R6 R7 Q12 R9
between leads 20 k 20 k 10 k 10 k 2 kΩ
Q11 R8 Q13 Q18 R10
Pen Q7 Q9 5.6 kΩ 90 Ω
Q8 Q10 Q17 6
Join diode leads case Q14 Q15 Output
onto wires
R11
Q5 Q6 Q22 Q21 Q16 240 Ω
Q19
− 2 Q1 Q23
Probe, construction of sensor Q27
Q25
Inputs R3
Q3 Q4 R14
50 kΩ 940
Q20
Q26 R13
+ 3 Q2 Q28 20 kΩ
R1 R2
2 kΩ 2 kΩ R12
Q24 320
Q29
R15 R16
60 kΩ 20 kΩ
4 V−
R19 R17 R15
6.4 kΩ 500 Ω 1 kΩ
Internal schematic of IC1 op-amp. (not needed to build or service instrument)
464 Measuring System Design
Computers have made simple circuitry design almost (1978), Soisson (1961), Studer (1963), Sydenham (1980),
trivial and more complex systems possible. Director (1974) Terman and Pettit (1952), Thomas (1967), Turner (1959,
and Szentirmai (1974) describe computer-aided design of 1963), Waller (1972a, 1972b), Wedlock (1969), Frenzel
circuits. This aspect has matured greatly and tool suites (1997), Gibilisco (1997), Grob (1996), Horowitz and Hill
are available to support efficient design – once the user (1989), Mano (2001).
is familiar with the tool and can afford to purchase the Texts continuously appear that describe new devices as
tool system. they arrive in the market. To the titles given here must be
As in electrical engineering, but probably more so, there added, in any search for information, the enormous amount
exist many works covering general electronic principles of application notes and internal publications provided by
and method. A selection spanning the ranges of depth and the component suppliers and marketing agencies. Electronic
of age (early texts contain useful conceptual information engineering is one of the areas of technology where
and theoretical derivations) is Brophy (1977), Carson tradition has given rise to sales backed by extensive
(1961), Churchman (1971), Delaney (1969), Langford- published assistance about product use. Some of these
Smith (1955), Lowe (1972, 1977), Millman and Halkias publications are included in the following list on devices
(1976), Mitchell (1951), Pender and McIlwain (1950), and components: Cleary (1969) and Thornton et al. (1966)
Ruiter and Murphy (1962), Squires (1967), Starr (1959), on transistors; Lancaster (1976) on the transistor–transistor-
Sydenham (1979b), and Waters (1978). Several of the logic assemblies, called TTL; and Lancaster (1977) on the
electronic equipment and component suppliers offer text advance on TTL, but not its replacement, the alternative
packaged courses, which often include aural recorded tapes, CMOS circuitry. Integrated circuits are covered in many
experimental apparatus. An example is Heathkit (1978). books (see Millman, 1972; Rosenthal, 1975; Turner, 1977).
Electronic circuits (these involve electrical circuits plus Other components are dealt with in Sheingold (1972, 1974).
specialized components) are the subject of Jones (1978), Turner (1978) is on field-effect transistors (FETs).
Millman and Halkias (1967), Sands and Mackewroth Thermionic valves are still used in a general way in
(1975), and Senturia and Wedlock (1975). Lowe (1974) some countries and will be found in older equipment; their
deals with drawing practice for electrical and electronic operation is explained in the numerous electronic subject
engineering: caution is, however, needed in using works on works compiled before 1960.
this aspect due to the existence of several different standards Often poorly considered when designing electronic
of practice. aspects of a measurement systems is the need to supply
Electronics, as did electrical engineering earlier, spawned adequately conditioned power. Texts on power supplies
numerous new words in its technical vocabulary. Dictionar- and on handling power include Csáki et al. (1975), Dewan
ies are often needed to understand terms used in written and and Straughen (1975), H-P (1973), Kepco (1966), Marston
spoken descriptions. Such definitions are given in Handel (1974), and Motorola (1961, 1966, 1967).
(1971), IEEE (1972), and Radio Shack (1975). There is a As will be explained in more detail below, one regime
surprisingly close, but not identical, agreement on terminol- of electrical and electronic systems is those systems using
ogy across the world in this discipline but very few people continuously varying signals that can take any level of
appear to make general use of the standard terminologies voltage or current within defined limits. These are called
that do exist, so minor localized differences often occur. analog signals and the components for use with such
It is difficult to separate electronic instrumentation from signals have become largely known as linear devices.
the total field of electrical and electronic engineering. For The operational amplifier is a salient linear circuit unit.
this reason, many of the above mentioned general texts A selection of works on the linear regime includes
may well provide the information sought. There are, addi- Burr-Brown (1964), Clayton (1971), Connelly (1975),
tional to these, many books available that indicate from Faulkenberry (1977), Graeme and Tobey (1971), Gupta and
their title that they relate to electronic instrumentation Singh (1975), Melen and Garland (1978), NS Corporation
and testing, the main theme of this article. As with all (1976), and Signetics (1977).
works, they must be consulted to establish their usefulness The other kind of signal regime uses a signal form that
in a given situation. The following list refers to those can only exist at two, or sometimes three levels, the former
specifically on instrumentation and testing: Banner (1958), being the more usual. This is known as a digital signal.
Booth (1961), Diefenderfer (1979), Edwards (1971), Gre- As will be explained below, several basic units exist for
gory (1973), Herrick (1972) IEE (1979), Kidwell (1969), building digital systems. They are variously described as
Kinnard (1956), Lion (1975), Maeison (1974), Malvino units or as combinations in Blakeslee (1975), Bouwens
(1967), Mansfield (1973), Norton (1969), Oliver (1971), (1974), Breeze (1972), Helms, Kaiser and Rabiner (1976),
Partridge (1958), Piller (1970), Prensky (1963), Regtien Kostopoulos (1975), Middleton (1977), Millman and Taub
Overview of Electrical and Electronic Technique 465
(1965), Namgoster (1977), Peled and Liu (1976), Rabiner DC signal level
and Roder (1973), and Wakerly (1976). General works on AC signal
electronic method, mentioned earlier, also discuss the two Signal +V
regimes of signal use. amplitude
Manufacturing aspects of electronic systems using 0 0
computer aids are covered in Cassell (1972). The works Time
Time
Dummer and Griffin (1962), Simpson (1976), and Waller
(1972c) deal with testing and reliability of components Combined these give
and materials. Servicing electronic systems has become
an important aspect due to the large size extent of
equipment used and the responsible position held by it in Signal AC signal having
amplitude + DC bias
the general integrity of a plant. Information about this topic 0
is given in Garland and Stainer (1970), Sloot (1972), and
Waner (1979).
Time
Finally, in this introduction to the literature of electrical
and electronic systems used for measurement, mention is Figure 2. Amplitude-time graph of an electric AC signal having
needed of the many books and booklets available from the DC bias.
commercial press and the marketing companies on the use
of electronic instrumentation in specific fields. These often may, in fact, always move only in one direction. This is,
repeat some of the basic information in their introductions however, still regarded as an AC system (possessing a DC
before moving onto the details of their specific topics. Two level) because the AC signal can be recovered, as can the
examples are Van Santen (1967) on weighing and Piller DC level. A DC level cannot carry information, unless it
(1970) on electromedical instrumentation. is modulated with an AC signal, but it is essential in many
forms of electronic arrangement.
Signals varying in time (the same concept applies to those
5 SIGNALS IN THE ELECTRICAL varying in space) can be broken down into their Fourier
REGIME frequency components. Sine wave signals are, therefore,
the most basic signal form. Many of the concepts and
Measurement is the process of obtaining meaningful infor- explanations arising in electrical and electronic engineering
mation about a topic. In the sense of relationship with rest upon the assumption that the AC signal in question is
the real physical world, this is achieved by interpreting sinusoidal. If it is not, then considerable error can occur in
received modulations of energy or of mass- transfer. The the inappropriate use of presented information.
entity that conveys the information is known as a signal. Where no qualification is given as to the waveshape or
Electrical and electronic systems make use of electrical to the averaging criteria used, it is usually safe to assume
energy for information transfer, so their signals occur in that the signal is indeed sinusoidal and that the root mean
some form of variation in the amplitude or phase rela- square (rms) value is that which is given.
tion of either voltage, current, or impedance. These three As amplitudes of the signals and the frequencies that are
entities are related according to Ohm’s law in linearly used in instrumentation range over numerous decades, the
operating circuits and through other relationships in nonlin- practice of compressing the scale from a linear expression
ear systems. For example, the system may use a constant is used extensively. The most usual form is the decibel
voltage-source impedance, variations in the, say, sensor method, which provides compression of magnitude on a
causing current modulations to occur at the receiver. Alter- logarithm basis.
natively, the current may be held constant, the impedance The range of frequencies involved in signals and systems
variations giving rise to voltage fluctuations. By suitable explanation and operation begins at the cycle period of
conversion, through the use of an impedance varying stage, thousands of years (as arises say, in considerations of
a constant-voltage supply system can be used to supply the movements of the orbiting planets), passing through
varying voltage signals. the most commonly encountered region of around 10 Hz
If the current flows only in one direction, it is called to 20 kHz (due to widespread interest in audio systems),
a direct current (DC) system. If the current alternates rising to much higher frequencies for communications and
backward and forward in the connecting link, it is known then into the lesser known and used areas above X-ray
as an alternating current (AC) system. Where the system frequencies. Cosmic rays occur in the 1025 Hz region. The
has DC bias added to the AC signal (Figure 2) the current most dominant energy frequency spectra used are those
466 Measuring System Design
of the electromagnetic spectrum and the audio spectrum. Csáki, F., Ganszky, K., Ipsits, I. and Marti, S. (1975) Power
Spatial frequency systems, as arise in optical systems, Electronics, Akadémiai Kiadó, Budapest.
closely follow many of the fundamentals expressed within Delaney, C.F.G. (1969) Electronics for the Physicist, Penguin
the material contained in electrical and electronic method. Books, Harmondsworth.
This account will now begin at the most fundamental Dewan, S.B. and Straughen, A. (1975) Power Semiconductor
system level at which most users of electronic and electrical Circuits, Wiley, New York.
systems will need to go, that is, at the components level. Diefenderfer, A.J. (1979) Principles of Electronic Instrumenta-
Systems will then be considered at progressively higher tion, Saunders, Philadelphia, PA.
levels of extent and sophistication. Director, S.W. (1974) Computer-aided Circuit Design: Simulation
and Optimization, Dowden, Hutchinson & Ross, Stroudsburg,
PA.
Dummer, G.W.A. and Griffin, N.B. (1962) Environmental Testing
RELATED ARTICLES Techniques for Electronics and Materials: Series on Electronics
and Testing, Vol. 15, Pergamon, Oxford.
Article 74, Basic Electronic Components, Volume 2;
Edwards, D.F.A. (1971) Electronic Measurement Techniques, But-
Article 76, Electronic Systems Design, Volume 2; Arti- terworths, Sevenoaks.
cle 78, Embedded Systems, Volume 2; Article 79, Test- Faulkenberry, L.M. (1977) An Introduction to Operational Ampli-
ing Electronic Systems, Volume 2. fiers, Wiley, New York.
Frenzel, L.E. (1997) Crash Course in Electronics Technology,
Butterworth-Heinemann.
REFERENCES Fox, J. (1952) Modern Network Synthesis, Wiley, New York.
Baitch, T. (1972) Electrical Technology, Wiley, Sydney. Fox, J. (1957) Nonlinear Circuit Analysis, Wiley, New York.
Banner, H.W. (1958) Electronic Measuring Instruments, Chap- Garland, D.J. and Stainer, F.W. (1970) Modern Electronic Main-
man and Hall, London. tenance Principles, Pergamon, London.
Graeme, J.G. and Tobey, G.E. (1971) Operational Amplifiers –
Blakeslee, T.R. (1975) Digital Design with Standard MSI and LSI,
Design and Application, McGraw-Hill, New York.
Wiley, New York.
Gregory, B.A. (1973) Electrical Instrumentation; An Introduction,
Booth, S.F. (1961) Precision Measurement and Calibration;
Macmillan, London.
Selected Papers on Electricity and Electronics (3 Vols), National
Bureau of Standards, Washington, DC. Grob, B. (1996) Grob Basic Electronics, Glencoe McGraw Hill.
Bouwens, A.J. (1974) Digital Instruments Course (several parts Gibilisco, S. (1997) Teach Yourself Electricity and Electronics,
issued), N. V. Philips Gloeilampenfabrieken, Eindhoven. McGraw-Hill.
Breeze, E.G. (1972) Digital Display Systems: Fairchild Applica- Gupta, K.C. and Singh, A. (1975) Microwave Integrated Circuits,
tion Note 212/1, Fairchild Camera and Instrument Corporation, Wiley, New York.
Mountain View, CA. Handel, S. (1971) A Dictionary of Electronics, Penguin Books,
Brophy, J.J. (1977) Basic Electronics for Scientists, McGraw-Hill- Harmondsworth.
Kogakusha, Tokyo. Heathkit (1978) Electronic-Courses for Learn-at-Home: Course
Bureau of Navy (1970) Basic Electricity, Dover, New York. 1 DC Electronics; Course 2 AC Electronics; Course 3 Semi-
conductor Devices; Course 4 Electronic Circuits; Advanced
Burr-Brown (1964) Handbook of Operational Amplifier Applica- Digital Technique Course; Microprocessor Course. Available
tions, Burr-Brown Research Corporation, Tucson, AZ. with experimental back-up and optional cassettes, Heath-
Burr-Brown (1966) Handbook of Operational Amplifier Active Schlumberger, London.
R.C. Networks, Burr-Brown Research Corporation, Tucson, AZ. Helms, H.D., Kaiser, J.F. and Rabiner, L.R. (1976) Literature in
Carson, R.S. (1961) Principles of Applied Electronics, McGraw- Digital Signal Processing, Wiley, New York.
Hill, New York. Herrick, C.N. (1972) Instruments and Measurements for Electron-
Cassell, D.A. (1972) Introduction to Computer-aided Manufactur- ics, McGraw-Hill, New York.
ing in Electronics, Wiley, New York. Hilburn, J.L. and Johnson, D.E. (1973) Manual of Active Filter
Churchman, L.W. (1971) Survey of Electronics, Rinehart Press, Design, McGraw-Hill, New York.
San Francisco, CA. Horowitz, P. and Hill, W. (1989) The Art of Electronics, Cam-
Clayton, G.B. (1971) Operational Amplifiers, Butterworths, bridge University Press.
Sevenoaks. H-P (1973) DC Power Supply Handbook, Hewlett-Packard, Palo
Cleary, J.F. (ed.) (1969) Transistor Manual, General Electric Alto, CA.
Company, Chicago, IL. Huelsman, L.P. (1977) Active R.C. Filters, Wiley, New York.
Connelly, J.A. (1975) Analog Integrated Circuits, Wiley, New IEE (1979) Electronic Test and Measuring Instrumentation: Test-
York. mex 79, IEE, London.
Overview of Electrical and Electronic Technique 467
IEEE (1972) IEEE Standard Dictionary of Electrical and Elec- Motorola (1967) Zener Diode Handbook, Motorola Inc., Phoenix,
tronic Terms, Wiley, New York. AZ.
Jones, M.H. (1978) A Practical Introduction to Electronic Cir- Namgoster, M. (1977) Digital Equipment Trouble Shooting,
cuits, Cambridge University Press, Cambridge. Reston, Reston, VA.
Kepco (1966) Power Supply Handbook, Kepco Inc., Flushing, Norton, H.N. (1969) Transducers for Electronic Measuring Sys-
New York. tems, Prentice-Hall, Englewood Cliffs, NJ.
Kidwell, W.M. (1969) Electrical Instruments and Measurements, NS Corporation (1976) Linear Applications Handbook, (Vol. I,
McGraw-Hill, New York. 1973; Vol. 2, 1976), National Semiconductor Corporation, Santa
Kinnard, I.F. (1956) Applied Electrical Measurements, Wiley, Clara, CA.
New York. Oliver, B.M. (1971) Electronic Measurements and Instrumenta-
Kostopoulos, G.K. (1975) Digital Engineering, Wiley, New York. tion, McGraw-Hill, New York.
Lancaster, D. (1976) TTL Cookbook, Howard W. Sams, Indi- Partridge, G.R. (1958) Principles of Electronic Measurements,
anapolis, IN. Prentice-Hall, Englewood Cliffs, NJ.
Lancaster, D. (1977) CMOS Cookbook, Howard W. Sams, Indi- Peled, A. and Liu, B. (1976) Digital Signal Processing, Wiley,
anapolis, IN. New York.
Langford-Smith, F. (1955) Radiotron Designers Handbook, Pender, H. and McIlwain, K. (1950) Electrical Engineers Hand-
AWV, Sydney (numerous editions). book: Electric Communication and Electronics, Wiley, New
York.
Leach, D.P. (1976) Basic Electric Circuits, Wiley, New York.
Piller, L.W. (1970) Electronic Instrumentation Theory of Cardiac
Lion, K.S. (1975) Elements of Electrical and Electronic Instru-
Technology, Staples Press, London.
mentation, McGraw-Hill, New York.
Prensky, S.D. (1963) Electronic Instrumentation, Prentice-Hall,
Lowe, J.F. (1972) Experiments in Electronics, McGraw-Hill,
Englewood Cliffs, NJ.
Sydney.
Lowe, J.F. (1974) Electrical and Electronic Drawing, McGraw- Rabiner, L.B. and Roder, C.M. (1973) Digital Signal Processing,
Hill, Sydney. IEEE, New York.
Lowe, J.F. (1977) Electronics for Electrical Trades, McGraw-Hill, Radio Shack (1975) Dictionary of Electronics, Radioshack (Tandy
Sydney. Electronics), USA.
Maeison, E.C. (1974) Electrical Instruments in Hazardous Loca- Regtien, P.P.L. (1978) Modern Electronic Measuring Systems,
tions, Instrument Society of America, Pittsburgh, PA. Delft University Press, Netherlands.
Malvino, A.P. (1967) Electronic Instrumentation Fundamentals, Rosenthal, M.P. (1975) Understanding Integrated Circuits, Hay-
McGraw-Hill, New York. den, Rochelle Park, NJ.
Mano, M.M. (2001) Digital Design, Prentice Hall. Ruiter, J.H. and Murphy, R.G. (1962) Basic Industrial Electronic
Controls, Holt, Rinehart and Winston, New York.
Mansfield, P.H. (1973) Electrical Transducers for Industrial Mea-
surement, Butterworths, Sevenoaks. Sands, L.G. and Mackewroth, D. (1975) Encyclopaedia of Elec-
tronic Circuits, Prentice-Hall, New Jersey.
Marston, R.M. (1974) Thyristor Projects using SCRs and Triacs,
Butterworths, Sevenoaks. Senturia, S.O. and Wedlock, B.D. (1975) Electronic Circuits and
Applications, Wiley, New York.
Melen, R. and Garland, H. (1978) Understanding IC Operational
Amplifiers, H. W. Sams, Indianapolis, IN. Sheingold, D.H. (1972) Analog–Digital Conversion Handbook,
Analog Devices, Norwood, MA.
Middleton, R. (1977) Digital Equipment Servicing Guide, H. W.
Sams, Indianapolis, IN. Sheingold, D.H. (1974) Non-linear Circuits Handbook, Analog
Devices, Norwood, MA.
Millman, J. (1972) Integrated Circuits: Analogue and Digital and
Systems, McGraw-Hill, New York. Signetics (1977) Analog Data Manual, available from N. V.
Philips Gloeilampenfabrieken, Eindhoven.
Millman, J. and Halkias, C.C. (1967) Electronic Devices and
Circuits, McGraw-Hill, New York. Simpson, A. (1976) Testing Methods and Reliability – Electronics,
Macmillan, London.
Millman, J. and Halkias, C.C. (1976) Electronic Fundamentals
and Applications, McGraw-Hill, New York. Sloot, W. (1972) Solid-state Servicing, H. W. Sams, Indianapolis,
IN.
Millman, J. and Taub, H. (1965) Pulse Digital and Switching
Waveforms, McGraw-Hill, New York. Smith, R.J. (1976) Circuits, Devices and Systems: A First Course
Mitchell, F.H. (1951) Fundamentals of Electronics, Addison- in Electrical Engineering, Wiley, New York.
Wesley, Cambridge, MA. Soisson, H.E. (1961) Electronic Measuring Instruments, McGraw-
Motorola (1961) Power Transistor Handbook, Motorola Inc., Hill, New York.
Phoenix, AZ. Spence, R. (1970) Linear Active Networks, Wiley, Chichester.
Motorola (1966) Silicon Rectifier Handbook, Motorola Inc., Squires, T.L. (1967) Beginner’s Guide to Electronics, Newnes–
Phoenix, AZ. Butterworths, Sevenoaks.
468 Measuring System Design
Starr, A.T. (1959) Electronics, Pitman, London. Turner, R.P. (1978) ABC’s of FET’s, H. W. Sams, Indianapolis,
Studer, J.J. (1963) Electronic Circuits and Instrumentation Sys- IN.
tem, Wiley, New York. Van der Ziel, A. (1977) Nonlinear Electronic Circuits, Wiley, New
Sydenham, P.H. (1979a) Measuring Instruments: Tools of Knowl- York.
edge and Control, Peter Peregrinus, London. Van Santen, G.W. (1967) Electronic Weighing and Pro-
Sydenham, P.H. (1979b) Electronics – It’s Easy (3 Vols), Modern cess Control, Philips Technical Library, N. V. Philips,
Magazines, Sydney. Gloeilampenfabrieken, Eindhoven.
Sydenham, P.H. (1980) Transducers in Measurement and Control, Van Valkenburg, M.E. (1974) Circuit Theory: Foundations and
Adam Hilger, Bristol. Classical Contributions, Dowden, Hutchinson and Ross,
Stroudsburg, PA.
Szentirmai, G. (1974) Computer-aided Filter Design, IEEE, New
York. Wakerly, J.F. (1976) Logic Design Projects Using Standard Inte-
grated Circuits, Wiley, New York.
Terman, F.E. and Pettit, J.M. (1952) Electronic Measurements,
McGraw-Hill, New York. Waller, W.F. (1972a) Electronic Measurements, Macmillan,
London.
Thomas, H.E. (1967) Handbook of Electronic Instruments and
Measurement Techniques, Prentice-Hall, Englewood Cliffs, Waller, W.F. (1972b) Electronic Testing and Measurement,
NJ. Macmillan, London.
Thornton, R.D., Linvill, J.G., Chenette, E.R., Boothroyd, A.R., Waller, W.F. (1972c) Electronic Component Testing, Macmillan,
Willis, J., Searle, C.L., Albin, H.L. and Harris, J.N. (1966) London.
Handbook of Basic Transistor Circuits and Measurements, Waner, W. (1979) Trouble Shooting Solid-state Circuits and Sys-
Wiley, New York. tems, Reston (Prentice-Hall), Englewood Cliffs, NJ.
Turner, R.P. (1959) Basic Electronic Test Procedures, Holt, Rine- Waters, F.J. (1978) ABC’s of Electronics, H. W. Sams, Indianapo-
hart and Winston, New York. lis, IN.
Turner, R.P. (1963) Basic Electronic Test Instruments, Holt, Rine- Wedlock, B.D. (1969) Electronic Components and Measurements,
hart and Winston, New York. Prentice-Hall, Englewood Cliffs, NJ.
Turner, R.P. (1977) ABC’s of Integrated Circuits, H. W. Sams, Willson, A.N. (1975) Nonlinear Networks: Theory and Analysis,
Indianapolis, IN. Wiley, New York.
74: Basic Electronic Components
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
mass, nucleus formed of protons and, perhaps, neutrons.
1 Materials Used in Electrical and Electronic The electron carries, by convention, the negative electrical
Regime 469 charge, the nucleus possessing positive charge that retains
2 Conductors and Insulators 469 one or more electrons attached to it.
3 Semiconductors 470 Normally, charges balance to neutrality, and they usually
attempt to establish this position in a material system.
4 Passive Elements 471
Removal of an electron leaves an atom positively charged;
5 Active Devices 472 this is the positive ion. Addition of an electron forms a
6 Supplementary Devices 474 surplus of negative charge; this is the negative ion. Atoms
Related Articles 475 are combined to form larger molecules, and these can form
References 475 myriads of compounds or mixtures that make up all solids,
liquids, and gases.
When material is formed of charge-stable arrangements
of similar, or different, elements, it becomes unclear as
1 MATERIALS USED IN ELECTRICAL to which atom is attached to which atom of the matrix.
AND ELECTRONIC REGIME This (simplified explanation) results in a matrix formed of
immovable atom nuclei between which there exist electrons
The features of materials and their use that places them moving around in what is known as the sea of electrons.
in the electrical and electronic regime are their electrical This primitive depiction of the make up of materials is
characteristics. Suitable components and assemblies, how- far from the complete, real situation but does provide for
ever, must be made of some form of substance, and design a superficial understanding of the classes of materials that
will always rest on possession of knowledge about other find different use in electrical and electronic technique.
properties of both electrical and nonelectrical materials.
For example, a transistor is formed of silicon, shaped and
arranged with various electrical properties, arranged in var- 2 CONDUCTORS AND INSULATORS
ious positions to which conducting leads are attached to
make electrical connection. The whole is then mounted A material in which the sea of electrons is very mobile,
in a suitable container. Thermal and mechanical require- in which any particular electron can easily be swept away
ments of the transistor are as important as electrical and another attracted into the space resulting, is known as a
characteristics. conducting material. Typical electrical conducting materials
Materials are formed of combinations of the basic ele- are the metals including silver, copper, aluminum, gold, and
ments. Atoms form the smallest level of discreteness that tin. It only requires a small energy level to cause a flow of
generally concerns the electronic and electrical component electrons, the current, in these materials, hence the name
designer. Atoms comprise electrons, of relatively small conductor. As well as solids can, certain liquids and gases
mass, that are in orbit around a central, much larger can be good conductors under certain conditions.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
470 Measuring System Design
Some materials require very high energy potential to where V is voltage potential applied to exert forces upon
obtain electron flow. These are the insulators or resis- the electrons, I is the electric current that flows, and R
tors. Rubber, wood, glass, most plastics, and ceramics are is the resistance to flow, a constant of proportionality for
typical insulating materials. Liquid and gaseous insulators the material.
also exist. The electrical resistance between two points on a piece
Conductors and insulators are at the two ends of the range of material is decided by its specific resistance, the cross-
of resistivities of materials. In reality, there are different sectional area of the material, and the distance between end
materials available with specific resistivities covering the connections.
range from one extreme to the other.
Electrical engineering in the main is (or was so in the
past) concerned with insulators and conductors and their 3 SEMICONDUCTORS
application. The former are used in suitable arrangements
These form a class of material in which current flow is
to contain charge (Figure 1) and its flow so that electric
not necessarily proportional to impressed electrical poten-
current flows in space where it is required. In a wire trans-
tial. They have resistivity values just above those of metals.
mission system, for example, conducting wires are insulated
The physics of semiconductors can be most complex. For-
to ensure that the electrical energy arrives at the desired
tunately, electronic systems development is more likely to
destination with as little loss as possible. In an electrical
need only understanding of what semiconductors can do
generator, insulators force current to flow in conducting
than of why and how they actually provide their char-
wires that are revolving in suitably placed magnetic fields.
acteristics. For this reason, it is not necessary to provide
Materials that form insulators and conductors generally
an in-depth study here. As time passes, the electronic sys-
obey the linear Ohm’s law given by
tems designer moves further away from such fundamental
V = RI (1) considerations, the only people concerned being device
Resistor
resists
current
flow,
Switch starts and produces
voltage Insulated wire
stops charge flow guides electric charge
drops
−
Primary cell Output current
battery
produces
+
electric
charge Input
current
Current
Capacitor
stores Active transistor
charge element provides
energy gain
Microphone converts Loudspeaker converts
sound pressure into electric charge flow
electric charge into sound pressure
changes
Figure 1. In a simple circuit (a sound amplifier is shown), selected materials provide electrical charge, controlling current flow into
the speaker where it is converted back into acoustic energy at greater than input level.
Basic Electronic Components 471
developers and research workers seeking new devices systems exhibit zero-order dynamic behavior; capacitive
and principles. and inductive systems can be used to realize higher-order
Semiconducting materials can usually be distinguished dynamic systems as their number is increased in suitable
from conductors by their property that thermal energy arrangements. The behavior of systems of passive compo-
will cause electron movement, which is not the case for nents forms much of the material of introductory electrical
conductors. This is usually (again simplistically) discussed engineering.
in terms of the energy gap existing in the semiconducting Calculation of circuit behavior for resistive systems is
material. If additional energy causes electrons to vacate relatively simple compared to those circuits having stor-
their positions, causing current flow, the material is denoted age elements. The simpler resistive circuit theory can,
as n-type. The spaces left are termed holes. A second however, often be applied to storage element systems,
form of current flow can occur in which adjacent valency provided the storage element is considered as a frequency-
electrons move to fill the holes. The materials in which this varying impedance.
happens are termed p-type. As electrons move, they leave The capacitive reactance (its apparent resistance) of a
holes, thus giving the appearance that the holes migrate. capacitor is given by
Suitable additions of impurity into extremely pure semi-
conducting materials (germanium was first used but silicon 1 1
XC = = (2)
is now more generally the basic material in semiconduct- 2πfC ωC
ing devices and with gallium-arsenide for high-temperature
and the inductive reactance of an inductor by
uses) can enhance the n-type and p-type behavior.
When a conductor is interfaced physically with a semi-
XL = 2πfL = ωL (3)
conductor, or two of the latter materials are so connected, a
flow of electrons occurs across the boundary, reaching some where f is the frequency of interest, in cycles per sec-
form of equilibrium situation. These connection regions are ond (Hz), ω the angular frequency in radians per second
called junctions. (rad s−1 ), C the capacitance in farad (F), and L the induc-
tance in henry (H).
The time-transient behavior of either a capacitor or an
4 PASSIVE ELEMENTS inductor combined with a resistor is a first-order response,
that is, the rise or fall of signal level to a step input follows
Subsystem components used to build energy conversion an exponential change given by the time constant τ of the
systems can be divided into passive and active kinds, combination in question; (τ = RC or R/L). Resistance acts
these two being combined into a system to provide the to alter the time constant and it gives rise to losses in RC
required response. and LC combinations. Such losses are not necessarily of
Passive components are those devices that operate upon importance. Much of electronic engineering makes delib-
signals to provide conversion between voltages and current erate use of practical arrangements in which loss occurs,
without being able to increase the output energy level this being quite unlike power electrical engineering where
beyond that of the input signal. The output signal energy reduction of losses is generally a prime design target. In
is always less, because of losses, than the input signal recent times, engineering of low energy level electronic
energy. A combination of resistors, for example, can be information signal systems has also concentrated on less
used to attenuate a signal’s energy but not to increase it. lossy arrangements as a means to increase the density of
It is, however, possible to transform the energy product components and to reduce the size and cost of the necessary
of voltage and current so that one is made larger than the power supply.
original form at the expense of the other that is then made Networks comprising only resistances can be converted
smaller. Examples of such transformations are the voltage into equivalent networks or be reduced to a single equiva-
step-up, or step-down, transformer, and the resonant circuit. lent resistance value between two terminals. Inductors and
The mainly used passive components are the resistor, the capacitors can only be reduced to equivalent resistive val-
capacitor, and the inductor. The resistor acts to dissipate ues at a given frequency of operation and when one kind
energy and to resist energy flow by wasting what it does not does not interact with others. Generally, this means only
allow to flow through it. The capacitor and the inductor, on series or parallel combinations of a like kind (of inductor
the other hand, can store energy; they do possess losses but or capacitor) are reducible. When the two forms interact
these are normally small compared with the storage energy to form a second-order or higher-order system, reduction
capability. It is the energy storage property that enables sig- is more complex as it must take into account the vectorial,
nal frequency processing to be obtained. Purely resistive amplitude, and phase nature of the two components.
472 Measuring System Design
Leading phase
As the windings of an electrical transformer form an
X L = 2pfL inductor, it is possible to resonate each winding with a
capacitor. This forms the tuned transformer, each or one
side only being used to increase the sharpness of the fre-
X L−X C ( = 0 at f r)
quency selection process provided by resonance. Selectivity
Reactance is quantified in terms of the Q-factor. The higher the Q-
0
magnitude factor, the sharper the resonance; its magnitude depends
Frequency
1
very much upon the resistance losses in the resonant system.
Xc = Such losses can be made up by the use of active amplify-
2pfc
fr ing elements.
While the resistor, the capacitor, and the inductor are
Lagging phase very common in electrical and electronic systems, there do
exist other passive elements. Some of these are the trans-
Figure 2. Behavior of inductive and capacitive reactance with former, rectifier, temperature sensitive resistor (thermistor),
frequency fr is the resonant frequency. light-generating device, the panel indicator, and other trans-
ducers. These, however, can each be reduced, in theory, to
When the two different forms of storage elements are combinations – called the equivalent circuit – of resistance,
combined, second- or higher-order systems are formed. inductance, and capacitance.
Losses arising from resistive elements can still be signifi- They are consequently generally regarded in theoretical
cant but they are, there, usually of secondary importance in circuit considerations in an equivalent form. A thorough
basic theoretical considerations. understanding, therefore, of the characteristics of resis-
Figure 2 shows how inductive and capacitive reactance tors, capacitors, and inductors, and their combination forms
vary with frequency. At a certain frequency, called the a most basic and important part of electrical and elec-
resonant frequency, where the two reactances XL , XC are tronic method.
equal in magnitude, their vector addition results in a com-
bined impedance of zero. This is because they each possess
an opposite phase angle. In terms of the complex algebra 5 ACTIVE DEVICES
description, the imaginary components cancel, leaving only
the zero, real component to resist current flow. Networks formed only of passive devices can be devised
Resonance, therefore, occurs at that will provide much of the signal processing needed
in electronic systems. However, their inability to maintain
1 the signal energy level at, at least, that driving the system
ωL = ωC fr = √ (4)
2π (LC) stage is a serious disadvantage for the signal progressively
degrades down to the noise energy level and becomes lost.
where ω is the angular frequency, fr the resonant frequency, Furthermore, the signal input energy may be insufficient
L the inductance in henry, and C the capacitance in farad. to drive the required output transducer. As an example,
In practice, no resonant system is free of loss; the real, a signal can be mathematically integrated reasonably well
resistive part has a finite value. This acts to limit the current with a simple RC, low-pass filter stage. To obtain accurate
in the resonant circuit at resonance: the less the resistive integration, however, it must also provide high attenuation.
component, the higher the series resonance current. Some method must usually be added to the output to restore
Resonance using an inductor and a capacitor can be the signal level – this is called an amplifier. (Early circuit
established as a series or as a parallel connection of the designs did just this but it is often more powerful to use
two, and each of these can be used in a series or a shunt the amplifier in an alternative less-obvious arrangement to
connection in a system. The availability of four options can obtain an improved overall performance.)
lead to confusion about how a resonant stage influences If the network has the ability to make up the circuit losses
the system. It is, therefore, necessary to verify which or increase the energy at the output to more than occurs at
connection is used. In electronic method, where power the input, the arrangement is known as an active system.
loss is often permissible, shunting is commonly used to Any arrangement that has active properties is capable
selectively remove signal frequencies that are not needed. of providing amplification. Amplifiers are able to provide
This apparently wasteful but useful practice in electronic larger energy signals at their output than occur at their
systems could not be tolerated in power electrical and input. At first sight, it appears that the network provides
electronic engineering. the extra energy needed, but in reality the amplifier acts as
Basic Electronic Components 473
Source of steady power 150 −I b(mA)
1000
800
100
Input signal
−I c(mA)
600
controls flow Output signal
of power into 400
output device
Active device 50
200
100
Implied common lead 50
0
10 20
Figure 3. Generalization of a three-terminal amplifier element.
−VCE(V)
a control that allows flow of energy from a supply into the Figure 4. Characteristic curves for a typical, small-signal p–n–p
output in accordance with some relationship with the input transistor. Ic is the collector current, Ib is the base current, and
Vcc is the collector to emitter voltage.
signal. The output signal is often not electrically connected
to the input signal. Figure 3 shows the generalization of
an amplifier. has some active elements available. The magnetic amplifier,
In electronic systems, amplification, also called gain, is the rotating electromechanical amplidyne amplifier, and the
used in many ways for many purposes. Each of the uses Ward–Leonard motor-generator arrangement are methods
has a different terminology. For example, a preamplifier for obtaining gain. They are, however, not practical at very
conditions a low-level signal ready for connection to a small power levels, the devices being too large compared
following stage. A stage inserted to obtain a more optimum to even valves for them to have been adopted in signal pro-
stage match is called a buffer amplifier; as the name implies, cessing in the widespread manner that transistors have over
it buffers one stage from the other so that they do not the past years.
interact. At the output end of the cascaded system, there The most basic amplifier, in the general sense, is a three-
is often a need for a power amplifier; this raises the signal terminal device having characteristics between the three
power providing the specific matching needed to suit, say, terminals that can be expressed along the lines shown in
the actuator coupled to the output. Figure 4 for a representative electrical component. Tran-
Another is the instrumentation amplifier. These are sistors, valves, and any other amplifier can be described
vaguely defined units that may carry out any, or all, of in this way. Gain occurs when signal input energy can be
these functions: the term generally implies that they are used to reproduce the same signal at a larger power level at
able to preserve signal purity and thus retain the meaning the output. Several other forms of characteristic describing
of measurement signals passing through the system chain. curve can be drawn.
The operational amplifier is an instrumentation amplifier The transistor is the most-used active electronic element
designed to use feedback in an analog computational mode: today. It is capable, in different forms, of handling very
they are popularly called op-amps. small (picowatt) signals through to very large (kilowatt)
Each of the above active amplifiers needs to be designed signals. It is not practical to use exactly the same actual
to suit the application in question. Although there are a cer- transistor assembly for all cases; the designer chooses that
tain number of basic designs in existence, the specification unit which is most suitable for the purpose in question.
of an amplifier can only be achieved properly and satisfac- Since the development of the transistor, there have appeared
torily when the task and the amplifier are well understood. many other active semiconductor devices that can provide
It is the availability and use of active devices for pro- gain. Each has its special virtues; devices include the silicon
cessing information that largely distinguishes electronics controlled rectifier (SCR), the field-effect transistor (FET),
from electrics. Prior to the thermionic valve era, that began the thyristor, and many more. Sophisticated junctions have
around 1910, it was very difficult indeed to obtain gain been devised using a variety of manufacturing methods,
in a system. The thermionic valve overcame that prime these including multiple junction devices. For most practi-
need, allowing numerous active networks to be devised and cal purposes, it is only necessary to learn about the actual
implemented. It enabled electronic technique to develop. physics of operation within the device on rare occasions.
Following the valve came the solid-state semiconductor Generally, use of makers’ published characteristic curves
transistor element. This performs the same role as did and numerical values suffices.
valves, operating it in quite similar ways, although not at As it stands, the transistor, as were valves, is not capable
the electron level of understanding. Electrical method also of amplifying AC signals centered at a zero DC voltage
474 Measuring System Design
level. This is because the basic transistor p-n-p, or n-p-n, example are new modules that are appearing as the result
junctions can only pass current in one direction. Only of electronic technique being used to a greater extent in
one polarity of the alternating current signal waveform mass-produced motor vehicles.
fed into the base connection will be amplified, the other Thermionic devices, although no longer used much for
being severely attenuated. It is, therefore, necessary to add basic amplification, still find some places where they have
extra passive components to the transistor to obtain practi- not yet been replaced by superior solid-state equivalents.
cal operation. The design of transistor amplifiers used to be Examples are the cathode-ray tube used in television moni-
very important to electronic system builders but the intro- tors and in the cathode-ray oscilloscope and the older forms
duction of low-cost, high-quality integrated-circuit ampli- of video camera. Thermionic devices are also used to obtain
fier modules enabled highly specialized amplifier designers spectral emission in gas analysis equipment. The electron
to apply considerable effort to the design and to use many microscope also requires electron emission from a cathode.
more active elements in the design than would have been Such supplementary devices are not normally built by
used in former times. Thus, it is common practice now to the less specialized user but are obtained ready to use. If
assemble systems using commercially available units cho- they fail, they are replaced rather than being repaired.
sen to suit the need. The user needs to know little about Another example of a supplementary device is the num-
internal amplifier design; only the external characteristics. ber and letter alphanumeric display device. Such devices
The active element can be used to produce gain for use again are purchased ready to wire into the circuit. Orig-
with continuous linear signal analog systems, or it can inally, these were manufactured by the user and were a
be used to provide gain through a switching action for significant part of the cost of a system. Today, they are often
use in digital systems. In early times (1840s onward), the integrated into the complete, ready-to-use, subsystem mod-
electromechanical relay was able to provide gain; it was a ule complete with any decoding and power drivers required.
prime component in enabling telegraphy to be developed Electronic systems need mechanical frames to support
so successfully. the various circuit boards, switches, variable controls, and
The design of amplifiers is covered in many texts. connections to other stages. These are called chassis in
Of those already listed, refer to Brophy (1977), Carson older terminology and more recently frames. Most net-
(1961), Clayton (1971), Cleary (1969), Csáki et al., (1975), work systems are today formed in their final version on
Delaney (1969), Faulkenberry (1977), Graeme and Tobey printed-circuit boards. These are boards in which all wiring
(1971), Heathkit (1978), Lowe (1977), Malvino (1967), connections between the commercially made components
Millman and Halkias (1976), Senturia and Wedlock (1975), are made by selective etching of a copper coating to form
Sydenham (1979), and Thornton et al. (1966). conducting tracks. Holes are then drilled in the copper
tracks, so formed, to take the leads of the components,
which are then soldered into the copper by hand, or solder-
6 SUPPLEMENTARY DEVICES flow methods. Wire-wrapped connections are also used,
especially where greater reliability is called for. Printed
To the range of passive and active devices mentioned so wiring boards, called p.w.b. boards, are also used with
far must be added a very wide range of supplementary wiring on both sides, selected holes being plated through
components and devices. These are combined with the to connect both sides where needed. The actual layout of
previously described active and passive elements to form components on a board more often is to suit manufacturing
electronic instrument (and other) systems. It is not practical, convenience and does not then follow the schematic block
or necessary, to cover all of these as a description of a few diagram layout.
will give the understanding needed here. Printed wiring boards are often connected into the com-
The variety available is most easily seen by reference plete system using plug and socket connectors. This makes
to electronic component suppliers’ catalogs. Electronics for easy repair and faultfinding. It also modularizes the sys-
always has had a popular following. It began with radio tem design.
in the 1920s, being added to by audio equipment and A smaller circuit assembly can be made by printing suit-
television interests. Popular interest then moved somewhat able materials onto a usually ceramic substrate – called
toward all manner of instrumentation for the domestic printed circuitry. These plus, vacuum-deposited conduct-
person’s entertainment and use. A typical suppliers’ catalog ing metal connecting tracks, make up the passive part of
has around a hundred pages of listings. hybrid circuits. To these are added unencapsulated semicon-
Systems are generally built assembling the commonly ductor active elements formed by the monolithic process.
available parts but design needs, where a very large market A complete hybrid circuit may then be totally encapsulated
is assured, will allow new products to be considered. An to provide environmental protection.
Basic Electronic Components 475
Still smaller assemblies are made by the monolithic need for skill and understanding arises if they do not func-
methods using masks to lay down progressively conductors, tion properly!
junctions, and components onto the usually silicon chip
base. A very large capital investment is needed to create
integrated circuits (ICs). The number of circuit elements RELATED ARTICLES
possible on a single IC now ranges from the original 15 or
so to greater than one million. Article 73, Overview of Electrical and Electronic Tech-
A myriad of major and minor supplementary components nique, Volume 2; Article 75, Electronic System Build-
ing Blocks, Volume 2; Article 76, Electronic Systems
are also used (see Figure 5). Some have already been men-
Design, Volume 2; Article 122, Operational Amplifiers,
tioned: the switch, plug and socket, printed wiring board.
Volume 2; Article 123, Instrumentation Amplifiers, Vol-
It is the ready, and relatively inexpensive, availability of
ume 2; Article 124, Specialized Amplifiers for Measure-
electronic components that allows people with relatively
ment Systems, Volume 2.
little training to form quite sophisticated electronic instru-
mentation systems. In many instances, the module required
can be purchased in kit form ready for insertion of the REFERENCES
components onto a ready etched and plated, printed wiring
board. Very little skill is needed to assemble these. The Brophy, J.J. (1977) Basic Electronics for Scientists, McGraw-Hill-
Kogakusha, Tokyo.
Carson, R.S. (1961) Principles of Applied Electronics, McGraw-
Hill, New York.
Clayton, G.B. (1971) Operational Amplifiers, Butterworths,
Sevenoaks.
Cleary, J.F. (ed.) (1969) Transistor Manual, General Electric
Company, Chicago, IL.
Csáki, F., Ganszky, K., Ipsits, I. and Marti, S. (1975) Power
Electronics, Akadémiai Kiadó, Budapest.
Delaney, C.F.G. (1969) Electronics for the Physicist, Penguin
Books, Harmondsworth.
Faulkenberry, L.M. (1977) An Introduction to Operational Ampli-
fiers, Wiley, New York.
Graeme, J.G. and Tobey, G.E. (1971) Operational Amplifiers –
Design and Application, McGraw-Hill, New York.
Heathkit (1978) Electronic-Courses for Learn-at-Home: Course
1 DC Electronics; Course 2 AC Electronics; Course 3 Semi-
conductor Devices; Course 4 Electronic Circuits; Advanced
Digital Technique Course; Microprocessor Course. Available
with experimental back-up and optional cassettes, Heath-
Schlumberger, London.
Lowe, J.F. (1977) Electronics for Electrical Trades, McGraw-Hill,
Sydney.
Malvino, A.P. (1967) Electronic Instrumentation Fundamentals,
McGraw-Hill, New York.
Millman, J. and Halkias, C.C. (1976) Electronic Fundamentals
and Applications, McGraw-Hill, New York.
Senturia, S.O. and Wedlock, B.D. (1975) Electronic Circuits and
Applications, Wiley, New York.
Sydenham, P.H. (1979) Electronics – It’s Easy, (3 Vols.), Modern
Magazines, Sydney.
Thornton, R.D., Linvill, J.G., Chenette, E.R., Boothroyd, A.R.,
Willis, J., Searle, C.L., Albin, H.L. and Harris, J.N. (1966)
Figure 5. A range of supplementary devices used to form elec- Handbook of Basic Transistor Circuits and Measurements,
tronic systems. Wiley, New York.
75: Electronic System Building Blocks
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
amplifier. These were taken up en masse because of the
1 Linear Units 476 versatility that they offered, when appropriate feedback
2 Amplifier Terms 476 connections are applied.
3 Digital Units 479
4 Gates 479
5 Astable Device 480
2 AMPLIFIER TERMS
6 Bistable Device 480
The satisfactory operational amplifier, the op-amp, must
7 Monostable Device 480 have open-loop gain (without feedback connections) of at
8 Inverter Device 480 least around 10 000, and draw a relatively insignificant input
9 Schmidt Trigger and Comparator 480 current. It must also be reasonably stable with temperature
10 Overall Features of Digital Systems 481 and time, for any drift of its DC levels internally will
Related Articles 482 provide a related output drift. Furthermore, it should be able
References 482 to handle both polarities of input signal. Simple integration
allowed designers to package a design (Figure 1) with
around 20 active junctions, plus some 12 passive resistors,
and a few capacitors. Inductors are generally avoided in
1 LINEAR UNITS integrated circuits due to their relatively large size in the
integrated form at low frequencies of operation. The cost of
Electronic systems are built using a relatively small number the IC op-amp is so low as to now be almost insignificant
of various kinds of building block units. They are divided when compared with assembly costs to place it into use.
into two main groups: those using analog signals and those Naturally, a general purpose op-amp may not be suitable
using digital operations. Here the former is considered. for more exacting uses and more expensive designs are
After many decades of research and development, a marketed for such cases. The selection of the op-amp must
satisfactory general purpose amplifier came into being in be made to suit the application. Guidance is generally given
the 1950s for use in the first electronic computers: they in texts on op-amp use and in makers’ application notes.
used analog techniques. The demand for these was rather To understand why the op-amp is so versatile, consider
specialized and their cost and availability were, therefore, the general case of an amplifier with a single feedback path
rather restricted. However, the introduction of integrated and multiple inputs, as shown in Figure 2. To simplify this
circuit semiconductor technology in which the active and explanation, only resistors are considered but in practice
passive components of systems that have many tens to the same derivation can be made for complex impedances
hundreds of elements in their construction and are made by in place of the resistors.
mass replication methods using photomechanical methods, It can be shown that the transfer characteristic between
led to the availability of a very cheap and stable DC the inputs V1 , . . . , Vn and the output Vo , if the amplifier
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Electronic System Building Blocks 477
Inverting input
+Vcc
Q8 Q9 Q12 Q13
Q14
Q15
Noninverting R6
Q1 Q2 R5
input 27 Ω
30 kΩ
Q18
Output
Q3 Q4 Q19 R7
R10 22 Ω
C1
30 pF Q21
50 kΩ
Q20
300 Ω
Q7
Q16 Q22
Q5 Q6 Q17
Offset Q10 Q11 Q23
null R1 R9 R 9 Q24
R3 R 11
1 kΩ R2 R4 50 kΩ 100 kΩ
50 kΩ
1 kΩ 50 kΩ
5 kΩ
−Vcc
Offset
null
Figure 1. Schematic layout of circuitry of an integrated circuit amplifier.
Ro feedback-connected op-amp is decided only by the ratio of
the input and feedback resistors. These can be made very
R1 stable and accurate in value. As the gain of the op-amp is
V1
not a critical parameter, it can be seen that variation of the
R2 −A Vo
V2 open-loop gain of the amplifier does not alter the feedback
connection gain, this leads to increased stability by the use
O of feedback. The gain of the op-amp is made very large to
Rn
Vn allow for closed loop operation.
The gain stability and selectivity features are but one
Figure 2. Generalized op-amp feedback connections.
virtue of op-amp feedback usage. It can also be seen
from the above expression that input voltages are summed
draws negligible current, is given by according to a multiplier factor for each input channel.
Furthermore, a single input signal appears at the output
Ro Ro Ro
Vo = − V + V + ··· + V with the opposite polarity. This is called an inverting
R1 1 R2 2 Rn n arrangement. Modern op-amps provide both inverting and
−1
1 R R R noninverting inputs. Their output can swing around a zero
× 1+ 1 + o + o + ··· + o (1) voltage level by a typical ±10 V. They are rarely used in
A R1 R2 Rn
open-loop due to the existence then of extremely high gain
This simplifies greatly if the gain A of the amplifier, in the (100 000 is typical), which has associated with it instability
open-loop state, is very much larger than unity, say at least that rises in magnitude as the gain of the feedback system
10 000. Then the above becomes simply arrangements moves toward the open-loop gain value.
By suitable application of feedback connections, the op-
n
Vi amp can be used to mathematically integrate, differentiate,
Vo = −Ro (2)
Ri multiply, divide, summate, subtract, convert from voltage
i=1
to current source, provide buffer amplifiers of many kinds,
This expression is, therefore, independent of the actual act as an analog signal filter, and more. The number of
amplifier gain and shows that the gain of the possible uses for linear circuit operation is enormous and
478 Measuring System Design
− −
Vin Vin
+ Vout + Vout
O O
(a) (f)
− −
O Vin
+ Vout + Vout
Vin O
(b) (g)
−
+ Vout
Vin
Vin 1 −
Vin 2 + Vout
Vin 3
O O
(c) (h)
R
Vref
−
(Constant)
+ Vout −
Vin + Vout
O
(d)
−
+
Iin
−
+ Vout
O O
(e) (i)
Figure 3. Some selected op-amp-based functional blocks (power supply connections not shown) (a) voltage amplifier, inverting;
(b) voltage amplifier, noninverting; (c) voltage summation with amplification, inverting; (d) Vout proportional to R; (e) current to voltage
converter; (f) integrator; (g) differentiator; (h) AC, voltage amplifier; (i) notch frequency filter.
grows continuously. Figure 3 shows some commonly used (1971), Huelsman (1977), Melen and Garland (1978), and
examples that illustrate this versatility of use. NS Corporation (1976). Most general electronic texts that
Op-amps are usually selected and designed according to originated after around 1965 include articles on op-amp
application notes provided by their makers and to designs use. Several manufacturing companies specialize in high
published in the, now many, texts on their use. Application performance op-amp production; they also provide relevant
notes are generally obtained on request from the op- users guides and design assistance.
amp marketing agencies. Suitable texts concentrating on The actual manufacture of integrated circuits is discussed
their use are Burr-Brown (1964, 1966), Clayton (1971), in most introductory texts about general electronics. It is
Connelly (1975), Faulkenberry (1977), Graeme and Tobey continuously changing in refinement and a bewildering
Electronic System Building Blocks 479
range of terms have been coined to describe the conceptual and then to solid-state semiconductor devices, such as the
basis of the new technology. As it is not particularly transistor and the diode as the basic switching element.
important to the majority of electronic systems generation, The digital family of devices comprises relatively few
for measurement applications it will not be discussed basic units, but as can be expected they are made in many
here. The user should not be daunted by a lack of forms. In the 1970s, a considerable degree of standardiza-
ability to understand just how an integrated circuit of any tion came about and two prime logic systems came to the
kind operates internally. Schematic diagrams are usually fore. These are the ones using transistor–transistor logic
published for ICs but these are of passing interest to (TTL) and those based upon complementary metal oxide
the user, becoming important only when the information semiconductor (CMOS) logic families of device. TTL sys-
provided elsewhere on the data sheet and application note tems are always used with high and low logic levels of
fails to satisfy the need. On the very rare occasion, it will be nominally 5 and 0.5 V, respectively, CMOS, however, can
found that internal connections may give rise to unexpected be used at varying levels of supply. The range of digital
operation when the IC is used in a system. devices also is often referred to as the logic range. Since
Op-amps and other amplifiers (buffers, servo, instrumen- then, all manner of logic types have been developed. For-
tation, interface, and converter) form only part of the range tunately, to build digital systems one does not need such
of linear devices. Others are voltage and current regulators, detailed internal knowledge. Basic digital building units are
analog-to-digital (A/D) and digital-to-analog (D/A) con- now described.
verter units plus sample-and-hold units all that are used
in Data Acquisition Systems (DAS). Other linear modules
include specialized radio, wireless, and audio integrated 4 GATES
circuits; video amplifiers for television use arrays of tran-
sistors and diodes, timers, numerous different purpose units Systems that perform logical AND, OR and the comple-
made for use in industry and automobiles (special experi- mentary negated functions NAND, NOR plus some special
mental runs are sold for original equipment manufactures cases, such as EXCLUSIVE OR, are called gates. As an
OEM to develop large volume products), display drives, example, two electrical switches in series will provide trans-
phase-lock loops, and voltage level comparators. Some of mission of a signal if both are closed, that is, if switch A
these transcend the linear and digital boundary enabling AND switch B are closed. This is denoted as A.B. Placed in
conversion from one signal domain to the other, a very parallel, A or B will give transmission; denoted as A+B.
common practice. The negated NOT form results when the opposite switch
state case is considered, the sense of the switches being
reversed; denoted A · B, A + B, A, B.
3 DIGITAL UNITS Modern practice no longer carries out gating func-
tions with mechanical contact switches (some will be still
Information can also be processed using digital signals encountered in special applications), the active semicon-
through application of logical principles. The foundation ducting junctions being used as equivalent switches instead.
of these principles was laid by philosophers of the late In use, logic ICs are provided with adequate buffering
nineteenth century, who had little inkling that one day and suitable interfacing such that the user only has to
their work would be applied to actual electrical hardware. first decide the logic gating needed and then connect the
Boolean algebra handles, in a mathematical sense, the units accordingly. For large sets of gates – hundreds to
interaction of logical situations. thousands – more advanced gate-array methods are used.
These are typified by the use of simply two-state con- The internal design of logic units is, again, of little
ditions, the most general and basic being connective state- importance to the user. The number of gate functions
ments, such as AND and OR. For example, a switch must available in a particular IC is decided by the number of
be turned on if input A AND input B are energized, but pins available for connection to external circuitry, two being
not for either one of them. It could also be that the switch needed to supply power to the IC. For example, a 14-pin IC
is on if A OR B is on. can provide 12 gate connections, which can be supplied by
This kind of thinking and problem solving was first the maker as either four, two-input/one-output gates; three,
developed to help philosophers conduct arguments on three-input/one-output gates; or six, one-input/one-output
strictly objective logical lines. Their methodology was signal inverters (these negate the logic polarity, but do not
applicable to electrical switching networks and the first digi- invert the signal polarity in the same way as does a linear
tal (computing) systems were formed using electromagnetic signal inverter). Logic ICs usually contain only one kind of
relays. These gave way to faster-acting thermionic valves function on the IC chip; basic functions may be combined
480 Measuring System Design
together to form a higher-level function. The exception Several forms of flip-flop IC are available, some include
occurs when a much greater scale of integration (large- gating into its inputs.
scale integrated circuits (LSI) to very large-scale integrated
circuits (VLSI)) is used.
A basic theorem about logic gates shows that all digital 7 MONOSTABLE DEVICE
logic systems can, in fact, be formed from just NAND gates.
In practice, however, such methods are not cost-effective If one intercoupling is made capacitive and the other
and makers subsequently offer a range of different ICs for a DC link (combination of the astable and the bistable
appropriate selection. To make full use of its available logic connections), the system will toggle over for an input
elements, the gates on one physical IC unit are connected to transition returning to the first state after a time period
different points of the circuit. Many of the functional units set by the circuit’s passive components. This is able
that are needed can be formed from gates alone, but again to provide a small time delay to digital signals or to
it is generally more practical to use purpose-designed units. restore the squareness of incoming signals as a repeater
Two active elements can be so interconnected that there for digital signal restoration. These are called one-shots, or
exist two inputs and two outputs with a large degree of monostables.
positive feedback between them. The interstage passive In each of the above multivibrator units, there are two
element coupling used between the two elements decides outputs, one being the negated or complementary output of
the resulting function. This family was originally described the other. Either output can be used depending on the logic
as the multivibrator set. The three forms are the astable, sense needed. This often allows an inverter to be omitted
bistable, and monostable. These provide three other basic to create the required logic gate system.
digital system building blocks.
8 INVERTER DEVICE
5 ASTABLE DEVICE
Already mentioned is another basic digital unit, the inverter.
If the interstage coupling between two amplifying units This is used to invert the logic sense of a signal. The need
(transistor) is capacitively, or AC, coupled to provide a for this often arises in digital system design.
large amount of positive feedback from one active element Inverters also can act to buffer one stage from those con-
to the other, the system will oscillate from one side in nected to it. Digital systems often require many connections
the low state, with the other at the high state, changing to an output. The ability of a unit to drive a given number
the states cyclically to form a square-wave output source. of other inputs is expressed by its fan-out. In TTL systems,
The parameters of the interstage circuit largely control the number refers to standard TTL inputs; other uses have
the frequency of the switching. These units are also to be decided on their merits.
called clocks (because they are used to pace in a digital
computational system) or square-wave generators.
9 SCHMIDT TRIGGER AND
COMPARATOR
6 BISTABLE DEVICE
When a varying level analog signal needs to be quantized
If the two stages are DC connected to provide the positive into the two standard digital levels, the Schmidt trigger
feedback to the other, the system will rest with one side can be used. This unit provides a rapid square-edge output
high and the other side low. A change of the state occurs if transition from a high to a low, or vice versa, state at a
one input is fed with sufficient energy to toggle the system given level of input of analog signal. The trigger level for
to the other state. Such a unit can divide pulses by two a rising signal may be higher than for a falling signal, the
as each output state only repeats a state after two input difference is called backlash. Backlash is often deliberately
transitions. In early bistable designs, true short-duration enhanced to give the trigger system a window in which it
pulses of energy were used; today, it is the transition energy will not keep retriggering until the signal level has changed
of a signal going from a high to a low state (or vice versa) its analog level by more than a given amount.
that operates these units. These are more commonly called A similar element is the comparator. In this, two voltage
flip-flops, binaries, two-states, and memories. They can be levels are compared. If one exceeds the other, the output
used to store a binary state for a chosen output that remains toggles. By fixing one as a reference voltage, a comparator
at a given state until toggled, or the power is disconnected. toggles at that level, as does a Schmidt trigger.
Electronic System Building Blocks 481
10 OVERALL FEATURES OF DIGITAL
SYSTEMS
The above listed units form the basic building blocks of
all digital systems. Figure 4 illustrates this point. Digital
systems design rests on understanding their characteristics
(a) (f)
at the conceptual level. Where costs justify it systems
formed of several of these units are manufactured as
large-scale integrated (LSI) or even larger very large scale
(VLSI) integrated circuits. With time, increasing numbers
of basic functions are being made available as more
sophisticated and powerful, higher level systems units. They
are the subject of Article 76, Electronic Systems Design,
Volume 2. For many projects, it is economical to design
(b) (g)
and make what are called application specific integrated
circuits (ASICs).
Digital systems convey their information primarily as
the timings of the level transitions; not by their signal
levels. The actual voltage, or current, levels of the square
waveforms are not of the same significance as they are
in analog systems. For this reason, rise and fall times
are kept as short as possible. Modern systems will make (c) (h)
the transition in fractions of microseconds, that is, in
nanoseconds. Circuit connections and component layout
can degrade the switching edge. For this reason, high- J
speed circuits must be carefully laid out and interconnected J1
J2 P
Q
Clock
inside their integration layout. Long interconnecting leads K2 Q
K1 S
are especially bad unless designed appropriately. The fast
K
rise times of the signals also lead to generation of stray
signals. These can influence other parts of the circuit (d) (i)
by electromagnetic induction. Furthermore, the energy
transients within a stage can cause influencing transients
KC Q
on the power supply line. This would affect other units Clock
in the system. For this reason, it is usual practice to add J Q
decoupling filters to each small group of ICs to prevent
J Q
signal unwanted transients disturbing the operation of others Clock
on the same mount. KC Q
The testing procedures for digital systems are quite (e) (j)
different from those of analog circuits. First, information
is only available about logic levels at various positions in Figure 4. Schematic diagrams of some typical integrated digital
the circuit. Display, using a logic analyzer, of numerous circuits: (a) quadruple two-input positive NAND gates; (b) hex
inverters; (c) eight-input positive NAND gate; (d) dual four-input
states is used to fault-find a digital system. Another feature
positive NAND gates; (e) triple three-input positive NAND
is that the system can be stepped, or clocked on, one step gates; (f) quadruple two-input positive NOR gates; (g) four-wide
at a time at any speed that is convenient. Analog systems two-input AND–OR invert gates; (h) monostable multivibra-
must act in real time to provide proper operation of many tor; (i) edge-triggered j–k flip-flops; (j) dual j–k master-slave
of the circuit functions. Timing between stages is important flip-flops.
in some digital designs and time delay may need to be
considered in what is called a race condition. occur in the first realized logic network design. In many
The design of digital switching systems can be made very applications, however, it is often less expensive overall to
sophisticated using well-established logical mathematical leave redundancies in the system; their location and elimi-
and graphical procedures. Design is largely aimed at reduc- nation can be very time consuming and requires an in-depth
ing the presence of redundant logic gates that generally understanding of switching theory.
482 Measuring System Design
Digital systems units are covered in Blakeslee (1975), Frenzel, L.E. (1997) Crash Course in Electronics Technology,
Bouwens (1974), Brophy (1977), Diefenderfer (1979), Butterworth-Heinemann.
Heathkit (1978), Kostopoulos (1975), Lancaster (1976, Grob, B. (1996) Grob Basic Electronics, Glencoe McGraw Hill.
1977), Middleton (1977), Millman and Taub (1965), Mill- Gibilisco, S. (1997) Teach Yourself Electricity and Electronics,
man (1972), Namgoster (1977), Sydenham (1979), Wakerly McGraw-Hill.
(1976), Frenzel (1997), Gibilisco (1997), Grob (1996), Graeme, J.G. and Tobey, G.E. (1971) Operational Amplifiers –
Horowitz and Hill (1989), Khalil (2001) and Mano (2001). Design and Application, McGraw-Hill, New York.
More detailed works exist on the various aspects, such Heathkit (1978) Electronic-Courses for Learn-at-Home: Course
as switching theory, Boolean algebra, and manufacture of 1 DC Electronics; Course 2 AC Electronics; Course 3 Semi-
digital systems. conductor Devices; Course 4 Electronic Circuits; Advanced
Digital Technique Course; Microprocessor Course. Available
with experimental back-up and optional cassettes, Heath-
Schlumberger, London.
RELATED ARTICLES
Horowitz, P. and Hill, W. (1989) The Art of Electronics, Cam-
bridge University Press.
Article 36, Systems in the Frequency Domain, Volume 1;
Article 73, Overview of Electrical and Electronic Tech- Huelsman, L.P. (1977) Active R.C. Filters, Wiley, New York.
nique, Volume 2; Article 74, Basic Electronic Compo- Khalil, H.K. (2001) Nonlinear Systems, Prentice Hall, New Jersey.
nents, Volume 2; Article 122, Operational Amplifiers, Kostopoulos, G.K. (1975) Digital Engineering, Wiley, New York.
Volume 2. Lancaster, D. (1976) TTL Cookbook, Howard W. Sams, Indi-
anapolis, IN.
Lancaster, D. (1977) CMOS Cookbook, Howard W. Sams, Indi-
REFERENCES anapolis, IN.
Blakeslee, T.R. (1975) Digital Design with Standard MSI and LSI, Mano, M.M. (2001) Digital Design, Prentice Hall.
Wiley, New York. Melen, R. and Garland, H. (1978) Understanding IC Operational
Bouwens, A.J. (1974) Digital Instruments Course (several parts Amplifiers, H. W. Sams, Indianapolis, IN.
issued), N. V. Philips Gloeilampenfabrieken, Eindhoven. Middleton, R. (1977) Digital Equipment Servicing Guide, H. W.
Brophy, J.J. (1977) Basic Electronics for Scientists, McGraw-Hill- Sams, Indianapolis, IN.
Kogakusha, Tokyo. Millman, J. (1972) Integrated Circuits: Analogue and Digital and
Burr-Brown (1964) Handbook of Operational Amplifier Applica- Systems, McGraw-Hill, New York.
tions, Burr-Brown Research Corporation, Tucson, AZ. Millman, J. and Taub, H. (1965) Pulse Digital and Switching
Burr-Brown (1966) Handbook of Operational Amplifier Active Waveforms, McGraw-Hill, New York.
R.C. Networks, Burr-Brown Research Corporation, Tucson, AZ. Namgoster, M. (1977) Digital Equipment Trouble Shooting,
Clayton, G.B. (1971) Operational Amplifiers, Butterworths, Reston, Reston, VA.
Sevenoaks. NS Corporation (1976) Linear Applications Handbook, (Vol. I,
Connelly, J.A. (1975) Analog Integrated Circuits, Wiley, New 1973; Vol. 2, 1976), National Semiconductor Corporation, Santa
York. Clara, CA.
Diefenderfer, A.J. (1979) Principles of Electronic Instrumenta- Sydenham, P.H. (1979) Electronics – It’s Easy, (3 Vols), Modern
tion, Saunders, Philadelphia, PA. Magazines, Sydney.
Faulkenberry, L.M. (1977) An Introduction to Operational Ampli- Wakerly, J.F. (1976) Logic Design Projects Using Standard Inte-
fiers, Wiley, New York. grated Circuits, Wiley, New York.
76: Electronic Systems Design
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
was also necessary to develop the display for the numerical
1 Approach to Electronic System Design 483 output as a distinct project facet.
2 Counters 484 Today, the same detail of task no longer exists for the
designer of measurement systems: complete counters like
3 Frequency Counters and Timers 485
that mentioned above are now offered for sale with integral
4 Registers 485
display. The most basic way to assemble the system now
5 Digital-to-analog Converters 485 is to make use of standard integrated circuits driving an
6 Multimeters 486 integrated digital display. The same unit is also available
7 Microprocessor 486 as a single LSI (Large Scale Integration) unit costing the
8 Interconnecting Stages and Signal equivalent of less than an hour of a technician’s time.
Transmission 487 Systems design is based on first having a sound con-
9 Grounding and Shielding 488 ceptual understanding of the task, this being committed to
Related Articles 489 paper in the form of block diagrams that lay down the signal
flows and show the characteristics needed. These are then
References 489
refined, until the first apparently workable system is ready
to assemble. A prototype is then put together using tempo-
rary assembly, with soldered or plugged joints. A number
1 APPROACH TO ELECTRONIC SYSTEM of proprietary assembly methods are available; many avoid
DESIGN the need to solder. When the prototype is thought to be
satisfactory (from actual tests) the circuit is made in the
Originally, electronic instrumentation systems were devised printed wiring, printed circuit, or hybrid integrated form.
using the basic active and passive elements, plus supple- Large organizations will make use of advanced design sup-
mentary devices, to realize whatever basic functional units port tool suites that can be used to create a design in the
were needed. Almost all early electronic design was carried computer, which will be almost perfect when actually man-
out using discrete components. Design then, therefore, first ufactured. While such tools do exist, they take time to install
needed to form the functional building blocks, which were and learn. For this reason, many electronic systems cre-
then interconnected to obtain the overall required perfor- ated for measuring systems are built from components, the
mance. design being tested as it is developed.
As an example, to build a fast reversible digital counter in It is generally the case that the first system design will
the 1960s, it was necessary to first develop a fast flip-flop not be entirely suitable. This occurs because design of
using more active elements than were commonly used at any technological system relies on the designer making the
the time for slower speed counters. A logical arrangement correct assumptions about a system that he or she is not yet
had then to be devised that would count as fast as possible, familiar enough to make. The actual act of assembly and
providing the output digital code needed for the display. It testing the prototype considerably improves understanding
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
484 Measuring System Design
of the need. The degree of success in arriving at a suitable 2 COUNTERS
initial design relates to the complexity of the problem
and the designer’s experience. Where possible, proprietary Many modules are now available from which a system can
modules are usually used to reduce the need for design be assembled. It is usually not necessary to build systems
knowledge. This eliminates many design decisions. If the using the basic linear and digital integrated circuits with,
modules are well tried and proven, then the whole is more perhaps, a few discrete components. This section considers
likely to be reliable. Digital systems tend to be easier to a few of these larger subsystems; the number constantly
develop than analog ones. grows as designs and demand appear, enabling integration
Interaction between stages is probably the greatest source to be used at a commercially viable level. A good set of
of prototype development difficulty. Stages may exces- maker’s catalogs and Web links need to be developed to
sively load the preceding one or transient signals can influ- support chip availability information.
ence other parts of the system. Race conditions, in logic A cascaded chain of flip-flops will divide a time sequen-
systems, can also occur if some part of the system acts tial string of digital logic levels according to the math of
faster than others where timing is an important parameter. binary division. By suitable arrangement, the chain can
Many of these problems cannot be dealt with by rigorous be used to count up, or down, or be reversible accord-
mathematical methods, unless a model-based tool is being ing to an external direction control signal. If the signal
used. The designer’s ingenuity is called upon to ensure that passes from stage to stage in sequential order, the pulse
as many of these likely problems as possible are circum- ripples through taking a considerable time (up to microsec-
vented, or allowed for in some way. onds) to pass through many stages. This form of counter
When a suitable prototype assembly has been developed, is called asynchronous for state changes happening in any
it must then be tested for operation in a real situation. Tem- stage of the chain are not necessarily made all at the same
perature, vibration, mechanical strain, humidity, and many time.
more influence quantities may prevent proper operation. By suitable interconnection of the stages, the ripple-
Because layout can be important, it will first be necessary, through time can be greatly reduced by causing an input
in some cases, to construct an assembly that is very close pulse transition to set each of the stages simultaneously.
to the actual construction method to be used. The prototype These are called synchronous counters. They are more com-
should never be destroyed or changed as the next version plex to interconnect, but, as they are rarely built from
is being made. It will often be found that the second, sup- discrete components, this is not a problem for systems
posedly superior, system exhibits faults that were not seen designers; it has been done in the manufactured IC. Coun-
before. A check of the first prototype may well show that ters can easily count at rates of 100 MHz and higher rates
the faults were also there in that model, thereby showing in the low GHz region, the reversible kinds reversing in as
that the design is at fault, not the change in layout. little as nanoseconds.
The design of integrated circuits is usually more rigor- Binary counters count in a binary sequence. To change
ously achieved but even there, where it is possible to devote the contained binary code numerical value into the often
more effort to a specific problem, similar methods are often more useful decade form, they must be decoded by a
utilized to produce a layout for integration. Prototyping of suitable conversion logic network.
this level only occurs in specialized groups. It is sometimes more convenient not to count in pure
A very common error is to regard a single, apparently binary, but to use what is called a binary coded deci-
satisfactory, assembly as being necessarily typical of a mal (BCD) system instead. In this system, four flip-flops
batch production run. Too often, production begins on are used to form a counting chain that passes through
the strength of inadequate testing and tolerance-spread only 10 and not the binary’s 16 states. Feedback between
checking. It is wise to expand from 1 to 10, to say, the counting stages is used to achieve this. The actual
100 systems, when very large numbers of an identical coding used within the four-element system decides the
unit are needed. Specialized functions such as reliability form of decoding network needed to drive the chosen
and thermal behavior are often overlooked in low volume decade number display. The individual BCD units are then
designs, with subsequent need for redesign. cascaded using simple connections to obtain the num-
This short article can only introduce electronic systems. ber of decades display needed. Many of these design
In practice, the electronic part of a measurement system variables need not be known by the use of ready-made
is usually but a portion of the whole, each module being LSI counter/display modules. In many instances, their
considered in turn, and then the whole, on other grounds price is so low as to make any thought of new devel-
such as reliability, production needs, and sales and service opment quite unrealistic. They are very well tried and
requirements. proven.
Electronic Systems Design 485
x tal 1 C1
4 Mag 33pF
R1 R2
2 3 2 3 3 3 3
1.5 k 1.5 k 9 12
14
1
IC1/1 3 4
IC1/2
6 5
IC2/1 IC2/2 IC3 11 1 IC4 12 1 IC5 12 1 IC6 12 1 IC7 12 10
7473 1 7473 7490 7490 7490 7490 7490 10 Hz
2 7400 C2 5 7400
047 mF 12 1 11 14 11 14 11 14 SW1/1 9
11 14
C9 6 2
100 rF 9 10 100 Hz
IC1/4 8
7400
13 12
8 IC1/3
7400
6 2 6 11
2 3 2 3 2 3 2 3
+ +5 V 9
R3
IC8/1 1 IC8/2 5 IC9 14 IC11 14 IC13 14 IC15 14
Q1 100 Ω
4.7mF
4.7mF
e 13 7473 8 7473 7490 1 7490 1 7490 1 7490 1
4.7mF
BC178
b
e + + 11 8 9 12 11 8 9 12
b c 11 8 9 12 11 8 9 12
C6 + C8
c Q2
BC178 C7
3 2 1 8 3 2 1 8 3 2 1 8 3 2 1 8
ZD1
6−12 V Battery
C3
4.3 V IC10 IC12 IC14 IC16
0047mF
R4 OV HP5082 HP5082 HP5082 HP5082
R5 1κ 7302 7302 7302 7302
470 Ω 1
c 4 5 4 5 4 5 4 5
b Q3
2N 3055
330 Ω
e MJE 3055
R6
12
SW 1/2
D1
7 EM401 11
Inhibit line
−
SWS
Figure 1. Schematic block diagram showing, as an example, how basic IC modules are used to form a general-purpose digital stopwatch.
3 FREQUENCY COUNTERS AND TIMERS end (serial input) or to each simultaneously (parallel
input); each flip-flop then records a respective digit of the
Addition of a clock source (a square-wave generator of number. Action of the clock-pulse input causes the num-
known frequency) and a suitable turn on, and off, gate ber to remain intact, yet be stepped sideways in either
system forms a method for measuring the frequency of direction. This is used as part of the hardware of sev-
a signal. The gate is, in one version, operated from the eral mathematical operations using digital binary num-
clock allowing the counter to accumulate one count per bers.
cycle of the signal being measured, for a known period
of time. This determines the frequency of the signal; the
count is displayed as frequency. If a count from the clock is
measured for the time during which the gate is held open by
the cycle time of the signal, then a period will be measured. 5 DIGITAL-TO-ANALOG CONVERTERS
Figure 1 shows the IC modules used to build such a unit.
Similar principles are used to count objects as part of,
A digital number is often needed in an analog equiva-
say, a batch control system.
lent form, for example, as might arise when an electronic
Timing is also possible using linear modules in which
digital calculation has been made and the resultant must
an output is held, say, high for a preset duration. In
operate an analog signal device. Several methods of con-
linear methods, mathematical integration is used to produce
version exist for this direction of conversion and for the
a time interval. Linear methods are not capable of the
high timing precision afforded by properly designed digital reverse, where analog signals are converted into digital
counter/timers. forms. These functions are rarely made any more from dis-
crete components; the cost of ready-to-use modules being
low enough to encourage selection for direct use in the
4 REGISTERS larger system. Although designers do not need to develop
these converters, it is very important that their internal oper-
A register is a form of cascaded flip-flops. A digital ation is understood as the performance can vary greatly with
binary number is passed into the digit stages from one application.
486 Measuring System Design
6 MULTIMETERS a basic digital computer system. It can be set up inter-
nally, through the use of external software programming,
As an example of an electronic instrument system, consider to perform an incredibly wide range of circuit functions.
the apparently simple high quality multimeter. This instru- It operates internally as a digital computing, but the inter-
ment has undergone extensive change in internal design faces used to connect it to the chosen application allow it
philosophy in recent years and exemplifies the general trend to be connected to analog devices when needed. Its low
in electronic systems operations. cost enables it to be used as a most basic single com-
A multimeter has the purpose of allowing the user ponent that is set by the user to perform specific tasks.
to measure voltages, currents, or resistances of a system It is set by software programming rather than by hard
and its components. Originally, these were always made wiring.
from an electromagnetic meter indicating movement to In reality, a considerable amount of peripheral equipment
which appropriate multipliers (series resistors – for volt- may be needed and the cost of this can far exceed the cost
age ranges), shunts (shunt resistors – for current ranges), of the microprocessor itself (less than a few man-minutes
and operational battery, and series resistor (for resistance) of time). Even so, the general-purpose nature of the module
were switched into circuit as needed by the operator. A makes it a first choice, when seeking to design a complex
multimeter of better quality would normally have overload control system. Rather than make connections with wires
protection to safeguard the meter from accidental damage the user writes a program. Once set, the microprocessor-
by inappropriate connection. based system can be reset reasonably easily.
The entire unit used analog methods and could also Use of microprocessors needs little understanding of the
incorporate a linear amplifier in the circuit to raise the concepts, basic modules, and building blocks of electronics.
terminal resistance to 1 M or more in order to remove Microprocessors have brought about a marked change in
loading effects in use. the design philosophy of electronic systems of the future.
The alternative design, now almost always used, contains Some predict that the cost of the microprocessor will fall so
ICs and some discrete components. It operates internally low that it will find use in numerous low-cost applications.
almost entirely using digital, rather than analog, signals. There is a running shoe with one in it. A modern automobile
The user must still select which variable is to be mea- has dozens in its control and instrumentation.
sured, but need not select the range; systems are autorang- With microprocessors, little electronic knowledge is
ing. The input signal is first converted into a digital equiv- needed to create the system: programming skills have been
alent using an analog-to-digital converter stage. The signal largely exchanged for network and system design expertise.
is then used with counters and digital display to show the Despite this, peripherals still need to be designed and
value in digital format. connected and in many situations, microprocessors and
Another example is the use of digital methods to pro- other marketed general-purpose computational designs are
duce a very accurate sample-and-hold module, over a long not able to match the performance possible from special-
period of time. In these, the analog storage capacitor that purpose designs.
stores the voltage level in the linear analog system has The programmable logic controller (PLC) can be placed
internal leakage that degrades the voltage level with time. in the same class as microprocessors. These are systems in
If the capacitor is used to sample the level and the value which the circuit wiring to provide the actual system task
so obtained is immediately converted to a digital equiva- needed (basically switching networks) is again done by a
lent form, it can be stored for an indefinite time period. form of direct programming into the computer-based sys-
Recovery uses a digital-to-analog converter. tem. Changes to the system operation are easily carried out
The relative cheapness, increased reliability, and often without the need to enter the electronic level of operation,
enhanced accuracy of digital alternatives easily outweighs or to be an expert at programming a computer.
the fact that internally their design is far more complex, and The operational power that can be provided by one, or
uses many more elements, than does the analog alternative. more, microprocessors has, as did valves in their more
primitive way, allowed the designer still greater freedom
of action in implementing conceptual schemes.
7 MICROPROCESSOR It is now normal to build electronic systems that are
self-checking, that can diagnose themselves, and that can
The concept of a totally general-purpose basic electronic make reconnections to overcome failures – these have
digital building block was taken a step further with the been coined the smart systems. A useful level of artifi-
introduction of the microprocessor. This is an LSI cir- cial intelligence has already been achieved in electronic
cuit, Figure 2, that contains the architectural features of systems. The low cost, small space, and power needs
Electronic Systems Design 487
Figure 2. Internal arrangement of a typical microprocessor IC.
of the microprocessor enable very extensive strategies to are considered as additional stages, then they are likely
be implemented on reasonable budgets of finance and to receive the attention that they must be afforded. Even
time. when modules are wired in close proximity, they must be
An original suggestion was that as the microprocessor connected appropriately.
was standard methodology, texts written on the micropro- It is necessary to consider the connections carefully,
cessor would be general purpose. To some degree this is for the link itself may act as a frequency filter, alter the
true, but usually there is the need to access information on amplitude of signals, pick up unwanted signals, introduce
the one being used in a design. Too many variations have phase shifts, or be mechanically unreliable.
emerged from this very powerful technology. Within closely assembled circuitry, simple printed and
insulated wiring will generally suffice unless high frequen-
cies are involved above around 1 MHz. If the distance
8 INTERCONNECTING STAGES AND between units exceeds a few meters, then it becomes nec-
SIGNAL TRANSMISSION essary to select a more suitable form of signal transfer
method. The open-wire, as seen on old telephone poles,
Whereas the modern electronic system designer has less to is usable to around 10 MHz, but beyond that the higher
consider now about the internal design of building blocks frequency signals need other transmission methods. The
and modules than in the past, it is still as important as ever open wires are very lossy and not particularly secure from
that they be interconnected properly. If the connecting links damage.
488 Measuring System Design
An alternative is the coaxial cable. This is formed from Analog signals are more prone to this, for their informa-
conductors suitably placed with a separating dielectric. tion is conveyed as an amplitude. Coding and the use of
They are made in circular, flat, and strip sections each digital signals are often better alternatives to use.
having its special application. They can convey signals Several connection techniques can be used to reduce stray
to around 5000 MHz. Multicore insulated cables could pickup in wiring. The principles apply to both long and to
be considered to be of this type, but they are not usu- even very short leads in low-level detection systems.
ally designed for high-frequency use, having excessive A first possibility is to transfer the information over the
losses. link at some frequency away from that of the expected
Above the usable range of signal frequency suited to noise. It is then theoretically possible to filter out the signal
coaxial cables, it is necessary to make use of waveguides at the receiver. This method can be satisfactory, but all
and open-space electromagnetic radio links. Laser-based filters allow some level of signal through, and the price
optical fiber links are also capable of conveying these high paid for sharp highly discriminating filters may not be
bandwidths and are used prolifically, see Wolf (1979) for acceptable; for instance, they tend to ring to rapid transient
details. They play a major role today as they are not prone changes of signal level.
to the same stray signal, pickup, effects as are electronic A better approach is to reduce the noise level at its source.
transmission systems. If that is not possible, then the wiring can be arranged such
Whichever kind of electrical communication line is
that two parts of it pick up similar noise signals, which are
used, certain basic generalizations can be made. First,
out of phase by 180◦ at the detection stage. They then cancel
a short line will possess small values of self, shunting,
to leave only the unbalanced contribution. This principle
capacitance, and resistive losses. As the line length is
is known as common-mode rejection. A twisted cable pair
increased, the capacitive effect becomes more dominant,
does this, as each wire has an equal signal induced in
the line then acts as a low-pass filter: this gives atten-
it by common radiation coming from the noise signal
uation and phase shift to the signal being sent over the
source; the two being induced in opposite directions in the
link.
The coaxial cable exhibits an interesting feature for detector input circuit. To obtain the best from common-
reasonably high frequencies of use. There the impedance mode rejection, connectors, and the input stage of the
that can be seen by looking into the end of an ade- active circuitry must also be differentially arranged in
quately lossless line of any length can be regarded as being this manner. Common-mode rejection should not be used
of fixed pure resistance. This is called the characteristic to remove noise that can be reduced at source, for the
impedance, Z0 , of the line. Common values made are 600, common-mode noise signals can require amplifiers with
100, and 75 . Thus, the actual length of the line matters unduly large dynamic ranges to avoid serious distortion
little. occurring.
When high frequencies need to be sent over trans- Shielding is also a method used to reduce signal levels
mission links, a serious effect occurs if the terminating induced into connecting leads. Leads form antennae in
impedance is not purely resistive and equal to the charac- which electromagnetic fields will induce voltages. Thus,
teristic impedance of the line used to make the connection. the shorter the leads, the lower their impedance. The better
If mismatched, reflections of signal energy occur, and time- they are shielded, the smaller will be the induced currents.
varying signals are generated in the links that feed on and Shields, and their necessary earthing connection, should
back, at each end of the link. At lower frequencies, the mis- not by themselves form a significant resistance closed
matched line acts to alter the signal magnitude and phase; electrical circuit, for the earth loop then produced will
reflections produce distortion and loss of power transfer. produce circulating currents that also induce unwanted
It is, therefore, always important to properly match high signals elsewhere.
frequency–signal terminations. This subject is covered in Radio frequency energy – electromagnetic interference –
Johnson (1950) and in Sinnema (1979). can also be bothersome. Often the problems are overcome
by preventing it escaping from the generating source by
careful scientific design of the enclosures. Alternatively,
9 GROUNDING AND SHIELDING the detecting apparatus is screened.
Several specific texts are available that provide detail
Another aspect of stage coupling is the need to preserve on long line–transmission systems, for example, Johnson
the signal-to-noise ratio (SNR). Ideally, a coupling should (1950). Low noise systems, and how to couple them, are
not degrade the system, but in practice there will always be the subject of Morrison (1977), Motchenbacher (1973), and
some reduction. Ott (1976).
Electronic Systems Design 489
RELATED ARTICLES Morrison, R. (1977) Grounding and Shielding in Instrumentation,
Wiley, New York.
Article 73, Overview of Electrical and Electronic Tech- Motchenbacher, C.D. (1973) Low Noise Electronic Design, Wiley,
nique, Volume 2; Article 74, Basic Electronic Compo- New York.
nents, Volume 2; Article 75, Electronic System Building Ott, H.W. (1976) Noise Reduction Techniques in Electronic Sys-
Blocks, Volume 2; Article 78, Embedded Systems, Vol- tems, Wiley, New York.
ume 2; Article 159, Designing and Building Software for Sinnema, W. (1979) Electronic Transmission Technology, Prentice
Measuring Systems, Volume 3. Hall, London.
Wolf, H.F. (1979) Handbook of Fibre Optics, Garland STPM
Press, New York.
REFERENCES
Johnson, W.C. (1950) Transmission Lines and Networks,
McGraw-Hill, New York.
77: Limits of Detection in Electronic Systems
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
also called Johnson noise after its prime investigator, has
1 Sources of Fundamental Physical Limits on been quantified as
Detection 490 υ 2 = 4kTRB (1)
2 Johnson Noise 490
3 Shot Noise 490 where υ 2 is the mean-square voltage generated, k is the
4 Flicker Noise 491 Boltzmann constant, T is the absolute temperature of the
5 Sources of Noise Resulting from Imperfect resistance, R is the resistance in question (in ohms), and B
Design 491 is the frequency bandwidth of the signal in the resistor.
Related Articles 492 This can also be expressed in terms of noise current
and noise power. Unlike the Brownian motion noise of
References 492
mechanical systems, which rarely is observed, this noise
is relatively easily detected by electronic systems and is
always a serious factor to consider when designing low-
1 SOURCES OF FUNDAMENTAL level (below microvolt) detection stages. Clearly, from
the equation, reduction of bandwidth, temperature of the
PHYSICAL LIMITS ON DETECTION resistor, and magnitude of the resistor reduces the Johnson-
noise level.
As in all of the energy regimes, electric energy systems also The frequency spectrum of this noise, in its purest form,
possess certain physical processes that prevent sensitivity is uniform, there being as much energy in a given band-
from being increased to infinity. No matter how well width at any place in the frequency spectrum. If this con-
contrived and constructed the apparatus is, the designer dition applies, the noise is said to be white. Gray and pink
must face the fact that these fundamental effects cannot noises are of this general form but they possess different
be eliminated. They provide an ultimate limit to detection frequency spectra. Gray, for instance, reduces in amplitude
for a given methodology. Three fundamental noise sources, in a linear manner as the frequency rises. Such noise types
and other controllable noise sources, are shown pictorially are close approximations to different noise situations that
in Figure 1. arise in practice and, thus, can be used as test noise sources.
2 JOHNSON NOISE 3 SHOT NOISE
Within the resistive components of any electric component, Wherever electric current flows, the corpuscular nature of
thermal energy causes the electrons to vibrate with increas- the electrons gives rise to discrete signal energy packets
ing activity as the thermal energy increases. It is related to called short noise. If there are enough of them, their sum
mechanical Brownian motion within matter. Thermal noise, averages to a steady current and the effect is not generally
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Limits of Detection in Electronic Systems 491
Fundamental source Despite considerable research, it is still not possible to
Flicker Controllable source entirely quantify this noise form on the generalized basis.
Single frequency The physical reason for this in valves and in transistors
Narrowband Gray is not proven. As a rule of thumb, an electronic system’s
Amplitude
of noise
flicker noise begins to become significant compared with
White (Johnson and shot)
sources the other fundamental limits at around 400 Hz, increasing
below that frequency to be at infinity at absolute zero
Pink
frequency. This is one reason for using carrier systems to
provide good signal-to-noise ratios in detection equipment:
the signal to be detected is generally modulated at well
Frequency
above 400 Hz, where this noise effect is lower than the
Figure 1. Frequency spectra of fundamental and other electronic other noise types.
noise sources. (Amplitudes of each type not shown to any strict Flicker noise has a direct counterpart in mechanical
scale.) and thermal systems, for the greater the integration time
that is used in an attempt to remove long-term, slowly
observable. If, however, each electron becomes an event in changing, drift effects, the lesser the rate of improvement
the detection system, as it does when very low-level signals obtained.
are concerned, then this discrete randomness makes its The above-mentioned noise sources are the three domi-
presence detectable. This was first investigated for electron nant sources that cannot be reduced beyond certain limits.
discharge in the former thermionic valves by using the The use of bandwidth-reduction techniques reduces their
assumption that the space charge was not filling the space level but at the usual expense of increase in response time.
available. The theoretical expression derived to explain Cooling is often employed to reduce the operational tem-
the effect quantitatively was thus an estimate; in most perature, including use of cryogenic stages. Electronic noise
applications, space-charge conditions alter the magnitude sources are well described in the literature (see Bennet,
of the effect. The usual basic expression quoted is 1960; Connor, 1976; Delaney, 1969; Usher, 1974; Van
der Ziel, 1954; Whalen, 1971; Wainstein and Zubakov
ī 2 = 2eIB (2) 1962).
where ī 2 is the mean-square current generated, e is the
electron charge, I is the average current passing, and B is 5 SOURCES OF NOISE RESULTING
the frequency bandwidth of the signal passing.
FROM IMPERFECT DESIGN
However, valves, transistors, and other active electronic
devices, each differ a little from this generalization. Suffice
to say that this effect is also detectable by quite simple Assuming that fundamental limits do not restrict the design,
electronic apparatus and, therefore, must also be allowed then the system must also be assembled, connected, and
for in the detection of stage design. Reduction in signal shielded such that additional noise pickup is kept as small
bandwidth and the through-current level reduces the level as is required. The common unit used to describe the
of this type of noise. Choice of detection device also has degradation that always occurs from connection and stage
an important bearing on the magnitude. This form of noise performance is the noise figure (NF). It is given by
is also spectrally white if it follows the given equation. SNRin
In both these cases, the formulations express the levels NF = 10 log dB (3)
SNRout
of noise signals generated within a component. The level
that is received by the following stage depends upon the A perfect stage would add 0 dB, that is, nothing, but in
matching conditions between the two stages. practice, around 1.5 to 3 dB is the lowest addition obtain-
able. Shielding, layout, wiring, device selection, connec-
tion modes, matching, and other factors must be con-
4 FLICKER NOISE sidered as possible means to reduce the NF of a stage.
The order that stages with differing NF are connected
A third important fundamental noise source of electronic in cascade is also an important factor. As a guiding
systems is called flicker noise. Since it varies approximately rule, the stage with lowest NF should be used closest
as the inverse of the frequency, it is often referred to as 1/f to the detection stage. Often, an initial stage is inserted
or hyperbolic noise. that provides little useful gain but has a low NF that
492 Measuring System Design
assists the overall NF to be minimized. As practical mea- REFERENCES
surement systems usually involve more than one energy
regime, the noise sources of each regime must be consi- Bennet, W.R. (1960) Electrical Noise, McGraw-Hill, New York.
dered. Connor, F.R. (1976) Noise, Edward Arnold, London.
Further information on noise in electronic signal systems Delaney, C.F.G. (1969) Electronics for the Physicist, Penguin
is provided in the Section Noise and Interference and Books, Harmondsworth.
Signal Recovery in the Presence of Noise. Usher, M.J. (1974) Noise and bandwidth. Journal of Physics
E: Scientific Instruments, 7, 957–61; (Reprinted in 1982,
Instrument Science and Technology, Vol. 1, (ed. B.E. Jones),
Adam Hilger, Bristol, pp. 110–115).
RELATED ARTICLES Van der Ziel, A. (1954) Noise, Prentice Hall, Englewood Cliffs,
NJ.
Article 56, Transfer Characteristics of Instrument Wainstein, L.A. and Zubakov, V.D. (1962) Extraction of Signals
Stages, Volume 1; Article 57, Static Considerations from Noise, Prentice Hall, Englewood Cliffs, NJ.
of General Instrumentation, Volume 1; Part 12, Ele- Whalen, A.D. (1971) Detection Signals in Noise, Academic Press,
ments: F – Signals and Noise, Volume 3. London.
78: Embedded Systems
Timothy Wilmshurst
University of Derby, Derby, UK
system, autonomous, or human or network interactive, oper-
1 Definition of An Embedded System 493 ating on diverse physical variables and in diverse envi-
2 Characteristics of Embedded Systems 493 ronments, and sold into a competitive and cost-conscious
market.
3 Design of Embedded Systems 495
4 Programming the Embedded System 496
5 Testing of Prototype Embedded Systems 497 2 CHARACTERISTICS OF EMBEDDED
References 497 SYSTEMS
The scale, complexity, and cost of embedded systems vary
1 DEFINITION OF AN EMBEDDED enormously. A typical system has some or all of the
SYSTEM elements shown in Figure 1.
At the heart of the system lies an embedded microcom-
The term embedded system implies one computer or com- puter. Most commonly this is a microcontroller, although
puters embedded within a system or product, which form larger-scale computers (e.g. embedded personal computers)
an important part of that system, generally undertaking the are also widely applied – see Article 158, Microcontroller
system control. Related functions, for example, data pro- Systems, Volume 3.
cessing, can also readily be undertaken. The embedded The embedded computer is itself made up of its hard-
computer may be hidden from view, and the user may not ware and software. Feeding into the embedded com-
even be aware of its existence. puter are input variables, taken from sensors and passing
Many definitions for embedded systems exist. The con- through input signal conditioning and a data conversion
sulting company Dedicated Systems Experts (http://www. stage.
realtime-info.be) offer the following definition: An embed- The computer may be networked with other systems, and
ded system is a computer system that is enclosed in another it may have a user interface, although neither of these is
system and makes an essential part of it. essential to the concept.
Wilmshurst (2001) uses the following alternative defi- The system is likely to generate output signals that are
nition, which explicitly evokes the control nature of most used to control system variables, for example, by driving
embedded systems: An embedded system is a system whose actuators through a power electronic interface.
principal function is not computational, but which is con- Figure 1 indicates the breadth of expertise that is required
trolled by a computer embedded within it. in the design and commissioning of an embedded sys-
Noting the characteristics of the embedded system, dis- tem, including sensors, data acquisition, microcomputer
cussed below, Wilmshurst also uses the following more hardware and software design, data communications and
descriptive definition: An embedded system is a microcon- networking, human interface, control, and actuators and
troller-based, software-driven, reliable, real-time control their drives.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
494 Measuring System Design
(User interface)
(display, keypad etc.)
Embedded computer
conditioning
conversion
Output
Signal
drive
Data
Input Output
variables Software
variable
Hardware
(Link to other systems)
Figure 1. The essence of the embedded system.
User
interface
Wind
speed
Multiplexer
Temperature
Micro- Satellite
ADC
computer link
Rainfall
Light
intensity Storage
memory
Power supply
Figure 2. An example of an embedded system, a remote four-channel automatic weather station.
An example of an embedded system, a remote four- dependent outputs. This is generally known as operating in
channel Automatic Weather Station is shown in Figure 2. real time.
Here four inputs are sampled periodically (not necessarily
with the same period), and the resultant data processed and
stored. It is also transferred via a telemetry link to a satellite.
2.2 Efficiency of hardware/software utilization
As it is designed for operation in the field, power supply is
a critical feature of the system.
Further important characteristics of the embedded system Embedded systems are often sold in markets that are
are as follows. extremely cost sensitive. Alternatively, or as well, many
such systems requirements are very space and/or power
conscious. For these reasons, the embedded system is
normally designed with restrictive hardware and soft-
2.1 Timeliness ware constraints.
The result of this, for example, is that the CPU may
Most embedded systems operate in an environment that is be only 8 or 16 bit, with limited computing power, and
strongly time dependent. They must respond in an appro- a restricted memory size may impose limits on program
priate timescale to external events, or must generate time length or data storage capability.
Embedded Systems 495
2.3 Reliability • Clarification of Need, and then moves through
• Conceptual Design, to
Many embedded systems now operate in situations having • Embodiment Design, and finally to
high demand on reliability. This includes, for example, • Detail Design.
many medical and motor vehicle applications. Appropriate
techniques for hardware and software reliability are there-
fore applied. An adaptation of this process to the embedded system
environment (Wilmshurst, 2001) is shown in Figure 3. This
is a form of system life cycle – see Article 63, Outline
2.4 Constraints of the operating environment of Systems Thinking, Volume 2 and other articles of the
subsection on ‘Engineering a measuring system’.
The operating environment of many embedded systems This model provides a sequence of design phases. The
places further demands on its design. Many are battery pow- principal characteristics of the system are identified in clar-
ered, and so must be designed from a low power viewpoint. ification of task, or requirements generation, that is then
Some operate in environments of hostile temperature, defined in the specification – see also Article 69, Require-
humidity, or vibration, for example, a motor vehicle engine ments Allocation, Volume 2. From this, the principle hard-
management system. ware and software functions emerge. These then continue
These characteristics, taken in summary, form the final as two interlinked design strands. These strands should pro-
definition given in Section 1. ceed in parallel, a process known as codesign.
Important aspects of the design process, whether or not
3 DESIGN OF EMBEDDED SYSTEMS this model is followed, are outlined below. Note however
that software program design issues are mainly covered
Pahl and Beitz (1995) have identified a design methodology in Section 4 below – see also Article 159, Designing and
for engineering products, which starts with Building Software for Measuring Systems, Volume 3 and
Parallel ports
Hardwired logic ADC, DAC
Microcomputer serial ports .... .
timers .... .
FPGA
.... .
PLC ....
. Choose
..... DSP micro- Detail
SoC Peripheral Hardware
..... Hardware functions computer(s) circuit strand
Solution layout design
Principal
hardware technologies
functions
Recognition Principal Memory Processing Prototype
Specification Word size
of need functions needs speed and test
Principal
software Issues of
functions data and
Issues of Write
computation
.... . timeliness Language Software
program
.... . .... . Program
Tasks code strand
Set targets, .... . structure
deadlines priorities C
and budget interrupts C + + .... .
latencies Java .... .
Assembler
Review targets, deadlines and budget
Clarification Conceptual Embodiment Detail
of Task Design Design Design
Figure 3. A design model for embedded systems. (Reproduced by permission of Institution of Electrical Engineers, with modifications.)
496 Measuring System Design
other articles of the subsection on ‘Software aspects of • microcontroller designed into a custom piece of hard-
measuring systems design’. ware.
Selection will be initially on device performance and
cost. Other aspects, however, must be considered. These
3.1 Solution technology
include cost of development tools (which can be significant
and take time to learn), and the level and quality of support
It is essential for the designer to be aware of the cur-
available for the chosen device.
rent technologies available for system implementation. The
There are a huge range of microcontroller suppliers,
conventional approach, of designing a microprocessor or
including Atmel (http://www.atmel.com), Hitachi (http:
microcontroller into a hardware system, is to a large extent
//www.hitachi.com/products/electronic/semiconductorcomp
being replaced by the opportunities of ASIC (Applica-
onent/index.html), Microchip (http://www.microchip.
tion Specific Integrated Circuit), Field Programmable Gate
com/), Motorola (http://e-www.motorola.com/webapp/sps/
Array (FPGA), and System on Chip (SoC) technologies.
site/homepage.jsp?nodeId = 03M0ym4t3ZG), and Philips
Thus, a microprocessor is not necessarily purchased as a
Semiconductors (http://www.philips.semiconductors.com
piece of silicon, but as a piece of Intellectual Property,
/products/standard/microcontrollers/).
(IP) which can be downloaded onto an FPGA, along with
other devices.
The SoC strategy takes this further, with CPU, mem- 4 PROGRAMMING THE EMBEDDED
ory, data buses, and an array of other system elements
available, in user-configurable form, on the chip. Xilinx SYSTEM
(http://www.xilinx.com/) and Altera (http://www.altera.
The usual requirement in an embedded system is for a
com/) are important suppliers in this field.
program with compact and reliable code, which executes
Other distinctions, such as that between the conventional
efficiently, and which can interact in a direct way with the
microprocessor and the DSP (digital signal processing)
hardware resources.
device (see Article 146, Principles of DSP Hardware
Programming normally takes place using an Integrated
Design, Volume 3), are also becoming blurred, with micro-
Development Environment (IDE). This runs on the host
processors and microcontrollers moving to take on DSP
computer, and develops programs for the target system (the
capability, and DSP devices taking their place in embed-
embedded system under development).
ded systems.
As a complete alternative, Programmable Logic Con-
trollers (PLCs) nowadays carry out sophisticated measure-
4.1 Programming language
ment, control, and network capability, and can themselves
form the basis of embedded control.
The requirements mentioned above make High Level Lan-
guages applied elsewhere less attractive.
One possibility is to use the native assembler of the
3.2 Hardware layout
processor. This leads to the most efficient code possi-
ble, and allows excellent interaction with the hardware.
This process identifies the principal blocks of the hardware
However, programming in Assembler is error prone, and
system, and their interconnection. It will include such
time consuming.
fundamental decisions as whether there are one or more
An excellent compromise is the language C. This leads
microcomputer elements, how data is transferred internally,
to compact and well-structured code, which can inter-
what input/output techniques are used (and hence what
act with the hardware, and which can be developed rea-
peripherals are required), and so on.
sonably rapidly. Techniques adapting C to the embed-
ded environment, including in high reliability applications,
are widely documented (e.g. van Sickle (2001), Hatton
3.3 Choice of microcomputer(s)
(1994), MISRA (1998). C compilers are available for all
microcontrollers, some with special extensions or restric-
Having evaluated both the hardware and software demands
tions, based on the controller architecture.
that will be placed on the microcomputer(s), a device
As larger-scale program memory (particularly with the
selection can be made. Alternatives include the
wide-scale adoption of Flash memory) is now becoming
• ASIC/FPGA/SoC options, mentioned above; available, C++ and Java are becoming increasingly used.
• externally supplied microcomputer subsystem; and Language subsets, removing the more memory-hungry
Embedded Systems 497
aspects of the languages, and targeted at the embedded MPLABTM ). Importantly, it holds values of all internal
world, include Embedded C++ (EC++) and Embed- processor registers and memory locations, which can be
ded Java. examined by the user. It can run the program continuously
or single-step.
MPLAB allows breakpoints to be set, which cause the
4.2 Multitasking in real time program execution to halt at a specified instruction. Its
Trace function maintains a record of register values up to
Most embedded systems are multitasking, and must meet the time of program halt. Although it does not operate
real-time programming demands, that is, they have a num- in real time, instruction or machine cycles can usually
ber of distinct functions, which must be performed more be counted, from which program execution time can be
or less in parallel, and which must meet certain deadlines. deduced. It is possible to simulate the external environment
These may be of different relative importance, and may or in only a limited way. Despite these shortcomings, once a
may not be synchronous with each other. program has run on a simulator, there can be a fair degree
A particular technique for dealing with this situation is of confidence that it will operate in the target hardware.
to apply a Real Time Operating System (RTOS). It is attractive to retain the facility of the software
An RTOS determines which task is given CPU time, and simulator when the program is downloaded to the target
for how long. This is called scheduling. It also controls hardware. This can be achieved by using an In Circuit
task synchronization, use of system resources that may be Emulator (ICE). This is a device that usually exists in the
needed by more than one task, and passing of data and form of a pod, which is inserted in the target hardware
messages between the tasks. in place of the microcontroller or processor. It remains
Various scheduling strategies exist. In Cyclic Schedul- connected to the host computer by an umbilical cable.
ing, each task runs until completion, before the CPU turns The ICE is designed to act precisely as the processor
to the next task. A Round Robin scheduler simply gives it replaces. However, it allows its internal registers to be
CPU access to each task for a fixed period in turn, irre- interrogated by the host computer, and generally it has
spective of its priority. This is an example of preemptive the breakpoint, trace, and other facilities of the software
scheduling, as task execution is interrupted before comple- simulator. The ICE is not without its drawbacks. It is a
tion. A prioritized preemptive scheduler gives priority to costly tool, and committed to only one processor, or family
tasks that have been designated as high priority; low prior- of processors. There can be subtle restrictions on its use,
ity tasks are not executed unless there are no high priority for example, it may have clock frequency or power supply
tasks waiting. Further details are given in many sources,
range constraints, or it may introduce awkward timing
for example, Simon (1999).
errors in high-speed systems. For these reasons, and because
digital electronic devices are becoming ever more complex
and compact, the ICE is slowly being replaced by other
5 TESTING OF PROTOTYPE EMBEDDED
test methods.
SYSTEMS Certain manufacturers now integrate diagnostic tools
onto their chips. A good example is the Background Debug
Embedded systems can be a particularly challenging type Mode (BDM) facility of many Motorola devices that
of product to test. Signals and logic states may be deeply allows values of internal registers to be read by the host
embedded within the system, and impossible to monitor. computer via a dedicated serial link. As the diagnostic facil-
Worse still, the system is under the control of a fast-moving ity is built into the device, it is by definition noninvasive,
program, which may instigate fault conditions under cir- and can have very valuable results.
cumstances that are then very difficult to replicate. A range Conventional electronic test tools, notably the oscillo-
of tools, of varying levels of sophistication, exists to address scope and logic analyzer, remain useful, especially in sim-
these challenges. pler designs. Certain products, for example, from Agilent,
It is highly advantageous to avoid a situation in which combine the functionality of both instruments into one.
untested software is run in untested hardware. This can be
avoided by the use of software and hardware simulation.
A software simulator is a computer package, usually part
of the IDE, which attempts to replicate the action of a REFERENCES
particular microcomputer under test, and which can run
a trial program. It is low cost, and sometimes provided Hatton, L. (1994) Safer C, McGraw-Hill International, ISBN
free of cost by microcontroller suppliers (e.g. Microchip’s 0-07-707640-0.
498 Measuring System Design
MISRA (Motor Industry Software Reliability Association) (1998) van Sickle, T. (2001) Programming Microcontrollers in C,
Guidelines for the Use of the C Language in Vehicle Based 2nd edn, LLH Technology Publishing, ISBN 1-878707-
Software, McGraw-Hill International, ISBN 0-952415-69-0. 57-4.
Pahl, G. and Beitz, W. (1995) Engineering Design, a Systematic Wilmshurst, T.J. (2001) An Introduction to the Design of Small-
Approach, 2nd edn, Springer-Verlag, London. Scale Embedded Systems, Palgrave Publishers, ISBN 0-333-
Simon, D.E. (1999) An Embedded Software Primer, Addison- 92994-2, Basingstoke.
Wesley, McGraw-Hill International, ISBN 0-201-61569-X.
79: Testing Electronic Systems
Patrick D.T. O’Connor
Consultant, UK
1.2 Digital circuitry systems
1 Circuit Test Principles 499
2 Manual Test Equipment 500
3 Automatic Test Equipment (ATE) 500 The basic approach to testing digital circuits is to check
that the output logic is correct for each possible input state.
4 Special Test Equipment 502
As a simple example of combinational logic, consider a 2-
5 Component Test 502 input AND gate (Figure 1). The possible input conditions
6 Design for Test 502 and the correct outputs for each of these is shown in the
7 Test Software 503 truth table. In principle, it is simple to test such a device for
8 Example of Manufacturing Test 503 all possible input conditions and faults. The fault conditions
References 503 in this basic approach are inputs or outputs stuck at 1 SA1
and stuck at 0 SA0.
In this case, four test conditions, or test vectors, are
sufficient to test all operating conditions. For any logic
1 CIRCUIT TEST PRINCIPLES network, the number of vectors necessary to show cor-
rect operation is 2n , where n is the total number of pos-
The term unit under test (UUT) is often used in the context sible input conditions. For a simple digital circuit, this
of electronics testing to refer to the item or system being presents no problem, but as device complexity increases,
tested. The term fault is used for any defect or malfunction. the time taken to perform such a test can become too
long to be performed practically and economically in mass
production.
1.1 Analog circuitry systems
In devices such as memories and processors, or complex
circuits that include such functions, the outputs depend
Testing analog circuit functions and parameters involves a
upon the inputs as well as on data flow and memory
range of measurement technologies and instruments. Some
locations of stored data. This is sequential logic. Alternative
of the most important and frequently applied tests are:
test methods have been developed for these devices, such
• current, voltage potential, and resistance as pattern sensitivity tests for memories.
• gain, impedance, waveforms, and other aspects of There are other practical limitations of the basic SA
AC circuits fault model approach to testing. There are other kinds of
• signal characteristics, such as frequency, gain, power, fault condition that can occur, such as a stuck at input
distortion, jitter, and so on, for audio frequency (10 Hz– condition, in which an output follows an input logic state.
10 KHz), radio frequency (RF) (10 KHz–1000 MHz), Fault occurrence can also be dependent upon other factors,
and microwave (1000 KHz–100 GHz) and high-speed such as timing, speed, interference, temperature, or voltage
digital circuits. supply, as described in O’Connor (2001).
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
500 Measuring System Design
A • special instruments, such as radio frequency testers,
Inputs O (Output) distortion meters, high voltage testers, optical signal
B
testers, and so on.
Truth table: Figure 2 shows some typical modern manual test
A B O instruments.
0 0 0
0 1 0
1 0 0
1 1 1 2.1 Computer-based testing
Figure 1. Truth table for 2-input AND gate. Computer-based testing uses software that enables personal
computers (PCs) to emulate test equipment. The PC is
connected to the circuit to be tested via a data acquisition
2 MANUAL TEST EQUIPMENT adaptor card. The PC screen becomes the instrument
display, and the tests are controlled from the keyboard
and mouse (Figure 3). Software such as LabVIEW –
Manual test equipment is used primarily during develop-
see Article 105, Modeling with LabVIEW , Volume 2
ment since, at that stage, it is not important to minimize
and LabWindows (National Instruments Corp. (URL:
test time and greater flexibility is possible.
http://www.ni.com) (Johnson, 1997)) ease the task of test
The main types of manual test equipment are:
design. Computer-based testing can reduce the costs of
test equipment since one PC can perform a range of test
• basic instruments, such as AVO meters (amps, volts, functions and test results can be easily stored, analyzed,
ohms) or digital multimeters (DMMs), power meters, and transmitted. The microprocessors in PCs also enable
and transistor testers; tests to be performed faster.
• oscilloscopes, spectrum analyzers, and waveform gener-
ators to measure, compare, and analyze voltage and cur-
rent levels and waveforms and to provide inputs so that 3 AUTOMATIC TEST EQUIPMENT (ATE)
circuit responses to input waveforms can be analyzed;
• logic analyzers, which are similar to oscilloscopes but Electronic equipment is tested as part of the manufacturing
specialized for displaying digital pulse streams; process to ensure that faults are identified and diagnosed
(a) (b)
(c) (d)
Figure 2. Manual test instruments: (a) DMM (copyright Fluke); (b) logic analyzer (copyright Agilent); (c) RF tester (copyright
IFR); (d) optical signal tester (copyright Agilent).
Testing Electronic Systems 501
Figure 3. PC-based oscilloscope with conventional instrument.
(Copyright National Instruments.)
and that shipped units are fault-free. In order to per-
form the tests economically, especially at high production Figure 4. In-circuit tester (ICT). (Copyright Agilent.)
rates and for complex circuits, most manufacturing test-
ing is performed using automatic test equipment (ATE).
relatively more common than faulty components. Figure 4
The main types of ATE for electronic assembly testing are
shows an ICT system.
given below.
3.1 Vision systems 3.3 Flying probe testers
These refer generically to inspection systems that acquire Flying probe testers (also called fixtureless testers) access
an image and then analyze it. These include automatic test points on the circuit, using probes that are rapidly
optical (AOI) and X-ray inspection (AXI). They do not moved between test points, driven by a programmed high-
actually test circuits, but they have become part of many speed positioning system. The advantage over ICT is the
production test sequences because of the great difficulty fact that the probe programming is much less expensive and
of human inspection of the large numbers of compo- more adaptable to circuit changes than expensive multipin
nents, solder connections, and conductor tracks on mod- ICT fixtures are, which must be designed and built for
ern circuits. each circuit to be tested. A consequent advantage is that
they can be used for testing during development as well as
in production.
3.2 In-circuit testers (ICT), manufacturing defects
analyzers (MDA)
3.4 Functional testers (FT)
ICT (In-circuit testers) tests the functions of components
within circuits on loaded circuit boards. It does not test Functional testers (FT) access the circuit, at the circuit
the circuit function. The ICT machine accesses the compo- board or assembly level, via the input and output connectors
nents, one at a time, via a test fixture (sometimes referred or via a smaller number of spring-loaded probes. Functional
to as a bed of nails fixture), which consists of a large num- testers usually include facilities for diagnosing the location
ber of spring-loaded contact pins spaced to make contact of causes of incorrect function. There is a wide range, from
with the appropriate test points for each component. MDAs low-cost bench-top ATE for use in development labs, in
(manufacturing defect analyzers) are similar but lower-cost relatively low complexity/low production rate manufacture,
machines with capabilities to detect only manufacturing- in-service tests, and in repair shops to very large high-speed
induced faults such as opens, shorts, and missing compo- high capability systems. The modern trend is for production
nents: justification for their use instead of ICT is the fact ATE to be specialized and focused at defined technology
that, in most modern electronics assembly, such faults are areas, such as computing, signal processing, and so on.
502 Measuring System Design
EMC and EMI, Volume 1 and Article 216, Test Facili-
ties, Volume 3.
5 COMPONENT TEST
Since electronic components are always tested by their man-
ufacturers, and since modern manufacturing quality is very
high, nowadays, it is very seldom appropriate or neces-
sary for components to be tested by equipment or system
manufacturers. We can consider electronic components to
be in two categories from the test point of view: discrete
components and integrated circuits or ICs.
5.1 Discretes
Figure 5. Functional tester. (Copyright Agilent.) Components such as resistors, capacitors, connectors, coils,
transistors, and other relatively simple types generally pos-
Some ATE for circuit testing during manufacture includes sess a relatively small number of performance parame-
combined ICT and FT; the ICT functions are performed ters. Therefore, testing them is relatively easy, quick, and
first, followed by FT. Figure 5 shows a typical modern inexpensive, and they are fully tested by the component
test station. manufacturers. The manufacturing processes for these com-
Electronics ATE is a very fast-moving technology, driven ponents are also relatively easy to control, resulting in high
by the advances in circuit performance, packaging and yields or low proportions of defectives. Therefore, testing
connection technology, design automation, and production by users is rarely cost-effective.
economics; O’Connor (2001) and Brindley (1991) provide
more information.
5.2 Integrated circuits
4 SPECIAL TEST EQUIPMENT ATE for IC testing must be capable of testing all or most of
the functions of the very complex circuits, at the maximum
There is also a wide range of specialized ATE, such as operating speeds. The problem of testing very large and fast
for integrated circuits, power supplies, cable assemblies, ICs is one of the major challenges of modern electronics
communication systems, and so on. technology. New methods for testing very large and fast
ICs are being developed in response to the challenges of
4.1 Hot rigs complexity and cost. ATE for IC testing is very expensive,
so testing by users is seldom cost-effective.
A ‘hot rig’ is a system that is used as a test facility
for the subassemblies from which it is constructed. Any
of the subsystems can be tested by being inserted into 6 DESIGN FOR TEST
a cabinet that contains the other subsystems and then
the system can be tested. ‘Hot rigs’ are often used for 6.1 Tasks of an ATE
testing in the manufacture of systems, particularly when
the quantities produced are relatively low so that test time In order to test a circuit or system, the ATE must be able
is not critical. They are also used in repair depots for testing to perform the following functions:
repaired items.
• Initialization. The ATE must put the UUT into known
initial conditions, for example, digital inputs and out-
4.2 Electromagnetic compatibility (EMC) testing puts at known logic states, parameters at known values,
and so on.
Testing for electromagnetic compatibility EMC and elec- • Control . Inputs must be generated to stimulate changes
tromagnetic interference EMI is described in Article 17, on the outputs.
Testing Electronic Systems 503
• Observation. The ATE must be able to detect the
Pass Pass ICT/ Pass
changes in output states and values. Assemble AOI MDA Ship
FT
• Diagnosis. The ATE might be required to perform
automatic or guided diagnostics to identify the origins Fail Fail Fail
or causes of faulty output states.
• Partition. It is often important that the ATE can test Diagnose/
relatively small parts of complex systems separately repair
to maximize test coverage, to enable effective fault
Figure 6. Typical electronic manufacturing test flow.
diagnosis, and to minimize test time.
These capabilities are affected not only by the performance many connections, particularly if they cannot otherwise be
of the ATE but also significantly by the design of the accessed for test by the ATE. The international standard for
UUT. Therefore, it is important that the system design the boundary scan approach is IEEE1149.1. Most modern
takes account of the need for test. This is called design digital ICs include boundary scan capabilities. The bound-
for test (DFT). ary scan method is described in Parker (1999), and infor-
mation is also available from JTAG Technologies (URL:
http://www.jtag.com).
7 TEST SOFTWARE
Test software must be developed for any circuit that is to be 8 EXAMPLE OF MANUFACTURING TEST
tested on ATE. Automatic test program generation (ATPG)
software is available to perform this task. The ATPG will For any manufacturing situation, the optimum test approach
create the test software from the circuit information on an must be planned in relation to technologies, economics, and
electronic design automation (EDA) system and indicate other factors. This topic is covered in Davis (1993) and
the fault coverage. For some systems, particularly aerospace O’Connor (2001). Figure 6 shows an example of a typical
electronics (avionics) and military systems, the abbreviated manufacturing test sequence.
test language for all systems (ATLAS) test language is used.
REFERENCES
7.1 Scan design
Brindley, K. (1991) Automatic Test Equipment, Newnes, London.
For very large digital electronic circuits, internal circuits Davis, B. (1993) The Economics of Automatic Testing, 2nd edn,
are often provided, which assist the ATE to perform McGraw-Hill, London.
the tests. The best-known approach is boundary scan. A Johnson, G.W. (1997) LabVIEW Graphical Programming, 2nd
boundary scan circuit receives test instructions from the edn, McGraw-Hill, New York.
ATE and then performs logical tests on the interconnec- O’Connor, P.D.T. (2001) Test Engineering, Wiley, Chichester.
tions between the digital circuits that it controls. This Parker, K. (1999) The Boundary Scan Handbook, Kluwer Aca-
capability is important for large IC packages containing demic, New York.
80:Principles of Fine Mechanics – Kinematic and
Elastic Designs
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
applies equally well to mechanical instruments that are a
1 Purpose of Mechanical Design in Instruments 507 form of machine.
2 Specification of Measurement Systems 508 Departure from the ideal concept will occur for many
3 Problems of Specification 510 reasons, for example, unavoidable elasticity of component
members, clearances and wear in joints, and ‘second-
4 Controlling the Mechanical Degrees of
Freedom 511 order’ influences such as thermal expansion of the material.
Design is the process whereby a satisfactory compromise
5 Kinematic Design Basics 511
is made between the range of envisioned parameters to
6 Elastic Design Basics 513 come to a single adequately satisfactory arrangement. The
Related Articles 513 designer strives to produce a design statement that will lead
References 513 to an article being made, which performs the set task within
the stated specifications of performance without the need to
remake the article after it is first assembled.
1 PURPOSE OF MECHANICAL DESIGN Mechanical design has two main stages of progression.
The first is where the kind of necessary mechanisms are
IN INSTRUMENTS
realized as a system. The following stage then selects a
Mechanics is the subject that deals with the study and appli- method of creating the chosen individual mechanisms as
cation of the motion and equilibrium states of material hardware that will suit the total task. For example, the
bodies. The aim, in the mechanical aspect of instrument mechanism seen to be needed may be a means of rotat-
design, is to create a specific mechanical system that pro- ing one shaft at a different speed to that of the other while
vides the necessary controlled movements in such a way maintaining a strict gearing ratio. This is the first step. Fol-
that predetermined moving and fixed relationships are pre- lowed by the study of the various ways of achieving this –
served under reasonable conditions of operation. toothed belts, friction gearing, spur gears, and others – to
Reuleaux, in his Kinematics of Machines of 1875, was see which suits the case in hand.
apparently the first person to see machine design as more No matter which type of mechanism is involved, the
than a complete design that was modified until it worked. fundamental issues behind the choices of the first step are
He defined a machine as ‘a combination of resistant bodies concerned with controlling the freedom of movement of
so arranged that by their means the mechanical forces of selected parts and points. This will be expanded later.
nature can be compelled to do work accompanied by certain The design of a measuring instrument, that a machine
determinant motions’. He further expanded that motion of uses for handling informational rather than energy quan-
parts is simply determined by the geometry of the system tities, generally requires specific attention to the fidelity
of parts (Burstall, 1970). Reuleaux’s generalizing definition of the input–output and other relationships over extended
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
508 Measuring System Design
periods of operational life. There is less interest in efficiency very close tolerances are required. Dimensional parameters
of energy transfer (although in mechanical design, losses in mechanical systems will often require parts per mil-
due to friction can cause thermal changes to components lion precision compared with parts per hundred in most
that give rise to error). Instrument design is generally dom- electronic techniques. If adequate performance is to be
inated by attention to deflections of the system and its achieved, it means that the various parts of a mechani-
components rather than to their strengths. cal system must be made precisely, or at least be capable
A general introduction to mechanical machine design of being assembled to close tolerances. Control of wear is
can be obtained from the widespread works describing the important since it may degrade performance.
design of machines that can transfer energy, examples being Electronic design, in contrast, can more easily compen-
Shigley (1988), Erdman, Sandor and Kota (2001), Juvinall sate for changes to the value of components, since they
and Marshek (2002), Mott (1998), Faires (1955), Beggs have been initially burned in to eliminate the few that will
(1955) and Lent (1970). These accounts will, however, only fail early.
take the user a limited distance into instrument design, for To its credit, mechanical design is more intuitive and
they generally lack specific detail about fine mechanism readily understood; wrong factors in a mechanical system
requirements. often show up very easily without the use of sensory
Work which discusses the general systems level of design aids because mechanical systems interface more directly
of instruments are few and generally now seen to be old by with the human sensors than concepts of other disciplines.
virtue of their date of publication. They are, however, most Mechanical resonant systems can also yield very high-Q
useful accounts, encapsulating considerable experience and oscillations for low component cost.
know-how. These include Glazebrook (1922) Vol. III, Rolt An important feature of mechanical systems, which may
(1929) Vol. II, Whitehead (1934), Strong (1938) and Brad- be either advantageous or deleterious, is the inertia of com-
dick (1954). This material is provided to help overcome the ponents. The necessary inherent mass of mechanical parts
difficulties in procuring information on this topic, now that is often the limiting parameter to dynamic performance
it has little research in place. when high frequency operation is needed. It can, of course,
Factors important to design include the methodology for be a useful factor providing time integration properties. In
controlling the design itself, cost of producing the design, a multiregime instrument, the mechanical parts will often
building the instrument and servicing it, the degree of dictate the upper frequency limit of operation. Generally,
serviceability, the ease of measurement in production, the unless specifically and carefully designed to be otherwise,
needs for calibration after manufacture, and ergonomics mechanical systems will be low pass in performance.
requirements (now often called the man–machine system –
MMS). These and others must all be properly thought out
so that the users get correct results. Many of these vitally 2 SPECIFICATION OF MEASUREMENT
important factors are often neglected in the initial design, SYSTEMS
being added by adaptation or modification at a later stage
when the design has progressed so far into production that There exists a maxim ‘$pecify $anely or $uffer’. It sums up
major change cannot be entertained. For example, a firm the importance of spending enough time on raising adequate
making fever thermometers only realized that their product and realistic specifications on which to base the design.
must fulfill certain calibration criteria when they were ready Mechanical design does not differ from any other design in
to launch the product on the market place! that to get started one must have a set of specifying state-
At present, instrument design is largely a case of some ments that will outline, in a quite rigorous and meaningful
intelligent application of science plus a very large com- manner, what the instrument must do. These statements
ponent of experience through years of actual involvement are called specifications. Too often, instruments just hap-
with similar products. pen with the specifications growing as the design proceeds.
Design procedures for each of the different physical A proper set of ‘specs’ at the beginning of the design will
regimes of an instrument have their own peculiarities. aid the designer in realizing what is ultimately required.
Mechanical design must allow for the fact that a wrong They should be set to paper before in-depth design begins,
choice is not as easily corrected as in, say, the elec- for once the designer has become immersed in the design,
tronic medium. There is, for example, no simple mechanical it is a fact of life that he or she can become very resistant
counterpart of the variable resistor of electronics, which can to change.
be placed into the system to find an optimum value with Specification should not be raised by the designer or
great ease and with little cost or performance penalty. To user alone but as a group effort. There should be a joint
obtain adequate performance from a mechanism, usually agreement between three parties – the system end user, the
Principles of Fine Mechanics – Kinematic and Elastic Designs 509
System end user System designer/Manufacturer Subsystem/Component vendors
Functional
performance, Procurement System
technical design
specification specification Items available
based on vendor
specifications
Operational Special vendor
requirements items per system
specification designer's
specification
System and
$
Pay hardware design
Pay $
Operational System Manufacture Purchased items
system delivered and
factory test
Installation,
commissioning
acceptance
testing
Figure 1. Vested interests in a typical system design process.
system designer and manufacturer (if they are of the same • Special subcontract specifications
group), and the subsystems and component vendors who • Incoming inspection specifications (system designer)
supply the parts that will be used, generally without modifi- • Incorporation of vendor specifications in system design
cation. Figure 1 shows how Wheeldon (1974) depicted this specifications.
process. His two papers expand the following points now
given. Various kinds of specifications exist, each having a The designer is the person who has to provide specific
different purpose of definition. detail to the factors presented by the various groups. A
Systems user and systems designer relationships realistic producible design at a satisfactory cost must be
achieved. Value judgments are needed about what the
• Operational requirements specifications customer says is needed (which is often what is then thought
• Functional and technical requirements specifications to be needed); what the vendor can offer (or what is said
• Design specifications (to implement and check design) to be available); the designer’s own ability and facilities
• Factory test specifications for design; and the permitted time and money available to
• Acceptance test specifications (Factory) complete the task.
• Documentation specifications Unfortunately, to specify a product completely before it is
• Packaging and delivery specifications made requires powers of prediction and enormous effort in
• Installation specifications attempting this virtual impossibility. To attempt to create a
• Commissioning specifications total specification that gives the user complete assurance of
• Acceptance test specifications (Field). being absolutely safe from faults and failures of compliance
Systems designer and component vendor relationships is an unreasonable expectation. The cost and time needed
to get even close to this are prohibitive. A compromise is
• Standard item specifications therefore struck that specifies the product in a short-form
• Modified ‘Standard’ items manner in reasonably standard ‘shorthand’ statements.
510 Measuring System Design
Specifications, therefore, attempt to portray what both the full encoder accuracy could not be used because the
the designer or maker and the purchaser or user expect to radar head was not designed for better than arc-minute use.
produce and receive, respectively. Owing to the small cost Specifications can also come to the designer as the result
level associated with the smaller systems and the stand- of the so-called ‘horse-camel’ syndrome. The end user
alone instrument, the verbal interface is poorly developed collects data sheets of the various products offered by
and exploited. In larger extensive instrument systems, the several different sources. From each of these he picks out
complexity there makes better assessment necessary, but the individual good points, adding these together to raise
then the complexity still prevents absolutely adequate state- his own specification list. This process entirely ignores
ments from being realized. the fact that these special features are often the result
of the designer trading off one feature to allow another
to be enhanced. The result of this combinational method
3 PROBLEMS OF SPECIFICATION is a specification of an instrument impossible to make.
Unfortunately, there are many suppliers in the instrument
In practice, dissatisfaction often arises between the parties manufacturing industry who lack enough experience to
to the contract and, in cases where the financial implica- recognize that such tenders may be impossible to make,
tions are large, independent arbiters are called in to make and they take them on only to find later (to their and the
judgments. The art and science of contract and specifica- purchaser’s regret) that they had been attempting a virtually
tion writing are now extensive; specialists exist to cover impossible task. The use of independent expert consultant
the needs. It is beyond the scope of this article to go much groups, a well-trained in-house team, study contracts, and
deeper, but here are some simple rules to follow. part-stage costing may help.
Try to use specifying statements that are the result of At all times, it is vital to ‘clear the air’ between all
past experience. Examples are Codes of Practice and the groups involved. If the user believes the designer can-
various Standards specifications. For instance, reference to not do what is suggested, then it should be challenged
the BS code for dimensional tolerances will show if those very early. Similarly, the designer should state what is
called for are reasonable for the size of the part. unclear early in the negotiations with the consumer. Change
Always watch for ‘minor’ details in a statement that becomes progressively difficult as the designer is grad-
may be more significant than they are thought to be. For ually forced to freeze the parameters in order to make
example, surface finish to be within a given limit may be concrete progress.
specified for a pressure cell capsule, yet nothing is given When writing specifications, considerable care must be
about the fluid tightness of the material. Further, never leave given to the provision of statements that are meaningful
important factors out because they seem so obvious. The and not ambiguous. Unfortunately, to write a reasonable
converse also holds; if they are not given, check them out. specification requires the use of abbreviated statements. For
All statements in a specification should be carefully example, statements of accuracy or linearity are used very
studied before signing an agreement; it is of little value loosely. Graphs could be used much more to define such
to look into it afterwards. Look carefully for fundamental parameters as frequency response, linearity, and the like.
inconsistencies. Does the frequency response called for When in doubt, seek further detail about the meaning of
exceed that of any mechanism yet produced? Does the a parameter.
required or specified noise level fall below the thermal There is no harm in verifying the self-quoted perfor-
control errors? Suppliers cannot make an instrument that mance of subcontractors and vendors.
refutes the known physical laws; but they have been known All too often, the user does not adequately think out
to offer such for sale – unwittingly, of course! the task for the instrument that is called for. The designer
Make clear what specifications apply to what, and at should be expected to ask poignant questions of the user,
what stage. Care is needed to prevent them ‘creeping’ in for the designer will often know the design problems more
the hands of enthusiastic clerks. There was a case of an closely than the user. Very often, discussions show that the
optical shaft encoder that had to read to arc-minute accuracy user really wanted something quite different from what was
when mounted on a radar tracking head. By the time thought to be necessary. Take, for instance, the person who
the specifications were transferred through several stages once asked for a thermometer to measure length changes
of contract writing and then into the maker’s plant, they of a component because he did not see that it could be
had crept to stated need for arc-seconds performance. The measured directly as a dimension.
manufacturer, not wishing to be caught out, went one better. Each case stands on its own merits. In measuring instru-
The result was that the encoder cost about a hundred times ment design, the spread of applications is so broad and
the price of one that would have done the job. Furthermore, the design becomes so complex that poor performance
Principles of Fine Mechanics – Kinematic and Elastic Designs 511
often results. The ongoing state of instrument design and At times, measuring machines are sometimes specified
manufacture is such that it is suggested that instrument users as, say, 3 12 axes, meaning that they have a rotation allowed
must constantly be on their guard against the supply of within the movement provided along one translation axis.
faulty instruments – see Moss (1978) for a relevant report It does not mean that the rule of six degrees of freedom for
illustrating the potential situation. an extended point has been broken.
Instruments, especially one-offs and those with mechan- Two extremes of design philosophy exist for realization
ical elements, are still in a state of low inherent quality of a suitable mechanical structural relationship. Each has
capability when compared with the performance that one its advantages and disadvantages and often the design
now expects and usually gets when using electronic sys- makes do with a compromise between the two extremes.
tems based on standard integrated circuits. If an IC works, These extremes are called kinematic and elastic designs; in
it is going to be in specification and continue to operate. between them is the semi-kinematic arrangement.
It will also be a very well designed component. In con-
trast, mechanical systems may continue to operate but with
deteriorating fidelity. 5 KINEMATIC DESIGN BASICS
When all parts of a structure are held in such a way that
4 CONTROLLING THE MECHANICAL they retain positional relationships without being in a state
DEGREES OF FREEDOM of variable stress except at their points of contact, the design
is said to be kinematic. In other words, effects such as a
Mechanical design is fundamentally concerned with pro- change in size due to temperature or material creep will
viding a system of materials arranged to allow controlled not stress the frame. Clerk-Maxwell in the Handbook of
movements, ranging from no movement in any axis, to all the Special Loan Collection of Scientific Apparatus 1878
six degrees of freedom of an extended object within a defin- wrote this:
ing space.
A theoretical point in space has three degrees of free- When an instrument is intended to stand in a definite
dom – the three translations that are generally expressed as position on a fixed base it must have six bearings, so
x, y, z in the Cartesian system; other coordinate methods arranged that if one of the bearings were removed the
direction in which the corresponding point of the instrument
can be used. The concept of rotation has no meaning for a would be left free to move by the other bearings must
theoretical point. be as nearly as possible normal to the tangent plane of
A point forming a position on an extended object, how- the bearing.
ever, can have six degrees of freedom for, in addition to
the three possible translations, there can be three axes of Strong (1938) defined the principle of kinematic design
rotation about the point. as being applied when a body has at least 6-n points in
The essence of good mechanical design is to devise a contact with a second reference body when it is to have n
system that restrains as many of the six freedoms as are degrees of freedom.
needed while allowing freedoms. For example, the slide of a A spherical ball sitting in a triangular hole, as shown in
traveling microscope must allow one translation, and while Figure 2(a), rests on three points of contact while it is under
doing so, constrain the other two translations and three the force of gravity (which acts as the ‘locator’). Thus, it
rotations as perfectly as are needed. Lack of straightness has three degrees of freedom constrained with respect to
of the slide will cause small translations and rotations to the datum block (the translations) yet is free to rotate about
occur as it slides. The degree of precision needed dictates three mutually perpendicular axes.
the necessary tightness in the tolerances of manufacture The classic example, demonstrating constraint of all
and the method by which this is accomplished in the face six freedoms, is the so-called ‘hole-slot-plane’ or Kelvin-
of wear, temperature effects, and other error sources. coupling arrangement. This is easily understood by refer-
Some other examples are the framework and sliding axes ence to Figure 2(b). The hole constrains three translations
of a 3-D measuring machine (three translations allowed (three contact points), the slot constrains two rotations (two
with rotations constrained) and the surveyor’s theodolite contact points), and the plane constraints the other rotation
(constrains all translations allowing the rotations). (one contact point). Addition of any extra contact points,
An equally important case is where all degrees of free- unless made perfectly enough (as in a ball-race bearing),
dom are to be constrained. This occurs when mounting a would add a ‘redundancy’ causing elastic deformation of
component to stay in a chosen position relative to another; the frame due to ‘overconstraint’.
for example, a method of locating a survey-leveling unit It is easy to see that changes to the size of the top plate
that is to be used repeatedly around a series of test lines. will cause the legs to move yet retaining the same stress
512 Measuring System Design
Ball Load bearing
area
x
x 3
x
mponent
Datum co
(a)
x x x1
2
Area within which load Unconstrained
should be bearing x x2 translation
(a)
x – contact point
3x
2 x x Hole
1
x
x
Slot Plane x
x x
(b) 2 x
1
x x2
Center lines remain
coincident as top
CL
plate changes size 3 rolling mechanisms
to reduce wear and
(b) friction
2 Figure 3. Kinematic design allowing one freedom of translation;
x x x (a) contacts subject to high wear; (b) contacts replaced with
2x x2 rolling devices.
x
120°
The principle of kinematic (or other alternative) design
(c) is applied whenever two assemblies need to be coupled.
x – contact point Gravity force acts as the locator It is necessary, therefore, to break down the whole system
into its basic couplings or mechanisms, tackling the design
Figure 2. The degrees of freedom in a kinematic design of a of each in a hierarchical order to ensure that overconstraint
coupling are equal to six less the number of contact points. (a) All does not occur unwittingly at a later stage by the addition
three translations constrained. (b) Hole-slot plane in which all six
freedoms are constrained. (c) Slot-slot-slot variation of case b. of a link in an incorrect way.
Features of the kinematic method are that its parts can
be made and installed or replaced without altering the posi-
relationships (provided the slotted plane contacts move tional system. Once made, the kinematic assembly will be
accordingly). Note that the hole-slot-plane does not, how- reassembled with the same dimensional performance. Sec-
ever, retain the center of the top member coincident with ondly, wear does not induce play in the coupling mode – it
that of the lower member should a relative dimensional may, however, introduce dimensional change. Another fea-
change occur between the two members. A variation of the ture is that the stresses exerted on all parts are well defined,
‘hole-slot-plane’ that uses three radiating slots (Figure 2c) and they can be designed virtually in isolation from the rest
overcomes the center shift defect. Figure 3(a) shows a kine- of the system once their contact loads have been decided.
matic form in which one degree (translation) is allowed, this Disadvantages are that there is a need for application
being proven because five contact points exist. Figure 3(b) of a locator force to hold parts together, such as gravity,
shows how this form is modified with rolling bearings, to magnetism, or small springs. The points of contact are also
retain point contact while reducing wear. It is typical of the heavily loaded and consequently operate under conditions
slide of a traveling microscope. of high stress. The pure principle is often degraded by
Many arrangements are possible depending on the need the use of extended contact surfaces to reduce the contact
to allow or control a given number of degrees of freedom. stresses – semi-kinematic design.
Sketches of various couplings are given in Whitehead A further difficulty arises when it is seen that the sys-
(1934) and Braddick (1954). tems result in triangular space frames when redundant links
Principles of Fine Mechanics – Kinematic and Elastic Designs 513
are not present. The three-legged stool is a good exam- of strains due to disturbances shows them to be adequately
ple of the possible undesirable features of purely kinematic stiff or compliant as the case dictates.
design. The stool always sits correctly on any reference The mechanisms of the two groups of design – kine-
surface without rocking, but place the load outside the cen- matic and elastic – will be dealt with in more detail in
tral triangle defined by the leg positions and the linkage is later articles.
soon lost. The locator should not be designed to withstand It is surprising that few texts on general machine design
such forces. actually discuss these basic design differences. In some of
the works that are specifically concerned with instrument
design, these concepts are expanded. Greater depth and
6 ELASTIC DESIGN BASICS more examples on the topic can be found in Glazebrook
(1922) Vol III, Whitehead (1934), and Braddick (1954).
The opposite philosophy is to join components rigidly, The choice of which philosophy to use and when is very
allowing them to flex within their elastic properties: a four- much a matter of training and experience. Where elastic
legged stool will still sit on any surface if the top is made methods can be used, they usually offer considerable cost
flexible. The elastic philosophy deliberately contravenes the and assembly advantages. For example, locating one part
‘zero redundancy’ law, doing so in a defined manner. relative to another with, say, one allowed translation of
For this method to be designable, the system must be so short stroke can easily be provided with a parallel-strip
constructed that the designer knows just what flexes and suspension. It will be virtually free of frictional effects. It
where. In cases where only limited translation and rotation will be cheaper to make but does suffer in that a changing
is desired, the elastic method is probably the best. Purely deflection force is generated as the ‘slide’ moves.
elastic design is almost an academic abstraction. In practice,
the method comes somewhere in between totally kinematic
and elastic, being called semi-kinematic. RELATED ARTICLES
As an example of the two extremes, consider the need to
Article 64, Executing A Measuring System Design, Vol-
swing a pendulum or the beam of a weighing balance. The
ume 2; Article 69, Requirements Allocation, Volume 2;
pivot may be made using knife-edges that rock on a rigid
Article 81, Principles of Fine Mechanics – Systems Con-
flat (kinematic) or it may rotate on a single or a crossed
siderations, Volume 2; Article 85, Elastic Regime of
flexure strip pivot (elastic). Another example might be the
Design – Design Principles, Volume 2.
need to support a sensitive tilt meter on a given surface
in such a way that the relative length changes, owing to
temperature effects on the body of the instrument with no REFERENCES
strain on the instrument body. This can be done with either
the kinematic hole-slot-plane concept or by the use of three Beggs, J.S. (1955) Mechanism, McGraw-Hill, New York.
elastic flexible legs that bend a little as the body changes Braddick, H.J.J. (1954) The Physics of Experimental Method,
length (see Figure 4). It is important to make specific Chapman and Hall, London.
calculations of the degree of flexing that is needed. What Burstall, A.F. (1970) A History of Mechanical Engineering, Faber
may appear to be a stiff rigid member can be comparatively and Faber, London.
flexible when the forces are high or the movements small. Erdman, A.G., Sandor, G.N. and Kota, S. (2001) Mechanism
The designer should regard all components at the onset as Design: Analysis and Synthesis, Prentice Hall.
being compliant, altering this belief only after calculation Faires, V.M. (1955) Design of Machine Elements, Macmillan,
New York.
Glazebrook, R. (1922) A Dictionary of Applied Physics, Vol. III,
All freedoms Macmillan, New York.
constrained
Juvinall, R.C. and Marshek, K.M. (2002) Fundamentals of
Machine Component Design, Wiley, New York.
Thin legs flex in
Lent, D. (1970) Analysis and Design of Mechanisms, Prentice-
calculable manner Hall, Englewood Cliffs, NJ.
as changes Moss, C. (1978) Quality of precision measuring equipment. NCSL
Points of contact Newsletter, 18(1), 23–24.
held to be constrained
in three translations
Mott, R.L. (1998) Machine Elements in Mechanical Design, Pren-
tice Hall.
Figure 4. Semi-kinematic design – (ANAC tiltmeter support Rolt, F.H. (1929) Gauges and Fine Measurement, Vol. II, Macmil-
system). lan, London.
514 Measuring System Design
Shigley, J.E. (1988) Mechanical Engineering Design (Mc- Wheeldon, R. (1974) Specifications – The Identifying Facts,
Graw-Hill Series in Mechanical Engineering), McGraw- Department of Cont. Education, University of New England,
Hill. Australia.
Strong, J. (1938) Modern Physical Laboratory Practice, Blackie Whitehead, T.N. (1934) Instruments and Accurate Mechanisms –
and Son, Glasgow. Underlying Principles, Macmillan, London.
Principles of Fine Mechanics – Systems
81:
Considerations
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
wear through use they retain the dimensional precision
1 Wear-true Philosophy in Mechanical Design 515 needed. He called it the wear-true philosophy.
2 Deciding the Right Size of a Fine The use of kinematic principles goes a long way toward
Mechanical Instrument 515 this end, and by appropriate choice of proportion and
3 Static and Dynamic Considerations for a Fine arrangements, it is often possible to achieve a linkage that
Mechanism 516 does not degrade in its accuracy as it wears or deforms.
Darwin illustrated this using the examples of the bearing
4 Mechanical Systems Modeling 517
arrangement of a wheelbarrow wheel. When the axle is
Related Articles 519 fixed in the wheel, the axle ends wear round, and the
References 519 bearing hole wears oval, yet still round enough to give
good location. But fix the axle in the shafts and the shaft
wears oval in the wheel and the wheel bearings wear very
1 WEAR-TRUE PHILOSOPHY IN large and sloppy.
There seems to be no basic statable rule for achieving
MECHANICAL DESIGN this and a situation that occurs often is when the designer
attempts to arrange wearing surfaces to wear evenly on each
This article continues the discussion given in Arti- contact face of the sliding surface. Another approach is to
cle 80, Principles of Fine Mechanics – Kinematic and make the wear or deformation occur at a geometric position
Elastic Designs, Volume 2 by adding systems issues. that has the least effect on the geometrical precision of the
Kinematic design is covered in Article 82, Kinemati- measuring component. Figure 1 shows two designs, one of
cal Regime – Members and Linkages, Volume 2; Arti- which does not adhere to the ‘wear-true’ philosophy.
cle 83, Kinematical Regime – Fasteners, Bearings, Vol- Darwin incorporated this principle in the design of his
ume 2 and Article 84, Kinematical Regime – Rotary very successful ‘rocking microtome’ instrument of the
Motion, Volume 2. Elastic design is dealt with in Arti- 1885 period. It was capable of retaining the ability to
cle 85, Elastic Regime of Design – Design Principles, cut microscope specimen slices to micrometer thickness
Volume 2; Article 86, Elastic Regime of Design – Spring for a half century of service – see Cambridge Instrument
Systems, Volume 2 and Article 87, Elastic Regime of Company (1945) and Braddick (1954).
Design – Plates and Bimorphs, Volume 2. Errors sources
in both kinematic and elastic deisgn regimes are the subject
of Article 88, Error Sources in Fine Mechanics, Vol- 2 DECIDING THE RIGHT SIZE OF A
ume 2. FINE MECHANICAL INSTRUMENT
In the Aeronautical Journal of July 1913, Horace Darwin,
cofounder of the Cambridge Instrument Co., expounded a Unlike the situation occurring in electronic and in optical
concept of design that states that as mechanical elements disciplines where the size of components usually dictates
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
516 Measuring System Design
Microdisplacement Strain measured is d / Ergonomic needs dictate that the unit must have manual
transducer controls that suit man-size fingers or be mounted at heights
that can be easily seen and reached. Whatever the reason,
d the designer should attempt to find objective bases for
choosing a given size. If not, there will always exist the
possibility of having to rescale the physical form at a
later date.
Wire forming the
length standard
Variation in flexure pivot
3 STATIC AND DYNAMIC
does not influence CONSIDERATIONS FOR A FINE
Mass providing tension to sensitivity
wire MECHANISM
(a)
The static regime of a measuring systems is covered in
d Article 57, Static Considerations of General Instrumen-
tation, Volume 1. The dynamic regime is introduced in
measured is
Article 59, Introduction to the Dynamic Regime of Mea-
Strain
b /a d
surement Systems, Volume 1. The transfer function of a
a
system block describes the input–output relationship at all
relevant frequencies of operation. It comprises a DC gain
b Changes in flexure length term multiplied by a frequency dependent term. Although
can alter gain ratio b/a
(b) there exists in reality a continuum of performance from
DC to the highest frequency of operation, descriptions of
Figure 1. Two balances used to tension a wire length-standard in the performance are generally split into two domains, the
earth-strain meters; (a) uses wear-true design principle, (b) does first being the steady state or static characteristics and the
not.
other the dynamic behavior that is described by the fre-
quency term.
the overall size of the instrument system, mechanical Mechanical systems are no different from others in that
instruments may or may not have clear-cut reasons for they possess both static and dynamic characteristics.
having any particular size. So, at the outset of design, The design of a direct-reading traveling microscope
a choice must be made if progress is to take place that requires virtually no understanding of the dynamics of
does not need to be completely revised later. Here are a movements that the operator imposes on its sliding parts
few considerations that may help decide the overall size unless it is large, in which case the inertia of the sliding part
of the instrument. Braddick (1954) provides specific case can be great enough to impart excessive forces restricting
studies to illustrate some particular reasons for a given the operator’s manipulation response. However, if the
size. readout is automated by closing the translation drive loop,
Certain functions, such as a size dictated by a necessary the dynamics of the whole system become important if
component, may decide the basic size to be used. Examples stable and rapid high-speed operation is to be obtained.
are the objective size and focal length in a telescope, the Apart from a very cursory statement saying that dynamics
length of translation of a measuring machine, the load and do occur in instruments, the majority of instrument design
volume to be supported or contained, and the need to match texts give little help with this aspect to the designer of
a given existing product. modern multiregime instrumentation. Instead, the designer
Portability may dictate low weight, but small size often must read much more widely in order to become familiar
means increased cost and difficulty with assembly and ser- with the basics of dynamical systems and then apply the
vicing. However, dynamic response rises as parts become general rules to instrument design where it is needed.
smaller. The physical size of a high frequency response Definition of static performance involves specifications
anemometer must be very small. being stated about such parameters as resolution of read-
If energy must be stored in mechanical or thermal out sensitivity to a given variable input, linearity, accuracy,
form, large rotating or static masses may be needed. precision, repeatability, and many more concepts. These
Examples are the wheel of a gyroscope and the bed concepts, which are unfortunately not totally standardized
of a measuring machine that must integrate short-term in terminology, are described in most works on instrumen-
temperature fluctuations. tation and therefore need not be developed here. Doebelin
Principles of Fine Mechanics – Systems Considerations 517
(1975), Foxboro (1972), and Bell and Howell (1974) are Thorough study of the dynamic performance of an
examples. They are terms generally describing only the instrument is far more complex and time consuming
static behavior, the results of tests measured after the sys- than is the case for the relatively simple static domain.
tems come to rest. During transients, the performance can Certain mechanical parts may not be stiff enough or
be very different indeed. In many cases, the manufacturer may possess too much mass when under the rapidly
provides no statement of dynamic behavior. changing movements existing in velocity, acceleration,
In the dynamic mode, the designer must be aware of the and jerk conditions. The first step to designing for the
effect of friction in its various forms, of the effects that the dynamic mode in a machine linkage begins with studies
various kinds of slop between couplings produce: backlash, that provide the designer with maximum values of velocity,
dead zones, and other nonlinear operation. Where the best acceleration, and jerk. The kinematics of linkages is a
acceleration is needed from a motor driving load, the load topic well covered in texts on the design of mechanism,
inertia must be matched to that of the driving source. A see, for instance, Beggs (1955) and Mabie and Ocvirk
gearbox is often needed to obtain this match. The transient (1958). This leads to the stress levels that occur from
which the deflections can be calculated (although often
performance of electric motors is generally compared by the
with considerable difficulty, for the maximum deflection
magnitude of their ratios of torque2 /inertia. This is called
and dimensional change points are not always in obvious
the ‘power-rate’ (West and Leonard, 1955, and Rothwell,
positions). Inertia force considerations calculated from
1966). The same term can be applied to any form of
knowledge of the moments of sections of components,
motor (spring motors, gravity feeds, and other mechanical
coupled with the velocity and acceleration data, lead to
motors), in a similar manner (Sydenham, 1969), giving the other dynamic considerations.
designer a chance to assess the relative merits of various
mechanical or electrical drives. Where high acceleration is
concerned, the diameters of rotating parts should be kept as 4 MECHANICAL SYSTEMS MODELING
small as possible; rotational inertia increases rapidly with
an increase in diameter, sometimes as the fourth power of The above description of the static and dynamic aspects
the radius. The problem of matching mechanical loads to of instruments is not based upon an overall unified sys-
sources is identical to the problems existing in electrical tematic understanding of the topics but is typical of the ad
power transfer. hoc approach that many designers use. Each of the above
When resonance is to be invoked, or appears unwanted, observations can be better explained and calculated using a
the designer needs to know how to enhance or reduce it. common fundamental mathematical basis for dealing with
The role of damping must be understood. In some instances the performance of physical systems. The mathematical
the designer attempts to design for the highest Q conditions description of mechanical linear systems is the same as that
so that the system constantly ‘rings’ by noise excitation (a for electrical, thermal, and fluid regimes, see Article 59,
relevant example being one form of gravity wave detector Introduction to the Dynamic Regime of Measurement
Systems, Volume 1. By the use of appropriate coefficients,
that uses large freely supported blocks of aluminum as
the performance equations of one regime can be interfaced
detecting resonators). This provides increased sensitivity to
with those of another. Figure 2 is an example showing
signals at or near the resonant frequency. The dimensional
how the moving coil meter can be expressed in a com-
changes, transduced to electrical signals, are then inverse-
mon electrical regime, which can then be studied as an
filtered to remove all but the varying amplitude of the
electric circuit to obtain its dynamic behavior. This means
frequency at which the system is resonating. This variation that electromechanical systems can be represented by a sin-
contains the signal to be detected. In this case, damping is gle mathematical expression. The actual performance will
deliberately kept to a minimum. closely be as predicted from this model (in so far as the
In many applications, however, the aim is to introduce real situation matches up with the model used).
damping deliberately to produce a certain response to This understanding is represented in the systems
input excitation. This is done by adding damping to approach to physical system – see Finkelstein (1977) for
the resonant system with suitable mechanical, or other its use in measurement. This is not to be confused with the
regime, dissipation elements. Consideration of the means much more widely applied ‘general systems theory’ that
to reduce the effects of mechanical hysteresis suggests that attempts to take in far more experiences, including such
the addition of damping is not always the best approach topics as biological and sociological phenomena. The latter
to use, because vibration to rest can often improve the is similar in concept but has to operate without the same
static precision of such elements as bellows, springs, degree of mathematical rigor and formal representation that
and diaphragms. is present in the former.
518 Measuring System Design
Source
Pointer Rs
Restoring spring Inertia J
of compliance D
N Magnet
Coil of resistance R
and inductance L
Damping G
(a)
Inertia Damping
R L
Source
connection Restoring
(low) Rs CM G spring
LM
Electrical coil Mechanical parts
(b) Conversion expressions LM = D × b 2
(in this form of analogy) CM = J × b−2
where b is conversion factor
Figure 2. Unified representational model, made in the electrical regime, of electromechanical moving coil indicator; (a) physical
electromechanical system; (b) electrical model.
The systems approach for physical systems has been These tools, however, still require the user to have a well-
steadily developing from roots emanating from the early developed sense of the metrology aspect of the system
radio era when microphones and loudspeakers were first being developed.
being developed. Various papers were written on what were Mathematical models, see Article 71, Modeling Metho-
later seen to be constituent parts of this integrated topic. dology, Volume 2 are only as good as the numerical
Eventually the basics were assembled together as a single values that give the workable boundaries to the system
methodology in such books as Olson (1943), where the of equations used. These are invariably obtained from
‘analogies’ were presented as a systematic approach to the experimental procedures; thus, there will always be a
design of linear dynamical systems. Initially, the approach decision to make about whether it is more expedient to
was developed for the design of electromechanico-acoustic generate a mathematical model or to build the device and
devices. It gradually found its way into the description test it.
of electromagnetical machines in general; see Fitzgerald Models may also take the form of scaled physical ver-
and Kingsley (1961) and Koenig and Blackwell (1961). sions or analog simulations in another regime. For example,
The material was later presented as a method to cover the an application required the development of a small linear
dynamics of all physical system regimes, Shearer, Murphy vehicle for robotic research. Testing its linear movement
and Richardson (1971) being a well-presented and extensive performance to check out the circuitry, the braking, accel-
work; it also lists other books of relevance. eration, and control circuits was more conveniently done
Each of those above mentioned works was in general using the drive system to operate, as the final stage, a
written for the study of energy conversion devices. Instru- rotary equivalent of the linear mass to be driven. Digital
ments are a class of those specifically concerned with the simulation is another choice for modeling the system.
conversion of information into knowledge, that is, using A word of caution must be given, however, on the use of
an energy conversion transmission path to convey meaning models. It is all too easy to become used to the idea that the
between points in a system. The foregoing theory of physi- model that has taken so much hard work to develop is a true
cal systems can be condensed somewhat when instruments and entirely faithful representation of the final instrument.
are the prime interest. Models are rarely more than just adequate because the effort
Modern finite element CAD tools are now able to needed to improve them beyond a certain point is not seen
support sophisticated design development, moving from to be further productive in low production rate instruments.
basic element representations, stress and strain calculations, The user often decides to compromise model precision in
and graphs, onto CNC software files for manufacture. order to make progress.
Principles of Fine Mechanics – Systems Considerations 519
A model, although it may well simulate the correct Dynamic Regime of Measurement Systems, Volume 1;
input–output relationships, may not do this by the same Article 80, Principles of Fine Mechanics – Kinematic
internal mechanisms as the original system. The field is and Elastic Designs, Volume 2; Article 84, Kinematical
covered by the discipline of ‘systems identification’ and Regime – Rotary Motion, Volume 2.
‘simulation’. This point is demonstrated by thinking about
the fact that there are many technological ways to achieve
an end: fuel gauges in a car can be electronic or entirely REFERENCES
mechanical. Models must be so constructed that any internal
data that are desired are available to the user of the model. Beggs, J.S. (1955) Mechanism, McGraw-Hill, New York.
The united ‘analogies’ approach to instrumentation also Bell and Howell (1974) The Bell and Howell Pressure Transducer
shows that the performance of an instrument system can Handbook, CEC/Instruments Division, Pasadena, CA.
be tailored by the use of any appropriate discipline. In Bosman, D. (1978) The Systematic Design of Instrumentation
setting up an instrument design, the degree of mechanical Systems. Journal of Physics E: Scientific Instruments, 11,
proportion needs consideration in the light of what can 97–105.
be achieved using alternative electrical, optical, and fluid Braddick, H.J.J. (1954) The Physics of Experimental Method,
regimens. For example, several years ago the design of Chapman and Hall, London.
gauges for dimensional measurement generally required the Cambridge Instrument Company (1945) 50 Years of Scien-
designer to incorporate as much mechanical sensitivity as tific Instrument Manufacture, Engineering, 159, 361–363,
possible in order to obtain an adequate mechanical signal 401–403, 461–463, 501–502.
output. Today, the electronic alternative allows virtually Doebelin, E.O. (1975) Measurement Systems: Application and
unlimited dimensional sensitivity: fractions of nanometers Design, McGraw-Hill, New York.
can be routine (Sydenham, 1972). Mechanical stages no Finkelstein, L. (1977) Instrument Science – Introductory Article.
longer have that constraint imposed on them; the trading Journal of Physics E: Scientific Instruments, 10, 566–572.
off of sensitivity usually allows higher frequency response. Fitzgerald, A.E. and Kingsley, G. Jr. (1961) Electrical Machin-
A second example of the use of alternative regimes arose ery – The Dynamics and Statics of Electromechanical Energy
Conversion, McGraw-Hill, New York.
when a radar-tracking turret was assembled and found to
have an unexpected resonance in a drive shaft. It was more Foxboro (1972) Process Control Instrumentation, Pub.10SE,
Foxoboro-Yoxall, Surrey.
easily overcome by adding electronic compensation cir-
cuitry to the control systems than by replacing the shaft Koenig, H.E. and Blackwell, W.A. (1961) Electromechanical Sys-
tems Theory, McGraw-Hill, New York.
with a stiffer design. Instrument design is no longer the
matter for a single approach, nor does electronics always Mabie, H.H. and Ocvirk, F.W. (1958) Mechanisms and Dynamics
of Machinery, Wiley, New York.
offer the best solution.
Design is the procedure whereby a mass of general facts Olson, H.F. (1943) Dynamical Analogies, Van Nostrand, New
York.
are gradually sifted to sort out which combination will
achieve the specification. There is seldom a unique solution Rothwell, E. (1966) The Characteristics of the Printed Motor.
and the designer is constantly required to make choices, Control, 10(93), 136–139.
knowing that a whole range of other possibilities are then Shearer, J.L., Murphy, A.T. and Richardson, H.H. (1971) Intro-
omitted from future considerations. Finkelstein (1977) and duction to Systems Dynamics, Addison-Wesley, Reading, MA.
Bosman (1978) discuss this in relation to instruments. Sydenham, P.H. (1969) Manual and Automatic Large-Scale
Metrology, PhD thesis, University of Warwick, UK.
Sydenham, P.H. (1972) Microdisplacement Transducers. Journal
RELATED ARTICLES of Physics E: Scientific Instruments, 5, 721–733.
West, J.C. and Leonard, J.L. (1955) The Necessary Torque
Article 56, Transfer Characteristics of Instrument Requirements for a Servo-motor. Journal of Physics E: Sci-
Stages, Volume 1; Article 59, Introduction to the entific Instruments, 32, 30–32.
82: Kinematical Regime – Members and Linkages
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
also deform at the points of contact, but the degree of
1 General Information on Kinematic yield is generally smaller than the component they might
Components in Fine Mechanism 520 be supporting.
2 Members in Fine Mechanism 521 The various kinds of mechanism that might be employed
3 Columns and Shafts Collapse Strength 521 include such devices as members, columns, bearings, link-
age arrangements, wobble pins, gears, belts, cams, fas-
4 Second Moments for Common Members
Forms 521 teners, flexures, shaft couplings, and clutches. There are
very many of these, and in the past they have always
5 Use of Shafts and Pins in Fine Motions 521
been ordered into logical groups. Today, it seems from the
6 Linkages 524 published material, no satisfactory classification has been
7 Design of the Ideal Linkage 524 achieved. Authors still differ widely in the way that their
Related Articles 525 works are organized.
References 525 In this account, they are ordered to reflect the design
process: how the designer moves from one stage to the next,
gradually completing the detail of the system being created.
Kinematic and semikinematic mechanisms are divided
1 GENERAL INFORMATION ON into members, linkage arrangements of members, fastenings
KINEMATIC COMPONENTS IN FINE that join members rigidly, bearings that are used at the join-
MECHANISM ing points of linkages to allow motion, and the transference
of motion across a linking point where rotation is required
Mechanisms act as links, called pairs, that couple parts of at a given gearing ratio. This article addresses members
a system together. Although not a classification generally and linkages.
used, it is convenient (and logical) to divide mechanisms Texts that contain information on the various types of
into two groups: those based mainly on kinematic concepts mechanisms are from the time when there was consid-
and those that operate via elasticity. For our purposes, links erable interest in fine mechanism. It is apparent from
that are strictly semikinematic because of their bearing inspection of the interiors of such equipments as print-
area being greater than a point contact, including those ers and copy machines, that this design expertise still
which are manufactured to perfect enough tolerances, will resides somewhere, but not in recent texts. The useful
be placed in the kinematic class forming the first group. source books include Beggs (1955), Braddick (1954) for
Both types do use elasticity to spread the initial load but do which a 1966 edition exists, Faires (1955), Mabie and
not rely on compliance to allow subsequent deformations Ocvirk (1958), which was reprinted in 1975, Peatfield
to take place. For example, a bolted joint deforms as it is (1951), Doughtie (1966), Reliance Gear (1972 and subse-
tightened but does not flex to the extent as would an elastic quent years), and Trylinski (1971). The many publications
cantilever beam that it could be holding. Ball bearings of Geary (1955, 1961, 1962, 1964) may also be of use in
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Kinematical Regime – Members and Linkages 521
the general selection of mechanism. Beggs’ work contains buckling as it bends to form one of several possible modal
a ‘repertory of mechanism’ that is provided as a source shapes of a bent member.
of ideas. Of these books, Trylinski’s work is the most This form of collapse is quantified by use of Euler’s col-
theoretical general work that is specifically on instrument umn formula (Timoshenko and MacCullough, 1959). The
mechanism. The date of publication of works on mech- formula gives the value of load at which the fundamental
anism is not very significant, for mechanisms have been mode, for the given end connections, begins to appear.
known for centuries. Little has changed in the principles Euler’s point of buckling depends upon the end condi-
used. They have just become more sophisticated in imple- tions of constraint that are applied to the two ends of the
mentation. Very few new mechanisms, in fact, have been beam. These may be so-called pin-connected (that allows
invented in recent decades. A key paper by Furse (1981) is rotation at the end in any direction) or be firmly clamped
relevant to kinematic design. so that perpendicular rotation is not possible at the point
of connection. The first case is also referred to as simply
supported, the latter as built-in or fixed-ends. Another con-
2 MEMBERS IN FINE MECHANISM dition of Euler buckling can also occur when the end is
not constrained in translation, being free to move sideways
Here, the discussion centers on three-dimensional materials under the load.
that are fashioned to have significant length over radius An interesting feature about the two sets of key beam
of gyration ratios. Included are columns, beams, plates, formulae (Eulers’ buckling and deflection) is that although
and shafts. each has a different basic expression, within each group
A slender member can be used in tension with the domi- the expressions vary only by their numerical coefficient.
nant failure mechanism being when it yields beyond its elas- The stiffness of a column or beam increases as the ends
tic limit in tension. The way it is fastened to another mem- are made more constrained and the end point loading made
ber is another design issue to be allowed as a connection more central. Figure 1 shows the various cases of collapse
(see also Article 83, Kinematical Regime – Fasteners, that can occur.
Bearings, Volume 2). As a guide, Euler collapse begins to occur when the
The elasticity and the strength of a slender member slenderness ratio rises to reach around 100. The value must
are calculated from the usual strength expressions relating be calculated for the smallest radius of gyration when the
Young’s modulus of elasticity to compliance and the mate- member has an asymmetric cross section.
rial parameters that define the limits of tensile load. Com-
pliance remains the same for all conditions of a material;
tempering and other treatment does not alter the elastic- 4 SECOND MOMENTS FOR COMMON
ity, only the ultimate point at which yielding begins; where MEMBERS FORMS
Hookean elasticity ceases.
There is an exception to this rule; that is, when the mate- Figure 2 provides information about the moments of typical
rial is formed as an extremely slender fiber; the mechanical sections of structural member as they are then needed.
properties may change significantly. It is good to remember These formulae are needed for many mechanical member
that the end connections may need to be of a much larger designs. They assist selection of cross-sectional shapes for
section than the member to keep bending stresses low. In suitable column and elastic members.
instrument design, where dimensional change is important,
the compliance of apparently adequately strong members 5 USE OF SHAFTS AND PINS IN FINE
may be too high and they need to be made well over the MOTIONS
strength needed. Calculations should be made to verify that
it is adequate from the viewpoint of its compliance. Rotating shafts also have a performance failure that can
be likened to the column collapse characteristic. A shaft
that is free to turn in bearings that do not constrain its
3 COLUMNS AND SHAFTS COLLAPSE rotation at the bearings will reach a critical speed at which
STRENGTH it begins to whirl – the central region attempts to move
outwards owing to centrifugal forces. It takes up a bent
In tensile use of a member, the longitudinal forces on it act profile while it rotates. This, too, depends upon the elasticity
to stretch the member; in this case the force system is stable. of the material and the slenderness ratio of the shaft. The
If, however, the force is reversed so that it now compresses expression giving the lowest whirling speed is given in
the slender member, the first failure mode to appear will Figure 1. Attachment of loads along the shaft alters this
not be compressive yield but collapse due to the column value (Faires, 1955; Mabie and Ocvirk, 1958).
522 Measuring System Design
Load P
Hinged and free
to move sideways Columns
Deformed
shape K = 0.25 General expression for least strength mode
of supportable maximum load
P critical = K · π E
2 I
P I2
Hinged but no
translation Where P cr initial load at point
of collapse
K = 1.0
E elasticity modulus
I least second moment
of cross section
P l length between ends
K = 2.0
Shafts
K = 4.0
Lowest speed at which whirl begins is
N = π EI
l 2l 2 w
Where E elasticity modulus I least second moment
Rotations Whirl shape of cross section
w mass per unit length of shaft
not constrained
of bearings l distance between bearings
Figure 1. Columns and shafts – collapse point.
Slender members might be used as shafts coupling each well developed for the various free- and clamped-end
motion, or as parts of a linkage to produce a given motion. connections.
They can also form wobble pins that transfer linear motion The formulation extends to cover the two-dimensional
without forcing any constraint on the freedom of translation plate having freely supported and built-in edges to which is
that occurs at the end of the members forming either side applied a central point load or a total area distributed load.
of the drive. These latter cases cover the needs of most diaphragms in
Wobble pins, Figure 3, can only be used to transmit pressure gauges. Intentionally designed deflection of shaped
compressive force, since the compressive force keeps the diaphragms is discussed in Trylinski (1971).
point contacts mating. Ends are usually formed to be No difficulty should be experienced in obtaining the var-
small radii hemispheres. Another use for slender mem- ious formulations. In the case of very complex loading,
bers is where thin legs are used to support a tiltmeter it may be necessary to develop the expression from the
while allowing freedom of expansion of its body. If in ten- basic approaches that are used to obtain the quoted equa-
sion, drive can be transmitted by relatively thin wires or tions; see Timoshenko and MacCullough (1959) for a full
rods, for in that situation the Euler’s collapse mode can- treatment of the equations of beams and other compo-
not occur. nents. Texts containing tables giving the common expres-
Slender members (often called beams) used in the hor- sions are Braddick (1954), Faires (1955), Neubert (1975),
izontal direction to support loads are the usual subject and Timoshenko and MacCullough (1959).
of texts on the strength of materials; origins are largely Beams that are curved and have to carry linear ten-
for civil engineering structures analysis. The various cases sile or shear loads, such as a hook, will experience high
of point load at a given position along the beam, a bending stresses in the region of the bend. Areas around
uniformly distributed load or of a mixture of loads are rivet and pin connections can be unexpectedly weak near
Kinematical Regime – Members and Linkages 523
Moment of inertia Radius of gyration
xy b bd 3 bd 3 d h
axis Ix = Ix = kx = kx =
12 3 12 3
x-axis d Rectangle
b 2d 2
x-axis Ixy =
4
r πr 4 r
x-axis Circle Ix = kx =
4 2
r1 π(r14 − r24)
r2
Tube I=
4
2b πbd 3 πb 3d d b
Ix = Iy = kx = ky =
x-axis 2d Ellipse 4 4 2 2
y-axis
x-axis d a l-section [bd 3 − (b − c)a 3)
Ix =
12
c
x-axis Semicircle πr 4 r kx = 0.264r
r y Ix = Ix = 0.11r 4 kx =
x-axis 8 2
4r
(where y = )
3π
bd 3 bd 3 h h
x-axis Ix = Ix = kx = kx =
y d Any Triangle 12 36 6 18
x-axis d
b (where y = )
3
x-axis Quarter circle πr 4 r kx = 0.264r
r Ix = Ix = 0.055r 4 kx =
x-axis 16 2
x-axis b Box section
d D BD 3 − bd 3 BD 3 − bd 3
Ix = kx =
12 12(BD − bd)
B
x-axis d D Two beams b (D 3 − d 3) D3 − d3
Ix = kx =
12 12(D − d)
b
In general I = Ak 2 Where A is area of cross section
and k is radius of gyration
Figure 2. Second moments of commonly met sections (composites can be built up by addition of these sections).
the joint if care is not taken to increase the section short in length to increase the frequency of resonance
size. when they transfer rotary motion in closed-loop sys-
Beams find use in instruments as structural members tems.
and as sensing members; they may, therefore, need to be The torsional elasticity of a shaft is calculated from the
appropriately inelastic or elastic. Successful design rests on torsional or shear modulus for the material, the process
choice of material, design of the member shape and size, of calculation being analogous to that for analyzing linear
and the connections used. elasticity of a member.
Shafts also can be made elastic or rigid depending Dynamic balancing of rotating parts is needed where
upon the need. They should be large in section and vibration levels must be kept low.
524 Measuring System Design
Exact alignment
of centers difficult to Micro-displacement electromagnetic
procure sensor with ferrite armature supported
by thin diaphragms
10 m quartz Quartz
tube wobble pin
Perforated diaphragm providing
some lateral constraint
Figure 3. A wobble pin can be used to couple a moving member to a more constrained point. It will buckle if too slender for the
compression force (Sydenham, 1974).
6 LINKAGES
P, moves in
straight line
Members connected together can be used to provide con-
version of motion of various kinds – from linear to rotary,
from linear to linear, and from rotary to rotary. The range (a)
of possible linkage functions is vast.
Linkages, Figure 4, may form a mechanism for geomet- Position, along arms of
equal masses a,b does
rical size conversion such as in the pantograph, or as the not alter balance
method by which a balance scale can be made insensi-
tive to the position of the load in the scale pan (seen in a
the so-called static enigma by Roberval in 1669, which b
appears to contravene the laws of the lever). Linkages have
found extensive use in the classic typewriter, pneumatic (b)
controllers, weighing scales, in mathematical calculating
instruments, in recorder pen drives, and numerous other
applications.
The method of analysis of linkages can also be used
to study the action of solid components, such as cams;
the instantaneous contact points form geometry that can
be considered to form hypothetical members. Linkages (c)
become structures under certain conditions of constraint.
Plate moves parallel and equidistant to lower plate
7 DESIGN OF THE IDEAL LINKAGE
The mathematical analysis of linkages has fascinated aca-
demics for centuries. Many of the defined fundamental
problems are still to be solved in theory. The rise and
(d)
development of the kinematics of linkage is described
in Ferguson (1962). To a large extent, analysis, the method Figure 4. Some linkages used in instruments: (a) Peaucellier’s
of breaking down the action of a given linkage and motions straight-line generator one form; (b) Roberval’s static enigma;
of its members, has succumbed to methodology that can (c) Beranger’s use of Roberval’s static enigma in a scale;
be used to ascertain the positional motion of linkages of (d) Sylvester–Kempe parallel translating linkage.
reasonably simple nature. As the number of members in
the linkage rises, the problem becomes more difficult. The As there does not seem to be a simple general theory that
four-bar linkage begins to present a significant theoreti- can be applied to decide the motion at the various parts of
cal challenge. a linkage system, the problem is often tackled by the use
Kinematical Regime – Members and Linkages 525
of graphical procedures in which the motion is generated invented during the history of man’s involvement with tech-
in incremental steps. To assist the designer, several authors nology. Pollard (1929), on the design of couplings, also
over the centuries have published hard copy catalogs that might help understanding of the procedures.
show the motion of variously constrained linkages. Today,
appropriate CAD tools can carry out this function.
What designers would rather have is the means to system- RELATED ARTICLES
atically synthesize – rather than analyze – a linkage system,
in order to obtain a required motional relationship. This Article 56, Transfer Characteristics of Instrument
is, however, still in development; only relatively simple Stages, Volume 1; Article 57, Static Considerations
cases can be handled. A computer solution of the equa- of General Instrumentation, Volume 1; Article 70,
tions involved has extended the use of synthesis methods. Measuring System Design Methodologies, Volume 2;
Usually, the designer resorts to the use of models (in the Article 80, Principles of Fine Mechanics – Kinematic
computer or real), makes use of the above-mentioned cata- and Elastic Designs, Volume 2; Article 81, Principles
of Fine Mechanics – Systems Considerations, Volume 2;
logs, or iterates toward a suitable linkage arrangement using
Article 88, Error Sources in Fine Mechanics, Volume 2.
analysis of each choice.
It is sobering to realize that the theoreticians sought for
many years, the theory by which a perfect linear action REFERENCES
can be produced by some point in a system of rotating
links. Unexpectedly, out of this and other similar inter- Beggs, J.S. (1955) Mechanism, McGraw-Hill, New York.
ests has emerged useful mathematics for other disciplines Braddick, H.J.J. (1954) Mechanical Design of Laboratory Appa-
(such as the Chebyshev functions now used in electronic ratus, Chapman & Hall, London (1st edn, 1954).
filter design, but first compiled for steam engine linkage Burstall, A.F. (1970) A History of Mechanical Engineering, Faber
behavior). This linkage was finally produced by Peaucellier and Faber, London.
(Figure 4a) in the 1870s. Doughtie, V.L. (1966) Elements of Mechanism, Wiley, London.
Many useful systems for instrument work have Faires, V.M. (1955) Design of Machine Elements, Macmillan
been invented. The ‘static-enigma’ balance mechanism Publishing, New York (1st edn, 1934).
made simple-to-use, accurate, top access scales feasible Ferguson, E.S. (1962) Kinematics of Mechanisms from the Time
(Figure 4b, c). of Watt, United States National Bulletin 228, Paper. 27, Smith-
The ‘Sylvester–Kempe translating linkage’ (Figure 4d) sonian Institution, Washington, DC, (pp. 185–230).
allows one platform to move parallel and equidistant from Furse, J.E. (1981) Kinematic Design of Fine Mechanisms in
a second surface. Instruments. Journal of Physics E: Scientific Instruments, 14,
264–272.
Johnson’s ‘converting motion’ provides two revolutions
from a single reciprocating translation. Geary, P.J. (1955) Knife Edge Bearings, Sira Institute, Chisle-
hurst.
Other systems related directly to linkages are the Geneva
stepping mechanism, the Scotch yoke, the Oldham cou- Geary, P.J. (1961) Instrument Ball Bearings, Sira Institute,
Chislehurst.
pling, and many more.
Geary, P.J. (1962) Fluid Film Bearings, Sira Institute, Chislehurst.
Ability to analyze a linkage might be necessary to estab-
lish the clearances needed as the members move, to assess Geary, P.J. (1964) Magnetic and Electric Suspensions, Sira Insti-
tute, Chislehurst.
the velocities and accelerations of the various parts and
points, to predict the input–output functions, and to study Lent, D. (1970) Analysis and Design of Mechanisms, Prentice
Hall, Englewood Cliffs, CO.
the action of systems like cams and gears.
Mabie, H.H. and Ocvirk, F.W. (1958) Mechanisms and Dynamics
The methods of analysis of linkages are discussed
of Machinery, Wiley, New York (3rd edn, 1975).
in Beggs (1955), Faires (1955), Mabie and Ocvirk (1958),
Neubert, H.K.P. (1975) Instrument Transducers – An Introduction
Zimmerman (1962), and Lent (1970). Trylinski (1971) uses to their Performance and Design, Clarendon Press, Oxford.
the procedures in his discussions of the kinematics of var-
Peatfield, A.E. (1951) Engineering Components and Materials,
ious mechanisms used in fine mechanics. Ferguson (1962) EUP, London.
includes several references to linkage theory. The jour-
Pollard, A.F.C. (1929) The Kinematical Design of Couplings in
nal Precision Engineering (first issue, January 1979) is Instrument Mechanisms, Adam Hilger, London.
a source of papers. Burstall (1970) discusses the history Reliance Gear. (1972) Catalogue RG5, Reliance Gear Co Ltd,
of linkage development giving several references to early Huddersfield. (Also similar catalogues by PIC Design Corp,
works. This book contains illustrations of linkages that were New York).
526 Measuring System Design
Sydenham, P.H. (1974) 2000 hr Comparison of 10 m Quartz-tube Trylinski, W. (1971) Fine Mechanisms and Precision Instru-
and Quartz-catenary Tidal Strainmeters. Geophysical Journal ments – Principles of Design, Pergamon Press, Oxford. Trans-
of the Royal Astronomical Society, 38, 377–387. lated by A. Voellnagel from Polish original.
Timoshenko, S. and MacCullough, G.H. (1959) Elements of Zimmerman, J.R. (1962) Elementary Kinematics of Mechanism,
Strength of Materials, Van Nostrand, NJ. (Numerous editions Wiley, London.
and printings exist since the 1st in 1935).
83: Kinematical Regime – Fasteners, Bearings
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
whole unit from a single piece of material. Examples are
1 Fastenings in Fine Mechanism 527 the manufacture of flexure pivots in which the flexure part
2 Permanent Joints 527 is formed by thinning down the piece of material. There
is, however, very little substantive evidence to prove that
3 Detachable Joints 528
this is necessary. To the contrary, it has been found that
4 Instrument Bearings and Pivots 529
remarkably crude joints, such as knotted metal wires, rings
5 Reduced Friction Pivots 530 placed over a shaft and in tension, small magnets holding a
6 Surveys of Instrument Bearing Information 531 part to a surface, and bolted joints (Jeffery and Sydenham,
Related Articles 532 1973), will each hold dimensional relationships across the
References 532 joint to less than micrometer level. Couplings working at
the very small tolerances of nanometers or less may well
exhibit instability, but it is far from easy to verify where
such a drift actually occurs. Errors of this magnitude can
1 FASTENINGS IN FINE MECHANISM come from so many sources. As an example of the many
forms of spring-gravimeter that have been made, units using
The purpose of fastenings is to join components together in bolted metal arrangements have shown as small a drift rate
a manner that will provide a firm constraint of all degrees
as others with mostly continuous parts.
of freedom at the joint. Available designs vary greatly, the
Joints forming part of a mechanical resonance loop
choice largely depending upon the magnitude of the forces
can, however, seriously alter the Q-factor because of the
and torques involved, whether the joint is permanent or
damping inherent in the jointed faces. Joints should be
needs to be detachable, the degree of vibration experienced,
avoided where high-quality factors are needed.
and the level of dimensional accuracy needed. The design
In instruments, the joints must provide adequate strength
of joints is as important as the other factors of a fine
but also they must usually be stiff in compliance. Because
mechanical system.
of the small sizes of the metal work in instruments, many
An introductory, nonmathematical catalog of fastenings
ingenious joints can be used that would not be practical for
can be found in Peatfield (1951). A thorough treatment of
large-size machine construction. Joints divide into perma-
many of the possible types is given in Trylinski (1971),
nent and detachable methods.
this being a key work of reference. Machine-design texts
such as Faires (1955), Lent (1970), Shigley (1988), Erdman,
Sandor and Kota (2001), and Juvinall and Marshek (2002)
generally treat only the joints used in power machines, that 2 PERMANENT JOINTS
is, bolted, riveted, and welded arrangements.
In some applications, the dimensional stability of the Figure 1 illustrates examples of some methods of making
joint can be very important. One view suggests that in a permanent joint. A commonly used permanent joint is
such cases, it is better to avoid using joints, making the formed by passing the metal of one part through a hole
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
528 Measuring System Design
Tang twisted or adhesives in instrument construction. Work at the former
bent over after Sira Institute in England has resulted in several publications
insertion
on the use of adhesives, Shields (1976) being a substantial
account. The slow setting rate of cements and adhesives
can make them unsuitable for mass production methods.
In some cases, using cements, those substances that fill a
gap without stressing the parts that are held together, may
be the most suitable method of securing components, for
(a) example, holding a lens in a holder.
When using glued or cemented joints, it is necessary to
ascertain whether the joint will have adequate heat trans-
End swaged to fer or electrical conduction across the joining layer that
hold shaft in is interposed between the parts. Cements and adhesives
hole
(b) are often hygroscopic; they may also creep in dimension.
In some instances, however, gluing may be the only sat-
isfactory method of making a fastening. A good design
Unsuitable loading attempts to form a fillet of glue over a large contact area.
directions
The glue should not be used to form a dimensional link but
Solder
should be used to adhere two creep-free materials in close
contact. Trylinski (1971) contains a large bibliography on
Suitable Loading
the subject.
(c)
Loading in any direction can
be used 3 DETACHABLE JOINTS
Spot- or
seam-welded zones Figure 2 shows some examples of detachable methods of
(d) jointing. These can be made using screw threads, which
hold flanges together or which are formed on the end of
Shaft held in disc by rods to be jointed; using pins holding a shaft in a hold
center piece moulded
after disc and shaft (tapered, parallel, roll-pin, shear-pin); using keys that hold
are made a shaft in a hole; and as press-fit combinations. Where the
jointing forces are low, such as the holding of a dust cap
(e) over a socket, the use of the lids and caps held by simple
turn-lock devices can be used. Trylinski also covers these
Figure 1. Some permanent fastening methods; examples of each in detail.
type: (a) bent tangs; (b) swaged after insertion; (c) soldered or
glued; (d) resistance welded; (e) molded in place. When bolts are used to hold a flange, the designer has the
choice of either making the bolt hole with close-fit tolerance
to ensure the correct dimensional relationship or using some
in the other and by swaging or twisting it to lock the two other method of location, using the bolt only as a clamping
parts together. The tang can be twisted, bent over, swaged device. Locating methods in use are spigots or taper and
flat, center punched, or crimped. Trylinski (1971) gives parallel cylindrical pins. It is not good practice to use a
illustrations of many types. large clearance hole in which the actual position is set up on
Force-fit joints that rely on friction and a certain degree of assembly. This may suffice for a one-off prototype, but in
cold welding as the parts are forced into unison can also be conditions of vibration or where service personnel may not
used. For these, the tolerances of the mating parts must be understand what is needed, the position can be lost. Where
carefully controlled. These give good dimensional control. positional accuracy is vital, the joint can be drilled and
Welded joints are often used, the smallest possible area pinned after assembly. A spigot is used where the location
of welding being at micrometer size. Fine pressure capsules is predetermined. Where the bolt is used to retain position,
are often welded using focused laser radiation. Gas or its shaft should define the location, not the thread diameter.
electric welding is also used. Distortion due to the heating To ensure that this is done, the thread is made with a smaller
must be allowed for. Not all materials can be welded. outer diameter than the shaft next to the head of the bolt.
Adhesives can be used to form permanent joints, epoxy Screw threads are used extensively and numerous forms
formulations having considerably changed the role of of thread profile and diameter are found in applications. The
Kinematical Regime – Fasteners, Bearings 529
Press fit, shoulder assists
location of gear blank at
right angles and in correct Disc must
position not reach
shoulder
Nut must exert force
Taper
onto disc via suitable
washer that clears
taper
(a) (b)
Clearance in Bolt to provide clamping
hole force
Pins to provide
location
(c)
Shoulder provides Clearance to allow bolt
location to hold plate
(d)
Spring type
(e) (f) retainer
Figure 2. Some detachable fastening methods – examples of each type: (a) press fit; (b) taper shaft; (c) pinned; (d) screw thread;
(e) clamp-hub; (f) retainer ring.
designer of a production instrument should strive to use a Selection of bearings begins by a definition of the role
standard thread where possible. Different disciplines have that the material support has to play. All bearings are
their own threads – cameras and microscopes being exam- imperfect and will add some measure of freedom at the
ples. Multiple start, special thread profile, taper threads, pivot-bearing point. System sketches realize the degrees of
and others are sometimes used in instruments. Threads can freedom that are needed; study of the system’s operation
be locked with some form of lock or tab washer or can decides the tolerances that are permissible. Other factors
be pinned. Where the joint must be secured against unau- to be considered will be the relative speed of the two sur-
thorized opening, the shaft outside of the nut is drilled faces involved, the kind of lubrication that can be used, the
to take a small wire that can be sealed. Special propri- coefficient of friction, the life needed, assembly require-
etary fluids are available to lock threads; paint is sometimes ments, cost and market availability, effect of temperature
used. Here again, Trylinski (1971) is a definitive reference changes, and the degree of stiffness needed from the joint
work. formed in the bearing. It is quite possible to produce a
usable bearing (using preloaded rolling balls) that is as stiff
and as backlash-free across the joint as is the rest of the
4 INSTRUMENT BEARINGS AND PIVOTS structure.
Supports that allow movement by virtue of elasticity
Joints between two parts that allow motion, such as rotation of a flexure are discussed in Article 85, Elastic Regime
or sliding in translation, are generally called bearings. of Design – Design Principles, Volume 2; Article 86,
Trylinski calls them supports. They are also referred to as Elastic Regime of Design – Spring Systems, Volume 2;
pivots. and Article 87, Elastic Regime of Design – Plates and
The relative motion may combine both rotation and Bimorphs, Volume 2.
translation through the bearing axis. They may be included Bearings for rotating shafts have been in use since the
either to limit end-float in a thrust bearing or to allow an arm earliest times. Rawhide was used by the Egyptians around
to turn about a fixed shaft pin. Figure 3 shows a collection 1500 B.C. Bronze bushes were used in wagons of Roman
of bearings that can be used in instruments. times. Crude ball bearings date to the same period. Caligula
530 Measuring System Design
Oil-impregnated Thrust
bearing material
Shaft
Preload may be used,
some forms are designed
with this in mind
Thrust
(a) force (b) (c)
Adjustable to
change end-float
and to assemble Shaft
Outer shell formed
over to hold needles
(d) (e) when shaft is not in place
(f) (g)
Frame
n s s n
‘Safety’ locator bearing
Outer cylindrical
with pin; not in contact
magnet s n n s normally
Inner magnet
Shaft being
(h) supported
Figure 3. Selection of passive bearings used in instruments: (a) plain bush; (b) radial ball race; (c) ball thrust race; (d) pivot type ball;
(e) needle roller; (f) jewel with thrust; (g) center support; (h) magnetic.
had a ship with turntables running on large bronze balls. The constraining moments of the bearing have less effect
These early ball systems were not strictly freerunning, for on the output in this arrangement. A number of well-known
they were constrained so that the balls also slid while instruments of that time used this bearing – Atwood’s
turning; Burstall (1970) gives more detail. falling machine, Michell’s and Nairne’s dipneedles being
Around 1770, Count Carburi used a system of metal examples (Turner, 1973; Sydenham, 1979).
tracks and balls to move a 1.5 × 106 kg granite mass across
the frozen Russian land (Turner, 1973). The tracks were laid
end to end as the statue was moved.
Four identical wheels
in bearings
Gravity force
nests shaft into
5 REDUCED FRICTION PIVOTS wheels
Friction was recognized as the limiting phenomenon in Shaft
many an elegant mechanical apparatus of the early times.
In those times, ball races, as we know them today, could
not be made to the degree of accuracy and quality control
that was needed. Many methods were used to minimize the
effects of friction in bearings. A particularly interesting one Magnetic dip
is the so-called frictionless bearing arrangement attributed needle
to Sully (1680–1728). This used a system of wheels and Plan Side elevation
shafts running on the periphery of overlapping support Figure 4. The Sully (c 1700) frictionless bearing arrangement
wheels held in the prime bearings (Figure 4) effectively as used by Michell in a late eighteenth century dip circle
to increase the torque that is placed onto the bearing areas. instrument.
Kinematical Regime – Fasteners, Bearings 531
6 SURVEYS OF INSTRUMENT BEARING Geary’s fourth report of the series on instrument parts
INFORMATION (Geary, 1961) covered instrument ball bearings in a similar
way, providing a specific account on the small series of
The design of bearings is a very well developed aspect precision-quality ball bearings. It is a key document where
of instrument construction. Considerable theory now exists information on the use and choice of small ball bearings is
to complement the highly developed fine mechanical pro- required. Harris (1966) is also relevant.
duction capability. Low-friction precision bearings can be Applications requiring zero clearance and a large mea-
as small as a millimeter in outside diameter. The designer sure of stiffness across the ball-bearing joint can use spe-
who wishes to be sure that his or her choice is justified cially designed ball races – the angular-contact series. In
and is within the requirements of the task has a wealth of use, the shaft and housing dimensions are chosen to force
theory and experimental data to fall back on. the balls to ride against one side of the race. This is known
Manufacturers provide considerable design material to as preload. Ball nuts, those drive nuts in which balls run in
assist application. Permissible load, run-out, clearances, contact with the lead screw thread, are often used together
speed, friction force, life, and assembly arrangements are in pairs, one being rotated against the other to preload the
all well documented. nut into the shaft. Ground leadscrew shafts can provide
Since the various forms of bearings are generally famil- remarkably stiff, yet resistance-free, joints.
iar, this account will only introduce each kind, giving For nonmagnetic applications, beryllium–copper ball
references to where in-depth information can be found. An races can be used. Stainless-steel bearings are also sold.
excellent work on each type is Trylinski (1971). It provides Bearings capable of 300 000 rev min−1 are used in den-
information about the small and miniature-sized bearings tist’s drills.
that is not generally given in machine-design texts. It cov- Each application dictates careful choice from within
ers jewel bearings; direct in the base bearings used in clock the various classes of bearings if best operation is to be
mechanisms; conical and tapered supports, which use a achieved. The fifth report (Geary, 1962) covered fluid-
tapered shaft end running in a hole in a suitable bush; spher- film bearings. This provides a general introduction to this
ical ends; hydrostatic designs; center supports in which a bearing type and gives details of liquid-lubricated bearings
spherical shaft end runs against a concave plate; cover-plate using external pressure generation, of gas-lubricated bear-
type thrust bearings, including the shock-resistant Incabloc ings, and of less well known bearings such as fluid-squeeze
arrangement; mercury supports; knife-edges; rolling-contact films, mercury globule, and acoustic radiation pressure
bearings, such as the very easy to use cup-and-cone minia- supports.
ture and subminiature ball bearings; and wire track bearings The frictional force of the different kinds of bearings
in which the rolling contact ball is used to form its mating varies considerably, ball-bearing systems exhibiting consid-
track by being pressed into soft metal wires. The chapter in erably less frictional torque levels than simple liquid-fluid
his text on ‘supports’ covers friction and bearing area stress bushes. As an indication of the magnitudes involved, oil-
calculation, errors in location, and many other theoretically filled gyroscope ball bearings having five balls running
based explanations of behavior. Eighty-two references are at 4-mm diameter might exhibit a breakaway torque of
included. A further chapter on guides covers the design of 8 × 10−6 Nm. Running torque is generally about 30 to 50%
sliding arrangements, which are usually of rectilinear form. less. Very small, three-ball pivot bearings breakaway at
Further in-depth works on instrument bearings include 1.0 × 10−7 Nm torque. Torque figures increase with load
the series compiled by Geary of the Sira Institute (then on the pivot. In contrast, a small plain bronze bush needs
called the British Scientific Instrument Research Associa- about 6 to 8 × 10−4 Nm to move from rest. A small radial
tion, BSIRA). His second booklet ( Geary, 1955) was a ball race will need around 3 × 10−5 Nm torque before it
bibliographic survey of knife-edge bearings. It gives an begins to rotate. These figures can only be used as a guide –
illustrated technical introduction, followed by 166 refer- the actual value varies greatly with bearing form, preload,
ences. Geary concluded then that knife-edges should be shear load, and shaft size. They must be calculated or mea-
considered as blunt or rounded edges, for, in practice, defor- sured for the support to be actually used. Misalignment
mation occurs, resulting in a less than ideal knife-edge. He can introduce considerable frictional loss, unless designed
also wrote that there was a general dearth of research and as a self-aligning arrangement. Suspensions for electrical
objective data available, especially in the English language, indicating instruments use knife-edges, jewels, bushes, and
and that the best-written works were of German origin. He elastic bands. Drysdale and Jolley (1924), Part 1, include
suggested that some of these works should be translated. an extensive discussion of these, most of the material given
(Trylinski’s work was later translated, helping to fill this being about jewel bearings. Geary (1964) covers magnetic
gap in available knowledge.) and electric suspensions. Permanent magnetic suspensions
532 Measuring System Design
for instrumental use were surveyed by Fremery and Doden Geary, P.J. (1955) Knife Edge Bearings, Sira Institute, Chisle-
(1978). hurst.
Geary, P.J. (1961) Instrument Ball Bearings, Sira Institute,
Chislehurst.
RELATED ARTICLES Geary, P.J. (1962) Fluid Film Bearings, Sira Institute, Chislehurst.
Geary, P.J. (1964) Magnetic and Electric Suspensions, Sira Insti-
Article 57, Static Considerations of General Instru- tute, Chislehurst.
mentation, Volume 1; Article 58, Description of Accu-
Harris, T.A. (1966) Roller Bearings Analysis, Wiley, New
racy, Linearity, and Drift, Volume 1; Article 59, Intro- York.
duction to the Dynamic Regime of Measurement
Jeffery, G.J. and Sydenham, P.H. (1973) Stability of Strain-
Systems, Volume 1; Article 80, Principles of Fine meter Mounts. Geophysical Journal of the Royal Astronomical
Mechanics – Kinematic and Elastic Designs, Volume 2; Society, 32(2), 185–193.
Article 81, Principles of Fine Mechanics – Systems Con- Juvinall, R.C. and Marshek, K.M. (2002) Fundamentals of
siderations, Volume 2; Article 85, Elastic Regime of Machine Component Design, Wiley, New York.
Design – Design Principles, Volume 2. Lent, D. (1970) Analysis and Design of Mechanisms, Prentice
Hall, Englewood Cliffs, CO.
REFERENCES Peatfield, A.E. (1951) Engineering Components and Materials,
EUP, London.
Burstall, A.F. (1970) A History of Mechanical Engineering, Faber Shields, J. (1976) Adhesive Handbook, Newnes-Butterworth,
and Faber, London. London.
Drysdale, C.V. and Jolley, A.C. (1924) Electrical Measuring Shigley, J.E. (1988) Mechanical Engineering Design (McGraw-
Instruments – Part 1 Commercial and Indicating Instruments, Hill Series in Mechanical Engineering), McGraw-Hill.
Ernest Benn, London. Sydenham, P.H. (1979) Measuring Instruments – Tools of Knowl-
Erdman, A.G., Sandor, G.N. and Kota, S. (2001) Mechanism edge and Control, Peter Peregrinus, London.
Design: Analysis and Synthesis, Prentice Hall. Trylinski, W. (1971) Fine Mechanisms and Precision Instru-
Faires, V.M. (1955) Design of Machine Elements, Macmillan Co, ments – Principles of Design, Pergamon Press, Oxford. Trans-
New York (1st edn, 1934). lated by A. Voellnagel from Polish original.
Fremery, J.K. and Doden, K. (1978) Active Permanent Magnetic Turner, G.L.E. (1973) Van Marum’s Scientific Instruments in
Suspensions for scientific Instruments. Journal of Physics E: Teyler’s Museum – Descriptive Catalogue, Noordhoff Interna-
Scientific Instruments, 11(2), 106–113. tional Publishing, Leyden.
84: Kinematical Regime – Rotary Motion
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
regarded in that light. It is important to see coupling sys-
1 Transference of Rotary Motion in Fine tems in this way, for there are many traps to fall into
Mechanism 533 because of imperfections in components and mechanism
2 Couplings 533 concepts. Covered here are the shaft coupling devices,
3 Gears 534 gears of various kinds, cams, flexible belt-type drives,
and electromechanical gearboxes. The general principles
4 Cams 536
of mechanical design are to be found in Article 80,
5 Flexible Connecting Drives 537
Principles of Fine Mechanics – Kinematic and Elastic
6 Belt Drives 537 Designs, Volume 2. The use of kinematic design methods
7 Chain Drives 537 for members and linkages is covered in Article 82, Kine-
8 Electromechanical Gearing 538 matical Regime – Members and Linkages, Volume 2
Related Articles 539 and for fasteners and bearings in Article 83, Kinemati-
References 539 cal Regime – Fasteners, Bearings, Volume 2. The design
method using the elasticity of materials to obtain con-
trolled motion is covered in Article 85, Elastic Regime
of Design – Design Principles, Volume 2 and Article 86,
1 TRANSFERENCE OF ROTARY MOTION Elastic Regime of Design – Spring Systems, Volume 2.
IN FINE MECHANISM Errors of all kinds is the subject of Article 88, Error
Sources in Fine Mechanics, Volume 2.
As in electrical systems, mechanical information machine The form and design of these components are well cov-
counterparts require means to transfer signals between sub- ered in the general machine design texts such as: Peatfield
units of the system. These signals may have to be transmit- (1951), Faires (1955), Beggs (1955), Mabie and Ocvirk
ted at the direct unity ratio or be operated upon to alter this (1958) and Lent (1970). Information specifically relating
ratio up or down. The motion may need to be converted to instruments is given in Trylinski (1971) and Reliance
into such motions as intermittent, quick return, and others. Gear (1972). Because of the extensive use of gears and
Transferring signals and forces over large distances in
other couplings in industry, there are many national stan-
mechanical equipment is more difficult than with electri-
dard specifications.
cal signals. Generally, in practice, we convert mechani-
cal information into the electronic form for long-distance
transmission. This was not always so, and there are still
some applications where direct mechanical transfer is more 2 COUPLINGS
secure, more reliable, and more easily serviced.
This article discusses those components that, while not When coupling two shafts together as a direct drive, the
generally being thought of as computing elements, nev- factors to be considered are the amount of misalignment
ertheless, might be used in that manner and should be that must be tolerated, the constancy of the velocity and
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
534 Measuring System Design
positional form transmitted within a revolution, the torque A solid or rigid coupling can only be used where the
level, the distance between supports to reduce shaft whirl, two shafts are aligned to exacting tolerances. If they are
damping of vibration, stiffness needed in closed-loop con- not, the coupling will ‘work’ on its screws or bend the
trol, allowance for end-float and more severe longitudinal shafts, causing looseness or failure to occur.
movement, ease of assembly, and such factors as avail- Coupling drives come in many forms; see Reliance Gear
ability and cost. A sample of the range is shown in (1972) and Trylinski (1971) for illustrations of the various
Figure 1. kinds of instruments. The latter definitive work presents the
Constant velocity couplings began with the design of geometrical theory for many designs. The following names
Hooke reported in 1674. The single Hooke joint, however, are used in the literature for various kinds of couplings; slip
does not give a constant velocity (or position) ratio between joint and slip couplings, membrane, spring and self-locking
the input and the output as the coupling rotates through a spring, antibacklash, bellows, molded plastic flexible, uni-
turn. However, two units can be used together in series versal lateral, Kudriavetz, Flex-thane multiple polyurethane
so that the modulation imposed by one is canceled by the pins, multijaw, sleeve, slip clutch, overrunning clutch, and
other. Other more compact forms of constant velocity joints phase adjust (a type where the relative shaft phase can be
were later developed for the automotive industry, examples slightly adjusted by a small screw rotation). There is no
being the Bendix–Weiss, Rzeppa, and Tracta joints. These, shortage of design possibilities; each has advantages and
in general, have not been produced at instrument size. They disadvantages and must be matched to the application. It is
all have the advantage of providing constant velocity from important to appreciate the motional and stress characteris-
a single coupling joint. The above coupling stages require tics of each in any given application.
the two coupled shafts to be positioned such that their Where the motion deliberately is not to provide con-
extended center lines intersect in the coupling center. A stant velocity or positional adherence to the input form,
double Hooke joint can, however, also take up translational intermittent drives may be used. These include the quick
as well as angular misalignments. return, Geneva, intermittent gear drives, and the various
Where two shafts are parallel but not longitudinally in forms of clockwork escapement mechanisms. Many cou-
line, the Oldham coupling can be used. This allows for plings make use of elasticity to allow for misalignment.
misalignment by a sliding, cross-shaped connector. In these instances, it is unlikely that they will be of use if
information fidelity or smooth unmodulated drive is needed
through the link. Today, industrial drives tend to use elec-
tronic control to implement such motions but there is still
place for mechanical solutions due to their lower cost for
low energy coupling situations.
Thin
diaphragm
3 GEARS
(a) (b) (c)
Although the toothed gear was in use in ancient times,
it was not until more recent times that it could be used
extensively because of the relatively low cost resulting from
Moulded Phase adjust mass production and good quality control of the form and
polyurethane screw size of the two mating parts involved.
(d) (e) (f) Gears are used to alter (or maintain) the ratio of speeds
between two turning or partly rotating shafts. Many forms
enable the directions of the two shafts to be placed at any
required angle and displacement. The main limitation of
gears is that the shafts involved must be reasonably close
Flexible pins
together; this can be a serious shortcoming. Idler gears and
(g) unity ratio stages can be used to couple stages that are
Flexible plastic
moulding not close together. A combination of gears is called a gear
(h) train; it is housed in a gearbox.
The simplest form of gearing uses two smooth disks run-
Figure 1. Selection of couplings used in instruments: (a) dou-
ble Hooke joint; (b) Oldham; (c) membrane; (d) rigid; (e) mol- ning in contact driving by friction force of the contact.
ded flexible; (f) bellows with phase adjust; (g) Flex-thane; Friction gears are necessary in many applications, for exam-
(h) Kudriavetz. ple, in converting the linear motion of running strip into
Kinematical Regime – Rotary Motion 535
rotation to drive an encoder, as parts of various mechanical in a more compliant servo-loop response. The backlash
computing devices, to subdivide a circular table, to drive spring must hold the gears in contact against the friction
paper through a chart recorder, or to drive a record turntable of the train. It is often better to use higher grade gears
or tape recorder capstan – see also Sydenham (1968), and for which backlash can be reduced by closer tolerances of
Jain and Sydenham (1980). Trylinski (1971) includes a manufacture.
chapter on friction gearing. Experience of Sydenham and The inherent inertia of rotating parts may be significant; it
Brandwijk (1966) indicates that friction gearing (Figure 2) must be considered when designing gear trains for dynamic
can hold positional division of a 16:1 ratio step-down to the use. Inertia rises as the fourth power of the gear radius
order of one part in 20 000. The variation in ratio is appar- and each stage adds inertia as the reflected inertia seen
ently random in nature, and absolutely fixed ratio gearing looking into it, multiplied by the gearing ratio. Multiple
and, thus, synchronism cannot be maintained over a large stage step-up gears soon accumulate significant inertia. The
number of revolutions. Best results are produced when the inverse obviously applies; step-down gear boxes see little
drive wheel has a rounded profile running on a flat surface, inertia. High-ratio step-down boxes, however, will have
when the pressure is as light as possible to provide enough increased backlash and the wind-up of the train becomes
driving torque, and when the two materials are very hard, more significant as the elasticity of the gears, shafts, and
such as steel. High friction coefficient drive surfaces, like bearings combine to form a more compliant through-link.
rubber, give good drive adherence but poor constancy of As a general rule, the best response servos and the lightest
gear ratio. action manual instruments result from the use of a low gear
Where absolute synchronism or high torques are needed,
ratio and low tolerance gear wheels.
the use of toothed gears becomes necessary. Spur gears are
The choice of more stages of lower ratio to obtain a given
made in numerous forms and are generally familiar.
overall ratio may be better than using a single large ratio
In instrument work, the design of instrument gears
step. Friction of step-down trains is generally of little con-
closely follows general gear practice – with a few excep-
sequence, but step-up units may load the drive excessively
tions. One is the need to obtain precision operation. Gear
because of the multiplier effect of the train ratio.
trains in instruments and precision, small-size, closed-loop
A gear train is only as good as the bearings and box
servo units will possess a certain degree of clearance
structure allow it to be. Considering that high quality
between the teeth. This small level of backlash can intro-
duce an error of measurement or instability of a servo gearing runs with around 5 µm runouts of concentricity and
loop. It can, however, be reduced by the use of antiback- with pitch errors of similar magnitude, it can be seen that
lash gears. These use two gears running adjacently that are the bearings and machining must be of the highest quality
spring loaded to cause the two halves to clamp around the and the box be very rigid.
mating teeth of the other gear. Whereas antibacklash gears Gears are graded according to many codes, examples
do eliminate backlash, this form of device can introduce being the AGMA (old and new – American Gear Manu-
a degree of elasticity through the train, possibly resulting facturers Association) Admiralty, British Standard Spec-
ification (BSS), Mil Specs, and more. Specifying odd,
nonstandard, gears can be expensive. It is better to choose
from standard ranges of diametral pitch and quality classes
wherever possible. Reliance Gear (1972), Trylinski (1971),
Mabie and Ocvirk (1958), Beggs (1955), Merrit (1975),
and Michalec (1966) each cover the theoretical and practi-
cal design of the various forms of gears. CAD tools are for
supporting gear designs in power gear trains, but are not as
useful for the specialism of fine mechanism gears.
The designer can usually make use of the marketed
instrument fine mechanism products that have been properly
designed to obtain efficient and smooth operation. Manu-
facturers provide tables and other information to assist the
designer to call up the correct centers, clearances, the lubri-
cation to use, and how to design gear trains with specified
backlash. Reliance Gear (1972) contains a practical step-by-
Figure 2. Friction gearing used to subdivide a circle in the step guide to the design of fine-pitch gear trains, suggesting
automatic manufacture of radial metrology gratings. the following hints be followed.
536 Measuring System Design
The accuracy of gears can be gradually relaxed as the generated over a small relative angular magnitude. Cams
sensitivity to error reduces. The slowest speed gears are the cannot provide large strokes or rotations unless made equal
most sensitive. Use spur gears where possible to make the or larger in size as the amplitudes needed. They find use
most of the production precision that is available. Keep the in timing, function generation, for compensation, and as a
number of meshes to a minimum to preserve transmission means of producing motion from a source.
fidelity. Preload bearings to reduce backlash caused by Fundamentally, cams are continuous surface forms of
bearing slop, which is not eliminated with antibacklash linkage. At any instance or position, the point of contact
gears. Do not use pinned hub gears for high accuracy work. between the two parts and the support forms a linkage.
Sometimes gears can be matched by correct phasing to The design of cams is well developed and they can be
reduce the overall eccentricity error. synthesized as well as analyzed using graphical and com-
The finest pitch gears, theoretically at least, must produce putational methods. Shigley (1988), Erdman, Sandor and
the most precise gear train but practice limits their produc- Kota (2001), Juvinall and Marshek (2002), Sclater and
tion. As a guide, avoid teeth finer than 120 DP or the finest Chironis (2001) Parmley (2000) and Mott (1998) are also
above this that is consistent with the load requirements. sources to use. Mabie and Ocvirk (1958) give a good
Avoid mixing pitches in a train. Cleanliness of assembly coverage of cam design; Beggs (1955) and Faires (1955)
and in use are vital. provide less extensive accounts. Rothbart (1956) is a text
High ratios are obtained in relatively small volume spaces devoted to cam design. Surprisingly, because cams are used
using worm and wheel gears. These, however, may be extensively in instruments, Trylinski (1971) does not cover
inappropriate where the gear train has to transmit motion in the subject.
both directions, as can occur in some servo drives. They can Cam design basically attempts to realize the correct
also be damaged easily in instruments where there are shafts conversion action from the two components, but it is
that can be unwittingly forced by the operator or service complicated by the fact that high-speed cams will impart
man. Where this might occur, a friction clutch can be used significant velocity, acceleration, and jerk to the follower.
to allow excessive torque to be dissipated without harm. This can cause the follower to leave the cam surface and
The use of gears is not restricted to full-circle wheels produce excessive wear. In other words, the correct func-
only. Part gears can be used where the motion needed tioning of the cam and its follower is lost unless the
is only part of a revolution or where intermittent motion shape is appropriate for the dynamics involved. Various
is needed. The Leibnitz carry mechanism used in classic cam shapes provide different velocity, acceleration, and jerk
mechanical calculators and mechanical automobile odome- characteristics. Typical forms used are parabolic, modified
ters is an example. Where precision length measurement parabolic, and simple harmonic for relatively slow actions.
or positioning is needed using gears, a rack and pinion is For higher speed applications where jerk becomes signifi-
generally adopted. The pinion must run with little backlash cant, the forms used are cycloidal, half-cycloidal, harmonic,
and the guides must be carefully manufactured to keep the
system aligned. The pieces of rack must by phased properly
when butted together to extend the length range. Precision
racks can provide relative readout or positioning to within
one part in 100 000 (or a little better) uncertainty.
4 CAMS
This is a two-part fundamental mechanism, in which one
part, the cam, imparts motion to the second part, the
follower, that stays in contact with the surface of the cam.
Cams take the shape of disks or plates turning on shafts,
as curved slots cut around a shaft, as eccentrics within
a yoke; they can also be formed as linear alternatives in
which a follower rises and falls (Figure 3) in response to
the shape of a linear plate profile moving relative to it.
Figure 3. Linear corrector plate used to provide automatic com-
They are used to provide rotary to rotary, or to linear pensation for calibration factor, in-scale nonlinearities and possi-
motional conversion, or linear to linear action. Cams are bly thermal errors in a wire length measuring device (Sydenham,
characterized by the feature that the output motion is only 1969).
Kinematical Regime – Rotary Motion 537
and eight-power polynomial. Each offers certain desirable slenderness ratios of at least 300 : 1 (ratio of radius of
features of operational acceleration and jerk magnitudes; bending to the thickness of the flexure in the direction of
see Mabie and Ocvirk (1958) for more detail. bending). This figure is only a guide. A calculation should
Cams may run directly against the follower and be con- be made of the stresses in the outer layer when the member
stantly in sliding contact. This situation can be designed is bent over the smallest radius, in order to see if the
for by the use of hardened steels and other suitable materi- material is not near to the yield point. There may be a
als. Alternately, a roller or hydraulic follower may be used. jockey wheel or a guide pulley added to keep the system
The follower can be a hinged rotating or sliding translating taut; often these are made too small in diameter. The stress
component. The cam in disk or plate form can be shaped level is as important as designing for adequate drive forces
to form a function required for applications of compensa- because excessive bending stresses will eventually cause
tion or function generation. In these applications, velocity, the drive material to fatigue and break.
acceleration, and the higher derivatives generally give no
problems. Cams are also used to generate a shape that is
‘followed’ by optical means. This is used in furnace and 6 BELT DRIVES
other temperature controllers where the temperature must
change with time in some specified manner. There are many alternatives available, each having partic-
Clearly, rapidly changing functions are not readily ular advantages and disadvantages. Flat belts made of a
accommodated by cams because the follower has a finite material having adequate tensile strength and friction coef-
size and will introduce, by virtue of its aperture width, some ficient can be used with flat pulleys (actually the pulleys
degree of integration. Cams are often required to rotate only should be slightly curved, rising in the center, to cause
in one direction. A reversed motion can become locked the belt to track into the middle). More usually, instru-
when the follower comes against a rise rate to which it ments would use small section vee-belts or continuous
cannot respond. Cam shaped gear systems also exist that round section belts made of reinforced plastics. O-rings
generate specific functions, but these are strictly gears, make excellent belts; as they are made to tight tolerances,
not cams. they provide smooth drives. If the forces to be transmit-
Cams are also made in which the follower responds to ted are high, the pulleys and belts should be designed after
the resultant of two inputs. These are referred to as 3-D calculations are made to decide the ability of the drive to
or space cams. These need a spherical ended follower that operate with a realistic tension that will transmit the force
is allowed to move in two translations as it is directed by without breaking. Round and vee-profile pulleys provide
the cam. Space cams are generally of complex 3-D shape increased friction drive force without needing high belt ten-
and can be made by sintered forming methods or be cut sions. Their design is laid down in texts on general machine
by numerically controlled machine tools. These cams are design and is well defined. Trylinski (1971) is a satisfactory
unable to impart much force into the follower so they are source on flexible instrument drives. Figure 4 gives a range
often used with a servo that slaves to the follower’s motion. of flexible drives used in instruments.
The dynamic behavior of cams can be complex because The friction drive belt depends upon the normal force
of the high value accelerations that can be present. These exerted between the belt and the pulley surface in order
can cause the cam to appear elastic and can induce to develop enough grip. These forces can sometimes be
resonance in the cam body and in the follower and its prohibitive and a better alternative may be needed. Further-
spring. Beggs (1955) reviews this problem. more, the drive may need to maintain a strict synchronized
positional relationship between the two pulleys. For these
cases, friction belt drives are not satisfactory.
5 FLEXIBLE CONNECTING DRIVES
This classification is given by Trylinski (1971) to that 7 CHAIN DRIVES
group of motional transference mechanisms that use flexible
elements to transmit motion. They may do this with or Synchronized flexible chain drive can be obtained by the
without conversion from rotary to linear regimes. Some use of toothed flexible belts running on appropriately
links use elasticity to provide the freedom of bending; toothed wheels. These are generally available only for
others use kinematic, pivoted principles such as those found comparatively large-size drives, the smallest being about
in a link chain. 10-mm wide. Metal roller link chains also can be used
Elastic types will not fail because of the bending stresses but these, too, are not generally made for the size drives
being excessive in their outermost ‘fibers’ if they have needed in the smaller instruments. Smaller special chains
538 Measuring System Design
a 15-N tensile load. There is also a toothed plastic belt
that will drive sprockets set at 90◦ to each other without
the need for belt twist. For illustrations of the above see
Figure 4.
(a) (b) Clearance needed When only a small distance of angular motion is required,
(c) the belt need not be endless; further, it can be fastened to
the two pulleys to obtain synchronized drive. Examples
are the mechanical drive of a tuning dial of a radio in
which a cord connects the tuning control with the translating
(d) (e) (f)
station indicator. Thin steel strips can also be used to couple
wheels. As these can only drive in tension, a continuous
strip or a biasing spring are needed. Trylinski (1971) gives
several illustrations of the possibilities.
Flexible drives can offer extended distances between
drive centers and the ability to provide gearing ratios;
(g) they can absorb vibration, attenuating it through the link
(but they may lack stiffness if this is so); they can allow
Figure 4. Cross sections of flexible connecting drives used in for misalignment between shafts, they can be designed to
instruments: (a) round, with or without reinforcement; (b) flat allow slippage if the load is too great (or the converse),
(on slightly domed pulley); (c) vee; (d) metal link roller chain;
(e) polyurethane molded – pinned belt; (f) as for 3-D pins;
and they can also be designed to give a continuously
(g) metal strip. variable gearing ratio by the use of tapered pulleys and
changing radius pulleys. They are virtually backlash-free
and will have relatively low inertia compared with gears
and toothed belts are made but their availability is not trains because they have less mass in motion. They can,
widespread. The user will have to search hard for a supplier however, introduce vibrations from the unsupported belt,
(see Reliance Gear, 1972, and the British Machine Chain which can modulate the drive throughput of force or
Co, Malden). Examples are the miniature oval-link chains information.
that use specially shaped drive wheels in which the chain
links lie keeping its crossed link shape, and the hook-link
chains that use sprockets whose teeth pass through the links
to provide drive. Fine metal chains can give backlash-free 8 ELECTROMECHANICAL GEARING
coupling at the force level needed.
Typical working forces for small chains are 1 N, with The present low cost of digital electronics and the general
links of 3-mm pitch. Small, plastic, positive-drive belts are availability of very reliable and inexpensive optical and
also manufactured. One system uses stainless steel or cord- other shaft encoders has given the designer the ability
reinforced polyurethane molded belts, which comprise a to make more use of electromechanical gearing and shaft
round belt having bars formed across it protruding to each motion transference.
side. Another form has its sides shaped like a ladder. The Combination of a shaft digitizer (Figure 5) that generates
bars nest into a single (or double) row of teeth on the drive electrical positional signals as the shaft rotates and a motor
pulleys. One Dacron-reinforced belt will take 1 N forces, that steps in synchronism with the generated signals forms
the unreinforced belt only 0.05 N. Some designs will carry the basis of an electromechanical gear box.
Rotary pulse Electronic pulse Stepping motor
generator divider
Input Output
shaft shaft
Distance between I/P and O/P units is not limited
Figure 5. Diagrammatic representation of an electromechanical gear box.
Kinematical Regime – Rotary Motion 539
Illustrating the method with the use of the optical REFERENCES
radial gratings, the disc attached to the input shaft can
conveniently generate around 1000 to 10 000 pulses per Beggs, J.S. (1955) Mechanism, McGraw-Hill, New York.
turn at rates rising to at least 106 pulses s−1 ; that is, shaft Erdman, A.G., Sandor, G.N. and Kota, S. (2001) Mechanism
speeds varying from 0 to 1000 rev s−1 for low resolution Design: Analysis and Synthesis, Prentice Hall.
or to 10 rev s−1 for high resolution can be accommodated. Faires, V.M. (1955) Design of Machine Elements, Macmillan
The digital signals are then passed through a digital divider Publishing, New York, (1st edn 1934).
arrangement to form the required divided pulse rate to Jain, P.K. and Sydenham, P.H. (1980) Radial Metrological Pat-
suit the speed needed from a stepping motor. Dividers are tern Generating Engine. Journal of Physics E-Scientific Instru-
available as integrated circuits (IC) in which preset switches ments, 13, 461–466.
or electronic signals can be used to set any division number Juvinall, R.C. and Marshek, K.M. (2002) Fundamentals of
needed. Two ICs can give ratios from unity to 255. Machine Component Design, Wiley, New York.
The advantages of this alternative form of ‘gear box’ Lent, D. (1970) Analysis and Design of Mechanisms, Prentice
are that the input and output shafts need not be in close Hall, Englewood Cliffs, NJ.
proximity or in any particular orientation, the power levels Mabie, H.H. and Ocvirk, F.W. (1958) Mechanisms and Dynamics
of the two sides of the link can be at any level needed, of Machinery, Wiley, New York, (3rd edn 1975).
more than one shaft signal can be combined or driven, and Merrit, H.E. (1975) Gear Engineering, Wiley, New York.
the division ratio is very easily changed under computer Michalec, C.W. (1966) Precision Gearing: Theory and Practice,
control. The ratio can be varied smoothly from one extreme Wiley, New York.
to the other. Accuracy can be at any level desired and Mott, R.L. (1998) Machine Elements in Mechanical Design, Pren-
backlash is not present. tice Hall.
This form of system, however, can be prone to electrical Parmley, R.O. (2000) Illustrated Sourcebook of Mechanical Com-
interference, but it is quite possible, with circuitry effort, ponents, McGraw-Hill.
to make the error rate as small as is desired by the use of Peatfield, A.E. (1951) Engineering Components and Materials,
coded digital communication transmission methods. EUP, London.
The cost of the electronic parts will be reasonably small Reliance Gear. (1972) Catalogue RG5, Reliance Gear Co Ltd,
and the mechanical parts needed are few – a shaft carrying Huddersfield. (Also similar catalogues by PIC Design Corp,
an encoder disk to couple to the output of a mechanism New York).
and an output shaft of the stepping motor. Such gear boxes Rothbart, H.A. (1956) Cams: Design, Dynamics and Accuracy,
are cheaper than conventional mechanical units, especially Wiley, New York.
where high precision is needed. These will, of course, need Sclater, A. and Chironis, N.P. (2001) Mechanisms and Mechani-
a source of electrical power, which may exclude them from cal Devices Sourcebook, McGraw-Hill.
some applications. Shigley, J.E. (1988) Mechanical Engineering Design (McGraw-
Hill Series in Mechanical Engineering), McGraw-Hill.
Sydenham, P.H. (1968) Linear and Angular Transducers for
RELATED ARTICLES Positional Control in the Decameter Range. Proceeding of the
IEE, 115, 1056–1066.
Article 57, Static Considerations of General Instru- Sydenham, P.H. (1969) Manual and automatic large-scale metrol-
mentation, Volume 1; Article 59, Introduction to the ogy, PhD thesis, University of Warwick.
Dynamic Regime of Measurement Systems, Volume 1; Sydenham, P.H. and Brandwijk, J.F. (1966) Production of Radial
Article 80, Principles of Fine Mechanics – Kinematic Line Gratings Using a Mechanical Photographic Apparatus.
and Elastic Designs, Volume 2; Article 81, Principles Journal of Scientific Instruments, 43, 380–382.
of Fine Mechanics – Systems Considerations, Volume 2; Trylinski, W. (1971) Fine Mechanisms and Precision Instru-
Article 85, Elastic Regime of Design – Design Princi- ments – Principles of Design, Elsevier, translated by A. Voell-
ples, Volume 2. nagel from Polish original.
85: Elastic Regime of Design – Design Principles
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
Uses and Benefits of MEMS, Volume 3; and Article 164,
1 Place of Flexure Elements in Instruments 540 Principles of MEMS Actuators, Volume 3.
2 Basis of Designing Flexure Elements 541 By appropriate choice of compliance, component shape,
and method of use, spring elements (often called flex-
3 Design Parameters of the Flexure Element 541
ures) can provide many of the functions provided by the
4 Carrying out the Design of a Flexure
kinematic forms. Elastic design deliberately makes use of
Element 543
elastic properties of materials to allow controlled movement
5 Specialized Formed Flexure Elements 543 where needed.
6 Commonly Found Beam Design Formulae 543 In many instances, a flexure method is to be preferred
7 Types of Flexure Elements 544 over a kinematic arrangement. This generally occurs when
8 Information on Flexure Systems 546 only limited movement between two parts is needed. For
Related Articles 546 example, a crossed-strip flexure hinge provides a limited
References 546 range angular hinging action and offers negligible backlash
and frictional force. It does not need lubrication and is not
affected by dust or dirt. It is very simple to make and often
does not require tight manufacturing tolerance.
1 PLACE OF FLEXURE ELEMENTS Flexure elements can be devised in numerous shapes and
IN INSTRUMENTS forms. Good design is always based on proper understand-
ing of the compliance of a flexure element in its various
Mechanisms used to assemble instruments are divided into directions of bending and compliance values of other asso-
two groups, namely, kinematic and elastic. Kinematic ele- ciated elements. In some cases, it needs to have a high ratio
ments are made sufficiently inelastic to be unchanging in of compliance in one direction with respect to another; in
shape as they are stressed by mechanical forces. Here, other cases, it needs to be the same. The addition of later
the general principles of the elastic group that deliber- items, such as protective covers, additional small loads, and
ately makes use of elasticity of components to achieve a further forces from sources other than those designed for,
given mechanism function are discussed. Flexing action could considerably alter the final performance of an elas-
produced by kinematic mechanism, such as a roller chain tic design.
or belt, is covered elsewhere. Detailed discussion of uses Flexure elements appear to be deceptively simple. They
of elastic elements is found in Article 86, Elastic Regime are easy to devise and install and they will often pro-
of Design – Spring Systems, Volume 2 and Article 87, vide quite adequate performance, with only the barest of
Elastic Regime of Design – Plates and Bimorphs, Vol- theoretical understanding being applied. However, a thor-
ume 2. Implementation in microchip form is covered in ough appreciation of the factors involved will assist the
descriptions of Microelectro Mechanical Systems (MEMS), user to understand their limitations; they are not perfect
Article 162, Principles of MEMS, Volume 3; Article 163, elements.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Elastic Regime of Design – Design Principles 541
The necessary relevant basic theory for carrying out that linear second-order and higher dynamic systems may
elastic flexures design pertains to that of the deflection and be produced (refer to Article 189, Basic Principles of Flow
stresses of beams and plates – commonly called strength of Measurement, Volume 3). The natural frequency of such
material or beam theory. systems becomes an important parameter where the system
To this, the knowledge of imperfections of real materi- has to follow dynamic inputs in a faithful manner (Syden-
als must be added because these introduce such effects as ham, 1982, 1983). Damping may be deliberately introduced
damping, friction, hysteresis, fatigue, fracture, and nonlinear to prevent resonances from building in amplitude or to
deflection versus load characteristics – subjects not nor- make the dynamic response at least critical so as to exhibit
mally covered in beam theory texts. response characteristics of first-order nature. It is feasible,
Fundamental material (which is of civil and mechanical by appropriate design and ingenuity, to alter the natural
engineering structures design) is to be found in such works frequency without altering the compliance. For the same
as Timoshenko and MacCullough (1959), Timoshenko and reasons, it may be necessary to keep the supported mass
Gere (1961), and Roark and Young (1975). Faires (1955) to a minimum so that the natural frequency is kept as high
provides a shorter but basically adequate account of the as possible.
strength of materials theory. Such texts, being written for
structure designers and not for instrument applications,
generally place greater emphasis on stress than on the 3 DESIGN PARAMETERS OF THE
deflections that are so important to instrument design. FLEXURE ELEMENT
2 BASIS OF DESIGNING FLEXURE Before considering practical flexures and their application,
ELEMENTS it is appropriate to discuss several basic design parameters
of flexure elements since these apply to all shapes and forms
Flexure design studies are usually based upon the assump- of system built with them.
tion that the force-deflection characteristics of the materials Flexures are formed by the use of a mechanical member
used are linear. Such a basis can, indeed, give the designer (or possibly, other spring form such as electromagnetism),
a very wide choice of workable ideas, but on occasion, it where design deliberately allows the member to deflect in
might be beneficial to make use of nonlinear materials and specific directions. It is usually imperative that deflection
processes. Eisley (1963) provides an entry point into the occurs only within the elastic region of the material’s
literature in this area of knowledge, giving about 90 ref- deformation (Figure 1). Although texts on materials give
erences to published papers. Schmidt and DaDeppo (1973) the general impression that the Hookean elasticity law
may also be useful in this area. applies in practice, there, nevertheless, is always a small
A considerable amount of published design information amount of hysteresis; a strained member does not come
is available but it is from earlier times with little being pub- back to exactly the same strain condition for a given
lished recently. This should not be taken to imply that the stress excursion. Special materials, those with a reduced
information has become outdated but merely that this field mechanical hysteresis, exhibit a lower level of this effect.
matured several decades ago. There is little likelihood of a In many applications of flexures in instruments, however,
basically new flexure element being invented, most having extreme reproducibility is not important. Tovey (1974)
been devised by now. This article provides a comprehen- discusses the materials used to form flexure elements.
sive overview of the many possibilities for elastic design. It is therefore important that the flexure members can-
Combination of several flexure mechanisms leads to more not be deflected beyond the chosen allowable elastic limit.
options, many of which are yet to be applied. As a rule, the smaller the deflection, the smaller the
The apparent simplicity of flexures can hide the fact that residual strain error. Plastic deformation, if allowed to
they are not perfect elements. As an example, a parallel occur, will introduce permanent set, with subsequent zero
strip support for a translating table can easily support a shift errors arising. For this reason, design often calls
mass of many kilograms, providing translational sensitivity for mechanical stops and other devices to be incorpo-
and stability of nanometer order without the backlash being rated to prevent overdeflection. Assembly and transport can
apparent. However, temperature effects may alter the inbuilt often give rise to the greatest deflections that might be
spring rate quite considerably, causing errors in other ways. encountered. As a rough rule of thumb, bending flexure
Choice of the wrong material can introduce significant elements should not be deflected into a radius of curvature
hysteretic backlash. smaller than around 300 times its thickness in the direc-
As flexure systems deliberately introduce spring compli- tion of bending. Beyond this level, the outer fibers will be
ance into a system where mass is involved, it is obvious stressed beyond their elastic limit, causing surface cracks to
542 Measuring System Design
Linear ‘elastic region’ Nonlinear, nonelastic
plastic region
To
failure
Maximum
stress
in member
due to
flexing Working region in which
flexure must remain to
retain settings. Size
depends upon residual Possible return path
strain allowable after excessive deflection
Practical return
path results in
some level of
residual strain
Strain deformation of flexure element
Residual strain magnitude
depends on strain excursion
and material properties
Figure 1. Stress–strain relationships for materials used to make flexure elements.
begin. If this is important, the bending moment caused by when in conditions of simultaneous corrosion effects),
expected flexing levels should be calculated for the material thermal coefficient of elastic modulus, and such other
used. parameters in any other energy domain. Rolling friction is
As a flexure deflects, it will provide a force reacting sometimes a parameter for which more information is pub-
with the direction of motion. This is the flexure system lished in Minato, Nakafuku and Takemura (1969). Gohn
compliance. It is produced fundamentally by the elasticity and Fox (1965) discuss stress relaxation in springs. Carson
of the material, the effect of which is modified by the total (1937) is about measurement of elastic drift in springs.
flexure system arrangement. As will be mentioned below, it When very small displacements are needed, a method
is possible to devise flexures that provide positive, negative, used by Jones (1967) might be applicable. It rests upon
and even zero forces, as well as nonlinear displacement the principle that an elastic material that is squeezed by
characteristics. a clamp will exhibit a small length change. Poisson’s ratio
Using strips in tension will increase the stiffness accord- links this action. This method gives fine control to clamping
ing to the load (Eastman, 1937). Conversely, they can be screws on threaded adjustments such as micrometer heads.
used in longitudinal compression quite satisfactorily, pro- The same principle was employed by Baird (1968) to
vided Euler buckling does not occur. The basic elements temperature-compensate an Invar rod with a steel wire
still undergo positive or negative deflections toward or away cover soldered to its outer surface.
from the elastic limit, but the net effect of the whole flexure When flexures are made from thin hardened and tem-
system is one of these options. pered strips, they can tear easily at connection points.
The fundamental design parameters of a flexure mate- Careful edge finishing is needed to reduce this failure mode.
rial are its elastic modulus, shear modulus, hysteresis level, The material must also be cut to suit any directional grain
elastic limit, endurance limit to constant flexing (especially that exists in the flexure material.
Elastic Regime of Design – Design Principles 543
Materials used for flexures are generally the metal carbon 5 SPECIALIZED FORMED FLEXURE
springs, stainless steels, copper, and nickel–iron alloys. ELEMENTS
In some applications, fused-silica (fused-quartz is a term
that is usually synonymous) forms the flexure element. In It is stated by some authors and users of flexures that
less exacting applications, plastic, rubber, paper, and fiber the applications requiring the greatest dimensional stabil-
elements might be suitable. ity should use flexures that have end mountings made in
The above parameters are concerned with the properties the same piece of material. (The usual method is to clamp
of the material used to form a flexure. Having selected the the flexure end between rigid plates.) This assertion has
material, it is then necessary to consider the properties of not been adequately proven by rigorous experimental pro-
the flexure system as a whole. Some factors that may be cedures and it must be left to the user to decide whether
relevant include the force-deflection characteristic, range of this additional cost is justified.
motion, required parallelity of pseudosliding action, stabil- A similar argument exists as to whether metal spring
ity of a center of rotation with angular deflection, and effect elements give better performance than fused-quartz springs.
of load on the system properties because the system may Experience with both has shown that poor reproducibility
stiffen up, with increasing load being supported. Another of the zero reading with time is more likely to be a factor
important design parameter is to be sure that the system of overall design than of the material used. It is possible
can be manufactured to the tolerances needed and that it to find evidence that suggests that either material group
can be assembled without damaging the flexure element in is the better to use for precision instrument springs. As
some way. Preform in flexure elements can be bothersome the problem reduces to one concerning the hysteresis and
at times. Elements may need to be heat-treated to stabilize temperature properties of the materials, it will be found
them before use and be assembled using jigs to prevent that certain metal alloys are far superior to fused-quartz in
overstressing. this respect.
A key parameter that tends to be overlooked in such
comparisons is the level of flexure excursion. Small move-
4 CARRYING OUT THE DESIGN OF A ments of a spring element, such as those occurring in a
FLEXURE ELEMENT tidal gravimeter or a tidal tiltmeter, will give rise to much
smaller levels of hysteresis when compared to the per-
Design of a flexure system, like that of a kinematic support formance of an exploration gravity meter or a borehole
system, should always consider the sources of error that tiltmeter. In the latter case, the flexure element must range
might arise in normal operation. over larger excursions and can be expected to be more
At present, and perhaps it will always be so, the design of severely handled.
a flexure system is initially approached as an art on the basis
of experience, with the possibilities open to the designer.
A possible arrangement is first selected that may fulfill the 6 COMMONLY FOUND BEAM DESIGN
need. This is then analyzed as a science to establish the FORMULAE
sensitivity, load-carrying capacity, limits of required action,
and so on. If the design is not quite suitable, it is then Figure 2 provides the usually needed deflection and stress
modified. It would be an obvious advantage if computer- expressions for a few of the commonly found elements.
aided design programs were available for use in optimizing Note that point load and distributed load cases each follow
standard design configurations (of which there are several the same general laws.
kinds) in the direction that a designer requires. For flexures to be theoretically designable, it is necessary
Design equations for most types have been reasonably that their stress levels be determinable for all parts of
well developed, but the designer usually must spend consid- the system. In the kinematic design procedure, forces are
erable time in locating the appropriate study. Many reports arranged so that they act at known, well-defined points of
on flexure elements usage do not include a theoretical interaction. In flexure design, deflection takes place over a
account, their design often being apparently raised by intu- distributed region.
ition and cut-and-try prototype building procedures. Of the To assist design, it is often expedient to structure the
references to general accounts given above, those includ- flexure elements of the system as members having dramatic
ing a well-developed basis of theory to support a design changes of size at given positions, so the compliance alters
are Chironis (1961), Eastman (1937), Andreeva (1966), very significantly. It can then be assumed that the flexure
Geary (1954, 1960), Thorpe (1953), Trylinski (1971), Wahl member is deforming elastically, with the end mounts being
(1963), and Weinstein (1965). adequately rigid.
544 Measuring System Design
Point loading Deflection d Maximum Stress s Type
P
I Cantilever with
PI 3/3E I PIh /I
d point load
P
Simply supported beam
PI 3/48E I PIh /4I
with central point load
I
P Point load applied to
center of beam with
3
PI /192E I PIh /8I
ends constrained
I rigidly
Uniformly distributed loading
p Cantilever with uniform
pI 4/8E I pI 2h /2I distributed load over
I whole length
p Simply supported
4 2
beam with uniformly
5pI /384E I pI h /8I distributed load only
I between supports
p Uniformly distributed
load applied to whole
pI 4/384E I pI 2h /12I beam with ends
I constrained rigidly
where I – second moment of cross section
l – distance
E – Youngs modulus of beam material
h – maximum distance of any point in beam from neutral axis
P – point load
p – load per unit length
Further expressions for other loadings, torsion on beams, and
diaphragms are variously available in Neubert (1975), Roark and
Young (1975), Timoshenko and MacCullough (1959), Trylinski (1971).
Figure 2. Expressions for commonly found beams.
From the viewpoint of fatigue strength, however, such arrangement. Too often, designers (and quoted formulae for
sudden geometrical changes in shape can give rise to loca- flexure arrangements) do not recognize this combinational
tions in the flexure where stresses can be raised. Flexure feature. For example, a simple single flexure strip being
designs needing high numbers of highly stressed cyclic used at the top end of a clock pendulum to allow it to
movements will require a more sophisticated treatment swing in one preferred direction also supports the mass of
to allow for graded compliance changes along the flex- the pendulum. The flexure element is, therefore, simulta-
ure element. neously stressed in both bending and tension. The tension
component can considerably stiffen the rotational compli-
ance in the bending direction.
7 TYPES OF FLEXURE ELEMENTS The helical coil spring is a torsional arrangement in
which torsional flexure is converted to provide linear
A flexure can provide controlled movement in a system of spring-force action. Torsional flexure systems can make use
flexure elements, using one or more of three basic modes of forms other than strip or other solid members. Two wires,
of flexure (Sydenham, 1984). placed apart, can produce a similar action – these are called
It can be deflected in pure transverse bending, in pure bifilar suspensions.
axial torsion, or it might be deformed elastically in its lon- The third class, wherein longitudinal tension or com-
gitudinal compressive or tensional mode (Figure 3a). In pression is used without transverse bending action of the
practice, combinations of these three actions occur to a flexure, is applicable when stiff flexures are needed. Sim-
lesser or greater extent depending upon the flexure system ple examples would be the use of a bolt to hold two
Elastic Regime of Design – Design Principles 545
z Tension or
compression
Bending
y
Cross section can be any
x suitable form; round, square,
rectangular, tubular, etc.
(a) Torsion
Assembled shape
Natural manufactured shape
Clock spring
Long flat strip
Relay
contact spring
Bent to obtain
(b) prestress when assembled
Figure 3. Fundamental flexure design options: (a) modes of flexure deflection; (b) natural shapes used either in some manufactured
shape or preformed.
parts together yet allowing some additional elasticity to A torque motor (that is needed to drive a mechanical clock)
enable it to cope with excess tension loads or a rubber can be produced by precurving the strip so that its natural
isolation mount. Torsional systems often stress the individ- shape is a small radius, that is, the unassembled spring rests
ual members this way. Such methods may dictate the use as a tight coil that is then unwound as it is deformed to
of materials with low values of Young’s modulus so that store energy. Springs holding electrical contacts in relays
they provide reasonable compliance from practical cross are usually preset to some extent so that their assembled
sections of material. Their use is, of course, dependent on state holds the contact in the correct position, applying a
the requirement and materials’ suitability. force to the other contact.
With the basic elastic kind of action being decided, the Yet another distinction can be drawn between the flexure
designer then has a further fundamental option available. system that is purposely intended to store energy in order
Many flexure systems use the element in a stress-free state to produce a force on demand and that which uses flexure
when the system is at its rest position. For example, a helical action as a mechanical linkage. In the latter case, it is
spring may be coiled so that the turns just nest together often necessary to design for minimum stored energy as
and with no extension, the spring material is in a relatively deflection occurs, storage and its resulting force-producing
stress-free condition (see Figure 3b). Elastic energy is then action being an unwanted parameter. Often, the flexure
stored as the spring is tensioned. system simultaneously provides both mechanical linkage
Another example is the use of clock spring to provide and energy storage functions; for example, a cantilever flat
drive energy for a clock. In its unassembled state, the strip spring provides return action and guidance.
natural shape of the flexure is a long flat strip. As it Through the use of a suitable design of spring, it is
is wound up, it takes on the curved shape seen in the possible to obtain mechanical magnification of small dis-
assembled clock. placements. Such springs are generally of the type in which
The alternative spring flexure arrangement is to prestress a small change in length of a tensional spiral spring pro-
the spring material into a natural shape that operates in the duces a rotational motion in the center or at the end of the
reverse manner to the obvious situation exemplified above. spring element. The rotation can then be used to operate a
546 Measuring System Design
suitable length output arm or the mirror of an optical lever. Design – Spring Systems, Volume 2; Article 87, Elas-
Torsional magnifiers are discussed in Geary (1960). tic Regime of Design – Plates and Bimorphs, Volume 2;
In practice, there are a very great number of possibil- Article 88, Error Sources in Fine Mechanics, Volume 2.
ities, so it is not feasible to list every combination and
arrangement that has been reported.
REFERENCES
8 INFORMATION ON FLEXURE Andreeva, L.E. (1966) Elastic Elements of Instruments, Israel
Program for Scientific Translations Ltd, Israel. Translation from
SYSTEMS Russian.
Baird, K.M. (1968) Compensation for Linear Thermal Expansion.
Collectively, the literature (when reviews included within Metrologia, 4, 145–146.
several of the citations are also recognized) provides a very Beggs, J.S. (1955) Mechanism, McGraw-Hill, New York.
large range of diagrams and applications of flexure elements
Braddick, H.J.J. (1954) The Physics of Experimental Method,
in the design of instruments. What constitutes an instrument Chapman & Hall, London.
is a matter of personal definition in this area, for flexure
Carson, R.W. (1937) Measuring Elastic Drift. Proceedings of
elements find as much use in power-supporting applications ASTM, 37, 661–675.
as they do in fine mechanisms.
Chironis, N.P. (1961) Spring Design and Application, McGraw-
Young (1989) is a definitive source of formulae for Hill, New York.
mechanical situations. The design of springs elements,
Dorien-Brown, B. (1971) Air Springs for Vibration Isolation,
which commonly concentrates upon helical and flat Harold Armstrong Conference on Production Science in
power springs (see Article 86, Elastic Regime of Industry, August, Institution of Engineers, Melbourne, (pp.
Design – Spring Systems, Volume 2), as opposed to 213–225).
instrument springs, is dealt with by Chironis (1961) and Dratler, J. and Block, B. (1972) A Wide-band Horizontal
Wahl (1963). Spring suspensions and flexure devices, for Accelerometer with Preliminary Earth Normal Mode and
which the emphasis is upon fine mechanism, are covered Seismic Investigations. Geophysical Journal of the Royal
in Eastman (1937), Geary (1954, 1960), Andreeva (1966), Astronomical Society, 27, 337–367.
Trylinski (1971), and Sydenham (1984). The definitive Eastman, F.S. (1937) The Design of Flexure Pivots. Journal of
reference work is Howell (2001). It covers the wide range Aeronautical Sciences, 5(1), 16–21.
of topics introduced here. Eisley, J.G. (1963) Nonlinear Deformation of Elastic Beams,
Geary (1964), Frazier, Gilinson and Oberbeck (1974), Rings and Strings, (Bibliography, 92 refs). Applied Mechanics
Reviews, 16(9), 677–679.
and Fremery and Doden (1978) cover another type of spring
suspension: that provided by magnetic and electric fields. Faires, V.M. (1955) Design of Machine Elements, Macmillan,
New York.
Dorien-Brown (1971) is one review of air springs.
Many other works include mention of flexure arrange- Frazier, R.H., Gilinson, P.J. and Oberbeck, G.A. (1974) Magnetic
and Electric Suspensions, Types of Suspensions Developed in the
ments in sections ranging from chapter length down to Charles Stork Draper Laboratory, MIT Press.
a few paragraphs. These include Rolt (1929), Whitehead
Fremery, J.K. and Doden, K. (1978) Active Permanent Magnet
(1934), Thorpe (1953), Braddick (1954), Beggs (1955), Suspensions for Scientific Instruments. Journal of Physics E:
Weinstein (1965), Jones (1967, 1968), and Neubert (1975). Scientific Instruments, 11(2), 106–113.
Hugill (1978) compares several spring-mass systems used Gast, T. (1974) Vacuum Micro Balances, their Construction and
in gravimeter designs. Equations of sensitivity are given Characteristics. Journal of Physics E: Scientific Instruments, 7,
in each case. Gast (1974) includes discussion on flexures. 865–875.
Dratler and Block (1972) is an interesting report. Sclater Geary, P.J. (1954) Flexure Devices – Pivots, Movements, Suspen-
and Chironis (2001) is a useful source of ideas. sions, Sira Institute, Chislehurst.
Geary, P.J. (1960) Torsion Devices, Sira Institute, Chislehurst.
Geary, P.J. (1964) Magnetic and Electric Suspensions, Sira Insti-
RELATED ARTICLES tute, Chislehurst.
Gohn, G.R. and Fox, A. (1965) Stress Relaxation in Metallic
Article 80, Principles of Fine Mechanics – Kinematic Springs and Conductors. Electrotechnology, 16, 85–92.
and Elastic Designs, Volume 2; Article 81, Princi- Howell, L.L. (2001) Compliant Mechanisms, Wiley, NY.
ples of Fine Mechanics – Systems Considerations, Vol- Hugill, A.L. (1978) Gravimeter Design, Int Rep DSS/AH/171,
ume 2; Article 82, Kinematical Regime – Members and September, Systems Science Department, The City University,
Linkages, Volume 2; Article 86, Elastic Regime of London.
Elastic Regime of Design – Design Principles 547
Jones, R.V. (1967) The Measurement and Control of Small Thorpe, A.G. II. (1953) Flexure Pivots-design, Product Engineer-
Displacements. Physics Bulletin, 18, 325–326. ing, 24(2), 192–200.
Jones, R.V. (1968) More and More about Less and Less. Proceed- Timoshenko, S.P. and Gere, J.M. (1961) Theory of Elastic Stabil-
ings of Royal Institution of Great Britain, 43(202), 323–345. ity, McGraw-Hill, New York.
Minato, K., Nakafuku, C. and Takemura, T. (1969) Rolling Fric- Timoshenko, S. and MacCullough, G.H. (1959) Elements of
tion of Metals. Japanese Journal of Applied Physics, 8(10), Strength of Materials, Van Nostrand, NJ. (Numerous editions
1171–1179. since 1935).
Neubert, H.K.P. (1975) Instrument Transducers – An Introduction Tovey, F.M. (1974) Material Properties in the Design of Trans-
to their Performance and Design, Clarendon Press, Oxford. ducer Flexibles, Ai Research Manufacturing Co, Phoenix, AZ.
Roark, R.J. and Young, W.C. (1975) Formulas for Stress and (plus illustrations to accompany above as separate).
Strain, McGraw-Hill, New York. (Editions since 1938). Trylinski, W. (1971) Fine Mechanisms and Precision Instru-
Rolt, R.H. (1929) Gauges and Fine Measurements, Vol. 2, Mac- ments – Principles of Design, Pergamon Press, Oxford. Trans-
millan, London. lated by A. Voellnagel from Polish original.
Schmidt, R. and DaDeppo, D.A. (1973) Variational Formulation Wahl, A.M. (1963) Mechanical Springs, McGraw-Hill, New York.
on Nonlinear Equations for Straight Elastic Beams. Journal of Weinstein, W.D. (1965) Flexure Pivot Bearings, Machine Design,
Industrial Mathematical Society, 23(2), 117–136. Part 1 June, 151–157, Part 2 July, 136–145.
Sclater, A. and Chironis N.P. (2001) Mechanisms and Mechanical Whitehead, T.N. (1934) Instruments and Accurate Mechanisms –
Devices Sourcebook, McGraw-Hill, London. Underlying Principles, Macmillan, London. (Also Dover Pubs,
Sydenham, P.H. (1982, 1983) Handbook of Measurement Science, New York, 1954).
Wiley, Chichester. Young, W.C. (1989) Roark’s Formulas for Stress and Strain,
Sydenham, P.H. (1984) Elastic Design of Fine Mechanism in McGraw Hill, London.
Instruments. Journal of Physics E: Scientific Instruments, 17,
922–930.
86: Elastic Regime of Design – Spring Systems
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
Trylinski (1971), and Wahl (1963) can be consulted for the
1 Helical Springs 548 theoretical design of the familiar cylindrical coil spring.
2 Spiral Springs 549 Howell (2001) is the definitive source in elastic mechanism.
Such springs are generally intended for use as pure tension
3 Flat-strip Springs 550
or pure compression flexure elements (Figure 1a and b), but
4 Managing the Degrees of Freedom in Elastic
they can also be used as torsional or bending elements.
Systems 552
The above references discuss the key spring design
5 Buckling Effects in Flexure Elements 552 parameters to obtain a spring with a given spring rate
6 Special Shapes of Flexure Elements 554 (compliance, stiffness), adequate range of deflection, ade-
7 Examples of Spring Flexure Elements 554 quate fatigue life, and given maximum stress levels in the
Related Articles 554 material, each applying for the range of materials normally
References 554 available. The more general machine design texts, how-
Further Reading 555 ever, do not usually consider such factors as the effect of
temperature on the spring rate or the need to reduce the
inherent mechanical hysteresis effects – two factors that are
vital to instrument design.
1 HELICAL SPRINGS Figures 1 and 2 show several of the many shapes and
forms that fine mechanism elastic spring elements may take.
Elasticity of mechanical components is used to obtain Helical torsion springs (Figure 1c) are formed into shapes
necessary degrees of freedom in the elastic regime of that are similar to the coil spring, but the forces are applied
fine mechanical design. The general principles of elas- to the coil as a torsion, rather than as longitudinal extension
tic design are given in Article 85, Elastic Regime of or compression. In this use, the basic mode of flexure of
Design – Design Principles, Volume 2. Comparison with the spring material is not torsional, as it is in the tension
the alternative kinematic methods is made in Article 80, helical spring, but occurs as bending action.
Principles of Fine Mechanics – Kinematic and Elastic In this use, the turns are required to rotate with respect
Designs, Volume 2 and Article 81, Principles of Fine to each other; practical design has to allow for any friction
Mechanics – Systems Considerations, Volume 2. Spring (the binding effect) that may occur when the coils slide
element design has regained interest in MEMS applications, against each other. Such springs are often made with spac-
see Article 162, Principles of MEMS, Volume 3. ing between turns. Another factor to be considered is that
The design of spiral (helical, coil) springs is well covered as the spring is twisted tighter in torsion the effective radius
in numerous commonly available sources because they form reduces. Long torsional helical springs may buckle; guides
part of the training of engineers in the general design of are often employed to support the spring, in which case
machines. For example, each of the following authors Chi- the added effects of friction must not be forgotten in sensi-
ronis (1961), Andreeva (1966), Faires (1955), Geary (1960), tive system design. Wahl (1963) suggests that buckling is
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Elastic Regime of Design – Spring Systems 549
Suggested reference and where all turns lie in a flat plane. The clock and elec-
sources tromechanical indicator restoring spring are in this group
Chironis (1961) of springs. These are generally referred to as hairsprings;
Andreeva (1966)
Trylinksi (1971) when the turns do not come into contact and are particu-
Wahl (1963) larly fine in dimension. Hairsprings can also buckle at high
(a)
deflections.
It is in this class of springs that the Neg’ator spring design
fits. This preformed spring was invented and reported in
as (a) 1952 by Votta (1952). It has the interesting and somewhat
hard to visualize characteristic that it can provide linear ten-
(b)
sioning action, exhibiting constant force with displacement
(as opposed to most springs in which tension increases with
displacement). By choosing a suitable bush size to support
as (a) it (Figure 1e), it is possible to tailor the design to provide
very slight negative or positive spring rates on the basis
of unusually high base loads. It is particularly useful as a
(c) means to counterbalance gravity-acting masses in instru-
ments. Arranged as a torque producing device, Figure 1(f),
instead of giving purely linear action, it forms a constant-
torque drive motor with a large range of output turns. It
as (a)
is marketed as the Tens’ator spring. Typical commercial
design data is available in AMETEK (1973).The dynamic
performance of this spring has been analyzed as a dynamic
motor in Sydenham (1969), where the commercially offered
width to thickness ratios and suggested spool diameters that
(d)
lie far from the optimum values needed to obtain minimum
Natural rotational inertia are shown. Given the optimum spring
Freely running shape AMETEK (1973) material thickness and width, the use of small diameter
bush Votta (1952)
Wahl (1963)
spools in such motors can provide very responsive torque
sources; energy stored can be dangerous while in use, and
(e) protection is needed to prevent damage, should a spring
break or come away from its end constraints.
The design of spiral springs is covered in Andreeva
as (e) (1966), Geary (1960), Trylinski (1971), and Wahl (1963).
Each of these sources covers some of the same material,
Free running
Output storage bush
but additionally it provides a somewhat different design
torque information. It is necessary to select the appropriate work
(f) for the task in hand. None of them appears to give a
coverage that caters for all likely needs; there are just too
Figure 1. Examples of coiled springs: (a) helical tension;
(b) helical compression; (c) helical torsion; (d) spiral torsion;
many spring options in existence.
(e) linear Neg’ator ; (f) torque Neg’ator . Trylinski (1971) concentrates on fine mechanism springs,
that is, those including hairsprings and main springs for
avoided if the spring is not turned at the torsional input by instruments such as clocks and watches. Geary (1960) is
more than 0.87 of a turn for length to diameter ratios that are useful for its general introduction and for the bibliogra-
short, the input rising to 1.25 turns for longer springs having phy provided. It provides the theory of torsional systems
the same diameter. This can only be taken as a guide; ref- using bifilar and multifilar suspensions. Wahl (1963) and
erences cited by Wahl should be consulted for more detail. Andreeva (1966) cover in general all types of springs.
Several highly sensitive scientific instruments make use
2 SPIRAL SPRINGS of a spiral form of spring, in which the turns are not in
contact and lie in a helix. Figure 2(a) gives an arrangement
Spiral springs are those wound springs in which flat mate- used in one gravity meter to suspend the proof mass.
rial is used (Figure 1d) that is curved in a decreasing radius Figure 2(b) is that of a somewhat similar use, but where
550 Measuring System Design
Suggested reference
sources
Detector
Source
Readout mirror
of optical
lever
Geary (1960)
Proof mass
(a)
Pressure into
fixed end
Trylinski (1971)
Mirror rotates on
closed end
(b)
Figure 2. Examples of coiled spring: (a) helical torsion in gravimeter; (b) helical Bourdon tube.
the form of a spiral quartz Bourdon tube is used for along the strip, increasing in amplitude as the force moves
ambient pressure measurement. Reference sources are given outward. The force-deflection characteristic depends upon
in the figures. the shape and thickness of the strip. It is nonlinear for a
parallel, uniform strip. To linearize the spring rate, it can be
tapered in thickness and width. Leaf springs used in vehicle
3 FLAT-STRIP SPRINGS suspensions are actually a very tapered cantilever spring
that is cut into sections and stacked in a pile. Their sliding
Flat strips used in bending and torsional modes of basic contact also provides damping. Early expensive vehicles
flexure, as with previously treated springs, can also be used further tailored the characteristic by tapering the spring
as energy storage elements or as supportive mechanisms. material in thickness as well as in width. The design of
In many applications, they serve both purposes together. these is straightforward. The strip can be regarded as a
With so many options reported, it appears most expedi- beam with one end (or two ends if needed) fastened rigidly.
ent, in this brief introductory review, to present schematic Another option for altering the rate characteristic is to let the
illustrations of a selection of options describing some of step progressively lean onto a shaped support as it deforms
them in detail to bring out certain design features. Figure 3 (Beggs, 1955). Figure 3(b) shows the strip used in the same
provides various views of flat-strip springs used to provide way as a beam with both ends supported. In this option, the
energy and support. Several applications use the flat strip ends can be simply supported so that rotation is allowed at
in a coiled fashion; it is not possible to entirely separate the ends. Deflections are not linear with load unless the
spring forms into unique groups. beam is suitably proportioned in thickness by the means
The first type shown in Figure 3(a) is the use of a mentioned above.
strip in which one end is secured so that rotation at When cantilever springs are used as small rotation
the fastening is prevented. Deflection progressively occurs locating mechanisms, the effective pivot point, called the
Elastic Regime of Design – Spring Systems 551
Suggested reference
sources
Force
Andreeva (1966)
Geary (1954)
Neubert (1975)
Fixed end Trylinksi (1971)
(a)
Force Eastman (1937)
Endreeva (1966)
Geary (1954)
Trylinksi (1971)
(b)
Flexure
(c)
as (a)
(d)
Geary (1954)
Jones (1951)
Jones/Young (1956)
Neubert (1975)
Flexures Thorpe (1953)
(4) Trylinski (1971)
Wahl (1963)
Weinstein (1965)
(e)
End blocks floating
on spring ends
Input drive to
Geary (1954)
top central limb
Jones (1951)
Jones/Young (1956)
Neubert (1975)
Trylinski (1971)
Member fastened
to frame
(f)
Figure 3. Examples of flat-strip springs: (a) cantilever; (b) beam; (c) single flexure pivot; (d) load cell; (e) parallel-strip slide
(uncompensated); (f) parallel strip (one form of compensation).
552 Measuring System Design
instantaneous center of rotation, moves in a curved path. length and is linear with width of spring material and
Cantilever flexures can also be used with the end of the elasticity modulus. Designs sometimes use two adjacent
beam being placed in torsion. A tensioned beam used this strips that are merged across into one strip. If the suspension
way with long length and narrow width is called a taut has to bear a downward acting load, the flexures can be
band suspension. stiffened with plates (to avoid Euler’s buckling) added to
Figure 3(c) shows a strip that bends while supporting a the central length of the strip (Geary, 1954).
load. As has already been mentioned, the bending stiffness Parallel-strip suspensions will deflect according to the
increases as the load supported is increased. Many expres- inverse of their stiffness values. Thus, a series connection
sions given in texts do not allow for this. The pivot action of a stiff unit driven by a more compliant unit acts to reduce
provides an instantaneous center of rotation that moves in motion in a positive, accurate, and stable dividing manner.
a circular path, as does a cantilever support. Geary (1954) Figure 4(a) shows the layout of strip flexures used to obtain
should be consulted for reference sources on this design. displacement reduction of 1000:1, with negligible friction
Trylinski (1971) and Eastman (1937) are also relevant. and force requirements. It was designed for displacing
Proof rings, which form the force to displacement con- an optical grating to a discrimination level of 0.08 µm,
version element of a load-cell type, are formed of shaped (Sim, Ross and Taylor, 1971). This application is typical of
rings, tubes, rods, and more, to which are attached a dis- many in which ingenuity is applied to obtain very effective
placement sensor. The sensor might be a simple mechanical mechanical mechanisms using flexures.
indicator or an electrical output device. In either case, the Having replaced the kinematic slide with a flexure equiv-
design aims to obtain linear or other specific conversion alent, at least for small amplitudes, it may seem obvious
characteristics by suitable choice of mechanical force load- that there might also exist a limited action replacement
cell element. Figure 3(d) shows a circular proof ring formed for the rotating bearing. This does exist and is known as
from a thick strip by machining it from a solid or by forming the crossed-strip pivot (there are numerous name variations
a flat plate. with ‘crossed’ used in each). Figure 4(b) shows the four-
strip hinge. This may take on the alternatives of three strips
when two are combined inside the center of the other two,
4 MANAGING THE DEGREES OF or a two-strip form in which they do not cross within their
FREEDOM IN ELASTIC SYSTEMS physical length, doing so at the theoretical extension of the
strips (see Geary, 1954).
Displacement of a table in one axis, while the other five The springs are generally arranged to cross at 90◦ , but
degrees of freedom are substantially held in constraint, can this is not vital. Considerable design options are available
be obtained by use of a parallel-strip suspension shown in with these. The instantaneous center of rotation moves such
Figure 3(e). These take several forms of which the simplest that the support can provide positive restoring or collapsing
variation is shown. Study of a given four-strip suspension moments (negative restoring force) when excursion reaches
will reveal that, as the amplitude of displacement rises, given points.
the table being supported does not follow a perfect linear Added to this feature is the possibility of preforming
translation but traces out a curved path in which it rotates the springs, altering the crossing point from the center
and rises and falls. These effects reduce as the length to higher or lower positions between the two surfaces,
of the strips increase with respect to the amplitude of and by forcing the springs into nonflat shapes by their
movement needed. clamping end conditions. Haringx (1949), Geary (1954),
Introduction of a symmetrical but oppositely applied and Trylinski (1971) give in-depth treatment of the crossed-
suspension, Figure 3(f), produces a differential error com- strip pivot. It would appear that only one company has
pensating system that can provide almost pure transla- marketed off-the-shelf crossed-spring assemblies as OEM
tion. These are called eight-leaf (or strip) parallel strip (Original Equipment Manufacturer) products, see Bendix
guides. Jones (1968), Trylinski (1971) and Neubert (1975) Corp (1975). It is necessary to design and manufacture them
show alternative physical arrangements of the compensated to order.
method. Jones (1951) and Jones and Young (1956) are basic
reading about parallel-strip suspensions.
It will be apparent that the suspension produces a reacting 5 BUCKLING EFFECTS IN FLEXURE
force as the table is displaced from the central rest position. ELEMENTS
The force to deflection ratio is constant over small ranges of
travel for both the four- and the eight-strip suspensions. The So far it has been suggested that care is needed to prevent
ratio follows cubic relationships for thickness and spring buckling through the application of a load greater than
Elastic Regime of Design – Spring Systems 553
Suggested reference
sources
Input
displacement Stiff strips
to lever arm
Sims et al (1971)
Single strip
pivot Attenuated displacement
of object holder
Weak strips
(a)
Moving body rotates
about dotted axis BENDIX CORP. (1975)
Geary (1954)
Haringx (1949)
Neubert (1975)
Trylinski (1971)
Weinstein (1965)
(b)
Eijk/Dijksman (1977)
Geary (1954)
Adjustment of sine spring
characteristic
(c)
Flexure strips Recess for
larger wheel
Flat surface for
rolling of smaller Geary (1954)
wheel Rolt (1929)
Tensioning
spring
Output
Input displacement
displacement
(d)
Figure 4. Examples of flat-strip springs: (a) displacement attenuator using strip suspension; (b) cross-strip pivot; (c) sine spring (used
here to reduce the restoring force of parallel-strip suspension); (d) rolling suspension.
the Euler buckling limit. This effect, however, can be put provided by a parallel-strip support; Geary (1954) also
to good use to form yet another spring element that can discusses these.
support considerable loads with very small spring rates Geary (1954) describes a buckled strip arrangement used
(including negative) without incurring the penalty of a large by the staff of the National Physical Laboratory. Their
initial deflection that occurs with a simple helical spring. application made use of the feature that a buckled beam
Figure 4(c) shows how Eijk and Dijksman (1977) used provides a rotating surface on which a mirror can be
the buckled-beam spring strip (also called a sine spring) mounted. As the end of the beam is buckled further by
to provide a negative restoring force to cancel the force translation, the mirror rotates to form an optical lever.
554 Measuring System Design
6 SPECIAL SHAPES OF FLEXURE A similar torsion action can be obtained using a tensioned
ELEMENTS taut band Figure 5(b).
A strip can also be employed to form a rolling suspension.
7 EXAMPLES OF SPRING FLEXURE
Figure 4(d) shows an example adapted from Rolt (1929)
(see vol. 1, p. 333 for greater detail). The strips here act ELEMENTS
to hold the roller onto a flat surface while allowing the
system one rotational degree of freedom. One spring strip Finally, in this short review of the innumerable possibilities
must be tensioned (by a spring or a gravity mass) to keep of spring systems it must be stressed that special shapes
the system taut. As is illustrated, the linear motion of the can often be devised for special purposes. Figure 5(c)
input is converted into a rotation that also can be given gain shows one case employed in a low-cost commercial humid-
by use of different diameter input and output rollers. ity meter to simultaneously provide tensioning force to
The bifilar suspension has already been mentioned; a stretched hygroscopic membrane and, as the membrane
Figure 5(a) shows a form of this. These take numerous forces the flexure to bend, to provide load-cell action by
forms in practice, ranging from the two round section wire driving four resistance-strain gauges mounted to its external
filaments shown to systems using multiple filaments. As a curved surfaces.
first approximation, these can be considered to have com- Flexure elements can play a very important part in
pletely flexible filaments, but a more detailed design would instrument design and offer an alternative to kinematic
be needed to consider their elasticity. Trylinski (1971) and design, that is not always the best option to choose.
Geary (1960) discuss their design in detail; numerous sci-
entific instruments have been built using such suspensions.
RELATED ARTICLES
Suggested reference
sources Article 80, Principles of Fine Mechanics – Kinematic
and Elastic Designs, Volume 2; Article 81, Principles
Filaments Geary (1960) of Fine Mechanics – Systems Considerations, Volume 2;
(May need to be Trylinski (1971) Article 85, Elastic Regime of Design – Design Princi-
regarded as having
elastic properties) ples, Volume 2; Article 162, Principles of MEMS, Vol-
ume 3; Article 164, Principles of MEMS Actuators, Vol-
ume 3.
Torsion
(a) REFERENCES
Tension applied to
retain band taut
Andreeva, L.E. (1966) Elastic Elements of Instruments, Israel
Program for Scientific Translations Ltd, Israel. Translation from
Russian.
as (a)
AMETEK. (1973) Design Data for Negator Constant Force
Torsion
Springs, Hunter Spring Division, Hatfield, PA.
Beggs, J.S. (1955) Mechanism, McGraw-Hill, New York.
(b) Bendix Corp. (1975) Free Flex Flexure Pivot Engineering Data,
Pub No 00U-6-752B, Bendix Corp, Electric and Fluid Power
Four resistance Div, New York.
strain gauges Drive for
for electrical mechanical Chironis, N.P. (1961) Spring Design and Application, McGraw-
output output Hill, New York.
Eastman, F.S. (1937) The Design of Flexure Pivots. Journal of
Tension and displacement input
Aeronautical Sciences, 5(1), 16–21.
(c)
Eijk, J. and Dijksman, J.F. (1977) Plate Spring Mechanism with
Constant Negative Stiffness, Internal Report, Fine-mechanics
Figure 5. Examples of flat-strip springs: (a) bifilar suspension; Group, Technische Hoschule, Delft.
(b) taut band torsion; (c) special design used in a relative humidity Faires, V.M. (1955) Design of Machine Elements, Macmillan,
meter. New York.
Elastic Regime of Design – Spring Systems 555
Geary, P.J. (1954) Flexure Devices – Pivots, Movements, Suspen- on Production Science in Industry, August, Institution of Engi-
sions, Sira Institute, Chislehurst. neers, Melbourne, (pp. 57–68).
Geary, P.J. (1960) Torsion Devices, Sira Institute, Chislehurst. Sydenham, P.H. (1969) Manual and Automatic Large-scale
Dimensional Metrology, PhD Thesis, University of Warwick.
Haringx, J.A. (1949) The Cross-spring Pivot as a Constructional
Element. Applied Scientific Research, A1, 313–332. Trylinski, W. (1971) Fine Mechanisms and Precision Instru-
ments – Principles of Design, Pergamon Press, Oxford. Trans-
Howell, L.L. (2001) Compliant Mechanisms, Wiley, New York.
lated by A. Voellnagel from Polish original.
Jones, R.V. (1951) Parallel and Rectilinear Spring Movements.
Votta, F.A. Jr. (1952) The Theory and Design of Long-deflection
Journal of Scientific Instruments, 28, 38–41.
Constant Force Spring Elements, Transactions of ASME, 74,
Jones, R.V. (1968) More and More about Less and Less. Proceed- 439–450.
ings of Royal Institution of Great Britain, 43(202), 323–45.
Wahl, A.M. (1963) Mechanical Springs, McGraw-Hill, New York.
Jones, R.V. and Young, I.R. (1956) Some Parasitic Deflexions in
Parallel-spring Mounts. Journal of Scientific Instruments, 33,
11–15.
Neubert, H.K.P. (1975) Instrument Transducers – An Introduction FURTHER READING
to their Performance and Design, Clarendon Press, Oxford.
Rolt, R.H. (1929) Gauges and Fine Measurements, Vol. 2, Thorpe, A.G. II. (1953) Flexure Pivots – Design, Product Engi-
Macmillan, London. neering, 24(2), 192–200.
Sim, P.J., Ross, H.A. and Taylor, R.J. (1971), Photographic Grat- Weinstein, W.D. (1965) Flexure pivot bearings, Machine Design,
ings for Linear Measurement, Harold Armstrong Conference Pt 1, June 151-7, Pt 2 July, 136–145.
87: Elastic Regime of Design – Plates and Bimorphs
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
as providing the net effect, summed in the center, of
1 Flat-plate Flexure Elements 556 one or more cantilever beams. The spiral case shown
2 Pressure-bellows Diaphragms 557 can be mentally unwound to form a long thin beam sup-
3 Capsules for Pressure Sensing 557 ported only at one end. In some designs, the plate is not
supported around the periphery but at points within the
4 Nonmechanical Elastic Systems 558
plate. There appears to be no published analytical design
5 Bimorph Flexure Systems 559 information on their design. Trylinski (1971) and Neubert
6 Multimorph Flexures 560 (1975) devote space to these guides but do not provide
Related Articles 560 analytical expressions for other than the simple continu-
References 560 ous plate with point or distributed loading. Gibbs (1957)
appears to be one of the few reports specifically devoted
to the subject, but it contains little more than is given
in Trylinski (1971).
1 FLAT-PLATE FLEXURE ELEMENTS
Thin plates used for this purpose are called slit mem-
Relatively thin plates can be used to provide precision brane guides or diaphragm spring guides. They are used
guidance of the center point of the plate; the plate can also extensively in such applications as loudspeaker moving-
simultaneously act to contain pressure of a fluid. coil supports, geophone suspensions, vibrator suspensions,
The simplest arrangement (Figure 1(a)) uses a flat, thin, and anywhere that relatively small amplitude, frictionless
flexible sheet of material that is fastened around the periph- translation is desired. By careful design, they can oper-
ery either as a fixed or a simply supported edge in which ate for millions of cycles at high frequency, providing
edge rotation is allowed. In the center, the member to high stiffness across the diaphragm and low stiffness in
be guided is fastened. As these provide a rapidly rising its plane.
restoring force with deflection, they are generally only They do not appear to be marketed as a stock item and
applicable to guide very small motions. Trylinski (1971) like most flexures have to be manufactured as needed.
provides expressions that show that those provide a linear Slits are formed by photomechanical etching methods,
force–displacement relationship for small movements. by spark machining, or by plain stamping where quality
Two basic methods may be employed to obtain a linear does not need to be very high and the production run
force-displacement characteristic over a larger range of is large.
movement with a reduced spring rate; that is, those that Slits may well provide greater compliance and range
are more easily deflected. but they remove the property of the plate to retain fluid
The first method is to slit the membrane in some appro- pressure. In cases in which guidance alone is needed, this
priate way so that it can deflect further. Figure 1(b) shows may be an advantage to prevent air pressure pumping. Some
several possible shapes. As a very approximate guide to applications may deliberately use a closed membrane to
estimating the compliance, designs can often be regarded exclude moisture and dirt.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Elastic Regime of Design – Plates and Bimorphs 557
Rigidly Force
supported
circumference
Deflected
shape
Force
Free to
rotate
(a)
Single start Multiple start Circular Radial
spiral spiral
Point fixing at •,
(b) not outer edge
Edge clamped Force may be exerted
as uniformly distributed
load or at a central point
(c)
Figure 1. Flexure membranes: (a) solid diaphragms, (b) slit diaphragms, (c) corrugated plates and capsules.
2 PRESSURE-BELLOWS DIAPHRAGMS simpler to implement; work previously needing much actual
prototype manufacture and testing has been bypassed by
the use of rigorous mathematical modeling of the pro-
If pressure-tightness is essential, as it is in pressure gauges,
files needed.
some other method of making the diaphragm more elastic
is needed. It can be obtained by pressing corrugations into
the plate. Several commonly used cross-sectional shapes
are shown in Figure 1(c). 3 CAPSULES FOR PRESSURE SENSING
Sealed diaphragms are also called membranes; Trylinski
(1971) calls them pressure controlled elastic elements. To obtain greater deflection for a given force, the
Presumably, because sealed diaphragms are used so diaphragms may be joined to form capsules and stacks of
extensively in pressure sensors, a product line that forms capsules, as shown in Figure 2(a). Bellows also provide
a large part of process instrument sales, there has arisen a similar increase in rectilinear flexibility (Figure 2(b)). As
the extensive need of pursuing their design in considerable the characteristics of the materials that must often be used –
depth. Design is reviewed in analytical terms in Trylinski because of temperature limits and the required high strength
(1971). CDA (1953) provides practical detail of applica- levels – do not always have the desirable low thermal
tion as do Beggs (1955) and BELL and HOWELL (1974). coefficient of elasticity, pressure systems often need some
Computer-aided procedures developed in the Department of form of thermal compensation built into the stack assembly.
Applied Physics of The City University, London have made Aneroid barometer bellows and capsules are sealed with
the design procedures of compliant plates considerably internal pressure; the choice of fluid, pressure, and quantity
558 Measuring System Design
rises sharply as the two magnetic poles of unlike kind
Displacement out
come closer together. Magnetic suspension systems need
Edges welded
or soldered
some form of stabilizing arrangement to make them stat-
ically stable. Earns Haw, in 1839, showed theoretically
that the inverse square law for force and distance fields
(magnetic and electric fields are such) cannot inherently
support a fixed magnitude pole in a stable manner (Geary,
1964). Automatic control is employed to alter the force
with distance as a closed-loop system. Further stabilizing
mechanisms may be needed to ensure that one pole does
Pressure in (fixed end)
(a) not slide away from the desired position. This can be visu-
alized when attempting to sit (levitate) a pole over the top
Displacement
of another pole of like polarity.
An advantage of magnetic springs is that they can be
Compensating made to act through sealed walls of containers. An example
Bellows spring
of an application might be as the connection and weighing
medium for a sensitive chemical weighing balance. The
process of measurement can be achieved by altering the
field strength of an electromagnet so that the balance returns
to a given datum in the manner depicted in Figure 3. The
Pressure in
(b) (fixed end) electric current needed to achieve the balance state is related
to the weight of the mass that is supported.
Figure 2. Flexure membranes: (a) capsule stack and (b) bellows. Electric fields also exert forces so they can be used as
elastic elements in the design of an instrument. This method
can be made to assist compensation; internal helical springs has been used in at least one gravimeter to restore the
are sometimes used. balance position.
In pressure sensors, diaphragm deflection is transduced Further detail on magnetic and electric suspensions is to
by direct linkage to a mechanical device or a sensor might be found in Geary (1964). Weighing balance applications
be actually attached to the membrane. In the latter case, are detailed in Gast (1968), Gast (1974), and Wolsky and
membrane design must allow for this. Zdanvk (1980).
Although not made from plates, the subject of this As the name implies, an air spring is a compliant move-
section, it is appropriate to include mention of Bourdon ment obtained by the use of air, contained in some way,
tubes. These are flexible tubes formed such that pressure to support a given load. The automotive pneumatic tire is
differences between the inside and the outside cause the the most obvious example of air spring being used in that
tube to deflect. Trylinski (1971) provides design formulae application to assist isolation of vibration.
for these. Deflection is harnessed to drive a pointer or an
appropriate electric sensor. Electromagnet
Error
Reference
4 NONMECHANICAL ELASTIC SYSTEMS +
N −
Magnetic field
Electric and magnetic fields have been used to provide
variable spring rates and so have columns of air. Fluids that
S
are easily compressed are more suitable (where springiness Datum
is needed) than the virtually incompressible fluids such as sensor
water and oil. Thus, an oil hydraulic bearing is very stiff, Freely suspended
having little compliance, while an air bearing will be softer. magnetic material
Magnetic suspensions are reasonably common in instru-
ments. They can be formed from permanent magnets
Evacuated shell
arranged in a suitable manner using additional electromag-
netic coils so that spring rate and stability can be controlled. Figure 3. Schematic of force-balance, magnetic suspension
In either case, it will be clear that the force of attraction weighing technique.
Elastic Regime of Design – Plates and Bimorphs 559
Air springs are capable of substantial load bearing with- Direction of bending
out large deflections, and they can easily provide low natu- with increasing
Passive side temperature
ral frequency, second-order suspension systems. They have
been used extensively in vibration isolation of machinery. Active side
The spring usually takes the form of a round flat flexible Principle
bag that is pumped up to a suitable pressure. Disk clicks upward to open
Air springs have no resonant effects within the spring electrical contact with fast
make-break action An application
itself (as do mechanical coil springs). Their rate can
be made to stiffen with increasing load, being nonlin-
ear devices. This means that second-order systems formed
with them remain at substantially the same natural fre- (a)
quency with changing load. They are, however, affected
by ambient temperature, and they require a relatively large + Bend up
or down
amount of maintenance work. They are generally regarded
as being suitable for only heavy load situations. The reader
should consult Jones (1967) and Dorien-Brown (1971) for Bimorph Multimorph
further references and an introduction to this form of elas- − side view section
(b)
tic system.
Bend one way
only
5 BIMORPH FLEXURE SYSTEMS
Magnetostrictive
If two materials (Figure 4(a)) having different thermal Nonmagnetostrictive
expansion rates are placed side by side with one end of the
Electromagnetic coil
parallel combination fixed securely, temperature changes to (c)
the element so formed will bend traversely due to differ-
Figure 4. Differential expansion bending and twisting devices:
ential expansion between the two elements. Such elements
(a) bimetals, (b) ceramic bimorph and multimorph, (c) magne-
are called bimetallic devices if made from metals. A simi- tostrictive bimorph.
lar device can be formed (Figure 4(b)) using two pieces of
piezoceramic that are fastened together and have deposited
electrodes placed along their outer edges and the central
Traverse deflection can be converted into rotation through
interface. These are termed bimorph-benders, or twisters.
the use of spiral spring forms. These also can be used to
Bimetals bend with temperature, converting tempera-
obtain greater deflection. However, as the deflection per
ture change into displacement. Bimorphs convert electrical
unit thermal change parameter is increased, it will usually
energy into displacement. Bimorphs involve a reversible
be found that the stiffness of the bimetal element reduces,
process – that is, deflection will generate electricity – but
making it less capable of driving a load without distortion.
bimetal elements are irreversible.
Preloading the bimetal element against a stop (prop) pro-
Bimetals are made by soldering or welding thin plates of
suitable material together, the final shape being stamped as vides a means of adjusting the datum at the temperature
strips, disks, or whatever shape is needed. The side having position at which the element begins to respond.
the higher thermal coefficient of expansion is called the The design of bimetal elements is covered in
active side; the reverse is known as the passive side. As detail in Timoshenko and MacCullough (1959) and
no material has an absolutely zero expansion characteristic in Trylinski (1971).
and metals generally possess only positive coefficients, the Bimorph devices are less common than bimetal ele-
net effect is always that of a combination of two positive ments and information about them is a little harder to
coefficient expansions. The forces of this expansion are locate. Neubert (1975) gives a short account of how piezo-
considerable, and bimetal elements can often directly drive electric sandwiches can be made to provide bending, twist-
an electrical contact or dial indicator. ing, and beam deflection action, as well as bending a
To obtain the greatest deflection effect, the metals used plate across the diagonal or lifting the free corner of a
need to be, for the passive side, one with the lowest practical three-corner mounted plate.
coefficient (which is usually Invar), and on the active side, Design information is usually available from the suppli-
one with a high coefficient (stainless steels, brasses, and ers of the materials, Brush-Cleveite being one source of
special alloys). application notes.
560 Measuring System Design
6 MULTIMORPH FLEXURES RELATED ARTICLES
A variation of the bimorph is the multimorph. These can Article 57, Static Considerations of General Instrumen-
provide greater elasticity and are said to be easier to tation, Volume 1; Article 58, Description of Accuracy,
manufacture. Instead of sandwiching two strips together, a Linearity, and Drift, Volume 1; Article 59, Introduc-
single strip is made with a row of fine holes formed through tion to the Dynamic Regime of Measurement Sys-
the centerline. These are filled with conductive material to tems, Volume 1; Article 80, Principles of Fine Mechan-
form the inner electrode. The outer faces are coated with ics – Kinematic and Elastic Designs, Volume 2; Arti-
electrode material in much the same manner as for the true cle 86, Elastic Regime of Design – Spring Systems, Vol-
two-slice sandwich bimorph. ume 2
Other forms of differential bending devices could be
formed on this principle but they do not appear to have been
used. For example, a nickel strip sandwiched (Figure 4(c))
REFERENCES
to a stainless steel side would bend in the same direction
with an increasing magnetic field of either polarity; the Beggs, J.S. (1955) Mechanism, McGraw-Hill, New York.
nickel, but not the stainless steel, is magnetostrictive. This
BELL and HOWELL. (1974) The Bell and Howell Pressure
is the magnetic equivalent of the ceramic bimorph. It would Transducer Handbook, Bell and Howell, CEC/Instruments Div,
also give reversible characteristics. Pasadena, CA.
Bimetal and bimorph devices can be used to provide CDA. (1953) Copper in Instrumentation, Pub No 48, Copper
compensation for the physical attribute that suits their input Development Association, Radlett.
energy form. Bimetal elements are often used, for example, Dorien-Brown, B. (1971) Air Springs for Vibration Isolation,
to correct diaphragm-type pressure gauges. They are also Harold Armstrong Conference on Production Science in
used to trip circuits when the ambient temperature produced Industry, August, Institution of Engineers, Melbourne,
by the direct true ambient or by the use of a heating coil or pp. 213–225.
direct heating causes the strip to deflect. Article 88, Error Gast, T. (1968) Conversion of Measuring Values for Mass,
Sources in Fine Mechanics, Volume 2 on error sources Density and Flow into Proportional Frequencies with the Aid of
and their reduction, provides more information about the Free Magnetic Suspension, Industrial Measurement Techniques
for On-line Computers, In IEE Conf Pub 43, June, Institution
use of thermal differential methods.
of Electrical Engineers, IEE, London.
The dynamic response of such differential devices may
Gast, T. (1974) Vacuum Micro Balances, their Construction and
be important. Bimetal devices are relatively slow to act
Characteristics. Journal of Physics E: Scientific Instruments, 7,
since they must transfer thermal energy to obtain deflection. 865–875.
They will be substantially first-order systems in response
Geary, P.J. (1964) Magnetic and Electric Suspensions, Sira Insti-
because the thermal storage and dissipation rate combi- tute, Chislehurst.
nation is much slower than the rate at which the strain
Gibbs, D.F. (1957) Spring Diaphragms. Journal of Scientific
energy transfers. Clearly, the smaller the element the faster Instruments, 34, 34–35.
the response, and physical size limits would generally pro-
Jones, R.V. (1967) The Measurement and Control of Small
vide these elements with time constants in the order of Displacements. Physics Bulletin, 18, 325–326.
seconds. Miniature thick and thin film devices are able
Neubert, H.K.P. (1975) Instrument Transducers – An Introduction
to respond much faster but at the expense of less power to their Performance and Design, Clarendon Press, Oxford.
drive capability.
Timoshenko, S. and MacCullough, G.H. (1959) Elements of
Bimorphs are inherently high frequency devices. As the Strength of Materials, Van Nostrand, NJ. (Numerous editions
rate of energy change for the electrical and the strain since 1935).
energies transfer is of the same order, these will provide Trylinski, W. (1971) Fine Mechanisms and Precision Instru-
second-order response with high natural frequencies in the ments – Principles of Design, Pergamon Press, Oxford. Trans-
kilohertz and higher region; low damping is a feature. They lated by A. Voellnagel from Polish original.
are not, of course, a replacement for the bimetal since the Wolsky, S.P. and Zdanvk, E.J. (1980) Ultra-microweight Deter-
two devices operate with different input energy forms. mination in Controlled Environments, Wiley, New York.
88: Error Sources in Fine Mechanics
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
Regime – Fasteners, Bearings, Volume 2; Article 84,
1 Fundamental Limits in Fine Mechanical Kinematical Regime – Rotary Motion, Volume 2; Arti-
Design 561 cle 85, Elastic Regime of Design – Design Principles,
2 Creep and Hysteresis in Mechanical Volume 2; Article 86, Elastic Regime of Design – Spring
Mechanisms 562 Systems, Volume 2 and Article 87, Elastic Regime of
3 Geometrical Error Factors in Fine Design – Plates and Bimorphs, Volume 2 it is now appro-
Mechanism 562 priate to discuss the error regime of designs.
4 Sine and Cosine Errors, and Centering Error In any instrument design, there exist certain design
in Fine Mechanism 563 factors that cannot be eliminated and must be worked
5 Influence Parameter Errors in Fine within. All physical processes inherently contain physical
Mechanism 564 phenomena that provide a limit to sensitivity or stability,
which ever is the issue. In the majority of mechanical
6 Temperature Error in Mechanical Systems 565
designs, this is not likely to be of great significance.
7 Reducing Thermal Errors in Fine
Research material on the detailed nature of fine mechanical
Mechanisms 565
errors reached its highs well before the advent of electronic
8 Thermal Compensation Methods in Fine
forms of publication. Much of the research material was
Mechanism 566
published before electronic solutions overtook mechanical
9 Temperature Error in Elastic Elements 567 means from the 1950s. Thus, sources may be hard to
10 Miscellaneous Errors in Fine Mechanism 568 locate – but they are often worth finding for they provide
Related Articles 568 a wealth of practical information on less popular topics.
References 568 In mechanical systems, the prime fundamental limit
to sensing detection is Brownian movement, a random
motion of the molecules forming the substance of the parts
concerned. The motions of the molecules are averaged by
1 FUNDAMENTAL LIMITS IN FINE the mass of the part, so larger components exhibit less
MECHANICAL DESIGN overall dimensional change due to this source than smaller
parts. It has been extensively studied in relation to electrical
Having covered the overall design issues relevant to the indicating galvanometers, in the period before the 1950s,
operational requirements of fine mechanical systems in and became of interest again in more recent times in the
Article 80, Principles of Fine Mechanics – Kinematic hope of detecting gravity wave radiation. Chapter 17 in
and Elastic Designs, Volume 2; Article 81, Princi- Bleaney and Bleaney (1957) provides a basic introduction
ples of Fine Mechanics – Systems Considerations, to the theoretical considerations. Braddick (1966) discusses
Volume 2; Article 82, Kinematical Regime – Members noise sources in experimental setups. Further references are
and Linkages, Volume 2; Article 83, Kinematical given in Sydenham (1979) where the importance of this
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
562 Measuring System Design
limit in the evolution of electromechanical instruments has Creep is a still more elusive parameter. Clearly releasing
been reviewed. locked-in stresses by annealing and other stress relief
The bibliography given in Logan (1973) contains refer- methods helps. Weathering for prolonged periods is also
ences to many of the published papers on gravity wave used where time permits, for it is known that the creep
detectors. Jones (1967, 1968) contains useful practical rate generally reduces with time. A part considered to be
accounts of the limits of measurement sensitivity. creep-free, however, may change its creep characteristics
To provide a rather crude estimate of the level of dimen- quite markedly if remachined or worked in some way.
sional noise that Brownian motion effects will generate Sydenham (1973) is concerned with the long-term stability
in a mechanical component, some published examples are of tensioned Invar and fused silica.
quoted. These, however, must be accepted only in the full The difficulty faced by the instrument designer is that
context as given in the quoted reports. Jones (1967, 1970) the vast majority of creep and hysteresis data is com-
estimates that the root-mean-square length change of a 10- piled for highly stressed parts, that is, those operated at
mm brass block of 1 mm2 area is likely to be about 10−16 m high temperatures and large stress excursions, such as are
when averaged over a 1 s integration time. This level of found in aircraft turbine blades. Instrument designers, on
change is fine for most requirements, but it will be a lim- the other hand, are generally interested in comparatively
iting issue for those mechanical sensing systems that are low stresses and well controlled, around ambient tempera-
taken to the limit of possibility. tures. The extreme difference between the two cases makes
As a second example, Usher (1973), a review of seis- extrapolation from one zone to the other a very unreli-
mometer designs contains an estimate that a seismometer able procedure.
with a mass of 10 gm, a natural period of 1 s, a mechanical As a guide, it is generally good practice to stress materials
quality factor, Q, of 100, and operated at room temperature (those that must be stable) to no more than a tenth of their
will possess a noise equivalent acceleration of 3 × 10−11 g. elastic limit. Brasses are considerably better than steels in
He shows that the factor for reducing this in that design as far as creep is concerned in instrument parts. It is not
is the damping, which should be kept low. The natural possible to provide any really worthwhile rule of thumb
frequency of the spring-mass system is not a factor in deter- guides for these two parameters.
mining the amplitude of the Brownian motion noise. In Both of these factors are given a very poor account in
these instruments, this noise source must be considered in instrument design texts, very little information is provided
order to create satisfactory instruments. on both these factors, presumably because of the lack of
Showing up in yet another form, in a sensitive electro- general knowledge about them. Trylinski (1971) discusses
magnetic galvanometer, the source there might give rise hysteresis in instruments, but not in numerical terms.
to noise equivalent currents of around 10−12 A. Further Creep is often finally accepted as a factor that cannot be
examples are to be found in Barnes and Silverman (1934). reduced to the levels really desired. Being largely (an often
incorrect assumption) systematic and adequately linear with
2 CREEP AND HYSTERESIS IN time, corrections can be made in the measurements using
a sensing mechanism that creeps. A prime example is the
MECHANICAL MECHANISMS exploration gravity meter in which an established, return to
start, routine is set up with respect to time and position of
Although not strictly a truly fundamental limit, the effects
readings; the drift can be assessed and removed from the
of creep and hysteresis of mechanical materials often act
data taken in the field.
as an effective fundamental limit. There are practical levels
to which any amount of better design, external influence
parameter control, and material processing cannot better.
Despite considerable research effort seeking to quantify
3 GEOMETRICAL ERROR FACTORS
the fundamental mechanisms of creep and hysteresis in IN FINE MECHANISM
materials, there does not yet appear to be available an
adequate model of the underlying process. Many texts are Attention is now turned to these features of a mechanical
available that summarize the current state of knowledge. design that can be controlled in order to improve the
Design to reduce hysteresis and creep is generally based performance and accuracy of the measuring apparatus.
upon experimental data about the material used, its strain A clearly identifiable group are those sources of error
history and the experience of the designer. that relate to the geometrical features of the mechanical
It is suggested that hysteresis can be reduced by limiting assembly. These are almost all parameters that can be kept
the mechanical stress excursion to the smallest allowable in control by careful design of the assembly and through
cycle amplitudes and by the use of low hysteresis materials. adequate selection of components, materials, and sizes.
Error Sources in Fine Mechanics 563
Mechanical design of instruments invariably rests upon Interval to be determined
the assumption that there exist points in the mechanical
space framework that are invariant in position with respect Unknown
standard
to each other. They are often merely drawn as crosshatched (a)
lines in schematic diagrams of the system.
In reality, all materials are elastic and change dimension Unknown
to some degree with force exerted on them. It is, there-
fore, important to ensure that such elements are, indeed, Travelling
sufficiently strong, inelastic, and stable. microscope a a
Stresses occurring in manufacture, testing, transport, and (b)
assembly are often far in excess of those experienced by
the component in its working state. For example, there is Figure 1. Abbe’s principle. In (a), the unknown and the standard
are compared in a direct manner but in (b), errors may exist in
need to finally align a measuring machine in its application,
the projections of the scale by the microscope.
for test bed results cannot be relied upon to be the same.
Bolting down stresses can change the geometric shape.
Transportation might have imposed distortion of the frame. texts about instrument design. Basically, it states that a
A second form of geometric error sources is the dimen- measurement is generally best performed by applying the
sional relationship change that will occur because of the measuring device to the object in as direct a manner
actual dimensions of individual parts lying in an error as possible. For example, in the field in which this
band given by the tolerances to which the parts have been was first expounded – mechanical dimension measuring
made. Tolerances should be selected for the various sur- equipment – it is preferable to determine the length of a part
faces of a component according to the assembled geometry using a comparison method that introduces as little error
required. There is an art in tolerancing; it comes largely due to the comparison process as is practicable. Figure 1
with experience. Whereas, to overspecify may well con- contrasts the use of a scale to measure the length of a part
trol the assembly dimensions more rigidly, it makes the with the method in which a traveling microscope head slides
instrument far more expensive. along a scale. Where the scale markings and the interval
Standards, such as BS 1916 on limits and fits for engi- that must be measured can be placed in actual contact,
neering, provide the starting point to the selection of realis- the reading cannot include any error from the comparison
tic tolerances for given situations. This aspect is generally process. (Of course, it could contain error of the scale or
included in courses on mechanical engineering. Tolerancing from its misreading.) In contrast, the traveling microscope
is usually a given feature of mechanical CAD tools. arrangement can introduce error by way of the slide not
Related to tolerance error is the gradual loss of accuracy being straight, causing the microscope to project the scale
that might occur if a wear-true philosophy has not been incorrectly on to the unknown.
adopted. Wear in pivot points and along slides will change Whereas Abbe’s Principle might be the ideal to aim for,
the effective geometry of the instrument, thereby possibly it is often not practicable to adhere to it. If not, then other
degrading the performance. geometrical errors must be considered.
Such matters, plus other general aspects of design, are
gained largely by direct experience. Several authors in
the past have endeavored to pass on that experience as 4 SINE AND COSINE ERRORS, AND
descriptive guides to general instrument design. These CENTERING ERROR IN FINE
include Braddick (1966), Glazebrook, Darwin and Mason MECHANISM
(1923), Rolt (1929), and Whitehead (1954).
Rather more specific are errors caused by assuming that Angular variations introduced by the mechanism embodied
location points, such as a center or a sliding perpendicular, in the traveling microscope example given in Figure 2 need
are provided in a perfect manner by a bearing or a slide, consideration. It depends upon the effect on the length being
respectively. determined or required to be held constant.
If the same microscope is used at each end of the length
and the slide is straight (Figure 2a), any fixed angular error,
3.1 Abbe’s principle α, in the mounting of the microscope head will not cause
error in the length determination because it is equal at each
Abbe’s Principle, although well known among practitioners, end. If, however, the microscope mount is perpendicular to
does not seem to have been defined in English language the slide at one end and the slide causes the head to rotate
564 Measuring System Design
l 4.1 Centering error
Unknown
α α Where the measurement of location depends upon a bearing
Microscope moved position 1 to 2 around which motion takes place, there may arise errors of
Standard centering. It is common knowledge that such errors can be
l reduced by reading the scale on opposite sides, taking the
(a) 1 2 average of these as the correct value. Whitehead (1954)
devotes space to this problem, pointing out that it is not
l
always necessary to use the best possible bearing tolerances.
He provided an analytical study of the magnitude of
α d errors. Many instruments are made to be read at least at
diametrically opposite places. Indeed, some arrange more
than this number of averaging processes to be carried out
in provision of the output value. Whitehead, however, only
l −d sinα
(b) considered the use of a centering bearing for a radial
True length = l cosα scale. In linkages, the error effect must be considered from
α
other viewpoints. Braddick (1966) also provides material
Indicate
d length
on centering errors.
,l Although the above error sources have been illustrated
(c) here in terms of the traditional, manually read instruments,
Figure 2. Sine and cosine errors (a) error of microscope mount-
it will be found that these error sources arise frequently in
ing not significant, (b) sine error occurs at one end due to rotation the design of the mechanical features existing in transducers
of microscope, (c) cosine error. of the modern form.
by angle, α, at the other end (Figure 2b), an error equal 5 INFLUENCE PARAMETER ERRORS
to dsinα will arise. As this error is directly additative, and IN FINE MECHANISM
because sines of small angles are significant quantities in
such cases, quite small angular changes can give rise to The very best of internal designs can only provide satis-
relatively important error magnitudes. factory operation if variations in external influences either
When measuring a length (Figure 2c) between two end lie within tolerable limits or can be handled by the internal
intervals instead, as arises with an inside screw micrometer design arrangement.
(the above sine error case occurs in vernier calipers), then All systems are influenced by changes in ambient condi-
a second kind of error source can occur. If the part to be tions to some extent. It is the relative magnitude of these
measured is not aligned with the central axis, the length effects that matters. Temperature, ambient air pressure, rel-
will be lcosα, not l as indicated. This is not as significant ative humidity, magnetic and electric fields, acoustic noise
as the sine error of the previous case for if the angle α is level, mechanical vibration, chemical attack, and more
small, which it usually is, the lcosα is close to the same as can each possibly give rise to error in operation of the
α. Another example where this error arises, is in the setting installed instrument.
up of a distance measuring ranger. The actual alignment It is common and established practice to test an instru-
need not be very accurate, because the error caused by ment toward the end of the design cycle for its tolerance to
misalignment will not be significant. such effects. Calibration is often accomplished taking into
Whereas cosine errors are generally easy to control, account given levels of vibration of influence parameters. It
they must, nevertheless, be calculated to ensure that the is, however, often overlooked that the test conditions may
approximation holds to the degree of accuracy that is not be reproduced in the final use situation, where yet more
required. Sine errors, similarly, must be controlled by influence parameters may be in existence.
ensuring that the angles remain within tolerable limits under The designer of an instrument must know which of
practical use. As d is reduced, in Figure 2b, it will be seen these influence quantities will exist in use and how to
that Abbe’s Principle is approached. allow for them. There are two approaches. One is to
Errors of this form are discussed but only briefly under control the internal environment of the instrument by the
the principle of alignment in Rolt (1929) and in Brad- use of a suitable enclosure; for example, the internal
dick (1966). evacuation in a vacuum weighing balance that avoids
Error Sources in Fine Mechanics 565
errors of buoyancy and provides some degree of thermal uniformity results as speedily as is required. In practice,
isolation. The alternative is to design each element of the these simple rules are very difficult to adhere to.
system, such that the overall system can tolerate significant Consider first the problem of thermal uniformity. The
enough influence parameter variations. An example here materials that are most suited for construction, such as
would be to use low thermal expansion materials plus the steels, aluminums, and brasses, do not have very high
certain mechanical structural arrangements (that will be thermal conductivities. They also possess comparatively
discussed below) to reduce thermal differential expansion high thermal expansion coefficients. Lower coefficient
error to within tolerable limits for the expected design materials generally have still lower thermal conductivities.
temperature variations. Heat is transferred by conduction, convection, and radia-
In practice, one or the other, or a combination of these tion. Knowing the constraints imposed by the materials that
two alternatives is used. Probably the hardest part of have to be used – good thermal conductors are not the best
the task is probably to properly and adequately decide of structural materials – it is possible to arrange for heat
what influence quantities are relevant and their magnitudes transfer to be improved through the use of design layout,
of change. shape, cross-sectional size, and similar design possibilities.
6 TEMPERATURE ERROR IN 7 REDUCING THERMAL ERRORS
MECHANICAL SYSTEMS IN FINE MECHANISMS
Without doubt, the most prevalent influence parameter in As a guide, design for good thermal uniformity will aim
the mechanical design of instruments is temperature change. to provide good conduction inside the instrument with a
Basic materials with absolutely zero thermal coefficient high thermal barrier attenuating external changes through
expansion do not exist. Furthermore, materials with the the use of some form of integrating enclosure. Enclosure
lowest values are not always suitable for the manufacture design can also be most sophisticated. Reference will be
of instrument parts. It is, however, possible to produce made to that later.
instruments from the relatively high coefficient materi- To obtain good internal uniformity, all parts should be
als, which are reasonably temperature insensitive. Before made to maximize conduction through the use of high
going into the methods that might be used, it is instructive thermal conductivity materials, large cross sections, good
first to examine the general thermal design of a mechani- thermally conducting joints, and optimal arrangement to
cal instrument. keep conduction paths as short as possible.
Thermally controlled jacketing and mechanical compen- Conduction and convection can often be improved by the
satory mechanisms are often used to overcome temperature use of a stirring arrangement such as a fan or stirrer paddle.
error. The value of each, generally, rests on an underlying The internal fluid, in such cases, will be a suitable liquid
assumption that the temperature is uniform within the or gas that will assist heat transfer. The transfer of thermal
instrument. This is often far from the case. It is not ade- energy is, however, often decided more by the heat transfer
quately recognized that thermal heat flow is not an instanta- film coefficient existing at the interfaces than it is by the
neous process. Any change of forcing temperature on some fluid’s internal properties.
part of an instrument will give rise to changes in temper- Sometimes the design aims to prevent convection taking
ature of all of its parts at varying rates. If, in exacting place, thereby aiming to maintain the temperature constant,
designs, the instrument is regarded as being a thermally if not uniform, within the instrument. It is a physical fact
dynamic rather than static assembly, then better design that the smaller the convecting cell size, the lower the
should result. rate of convection. Below a certain cell size and shape it
If the external temperature does not change, then only ceases completely. Because of this, it is often helpful to fill
internal sources of heat, such as arise from electronic cir- interspaces of an instrument with loose thermal insulation
cuitry enclosed in the housing or from friction between so that the internal convecting spaces are made smaller.
moving parts, can cause thermally produced differential Heat transfer by radiation can be increased by suitable
dimensional errors. Heat source changes due to adiabatic use of surface treatment, blackening being the most used.
action on air pressure may be relevant. Design, therefore, The actual surface treatment used will be decided by the
should aim to keep internal and external sources of temper- thermal wavelengths involved. Within the system, parts are
ature change as small as necessary. They, however, cannot blackened and the whole is then surrounded by a thermally
ever be held at an absolute stability. Internal heat trans- reflecting enclosure. The silvered vacuum flask is a good
fer mechanisms should be arranged so that temperature example of this.
566 Measuring System Design
It will be appreciated from this account of the possi- approximate the first-order linear response) for a vacuum
bilities that the thermal design of an instrument can be flask is around 2 to 3 h; a solid house brick box is similar.
very complex. Perhaps that is why this aspect of design When using such methods to average out daily variations,
of instruments has not been adequately researched or orga- they are barely effective. Daily variation smoothing requires
nized. Braddick (1966) gives brief mention to the problem. several meters of rock cover.
Considerable practical information exists in primary journal Active systems, those in which temperature is actively
literature on temperature controllers, but the basic princi- controlled, can be far more effective. Long time constant
ples, theory, and design do not appear to have yet been passive surrounds also suffer in that they must be given
ordered in a didactic manner. time to stabilize before the internal instrument can be used
Basic texts on heat transfer, such as Eckert (1950) and in its intended state. As thermal mass is the dual of thermal
Jakob and Hawkins (1957), provide general design princi- insulation, it is possible to construct first-order thermal
ples. Publications relating to cryostat design (White, 1959) enclosures from either conducting or insulating material.
and use also delve in these problems. Abachi, Molenat and Each has its advantages. Use of both kinds together raises
Malbrunot (1979) provide a useful bibliography of reported the order of the enclosure’s thermal time response and,
designs. Designers of microbalances have given thermal therefore, if designed properly, can provide better results.
effects considerable attention, Wolsky and Zdanuk (1980) In the practical construction of thermal controllers, it
being a key review work. Wolsky et al. (1967) discusses is far easier to produce heat energy than it is to remove
how gold plating a quartz balance arm can reduce thermal it (compare electric heating elements with Peltier cooling
expansion errors through subtle heat transfer mechanisms. cells, for instance). For this reason, thermal surrounds
Sporton (1972) is probably the best available paper to con- often are operated at some margin above the ambient
sult when beginning a design of an enclosure. temperature so that natural heat loss acts to provide the
Thermal enclosure design is often needed in the devel- heat reduction mechanism.
opment of the total instrument system. Some applications It is the author’s experience that a contributing factor to
require the instrument to operate within temperature limits the mechanical instability of materials is the stress brought
as small as a millikelvin variation. Design aims to form about by temperature variations of the part. It appears that
an enclosure in which the walls seen by the contained the more controlled temperature variations are, the better
instrument are uniform in temperature and of the lowest the ultimate stability of creep exhibited. This is further
practicable thermal impedance. supported by the practice of thermostating gravity meters
Simpler designs generally use an almost point source with a temperature-controlled jacket that is kept energized
of heat, which is controlled from a single point on–off even when the instrument is not being used.
temperature sensor. Clearly, this does not come close to the
rule of maintaining uniformity throughout and, therefore,
only provides crude control. Far better are the designs in 8 THERMAL COMPENSATION
which the heater and the sensor are distributed throughout METHODS IN FINE MECHANISM
the wall area of the enclosure. Proportional control is
generally assumed to provide the best control loop but this The alternative method for reduction of thermal mechanical
may not always be the case. error is the use of compensatory mechanisms.
Parameters of thermal enclosure design are the external Most constructional materials possess a positive coeffi-
variations and frequency to be controlled, the size and cient of thermal expansion; that is, they expand in length
shape of the instrument (clearly smaller instruments are with increase in temperature. Owing to the shortage of suit-
easier to control to a given limit than large structures), able, large enough, negative coefficient materials (they do
the limits of control variation required, the internal heat exist – carbon fiber and the silicas), it is not generally
sources, the enclosure design (sandwiched heat conducting possible to counteract the positive expansion with a neg-
and insulating walls), plus judicious use of air spaces have ative part as shown in Figure 3. This kind of component,
proved very effective (Sporton, 1972). however, using two different positive coefficient materi-
In many applications, the aim is to integrate the external als, can provide an adjustable positive coefficient rod by
thermal forcing function to an adequately small internal
variation. Thus, it seems an obvious step to use an enclosure
that has an adequately long thermal time constant. In d
practice this can be, with passive methods at least, less + −
effective than might be expected. The thermal time constant
(that for a 64% change if the system is assumed to Figure 3. Thermal compensation by cascaded bars.
Error Sources in Fine Mechanics 567
d Invar bar
Corner cube
Steel wire, soft soldered Section
to bar
Brass Figure 5. Thermal compensation by counterstressing.
Cast iron
the Invar, causing it to be tensioned outward. This, in turn,
Length and materials varied via the Poisson ratio shape–volume change effect, causes
to obtain compensation needed the Invar bar to reduce in length. The reverse holds true.
He provided the theoretical considerations for obtaining the
degree of compensation needed for a pair of materials.
Clearly, this method can be applied by plating layers or
by pressing tubes over the base bar. The thickness and the
Figure 4. Thermal compensation by reentrant bars. external material used are the key design factors.
A third method for compensating thermal errors is to
clamping it with distance, d, spanning the ratio of the two measure the temperature change in a part, applying a
lengths required. generated appropriate length change where needed. Many
This problem can be completely overcome (within limits) methods exist for the provision of small length changes
by the use of two positive coefficient materials arranged as a under electric control. They include piezoelectric crystals,
reentrant compound structure. Figure 4 shows the mounting magnetostrictive rods, and thermal heating of rods and
arrangement used to maintain the reference arm distance, electromechanical devices. Measurement of the temperature
d, in a Michelson optical interferometer constant with of a part can be difficult because of transient thermal effects.
temperature change to the mount and corner cube. As the The control loop, however, may also introduce phase shifts
cast iron expands outward with temperature rise, the brass that distort the required behavior.
expands in the reverse direction at a larger rate, giving a
net change of zero to distance, d. Concentric tubes can also
be used to form a zero length change bar. Bimetal strips, 9 TEMPERATURE ERROR IN ELASTIC
in which the two materials are joined, can also be used to ELEMENTS
change dimension in some way through a forcing action.
The choice of materials being limited means that the Temperature changes to elastic elements will give rise
design of reentrant arrangements must be proportionate to to length changes according to the thermal coefficient
the relative lengths in order to obtain the zero (or other) of length; additionally, however, the elastic modulus will
state using the small range of thermal expansion values change.
that will be found to be usable. In the above mount, the The combined effect is to alter the elastic rate (scale
design also allowed for the expansion of the glass of the error) of the spring element and also to alter its free
corner cube. length, giving rise to a zero position shift (zero error).
The assumption that both parts of a reentrant link If the component is designed to be a flexing spring,
change (track) temperature together may not hold; during then these combined effects can be very serious. Exactly
temperature transients the compensation may not operate how the spring’s combined scale and zero behavior varies
satisfactorily. with temperature change depends upon the spring’s form
A second method, one that is not used as commonly and geometry. Producing a spring that has negligible
as the reentrant arrangements (because, it seems, it is not temperature error does not occur merely by selecting a zero
well publicized) is to make use of the stresses set up by elasticity modulus material, nor will it occur by using a zero
temperature change in an element. Baird (1968) used this thermal coefficient of expansion material.
principle (Figure 5), to compensate a bar of Invar. It can be shown (see Gitlin, 1955a) that for the common
In this method, steel wire was wrapped around an Invar cantilever, helical, and torsional forms of spring the error of
rod. It was then soft-soldered to it, so that axial stresses in spring rate change (scale error) is not a factor of the physical
the steel wire, caused by temperature changes, are coupled dimensions but it is dependent upon the thermoelastic and
into the circumference of the Invar bar. With temperature linear expansion coefficients and the magnitude of the
rise, the steel attempts to form a larger diameter ring around temperature excursion. Thus, once the material is chosen,
568 Measuring System Design
a given level of scale error exists for all designs with aspect of mechanical instrument design is sadly neglected in
that material. Beryllium copper, for example, possesses texts, in courses, and in practice. Scientists involved in the
a scale error of −0.03% per degree Centigrade for an manufacture of their own experimental and measurement
increase in temperature, that is, the spring softens with apparatus often reach these barriers and they, therefore, are
temperature rise. aware of the errors that can arise. Other designers, however,
Shift in zero with temperature is dependent upon the often do not realize that their designs also may suffer from
spring’s form. The three articles by Gitlin (1955a, b, such effects; it is necessary to eliminate them as design
c) are key reference accounts about the thermal errors parameters by reasonably rigorous study before they can
in instruments. Trylinski (1971) includes discussion of be honestly regarded as insignificant. They are often easily
this problem. reduced at the initial design stage.
The main error source is temperature and it affects
just about every component. It always needs consideration
9.1 Error due to ambient pressure before any measurements are made or instruments designed.
The mechanical components of an instrument change di-
mension according to the external pressure exerted upon
RELATED ARTICLES
them by their ambient environment. This has a very small
effect, but on occasion it can be significant. It is calculated
from the Poisson ratio for the material, which expresses the Article 16, Reduction of Influence Factors, Volume 1;
longitudinal change in length for transverse stress variation. Article 48, Common Sources of Errors in Mea-
As a guide, metals will lengthen by around 1 part in 108 surement Systems, Volume 1; Article 80, Principles
for a typical daily ambient barometric pressure variation. of Fine Mechanics – Kinematic and Elastic Designs,
Error from this source can be reduced by the use of Volume 2; Article 81, Principles of Fine Mechan-
ambient pressure control, which need not necessarily be ics – Systems Considerations, Volume 2; Article 82,
at the evacuated level as long as the pressure remains Kinematical Regime – Members and Linkages, Vol-
constant enough. Alternatively, it can be compensated in ume 2; Article 85, Elastic Regime of Design – Design
some way through the use of some form of pressure- Principles, Volume 2
sensitive device. Sealing an instrument inside a container
does not necessarily provide constant pressure inside the
container; temperature changes to the gas inside it will give REFERENCES
rise to significant pressure changes.
Abachi, H., Molenat, J. and Malbrunot, P. (1979) Continuously
Variable Cryostat Giving Accurate Settings at any Temperature
between 77 and 300K. Journal of Physics E:Scientific Instru-
10 MISCELLANEOUS ERRORS IN FINE ments, 12, 706–711.
MECHANISM Baird, K.M. (1968) Compensation for Linear Thermal Expansion.
Metrologia, 4, 145–146.
The above discussion has dealt with sources of errors occur- Barnes, R.B. and Silverman, S. (1934) Brownian Motion as a
ring in mechanical components. Also of importance might Natural Limit to all Measuring Process. Review of Modern
be those cross-coupled, mechanically produced effects that Physics, 5, 162–192.
transduce mechanical energy into another energy domain. Bleaney, B.I. and Bleaney, B. (1957) Electricity and Magnetism,
An example is thermoelectrical current, which is gener- Clarendon Press, Oxford.
ated by the mechanical materials used in combination with Braddick, H.J.J. (1966) The Physics of Experimental Method,
thermal effects. Piezoelectric and magnetostriction are elec- Chapman & Hall, London, (1st edn, 1954).
trical resistance-strain changes that are each brought about Eckert, E.R.G. (1950) Introduction to the Transfer of heat and
by mechanical input parameters. They must, therefore, be Mass, McGraw-Hill, New York.
considered in the mechanical design stages of an instrument Gitlin, R. (1955a) How Temperature Effects Instrument Accuracy.
as possible error sources in the total system. Control Engineering, April 70–78.
It can be seen that there are many sources of error in, Gitlin, R. (1955b) What’s Available for Compensating Instru-
or caused by, the mechanical components of an instrument. ments for Temperature Changes. Control Engineering, May
If good mechanical performance is to be obtained, these 70–77.
must be considered and reduced to satisfactory levels of Gitlin, R. (1955c) How Temperature Compensation can be used.
magnitude. This point needs to be stressed because this Control Engineering, June 71–75.
Error Sources in Fine Mechanics 569
Glazebrook, R., Darwin, H. and Mason, C. (1923) Instruments, Sydenham, P.H. (1973) Nanometre Stability of Invar and Quartz
Design of Scientific, A Dictionary of Applied Physics, Vol. III, Suspended in Catenary. Journal of Physics E:Scientific Instru-
Macmillan, London, 445–450. ments, 6, 572–576.
Jakob, M. and Hawkins, G.A. (1957) Elements of heat Transfer, Sydenham, P.J. (1979) Measuring Instruments, Tools of Knowl-
Wiley, New York. edge and Control, Peter Peregrinus, London.
Jones, R.V. (1967) The Measurement and Control of small Trylinski, W. (1971) Fine Mechanisms and Precision Instru-
Displacements. Physics Bulletin, 18, 325–336. ments – Principles of Design, Pergamon Press, Oxford (Trans-
lated from Polish by A. Voellnagel).
Jones, R.V. (1968) More and More about Less and Less. Pro R
Usher, M.J. (1973) Development in Seismometry. Journal of
Instn Gt Br, 43, 323–345.
Physics E:Scientific Instruments, 6, 501–507.
Jones, R.V. (1970) The Pursuit of Measurement. Proceeding of
White, G.K. (1959) Experimental Techniques in Low-temperature
the IEE, 117, 1185–1191. Physics, Oxford University Press, Oxford.
Logan, J.L. (1973) Gravitational Waves – A Progress Report. Whitehead, T.N. (1954) The Design and use of Instruments and
Physics Today, 26(3), 44–52. Accurate Mechanism, Dover, New York (1st edn, 1933).
Rolt, F.H. (1929) Gauges and Fine Measurements, Vol. II, chap- Wolsky, S.P. and Zdanuk, E.J. (1980) Ultra Micro-weight Deter-
ter XIV, Macmillan Publishing, London. minations in Controlled Environments, Wiley, New York.
Sporton, T.M. (1972) The Design of a General-purpose air Wolsky, S.P., Zdanuk, E.J., Massen, C.H. and Poulis, A. (1967)
Thermostat. Journal of Physics E:Scientific Instruments, 5, On Gold Coated Quartz Arms to Reduce Thermal Expansion.
317–321. Vacuum Microbalance Technique, 6, 37.
89: Optical Materials
Pak L. Chu
City University of Hong Kong, Hong Kong, China
electronics. In this emerging century, microminiature optics
1 Blank Materials 573 will feature strongly.
2 Abrasives 575 The eventually realized laser source, in the 1960s,
3 Tools for Optical Manufacturing 575 provided more capability for optical methods to be
used in more kinds of measurement. In the solid-state
4 Blocking Materials 575
form, the laser source has revolutionized many measuring
5 Polishers 576 instruments.
6 Cleaning Agents 576 Several texts can be useful here – see Musikant (1985),
7 Optical Cements 576 Ahmad (1997), Hobbs (2000), Smith (2000), Weber (2003).
8 Material Properties 576 Optical instruments have been extensively used for var-
References 578 ied applications in diverse areas and thus have immensely
contributed to the science of measurement. During the
latter part of the twentieth century, the interest in this
branch of physics has further been stimulated and sus-
1 BLANK MATERIALS
tained with the availability of high power–coherent light
Optical technique has long been important for measuring sources and detectors. This era has thus also witnessed
and observing instrument design, making its first impact many original approaches toward developing sophisticated
as the telescope and microscope of the sixteenth-century optical instruments and systems and associated technolog-
Galilean times. Combined with fine mechanics develop- ical advancements in meeting intricate measurement needs
ment, it provided a vast array of mechanical-optical and, for different sectors of the economy. The quite relative
since the early twentieth century, electro-optical instru- newcomer, the miniature optical-array detector, has also
ments. They are vital to modern telecommunications in opened up remarkable possibilities for faster and better
optical fiber technology, in database storage devices, audio measurement.
recording and playing equipment, and so on. While the This article primarily aims at elaborating the fundamen-
pace of development of the technologies involved is truly tal issues connected with optical technology and under-
amazing, the basics of the optical parts remain as it has standing the basic principles underlying the construction
been for centuries. Over time, the advances have been in of important optical measurement instruments and their
understanding the optical theory and in using computers to applications. Mass-produced optical elements still need
support design. to use many of these centuries’ old methods. Small-
Optical methods made significant strides when machine scale production, typical of measuring systems, has to
computation reached a state where the previously needed employ them as a handcraft needing considerable skill and
person-years of design effort have been reduced to hours. A know-how.
further boost came when the traditional optical-mechanical Raw materials used in the optical shop are considered
industries combined their expertise with twentieth-century under the categories blank materials, abrasives, optical
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
574 Measuring System Design
tooling, blocking materials, polishers, cleaning agents, and therefore, have considerably less dispersion when compared
optical cements. to flints. This combination is useful for it allows corrected
The stock material used to form optical elements is lenses to be formed by the use of two elements placed
known as a blank . Glass is the most commonly used blank together.
material for the production of optical elements but other Optical glass contains silica, which, when mixed with
materials, including plastics, are also used. the oxides of potassium, calcium, barium, lead, and more
Glass is a transparent dielectric substance, in which the in different proportions, yields a wide range of optical
molecular structure is characterized by an extended network glasses of different properties. Table 1 gives some typical
lacking in periodicity and symmetry, but with atomic forces compositions.
comparable to those present in crystals. Glasses of optical A wide choice of optical glasses varying in refractive
grade, commonly known as optical glass are used for index and ν value is available commercially. The end use
quality optics. and design considerations govern the choice of appropri-
As optical materials are commonly used to diffract ate material.
radiation passing through an element, the effects that Testing of optical glass involves evaluation of refractive
will cause distortions of images transmitted need to be index, ν value, striae, strain, and bubble contents. The test
identified and kept within limits. Mirror elements, where methods are simple and can easily be adopted in actual
the radiation is reflected from the surface, are less impacted practice by following standard specifications.
by defects but they still need to keep their form and Some other blank materials employed to make best use
shape as residual stresses relax over time or with thermal
of the different radiation bands of the EM spectrum are
cycling in operation. Optical glass must be extremely
quartz (both crystalline and amorphous varieties), rock salt
homogeneous, must not absorb gases and liquids and must
(NaCl), sylvine (KCl), fluoride (CaF2 ), calcite, KBr, ZnS,
have negligible residual strain, bubbles, and striae. Striae
calcspar, Iceland spar, solid solutions of thorium bromide
are optical lines seen in glass due to variations of its
with thorium iodide, Corning 7940 glass, Mylar, sapphire,
refractive index.
and Irtran. All these possess quite different transmission and
The specifications of A grade optical glass state that
reflection properties making them necessary for different
it should not possess any visible striae or cord, with
wavelength operations.
birefringence less than 10 µm cm−1 , that is, no color beyond
light gray should be seen in the crossed Nicols test and it Curves illustrating transmission of light through the
should be totally free from bubbles larger than 0.01 mm atmosphere, as also through some selected materials, can
in diameter. This kind of glass is refined by heating and be found in manufacturers data sheets and in works
cooling many times to remove the defects. such as the classic work by Hardy and Perrin (1932) and
Optical glass is characterized by its refractive index and Weber (2003).
the constringence or Abbe number (or simply ν value). This The spectral characteristics of optical glasses render them
is defined as (Nd − 1)/(NF − NC ), where NF and NC are the largely unsuitable for incorporation in infrared or ultraviolet
refractive indices of the material corresponding to the F and imaging systems: other special materials are available for
C lines of the spectrum of hydrogen having wavelengths use at these wavelengths.
4861 and 6563 Å respectively, and Nd pertains to the helium Plastics of many types are frequently used in optical
yellow line at wavelength 5876 Å. technology particularly for making inexpensive and not too
The physical significance of ν is that it relates to the critical optical elements. Polymethyl methacrylate, in short
dispersion caused by the material: the higher the ν value, PMMA or commercially called Perspex , has properties like
the lower is the dispersion. In simple terms, this means crown glass, and polystyrene is equivalent to flint type. The
that the higher the ν value is the more the radiation is principal drawbacks in their usage are the nonavailability
bent as it changes from one index to the next. Glasses of materials with wider choice of refractive index and ν
with higher ν are customarily called crown glass and those value, lack of homogeneity, lower softening point, fragile
with lower ν are known as flint glass. Crown glasses, nature, and proneness to scratching.
Table 1. Percentage composition of some typical optical glasses.
Glass type SiO2 B2 O3 Na2 O K2 O CaO BaO ZnO PbO Al2 O3 Fe2 O3
Borosilicate crown 71.0 14.0 10.0 – – – – – 5.0 –
Crown 74.6 – 9.0 11.0 5.0 – – – – –
Light flint 62.6 – 4.5 8.5 – – – 24.1 – –
Barium flint 45.2 – – 7.8 – 16.0 8.3 22.2 – –
Optical Materials 575
2 ABRASIVES Table 2. Abrasive gradation.
Mesh size Other Elutriation Approximate
To create optical elements, the blank must be shaped into designation time in average size
flats or curves. The traditional method has been to grind minutes (µm)
blanks with abrasives and a form. It is possible to cut 60 M 60 – 290
the material into shape using computer-controlled machine 100 M 100 – 150
tools with micrometer tolerances, but they are still not 180 M 180 1 85
220 M 220 – 75
capable of replacing grinding in most applications. Plastics 700 M 302 5 20
can be molded as is done in spectacle lenses but again 850 M 3021/2 10 16
a mold shape is needed and the method is not capable 1000 M 303 20 13
of producing geometry of adequate accuracy for exacting 1300 M 3031/2 40 10
optical systems. 1800 M 304 60 5
3000 M 305 – 3
Glass is quite a hard substance and for working on its
surface, special abrasives are used. Obviously, an abrasive
must be hard compared to the surface being ground. On 3 TOOLS FOR OPTICAL
Moh’s scale, the hardness numbers of diamond, ruby, topaz,
and quartz are, respectively, 10, 9, 8, and 7.
MANUFACTURING
Carborundum (SiC), emery, in natural as well as artificial
forms (fused Al2 O3 , aloxite, alundum), and boron carbide Turning, smoothing, and polisher forming tools are nor-
(B4 C) are some of the abrasives commonly used in optical mally made of cast iron. Molten iron treated with meehanite
workshop practice. These are mostly used in powder form (CaSi) gives a fine structure free from porosity and other
suspended in a water medium. Diamond powder is available defects. Polisher-holder tools may be made of aluminum.
in almost all grades and is used for impregnation on Gunmetal has also been used.
working tools particularly when large stock removal is A radius turning attachment is used for providing the
necessitated. desired curvature on the tool surface. As the turning process
The average grain sizes of abrasives used in the optical is quicker compared to lapping, it is desirable to produce,
shop range between 3 to 300 µm. Rough work requires a as accurately a curved surface from the tool as is possible
faster cutting rate, which can be achieved by using coarser by turning.
grains. Silicon carbide or carborundum powder is most Templates and gauges are generally made of brass, 2-mm
suitable at this stage. Once the desired shape and size thick, the edge reduced to about 0.5 mm. Collets for holding
have been given to the blank, finer grades of abrasives are the job during curve generation on automatic machines may
progressively employed to obtain better finish. Aluminum be made of brass (aluminum is also used). Clamping bells
oxide or emery is normally used for this purpose. required in centering and edging operations are normally
Grain size grading of abrasives is done in several ways. made of brass.
An example is the elutriation time taken, in minutes, for Test plates, test spheres, polygons, standard angles,
the emery to settle through a column of water contained in optical flats, and the like are made to the required tolerance
a vessel 30 cm in diameter and 1-m high, using the sieve and durability requirements, from specified materials such
or mesh size through which that particular grade can pass. as white plate glass, ophthalmic glass, optical glass, or low
Table 2 gives the commonly used nomenclature of parti- thermal expansion coefficient materials such as Pyrex, fused
cle size grading and the corresponding average grain size. silica, CerVit, and Corning ULE.
The quality of uniformity in grain size is often indicated
by numbers 0, 1, 2, 3, . . . , 10; 0 being perfectly graded and
10 being badly graded. 4 BLOCKING MATERIALS
For the polishing operations, extremely fine-grain hard
powder free from impurities is used. Rouge (Fe2 O3 ), cerium After smoothing, several optical elements are normally
oxide (CeO2 ), putty powder (SnO), chromium oxide, and blocked over a tool and later operations performed upon
diamond paste are some of the common polishing agents. them in situ. This requires a material that holds the optical
Tripoli, diatomaceous earth, chalk, talc, slate, and pumice elements rigidly during working. The blocking material
powder are other abrasive materials. Garnet optical pow- should be hard enough to withstand the pressure and
ders have also been used for the particular purpose of friction of the smoothing and polishing tool, yet be soft
ophthalmic (spectacles uses) machines employing autofeed enough for easy extraction of the secured optical elements
slurry pumping systems. whenever needed.
576 Measuring System Design
Pitch is the most common blocking material as it is is very useful for degreasing. Methylated spirit, isopropyl,
soft when warm, but hardens on cooling. Normal tar and anhydrous alcohol are commonly used for general
can be hardened to the desired degree by mixing with purpose cleaning of optical surfaces. Benzene, benzol,
it additives such as sawdust and cotton wool. Blocking and trichloroethylene dissolve pitch and are used during
material should be cloth filtered to eliminate undesired hard the deblocking operation. Kerosene dissolves beeswax
foreign particles. Beeswax, plaster of paris, and hydrated efficiently. Acetone is an excellent solvent and is used for
lime are used for blocking optical flats, prisms, and similar cleaning polymerized HT cements.
elements. Low melting–point alloys are also used as
blocking material.
7 OPTICAL CEMENTS
5 POLISHERS Optical cement must be transparent, reasonably color free
and must not develop much strain on the cemented sur-
A polisher is an accurately shaped cast iron tool upon
faces. These are normally nonadhesive, but join two glass
which the polishing medium remains mounted. Polishing
surfaces after some physical or chemical operation has been
of optical surfaces is achieved by lapping the job with a
performed upon them. Decementing may or may not be
preformed polisher of desired shape and size and applying
possible, depending upon the type of cement used. Some
polishing powder. The accuracy of the surface quality
cements set on heating to a particular temperature followed
generated on the finished product largely depends upon the
by curing at another temperature.
quality of the polisher. A polisher material should be hard
Canada balsam, a natural oleoresin with refractive index
enough to retain its form during polishing, yet it must be
1.52, is commonly used in optical cementing work. Its
amenable for alteration as and when needed. Pitch, as stated
unique characteristic is that it has very little tendency for
earlier, possesses these properties: a polisher is formed of
granulation or crystallization on drying from a solution.
warm pitch of the desired shape and size. On cooling, it
Cellulose caprate (refractive index 1.47–1.49), which is
becomes sufficiently hard to retain its character.
another that is widely used, is a thermosetting cement.
Polisher pitch is softer than blocking pitch. Mixing tur-
Thermosetting cements readily release the cemented
pentine with hard pitch produces polisher pitch. Polishing
components upon heating. Many other types of commer-
grade pitch is commercially available. As the hardness
cially available optical cements, on exposure to ultravio-
of pitch depends upon the temperature, the viscosity and
let light, are set by a polymerization process. These are
hardness of blocking and polishing pitch should be main-
extremely fast drying, but once cemented, the reopening
tained in accordance with the ambient temperature of the
process becomes quite tedious.
shop floor.
Felt, cloth, and wax are some other materials used in
preparation of polishers. Wax polishers cause less streaks
and scratches but are slow compared to pitch and do not 8 MATERIAL PROPERTIES
produce accurate surfaces so readily. For hand polishing,
wood pitch is considered to be best. In an optical material, its transmittance is an important
For better results, polisher pitch might be prepared to a characteristic to be known because it measures how much
consistency such that it is readily but not deeply indented light can pass through the piece of material. Technically, it
by the thumbnail. Thus, it should be neither too soft nor is defined as the ratio of the transmitted optical power to the
too viscous. Loaded polishers are made by addition of wood incident optical power when an optical beam is launched
flour, cotton wool, or yellow felt to the pitch. These help into the piece of material. Obviously, the difference, instead
maintain the shape of polishers. Addition of beeswax to of the ratio, between the input power and the output power,
pitch reduces its tendency to scratch the working surface. accounts for the loss or attenuation of the material. The
Teflon has also been tried with success as a polisher. attenuation consists of reflection, scattering, and absorption
Different types of polishing pads are also available com- of light by the material. The magnitudes of these quantities
mercially, and are mainly used for medium grade flat work. depend on the wavelength of light in operation. Since
the optical material is made of molecules such as silica
dioxide (SiO2 ) in glass and CH4 in PMMA polymer,
6 CLEANING AGENTS these molecules can be regarded as spheres with certain
diameters. If the wavelength of light is smaller than these
At various stages of the work, the optical surfaces are diameters, it will see the molecules as spheres and when it
cleaned to remove grease, pitch, or wax. Teepol solution impinges onto the sphere, it will be scattered into directions
Optical Materials 577
other than the original direction of propagation. This kind It also has harmonics at 4.5 and 2.25 µm and so on. Its
of loss is called scattering loss and is typically described long absorption tail extends well into the visible wave-
by Rayleigh’s law of scattering: length range causing extra losses. This is called infrared
absorption.
K Roughly speaking, the absorption spectrum of an optical
Lscattering = (1)
λ4 material can be represented by a V-curve as shown in
where K is a constant dependent on the diameter of Figure 1.
the sphere and λ is the wavelength of light. Thus, we The left arm of the V represents ultraviolet absorption
can see that as the wavelength shortens, the scattering and the right arm of the V represents infrared absorption.
loss increases dramatically. Of course, if the wavelength The apex of the V therefore represents the minimum
of light is greater than the diameter of the sphere, the absorption loss and the wavelength at which this occurs.
propagating light will not see individual spheres in the For silica glass, this minimum is 0.15 dB km−1 at 1.55 µm.
optical material. In this case, the Rayleigh type scattering It forms the basic material for communication-type optical
will be zero. However, if the material is not homogeneous, fiber. Figure 1 also shows that fluoride glass has a much
the resulting nonuniformity will also cause scattering if lower absorption loss of 0.01 dB km−1 at 2.7 µm, while the
its period of variation is smaller than the wavelength KCl glass has a still lower loss of absorption at 0.001 at
of light. 4.7 µm. However, these glasses tend to absorb water and
Now, the reflection of light at the surface of the optical are not strong mechanically.
material also depends on its smoothness. The simplest Figure 2 shows the transmittance of several infrared
relation describing the reflection is the reflectance defined transmitting glasses.
as the ratio of the optical power reflected by the surface and Another important property of optical material is
the optical power incident to the surface. Mathematically it the wavelength dependence of its refractive index. The
is defined as
n−1 2
R= (2) 103
n+1
102
Loss (dB/km)
10
This formula assumes that the interface is formed between 1
air and the optical material with refractive index n. It is 10−1
also assumed that light enters the interface perpendicu- 10−2
10−3
larly. In fact, if light enters at an angle, the reflectance 10−4
can be calculated by means of a complicated formula nor-
mally found in textbooks of optics such as Hardy and 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5
Perrin (1932). l (µm)
As light passes through the optical material, it is not
only scattered but can also be absorbed by the material. Figure 1. Absorption spectra of SiO2 , fluoride, and KCl glasses.
The absorption can be divided into two parts: the ultra-
violet absorption and the infrared absorption. For ease of
explanation, let us consider the absorption of a pure sil- 1. Silicate glass
2. Germanate glass
ica (quartz). It is made up of molecules of SiO2 . The Si 3. Calcium aluminate glass (T = 2 mm)
atom can oscillate at several wavelengths, that is, its reso- 100 4. Tellurite glass (T = 1 mm)
nant wavelengths. Two of them are particularly important 90 5. Lead bismuthate glass (T = 1 mm)
Transmittance (%)
80
to the absorption loss of glass, that is, 0.068 and 0.116 µm. 70
Light with these wavelengths entering into the glass will 60
be completely absorbed to give rise to the oscillations of 50
the Si atoms. Not only so, light with wavelengths close to 40
30 1 2 3 4 5
the resonances will also be absorbed although the inten- 20
sity of absorption decreases as the difference between them 10
increases. The wavelength range of visible light varies from 0
3 4 5 6 7 8
0.4 to 0.8 µm. Thus, the influence of the Si atom res-
Wavelength (µm)
onant absorption is significant. This is called ultraviolet
absorption. On the other hand, the molecule SiO2 also has Figure 2. Transmittance spectra of infrared glasses. (Reproduced
several resonant wavelengths. An important one is at 9 µm. by permission of Marcel Dekker.)
578 Measuring System Design
1.70 dn
(dispersion)λ = (3)
dλ
1.65 Because of the dispersion, a white light passing through
Dense flint glass a block of glass will spread into many colorful beams.
In fact, this is a serious problem in optical fiber because
Refractive index
1.60 the dispersion reduces the information carrying capacity of
Light flint glass
the fiber. Figure 3 shows the refractive index spectrum of
several common glasses.
1.55
Borosilicate glass REFERENCES
1.50
Ahmad, A. (1997) Handbook of Optomechanical Engineering,
Hydrogen F Sodium D Hydrogen C CRC Press, Boca Raton, FL.
1.45 Hardy, A.C. and Perrin, F.H. (1932) The Principles of Optics,
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 McGraw-Hill, New York.
Wavelength (µm)
Hobbs, P.C.D. (2000) Building Electro-Optical Systems: Making
Figure 3. Refractive index spectra of common glasses. (Repro- It All Work, Wiley-Interscience.
duced by permission of Marcel Dekker.) Musikant, S. (1985) Optical Materials, Marcel Dekker, Inc., New
York.
Smith, W.J. (2000) Modern Optical Engineering – the Design of
dispersion of the material is defined as the rate of change Optical Systems, McGraw-Hill Professional, New York.
of its refractive index as a function of the change of Weber, M.J. (2003) Handbook of Optical Materials, CRC Press,
wavelength, mathematically, Boca Raton, FL.
This article is based in part on a chapter by J. Prasad and G. Mitra originally published in Handbook of Measurement Science Vol 2,
P.H. Sydenham, Wiley, 1983.
90: Optical Elements
Pak L. Chu
City University of Hong Kong, Hong Kong, China
Many other miscellaneous optical elements, for example,
1 Refracting Elements 579 graticules, filters, fiber-optic elements, polarization compo-
nents, and optical modulators, are also employed for varied
2 Reflecting Elements 581
purposes in the optical system design.
3 Gratings and Filters 581
This article deals with basic understanding of the function
4 Optical Fibers 582 of the various types of basic optical elements.
5 Polarizers 583 Refracting elements are made up of an optically trans-
6 Modulators 583 parent homogeneous medium bounded by two or more
7 Optical Geometric Element Design 584 surfaces. When a ray of light crosses from one medium
8 Interferometry 585 to another, the path of propagation of the ray is altered in
accordance with the well-known Snell’s law, which essen-
9 Optical Transfer Function (OTF) 588
tially forms the fundamental basis for developing designs
10 Holography 590 of the varied types of refracting elements.
References 592 The lens and the prism are two basic types of refractors.
1 REFRACTING ELEMENTS 1.1 The lens
According to the ray concept, a beam of light travels The purpose of a lens is to form an image of an object,
in rectilinear path inside any homogeneous medium. The that is, a lens is essentially an image-forming element. The
optical elements interposed in the light path cause a change image may be
in the course of the original path by means of
• real
• refraction • virtual
• reflection • formed at infinity.
• diffraction.
A lens that forms a real image is called a positive
On the above basis, the optical elements that can be lens; similarly, a virtual image–forming lens is called a
regarded as the building blocks of any optical system can negative lens.
be broadly considered under three categories: When a lens forms the image at infinity, it is called
a collimating lens since it transmits a parallel beam
• refracting elements after refraction.
• reflecting elements A single lens element is characterized by its two radii
• diffraction gratings. of curvature, axial thickness, diameter, and type of glass.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
580 Measuring System Design
a
F′
F F H′
b
Figure 1. Lens definitions.
I
The line joining the two centers of curvature is called the
optical axis of the lens.
Associated with each lens are three pairs of points called
cardinal points, which are of immense value in analyzing Figure 2. Ray deviation in prism.
its behavior. These are the
• focal point From Snell’s law, it follows that
• principal point
−1 N
• nodal points. Ic = sin (1)
N
When an incident parallel beam of light traverses through where N and N are the refractive indices of the incident
a lens, it either converges to an axial point (in case of and refracting media respectively. This equation forms the
positive a lens) or appears to be diverging from a point (in basis of design of many deviating prisms.
case of negative a lens). The angular deviation β suffered by a ray traversing
This situation can be modeled both for a bundle of through a prism (Figure 2) immersed in air, with vertex
rays traveling from left to right, or vice versa (Figure 1). angle α, can be shown to be given by
These convergence points of parallel incident rays are called
focal points F and F . The planes perpendicular to the β = I − α + sin−1 [(N 2 − sin2 I )1/2 sin α − cos α sin I ]
optical axis passing through the first and second focal (2)
points are known as the first and second focal planes where I is the incident angle and N is the refractive index
respectively. Obviously, rays passing through focal points of the prism material.
become parallel to the optical axis after refraction. The In the minimum deviation position (β = βm ),
surface generated by the locus of the point of intersection
sin 12 (α + βm )
of the incident and refracted rays is known as the principal N = (3)
plane. Like focal planes, there are two principal planes, sin(α/2)
and their axial intersection points are designated as first which is a very suitable equation for refractive index
and second principal points H and H . measurement.
The effective focal length f (or simply focal length) is As the refractive index corresponding to each color
defined as the distance from the focal point to the cor- (wavelength) is different, the deviation suffered by each
responding principal point, while the distance from nearest of them is different; this causes wavelength dispersion of
lens surface to the focal point is called the back focal length. the incident beam into the constituent colors after refraction.
Nodal planes are a pair of conjugate planes corresponding The resolving power of a prism is a measure of its capability
to unit angular magnification. Nodal points are the axial for separating adjacent spectral colors.
intersection points of these planes. Nodal planes coincide For erection of an inverted image without angular devia-
with the corresponding principal planes when the lens is tion, the prisms commonly used are Leman, Goerz, Hensolt,
immersed in air. Porro types 1 and 2, and Abbe (or Konig). The Abbe prism
does not cause lateral displacement. The Schmidt prism
erects an inverted image and deviates it by 135◦ (or 45◦ ).
1.2 The prism Penta and Amici prisms cause a 90◦ deviation. A special
feature of the penta prism is that the transmitted beam
A prism is essentially a non–image-forming optical element remains stationary even when the prism is turned. Hence,
used for two main purposes, namely, to deviate the light it is also called an optical square. A Dove prism inverts
path to any desired direction and to create wavelength without deviation, and on rotation the transmitted image
dispersion of incident light into its constituent colors. rotates synchronously.
The incidence angle corresponding to which the refrac- The actual configurations of the various types of prisms
tion angle becomes 90◦ is termed the critical angle, Ic . can be seen in standard textbooks such as Ahmad (1996),
Optical Elements 581
Jacobs (1943), Levi (1968), Smith (1966), Smith (1978), where m is the order of diffraction and N is the total number
Smith (2000), Hobbs (2000), and Weber (2002). of lines.
In this field, much of the detail was defined a considerable In recent years, holographic gratings have also
time ago. At that time, more detail was often published than become commercially available. Their development greatly
what is easily available today. For this reason, older titles extended grating applications and performance.
often contain useful information not available elsewhere. Another form of manufacture has been developed – the
holographic grating. Holographic gratings are produced by
recording the optical interference patterns created between
2 REFLECTING ELEMENTS two parallel laser beams obliquely set relative to each
other. Relative obliquity of the interfering beams enables
Reflectors are those elements that reflect back the incident control of line density. Compared with the ruled grating
ray in the original medium; they are called mirrors or reflec- the holographic grating remains free from periodic error.
tors. A mirror may be flat, concave, or convex. Concave With proper choice of the recording medium, blazed
and convex mirrors act like positive and negative lenses holographic gratings can also be made. Other advantages of
respectively regarding their imaging properties. holographic gratings include a greater number of grooves
In many specialized applications, particularly in astro- being possible, up to 6000 grooves per millimeter; large
nomical telescopes, reflector elements of parabolic, ellip- dimension, up to 400-mm diameter; no ghosting; very low
tical, or hyperbolic shape are used. Parabolic mirrors are level of stray light; recording being possible on concave,
also extensively employed in lighthouse and other projec- toroidal, and aspheric blanks; and a possibility existing for
tion systems. making aberration correction.
Beam splitters, which permit only a specified fraction A grating formed with concentric circular grooves (or
of the incident light to pass through and maintain a known obstructions) is called a Fresnel zone plate. It has the
definite intensity ratio between the reflected and transmitted property of a lens but with several focal lengths for the
beams, can also be considered as a class of reflectors. various wavelength components.
Hot, cold, partial, pellicle, and multilayer dielectric Metrological gratings, used to determine length, carry
metal mirrors are some other useful reflecting components identical opening and obstruction widths and are usually
employed in optical system design. much coarser (tens of lines per millimeter). The con-
structional details of a versatile, low-cost, radial pattern-
generating engine suitable for the manufacture of medium
3 GRATINGS AND FILTERS accuracy masters of metrological patterns as scales, line
gratings, absolute multitrack encoder disks, spirals, circles,
Diffraction gratings essentially consist of a large number zone plates, and variable density and variable width circular
of closely spaced, equidistant, lines or rulings formed on tracks has been reported (Jain and Sydenham, 1980).
plane or concave optical surfaces like a prism or filter; it The Fiber Bragg grating is an important optical signal
causes spectral separation of the incident light beam. processing component. It finds applications in optical
The efficiency of a grating is determined by the amount communications and in optical sensing. In this grating, the
of energy diffracted in any desired direction. This is refractive index of the fiber core experiences a periodic
enhanced by control of the groove shape. Sawtooth groove variation with a period . Any light with a wavelength λ
patterns of different step angle enable light concentration satisfying the following relationship
in the preferred direction (called blazing).
Linear gratings with straight apertures are traditionally λ = 2N (4)
produced by mechanically ruling precise parallel grooves,
with a diamond tool, on a suitably worked blank. They can will be reflected by the grating, where N is the average
then be replicated from a master ruled grating by molding refractive index of the fiber core. The reflectivity increases
methods using plastics. These may be of transmitting, or with the index variation and length of grating. A reflectivity
reflecting type. close to 100% can readily be obtained in single-mode
Concave reflection gratings are frequently used for effi- optical fibers. The creation of this grating depends on the
cient spectral separation. Spectroscopic grade gratings carry photosensitivity of the glass making up the core of the
extremely fine grooves (of the order of 400 lines per mil- fiber. Boron- or phosphorus-doped germanosilicate glass is
limeter). normally photosensitive, whereas a simple germanosilicate
The resolving power of a grating, that is, its ability to glass is not. However, the latter can be made photosensitive
separate adjacent spectra, is given by the relationship mN, when hydrogen is introduced into it. To create the grating,
582 Measuring System Design
a phase mask with rulings of period is placed close to filters are widely used in photography and colorimetry and
the fiber and the mask is illuminated by an optical beam of also in the production of interference filters for blocking
appropriate wavelength at which the core glass of the fiber extraneous passbands. Plastic filters are available both in
is photosensitive. sharp-cutoff and intermediate bandwidth types.
Thin film interference filters operate in the same manner
as a Fabry–Perot interferometer. These are usually designed
3.1 Graticules for normal incidence but may be constructed for specific
nonnormal applications. Nonnormal incidence results in a
Graticules are fine measuring scales or well-defined patterns shift of the passband towards shorter wavelengths. Two
placed in the focal plane of an optical instrument. These are basic types of interference filters are available – metal-
employed for determining size, distance, direction, position, dielectric and all-dielectric. These are manufactured by
or any other attribute of the object that is viewed coincident thin-film, vacuum deposition techniques using special man-
with them. Graticules are also often denoted by terms ufacturing plant. The experimental techniques adopted are
such as diaphragm and web or in accordance with their described in Horne (1974), Macleod (1969), Smith (2000),
use, examples being sighting scales, cross-lines, (optical) and Weber (2002).
micrometers, and stage micrometers. Under spatial filters are included components, which,
Since a graticule is usually required to be viewed in in a functional sense, are used to modify the spatial
the instrument under high magnification, tolerance require- characteristics of the incident light. A typical example is the
ments in the manufacture of graticule blanks are generally fine pinhole (about 5–10 µm in diameter) used in a Fourier
very severe. For example, scratch width and dig or pit system for filtering out the unwanted portion of diffraction
diameter should not exceed 0.01 and 0.05 mm respectively; caused by the circular aperture of the microscope objective
surface accuracy should be within two fringes and spheric- lens when used in a laser beam.
ity should be within 0.5 of a ring; thickness should be within
0.05 mm; and decentration should not exceed 0.5 to 1 min
(Prasad and Singh, 1970).
Graticules are made by several different techniques, 4 OPTICAL FIBERS
namely, ruling and etching, ruling followed by vacuum
deposition of chromium or kanthal, photographic pro- Fiber-optic methods are based on the ability of thin smooth
cesses, and photo etching. Details of the procedures fol- strands of transparent materials to convey incident signal
lowed in graticule manufacture and compositions of some by the total internal reflection process, without interference
photoresists suitable for this work are given elsewhere from neighboring fibers. It is achieved by drawing thin (of
(Horne, 1974). the order of 10–100 µm), coated optical fibers having a core
Health and safety regulations now prohibit the use of and cladding made of different refractive index materials.
many traditional photoresists, so it is important to keep up Fiber-optic elements can broadly be grouped into two
with the latest information. types, noncoherent and coherent. Noncoherent fiber bundles
consisting of thousands of thin fibers not arranged in any
order are used for transporting light and are often called
3.2 Filters light guides or light cables. In coherent fiber bundles, the
fibers at the end faces are arranged in the same manner
Filters types fit into the three broad categories: so that optical image can be transmitted without suffering
• absorption or color filters distortion. The fiber bundles are usually made of fibers with
• interference filters relatively high loss. Hence the length of each bundle is
• spatial filters. about 1 m.
The flexible coherent fiber bundle is the basic component
Absorption filters derive their performance from bulk of many endoscopic medical and industrial inspection
interaction between light and the filter media; the trans- instruments used in remote examination of internal cavities.
mission generally being a smoothly decreasing function of The rigid form of a coherent fiber bundle, for example, a
thickness. Such filters are made in a variety of base mate- faceplate, is the essential component of image intensifiers,
rials: gelatin, glass, and plastic are commonly used. enabling high optical efficiency in image transfer. These
Selective color transmission is obtained primarily by components have also been used as windows of cathode-
ionic absorption or by selective scattering. Gelatin filters ray tubes built for direct recording of high-speed computer
are low in cost and provide a wide color choice. Glass printout or other data recording.
Optical Elements 583
change in polarization characteristics upon reflection (at,
and near, Brewster’s angle).
Retardation plates (quarter, half, and full wave), the
Savart polariscope, Soleil–Babinet compensator, dichroic
polarizers (sheet type), laser-polarizing beam splitters, and
(a) Cornu depolarizer are some of the important components.
6 MODULATORS
In communication, display, data recording, and measure-
(b)
ment, it is often desirable to control the amplitude, phase,
Figure 3. Grin Lens (a) Quarter Pitch Grin Lens, (b) Half Pitch frequency, and state of polarization of the light beam. Opti-
Grin Lens. cal modulators are the devices variously used for these
purposes. They are generally based on basic electro-optic
(Pockels, Kerr), magneto-optic (Faraday, Cotton–Mouton)
Low loss (attenuation approaching 0.2 dB km−1 at a
or acousto-optic (Debye–Sears, Raman–Nath) effects.
wavelength of 1550 nm) single-mode fibers, for use in
With the advent of the laser source of radiation, optical
optical communication, have also been developed.
modulators have assumed much importance (Hartfield and
Another fiber-optic component is the gradient index lens,
Thompson, 1978). Table 1 gives some basic information
also called grin lens (Figure 3). It is essentially a short
about the various types of optical modulators. A popu-
length of graded index fiber (about 5–10-mm long) with
lar high-speed optical modulator that finds application in
large core diameter (between 1 and 2 mm). The imaging
optical fiber communication is the electro-absorption mod-
properties of this lens depend on its length. A quarter pitch
ulator. It is a reversed biased pn junction similar to a pho-
lens forms an inverted, real, and magnified image of an todetector. Light impinging onto the junction is absorbed
object at infinity on the opposite end of the lens. A half and the absorption varies in accordance with the bias volt-
pitch lens simply inverts the image of an object from one age. Hence the modulating signal is the electric bias voltage
end face to another without magnification. and the light leaving the device is intensity modulated.
An electro-absorption modulator is rarely used freestand-
ing because of the high loss associated with the device.
5 POLARIZERS Instead, it is part of an integrated optical transmitter, that is,
a semiconductor laser and the modulator are built together
Polarizers are used for the production, control, and anal- on the same chip, with the output of the laser being fed
ysis of light in special polarization states. Their opera- directly into the modulator. In this configuration, the addi-
tion depends on the properties of birefringence (double tional loss caused by the modulator is only of the order of
refraction), dichroism (a special case of birefringence), and 1 dB or so.
Table 1. Types of optical modulators.
Type Principle Typical materials Typical applications
Electro-optic Induced birefringence on ADP, KDP, LiNbO3 , ADA Amplitude, phase, frequency and
application of external polarization modulation, high-speed
electric field shutter, laser Q-switching, mode locking,
cavity dumping
Magneto-optic Induced optical activity on Fused silica, dense glasses, Deflection, scanning, spatial modulation,
application of external H2 O, CS2 , P information processing
magnetic field
Acousto-optic Diffraction and scattering of Fused silica, GaAs, YAG, YIG, Scanning, optical delay line, heterodyning
light waves by sound TiO2
Mechanical Chopping of light beam by Signaling, IR detection
rotating sector disc –
Electro-absorption Light absorption in going Semiconductor material such as High-speed optical communication systems
through a reversed biased InGaAsP/InP, dependent on
pn junction depends on the operating wavelength
biasing voltage
584 Measuring System Design
7 OPTICAL GEOMETRIC ELEMENT Any incident polychromatic beam suffers dispersion due
DESIGN to the wavelength dispersive nature of the material of
the refracting element. Thus, with incident white light, a
In geometrical optics, diffraction effects are neglected and point object has different image locations for its different
the ray aspect of light is considered. Cophasal surfaces colors. The axial and transverse distances between the violet
orthogonal to a set of rays are defined as geometrical and red foci are called longitudinal and lateral chromatic
wavefronts (or simply wavefronts). aberration respectively. General accounts of lens types and
An ideal lens produces an image exactly alike the object. aberrations are available in Levi (1968) and Smith (2000).
However, an ideal lens is only a theoretical postulate since,
in reality, there are certain parameters inherent in the
7.2 Elements of lens design
refraction process that cause aberrations and deviate the
image structure from its ideal shape. The purpose of optical
It is possible to reduce aberrations of an optical system by
system design is to work out technical specifications of
choosing suitable combinations of two or more lens ele-
each of its constituent elements, tailoring the geometry such
ments. Consistent with the stipulated focal length, f num-
that the entire system gives an adequately aberration-free
ber, image definition, and other requirements, the selection
acceptable image.
of proper lens combination, their glass types, radii of curva-
ture, thickness, axial separation, and stop specifications falls
within the purview of lens design. Detailed analyses of lens
7.1 Aberrations design are available in several standard treatises (Buchdahl,
1954; Conrady, 1957, 1960; Cox, 1964). The relationship
There are five monochromatic and two chromatic aber-
between ray and wavefront aberrations and their evaluation
rations. Spherical aberration, coma, astigmatism, petzval
has been extensively covered in Hopkins (1950). Zoom lens
curvature, and distortion fall in the first category, with lon-
are covered in Clark (1973). Today, much of this foun-
gitudinal and lateral (or transverse) chromatic aberrations
dational theory is embodied into lens and optical system
forming the second.
design tools.
Spherical aberration arises due to the fact that rays
The analysis of an optical system requires tracing a large
originating from an axial object point and intersecting the
number of rays through it and consequently determining
refracting surface at different heights do not meet at a single
the residual aberrations, which become the basis for further
point after refraction. In other words, the paraxial focus
refinement. Depending upon the computational technique
is different from the marginal focus; the axial distance
adopted, several ray tracing schemes have been developed,
between them is called longitudinal spherical aberration
Smith (1966) being an example.
(l ).
Ray tracing schemes require numerous calculations,
When rays from an off-axial object point pass through
of simple nature, the whole process being very time
the lens, the paraxial and marginal bundles again have
consuming and tedious by traditional methods. Electronic
different foci. A comet-like flare is then seen in the image
computation has been applied to this with great effect.
plane, which is generated by the partial overlapping of
In either case, considerable experience is required to
circular light patches of increasing diameter. This off-
design even a simple corrected lens system. For this
axial aberration is called coma. It causes asymmetry in reason, an instrument designer would normally call upon
the image. the services of a lens designer or make use of commercially
Astigmatism occurs because rays from an off-axial object available proprietary lens systems when lens design is a
point lying in the meridional and sagittal planes focus at critical feature of the whole system.
different points. The distance between these two foci is
called astigmatism.
An image of a straight object perpendicular to the optical 7.3 Stop and pupil
axis is formed on a curved surface instead of a plane. This
is due to petzval curvature. To control the illumination and field coverage, certain
Owing to unequal lateral magnification throughout the diaphragms are used in optical systems. An aperture stop
field, the appearance of an image changes from the original determines the maximum cone of light that passes through
object; this occurs even when the above aberrations are not the system; its location has a marked influence upon off-
present. Pincushion distortion means that the outer zones axial aberrations of the system, though spherical aberration
are more magnified compared to the inner; the reverse and longitudinal chromatic aberration remain independent.
manifestation is called barrel distortion. A field stop is the diaphragm used to control the image size.
Optical Elements 585
Entrance and exit pupils are defined as the images of 8 INTERFEROMETRY
the aperture stop formed by the elements placed on the left
and right of the stop respectively. The ratio of the pupil Interferometric measurements serve as a very powerful
diameters represents system magnification. measuring tool yielding a high degree of precision (within
fractions of a wavelength of the radiation used).
Under suitable conditions, two light beams interfere with
7.4 Field angle each other, producing a resultant pattern consisting of alter-
nate bright and dark bands (or fringes). Shape and location
Field angle is defined as the angle subtended by the object
of these bands depends upon the difference in total optical
at the entrance pupil; it specifies the maximum object size
paths traveled by the two interfering beams. Accordingly,
that can be imaged by the system. As the field angle
length and phase variables may be measured directly in an
increases, the effects due to off-axial aberrations become
interferometer, while refractive index, displacement, veloc-
more pronounced, with consequent increased difficulties in
ity, and more can be evaluated by converting them into
the design process.
corresponding optical path difference OPD.
In the following, we first outline the fundamentals of
7.5 Effects of non–image-forming elements interferometry, followed by a brief description of common
types of interferometers.
Non–image-forming elements, such as plane parallel plates
and prisms, also introduce aberrations when placed in non-
collimated light beam. Such elements are often employed 8.1 Conditions of coherent radiation interference
as a corrector for compensation of residual aberrations of
the system. In Prasad, Mitra and Jain (1975), the aberra- Consider two light waves A1 and A2 of the same angular
tions produced by a single plane parallel plate immersed in frequency ω, given by
air and inclined at an angle with the vertical are discussed.
They provide equations for calculating the lateral displace- A1 = a1 sin(ωt + φ1 ) (5)
ment, tangential coma, and astigmatism of the system. A2 = a2 sin(ωt + φ2 ) (6)
A parallel plate can also be used to convert large
angular displacements into small translational equivalents where a1 , φ1 and a2 , φ2 are the corresponding amplitude
that can be used on a position sensitive photodetector and phase respectively, at any instant of time t.
to sense movement. This has been used in metrology Straightforward mathematical analysis shows that super-
(Sydenham, 1969). imposition of A1 and A2 produces a resultant amplitude A
given by
A = a sin(ωt + θ ) (7)
7.6 Performance analysis
where
Performance of the system can be expressed in terms of
residual aberration characteristic curves corresponding to
different field angles or in terms of a spot diagram depicting a = [a12 + a22 + 2a1 a2 cos(φ1 − φ2 )]1/2
the image plane intersection points of several rays passing
and
through various zones of the system. The resolving power a1 sin φ1 + a2 sin φ2
test is based on Rayleigh’s criterion of resolution; the θ = tan−1 (8)
a1 cos φ1 + a2 cos φ2
Strehl intensity ratio and, as is discussed later, the optical
transfer function (OTF) technique are other methods of From this, it can be seen that the resultant intensity
image quality assessment. pattern is maximum (Imax ) or minimum (Imin ), when φ1 −
φ2 = 2nπ or (2n + 1)π respectively, where n is zero or
an integer.
7.7 Optical modules Fringe visibility (or contrast) is defined as (Imax −
Imin )/(Imax + Imin ). For best visibility (or contrast), a1 and
Illustrations of configurations of often met optical systems
a2 should be equal or nearly equal.
are given in Figure 4. Their design clearly becomes pro-
Summarizing the conditions of interference,
gressively more complex as the number of elements rises.
Zoom lenses, for instance, were only made economic by 1. the two interfering beams must be of the same wave-
the advent of electronic computer methods of design. length monochromaticity;
586 Measuring System Design
(a) (b) (c)
(d) (e) (f)
(g) (h) (i)
(j) Focusing movement
Figure 4. Configuration of some optical systems: (a) Huygen eyepiece; (b) Ramsden eyepiece; (c) Kellner eyepiece; (d) Lister-type
microobjective; (e) Amici-type microobjective; (f) oil-immersion microobjective; (g) Cooke triplet; (h) Tessar lens; (i) double Gauss;
(j) zoom television lens-Taylor Hobson (Clark, 1973). (Reproduced from Zoom Lenses, Clark A.D., Copyright (1983) Adam Hilger.)
2. the phase difference between them must remain closely The principle of interferometry is used in many measur-
constant over the period of observation (otherwise, the ing instruments. Such use is discussed in Dyson (1970),
fringe pattern would dance around, causing a blurred Ahmad (1996), and Smith (2000)
patch – the coherence condition); and
3. the state or plane of electromagnetic vibrations in both
beams must be the same Fresnel–Arago law. 8.2 Michelson interferometer
To satisfy the above conditions, light sources used in The Michelson interferometer (Figure 5) is used for the
interferometry generally have a high degree of monochro- measurement of wavelength of light, slow air drift, calibra-
maticity and are highly coherent so that a large path dif- tion of a standard length, and other scientific applications.
ference can still produce a meaningful interference pattern. Light from an extended source S (white light or
The so-called coherence length is a measure of how long monochromatic) is divided into two paths by a beam split-
the beam length difference can be to retain coherence. ter B (Figure 5). The end mirrors M1 , M2 reflect back the
Further, all interferometers, to a greater or lesser degree, respective coherent beams 1, 2, which recombine to form
remain sensitive to stray mechanical vibrations. Adequate fringes in the viewing telescope T. In one of the arms, a
precautions are needed during their installation to minimize glass plate P identical in construction to the beam splitter
these effects. This makes use in severe vibration condi- is introduced to compensate for the inequality of glass path
tions difficult. traversed by the interfering beams. In practice, corner-cubes
Optical Elements 587
M1
S
P
I
O
S 2
C
B P M2 B
C
T
Figure 5. Michelson interferometer: S, source; C, collimat-
F
ing lens; B, beam splitter; P, compensator; M1 , M2 , mirrors;
T, telescope.
T
can be used instead of flat mirrors; this makes adjustment
considerably easier.
The Twyman–Green interferometer is a versatile instru- Figure 6. Fizeau interferometer: S, source; P, pinhole; B, beam
splitter; C, collimating lens; F, master surface; T, test surface; O,
ment that is particularly suited to optical shop-floor appli- observation point.
cations. It is similar in construction to the Michelson
interferometer. Here, the incident light is formed as a well- of linearity of the fringes expresses the surface error in
collimated monochromatic beam and the compensator plate terms of fractional fringe width (each fringe corresponds to
is dispensed with. The optical job under test is placed in one λ/2 departure).
arm of the interferometer, while the other arm provides the Numerous modifications of this classical instrument
reference beam. In cases of a lens, prism, and the like, the exist. Concave and convex surfaces are tested in a diverg-
beam passes twice through the test piece; hence the fringe ing and converging beam respectively, the master flat then
pattern depicts twice the amount of total error, which is being replaced by the complementary master surface. Using
inclusive of surface defects, material inhomogeneities, and a laser source, a spherical surface may be tested against a
aberrations. master optically flat surface. In this interferometer, a con-
The invention of the laser source enabled development of verging beam is focused at the center of curvature of the
an unequal path Twyman–Green interferometer of different test surface. Another set of fringes is obtained when the
geometry that is suitable for testing large optics in a beam is focused at the vertex of the test surface, and as
production line. such, the radius of curvature can also be measured.
8.3 Fizeau interferometer 8.4 Rayleigh interferometer
A Fizeau interferometer is suitable for qualitative as well The Rayleigh interferometer is a very useful instrument
as quantitative assessment of surface quality. In its classical for the measurement of the refractive indices of gases and
form, it is used (Figure 6) for flat surface testing. A liquids and for the control of their composition. Light from
collimator C sends a parallel beam of light from source a slit source is collimated and made to fall upon two tubes
S onto the test surface T. A good quality plane parallel placed side by side, one containing the sample under test
glass plate F, whose lower surface is of a high degree of and the other a standard sample. Fringes are viewed through
flatness (master surface), is placed over the test piece with a focusing lens and a magnifier. Fringe displacement is
an air gap. The collimator focuses the return beams onto the measured by tilting a compensator plate placed in the light
observation point O that is suitably located using a beam path passing through one of the tubes. White light sources
splitter B. The interference pattern is formed between the may be used enabling easy measurement of the fringe
rays reflected from the master and test surfaces. Departure displacement.
588 Measuring System Design
The Michelson stellar interferometer employs two widely
separated slits whose interdistance can be varied. Each slit L1
selects different parts of the wavefront. From the change in
visibility of the fringes seen through the telescope, a stellar
diameter can be evaluated. L2
8.5 Jamin interferometer
Used in refractometry, a Jamin interferometer has a thick (a) (b)
plate that splits the incident beam into two using reflections
at the front and back surfaces. An identical plate recom-
bines the beam. The Mach–Zehnder interferometer is a
modification of this. Using beam splitters for beam division DP
and recombination, the separation between the interfering
beams is made large. It is extensively used in hydrodynamic
measurements.
(c)
8.6 Point-diffraction interferometer Figure 7. Shearing techniques: (a) lateral shear; (b) radial shear;
(c) rotational shear produced by counterrotating Dove prisms (DP).
Generation of an aberration-free reference wavefront with-
out using high quality optics is achieved in a point- beams are laterally, rotationally, or radially sheared relative
diffraction interferometer (Smartt and Strong, 1972). A to each other so that different parts of the two identi-
pinhole on a transparent substrate is located slightly off-axis cal wavefronts interfere. There exist several techniques for
at the focal plane of the lens system under test. Diffracted achieving the desired shear. Figure 7 gives simple exam-
wavefronts from the pinhole produce reference wavefronts. ples. By employing birefringent crystal elements, such as
An absorption coating is placed over the pinhole substrate to the Wollaston prism or Savart polariscope, shear is readily
enable matching of the intensities of the interfering beams. obtained. This type is generally called a polarizing inter-
ferometer.
A shearing interferometer is less sensitive to vibrations
8.7 Scatter fringe interferometer
(relative to Twyman–Green interferometer) and is suitable
Large concave mirrors, as used in astronomical telescopes, for testing a lens, wedge angle, homogeneity of transparent
are effectively tested in a scatter fringe interferometer. samples, optical flatness, and the optical transfer function.
A scatter plate placed perpendicular to the optical axis
passing through the center of curvature of test surface splits 8.9 Multiple-beam interferometer
a converging beam into two components. The specularly
transmitted beam, after reflection through a small segment If the test surface and the inner surface of the master
of the mirror, serves as the reference beam, while the plate of a Fizeau interferometer are partially reflecting,
scattered component fills the whole test surface and is each ray splits into several components. Consequently,
the test beam. A semireflecting mirror is interposed to a number of beams of varying path length take part in
superimpose both the beams onto another scatter plate forming a multiple-beam interference pattern, where high
(identical to the first one). The instrument is comparatively finesse fringes are formed. High precision measurement
less sensitive to vibrations since the interfering beams are of thin film thickness, surface contour, and separation of
closely confined to each other. neighboring wavelengths are performed in a multiple-beam
interferometer.
8.8 Shearing interferometer
9 OPTICAL TRANSFER FUNCTION (OTF)
In a shearing interferometer, the test beam is split into two
beams, which are compared against each other, thereby Various criteria exist to express the performance charac-
doing away with any reference beam, as is needed in teristics of a lens system. Measurement of the individual
comparison-type interferometers. When superimposed, the aberrations gives clues to the designer in respect of its
Optical Elements 589
defects. A standard interferogram depicts the phase error where O(ξx , ηy ) is the OTF; ξx , ηy are frequency coor-
across the pupil. The resolving power test indicates the final dinates in the image plane; M(ξx , ηy ) is the modulation
resolution limit attainable by the system. However, none of transfer function (MTF) and φ(ξx , ηy ) is the phase transfer
these attributes yields a completely objective assessment of function (PTF).
the quality and image-forming capabilities of the optical The OTF comprises both the MTF and the PTF. The
system under test. significance of the phase term is that the image does not
The OTF has been used as a merit function that, to a large remain located at the ideal conjugate point but is displaced
extent, serves this need, even though it must be appreciated by the amount given by the PTF.
that this technique also is not the final solution to all lens The above equation further reveals that for complete
evaluation problems. evaluation of system performance, both the MTF and
PTF are required. However, it has been seen that for
a moderate range of object frequencies, the MTF gives
9.1 OTF representation sufficient information of practical interest, and is therefore
the quantity that is most commonly evaluated.
OTF is basically a resolving power test, but unlike the
classical resolution test, it expresses the image contrast
at different spatial frequencies of a sinusoidal object. It 9.2 OTF evaluation – scanning methods
is, in this respect, similar to the time frequency response
tests associated with electrical circuits (Coltman (1954); This method (Murata, 1966) involves scanning the image
Schade (1964)). intensity distribution of a known object, which may be
For elucidation of the OTF concept, consider two bright a sinusoidal grating of variable frequency, square-wave
object points separated by a dark space. Theoretically, grating, single slit, two square-wave gratings rotating in
the image of this object assembly should be two bright opposite direction (producing Moiré patterns of varying
points separated by a dark region. In such a situation, the frequency), or of any other arbitrary shape.
object is said to be well resolved, and both the object The basic experimental setup is given in Figure 8. When
and the image possess unity contrast. However, since any the object O is composed of sinusoidally varying intensity
real system is afflicted with aberrations and associated gratings of different frequencies, the response of P remains
diffraction effects, in actual practice, some light would proportional to the corresponding intensity in the image.
encroach into the dark space, with consequent reduction The image scan therefore directly provides Imax and Imin ,
in image contrast. As the two object points are brought and hence the contrast. In the case of square-wave gratings,
closer, the contrast decreases until it becomes zero; this is an intermediate electronic subsystem is introduced, which
taken as the resolution limit. allows only the fundamental of each frequency to be passed,
According to Fourier mathematics, any periodic function thus indirectly achieving generation of sine-wave targets.
can be expressed in the form of a series containing Since, for good performance, the object contrast at all
sinusoidal terms. frequencies must be identical, the actual realization of
Upon combining the above two concepts, the OTF variable frequency sinusoidal or square-wave gratings is
can be taken as a functional representation of the image in itself a cumbersome task. Furthermore, in both cases,
contrast and location corresponding to various spacings of since the object frequency range is limited with regard
sinusoidally varying objects. to generation of higher frequencies, a relay lens located
The OTF and the point spread function are interrelated between O and L must be employed. This introduces its
according to Fourier transformation rules, each being trans- own defects into the final result, and hence, it should have
formable into the other, the point spread function being much better corrections compared to those of the test lens.
the representation of intensity variations in the image of a
point object. Applying Fourier transform rules, it can fur-
ther be shown that the OTF is the autocorrelation of the
P
pupil function, where the pupil function is the intensity
distribution of the lens system at the exit pupil and auto- O L S C
correlation is a standard mathematical operation (Francon,
1963; O’Neill, 1963). M D
Mathematically, the OTF can be expressed in the form Figure 8. Scanning method of determining OTF: O, object; L,
test lens; S, scanning slit; C, condenser; P, photomultiplier tube;
O(ξx , ηy ) = M(ξx , ηy ) exp [−ikφ(ξx , ηy )] (9) M and D, detection and display.
590 Measuring System Design
Table 2. Comparison of test objects in OTF evaluation.
Object shape Advantages Disadvantages
Sine-wave grating Direct reading Difficult to make; limited range of spatial frequencies;
intermediate relay lens needed for obtaining higher
frequencies
Square-wave grating Direct reading Same as in (1). Additional electronic subsystem needed for
filtering fundamental of each frequency
Two counter rotating gratings Rapid direct display Same as in (2)
Pinhole or single slit Easy to make; wider range of Low light efficiency; computer needed for data conversion;
spatial frequencies correction factor needed for finite width
Knife-edge Same as in (4). Also, better Computer needed for data conversion; stray light affects the
light efficiency compared to OTF measurement to the largest degree; very low signal
pinhole or single slit to noise ratio at high spatial frequencies
The difficulties are met to the same extent by choosing
objects of simple geometry, like a single slit, a pinhole, or A
B
a knife-edge, but these have other disadvantages. C
Table 2 summarizes the relative advantages and disad- D
vantages of commonly used test objects. E
9.3 OTF evaluation – interferometric methods MTF
The desired autocorrelation (or self-convolution) is Spatial frequency
achieved in lateral shearing interferometers; this provides
the basis of an analog technique of OTF measurement Figure 9. Qualitative representation of MTF curves: A, diffrac-
(Hopkins, 1955). It is known that if the path difference tion-limited system; B, with central obstruction; C, defocusing
between the two laterally sheared beams is changed, the effect; D, E, different designs of the same optical system.
total light flux across both of the sheared beams varies
sinusoidally. Further, the shear magnitude represents the respectively; and D, E are representations of different
spatial frequency. Hence, by variation of the shear and practical designs of the same optical system.
measurement of the corresponding contrast, the OTF can It might be noted that while D has better performance at
be evaluated. lower spatial frequencies, the quality of E is superior in the
Polarizing type, lateral shear interferometers for OTF high frequency domain. Thus, MTF curves give a reliable
measurement have also been developed (Francon, 1966). quick answer and enable fast selection of a good lens for a
particular optical system.
9.4 OTF applications
10 HOLOGRAPHY
OTF analyzers have been applied with practical advantage
in many interesting situations. Examples are as a go/no- In the conventional image recording process, the photo-
go gauge for final checking of optical elements, giving graphic emulsion records the time average square modulus
the cumulative effect of residual aberrations, diffraction, of the complex amplitude function. Hence, the phase infor-
and inaccuracies caused during manufacturing processes; mation contained in the object wavefront is lost. In 1948,
providing means of evaluating performance of complex Gabor originated the basic ideas of wavefront reconstruc-
cascaded electro-optical systems; effecting improvement tion through which both the phase and amplitude informa-
in the quality of optical system at the design stage; and tion could be obtained.
many more. It involves two steps. First, a complex interferogram
As an illustration, qualitative representation of MTF (called a hologram), formed by interference between the
curves is made in Figure 9. Curve A is the plot of a object and reference wavefronts, is recorded. Second, the
diffraction-limited (theoretical best) system; B and C show hologram is suitably illuminated to reconstruct the original
the effects of adding a central obstruction and of defocusing object beam in space.
Optical Elements 591
Holography remained in a dormant stage until the early forms a real image in front. Angular separation between
1960s when highly coherent (long coherence length) laser them, and from the axially traveling beams, depends upon
sources became available. Considerable improvements in the angle between the object and reference beams during
the process were effected. With the laser becoming an recording. The reconstructing beams need not be identical
integral part of the practice of holography, this subject to the recording reference beam. For example, use of a
developed very fast into an active field that has provided longer wavelength and more divergence give rise to image
new measurement techniques and solutions to a wide range magnification.
of difficult problems. In the case of a reflection hologram, reconstructed
wavefronts are obtained on reflection under white light
illumination. Holographic images are three dimensional in
10.1 Recording of a hologram nature, retaining the parallax effect.
In principle, holography is a kind of lensless photography.
The object is illuminated by laser light. The transmitted 10.3 Applications of holography
or scattered object beam is allowed to interfere with a
reference beam derived from the same source. A suitably The basic concepts of holography were initially introduced
placed photographic plate records the hologram. to achieve very high magnification in microscopy (Gabor,
Any geometry employed in a comparison-type inter- 1948). Using a shorter wavelength, for example, an electron
ferometer, with suitable modification to ensure that the or X-ray beam, for recording, carrying out reconstruction
reference beam subtends a certain minimum angle with the with a visible radiation wavelength, provides a very high
object beam at the plane of recording, can be used in mak- order of magnification in addition to large depth of focus.
ing holograms. Holograms, because of the three-dimensional imaging
The quality of a hologram is influenced by several properties, are used for display and demonstration purposes.
factors, namely, the various modes present in laser light, Three-dimensional motion and television pictures by holo-
diffraction caused by the laser aperture speckles, and stray graphic principle are potential applications, though yet to
vibrations. Their effects are minimized to some extent by be fully developed.
applying spatial filtering of the incident laser beam using a Holographic interferometry utilizes the interference pat-
pinhole placed at the focus of a microscope objective; by terns(s) formed by making multiple exposures on the same
properly attenuating the reference wavefront for intensity photographic plate, corresponding to different stages of the
matching; by mounting the experimental setup on a good object wavefront in a dynamic situation. Vibration analyses
vibration isolation table; and by employing high-speed, have thus been possible by making a continuous exposure
high-resolution photographic emulsions (nonconventional of the vibrating object. Instead of photographic plates, the
media are also used) for recording of the hologram. interference patterns can be recorded in a computer, and
When the interfering beams fall on the same side of the online monitoring of the vibrating object is now possible.
recording emulsion, a transmission hologram is obtained. Phase change in the surrounding medium due to the pas-
To record a reflection hologram, they are introduced from sage of a high velocity missile, thermal gradients of objects,
opposite sides of the emulsion, enabling the interference to hot spots in electronic integrated circuits, mechanical strain,
take place within the depth of the photosensitive layer. and particle size distribution are some of the applications
of this technique.
The holographic technique has also been applied in the
10.2 Wavefront reconstruction fabrication of diffraction gratings and simulated optical
elements and in data storage and information processing.
When a transmitting hologram is illuminated by a reference Standard holographic technique is very sensitive to envi-
wavefront (spatially filtered as before), the object wavefront ronmental disturbance, making the measurement difficult. A
and its complex conjugate are reconstructed in space modified technique called shearography (Hung and Durelli,
together with two more beams of light travelling near the 1979) is robust and free from interference from the sur-
transparency axis. The latter two beams consist of the rounding. This technique compares the deformation of one
directly transmitted reference beam and a beam whose point B in the object with respect to another undeformed
intensity is proportional to the object intensity; these are point A in the same object (see Figure 10). A coherent
of little consequence. The object wavefront produces a optical beam from the light source is split into two beams,
virtual image of the original object behind the hologram (as one hits the object at B and the other at A. We assume
seen through it), while the complex conjugate component the surface of the object is reflective. The reflected beams
592 Measuring System Design
Hobbs, P.C.D. (2000) Building Electro-Optical Systems: Making
Coherent light B It All Work, Wiley-Interscience.
source
Hopkins, H.H. (1950) Wave Theory of Aberrations, Oxford Uni-
versity Press, London.
Beam splitter Hopkins, H.H. (1955) Interferometric Methods for the Study of
A Diffraction Images. Optica Acta, 2, 23.
Horne, D.F. (1974) Dividing, Ruling and Mask-making, Adam
Object Hilger, Bristol.
Hung, Y.Y. and Durelli, A.J. (1979) Simultaneous Measurement
Computer
Photodetector of Three Displacement Derivatives using a Multiple Image-
shearing Interferometric Camera, Journal of Strain Analysis,
Figure 10. Shearographic setup for monitoring defects. 14(3), 81–88.
Jacobs, D.H. (1943) Fundamentals of Optical Engineering,
from A and B fall onto the photodetector, which is a square McGraw-Hill, New York.
law detector. Hence, interference fringes are obtained and Jain, P.K. and Sydenham, P.H. (1980) Radial Metrological Pat-
recorded in the computer. Now, if B undergoes a deforma- tern Generating Engine. Journal of Physics E: Scientific Instru-
tion while A remains unchanged, a new set of interference ments, 13, 461–6.
fringes is obtained and recorded in the computer. The com- Levi, L. (1968) Applied Optics, Vol. 1, Wiley, New York.
puter then subtracts the two sets of fringes. The result is a Macleod, H.A. (1969) Thin Film Optical Filters, American Else-
set of new fringes due to the deformation at B. Instead of vier Publishing Company, New York.
using the computer for signal processing, a photographic Murata, K. (1966) Instruments for the measuring of optical
film can be used. In this case, the film has to undergo a transfer functions, in Progress in Optics, Vol. V, (ed E., Wolf),
double exposure before development. North-Holland, Amsterdam.
O’Neill, E.L. (1963) Introduction to Statistical Optics, Addison-
Wesley, Reading, MA.
REFERENCES Prasad, J., Mitra, G. and Jain, P.K. (1975) Aberration of a
System of Arbitrarily Inclined Planar Surfaces Placed in
Ahmad, A. (1996) Handbook of Optomechanical Engineering, Non-collimated Light Beam. Nouvelle Revue d’Optique, 6,
CRC Press, Baco Raton, FL. 345.
Buchdahl, H.A. (1954) Optical Aberration Coefficients, Oxford Prasad, J. and Singh, R. (1970) Tolerances in the Manufacture
University Press, London. of Precision Optical Components. Research and Industry, 15,
15.
Clark, A.D. (1973) Zoom Lenses, Adam Hilger, Bristol.
Schade, O.H. (1964) An Evaluation of Photographic Image Qual-
Coltman, J.W. (1954) The Specifications of Imaging Properties by
ity and Resolving Power. Journal of the SMPTE – Society of
Response to a Sinewave Input. Journal of the Optical Society
Motion Picture and Television Engineers, 73, 81.
of America, 44, 468.
Smartt, R.N. and Strong, J. (1972) Point-diffraction Interferome-
Conrady, A.E. (1957 and 1960) Applied Optics and Optical
ter. Journal of the Optical Society of America, 62, 737.
Design, Parts I, II, Dover, New York.
Smith, W.J. (1966) Modern Optical Engineering, McGraw-Hill,
Cox, A. (1964) A System of Optical Design, Focal Press.
New York.
Dyson, J. (1970) Interferometry as a Measuring Tool, Machinery
Smith, W.J. (1978) Image Formation: Geometrical and Physi-
Publishing Company, Brighton.
cal Optics, in Handbook of Optics, (eds W.G., Driscoll and
Francon, M. (1963) Modern Applications of Physical Optics, W., Vaughan), McGraw-Hill, New York.
Wiley, Chichester.
Smith, W.J. (2000) Modern Optical Engineering, McGraw-Hill
Francon, M. (1966) Optical Interferometry, Academic Press, New Professional, New York.
York.
Sydenham, P.H. (1969) Position-sensitive Photocells and Their
Gabor, D. (1948) A New Microscope Principle. Nature, 161, 777. Application to Static and Dynamic Dimensional Metrology.
Hartfield, E. and Thompson, B.J. (1978) Optical modulators, in Optica Acta, 16, 377–89.
Handbook of Optics, (eds W.G., Driscoll and W. Vaughan) Weber, M.J. (2002) Handbook of Optical Materials, CRC Press,
McGraw-Hill, New York. Baco Raton, FL.
This article is based in part on a chapter by J. Prasad and G. Mitra originally published in Handbook of Measurement Science Vol 2,
P.H. Sydenham, Wiley, 1983.
91: Light Sources and Detectors
Miroslaw Jonasz
MJC Optical Technology, Beaconsfield, Quebec, Canada
1.1 Incandescent filament lamps
1 Light Sources 593
2 Light-emitting Diodes (LEDs) 595
Incandescent filament lamps generate electromagnetic radi-
3 Lasers 595 ation by resistively heating a filament, made usually of
4 Detectors 597 tungsten, to about 3000 K. The filament temperature, and,
5 Thermal Detectors 597 thus, the maximum of the spectral output (Table 1) depend
6 Photon Detectors 599 strongly on the operating voltage, typically in a range of few
Related Articles 602 volts to a few hundred volts. Various configurations of the
filament, usually coiled-coil, and the envelope are widely
References 602
available, including designs featuring integrated ellipsoidal
Further Reading 602 dichroic reflectors, transparent in the infrared (IR), that
increase the visible output power.
Spectral output of these lamps (Figure 1) resembles,
1 LIGHT SOURCES but is lower than, thermal (blackbody) radiation spectrum
with about 90% of the spectral output in the infrared.
Light sources may be selected on the basis of several The spectral output of the filament is modified by the
criteria, including: transmission spectrum of the envelope (glass or quartz).
• power Being roughly omnidirectional, extended-area light sources,
• spectral range these lamps make it difficult to collimate or focus
• wavelength tunability light output, for example, launch light into an optical
• duration of emission (continuous, also referred to as fiber.
CW for continuous wave or pulsed) Incandescent tungsten lamps are available as vacuum
• coherence lamps, inert gas filled lamps, and halogen lamps, in the
• focusability. order of the increasing filament temperature. In halogen
Two groups of light source types have evolved commer- lamps, a small quantity of iodine or bromine compounds
cially added to the inert gas fill used maintains the lamp out-
put at close to 100% throughout the lamp life, provided
• low-coherence broad-spectrum sources (incandescent the lamps are operated at their nominal voltage to main-
filament lamps, arc lamps, pulsed arc lamps, infrared tain the envelope temperature required for the halogen
incandescent sources, light-emitting diodes, i.e. LEDs) cycle.
• high-coherence, nearly monochromatic sources (lasers). Owing to a relatively large thermal mass of the fila-
Key parameters of light sources are summarized in ment, incandescent lamps are used generally as continuous,
Table 1. Irradiance spectra of representative light sources steady-output light sources. For high-stability applications,
are shown in Figure 1. stabilized DC power supplies are required.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
594 Measuring System Design
Table 1. Key characteristics of electromagnetic radiation sources (CW = continuous output).
Approximate Approximate maximum Approximate Approximate pulse Light source type
spectral range output wavelength power range length range
150 nm–2.5 µma,b 500 nm 1–300 kW 1 ns–1 ms Gas arc lampc
0.1–10 Jg
150 nm–20 µme Variesh 0.1 mW–10 kW 10 fs–10 ms Laserd
1 pJ–20 Jg
250 nm–2.5 µmb 0.8–1.2 µmf 1–10 kW – Incandescent filament lampc
460–950 nm Varies 0.1–1 mW 1 ns–CW LEDc
1–30 µma Varies 1–100 W – Infrared light sourcesc
0.1–100 µma Varies 0.1–1000 W – Blackbodyc
a
Broad, thermal spectrum.
b
Sharp spectral peaks at wavelengths dependent on the fill gas.
c
Incoherent light source, difficult to focus and collimate.
d
Coherent light source, easy to focus and collimate.
e
Nearly monochromatic output at one or more wavelengths within this range.
f
Depends on the filament temperature and lamp type (halogen, gas filled, or vacuum).
g
Pulsed output.
h
May be wavelength-tunable.
1000 1.2 Arc lamps
Irradiance at 0.5 m, mW m−2 nm−1
100
1.2.1 CW arc lamps
10
In arc lamps, optical radiation is generated by an electric
1 discharge (as an arc) through ionized gas.
The spectral output of an arc lamp (Figure 1) consists
0.1 of a broad, blackbody-type output and sharp peaks at
wavelengths that depend on the gas fill. Gases used in these
0.01
lamps are typically deuterium, mercury vapor (Hg), xenon
0.001 (Xe), and a Hg–Xe mixture. The deuterium lamp outputs a
smooth spectrum usable from about 160 (envelope limited)
0.0001 to about 400 nm with spectral peaks in an essentially
100 1000 10 000 100 000
Wavelength, nm unusable, owing to low spectral power density, visible
spectral range. The Hg and Xe lamps (envelope-limited
Deuterium Hg arc (peaks in UV and VIS)
TH TH envelope useful range of 250 to 2500 nm) have major peaks at 365.0,
N laser (337 nm) Ar-Kr laser (488-647 nm) 404.6, 434.8, and 546.1 nm (Hg) and between 750 and
CO2 laser (10200 nm) 1550 nm diode laser
1000 nm (Xe). Owing to a high gas pressure in these lamps
Xe arc (peaks in IR) (of the order of 10 to 100 atm at the operating temperature),
SiC
He-Ne laser (633 nm) the spectral peaks are significantly collision broadened (on
the order of several nanometers).
Figure 1. Spectra of representative light sources. Thermal
(broadband) light sources spectra are based on the data The generally small arc (on the order of 0.1 to 1 mm)
from a catalog of Oriel Instruments (now Spectra Physics, allows relatively efficient collimation or focusing of the
www.spectra-physics.com): 150-W Xe arc lamp, 200-W Hg arc lamp radiation. Compact Xe arc lamps containing an inte-
lamp, 30-W deuterium lamp, 100-W tungsten halogen (TH) lamp grated sealed-in paraboloidal reflector that produces a ‘col-
(dashed line above about 5000 nm represents the output of the limated’ beam with a moderate divergence are commer-
lamp envelope), and 50-W silicon carbide envelope-less infrared
element. Wavelength ranges of the lamps’ spectra are limited
cially available (CeraLux, Luxtel, www.luxtel.net). The
by the envelopes – in the case of a quartz envelope to a range arc position in an arc lamp varies in time (arc wander) and
of about 200 to between 2500 and 3500 nm. Spectral lines of may cause variations in the output power, especially when
few major fixed-wavelength laser types are indicated symbolically this output is collimated or focused.
with short dashed vertical lines. The vertical range of these lines
does not represent the magnitudes of the irradiances that are
many orders or magnitude higher than the maximum of the graph 1.2.2 Xe flash (pulsed) lamps
scale. For example, a typical 1-mW HeNe laser (λ = 633 nm,
λ = 0.0002 nm, beam diameter 1 mm2 ) produces irradiance on These arc lamps range from small flashlamps for instru-
the order of 5 × 109 mW m−2 nm−1 . ments, to large flashlamps such as those used as pump
Light Sources and Detectors 595
sources in lasers. The flash (of the order of microseconds a discrete wavelength and irradiance pattern in a plane
long) is produced by discharging a capacitor. Flash repeti- perpendicular to the optical axis of the cavity. The gain
tion rates are on the order of 100 pulses/s. medium is selected so that its absorption band overlaps with
the wavelength of at least one of the cavity modes and that
the radiative energy transition is the most probable transi-
1.3 Infrared incandescent sources tion corresponding to that absorption band. In lasers that
employ gain media with wide spectra, prisms or gratings
The radiating elements of these sources are made of high-
are used to select a subrange of the gain medium spectrum.
temperature-resistant materials such as silicon carbide (Glo-
By ‘pumping’ the gain medium with an electric dis-
bar), and a sintered mixture of oxides of zirconium, yttrium,
charge, current, or light, one excites electrons of that
thorium, or erbium (Nernst glower). These elements, with
medium to an energy state. Electrons falling (initially spon-
sizes on the order of several millimeters, are heated resis-
taneously) from that state to a lower one emit light of
tively to a temperature on the order of 1000 to 2000 K. The
various radiation modes. Competition, via stimulated emis-
radiating elements are operated in the air with no envelope,
sion, of modes supported by the cavity for energy supplied
limiting the spectral ranges of these sources (Table 1).
to the gain medium quickly leaves a few dominant modes
Incandescent IR sources have spectral outputs (Figure 1)
inside the cavity. The simplest and commonly provided
similar to that of the blackbody, but their wavelength- mode, TEM00 , has an axially symmetric Gaussian irradi-
dependent emissivity is less than unity, typically on the
ance profile that is frequently associated with the laser beam
order of 0.9. Blackbody sources operable at temperatures in general. The process of generation of light by a laser is
ranging from 400 to 3000 K are brighter (emissivity >0.99), more akin to that used in an electronic frequency generator
but an order of magnitude more expensive alternatives to than to optical noise generation characteristic of a thermal
these IR emitters. light source.
The cavity is deliberately made slightly leaky (<1% loss)
2 LIGHT-EMITTING DIODES (LEDs) in order to let laser radiation out. In pulsed lasers that use
Q-switching technique for generation of high-power mod-
The light-emitting diodes are relatively narrowband erately short pulses (on the order of nanoseconds long),
(20–50 nm), low-voltage (2 V), and low-power sources the cavity quality (Q-factor) is spoiled for a sufficient time
(Table 1) available in a wide spectral range spanning the to build a significant population of electrons in an excited
visible and near IR. White-light LEDs that use a phosphor state and then it is temporarily restored to generate a short
to convert the base emission wavelength range into a broad pulse of radiation. Trains of much shorter pulses (several
range perceived as white light are also available. LEDs femtoseconds) can be obtained by mode locking, that is,
convert electric current (∼20 mA) to light by recombining introducing a fixed phase relationship between the cavity
charge carriers in semiconductors. The semiconductors used modes. The pulse train is a result of interference between
include SiC, GaN, ZnSe, GaInN (blue), GaP (green), GaP:N the locked modes.
(yellow), and AlInGaP, GaAsP (red). The cooperation of the resonant cavity properties, gain
LEDs are available in a wide range of packages, ranging medium absorption profile, and the stimulated emission
from plastic encapsulation to TO-type metal transistor gives the laser beam its directionality, high monochromatic-
cans with flat or lensed windows. The emitting area is ity, and coherence. Stabilized frequency lasers have line
on the order of a fraction of 1 mm in size. Plastic and widths <1 × 10−11 nm (3 MHz), and coherence lengths on
glass lensed packages provide rough collimation of the the order of 1 km. Key characteristics of representative laser
output, but a better result in collimating or focusing a types are summarized in Table 2. Wavelengths of major
LED output can frequently be obtained with flat-window laser types are shown in Figure 1.
packages that generally provide a nearly Lambertian spatial
radiation pattern.
3.1 Gas lasers
3 LASERS These lasers span a wide wavelength range from UV
(excimer, for example, XeF, XeCl; nitrogen), through vis-
The laser (for example, Hecht, 1992) consists of an ible (argon, argon/krypton, copper vapor, helium–neon,
optical resonator cavity, created by two opposing, high- helium–cadmium) to far IR (CO2 ). The gain medium is
reflectance mirror surfaces (generally slightly concave), and gas excited by an electric discharge. The gas lasers provide
gain medium inside that cavity. The cavity supports a set good-quality low-divergence beams with the diameters on
of optical radiation modes (resonances) each embodying the order of 1 mm.
596
Table 2. Basic data on major laser types listed according to the increasing first wavelength of the spectral range. CW denotes continuous output. This table is mainly
based on the Laser FocusWorld 2003 Buyer’s Guide. Additional data obtained from specifications of selected manufacturers have been also used. Users of this table should
exercise caution when applying the data given.
Wavelengths (nm)m Laser type Power (W) or pulse Pulse repetition Pulse width (s) Beam Beam
energy (J) rate (1 s−1 )l sizea divergence
(mm) (mrad)
Measuring System Design
154–532f , (1064), 2950 Nd:YAGc (pulsed) 0.001–180 J 1–50 × 103 20 × 10−12 –10 × 10−3 1–14 0.5–130
157, 193, 248, 308, 351 Excimerb (pulsed) 0.001–0.2 W 20–2 × 103 0.01–30 × 10−9 2.5 × 22 0.3–4
190–18000h,k Tunable solid state (pulsed) 0.1 × 10−9 –100 J 0–90 × 106 17 × 10−15 –100 × 10−9 3–10g 0.5–1.5g
197–4500h Tunable dye (pulsed)g 30 × 10−6 –0.2 J 1–10 × 103 5 × 10−12 –20 × 10−9 1–3 × 6 5
202–1090i (514) Argonb (CW) 2 × 10−3 –25 W – – 0.6–3.5 0.05–3
266–532f (1064), 1320 Nd:YAGc (CW) 0.01–5.4 × 103 W – – 0.7–6 1–200
(337) Nitrogenb,g (pulsed) 0.1 × 10−3 –0.5 × 10−3 J 1–100 ∼1 × 10−9 1×2 2×3
375–760h Alexandritec,g (CW) 0.01–0.15 W – – 1 0.1
375–2900h (500–1550)h Semiconductor diode (CW) 0.1 × 10−3 –4 × 103 e W – – – 1–17 ×
1–17j,d ,
87 × 870d
380–1935h (600–1550)h Semiconductor diode (pulsed) 5 × 10−12 –102 e J 100–100 × 106 15 × 10−12 –0.8 × 10−3 – 0.3j , 87 × 785d
543, (632.8), 1152, 3392 HeNeb 0.1 × 10−3 –35 × 10−3 W – – 0.5–2.7 0.7–4.5
690–2000h Ti: sapphirec,g (CW) 0.05–3.5 W – – 0.6 1.7
694 Rubyc,g (pulsed) 0.08–20 J 0.005–0.1 20 × 10−9 –2.5 × 10−3 7–25 0.2–1
3000–25 000 Lead salt diodeg (CW) 0.1 × 10−3 W – – – 870
5200–6500h COb,g (CW) 10–1 × 103 W – – 5–16 4
9000–11 400 (10 600) COb2 (CW) 0.04–12 × 103 W – – 0.5–22 0.5–9
9000–11 400 (10 600) COb2 (pulsed) 0.5–60 J 0–130 × 103 0.3 to over 10−3 1.8–30 × 30 1–10
a
If one value is given, it is the TEM00 (Gaussian) beam diameter.
b
Gas laser.
c
Solid-state laser.
d
In two perpendicular dimensions.
e
The upper end of the range refers to the output of laser stacks.
f
Through harmonic frequency conversion.
g
Few data.
h
Continuously tunable, may be tunable in a subrange of the stated range.
i
Line selection.
j
Diode lasers equipped with collimating optics.
k
Line widths from several GHz down to 1 kHz.
l
A repetition rate of 0 denotes a single-shot mode.
m
Dominant wavelengths in parentheses.
Light Sources and Detectors 597
3.2 Semiconductor (diode) lasers 3.4.2 Tunable lasers
Tunable solid-state lasers, available mainly as pulsed lasers,
These lasers (for example, Camparo, 1985) are similar
operate by using principles similar to those employed in
to LEDs in that they use a recombination of charges in
dye lasers except that the gain medium is solid state. In
semiconductors to generate emission of light. Owing to a
particular, a tunable titanium-doped-sapphire (Ti:sapphire –
very small emitting area of unequal sizes (on the order
the broadest range of any tunable laser: 660–1180 nm) laser
of 10 µm) in two perpendicular directions, these lasers
has become a widely used replacement for tunable dye
typically emit a divergent astigmatic light beam and require
lasers. However, the Ti:sapphire laser has the noise lower
complicated optical systems to either collimate or focus it.
by a factor of 10 as compared to the dye laser and also has
In fiber-optics applications, the small emitting area can be
a much narrower line width (down to 1 kHz). As with other
simply butt-coupled to a fiber.
SSLs, up-conversion is used to provide tunable light in the
Semiconductor diode lasers are also available in verti-
UV and visible range.
cal emitting cavity (VCSEL) configuration that emit low-
divergence, circular beams. Although these lasers are more
difficult structures to grow on a semiconductor wafer, 4 DETECTORS
they do not need to be cleaved to produce the resonant
cavity and, thus, can be fabricated solely by using stan- Developments in electro-optical detector technology (for
dard semiconductor processing techniques (for example, example, Donati, 2000) combined with advantages of direct
Lang, 2002). electronic signal processing have long displaced the human
eye and photographic film as detectors of light in all but
marginal applications. Two basic groups of electro-optical
3.3 Dye lasers detectors are discussed here: thermal and photon detectors.
An electro-optical detector of light is selected on the basis
Dye lasers use as their gain media various dyes (including
of one or more of the characteristics listed in Table 3. Some
coumarin, fluorescein, and rhodamine) covering a wave-
detector types may have additional selection criteria, for
length range from about 190 to 4500 nm. These lasers are
example, after-pulsing for photomultipliers.
pumped with either flashlamps or other lasers, for example,
The sensitivity is frequently the dominant selection crite-
nitrogen lasers. The wide absorption bands of the dyes,
rion. Hence, the ranges of light power that can be measured
as compared, for example, to narrow absorption spectra of
with the various detector types are listed in Table 4 and
gases, support the generation of very short light pulses (sev-
shown in Figure 2 for major detector types.
eral femtoseconds long).
3.4 Solid-state lasers 5 THERMAL DETECTORS
3.4.1 Fixed-wavelength lasers In a thermal detector, the active element is heated via
absorption of light. This heating induces electrically sensed
Solid-state lasers (SSLs) utilize a solid gain medium, per- changes in that element’s properties. Thermal detectors
mitting high-quality beams, and various pumping mech- have a flat spectral response in a broad spectral range, deter-
anisms. The first solid-state laser used the ruby crystal mined by the absorption spectrum of the sensing element
as the gain medium and a flashlamp as the pump. SSLs material or its absorptive coating that converts the energy
are also pumped by other lasers and by semiconductor of the incident radiation to heat. The detector spectral range
diode lasers (diode-pumped solid-state lasers – DPSSL). is limited/modified by the detector window/filter. A weak
In addition to ruby, the gain media include neodymium- thermal link to a heat sink allows the detector to correctly
doped (Nd:YAG, Nd:YLF, Nd:YVO4 ) and holmium-doped respond to the incident radiation power. Thermal detectors
(Ho:YAG, Ho:YLF) crystals and glasses, alexandrite and are generally less expensive than photon detectors, and do
titanium-doped sapphire (Ti:sapphire). As in dye lasers, the not require cooling when operated in the infrared. However,
wide absorption bands of the solid gain media support the thermal detectors are typically slower and have a lower
generation of extremely short light pulses. sensitivity than photon detectors.
These lasers emit in the red and near IR range. Radiation The basic types of thermal detectors are:
in the near-UV and blue/green region of the visible spec-
trum can be generated by the harmonic up-conversion (for • thermocouples
example, 532 nm in an ND:YAG laser). • thermopiles
598 Measuring System Design
Table 3. Key characteristics of radiation detectors.
Name and typical symbol Typical units Definition
Responsivity, R A/W, V/W, V/Ja Ratio of the detector output to the incident light power dose
Quantum efficiency, QE b Nondimensional Number of electrons generated by the detector in response to an
input of one photon of radiation
Dark current, ID , or dark voltage, VD A, V The baseline output of the detector in the absence of incident light
Response time, tR s Time interval required for a detector to generate an output
commensurate with the input light power (dose) when that detector
is illuminated by a step increase in the light power (rise time) or
when the illumination ceases abruptly (fall time)
Linearity range W A range of the input light power within which the detector output is
√ linearly proportional to the input power
Noise equivalent power, NEP W, W/ Hz Light power that generates a detector output equal to the RMS
detector noisec
a
For detectors sensitive to the radiation dose.
b
Applicable to photon detectors only.
c
With the detector noise being a function of the measurement (electrical) frequency bandwidth, NEP is usually given at a frequency of 1 kHz and for a
1-Hz bandwidth. An amplifier, if needed, may significantly increase the NEP of a complete detection system.
Table 4. The first order of magnitude incident radiation power ranges that can be measured by the various detector types. The minimum
power is the detector noise equivalent power at the wavelength of the maximum responsivity. PMT = photomultiplier, APD = avalanche
(semiconductor) photodiode.
Approximate power rangeh , W Detector type Approximate spectral rangeb
Minimum Maximum
10−19 a 10−12 PMT: photon countinge 100–900 nm
10−19 a 10−12 APD: photon countinge 200–2500 nm
10−15 10−9 c PMT: analog detectione 100–900 nm
10−13 10−6 APD: analog detectione 200–2200 nm
10−12 10−1 Semiconductor photodiodee 0.1 nm–5 µm
10−12 1 Semiconductor photoconductore 400 nm–150 µm
10−9 10−3 Thermopilesf 200 nm–200 µm
10−9 10−3 Pyroelectricf 200 nm–1 mm
10−8 d 1 Bolometerf 0.1 nmg –5 mm
a
This power corresponds to ∼1 photon/s in the visible and near infrared.
b
Various detectors of the stated type (except thermal detectors) may have their maximum sensitivity located at different wavelengths in this range and
be sensitive each in a subrange of this range.
c
A much higher maximum power (about 7 mW) was measured with short (∼1 µs) light pulses (Lasher and Redhead, 1963) at the cost of high mortality
of the PMTs.
d
down to 10−17 in the far infrared at a temperature close to 0 K.
e
Photon detector.
f
Thermal detector.
g
Can be used as a photon counter/spectrometer for X rays.
h
The dynamic range of a detector may be smaller than the min–max range.
• pyroelectric detectors by a junction of dissimilar metals, usually bismuth and
• bolometers. antimony. The thermocouple consists of two such junc-
Golay cells (pneumatic detectors) were the mainstay tions connected in series in reverse in order to cancel
thermal detectors in the past but have been all but displaced the effect of the ambient temperature – see Article 196,
by other thermal detector types. Thermocouple Temperature Sensors, Volume 3. One of
these junctions is coated with a thin light-absorbing layer.
The other is shielded from the incident light, that is,
5.1 Thermocouples and thermopiles remains at the ambient temperature. The voltage generated
by a thermocouple is low (in the mV range); therefore,
Thermocouples and thermopiles (a series of thermocouples) several thermocouples are typically connected in series
utilize the temperature dependence of voltage generated (thermopiles).
Light Sources and Detectors 599
1.E− 08 expands. This changes the polarization of this element and
1.E− 09
causes current to flow in a circuit connected to it. Pyro-
electric detectors thus respond to changes in the incident
1.E−10
light power, that is, to chopped or otherwise modulated
NEP, W Hz1/2
1.E−11 radiation power.
1.E−12
1.E−13
1.E−14
5.3 Bolometers
1.E−15
Bolometers utilize changes in the conductivity of a resistor
1.E−16 to sense its temperature that is affected by absorption
100 1000 10 000 100 000
Wavelength, nm of radiation. Room-temperature bolometer arrays allow
inexpensive imaging of IR radiation, as compared to other
Cs-1 PMT (120 nm) GaAs(Cs) PMT (720 nm)
Si PD G = 1e8 (960 nm) Si CCD at 273 K (710 nm) IR imaging arrays.
PbS PC at 77 K (2400 nm) PbSe PC at 77 K (4020 nm) When operated at room temperature, these devices are
InSb PD at 77 K (4700 nm) HgCdTe PC at 77 K
(17000 nm)
relatively poor detectors. Chopping, or modulating the inci-
dent radiation power, can be used to reduce the effect
Si APD G = 100 (840 nm)
Ge PD at 298K (1500 nm) of slow-changing ambient temperature. Room-temperature
In GaAs PD at 298 (2200 nm) bolometer detectors may also utilize two identical bolome-
ters (one exposed to the incident radiation, the other
Figure 2. Sample spectra of the noise equivalent power (NEP)
for major types of photon detectors (the wavelength of the min- shielded from it) in two arms of an electrical bridge. This
imum NEP is given in parentheses in the curve label). They arrangement cancels the effect of the ambient temperature
indicate approximately the minimum power that can be mea- on the detector output. However, the full potential of these
sured by a given detector type at a signal-to-noise ratio of devices is realized by operating them at temperatures close
1. The NEP of a specific detector of that type may vary, to absolute zero, which reduces their inherent thermal noise.
depending on the detector area and temperature. Abbreviations:
APD = avalanche photodiode, LAAPD = large-area avalanche Cryogenically cooled bolometers operating in the far IR are
photodiode, G = gain, PC = photoconductor, PD = photodiode, by several orders of magnitude more sensitive than other
PMT = photomultiplier. PC and PD detectors require external sig- detectors of electromagnetic radiation in that spectral range.
nal amplifiers that typically significantly increase the total NEP
of the detection system. Long-term signal integration (on the
order of seconds) may reduce the total NEP in such a case.
The field of view of the far IR detectors is limited to 60◦ . 6 PHOTON DETECTORS
The NEP is generally specified at a modulation frequency of
the optical input of about 1 kHz, with a bandwidth of 1 Hz. Photon detectors convert the energy of incident light
However, an external amplifier with a high gain required for photons to electricity directly. These detectors work in
the PDs may significantly limit the maximum modulation fre-
relatively narrow spectral ranges, with a distinct long-
quency. Detector areas are on the order of 0.01 cm2 , except
those of the PMTs (∼0.5 cm2 ) and the LAAPD (∼0.1 cm2 ). Pyro- wavelength cutoff, but can be extremely sensitive, up to
electric, thermopile, and bolometer detectors (not included) have the detection of single photons. Biological detectors of light
a flat spectral response with NEP ∼10−9 to 10−11 (at several are nearly as sensitive. The human eye can detect a burst
Hz) typically from UV to far IR. This spectral range is lim- of several photons and the rat eye can reliably detect single
ited/modified by the window/filter transmission range. The graph photons (Hagins, Penn and Yoshikami, 1970).
is based on the data of the following manufacturers: Si PD and
the PMTs – Hamamatsu (1994), www.hamamatsu.com, Si APD –
Perkin Elmer, www.perkinelmer.com, Si LAAPD – Advanced
Photonics, www.advancedphotonix.com, PbS and PbSe PC – New 6.1 Photoemissive detectors
England Photoconductor, www.nepcorp.com, Ge, InGaAs, InSb
PDs, and HgCdTe PC – Judson Technologies, www.judtech.com.
Photomultipliers (PMTs; BI, 1980; Hamamatsu, 1994) are
by far the most commonly used detectors in this class,
5.2 Pyroelectric detectors followed by image intensifiers.
Pyroelectric detectors use materials, typically lithium tan-
6.1.1 Photomultipliers
talate, that maintain electrical polarization even when no
voltage is applied to the detector. When an active element A classical photomultiplier is a vacuum tube with a photo-
of a pyroelectric detector is heated via absorption of light, it sensitive cathode (Table 5) and on the order of 10 dynodes
600 Measuring System Design
Table 5. Basic characteristics of typical photocathodes.
Photocathode Photosensitive Spectral range Wavelength of Responsivity Maximum
type material (nm) maximum (mA/W) quantum
responsivity (nm) efficiency
S-1 Ag–O–Cs 400–1100 350–800 3.5 0.012
S-4 Cs–Sb 160–550 300 50 0.21
S-11 Cs–Sb 290–600 450 55 0.15
S-20 Na–K–Cs–Sb 160–720 400 50 0.15
– GaAs (Cs) 160–850 400 37 0.22
that collect and multiply the photocurrent generated at the the MCP multiplier, the imaging of the input electron irra-
cathode. A high voltage of the order of 1000 V is dis- diance distribution is ensured by restricting each electron
tributed between the electrodes of a PMT by a voltage cascade to its own microchannel and by placing the photo-
divider network. A photon striking the photocathode ejects cathode, MPC, and the anode-phosphor screen close to each
an electron with the quantum efficiency (QE) of less than other (<1 mm). This creates compact devices with imaging
one-fourth. That electron is accelerated by the potential dif- resolutions of the order of 50 line pairs per mm. Photon
ferences between the cathode and the following electrode gains up to the order of 109 are possible with devices that
and – upon impacting that electrode – ejects several next- employ stacks of two or three MCPs.
generation electrons. This electron-multiplication process The IMI is by itself an optical-to-optical converter and
continues for each following dynode until the anode, where requires an array detector such as a CCD (discussed
electrons are collected, and results in a total electron gain on in Section 6.2) to convert an optical image to electron-
the order of 106 to 108 . A high gain and a low dark current ically processable image. Phosphor-less versions of an
on the order of 1 photoelectron/s (and less when cooled) image-intensified CCD are also available. In these imaging
enable the detection of single photons of the incident light. detectors, the phosphor screen is replaced with a CCD
Classical PMTs are available generally in two forms: embedded inside the vacuum tube.
• side-on (with a reflective photocathode)
• end-on (with a semitransparent photocathode).
6.2 Solid-state detectors
The side-on form is typically a faster detector (rise time
on the order of a fraction of 1 ns), due to a more com- The sensitivity spectra of solid-state detectors span a wide
pact electron multiplier design. Supercompact multianode wavelength range from the soft X rays to far IR. Spectra of
PMTs that enable a low-resolution imaging of the incident major types of solid-state detectors are shown in Figure 2.
irradiance distribution (∼10 × 10 ‘pixels’) are also avail-
able. The sensitivity spectra of PMTs span the near-UV to 6.2.1 Photovoltaic detectors and detector arrays
near-IR wavelength range (Figure 2).
A radical variance of the vacuum-tube design is a In a photovoltaic semiconductor detector, the semiconduc-
multichannel-plate PMT (MCP PMT), where the multiplier tor junction separates charge carriers generated by absorp-
section is a glass plate, about 0.5-mm thick, with closely tion of photons of the incident light and produces current
packed narrow (on the order of 10 µm) cylindrical chan- in an electric circuit connected to the detector. Photovoltaic
nels, each acting as a set of dynodes. The MCP offers the detectors can be used in the low-speed short-circuit photo-
most compact PMT design allowing rise times on the order voltaic mode (no dark current and the related noise) and
of 100 ps. in the high-speed photoconductive mode with moderate
bias current and resulting higher noise. A wide variety
of semiconductor photovoltaic detectors are available com-
6.1.2 Image intensifiers
mercially, including position-sensitive detectors and imag-
Image intensifiers (IMIs) use various forms of imaging elec- ing (array) detectors. These detectors are typically made
tron multipliers, typically the MCP, to map the distribution of Si, GaAs, Ge, InGaAs, InAs, and InSb semiconductors
of photoelectrons exiting the photocathode to that at the that absorb light in subranges of a wide range between X
anode (for example, Johnson and Owen, 1993). The anode rays and far IR (Table 6). Detectors specialized for either
is a phosphor screen (metallized to conduct electric current) low-noise, low-speed operation or high-speed operation
that emits visible light when impacted by electrons with (>100 GHz) are available. Semiconductor detectors have
energy sufficient to penetrate the screen metallization. With QE on the order of unity, much higher than that of PMTs,
Light Sources and Detectors 601
Table 6. Basic characteristics of typical photovoltaic semiconductor detectors at room temperature, unless indicated otherwise.
Photosensitive Spectral range Wavelength(s) of Maximum Maximum
material (µm) maximum responsivity, quantum
responsivity (µm) A/W efficiency
Si 0.2–1.2 0.8–0.95 0.7 0.9
GaAs 0.3–0.75 0.6–0.72 0.4 0.7
Ge 0.8–1.8 1.55 0.9 0.7
InGaAs 0.7–1.7 1.6 1 0.8
0.7–2.6 2.3a 1.1 0.6
InAs 1–3.1 3b 2 0.8
InSb 1–5.5 5.2b 2 0.5
a
At 253 K.
b
At 77 K (liquid nitrogen).
that is relatively insensitive to temperature changes, except quantum efficiency of the CCD array as compared to that
near the long-wavelength cutoff (Figure 2). of the photodiodes.
Photovoltaic detectors are available both as: Back-illuminated operation of the CCDs compensates
somewhat for this reduction. CCDs tend to suffer from
• unity gain devices (photodiodes, photodiode arrays, ‘blooming’ caused by spilling charge from an overfilled
CCDs, CMOS arrays) pixel to the surrounding pixels (Janesick et al., 1987).
• detectors with internal amplification: avalanche photo- Recently, electron-multiplying CCDs (EMCCDs) have
diodes (APDs) and APD arrays. been commercialized (e2v technologies, www.e2v.com).
These CCDs utilize, in the readout section, an electron-
In the latter detectors, the cascade multiplication of multiplication technique implemented in the APDs. This
photoelectrons is induced by applying a high voltage on allows high gain (single-photon counting) with the low-
the order of several hundred to several thousand volts. In noise characteristic of the CCDs.
the lower part of the high-voltage range, the number of The CMOS imaging arrays utilize silicon as photosensi-
electrons produced is proportional to the applied voltage, tive material and have a similar spectral response to that
as it is in the PMT. In the high part of the voltage range, of silicon CCDs. However, the CMOS arrays use a differ-
the cascade multiplication becomes free running. This latter ent approach to sensing the incident light power and also
mode of operation (Geiger mode), is used for single-photon different manufacturing techniques. In a pixel of a CMOS
counting. In that mode, the electron cascade is quenched to array, a photodiode typically discharges a storage capaci-
enable photon counting at a rate on the order of up to 106 tor, thus preventing blooming. The capacitor charge can be
photons/s (active quenching) limited by dead time and APD directly read out via an integrated amplifier.
heating (for example, Dravins, Faria and Nilsson, 2000). This allows addressing selected sections of the array and
Semiconductor array detectors are the mainstream of enables a high-speed readout. The CMOS arrays employ
imaging detectors. Arrays with pixel sizes on the order of technology used for microprocessors. This permits inte-
10 × 10 µm are available in the photodiode array, CCD, gration of preprocessing electronics, such as an analog-to-
and CMOS formats. The photodiode arrays have individu- digital converter, on the array chip. However, additional
ally addressable elements and offer high quantum efficiency semiconductor/metal structures implementing these func-
but suffer from a substantial read-out noise (for exam- tions take up a significant portion of the area available
ple, Johnson, Edwards and Mendenhall, 1994). CCD arrays for light collection. Complexity of these structures also
have a much lower readout noise but also a lower quan- increases the manufacturing defect rate. Such disadvan-
tum efficiency. In a CCD array, the photogenerated charges, tages have generally limited applications of CMOS arrays
instead of being in real time removed from the sensitive vol- to low-cost and low-quality consumer applications (Tay-
ume as it is done in the photodiode, are stored in a potential lor, 1998).
well until they are transported out with an extremely high
efficiency, along a column or row of pixels that perform
6.2.2 Photoconductive devices
the role of an analog shift register.
Charge integration and transfer are achieved at the rate Photogeneration of charges in semiconductors significantly
of an external clock signal by varying potentials of an array affects its conductivity. This is utilized in junctionless
of electrodes and semiconductor gates located at the CCD semiconductor detectors, photoconductors, for the detection
chip surface. It is these surface structures that reduce the of electromagnetic radiation. Typical semiconductors used
602 Measuring System Design
Table 7. Basic characteristics of typical photoconductive semi- Camparo, D.C. (1985) The Diode Laser in Atomic Physics.
conductor detectors at room temperature, unless indicated Contemporary Physics, 26, 443–477.
otherwise. Donati, S. (2000) Photodetectors: Devices, Circuits and Applica-
Photosensitive Spectral range Maximum tions, Prentice Hall, NJ.
material (max sensitivity) responsivity Dravins, D., Faria, D. and Nilsson, B. (2000) Avalanche Diodes
(µm) (V/W) as Photon-counting Detectors in Astronomical Photometry,
in Optical and IR Telescope Instrumentation and detectors,
CdS 0.4–0.75 (0.56) –b
(eds I. Masanori and A.F. Morwood); SPIE Proceedings, 4008
CdSe 0.6–0.9 (0.71) –b
298–307.
PbS 1–2.7 (2.2) 6 × 105
1–4.2 (3.1)a 9 × 105 a Hagins, W.A., Penn, R.D. and Yoshikami, S. (1970) Dark Current
PbSe 1–5 (4) 7500 and Photocurrent in Retinal Rods. Biophysical Journal, 10,
1–6 (5)a 55 000a 380–411.
HgCdTe 1–4a – 24a 105 a –200a Hamamatsu. (1994) Photomultiplier Tube: Principle to Applica-
a
At 77 K (liquid nitrogen). tions, Hamamatsu Photonics KK, Hamamatsu, Japan.
b
These detectors are used mainly as light switches, because of the Hecht, J. (1992) The Laser Guidebook, McGraw-Hill, New York.
sensitivity of their response to their light-exposure history.
Hudson, R.D. (1969) Infrared System Engineering, Wiley Inter-
Science, New York.
in these photoconductors are CdS and CdSe (visible), PbS Janesick, J.R., Elliot, T., Collins, S., Blouke, M.M. and Free-
and PbSe (near IR), and HgCdTe (IR) (Table 7). Note man, J. (1987) Scientific Charge-coupled Devices. Optical
Engineering, 26, 692–714.
that the semiconductor conductivity is also a significant
function of temperature, so that the detector temperature Johnson, J.B., Edwards, G. and Mendenhall, M. (1994) Low-cost,
High-performance Array Detector for Spectroscopy Based on
must be stabilized.
a Charge-coupled Photodiode. Review of Scientific Instruments,
The literature on optical engineering usually covers 85, 1782–1783.
sources and detectors. A useful general book is OSA (1993).
Lang, R.J. (2002) Semiconductor Lasers: An Overview of Com-
A useful reference work on infrared engineering is Hud- mercial Devices, in Photonics Handbook, Laurin Publishing,
son, 1969. Pittsfield, MA, (pp. H215–H218).
Lasher, D.R. and Redhead, D.L. (1963) High-current Operation
of RCA Type 2059 Photomultiplier Tubes. Review of Scientific
RELATED ARTICLES Instruments, 34, 115–116.
OSA. (1993) The Handbook of Optics, McGraw-Hill, New York.
Article 89, Optical Materials, Volume 2; Article 90,
Taylor, S.A. (1998) CCD and CMOS Imaging Array Technologies:
Optical Elements, Volume 2; Article 92, Optical Mea- Technology Review , Technical Report EPC-1998-106, Xerox
suring Instruments, Volume 2; Article 93, Testing Opti- Corporation, Cambridge, UK, (p. 15).
cal and Other Radiation Systems, Volume 2.
FURTHER READING
REFERENCES
Johnson, C.B. and Owen, L.D. (1993) Image Intensified Elec-
BI. (1980) Photomultiplier Handbook, Burle Industries, Inc, tronic Imaging, in The Handbook of Optics, McGraw-Hill, New
Lancaster. York, (pp. 21.1–21.32).
This article is based in part on a chapter by J. Prasad and G. Mitra originally published in Handbook of Measurement Science Vol 2,
P.H. Sydenham, Wiley, 1983.
92: Optical Measuring Instruments
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
An Abbe refractometer consists of two 90◦ − 60◦ − 30◦
1 Refractometers 603 prisms, a pair of Amici prisms, and a telescope. One
2 Curvature Measurement 603 right-angled prism is used for illumination, while the other
serves as the test specimen holder, which is placed over
3 Autocollimators 604
it with a liquid of high refractive index at the contact
4 Focometers 604 area. The Amici prisms are rotated in opposite directions
5 Microscopes 604 to compensate for the dispersion introduced by the test
6 Telescopes 604 specimen and stage prism. The telescope can be swiveled
7 Projectors 606 along an arc to pick up the total internally reflected light
8 Spectrophotometers 606 and its position is read on a scale directly calibrated in
terms of refractive index. The instrument also incorporates
9 Opto-medical Instruments 607
the provision for measuring the refractive index of liquid
10 Laser-based Instruments and Systems 607
samples.
Related Articles 609 In the Pulfrich refractometer, the emergent angle I of
Further Reading 609 the grazing incident rays, as measured by the telescope,
is used for the determination of the refractive index N
of the test samples, according to the equation N = (n2 −
sin2 I )1/2 , where n is the refractive index of the supporting
1 REFRACTOMETERS
glass block.
This article shows how the building blocks of optical
systems – materials, Article 89, Optical Materials,
Volume 2; elements, Article 90, Optical Elements, 2 CURVATURE MEASUREMENT
Volume 2; and sources and detectors, Article 91, Light
Sources and Detectors, Volume 2 – are assembled to Measurement of radius of curvature r of optical surfaces
form whole optical systems. Testing of optical systems is is conventionally made employing mechanical or optical
found in Article 93, Testing Optical and Other Radiation spherometers.
Systems, Volume 2. In mechanical spherometers, the sag of the test surface
Refractometers are employed for refractive index and is obtained by locating the correct contact point with the
Abbe value measurement. Depending upon the precision help of a plunger. This quantity in conjunction with other
needed, several instruments are available commercially. known parameters enables calculation of r.
Abbe and Pulfrich refractometers are the most commonly The optical spherometer is based on the autocollimation
used instruments for precise measurement. The design of principle. The instrument is first focused on the test surface
both these instruments is based upon the total internal and then at its center of curvature. The distance between
reflection principle. the two image positions determines r.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
604 Measuring System Design
The well-known Foucault knife-edge test method is usu-
ally employed for testing long radius of curvature, concave
surfaces, and assessing sphericity errors at different zones.
3 AUTOCOLLIMATORS
Parallelism of a glass plate, prism and wedge angles, and
surface flatness can be efficiently tested using an autocolli-
mator. It is also frequently used for alignment purposes.
When the surface under test is not exactly normal to the
incident rays, the reflected image is transversely displaced.
This image displacement is measured (manually and pho-
toelectrically) precisely, yielding test surface inclination.
In the angle Dekkor method, the test specimen is placed
horizontally upon a flat base, while the autocollimator can
be swung in a vertical plane. A precision goniometer is
essentially a spectrometer with an in-built autocollimat-
ing telescope.
Figure 1. Stereozoom microscope. (Copyright Bausch &
Lomb.)
4 FOCOMETERS
normal illumination), a special objective, carrying a beam
The two standard techniques conventionally used for focal splitter at its back, is used.
length measurement of an optical system are the focal When the object has only variation in phase or refrac-
collimator and nodal slide methods. For better precision, tive index and no differential absorption, a phase-contrast
an autocollimating microscope can be used. microscope is used. With the help of a phase-shifting
A wide variety of optical instruments are available for device, the variations in the object are converted into inten-
various measurements in the industry. Brief details of some sity variations in the image. On phase retardation, a denser
common types of instruments are now discussed. medium appears brighter and vice versa in the case of phase
advancement.
In fluorescence microscopy, the specimen (usually bio-
5 MICROSCOPES logical) is dyed with an appropriate dye and observed, using
filters, under ultraviolet light.
Study of the crystals is done on a polarizing microscope.
A microscope consists essentially of a condenser lens to
Plane polarized light, after passing through the specimen,
illuminate the object, an objective lens for image formation,
changes its state of polarization and the output beam is
and an eyepiece optical system for viewing. Opaque objects
studied by an analyzer.
are illuminated from the top, either by a beam splitter placed
Interference objectives are basically Twyman–Green
below the objective or by fixing miniature lamps by the
types of microinterferometers, which are coupled with
side of the objective. An industrial version of a stereozoom
a microscope to measure the microsurface structure of
microscope is shown in Figure 1.
polished reflecting surfaces.
A tool-room microscope eyepiece can be fitted with a
special graticule pattern. Microthreads, tool shape, and the
like are checked by matching their images with respect to
a corresponding ideal pattern contained in the graticule. 6 TELESCOPES
Dimension-measuring traveling microscopes are mounted
on a high-precision translating rail. The travel distance is Basically, a telescope has an objective and an eyepiece.
read from a scale using an optical vernier. Astronomical telescopes of the reflecting type contain at
In the case of metallurgical specimens (where it is impor- least a large primary mirror and a small secondary mirror.
tant to study fine scratches in polished surfaces, faults, and Terrestrial telescopes are often of the refracting type fitted
inclusions, which, due to glare, are usually not visible under with additional inverting optics for image erection.
Optical Measuring Instruments 605
Figure 3. Universal automatic level. (Copyright Wild Heer-
brugg Ltd.)
and the working range is ±2 min of arc. Graduation inter-
vals of the horizontal and vertical glass circles is 1◦ of arc
with direct reading on the micrometer to 6 s of arc and by
estimation, 3 s of arc.
In the case of the Wild universal automatic level shown
in Figure 3, the compensator setting accuracy is enhanced
to ±0.3 s of arc and working range to ±15 min of arc.
For precise leveling, a parallel-plate micrometer, reading at
0.1 mm direct and 0.01 mm by estimation, is attached to the
basic instrument.
Endoscopes are often employed in industry for obser-
vations inside high-temperature enclosures or other inac-
cessible areas. Figure 4 shows a commercial model of a
portable endoscope for photographic and visual examina-
tion of the inside walls of operating furnaces and boilers.
Photo-visual endoscopes suitable for use in very high pres-
sure (25 kg cm−2 ) chambers with corrosive atmospheres are
Figure 2. Micrometer theodolite. (Copyright Wild Heerbrugg
Ltd.)
The precision theodolite and sighting level are another
important class of telescopic instrument. They are exten-
sively used in land surveying, geodesy, and tacheometry.
Figures 2 and 3 illustrate, as examples, the optical config-
urations of standard versions of the micrometer theodolite
and universal automatic level.
As an example of an optical system, consider the techni-
cal data of a Wild micrometer theodolite shown in Figure 2.
Telescope magnification with a standard eyepiece is 30×.
The field of view at 1000 m is 27 m with a shortest focusing
distance of 1.7 m. Bubble sensitivity per 2 mm run, for the
circular level is 8 min of arc and for the plate level 30 s
of arc. Liquid compensator setting accuracy is ±1 s of arc, Figure 4. Portable endoscope. (Copyright S.R. Clave.)
606 Measuring System Design
also available. A wide range of forms for medical use is
covered in Section 9.
7 PROJECTORS
Projection equipment consists essentially of a high-power
lamp, a suitably designed condenser lens system and a well-
corrected projection objective lens. The condenser fully
illuminates the entrance pupil of the objective, which forms
a magnified image on a distant screen. The diascope is used
for the projection of transparencies (the overhead projector)
and the episcope for opaque objects.
Profile projectors are made epidiascopic in design so (a)
that both opaque and transparent objects may be imaged. S
A set of projection objectives, of different magnifications,
is mounted on a revolving turret. The test specimen is
mounted on a glass platform provided with X–Y trans-
lational motion. The image is received on the rear of a
ground glass screen. Reflecting mirrors are used to fold the
optical path to make the instrument more compact.
Optical micrometric readers are incorporated on several
machine tools; these help in accurate positioning of the
workpiece. An optical dividing head is used for precise
angular and displacement measurements.
G
D L2 L1 Se F
(b)
8 SPECTROPHOTOMETERS
Figure 5. (a) Spectrophotometer; (b) schematic layout of mono-
Spectrophotometers are used for a variety of applications in chromator comprising: S, source; G, holographic grating; F,
industry, including routine batch composition analyses and removable filters; Se , exit slit; L1 , L2 , lens; and D, detector. (Copy-
pollution control. right Jobin Yvon.)
The principal component of the instrument is the mono-
chromator, an optical source, which provides a selectable its composition, each substance has its own characteristic
continuous range of wavelengths in the specified spec- IR absorption band(s), which serves as a fingerprint for its
tral region. identification and quantitative evaluation.
A commercial model of the instrument together with the Optical pyrometers are used for the temperature measure-
schematic layout of the optical system of the monochroma- ment of industrial furnaces wherein the flame luminosity
tor used is given in Figure 5(a) and (b). The specifications, is compared with that of an incandescent filament under
as given by the manufacturer, are wavelength range 0.35 varying electric power input conditions until complete pho-
to 0.80 µm; source, a tungsten lamp; dispersing element, a tometrical matching is obtained.
holographic concave grating with 1200 grooves/mm; and The saccharimeter is used to determine sucrose concen-
wavelength separation accuracy better than 0.02 µm. The tration in the sugar industry: in these the property of optical
detector is a silicon photodiode. rotation is used.
A colorimeter basically employs optical filters instead of Photometers are used for determining the light output of
the dispersing element. various sources and illumination measurements.
A large number of other instruments based on optical Moiré fringe methods use two sets of ruled lines overlaid
principles are used for quality control and other purposes at a slight angle to form fringes that move transversely to
in industry. A selection of examples, which have not been the relative motion of the grating and its index. Movement
dealt with already, includes the infrared-based instruments of a fringe at a fixed location is counted with a detector
such as the IR (infrared) spectrophotometer, IR gas ana- driving a high-speed digital electronic counter. They are
lyzers, and noncontact-type online IR moisture analyzer, used extensively by the machine-tool industry for precise
which are based upon the principle that depending upon determination and control of linear and radial movements,
Optical Measuring Instruments 607
and in many aspects of gauge-room inspection. Lastly, in
these examples, liquid and gas chromatographs are used
for rapid estimation of sample purity and for conducting
chemical analyses.
9 OPTO-MEDICAL INSTRUMENTS
For diagnostic purposes, observation of different places
inside the human body is made by the endoscopic class Figure 6. Hemoglobinometer. (Copyright American Optical.)
of optical instruments. The cystoscope, laparoscope, arthro-
scope, bronchoscope, gastroscope, colonoscope, auriscope Other opto-medical instruments not covered above are
are some of the instruments belonging to this family. These the ear microscope, slit lamp, trial sets and refractors,
vary mainly in their length, diameter, and other configura- optometer, keratoscope, projection perimeter, Nagel anoma-
tion parameters, to suit the particular application. loscope, photocoagulators, focimeter, noncontact tonome-
An endoscope is essentially a tele-microscopic system ter, IR body temperature ear probe, and colony counters.
with a large number of intermediate image transfer relay
lenses or coherent optical fiber bundles. Usually a low-
wattage miniature lamp, fitted at the viewing side, provides
10 LASER-BASED INSTRUMENTS AND
illumination. Use of a higher-wattage light source, pho-
tography, and electro-optical sensor arrays is feasible with SYSTEMS
optical fiber–based endoscopes.
Ophthalmic instruments are designed to examine and The highly coherent, high-power, well-collimated beams
measure various parameters of the human eye. Retinal obtainable from lasers have rendered them extremely useful
observation is done by an ophthalmoscope. It has a minia- in different sophisticated instruments. Both accuracy and
ture low-wattage light source, which is focused by a con- range of many conventional systems are enhanced using
denser onto a tiny mirror and the light reflected from the the laser radiation source. Shop floor, as well as field,
mirror illuminates the retina. During observation, any of the alignment problems from micrometers to kilometers can
number of small lenses of different dioptric power, mounted easily be accomplished precisely with a laser beam.
on a rotatable disc, can be brought into the return light As an example, the schematic diagram of an early mil-
path. A lens introduced in the correct focus position gives itary tank, laser-based, range-finder system is shown in
the refraction error of the eye. A retinoscope measures the Figure 7. The laser transmitter–receiver system and optical
retinal curvature and is based on the Foucault knife-edge sight are housed in the tank. The line of sight in elevation
test principle. is aligned to the axis of the gun by means of a precision
Retinal photography is taken by a fundus camera. It parallel linkage and in azimuth by coincident turret mount-
focuses the filament of a lamp onto the pupil, while the eye, ing. Bore sighting is achieved using controls on the unit
to be examined, looks at a small steel ball placed suitably with either a muzzle boresight or muzzle reference system.
in the light path. The same objective, used for illumination, Gunlaying is achieved through a ballistic graticule.
focuses the image of the retina onto a photographic plate. A laser speckle interferometer is used to measure small
After initial adjustment is made, a photograph or photo displacement, vibration, deformation, turbulence, contour
array image is taken under electronic flash illumination. generation of three-dimensional objects, and for thermal
A cold light source, which is basically a noncoherent fiber mapping; it is analogous to holographic interferometry. It
bundle transmitting only visible light (IR content is filtered is based on the principle that scattered laser light from a
out by thin film device) is often used in retinal surgery. grainy surface (grain size larger than the wavelength of
An ophthalmometer is used to measure corneal curvature; light employed) gives rise to an interference pattern (called
the size of the reflected image from the cornea is measured speckle). Overlapping of two speckle patterns produces
by a variable image-doubling technique. Moiré fringes.
An Hbmeter (Figure 6) determines the hemoglobin con- Using a laser source and flexible optical fiber as a
centration in blood. The instrument compares the absorption probe, it has been found possible to measure vibrations
of light by hemoglobin in a carefully defined depth of inside a closed chamber or of other places where nor-
hemolyzed blood with a standard glass wedge of similar mal optical techniques may not be suitable. The fiber
absorption characteristics. bundle carries the object beam forming one arm of a
608 Measuring System Design
Transmitting axis
Laser Sighting
Photo Laser transmitting receiver
Driver detector cavity optics optics
Sighting
Receiver
axis
axis
Power supply
unit
Photo Optical
Driver Amplifier multiplier beamsplitter
Computer fire Aiming mark
control system projection system
Electronic
unit (ranging) Commander's display
Gunner's display
Figure 7. Schematic diagram of tank laser range-finder system. (Copyright Barr & Stroud.)
3
A 2′
G2 2
f4 m7
f3 m4
m3
m5
G1 m2
f2 m1
1
f1
m6
1′
Figure 8. Laser Raman spectrophotometer: A, optical diagram; B, basic unit; R, detection system; 1, entrance beam path; 1 , alternate
entrance beam path when m6 is present; f1 , f2 , f3, and f4 , stepping-motor controlled slits; m6 , kinematically mounted plane mirror, which
permits use of 1 as entrance path; 2 , alternate exit beam path when m7 is present; 3, grating holders and shaft; G1 , G2 , holographic
gratings; m1 , m2 , m3 , m4 , m5 , plane mirrors, and concave mirror for imaging exit slit of first monochromator f2 on entrance slit f3 of
second monochromator; and m7 , kinematically mounted plane mirror to permit use of 2 as exit path. (Copyright Jobin Yvon.)
Twyman–Green-type interferometric setup. The laser inter- A laser Raman spectrophotometer is based on the prin-
ferometer has also been used in the measurement of refrac- ciple that a change in radiation frequency occurs due to
tive index variation, plasma density, and fluid velocity with molecular scattering. This has become a very powerful tool
the laser velocimeter. for studying vibrational and rotational energy levels of a
Optical Measuring Instruments 609
test sample. A commercial model of a laser Raman spec- and Detectors, Volume 2; Article 93, Testing Optical and
trophotometer, together with its optical diagram, is shown in Other Radiation Systems, Volume 2.
Figure 8. The instrument employs two concave, aberration-
corrected, holographic gratings with 2000 grooves/mm and
offers a resolution better than 0.5 cm−1 at 0.5145 µm. FURTHER READING
Ahmad, A. (1996) Handbook of Optomechanical Engineering,
CRC Press, Boca Raton, FL.
RELATED ARTICLES
Hobbs, P.C.D. (2000) Building Electro-Optical Systems: Making
It All Work, Wiley-Interscience, New York.
Article 89, Optical Materials, Volume 2; Article 90, Smith, W.J. (2000) Modern Optical Engineering, McGraw-Hill
Optical Elements, Volume 2; Article 91, Light Sources Professional, New York.
This article is based in part on a chapter by J. Prasad and G. Mitra originally published in Handbook of Measurement Science Vol 2,
P.H. Sydenham, Wiley, 1983.
93: Testing Optical and Other Radiation Systems
Alan J. Cormier
Sira Technology Ltd, Chislehurst, UK
focusing to an image point the size of which is limited only
1 Basic Parameters of Optical Instrumentation 610 by diffraction effects. When this spot size is achieved, the
2 Principles Used in Testing of Optical system is said to be diffraction limited. Real optical sys-
Instrumentation 610 tems generally have aberrations that degrade the wavefront
3 Equipment Used in Testing of Optical leading to larger spot sizes and lowered resolution.
Instrumentation 611
4 Limitations to Testing of Optical Afocal systems: For an afocal system (not image point
Instrumentation 613 forming), a plane output wavefront is implied. In the case
of visual systems, however, a departure from this condition
5 List of Global Optical Calibration Facilities 614
may be used to correct refractive error for the users.
References 615
Further Reading 615
1.2 Overall image quality
1 BASIC PARAMETERS OF OPTICAL In addition to wavefront quality, this depends upon factors
such as system transmission losses, chromatic aberration,
INSTRUMENTATION
image distortion, and the effects of stray light. Wavefront
vignetting due to internal aperture sizes and positions will
In general, optical instrumentation may be classified as
also affect final image quality (by modifying the numeri-
either imaging (e.g. camera lens) or afocal (e.g. telescope),
cal aperture) as will the method of detecting the image. In
where the term ‘optical’ strictly implies operation within the
visual systems, the eye may limit image quality; with con-
visible band (∼400 nm to ∼700 nm) although the ultraviolet
ventional camera systems, film grain size and the processing
and infrared spectral bands are also included here.
method may be the limiting factor. For digital cameras
There are a large number of basic design parameters
using CCDs (Charge Coupled Devices), resolution is lim-
associated with an optical instrument, such as focal length,
ited by the pixel size.
numerical aperture, magnification, and distortion; how-
ever, here we consider more specifically wavefront and
image quality.
2 PRINCIPLES USED IN TESTING OF
OPTICAL INSTRUMENTATION
1.1 Wavefront quality
2.1 Interference
Imaging systems: Light from an object point, emerg-
ing from the exit pupil of an imaging optical system, will Wavefront amplitude and phase: Optical radiation prop-
ideally be in the form of a spherical wavefront capable of agates in the form of a wave. A converging wavefront must
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Testing Optical and Other Radiation Systems 611
Plane wavefront Wavefront aberration assess image quality using an OTF Optical Transfer Func-
tion bench. An associated function, the ESF, can also be
used to determine OTF. This function is derived by record-
ing the energy passed by an edge boundary moved across
the image plane (Williams, 1999).
Center of reference
sphere 2.3 Optical transfer function
Aberrated
Test lens wavefront We can consider image quality measurement in terms of
Reference both ‘spatial’ (e.g. millimeters) and ‘spatial frequency’
sphere
(e.g. cycles per mm) units. These two domains are related
Figure 1. Wavefront aberration related to the reference sphere. through a Fourier Transform. By performing a Fourier
(Source: Reproduced by permission of Sira and IoP Publishing.) Transform of the (spatial) image, LSF, we obtain the
OTF having both amplitude modulation (MTF or Modula-
be perfectly spherical if the minimum focal spot size is to tion Transfer Function) and phase (PTF or Phase Transfer
be achieved. By comparing a real wavefront with a perfect Function) terms. MTF is of particular significance, giving
reference wavefront, we can specify the wavefront error, information on image contrast (Williams, 1999).
Figure 1.
3 EQUIPMENT USED IN TESTING OF
Two-beam interference: If both reference and test wave-
fronts are ‘coherent’, then interference fringes will occur. OPTICAL INSTRUMENTATION
Maximum fringe intensity occurs where the local phase dif-
ference in radians is given by, 3.1 Radiation test sources
φ = n2π radians(n = 0, 1, 2 . . .) (1) Incandescent sources: One of the most useful is the
quartz tungsten halogen lamp where a tungsten filament
is enclosed in a quartz envelope (see also Article 91,
Light Sources and Detectors, Volume 2). The halogen
2.2 Diffraction gas and quartz envelope allow the lamp to be run at high
temperatures, thus shifting the continuous spectral curve to
The diffraction effect: When a beam of light touches an shorter wavelengths (down to ∼350 nm).
aperture boundary, the beam is spread slightly (diffracted).
For a circular aperture, this spreading causes the beam Gas discharge sources: At low pressures, gases such as
diameter to increase. The amount of spreading is related Na and Ne, and Hg vapor, will radiate just at fixed resonant
inversely to the wavelength of light used. wavelengths. Such spectral line lamps are useful for wave-
length calibration purposes. As pressure is increased, so will
Airy disc and point spread function (PSF): In the case light intensity; however, the lines will gradually broaden
in which diffraction effects were absent, a perfect spherical and be superimposed on a spectral continuum. Mercury
wavefront could produce an infinitely small focal spot. lamps are widely used, with the 546.1 nm spectral line being
Diffraction effects, however, limit the minimum possible a common optical standard.
spot diameter to the Airy disc (a fact made use of in Knife-
edge and Star Tests). This diameter increases when optical Arc lamps: Two closely spaced electrodes located
aberrations are present. The distribution of intensity in the within a quartz envelope filled typically with high-pressure
image plane is represented by the PSF. xenon gas or mercury vapor. Very bright concentrated
sources are produced with resonant lines superimposed on
Line spread function (LSF ) and edge spread function (ESF ): a thermal radiation continuum. Compared with tungsten
By scanning a narrow slit across the PSF we obtain the halogen sources they produce more ultraviolet light and
LSF. For practical purposes, the LSF is usually obtained much higher brightness in the visible band.
by scanning a slit aperture across the image of a slit source
(or alternatively by imaging the slit source onto a 2D Lasers and Light emitting diodes (LEDs): A laser
array detector). This function is useful when we wish to source has the special property of ‘coherence’, which is
612 Measuring System Design
put to good use in the interferometer. Laser sources also −6 −5
have intrinsically very high brightness and narrow band- 1
widths. For optical testing purposes, the He–Ne (633 nm) 2
laser is frequently used; others are available such as the 2
CO2 laser for testing in the infrared (10.6 µm). The LED is
similar to the laser except that it does not have an optical −4 −3 3
3 2 1
cavity; consequently, the spectral bandwidth is broader and 2
3 −2 −1 3 4
the coherence lower. 2
3
4
4
5
4 5
6
6
5 −4 5
4 6 1
Blackbody infrared source: An ideal blackbody source 6
completely absorbs all radiant energy striking it and, there- SINE PATTERNS
fore, appears perfectly black at all wavelengths. A perfect 5 −6
blackbody source emits radiation with an emissivity of
1 (see Burnay, Williams and Jones, 1988). Real sources, 1
using controlled high temperature cavities, can emit wave- 6
lengths between 1 µm and >25 µm. The shorter wave-
lengths require higher temperatures. When used with a Figure 2. USAF 1951 target. Public domain. (Source: Copyright
pyroelectric detector, the radiation source must be used with Sine Patterns LLC.)
an optical chopper.
Spectral filter
turret
3.2 Test targets S1
L4 Integrating sphere
USAF 1951 target: A significant number of targets
formed from opaque line patterns deposited on substrates L1 L2
Entrance port
Lamp
such as fused silica are available. These are used to assess L3 Exit port
Variable Fixed ND
resolution limit directly. One of the most common is the ND filter filters
USAF 1951 target, shown in Figure 2. Use of this and other Target pattern
similar targets is considered in the Military Standardization Turret wheel
Handbook (MIL-HDBK-141, 1962).
Slit and bar targets: Formed either by deposition tech-
niques, or machined into a solid opaque substrate. In the Figure 3. The integrating sphere as part of Sira’s helmet mounted
display test facility. (Source: Reproduced by permission of Sira
latter case, substrate transmission is not an issue. These tar-
and IoP Publishing.)
gets are frequently used to produce an LSF from which the
system OTF can be calculated. Bar targets are generally
multiple, wide slits with equal mark-space ratio. 3.4 Knife-edge and star tests
Knife-edge or Foucault test: As shown in Figure 4, a
3.3 Integrating sphere sharp edge is positioned close to a beam focus, produced
by a lens or mirror, with the eye immediately behind and
Used to produce an output beam that is insensitive to any focused at the exit pupil position. By moving the edge
spatial, angular, or polarization changes occurring in the across the beam, sections of the exit pupil are seen to go
input beam. They do this by collecting input light through dark in correspondence with the blocked rays. From the
one aperture, which, after multiple reflections and scattering pupil patterns seen, experienced observers can deduce the
at a high reflectivity surface, eventually emerges through a aberrations present (Malacara, 1978).
second aperture at right angles to the first (see Figure 3).
Integrating spheres are used to overcome problems such Star test: For this test, the optical system is used with
as spatial and angular response variations exhibited by a point object (similar to a star) to produce a geometrical
many detectors and the effects of input beam position and image that is smaller than the Airy disc. By using an eye-
profile changes. piece to examine the image, the experienced observer can
Testing Optical and Other Radiation Systems 613
Surface under test Axis of rotation
Plane mirror
Pinhole Test unit
Focal slide
Reflecting
prism
Mirror
Knife-edge collimator
Image
Object analyzer
Figure 4. Knife-edge test configuration. (Source: Reproduced by generator Rotary
permission of Sira and IoP Publishing.) Rotary
X, Y, Z stage
stage
stage
Field slide
Slideway
again learn a great deal about the aberrations present that
affect the shape of the diffraction pattern (Malacara, 1978). Figure 6. Typical OTF test bench configuration. (Source:
Reproduced by permission of Sira and IoP Publishing.)
3.5 Interferometer
3.6 Camera/OTF bench
A number of different types of interferometer exist; how-
The camera bench is an important piece of basic test
ever, in each case, coherent beams derived from the same
equipment and is used in the measurement of fundamental
source are made to interfere in the optical domain.
system parameters, such as focal length, magnification, and
Perhaps the most common type is the He–Ne Fizeau.
distortion. When used with a single element detector behind
The laser beam passes into an air gap formed between
a slit aperture and a (scanned) slit source, an OTF bench
two surfaces one of which is a plane (flat) reference
is formed, Figure 6. Because the detector axis direction
surface. The other surface is the plane test surface. An
usually remains fixed, test limitations can occur at larger
interference fringe pattern is produced, which maps the
field angles in cases in which the image slit cannot be
errors on the test surface. Spherical test surfaces may
placed close to the detector surface. This is due to the
also be examined when the interferometer is fitted with
signal falling outside the detector boundary. OTF benches
a special lens called a transmission sphere (see Figure 5).
often use an area array detector in place of the originally
In this case, the last concave surface acts as the reference
used image slit, which overcomes this problem (Williams,
surface.
1999).
Because two-beam interference effects repeat for every
2π phase difference, the interferogram will be ambigu-
ous outside this range. Phase shifting of the reference
4 LIMITATIONS TO TESTING OF
beam is commonly used to overcome this problem (Zygo
Corp, 1998). OPTICAL INSTRUMENTATION
Limitations are broadly related to
Fizeau interferometer
1. accuracy of the measurement equipment,
2. suitability of the measurement configuration, and
3. environmental effects.
Laser
In some cases, the limitation may be absolute, in other
cases the limitation may be reduced by careful recalibration
of measurement equipment or changing the measurement
Detector
configuration. Any tests made must take into account the
Part under test instrument’s intended mode of use. This is particularly
Transmission sphere true in the case of visual instrumentation where the users
(TS)
perception of image quality is a key factor (Haig and
Figure 5. Fizeau interferometer test configuration using a trans- Burton, 1987).
mission sphere. (Source: Reproduced by permission of Sira and In the case of an imaging system, such as a film camera, a
IoP Publishing.) common test would be to photograph a resolution test chart;
614 Measuring System Design
however, the results are generally subjective. It is there- In the above text, we have seen that thermal and stray
fore usually necessary to test the lens objectively. When light effects can act to limit the accuracy of test mea-
performing OTF tests, measured image quality is limited surements. Other environmental effects, such as vibra-
through aberrations introduced by the optical components tion, shock, and humidity, can similarly limit performance.
used to provide the input beam (although it is usually easy Vibration, for example, can affect significantly the accu-
to make them insignificant). Wherever possible, additional racy of interferometric measurements. This is particularly
components should not be used to relay the test image to the case when using phase-shifting hardware.
the detector. Where this is necessary (for example, when A very wide range of optical instrumentation exists, from
testing afocal systems), the relay optics should be diffrac- complete imaging systems to instruments that are them-
tion limited. selves used in the testing of other optical instruments. As
The use of interferometers, in conjunction with phase- a consequence, a wide range of test requirements will also
shifting techniques, has the potential for high measure- arise. By understanding the physical limitations involved
ment accuracy. However, poor agreement between different and adopting environmental control procedures and regular
instruments and operators has been a source of concern equipment recalibration, an acceptable result can usually
(Briers, 1999). Special techniques are also required when be achieved.
using high aperture (e.g. NA 0.5) transmission spheres (Sel-
berg, 1990). 5 LIST OF GLOBAL OPTICAL
For visual system tests, it is usual to perform photometric CALIBRATION FACILITIES
measurements where an observer’s response is standard-
ized through a combination of detector and matching filters. The foremost optical metrology laboratory in the United
While the overall detector spectral response achieved may Kingdom is the National Physical Laboratory (NPL):
appear close to the ideal form, small localized variations can
result in significant measurement errors (Ohno and Thomp- National Physical Laboratory (NPL)
son, 1998). These errors may be increased through effects Queens Road
such as detector ageing and filter degradation, uncorrected Teddington, Middlesex TW 11 OLW
thermal dependence, and stray light. Periodic recalibration Phone: (44 181) 977 3222
of such detection equipment is crucial if the highest mea- Fax: (44 181) 943 6458
surement accuracy is to be achieved. E-mail: [email protected]
Testing of thermal (or heat) imaging systems operat- Other key laboratories include
ing in the mid- and longwave infrared presents additional
problems. In particular, the optical elements and support France: Bureau National de Metrologie
structures are themselves a source of radiation that can 1, rue Gaston Boissier
lead to reduced image contrast. A common practice is 75 015 Paris – France
to chop the infrared source, which can then be separated Phone: (33 1) 56 56 88 63
electronically from the DC background – see Article 184, E-mail: [email protected]
Photon Counting, Volume 3. This technique is effective Germany: Physikalisch-Technische Bundesanstalt (PTB)
when performing OTF measurements; however, limitations Bundesallee 100
will occur when accurate radiometric measurements are Postfach 3345
required, unless the temperature and emissivity of the chop- D-3300 Braunschweig
per blades are known. The limitation may be reduced Phone: (49 531) 592–3005
by using a chopper with mirror blades, which allow a Fax: (49 531) 592–4006
calibrated blackbody to be imaged during the obscura-
tion period. Italy: Istituto Elettrotecnico Nazionale
Precise knowledge of target and background plate tem- Dept. of Photometry
Strada delle cacce 91
perature and emissivity is also required when undertaking
10135 TORINO, Italy
other forms of differential temperature imagery (e.g. Min-
Phone: (39 11) 348 8933
imum Detectable Temperature Difference, or MDTD, and
Fax: (39 11) 650 7611
Minimum Resolvable Temperature Difference or MRTD)
as thermal imagers respond to radiance and not temperature United States of America:
differences. Further consideration is given to this aspect in National Institute of Standards and Technology (NIST)
(Burnay, Williams and Jones, 1988)). Optical Technology Division
Testing Optical and Other Radiation Systems 615
NIST, 100 Bureau Drive, Stop 3460, Gaithersburg, Haig, N.D. and Burton, G.J. (1987) Effects of Wavefront Aber-
MD 20899-3460 ration on Visual Instrument Performance and a Consequential
Test Technique. Applied Optics, 26(3), 492–500.
E-mail: [email protected]
Phone: (301) 975-NIST (6478) or TTY (301) 975–8295 Malacara, D (ed.) (1978) Optical Shop Testing, John Wiley &
Sons, New York.
Canada: National Research Council Canada MIL-HDBK-141, ed. (1962) Military Standardization Hand-
Institute for National Measurement Standards book – Optical Design, Department of Defense, Defense Supply
Montreal Road, M-36 Agency, Washington, DC.
Ottawa, Ontario K1A 0R6 Ohno, Y. and Thompson, A.E. (1998), Photometry – The CIE
Phone: (613) 993–9326 V(λ) Function and What Can be Learned from Photometry,
Fax: (613) 952–5113 International Symposium on Measurements of Optical Radiation
Hazards, 1–3 September, NIST, Gaithersburg, MD.
E-mail: [email protected]
Selberg, L.A. (1990) Interferometry Accuracy and Precision, SPIE
Other countries: Optical Fabrication and Testing, Vol. 1400, SPIE (pp. 24–32).
A global list of metrology laboratories may be found Williams, T.L. (1999) The Optical Transfer Function of Imaging
on the US National Institute of Standards and Metrology Systems, IOP Publishing.
(NIST) Web site at http://www.nist.gov/oiaa/nat pg2.htm Zygo Corp (1998) Application Note – Fringe Analysis versus
Phase Measuring Interferometry, Zygo Corporation, Web site
http://www.zygo.com.
REFERENCES
Briers, J.D. (1999) Interferometric Optical Testing: Considera- FURTHER READING
tions for the Proposed New International Standard, SPIE: Opti-
cal Fabrication and Testing, Vol. 3739, SPIE (pp. 496–507). Labsphere (2002) A Guide to Integrating Sphere Radiometry and
Burnay, S.G., Williams, T.L. and Jones, C.H. (1988) Applications Photometry, Labsphere Inc, Web site http://www.labsphere.
of Thermal Imaging, IOP Publishing. com.
94: Human Factors Engineering
Nicholas I. Beagley
DSTO, Canberra, ACT, Australia
Engineering (HFE) have increased in use – for more on
1 Ergonomics and Human Factors Engineering 619 HFE see Article 97, Design Methodology, Volume 2.
2 Task Capture and Design 620 Despite hot debate about the appropriate use of terminol-
ogy, the terms Ergonomics and Human Factors Engineering
3 Characterizing the User Population 621
are frequently used interchangeably. There are numerous
4 Establishing Context 621 definitions that seek to encapsulate all aspects of the disci-
5 Training Specification 622 pline, each serving a purpose in setting out the scope for
6 Modeling and Prediction 622 the discipline. In trying to cover this range, they tend to be
7 Prototyping and Testing 622 relatively long and wordy. An alternative method of com-
References 623 municating the scope and purpose of this area is to convey
the underlying philosophy of the discipline. This is put suc-
cinctly in the statement that Ergonomics is about ‘equipping
the man, not manning the equipment’.
1 ERGONOMICS AND HUMAN FACTORS This conveys the idea that in order to get the most
ENGINEERING from the total system, designs should be tailored to the
capabilities of the human rather than relying on people to
Translated literally from the Greek words ergon (work) and adapt to unsuited designs.
nomos (science/laws), Ergonomics means the ‘Science of The practice of Ergonomics/HFE draws on a number of
work’. broad activities that combine to improve our understanding
The study of work has been traced back to a publication of human/system performance and ensure that this knowl-
in 1713 by Bernardino Ramazinni on work related injuries. edge is available to positively influence system design in
Before World War II, the focus on workplace studies was the goal of achieving more efficient and safer systems.
on efficiency and productivity through time and motion Ergonomics is underpinned by a knowledge base infor-
studies by Taylor (1911) and Gilbreth (1911). Concentration med by research. Fundamental research in areas such as
on the design of military systems prompted a shift in focus vision, memory, and motor skills provides data that can
to human performance, behavior, and safety. often be generalized across a wide range of contexts.
The term Ergonomics was proposed by Prof. Hugh Mur- At a higher level of human function where basic func-
rell in 1949 and first applied to this field through the naming tions, such as the senses and cognition, combine, for
of the UK’s human research focused Ergonomics society example, human behavior, the influence of functional com-
that year. It is this broader description of the field for binations and context increases.
which the term Ergonomics has been adopted. As the study The consequence is a reduction in our ability to apply the
of the human in the context of their working environment findings of generic research for the purpose of prediction.
expanded so other terms such as the more general Human Given the limitations on prediction, final confirma-
Factors (HF) and the more design specific Human Factors tion of design suitability is normally based on testing
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
620 Measuring System Design
within a close representation of the system’s intended Given the central role of the activity of the human to
context. the practice of Human Factors, it is not surprising that the
To effectively influence design, it is important to package documentation and analysis of these activities lies at the
the accumulation of research knowledge and experience in heart of many human factors studies.
a way that is accessible to multidisciplinary human factors The process of task description is often one of the start-
and engineering design communities. ing points of any human factors analysis. The capture of a
As with all areas of science, the published research lit- baseline description of a person’s activities within the con-
erature is the main repository for knowledge. The level of text of a system’s function is an essential part of assessing
domain knowledge required to retrieve, review, and apply opportunities for, and the achievement of, progress through
this knowledge limits its widespread applicability. Counter- design and development. It is not possible to demonstrate
ing this problem, a large number of documents have been that a design change has improved a situation without estab-
developed to bring together principles of design, guides to lishing what went before.
best practice and agreed standards. Key documents in this Task description of an existing process can be a use-
area include Def Stan 00–25, UKMOD (2000); Mil Std
ful baselining diagnostic tool for revealing parts of an
1472F, USDOD (1999); and ISO 9241, ISO (1993).
activity that are inefficient or error prone. At this point,
In the absence of an all-encompassing deterministic pre-
there is a blurring between the terms task description and
dictive model of human/system performance, it is necessary
task analysis.
to study the human/system in context.
The widespread application of task analysis within the
A wide range of tools and techniques has emerged from
the human factors community over the years to support this fields of ergonomics and HFE became popular in the
activity. These tools include techniques for measurement, 1950s. It has since been modified and tailored to a variety
analysis, and for influencing the design cycle. Tools such of objectives.
as task analysis, that is, the systematic decomposition of Task analysis has a key role in process and product
an activity to support understanding and redesign, are design. This concept is akin to a ConOps – see Article 67,
a fundamental activity for most ergonomists seeking to Principle of Concept of Operations (ConOps), Volume 2.
understand the goals and associated activities carried out It can be used to aid the allocation of function, product
in pursuit of a system objective. testing, and for the evaluation of workload. Accidents
Human/system performance measurements, using a range and assessments of the risk of accidents can benefit from
of objective and subjective measurement tools and tech- the use of task analysis. A more complete description
niques within designs sensitive to the human/system vari- of task analysis techniques is available in Kirwan and
able, are a vital part of design testing. Ainsworth (1992).
At the top-level process, methods such as the Human Fac- Task analysis methods are often distinguished by their
tors Integration (HFI) Program (UKMOD, 2000; UKMOD, method of subtask separation and organization. A widely
2002) – see other documents on http://www.ams.mod.uk/ applied approach to task analysis is called hierarchical task
ams/content/docs/hfiweb/hfihome.htm, and the US Man- analysis (HTA). It starts with a high-level objective oriented
power Integration (MANPRINT) Program (see documenta- description of the activity and proceeds to decompose the
tion on http://www.manprint.army.mil/manprint/guides/mp- activity into goal focused tasks and subtasks.
ref-guidebook.asp), are followed to ensure consistency of The approach to HTA descriptions can be freely adapted
analysis and to provide an audit trail of the human science to capture specific information relevant to the objective
support to the design process. of the analysis such as information flows, or associated
technologies. Frequently, the sequence and logic underlying
the conduct of a task sequence is represented within the
2 TASK CAPTURE AND DESIGN
HTA box diagram.
Ergonomics at its most fundamental remains the science A baseline task description of an existing process, when
of work. It is not the study of the human in isolation, nor analyzed systematically, can reveal opportunities to ratio-
is it the study of the world around the person, although it nalize the number, or order, of steps to gain time savings
draws on each of these disciplines. At its core, ergonomics or reduce errors.
is concerned with the activities of people, the design of the The form and content of the task description should be
systems that define these activities, and an understanding based on the objectives of the method. If the issue of con-
of how the capabilities of the human and the demands of cern is the mental workload placed on the operator, the task
the environment impact on the way people interact with the description should capture and represent the information
world around them. flows over time.
Human Factors Engineering 621
This information, when applied to user walk throughs knowledge, and experience considered to be beyond the
using representative scenarios designed to test the limits reasonable training expectations of the current population.
of the system, can provide a means of assessing operator If the requirement for skills and training is noted at an
loading and the associated potential for failure. early stage, the parallel processes of training and recruit-
An alternative form of task description and analysis is the ment can be put in place over the timescale for system
operational sequence diagram (OSD). A representation of procurement.
the task sequences and the logical links between tasks can It should, however, be recognized that in an increasingly
be a useful tool for examining the implications of changes technological world, the number of trades that will be
to a system on its performance. This approach to modeling competing for the pool of highly skilled individuals could
forms the basis of computer-based task-modeling tools also be expected to increase.
such MicroSaint and the Integrated Performance Modeling For the purposes of ergonomic design, the parameters of
Environment (IPME). the population, including anthropometric, that is, body size,
Occasionally, it can be useful to consider the linkages dimensions, should be captured.
between subcomponents of a system. An example might be The population ranges normally captured in databases
the analysis of communications between different people and tables and formulae include stature, weight, strength,
within a control room. A diagrammatic representation of the functional reach, leg length, sitting height, and hip breadth.
control room overlaid with lines depicting the distribution A wider list of anthropometric measures should be pro-
of communications amongst people within the control room vided when the design has a large human/equipment inte-
can be a useful tool for examining the location of work- gration component.
stations and the information necessary to meet the control System designs are normally required to safely and
room’s operational role. This method of system description comfortably accommodate potential users within the 5th
is called link analysis. to 95th percentile, that is, the percentage of the target
population that can fit, reach, and operate the system.
In addition, cognitive attributes of the target population
3 CHARACTERIZING THE USER have become increasingly relevant for systems designers.
This may be specified by selection tests applied to people
POPULATION
within the target group. Selection criteria may provide
an additional source of information regarding the group’s
If a design is to take account of the characteristics of its
normal spread of knowledge, experience, and skills as
intended user population, the key characteristics relevant
relevant to the design under consideration.
to a system’s emerging concept must be captured as a
The TAD is simply a method of formalizing and reporting
reference for the designers and decision makers.
the capture and consideration of user attributes. Small-
If assumptions made about the intended population, such
scale design programs may not justify the development of
as the ability to recruit and retain highly skilled engineers
a full TAD. However, the process of capturing and con-
as users, are unfounded, then the procuring organization
sidering these attributes at an appropriate level of detail
could be left with an expensive system that cannot be
has a place in all human system procurement and develop-
operated due to insufficient numbers of capable operators.
ment activities.
Within the UK’s military procurement process the tool
for characterizing the user is called the Target Audience
Description (TAD). 4 ESTABLISHING CONTEXT
The TAD is a document that is created early in the
design concept phase of a procurement program. The TAD The study of human centered systems is inextricably
has become a standard tool whose structure prompts the linked with the context in which they are placed. It is
gathering of information likely to be relevant to the pro- on this premise that the ecological approach to human
curement program. factors research is based. The ecological approach starts
The document starts with a general description of the from the position that there is a reciprocal coupling
current group who would be intended to use a system. between the person and their environment. It takes the
A reasonable assumption for the manning of a system is view that the environment is both an opportunity for
that the future users will have the same level of skills and human action (or affordance) and a constraint. For exam-
capabilities as the existing population. ple, a maze affords a path for progress while at the
There is a danger in specifying a highly technologically same time constraining the path’s direction. From this
sophisticated replacement for an existing low tech system perspective, it is clearly of particular importance to ana-
when the new system would require a range of skills, lyze it.
622 Measuring System Design
An analysis of the environment should seek to identify through performance prediction and modeling will have
those aspects of the environment that exert an influence over obvious savings in costs and timescales.
human behavior and contrast them against those behaviors Human modeling and performance prediction is already
that are specific to the individual. used as a tool to support the design and analysis of human
systems. There is a range of commercially available models
that can be used to predict human performance in a number
5 TRAINING SPECIFICATION of ways. Stochastic, that is, based on probability, task mod-
els such as MicroSaint and The Integrated Performance
As the systems we expect people to use become increas- Modelling Environment provide the researcher with the
ingly complex, there is a need to pay special attention to the ability to specify task sequences and their associated pro-
demands this places on providing an effective mechanism cedural logic and performance parameters. By defining a
for teaching the knowledge and skills necessary to support concept task in this form, it is possible to make predic-
these interactions. tions about the likely performance of the task in terms of
Training is a necessary overhead to effective system defined parameters such as completion time. The addition
performance. This training must be considered alongside of other parameters such as information flow or physical
the more traditional cost estimation parameters of purchase load can be used to reveal peaks in workload that could
and maintenance. Training programs should be designed have consequences for task performance.
to ensure that the necessary skills are transferred from the There are several commercially available body modeling
training environment to the operational environment. It is tools on the market including JACK , Safework , and
through this demonstration of transfer of training that the Sammie . These tools allow the user to specify the key
training program can be formally evaluated. anthropometric dimensions of virtual mannequins.
The normal approach to demonstrating a transfer of train- Most of these tools allow the user to import virtual Com-
ing is to compare the performance of a trained group against puter Aided Design (CAD) models of concept equipments
an untrained control group to demonstrate a statistically for evaluation within the virtual environment. To these are
significant difference between the performances of the two added the mannequin with the needed anthropometric data;
groups. Under some circumstances, it is possible for a train- the two then can be made to interact.
ing program to have a negative transfer. An example might These techniques have demonstrated benefits for the eval-
be the training in a flight simulation with a slightly differ- uation of issues of fit, reach, and vision. Some include
ent interface style to the real aircraft. Errors might arise basic modeling of the physiological stresses that might
because of the interface differences that have the effect of accompany varying posture and loads, although the true
surprising the operator when they encounter the different complexity for the human musculoskeletal system pre-
real system. vents these models from offering validated predictions of
Training systems cannot be solely assessed on their human biodynamics, that is, the study of dynamic forces
ability to support a transfer of appropriate skills to the real on the body.
world. While real flight might offer a better transfer of
training than the use of flight simulation, the cost of this 7 PROTOTYPING AND TESTING
approach can, in some circumstances, be prohibitive.
Training through simulation has the added advantages of The objective of HFE is often to draw conclusions about
graded, and total, control over the scenario and the ability the merits of concept systems. The difficulty associated
to automatically gather performance measures with which with this form of analysis is the fact that these systems
to set up trainee development. do not yet actually exist. Human Factors has developed a
wide range of tools and techniques focused on performance
measurements that support the comparative evaluation of
6 MODELING AND PREDICTION concept human/systems.
When a system component exists only as a concept,
The complexity of human interaction within the real world the researcher must be inventive in testing the projected
reduces our ability to predict outcomes. This leaves the capabilities. The normal approach to this is prototyping,
practice of Human Factors with no alternative other than to that is, the creation of a model or simulation of a product.
conduct user testing in representative environments. If the availability of a new technology is anticipated
The time and logistical costs of running a user experiment within a few years, there may be a requirement to examine
is compounded by the time load on the subjects. Any the potential issues that may arise in its usage or the
opportunity to achieve the design benefits of user testing implications for the normal approach to associated tasks.
Human Factors Engineering 623
If an early prototype of the technology is available, this REFERENCES
can form the basis of product testing to examine these
issues. In many instances, the early prototypes of new Gilbreth, F.B. (1911) Motion Study, Van Nostrand, Princeton, NJ.
technologies are extremely limited in their functionality or ISO (1993) Ergonomic Requirements for Office Work with Visual
are unduly bulky. In these cases, there is a danger that the Display Terminals (VDTs), European Standard 9241 (Parts 1 to
generic capability illustrated by the prototype technology 17), International Standards Organisation, Geneva.
will be missed by the user who becomes distracted by the Kirwan, B. and Ainsworth, L.K. (eds) (1992) A Guide to Task
limitations of the prototype. Analysis, Taylor and Francis, London.
In cases in which the prototype technology is too imma- Taylor, F.W. (1911) The Principles of Scientific Management,
ture for useful user testing, it may be possible to simulate Harper & Bros, New York (Reprinted 1967 by W.W. Norton
the performance of the system. An example of simulat- & Co.).
ing an immature technology is the use of a researcher to UKMOD (2000) Human Factors for Designers of Military Equip-
take the place of a proposed artificial intelligence system ment (Parts 1 to 14). Defence Standard 00–25, UK Ministry of
Defence.
intended to aid the user in a task. The subject of the eval-
uation is not made aware of the method used to provide UKMOD (2002) A Practical Guide to Human Factors Integration:
A Guide for Integrated Project Teams and Capability groups,
this functionality thus allowing the researchers to examine Version 0.9.3, 19 July 2000, UK Ministry of Defence.
issues such as modes of interaction and user confidence in
USDOD (1999) Department of Defence Design Criteria Standard
the system. This approach allows conclusions to be drawn Human Engineering, Military Standard 1472F, US Department
that will contribute to design decisions that will have to be of Defense.
taken when the prototype technology reaches maturity.
95: Human–Machine Interface
Nicholas I. Beagley
DSTO, Canberra, ACT, Australia
accuracy of motor movements, such as, hand control,
1 Comparing Humans and Machines 624 through laboratory testing (Fitts, 1954).
2 The Interface 625 A wide body of research reveals many parameters of
human performance in the areas of speed, strength, and
3 The Visual Interface 625
endurance as well as sensitivity to stimulus and the capacity
4 The Auditory Interface 626 for mental storage, retrieval, and processing.
5 The Other Senses 626 In most cases, these parameters vary between individuals
6 Control Devices 626 across the population and with the variations in both the
References 627 task to be completed and the demands of the environment.
Despite the appeal of a reductionist model of human
performance, the ability to accurately predict human per-
formance within a conceptual system remains elusive.
1 COMPARING HUMANS AND The relationship between man and the engineering design
MACHINES of machines has been discussed in Sydenham (2003).
While the pursuit of this ideal continues, Human Fac-
This article continues the development of the topic of tors (HF) practitioners are left to apply a combination of
Human Factors begun in Article 94, Human Factors Engi- modeling and human testing within an iterative design
neering, Volume 2. There has been an uneasy alliance cycle to achieve the objectives of true human/system
between people and machines over the course of history. integration.
The use of simple tools is cited as one of the first steps in Fitts contributed further to the toolset of the HF practi-
the civilization of the human species. tioner with his list comparing the performance of humans
Subsequent developments of agriculture and weapons are with machines (Fitts, 1951).
the basis on which civilization in its current forms evolved The original Fitts list consists of two columns, comparing
(Diamond, 1999). the strengths of the person versus strengths of the machine.
With each technological advance, people have been able As technology continues its relentless advance, the ability
to achieve goals in increasingly efficient and effective ways. to establish clear distinctions between the capabilities of
This is not to say that technology is consistent in delivering people and machines is diminishing.
advantages. However, the underlying trend is a world in Opportunities exist to overcome human weaknesses such
which the traditional labors of the human race have been as physical and mental fatigue with automated solutions
reduced or replaced by the application of technology by capable of quickly and accurately completing repetitive
those able to afford to develop and acquire them. tasks.
Our understanding of human performance is based on Advances in computer technology have allowed
fundamentals revealed through human physiology, for machines to outperform people in areas originally
example, Paul Fitts established a law for the speed and considered to be the domain of the human. As a result, the
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Human–Machine Interface 625
original Fitts list serves little purpose beyond illustrating Many issues face the interface designer. The task is to
the rapid progress of technology. clearly and unambiguously present the person with the
That said, the underlying concept holds true. There necessary information without overloading them and to
remains value in considering the allocation of function make it as easy as possible for them to implement the
between the different components of a system, including actions they decide upon in a timely and error-free manner.
the person. An example of a modified Fitts list is shown Measuring systems are there to interface the use to the
in Table 1. state of selected physical variables; human-factors engineer-
ing is therefore part and parcel of measuring instrumenta-
tion design.
2 THE INTERFACE
People interface with the world around them by physical 3 THE VISUAL INTERFACE
actions directed by the nervous system based on sensory
inputs. Our ability to sense our posture in relation to the Human vision is the primary source of input for most human
earth’s gravitational force and coordinate our muscular tasks. The eye shares many similarities with the camera
response accordingly is the basis by which we are able focusing light with a lens on the retina.
to stand erect. The rapid feedback loop, from our sense of The retina is made up of approximately 100 000 low light
vision, balance, touch, hearing and so on, allows us to walk, sensitive rods and 60 000 color sensitive cones that provide
run, and jump in a dynamic environment. The evolution of daylight vision, which in combination offer a range of light
human capabilities is based on the demands of survival sensitivity from 1 billion to 1.
in the natural world. It is against the framework of these Photoreceptors are not evenly distributed on the retina.
capabilities that we must consider and tailor the interfaces In the region of the fovea, the high concentration of pho-
of the modern world. toreceptors provides a resolving power of 1 min of arc as
Many of these fundamental parameters of human capa- compared to several minutes of arc in the periphery of the
bility have been well established. retina. As a consequence of this distribution, visual atten-
Wavelengths of energy visible to the human eye are tion is based on head and eye movement focusing the area
defined (400 to 700 nm). of interest on the fovea with the peripheral vision providing
Frequency and intensity of sound waves comfortably an additional cue to visual attention through the detection
heard by the human ear have been established (20 to of movement. Movement of the eye is finely controlled and
20 000 Hz with a sensitivity ratio of 1 : 1 billion). can range from slow to extremely fast (over 400 degrees
Resolution of tactile sensation that can be felt over per second).
different areas of the human body has been mapped out It is important to recognize the capabilities and limita-
(2 to 3 mm at the finger). tions of the human visual system when designing interfaces
Across this range of sensory capabilities some modes to take advantage of this primary sense.
are better suited to communicating certain information A key factor to recognize in the design of visual inter-
than others, depending on the context of the task and the faces is the singular attention of the human. People can
environment. look quickly between different locations but can only focus
Variation in sensory capabilities across a user popula- on one point at any one time.
tion places further limits on the method of information Efficient design of displays requires a consideration of
presentation. the task they are supporting and the load the design places
on the operator. If there are a number of display elements
Table 1. Modified Fitts list. that need to be monitored at the same time, they should be
colocated in a logical fashion to reduce the time taken to
Human Machine
search between locations.
Inductive reasoning Deductive reasoning A good example is the design of the head up display for
Pattern detection Parallel processing fast jet aircraft cockpits. The display places the key infor-
Improvisation Processing capacity
Task shedding when Accuracy of calculation
mation required by the pilot to operate the aircraft and its
overloaded supporting systems, as an overlay on his view of the world.
Relational memory retrieval Speed of response This allows the pilot to spend more time ‘head out’, observ-
Integration of multiple senses Absence of fatigue or boredom ing the world through which they are flying thus raising the
Adaptive learning and skill Supra-human sensory opportunities for developing situational awareness.
capabilities
Color should be used with care. Although color can help
Mobility Data storage and search
highlight important features of a display, or indicate the
626 Measuring System Design
Table 2. Color associations. In these circumstances, we tend to use visual cues such as
Color Conventions Other associations the movement of the other person’s mouth as well as an
anticipation of what might be said, based on the subject of
Red Danger Stop
the discussion, to fill in gaps in the information masked by
Emergency Fire
Failure Hot the noise.
Blue Mandatory Water
Hearing can be a useful alternative to vision for the
Sky communication of information, particularly when we do
Off not wish to be distracted from a visual task. For example,
Cold it is possible to conduct a simple visual search task while
Green Safe Safe passage, maintaining a conversation.
Exit It can also be a useful medium when the display of visual
Continue information would be difficult for example, at night. Audi-
Yellow and black Caution Ambient tory warnings are used extensively to attract our attention
Delay
away from a visual task. The characteristics of the sound
Warm
can be tailored to convey the importance of the message
thus increasing the amount of information a relatively sim-
grouping of common elements, its overuse can have the ple warning sound conveys.
opposite effect, distracting the user instead.
Inappropriate color combination can reduce the contrast
between different elements of a display disguising infor- 5 THE OTHER SENSES
mation and reducing the effectiveness. It should be noted
that people link colors with particular attributes or task While vision and hearing are the main sources of infor-
specific associations. In addition, it should be noted that mation to the human user of a system, we are sensitive to
individual variations amongst the population, for example, many other attributes of our environment.
color-blindness, and changes to the ambient light levels can Walking along the street draws on visual input to judge
dramatically alter the perception of a color scene. where to place our foot when taking the next step. Audi-
Table 2 includes a list of common associations for colors tory information warns us if there is traffic approaching
and their combinations. from behind. Our vestibular system continually updates our
orientation without which balance would be impossible.
Proprioceptors in our feet tell us when our feet are in con-
tact with the ground. Our linked senses of smell and taste
4 THE AUDITORY INTERFACE might play a part in directing our destination in search of
somewhere to have lunch.
Hearing is the next most important source of information The ability of the human to filter and integrate the vast
input after vision for most tasks. Hearing and speech is a amounts of information continually received from our broad
highly evolved method of communication. range of sensory capabilities and our ability to adapt our
When we talk to another person, we are using an interface processing of this information in a changing environment
in which participants sense and interpret the speech and sets us apart from machines.
expressions of the other person, forming a response and
then responding through the formation of speech or some
other action. 6 CONTROL DEVICES
Humans learn to speak through a process of training,
trial and error, when growing up. It is a highly effective The human ability to sense our environment is just one
and adaptable form of communication containing subtle part of the story when considering the human, or the Man
nuances that convey information at a number of levels Machine Interface (MMI).
that are received and recognized in a way that is currently As mentioned above, the information we receive is pro-
impossible by machines. cessed at several levels in our central nervous system
When communicating with a person who does not share from reflex responses up through lower brain control to
your language, it is often possible to adapt your technique higher cortical control. Having decided on a response to
using gestures and a few shared words to convey meaning, the information gathered by our senses and their relation
but the speed and quality of the information is degraded. to other memory and cognitive processes we are able to
At a lower sound level, the introduction of noise may influence our environment. Our methods for influencing
mask some of the information leading to misinterpretation. the environment are based on bodily control through the
Human–Machine Interface 627
movement of our limbs and our ability to communicate Where clear conventions exist, such as turning a wheel
using speech. clockwise to move a vehicle to the right, they should
From the most basic tools, such as a rock used to open a be adopted.
nutshell, through to a pilot’s yoke linked to a sophisticated Where there is some doubt over the natural physical
fly-by-wire jet aircraft, we convert physical movement, relationship between the cause and effect, for example,
through gripping objects, into a desired outcome. Our using a horizontally mounted joystick to move a cursor up
ability to tailor these tools to our requirements is central or down a screen, the stereotype of the population should
to our ever-advancing technologies. be explored.
The shaping of the earliest flint tools to the hand size of As the design of interfaces shifts from physical design
its intended user illustrates our early skills in tailoring to our toward one of human computer interaction design, this
requirements. In the case of aircraft controls, the need to tai- tailoring of the controls has developed a new emphasis
lor the design to the capabilities of potential pilots continues on information flow and the associated mental workload.
to hold true with the need to avoid the exclusion of users The basic objective of ensuring that the capabilities and
on the grounds of hand size or grip strength, and so on. limitations of the potential user population are respected
The usability of controls extends beyond the person’s continues to hold true across this range of technologies
ability to physically grip or manipulate the control device. from the very simplest through to future technologies yet
Our ability to use controls is based on an expectation to be conceived.
of their function. We are able to use a hammer because This article has provided an introduction to the Man
we have past experience of hammers and their use. The Machine Interface. The topic is extended by explaining the
hammer itself provides us with cues to its function with an six human factor domains in Article 96, The Domains of
obvious area to grip and a flattened metal area for hitting Human Factors Integration, Volume 2; and Article 97,
the nail. Design Methodology, Volume 2 covering how to go about
More advanced tools/machines rely on the same compo- designing for Human Factors.
nents of knowledge and cueing to convey their function to
the human operator. For complex systems such as an air-
craft flight deck, this places a requirement for consistency
REFERENCES
in design and operation without which pilots experienced
in flying planes with different controls will require retrain- Diamond, J. (1999) Guns, Germs, and Steel; The Fates of Human
ing and, even once trained, may make errors owing to Societies, W.W. Norton & Company.
a tendency to revert to the operational expectations of Fitts, P.M. (1951) Human Engineering for an Effective Air Navi-
other systems. gation and Traffic Control System, National Research Council,
This expectation of how something operates can be WA.
seen on a wider scale with the existence of population Fitts, P. (1954) The Information Capacity of the Human Motor
stereotypes such as association with twisting a tap head System in Controlling the Amplitude of Movement. Journal of
clockwise to turn it off. When designing a system that may Experimental Psychology, 47, 381–391.
be used by a multinational population, these stereotypes Sydenham, P.H. (2003) Systems Approach to Engineering Design,
may change with a consequent risk of error. Artech House.
96: The Domains of Human Factors Integration
Nicholas I. Beagley
DSTO, Canberra, ACT, Australia
to continue is attained – see Article 66, Phases of Sys-
1 Human Factors Integration and MANPRINT 628 tem Life Cycle, Volume 2 and Article 69, Requirements
Allocation, Volume 2.
2 Manpower 629
This process is well suited to the specification and
3 Personnel 629 testing of tightly defined engineering solutions designed
4 Human Factors Engineering 630 to meet a clear functional requirement – see Article 69,
5 Training 630 Requirements Allocation, Volume 2.
6 Safety 631 However, where human performance contributes to total
7 Hazards 631 system effectiveness, the users’ capabilities and limitations
References 631 must be taken into account. At the same time, the added
complexity of human involvement reduces the predictability
of design alternatives taken within the development process,
thus limiting the degree to which the requirements can be
tightly specified or met.
1 HUMAN FACTORS INTEGRATION AND
Despite these difficulties, accounting for the human com-
MANPRINT ponent must not be overlooked. To the contrary, the need
was recognized for a more effective way of inserting
Large-scale acquisition, for example, military procurement, Human Factors Analysis into the defence procurement
is normally approached at a high level with the aim of lifecycle. From this requirement, emerged the Manpower
achieving a general capability. The lower level objectives Integration (MANPRINT) program, developed by the US
and the functionality required from any system designed to Air Force in support of their system acquisition program,
fulfill this capability are often unclear at the outset. MANPRINT (2003).
Procurement programs generally set out a vision of the MANPRINT sets out the necessary Human Factors inputs
capability and then set about establishing, in ever-increasing to the procurement process in the form of prescribed analy-
detail, the system design solutions necessary to achieve sis and reporting for each stage of a system’s procurement.
the vision – see Article 63, Outline of Systems Thinking, Further guidance was provided through its positive direc-
Volume 2 and other articles within the section Engineering tion to design teams to consider defined areas of Human
a Measuring System. Factors assessment, called domains. These are consid-
Commonly adopted procurement processes share similar- ered here.
ities in their approach, stepping through defined sequences UKMOD adapted the MANPRINT framework to map
of specification, design, refinement, acceptance, delivery, onto their own updated procurement process (UK MOD
through life support, and finally disposal. 2000). The UK’s Human Factors Integration (HFI) program
The process is normally interspersed with milestones by defines the Human Factors specific roles and responsibilities
which the progress of the program is assessed and approval of the Capability Working Groups and Integrated Project
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
The Domains of Human Factors Integration 629
Teams prescribed by the overall procurement process thus that must be planned for in the full life considerations
giving HF a formal place in the development cycle. of setting up and maintaining a capability. Similarly so
Central to the HFI process is the activity of Early Human with complex measuring systems such as are found in
Factors Analysis, in which concepts are reviewed within the aircraft, transport systems, industrial plants, and medical
domains of HFI. This is started at the earliest opportunity to health systems.
avoid where possible costly changes to the final design due Where a system relies on a particular set of skills at
to the discovery of incompatibilities with the human user. different levels within the system, its design should consider
The specified activity of Early Human Factors Analysis the program’s phasing demands to supply the right people
compiles a log of the issues identified within each domain, with the right level of skills at the right time along with
the plan for each issue’s resolution, and the process toward a sustainable succession plan over the lifetime of the
an acceptable design. system.
As defense equipment includes numerous measuring sys-
tems that have to perform in demanding conditions, MAN-
PRINT methodology will be found useful for measuring 3 PERSONNEL
systems design of all kinds.
The Personnel domain of HFI is concerned with the charac-
teristics of the people in relation to the requirements of the
2 MANPOWER total system, in pursuit of the optimal system performance.
People’s characteristics vary considerably; it is necessary
The Manpower domain of HFI is concerned with the num- to recognize their capabilities and limitations in the context
ber of people required to operate and maintain a system. of the system’s demands on the person. It is also important
This consideration goes beyond the issue of the person from the point of selection and training and their associated
sitting at the controls of a piece of machinery, extending to costs to recognize when and where, in which subsystem,
the people required to train that person, maintain the system, a system design is likely to demand particular aptitudes
relieve that person when they become fatigued, and so on. or knowledge.
When considering large system structures involving many In some cases, for instance in aircraft and race car cock-
people, such as the manning of a ship, the logistical issues pits, the system places physical constraints on the individual
of manpower can become critical to the sustainability of the such as body size, physical strength, or a resistance to
system. For example, the total cost of a naval ship owes motion sickness.
far more to the manning and sustainment cost than it does Whilst training and experience can help develop candi-
to the hardware purchase costs. dates to meet a system’s requirements in terms of physical
This situation is commonly met in many of today’s strength, endurance, and motor skills, some attributes such
systems. A design or operational approach that efficiently as anthropometry are fixed.
uses the capabilities of the people to reduce the total number Some trades require certain physical criteria for selection
of people required to operate a large system can represent designed to ensure that the people selected are capable of
a significant financial saving. meeting the requirements of the systems they support. For
Reducing the number of people available for operating a example, selection into the army demands a level of phys-
system is likely to raise the responsibilities and associated ical fitness that allows the designer of supporting systems
workload placed on those remaining. The sustainability of to make certain assumptions about the target population’s
this approach must be considered across a range of scenar- physical abilities.
ios. For example, it may be possible to conduct the normal The mental aptitude of the user group can greatly decide
operations of the ship under a peaceful scenario with a given their ability to develop the cognitive skills necessary for
staffing level. If, however, a wartime scenario places raised contributing to the performance of a system. For example,
demands on the crew and their ability to consistently per- market traders must demonstrate a capacity for extracting
form at the required level cannot be sustained, this situation relevant information from a multitude of market informa-
must be accounted for. tion sources and apply this information, along with relevant
Important implications can arise if the design demands a knowledge and experience, to make informed decisions on
higher entry requirement for the people chosen to operate the purchase and sale of assets. The personal attributes nec-
the ship than those currently employed in similar roles. essary to be successful in this occupation are not common
If this is the case, then the availability of these people across the whole population.
in the workforce and the ability to motivate them into The design of any system should clearly establish the
working on the future ship will be important assumptions assumptions about the personnel it is making with regard
630 Measuring System Design
to the skills, limitations, and expectations of its intended 5 TRAINING
user population; this will help ensure a significant mismatch
does not arise. For large-scale system acquisitions, this is The Training Domain of HFI is concerned with the knowl-
normally captured in a document called a Target Audience edge and skills required for operating and maintaining
Description (TAD). the system. The importance of training increases with the
complexity of a system and with the negative consequences
of system failure. Some instrument systems are easy to learn
4 HUMAN FACTORS ENGINEERING
to operate but some require specific training to be used cor-
The Human Factors Engineering (HFE) domain of HFI rectly. For example, using a simple digital voltmeter should
is concerned with the physical and operational integra- not tax any competent technical person but setting up an
tion of the user with a system’s process and associated automated test equipment console is another matter.
technologies. These considerations extend to all aspects of Consideration of the training requirements of a system
human/system design and include guidance documentation, should produce recommendations regarding what should be
and a range of tools and techniques designed to ensure the taught and the method that should be adopted for teaching.
full integration of the system’s characteristics with the capa- There exist many approaches to training and many dif-
bilities and limitations of the human operator, the demands ferent scales of training appropriate to the system under
of the task, and the attributes of the environment. consideration. Training for a simple system such as a com-
HFE within the HFI process is normally concerned with mon consumer product, for example, a television set, might
the specification of mandatory criteria such as are given in simply take the form of a reference leaflet. People with
Def Stan 00–25 (A UK MOD Defence Standard) and Mil experience of similar systems might not require the leaflet
Std 1472F (A US DOD Standard) against which designs to access the regular features of the television, using it only
will be judged. These not only prescribe documents but to explain unfamiliar features.
also offer guidance for specifying both the known user Complex systems, for example, large software applica-
specific–design constraints and the appropriate methodolo- tions, such as setting up an exhaustive in-flight test plan for
gies for ensuring that user requirements are established and a new car, railway, or aircraft, are often accompanied by a
accounted for throughout the design process. training course that leads potential users through the avail-
An increasing concern of HFE is in the area of Human able functionality and provides worked examples to ensure
Computer Interaction (HCI). An analysis of the underly- an understanding of the concepts on which the software’s
ing task is frequently the starting point from which concept functionality is based.
designs are created and iterated on the basis of subjective Techniques, such as computer-based training and self-
user feedback, workload assessments, and objective mea- instructional packages, have increased in popularity given
sures of human performance. their ability to support training at a pace determined by the
The traditional area of ergonomics also begins with the trainee without the overheads of a human trainer.
need to understand the task and the characteristics of the Computer-based training tool development should be
user to ensure an appropriate fit. Human body modeling, based on a careful analysis of the knowledge and skills
prototype development, and user-based testing and evalua- required by the trainees, and so are not suited to all training
tion, all contribute to the activity of ensuring the fit between demands for a given system.
the human and the system. The knowledge, experience, The use of simulation as a training tool has expanded
and skills, both physical and cognitive, must also be deter- the range of tasks that can be trained for using a computer.
mined to inform the design process. All of these processes A current limitation to computer-based training is that it is
have a large supporting literature – Chapanis (1996), Jacko only able to provide limited feedback. Humans still remain
and Sears (2002), Sanders and McCormick (1993), Tilley the most effective trainers across a wide range of contexts,
(2001), Wickens, Gordon and Liu (1997). offering insights based on a deep base of knowledge and
Human Factors analysis techniques must be sensitive to experience combined with the power of inductive reasoning
variations in the user population and the different contexts and wisdom to appropriately guide the trainees.
of operations. Designs must take into account different task The normal form of training analysis and reporting
demands dictated by possible scenarios, in which the system applied to large-scale procurements is called the training
might be required to function. needs analysis (TNA). Development of a TNA is con-
The impact of different environments on the user popula- ducted early in the design and development process setting
tion’s ability to use the system in the intended manner needs out anticipated training system requirements such as the
consideration to allow appropriate changes to be made to required facilities, the necessary fidelity, the time require-
the design or the intended operational procedure. ment for trainees, and so on.
The Domains of Human Factors Integration 631
6 SAFETY domain of HFI is concerned with the conditions of opera-
tion inherent in the system and their impact on the health
An important responsibility of the system designer is to and safety of the human operator.
safeguard the well being of people using or coming into At the heart of hazard analysis is the consideration of the
contact with the system. The System Safety domain of HFI system’s operational characteristics and the environment in
is concerned with the risks to human health that might arise which it operates.
from system failure or use in an unintended manner. Factors such as noise and vibration can have health
Consideration of system safety is in reality an analysis implications for its operators and those around
of risk. This requires consideration of all aspects of system them.
design and operation, in which the demands of the task Noise dose limits are set in legislation on the basis
place the operator, or bystanders, under a risk of injury. of dose limits worked out as a combination of intensity
This analysis must include opportunities for error in the and frequency range. Vibration levels can have significant
operation of the system. implications for the health of the individual with the link
In each case, a standard risk assessment is normally between prolonged doses of mid to high frequency vibra-
undertaken, in which the likelihood of the risk occurring tions to back pain arising from musculoskeletal injury. The
and the consequence for human safety are rated. combination of vibration, repetition, and wrist deflection
Issues revealed by this process can have additional impli- when using power tools, has been linked to high incidence
cations for system performance that may not represent a of repetitive strain injuries (RSI) such as vibration white
significant risk to user safety, but may still result in the finger and carpal tunnel syndrome.
failure of a system to meet its objectives. Many mechanical systems include an inherent risk to
The output of a safety assessment and its risk analysis is the operator. Many systems contain toxic substances or
usually a hazard log that characterizes the risk and draws produce toxic fumes from which the operator must be
on relevant expertise to establish measures that should be protected.
taken to mitigate the risk, for example, the mitigation of Similarly, moving parts present a risk to the opera-
the risk of sailors falling from their yachts in high seas tor with opportunities for cutting, trapping, and crushing
is usually a combination of harnessing, lifejackets, and limbs and digits. Well-established methods of guarding the
location beacons. operation of mechanical systems should be in place and
In some cases, a conflict will arise between the proposed should be tested against a range of potential operational
safety measure and the performance of the system. On rare scenarios including the risk of circumvention by the oper-
occasions, users may be faced with a potential compromise ators themselves.
between the risk of injury against the performance of Inherent risk to the vision of the operator can arise. Activ-
the system. ities that require long periods of visual attention should be
When failure of the system has significant implications broken up into shorter periods to reduce fatigue. Some sys-
for the welfare of the individual, for example, in some mil- tems such as the use of high-magnification optics require
itary systems, there may be circumstances in which the safeguards ranging from training and procedures through
individual may elect to bypass a safeguard in the interest to reactive optical filters that prevent operator blindness
of system performance. In such cases, it is important for caused by pointing the device at a high luminance source
designs and procedures to be developed that ensure safe such as the sun or a laser.
use under normal circumstance and a reasonable balance Extremes in temperature are a hazard to the operator
of risks in life-threatening situations. Within the guides of whether they are a result of the external environment
relevant legislation, the design should allow the user to given the scenario or because of the heating or cooling
make these decisions when necessary. Safety in systems of a component.
is also covered in Article 99, Safety Organization, Vol- Where a risk is identified, consideration should be given
ume 2; Article 100, Safety Analysis Methods, Volume 2; to guarding against the risk, including an assessment of
and Article 101, Assessing and Demonstrating Safety, whether the system’s operational procedure raises or lowers
Volume 2. the individual’s exposure to the risk.
7 HAZARDS REFERENCES
Whereas the Safety domain was concerned with risks Chapanis, A. (1996) Human Factors in Systems Engineering, John
related to the use and abuse of a system, Health Hazards Wiley.
632 Measuring System Design
Jacko, J.A. and Sears, A. (2002) The Human-Computer Inter- Tilley, A.R. (2001) The Measure of Man and Woman: Human
action Handbook: Fundamentals, Evolving Technologies and Factors in Design, John Wiley.
Emerging Applications (Human Factors and Ergonomics), UKMOD (2000) A Practical Guide to Human Factors Interaction:
Lawrence Erlbaum Associates. A Guide for Integrated Project Teams and Capability Groups,
MANPRINT (2003) MANPRINT documentation is only avail- Version 0.9.3, 19 July, Ministry of Defence, UK, also available
able as a web document on http://www.manprint.army.mil/ on http://www.ams.mod.uk/ams/content/docs/hfiweb/hfihome.
manprint/guides/mp-ref-guidebook.asp. htm.
Sanders, M. and McCormick, E.J. (1993) Human Factors In Engi- Wickens, C.D., Gordon, S.E. and Liu, Y. (1997) An Introduction
neering and Design, McGraw-Hill Science/Engineering/Math. to Human Factors Engineering, Addison-Wesley.
97: Design Methodology
Nicholas I. Beagley
DSTO, Canberra, ACT, Australia
In either case, the role of the user, as the foremost expert
1 The Objective of Design 633 in their requirements, is vital in guiding the design process.
2 Requirements Specification 634 The challenge facing the human factors practitioner is to
ensure that users are provided with appropriate levels of
3 The Design Life Cycle 634
information and experience of emerging concepts, and that
4 Accounting for User Variation 634 any bias within the population are recognized and balanced
5 Design Standards and Guidelines 635 where appropriate.
6 Concept Modeling 635 When designing a system, many attributes must be con-
7 Product Evaluation 636 sidered on the way to meeting its associated capability.
Further Reading 636 Designs are a trade-off of attributes such as functional-
ity, ease of use, cost, maintainability, training demands,
and so on. For example, if you are designing a consumer
product such as a kettle, the product’s overriding objec-
1 THE OBJECTIVE OF DESIGN tive is to appeal to the consumer market and thus achieve
sales. The attributes necessary to achieve this go beyond
The underlying objective of the design process is to pro- the technology required for boiling water – to the cost of
vide a new or enhanced capability. Design can be either production, ease of use, and aesthetics of the design. Other
evolutionary or revolutionary in nature. issues such as brand styling and reputation may place addi-
An evolutionary design takes discrete steps forward from tional demands on a design team.
an existing design’s form and functionality. Any design can be judged on the basis of its comparative
The alternative is a revolutionary design in the capa- achievement in providing a defined capability, that is, it
bility that is normally associated with the availability of must be fit for the purpose. It must not only provide the
a new technological development that will support a new necessary functionality but also take account of the users’
capability or an ability to do something in a new and capabilities, the demands of the task and the restrictions of
improved way. the environment.
One difficulty associated with evolutionary design is that As a consequence, designs should be considered in the
the expectations of the user are firmly aligned to the current broad context of their operating environment. Setting up
product. This can have the effect of coloring their opinion the boundary limits of a project are covered in Arti-
of prototype systems that operate in a slightly different way. cle 68, Setting the System Boundaries, Volume 2. If a
Revolutionary design faces a different problem in that design does not account for the user’s requirements, it
with no model to relate to, the potential users of a revolu- will at best be inefficient and at worst be rejected out-
tionary system may have difficulty in grasping the role and right by its intended user. Phases of the life cycle of a
function of the new system sufficiently to provide useful development are covered in Article 66, Phases of System
feedback for the system’s design and evaluation. Life Cycle, Volume 2. More on requirements generation
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
634 Measuring System Design
is found in Article 69, Requirements Allocation, Vol- 3 THE DESIGN LIFE CYCLE
ume 2.
The design life cycle is a process of concept refinement
to the point of production of a solution that can be imple-
2 REQUIREMENTS SPECIFICATION mented. The design of a new car or a new training program
each start with a concept that brings together particular
Specification of large-scale systems is frequently conducted attributes perceived to present the intended product with
as a two-stage process starting with a capability focused a niche in the marketplace.
user requirement that is then refined to become a more The design life cycle should extract and question the
detailed, functionality focused, systems requirement. assumptions surrounding these perceived qualities whilst
This is essentially driven by engineering-based, quality seeking to embody them in the final design.
control processes. Guidance for the input of Human Factors The scale of the design process should be tailored to
into this process has been developed in the form of the an appropriate level given the available resources and the
UK’s Human Factors Integration (HFI) Process and the consequence of design failure.
US’s MANPRINT program. In these programs, human Safety critical systems rely on stringent safeguards that
analyses and their deliverables are mapped onto a defined prolong the design cycle whilst some low cost commercial
acquisition process, in which the various customer and products can afford to fail in the market place as long as
supplier responsibilities are prescribed, and the milestones lessons are learned that improve the chances of success for
for acceptance, delivery, and payment are set out within the subsequent products.
framework of the design and development process. Design should be considered to be a competitive process.
The user requirements document normally sets out the There are costs associated with the design process in time
characteristics of the current capability, its users and the and money as well as disruption to the intended user of
procedures for use whilst describing the enhanced concept, the new design, particularly when it is intended to replace
its associated functions, and the context of operation. an established system. The value of the proposed design
Also defined is an initial assessment of the constraints, enhancement should always be compared to a baseline
such as user recruitment, that might be placed on the taken from what has gone before. Where available, a
future system and the measures of effectiveness, (MoE) comparison should be made with alternative designs.
by which a system could be evaluated. MoEs should be If the life cycle process of a design raises the costs
sensitive to all aspects of the concept capability and are beyond the resources of the potential purchaser, or delays
particularly important for the evaluation of design options. the delivery to a point that it becomes obsolete before
Example measures include subjective measures such as user delivery, the life cycle process will have failed.
acceptance and objective measures expressed in terms of There has been increasing pressure to reduce the time
system performance and usage errors. and costs associated with the design life cycle. This, in
For large scale systems, the user specific information parallel with recognition of a user’s inability to completely
should include an overview of the likely capabilities of specify their requirements in the early stages of the design
the target population, available numbers, existing training process, has led to a shift toward a process of concept
demands, career, retention issues, and so on. prototyping and design evolution through user testing and
The system requirement document (SRD) is normally virtual prototyping.
a more tightly defined description of the required system Within this process, there remains a need to establish
and its components. It is derived through a process of clearly defined criteria for acceptance with which to evalu-
user based–concept refinement using prototypes developed ate the progress of a design and to manage the contractual
in response to the user requirements document (URD). obligations between the customer and the supplier.
This document sets out clearly defined functions and their At the same time, there has been recognition of the risk of
associated performance criteria. failure associated with design and the need for a closer part-
Human specific performance criteria should relate to nership between the user/specifier and the supplier/designer
issues of safety, operability, reliability, maintainability, in moving toward a product that achieves the best balance
training, and so on, and might include such performance between the inevitable compromises of design.
standards and metrics as completion times, error rates, lift-
ing limits, workload limits, and so on.
It will specify the necessary standards to adopt and 4 ACCOUNTING FOR USER VARIATION
should include appropriate Human Factors standards for
design such as the UK’s Def Stan 00-25 or the US’s Mil Human variation must be recognized when designing for
Std 1472F. a heterogeneous population. Each person is an individual
Design Methodology 635
in the whole population and it is important to recognize out the expectations of the procurer by defining acceptable
and design for the range of capabilities in a targeted levels of performance of a system.
user population. This is particularly important for areas such as human
Variation in size, weight, strength, sight, hearing, mem- health and safety, in which design standards play an essen-
ory, experience, skills, and so on, can each influence our tial role in safeguarding future users of a system.
ability to use an interface or play a part as a component of Several organizations develop standards for engineer-
a wider system. ing design, including the International Standards Organiza-
From the perspective of design, there are four main tion (ISO), the UK’s Defence Standardization Organization
approaches for accounting for human variation. These are (Dstan) and the US Department of Defence (USDOD).
product tailoring, self-adjustment, multiple versions, and Key Human Factors standards from these organizations
one size fits all. include ISO 9241, Def Stan 00-25 and Mil Std 1472F.
The tailoring of a suit to our body shape has similarities Whilst some are created for defense application, they are
with bespoke software and product design offering confi- also useful to all human factors uses. Standards from the
dence in a product’s fitness for purpose, at a price. government military and some standards organizations are
The approach of allowing self-adjustment is highly suit- freely available on their Web sites. The ISO standards are
able for many products. Car and office chairs must accom- available in some libraries, or can be purchased on-line.
modate a wide range of the population and must therefore Standards such as the UK’s Def Stand 00-25 cover a
include the capability for adjustment, without which large broad range of human, task, and environmental attributes
proportions of the community would find the product unus- and their design implications.
able. Many of the mainstream software applications allow Def Stan 00-25 assists the designer by defining known
a degree of user adjustment to suit their working practices. parameters of the military user population it serves pro-
It should be noted that too much adjustment capability can viding the designer with reference tables for operator size,
bring with it costs in adjustment time or even weaknesses strength, sensory capabilities, and so on. It also provides
in the structural integrity of a product. guidance on appropriate strategies for design and testing to
The approach of producing multiple versions is familiar ensure that the government’s requirements for integration
to us all through the purchase of off-the-shelf clothing. of the human users are met.
This term off-the-shelf has been adopted more widely
to describe products that have already been developed
and marketed. The two usages bear similarities in their 6 CONCEPT MODELING
associated cost savings.
The final approach of one-size fits all is appropriate for It can be quite difficult to tightly define specifications for
a subset of products depending on their intended purpose. a concept early in the design cycle. Without a product or
The term is frequently applied to T-shirts that will look system on which to base their assessment, users and design-
like a tent on some people and stretch to fit others. This ers alike encounter problems in specifying and refining
design compromise frequently offers considerable savings their concepts.
at the cost of poor fits for proportions of the potential A common approach to making these concepts more
user population. concrete as a tool for evaluation or design refinement is
through modeling.
A model of a system can take many forms. The form
5 DESIGN STANDARDS AND chosen should reflect the purpose of the intended model. A
GUIDELINES simple paper-based prototype of a series of screen interface
designs can be a low cost, quick to develop, and flexible tool
The normal repository for the knowledge gathered through for communicating a concept and seeking user feedback.
research on human attributes is the published literature High-end prototyping applications can be used to simulate
and teaching textbooks; it is widely dispersed. The broad the design and intended functionality of a proposed sys-
application of human sciences findings to system design tem. This might be a logical progression from the use of
and their implications for system success has led to the paper-based prototypes to refine a concept. The degree to
need for design standards and guidelines that are accessible which these methods are used depends on the complex-
to the design community. ity of the system being designed and the consequences of
These documents often have the dual role of offering inappropriate design features remaining in the final design.
advice to a designer as they seek to develop a product that Given the costs associated with altering physical designs,
integrates with the human, and at the same time setting there is an obvious emphasis on prototyping to reveal
636 Measuring System Design
design issues prior to the final production run. Physical For example, the down selection between a number of
prototypes can take many forms from cardboard mock- new cars might be made on the basis of their promotional
ups through to machined pieces. Computers are playing literature but the final selection before purchase normally
an increasing role in the modeling process with computer involves a test drive of the remaining options. If you were
aided design techniques used to refine design concepts buying a fleet of cars, there may be value in arranging for
and to provide visualizations. Human body modeling tools a longer term loan of a vehicle to consider other issues of
offer the designer the ability to examine issues of popu- suitability within your context of intended operation.
lation fit and reach at the very earliest stages of concept The process of comparative evaluation is similar with the
formulation, helping them to weigh up alternative design availability of many tests possible for comparing competing
strategies and avoid the cost of large changes that would products against a priority list of desirable features and
have been associated with discovering these problems fur- negative attributes to be avoided.
ther along the design path. More recently, the use of stereo There exist a number of tools and techniques that can
lithography has increased to provide physical outputs to be used to help in the development of a priority list of
computer designs providing further realism to the process attributes in situations in which different users can be
of evaluation. expected to hold different opinions. Approaches include
workshops, ballots, and a tool called the repertory grid,
in which attributes are compared individually and analyzed
7 PRODUCT EVALUATION to produce a ranked list.
With all approaches, there is a need to ensure that users
The evaluation of designs can be conducted at several are given sufficient access to the designs under consid-
levels. Evaluations might be an assessment of the design’s eration of representative circumstances. Where required,
fitness for an intended purpose or it might seek to compare training should be provided and the implications of skill
two existing designs to determine which is the best, or the development with training over time accounted for in the
least bad, design for the intended purpose. analysis and the interpretation of the user’s subjective
In the process of either evaluation, it is possible to responses to a system.
reveal attributes of a design that are particularly suitable or
unsuitable thus providing a third option by which desirable
components of competing systems are brought together, FURTHER READING
or unsuitable components are enhanced in the process of
developing a new and improved system. DOD-HDBK-763, Human Engineering Procedures Guide.
It should be recognized when evaluating a design’s ISO 9241, Ergonomic Requirements for Office Work with Visual
fitness for a purpose, that it may be that the design was Display Terminals, Parts 1 to 17.
developed to meet a slightly different requirement of an MIL-STD-1472E, Department of Defense, Design Criteria Stan-
alternative user group. dard: Human Engineering.
Variations in user, task, or environmental context can MIL-H-46855B, Department of Defense, Handbook: Human
have important implications for the success of a design. Engineering Guidelines for Military Systems, Equipment, and
Facilities.
Whilst there is much that can be done to assess the attributes
of a potential system through basic analysis, the findings MIL-HDBK-759C, Department of Defense, Handbook for Human
Engineering Design Guidelines.
cannot assure one that it will provide absolute assurance of
suitability. MIL-STD-1801, User/Computer Interface, 29 May 1987.
Prior to final acceptance a design should be tested in its Ministry of Defence, Defence Standard 00-25, Human Factor for
context, or perhaps within a close representation of that Designers of Equipment, Parts 1 to 13.
context, to reveal issues of suitability. NASA-STD-3000, Man-Systems Integration Standards (MSIS).
98: Reliability and Maintainability
Patrick D.T. O’Connor
Consultant, UK
The MTBF measure implies that the rate at which fail-
1 Concepts of Reliability and Maintainability 639 ures occur is, on average, constant over the time con-
2 Causes of Failure 640 sidered, and then MTBF is equal to the inverse of the
failure rate.
3 Design for R&M 641
In the case of components that can only fail once, the
4 Testing for R&M 642
expression hazard rate is used. It is often expressed as
5 Reliability Data Analysis 642 failures per million hours (fpmh). Failure rate and hazard
6 Related Concepts 644 rate are conventionally denoted as λ.
Related Articles 644 In the case of a constant failure/hazard rate, the reliability
Reference 644 (R = probability of no failures over a time interval t) is
Further Reading 644
R = (1 – probability of failure) = EXP[–λt] (1)
1 CONCEPTS OF RELIABILITY AND If the hazard/failure rate is not constant, other relationships
MAINTAINABILITY apply. The reliability formula most often used is the Weibull
distribution:
1.1 Reliability
(t – γ ) β
R = EXP – (2)
η
Reliability is the ability of an item or system to keep
operating, or to be available for operation, over a period
of time without failing. Reliability can be defined as a where γ is the failure-free life (durability), η is the location
mathematical probability of not failing or of being available parameter, and β is the shape (or slope) parameter.
for use. If β = 1, then the constant hazard/failure rate is repre-
Durability is the ability of an item to withstand wear out sented, and η is the MTTF/MTBF. If β > 1, it represents a
mechanisms such as fatigue, wear, and so on. Durability is hazard/failure rate that is increasing over time (wear out).
usually expressed as a minimum time before the occurrence If β < 1, it represents a hazard/failure rate that is decreas-
of wear out failures. ing over time (infant mortality).
Reliability is also expressed as the mean time to failure Figure 1 shows the three basic patterns of failure that
(MTTF) (for an item that can only fail once, like a light typically occur. This is often called the bathtub curve. As
bulb, microprocessor or interplanetary spacecraft), or the shown, it applies best for electronic systems but not so well
mean time between failures (MTBF), for an item that can for mixed ones with mechanical components.
fail and be repaired a number of times, like a car or an Reliability statistics and methods are fully described
electronic assembly. in O’Connor (2002).
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
640 Measuring System Design
Total hazard rate ‘bathtub’ Wearout
Hazard rate
failures
of ‘good’
items
Probability
Externally
induced
Failure of weak items failures
Infant mortality Useful life Wearout
Time
Figure 1. Patterns of failure: the bathtub curve.
0
L S
1.2 Maintainability and availability
Load−strength
Maintainability is the property of an item to be returned Figure 2. Stress and strength variation.
to a working condition after failure (corrective mainte-
nance, CM), or to be kept in a working condition by
performing preventive (or planned) maintenance (PM). tail of the stress distribution is applied to an item
CM is expressed as the mean time to repair (MTTR). in the weak tail of the strength distribution so that
PM can be expressed as total time (e.g. man-hours) over there is overlap, then failure will occur. Obviously, it
a period. becomes more difficult to design against failure, or to
Availability is the probability that an item or system will predict reliability, in this kind of situation. We can no
be available for use at a particular time, or the ratio of longer rely on simple, deterministic, values of load and
available time to total time. In the steady state, assuming a strength. We now need to know how these values vary,
constant hazard/failure rate and no PM: particularly in the high tail of the stress distribution
and in the low tail of the strength distribution. Values
MTBF
Availability(A) = (3) near the average, or in the opposite tails, are of little
(MTBF + MTTR) interest to engineers when considering this kind of
failure, during design and test.
4. Failures can be caused by wear out. Figure 3 shows
2 CAUSES OF FAILURE
how time-dependent strength variation affects reliabil-
There are many reasons and causes why a product might ity. The level of uncertainty in predicting reliability
fail. The main causes of failures are the following: increases considerably over time.
5. Failures can be caused by other time-dependent mech-
1. The design might be inherently incapable. The list of anisms. Battery run-down, material creep, and progres-
possible reasons is endless, and every design problem sive drift of electronic component parameter values are
presents the potential for errors, omissions, and over- examples of such mechanisms.
sights. The more complex the design or difficult the
problems to be overcome, the greater is this potential.
2. The item might be overstressed in some way. If
Strength
the stress applied exceeds the strength, then failure
will occur.
3. Failures might be caused by variation. The actual
strength values of any population of components will
vary: there will be some that are relatively strong,
others that are relatively weak, but most will be of
Load
nearly average strength. Also, the stresses applied will
be variable. Figure 2 shows this general stress–strength
situation. Failure will not occur if the applied stress
t ′ Time/load cycles
does not exceed the strength. However, if there is Log scale
an overlap (interference) between the distributions of
stress and strength, and a stress value in the high Figure 3. Time-dependent strength variation.
Reliability and Maintainability 641
6. There are many other potential causes of failure. Gears Reliability prediction: The expected reliability of the
might be noisy, oil seals might leak, display screens new design is estimated, using data and assumptions regard-
might flicker, operating instructions might be wrong ing application environments, and so on. Reliability data is
or ambiguous, electronic systems might suffer from published for many component types, most notably the US
electromagnetic interference, and so on. Military Handbook 217 MIL-HDBK-217 (Reliability Pre-
diction for Electronic Systems). However, this approach is
The important point to appreciate is that failures are by no nearly always misleading, on several grounds. The most
means always clear cut and unambiguous. They are often notable are the following:
open to interpretation and argument. They also differ in
terms of importance (cost, safety, other effects). Therefore, • Engineering components have no intrinsic, quantifiable
we must be careful not to apply conventional scientific, property of failure rate.
deterministic thinking to the interpretation of failure data • The mathematical models used to represent the effects
and the estimation of reliability. of application conditions (temperature, etc.) on failure
For example, a mere count of total reported failures of rate are not always consistent with the physical and
a product is seldom useful or revealing. It tells us noth- other causes of failure.
ing about causes or consequences, and therefore nothing • The data provided in such sources is nearly always
about how to improve the situation. This contrasts with a of dubious quality and several years out of date in
statement of a physical attribute such as weight or power relation to the technology and capability, thus giving
consumption, which is usually unambiguous and complete. pessimistic/conservative predictions in comparison with
what can actually be achieved with adequate engineer-
ing effort being applied.
3 DESIGN FOR R&M • Failures do not generally occur at constant average rates
over time. They can be time-dependent, cyclic-stress
Manual and computer-driven analyses are usually cheaper dependent, and so on.
than testing, and testing also adds to project timescales. • Component failures do not necessarily cause system
Therefore, designs should be analyzed as far as is practica- failures, and system failures are not always the result
ble, (see Article 71, Modeling Methodology, Volume 2) of component failures.
to obviate or to reduce the need for testing.
Failure modes and effects (and criticality) analysis:
Analysis can also show up which aspects of the design
might need to be further investigated or confirmed by test- Failure modes and effects analysis (FMEA) or failure
modes and effects (and criticality) analysis (FME(C)A) is a
ing, by highlighting aspects that present risks or uncertainty.
method for tabulating all of the components (or functions)
Reliability, durability, and performance aspects that can
within a design, and asking the following questions of
be affected by variation are difficult to analyze effectively,
so further testing is nearly always necessary to confirm each one:
these aspects. 1. How can it fail in the application? (failure mode).
Designs can be analyzed manually, with mathemat- 2. How likely is each failure mode? (failure probability
ics software, or with computer aided engineering (CAE) or failure rate).
software. These include computer aided design (CAD), 3. What would be the effect of each failure mode?
finite element analysis (FEA), electronic design automation (failure effect).
(EDA), electromagnetics (EM) analysis, system simulation, 4. How critical is the effect? (failure effect criticality).
and other tools. 5. What should or could be done about it?
The tools enable engineers to create and analyze complex FME(C)A is widely used in many industries, particularly
and difficult designs quickly and economically. However, in those for which failures can have serious consequences,
it is not safe to rely on the software to replace the need for such as military, aerospace, transport, medical equipment,
testing. All CAE methods involve assumptions and simpli- and so on.
fications that can, to varying degrees, generate erroneous Some industries have established standardized appro-
or misleading results. No design analysis software can deal aches (the US Military Handbook is MIL-HDBK-1629, and
with the whole range of possible operating stresses, envi- the US Society of Automotive Engineers (SAE) has also
ronments, variations, and degradation mechanisms that can produced a guidance document).
cause failures.
Some methods are used specifically for R&M analy- Variation analysis: Variations (tolerances, environmental
sis of new designs. The methods most often used are conditions, interactions, etc.) should be analyzed to ensure
the following: that they can be controlled and their effects are minimized,
642 Measuring System Design
see Article 55, Uncertainty Determination, Volume 1; Table 1. Accelerated test/DoE selection.
Article 52, Error Models, Error Budgets and their Cal- Important variables, effects, etc. DoE/Accel. test
culation, Volume 1.
Parameters: electrical, dimensions, etc. DoE
Effects on measured performance DoE
Maintainability analysis: Maintenance tasks that might parameters, yields
be necessary (failure diagnosis and repair, and PM tasks Stress: temperature, vibration, etc. Accel. test
such as lubrication, cleaning, replenishment, calibration, Effects on reliability/durability Accel. test
etc.), must all be analyzed to estimate the effort involved Several uncertain variables DoE
and to ensure that they can be performed correctly by the Not enough items available for DoE Accel. test
Not enough time available for DoE Accel. test
people likely to be involved.
Note: These are by no means clear-cut criteria, and there will often be
Conclusions: Since the analytical methods indicate the shades of gray between them. We must decide on the most appropriate
method or combination of methods in relation to all of the factors: risks,
aspects of designs that are critical or might present risks, knowledge, costs, and time.
they provide an excellent basis for optimizing development
and manufacturing tests. Therefore, the results of the anal- relation to expected variations. The design process is explic-
yses should be used to help plan and prioritize the tests, itly considered to include the design of the production
and the engineers involved should be part of the test team. methods and their control.
R&M design analysis methods are described in Statistical experimental methods of optimization in engi-
O’Connor (2002). Information on the most popular soft- neering design can be effective and economic. They can
ware for design analysis, with Internet contact information, provide higher levels of optimization and better understand-
is provided on the homepage of O’Connor (2002) see also ing of the effects of variables than is possible with purely
http://www.pat-oconnor.co.uk/practicalreliability.htm. deterministic approaches, when the effects are difficult to
calculate or are caused by interactions.
4 TESTING FOR R&M
4.3 Accelerated test or DoE?
4.1 Reliability testing
Accelerated test and DOE are complementary approaches in
Generally, functional and compliance testing of a new development testing for reliability. Table 1 (from O’Connor
design needs to be performed only once, or maybe a small (2002)) gives some guidance on which approach to select
number of times or on a small number of items. for particular situations.
However, reliability/durability/variability testing usually
involves multiple tests and test items, because of the wider
uncertainty associated with these aspects. Another impor- 4.4 Maintainability test
tant difference is that we do not deliberately plan to cause
failures in the first category, but we should do so in the sec- Testing for maintainability involves the timed performance
ond, using the methods of accelerated testing, as explained of specified tasks. The maintainability parameters can
in Article 214, Accelerated Testing, Volume 3. be estimated using weighted averages, based upon the
expected rates of occurrence of the tasks. See O’Connor
(2002).
4.2 Statistical design of experiments DOE
In a statistically designed experiment, a range of tests is 5 RELIABILITY DATA ANALYSIS
performed in which every variable is set at its expected
high and low values (and possibly also at intermediate 5.1 Reliability demonstration
values), and the effects are recorded for each set of values.
The results are analyzed using the technique of analysis Reliability demonstration testing (RDT) is based upon the
of variance (ANOVA), which determines the magnitude method of probability ratio sequential testing (PRST).
of the effect of each variable, and of the interactions The equipment being evaluated is operated for a period
between variables. of time expressed as multiples of the specified MTBF,
Taguchi developed a framework for statistical DOE and as failures occur they are plotted against the operating
adapted to the particular requirements of engineering de- time. Testing is continued until the staircase plot of failures
sign, with the objective of making the design robust in versus time crosses a decision line. The decision lines are
Reliability and Maintainability 643
constructed from the test criteria that have been decided and statistical interpretations, instead of concentrating
beforehand. These are as follows: on the engineering.
• The upper test MTBF, θ0 , the level considered to
be acceptable. 5.2 Reliability growth monitoring, RGM
• The lower test MTBF, θ1 , the agreed minimum value
to be demonstrated. Reliability growth monitoring (RGM) is the term used
• The design ratio, d = θ0 /θ1 . to cover a range of statistical methods that have been
• The decision risks, expressed as percentages. The pro- developed for comparing the MTBF measured during tests
ducer’s risk, α, is the probability that equipment with an or in service with targets or requirements. The best known
actual MTBF that is higher than required to pass, will is the Duane method. The cumulative MTBF, measured
by chance fail. The supplier’s risk, β, is the opposite. at intervals, is plotted and compared with the target or
The tests are required to be performed using operating required value, using logarithmic scales. The slope gives an
and environmental conditions that represent expected in- indication of the rate of MTBF growth, and extrapolations
service conditions. indicate when the target or requirement might be achieved.
The method is the basis for US MIL-HDBK-781 and Figure 5 shows an example.
other standards, as described in O’Connor (2002). Figure 4 These methods are subject to most of the same criticisms
shows an example of a PRST plot. that apply to reliability demonstration tests. RGM should
Care should be exercised when considering its use. It never be used for monitoring reliability growth during
should not be used during development testing for the development testing, since the objective in this phase should
following reasons: be to generate failures as quickly as possible so that
real reliability growth can be achieved by identifying and
• It implies that all failures are of equal importance, so implementing improvements. It can be used as part of an in-
that simple summation can provide the main criterion. service monitoring process, provided that it is supplemented
• It implies that all failures occur at constant average rates by other methods that can help to identify failure causes
in time, in particular, that there are no significant wear and priorities.
out failure modes which could cause increasing numbers
of failures later.
• It ignores action taken to correct causes of failures that 5.3 Trend analysis
have occurred.
• It ignores failures that have NOT occurred, but that Times to failure of items on test or in service can be ana-
might occur in future. lyzed using a range of methods, to determine the underlying
• By being based upon simulated instead of accelerated trends. The most useful of these are
stress conditions, it does not effectively or econom-
ically stimulate failures to identify opportunities for • probability plots and hazard plots,
improvement, as described in Article 214, Accelerated • time series analysis,
Testing, Volume 3.
• It generates an attitude that discourages the occur-
rence of failures, since the objective is to pass the
test, rather than to generate improvement opportuni- , q1
MTBF
ties. The parties concerned argue about aspects such ous
ane
tant
as whether or not particular failures should be counted Ins , qc
TBF
lative M
log MTBF
u
Cum
e
ct lin
Reje Truncation
tan−1 α
Failures
s ting lines qo
ue te
Contin
e
pt lin
Acce
Test time (multiples of specified MTBF) To log T
Figure 4. Probability ratio sequential test (PRST). Figure 5. Duane reliability growth plot.
644 Measuring System Design
• proportional hazards analysis, Therefore safety and reliability effort and management
• M(t) method. should be integrated. See Article 99, Safety Organiza-
tion, Volume 2; Article 100, Safety Analysis Methods,
These methods are described in more detail in O’Connor Volume 2; Article 101, Assessing and Demonstrating
(2002). Computer software is available for performing the Safety, Volume 2 and O’Connor (2002).
analysis methods described. Software is listed in
the book’s homepage – http://www.pat-oconnor.co.uk/
practicalreliability.htm.
RELATED ARTICLES
Article 71, Modeling Methodology, Volume 2; Arti-
6 RELATED CONCEPTS cle 99, Safety Organization, Volume 2; Article 214,
Accelerated Testing, Volume 3.
6.1 Manufacturing quality
Manufacturing quality is just as important a driver of REFERENCE
reliability as is design. Therefore, the reliability effort must
include this aspect, see O’Connor (2002). O’Connor, P.D.T. (2002) Practical Reliability Engineering,
Wiley, Chichester.
6.2 Safety
FURTHER READING
Safety and reliability are closely linked. The anal-
ysis and test methods used are mostly the same. O’Connor, P.D.T. (2001) Test Engineering, Wiley, Chichester.
99: Safety Organization
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
a commitment to good environmental practices. Continual
improvement will be achieved by implementing the Group’s
1 Definitions of Safety 645
HS&E Management System and related standards, which
2 Concepts of Safety and its Application to will include the setting of objectives and targets. As a min-
Measurement Systems 646 imum, the standards will meet relevant legal, industrial, and
3 The Safety Conscious Culture 647 other requirements. . . . etc.’
4 Responsibility for a Safe Design 647
5 Planning for Safety 648 The statement continues, over a page, to address key
aspects including how it will do this, who is responsible, the
6 Management of Safety in a Change
role of employees, its internal audit and reporting, training
Environment 649
commitment, and, at the end explains how each person
7 Communicating and Coordinating Safety
involved is responsible for implementing HS&E.
Issues 650
Safety statements cannot be formally precise in their
Related Articles 650 expression; the concept of safety is necessarily subjective.
References 650 Greater appreciation is seen by considering the conse-
Further Reading 650 quences of failure of a system that leads to death, injury to
persons, and damage to property.
Safety cannot be designed into a system by dictat-
ing specific solutions but has to be developed by effec-
1 DEFINITIONS OF SAFETY tive risk estimation and consequent risk reduction through
good design.
Safety is simply defined as ‘Freedom from acceptable risk’ Of the many risk factors that can be attributed to a
(Railtrack, 2000). system, it is those that cause catastrophic failure that are
This is a high level intention that needs elaboration. most closely associated with getting the safety aspects right.
What the concept means to an organization that designs The design and operation of measuring systems need
and/or operates a system is better understood from a typical safety consideration just as much as the larger systems
Health, Safety, and Environment Policy statement, the one aspects do.
chosen here being that of Rolls-Royce (2003), the designer Issues affecting safety are not proportionate to the cost
and maker of aircraft engines, nuclear reactors, and other or size of a component or subsystem. A critical element
highly sophisticated engineering equipment that need the of any size and cost, such as a small hole in a test valve,
highest levels of safety in their manufacture and operation: could be the cause of a catastrophic failure, as it was for
an early submarine that sank on its maiden voyage losing
‘The Rolls-Royce Group will make proper provision for the
health, safety and welfare of its people, visitors and con- all those aboard her when the hole, being filled solidly with
tractors and those in the community who may be affected paint, did not show that the torpedo tube was open to the
by its activities. It will care for the environment through sea when the test valve was opened.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
646 Measuring System Design
Measuring systems are too often seen as a minor part • acceptably unsafe (this is the usual principle applied).
of the larger systems in which they are placed. They are, Benefits outweigh the risks is a criterion; different
however, as critical as other subsystems in themselves, groups of people have different views on what this is. It
but further, they can be part of a critical safety assess- is usually enshrined in the law of the country involved.
ment system such as a smoke alarm in an aircraft engine For example, the UK law is the Health and Safety at
compartment. Work (H&SW) Act 1974.)
First, to state it as it is. Safety could only be totally
assured if absolutely every aspect and future circumstance A large systems development organization will certainly
were known and fully understood in advance. The prac- have its own policies and practices in place for safety
tical reality is that even the best design effort might still aspects of systems under its development. Smaller orga-
miss making a critical assumption. The safest designs are nizations may not have their own safety policy and may
those that have their safety and operation systematically have to rely on national laws for guidance.
addressed by very experienced persons, within a continuous What is acceptable depends on whom you ask:
improvement regime that learns and upgrades the system as In the United Kingdom, the ALARP (As Low As Rea-
an ongoing process. sonably Practicable) principle is often applied. In France,
the GAMAB (Globalement Au Moins Aussi Bon) principle
based on ‘. . . a level of risk globally at least as good as the
2 CONCEPTS OF SAFETY AND ITS one offered by any equivalent existing system’ is applied.
APPLICATION TO MEASUREMENT In Germany, they use the Minimum Endogenous Mor-
tality (MEM) principle in which ‘. . . hazards introduced by
SYSTEMS the system should not significantly increase the mortality
rate, due to technological facts, above the MEM’.
The issues involved in safe systems are many. Here are a
In the United States, numerous safety principles state-
few of the salient ones:
ments exist. For example, the Threshold Level Value (TLV)
Possibly the first question to be resolved is where does
used by the US Government Conference of Hygienists as
safety start in a design? Can it be added at some later
the ‘safe’ level states it is the
date when the system and its use are better understood?
At what stage of the system development life cycle should ‘. . . airborne concentration of a material to which nearly
it be addressed? all persons can be exposed day after day without adverse
The answer here is categorically as soon as develop- effects’.
ment commences.
It will only be fulfilled well enough if seen as a ‘whole Note that all of these statements are subjective – it is not
of life cycle’ issue. It is a matter for all to take most possible to define safety in a strictly formal way.
seriously. It cannot all be left to the person designated The public has very strong views on safety risk, but
to take responsibility for setting it in place. The whole again with much subjectivity being apparent. The public
organization must take it seriously and develop a culture perception of safety risk must be taken into account to suit
that makes designers constantly conscious of the need to the area of application of the measuring system.
assure they create safe systems. For example, in rail travel, the general public will tolerate
Key issues of system safety are many single fatality accidents much more readily than
a single multiple fatality incident. The level of safety
• the definition of safety,
acceptable for road travel is way below that for air travel.
• safety acceptance levels,
A sensor system built to protect a small electric hand
• hazards, accidents, and causal factors,
tool from overheating can use a more relaxed approach to
• organizational fundamentals,
safety design than a sensor used to detect overheating of
• management considerations,
an aircraft engine.
• standards of safety and safety assessment,
It also depends on whose safety is threatened; some
• safety planning.
groups are more tolerant than others:
Safety as a concept has many interpretations (Mugridge,
2002) such as • Armed forces: becoming more safety conscious for its
friendly users but a hard area to make safe under
• totally zero rate of dangerous defects (a nice idea, but all conditions.
impractical), • Workers in industry: now much more safety conscious
• absence of all danger (‘cotton wool’ mentality), than at prior times.
Safety Organization 647
• Civilians, for example, commuters: often seen as sec- • effective communication and coordination of safety
ondary to the system they use but are equally in need related issues to all who should know;
of safety as are those who build or run it. • ability by those who need to know of how to develop
• Children: sentimental views easily come to the fore a safety case for a system in terms of requirements,
more than for the above classes. purposes, and detailed engineering;
• understanding and application of relevant standards;
It also depends on what type of system application
• ability to plan safety into a system.
the company design work is directed toward. If it is for
electronic games equipment, then it will not need the same
level of attention as it does for sensors to be placed to
4 RESPONSIBILITY FOR A SAFE DESIGN
monitor a nuclear reactor vessel.
More detail on safety is provided in Brauer (1994), Ham-
Even though the law will point to one, or a few, per-
mer (1993), McPartland and McPartland (2002), Roland
sons as being finally responsible, safety is not one person’s
and Moriarty (1990) and Roughton and Mercurio (2002).
responsibility alone. To reach good safety levels in the
Web sites exist where safety information and group activi-
development of a project requires the operation of an effec-
ties are available, Safety US (2003).
tive safety organization. Different types of responsibility
A useful guide to the issue of safety engineering is
exist. They need to be defined, recorded, and formally
Railtrack (2000). Although compiled for the UK railway
accepted by signed agreements. Delegation of safety duties
industry, it is written as a general guide and in a language
can be made – but not the responsibility.
that is easily understood.
A key organizational need is for some form of authority
When preparing a measuring system design, however, it
such as a Board of Safety Management (BSM). This group
is at the least necessary to adhere to the safety standards in
needs to have excellent communication with all necessary
force for the project.
parts of the organization, its contractors, and its system
With so many variations being in use, it is necessary
users. After a tragedy occurs is not when such a board
to adhere to those defined as being needed by the system
should be set up – as is seen in the press so often – but it
development in which the measuring systems are placed.
should be set up as soon as the project begins. If it does
It is also not usually feasible to cater to all standards
not have one, it should immediately be formed.
at once, so the measuring system design built for one
Responsibilities suggested in Railtrack (2000) are
application may not be deemed satisfactory for another.
• setting, maintaining, and monitoring a safety policy;
• ensuring that the ESM is effectively implemented
3 THE SAFETY CONSCIOUS CULTURE and maintained;
• agreeing to safety classifications for projects;
‘Safety’ then, is much more than making sure a design is • endorsing key safety documentation;
safe in the engineering sense and that it will be used safely. • monitoring ESM performed on each project;
Engineering Safety Management (ESM) is a systemic (all • appointing a system of auditors and assessors.
encompassing) and systematic (follows rules) process and a
state of awareness that collectively leads to a system being Allocation of safety responsibility is a serious task.
sufficiently safe in its design and in its likely applications. It should be done in a clearly traceable manner, which
For the safety culture to develop, several important usually implies a top-down military type line-management
aspects need attention. These are structure.
Everyone should have access to that structural informa-
• understanding of the nature and language of safety; tion in order to know who is responsible for what.
• clear expression of safety responsibilities; Delegating safety responsibility to outside organizations
• existence of a sound safety culture; and individuals can be fraught with weakness if not firmly
• competence in safety issues by staff and users; handled. As people get removed from the safety structure
• maintaining that culture with all concerned including set up, the messages about being safety conscious atten-
contractors and users; uate with regard to the whole, but can enhance for the
• tight management of system changes to ensure safety individual concerned.
levels are not compromised as systems are repaired and When commissioning systems, all the necessary docu-
upgraded (there was the instance of an explosion-proof mentation and data on safety needs to be passed into a
instrumentation enclosure being left open after some sound depository and a clearly capable safety organization
maintenance was done!); must exist for that system.
648 Measuring System Design
More detail is to be found in Railtrack (2000) and for the the major parts of the safety plan are formed as the project
various safety standards – see Article 100, Safety Analysis gets better defined. Safety issues can have a great impact
Methods, Volume 2. on engineering design.
It is recommended that once the Preliminary Safety Plan
is ready, it should be submitted to the safety authority that
5 PLANNING FOR SAFETY would have been set up by this time.
The next major activity is to create a Hazard Log. This
Some of the task of planning will already be apparent. log records details of expected hazards on the basis of
As the Systems Engineering life cycle is traversed – see relevant past experience for a newly started development
Article 66, Phases of System Life Cycle, Volume 2, a and as they arise in an ongoing one. Not only is this a
safety system must be running alongside as a whole life source design data but it is also an important part of the
activity. necessary burden of proof that will invariably be needed to
It is rarely possible to economically bootstrap a project defend any legal actions that may result.
with safety systems, that option getting increasingly harder With this log in place, it is possible to conduct var-
as the life cycle stages move on to production. At the ious forms of safety analysis – see Article 100, Safety
production stage, the cost of redesign and rework just gets Analysis Methods, Volume 2 and Article 101, Assessing
too great to bear. and Demonstrating Safety, Volume 2. This stage is often
Figure 1 shows a suggested framework of plans and their called the Hazard Identification and Analysis (HIA) stage;
timings (Railtrack, 2000). it permits the all-important Risk Analysis to be done.
For a new development, or to upgrade, the final Safety Risk analysis allows Safety Requirements to be prepared
Plan usually has to pass through a Preliminary Safety Plan as part of the overall system requirements engineering task.
stage for, at the Concept Stage, Article 67, Principle of The required full Safety Plan can then be prepared and
Concept of Operations (ConOps), Volume 2, things are authorized to be placed alongside the System Engineering
still far too fluid to start to dictate issues of detail. However, Plan, and the Test and Evaluation Plan.
Endorsements by Project Tasks
Safety Board
Prepare Preliminary
Safety Plan
Establish Hazard Log
Preliminary Safety
Plan Identify and analyze
hazards
Assess risk
Establish Safety
Requirements
Safety Requirements
Prepare Safety Plan
Independent Safety
Safety Authority
Safety Plan Assessor
Implement Safety
Plan
Project
Perform safety
Commission safety assessment
assessment
Issue Safety
Prepare Safety Case Assessment report
End
Safety Case
Safety Approval granted
Customer
Transfer safety
End End responsibility
End
Figure 1. Suggested framework of safety activities. (Reproduced from Railtrack (2000) Yellow Book – 3, Railtrack PLC, Euston Square,
London, distributed by Praxis Critical Systems Ltd, 20 Manvers Street, Bath BA1 1PX, UK, pp. 1–4, Fig 1-1 of Yellow Book.)
Safety Organization 649
Once all this has been done, it provides the substance A Change Management Committee, or the like, is needed.
needed for a Safety Case to be prepared, which, after receiv- They review proposals for change that is put to them. No
ing endorsement, allows Safety Approval to be granted. change is permitted without correct approval.
This may seem to be a large overhead on a project. Its All changes are accounted for within the configuration
extent is tailored to suit the type and size of project. The management process; each has to have its impact on safety
driving force for doing it well – and beyond just profes- assessed. This process needs to record
sional responsibility – is the fear of legal repercussions. A
catastrophic failure can lead to very large legal settlements • identification of versions,
and perhaps to imprisonment. Even a low level breach of • history and status of versions,
safety could lead to product recall for a simple modifica- • parts forming items,
tion, a process that has the potential to bankrupt a company • relationships between items.
due to the cost of the recall and fixes, without the cost of
legal action. Figure 2 is a change proposal management process sug-
gested in Sage and Rouse (1999). The Engineering Change
Proposal (ECP) is created and subsequently put to the
Change Control Board (CCB) for approval before it can
6 MANAGEMENT OF SAFETY IN A be put to use.
Software tools are available to assist the task of config-
CHANGE ENVIRONMENT
uration management, but its use will need safety approval
in safety critical systems developments.
Safety is not an issue to be addressed at the start of a Lack of sufficient safety is an insidious parameter for it
development and then ignored. is usually not one event in isolation that leads to failure
Each time a design change is made, its impact on safety but a complex chain of repercussions that lead, often
needs to be investigated to see if the level of safety reached unexpectedly, to failure. Reasons become blatantly obvious
has been compromised. This is not always easy to establish. only after the event!
Each engineering change has to be assessed via the Somewhere, in one or more places in a chain, unpredicted
change management process adopted for the development. events arise. Once they have been seen – usually too late
That adding safety features will change the level of safety at the first occurrence – they can be avoided. Not knowing
is obvious; but so also can removing them. about them leaves the situation waiting for that particular
Awareness of change
Classification of change impact
Create engng change prop ECP
Evaluation of ECP by tech staff
Yes
Emergency ECP? Awareness of change
No
CCB evaluation and approval
No CCB
Sent to tech staff for
further evaluation approval?
Yes
Implementation and verification
Figure 2. Change proposal management process. (Reproduced from Handbook of Systems Engineering and Management, Sage and
Rouse, Copyright (1999), Reprinted with permission of John Wiley & Sons, Inc.)
650 Measuring System Design
set of circumstances to arise. Because something has never resolved by this stage. Use all forms of communication,
happened before, does not demonstrate it never will. especially those with formal records once the issue is
The software component of safety critical systems is seen to need more central action. Formal records – the
particularly risky and has to have far more effort expended auditable paper trail – tend to be heeded more than dis-
on its development than do run-of-the-mill applications. cussion does.
In these cases, even the tools used to check the software It is to be expected that organizations will naturally set
have to be understood to the point at which their standard up liaison groups between their respective safety boards.
of operation will not compromise safety levels by, for As safety issues are risks, the use of the risk control
instance, not finding potential faults. For this reason, safety register should be the database and control mechanism
critical system developers will often build their own safety for safety.
examination tools. Formal committee type units usually have associated
The safest system could well be the one that is not used! with them long delays in getting issues resolved. As with
It is not possible to build a useful system that is perfectly any committee of importance, there needs to be an exec-
safe. The cost to provide it would not be practical. Such a utive group who can act rapidly when safety issues need
system would be so overburdened with safety features as to urgent attention.
be not workable. Safety then is a gradually improved facet
of a systems development.
Swapping of safety cases and data is recommended as RELATED ARTICLES
that increases the pool of event situations; but organizations
are not always willing to do this because of legal and media Article 63, Outline of Systems Thinking, Volume 2;
ramifications of things they may wish to keep low key. Article 98, Reliability and Maintainability, Volume 2;
Article 100, Safety Analysis Methods, Volume 2; Arti-
cle 101, Assessing and Demonstrating Safety, Volume 2;
7 COMMUNICATING AND Article 102, Introduction to the Legal Process, Vol-
COORDINATING SAFETY ISSUES ume 2.
Given that a project development often involves many
people who would normally never meet for discussion of REFERENCES
common problems, it is necessary to ensure that issues
Brauer, R.L. (1994) Safety and Health for Engineers, John Wiley,
that arise, no matter what the source, are communicated New York.
to whoever needs to know.
Hammer, W. (1993) Product Safety Management and Engineer-
This can be a difficult issue to address, as people are ing, American Society of Safety Engineers.
often loath to be critical of the work of others, or to offer
McPartland, J.F. and McPartland, B.J. (2002) McGraw-Hill’s
too much about aspects of their own work that could be National Electrical Code Handbook, McGraw-Hill/TAB Elec-
construed as a weakness on their part. Good, effective tronics,.
communication is essential between all involved. Mugridge, C. (2002) Notes on Safety for a Short Course, Private
Usually, the project will operate under many mandatory Communication, Witchwood Engineering, UK.
issues. If so, one will assuredly be that all persons have Railtrack (2000) Yellow Book – 3, Railtrack PLC, Euston Square,
the obligation to report safety matters to a central place London, distributed by Praxis Critical Systems Ltd, 20 Manvers
for action. All parties involved, including contractors, are Street, Bath BA1 1PX, UK.
subject to this ruling. An example is found in the UK Roland, H.E. and Moriarty, B. (1990) System Safety Engineering
Engineering Council ‘Code of Professional Practice on Risk and Management, Wiley-Interscience, New York.
Issues’. In essence, it states the following: Roughton, J.E. and Mercurio, M.J. (2002) Developing an Effec-
tive Safety Culture: A Leadership Approach, Butterworth-
• Communicate up and down the chain of responsibility. Heinemann.
• Seek to raise awareness of potential hazards and risks Sage, A.G. and William B.R. (eds) (1999) Handbook of Systems
with colleagues. Engineering and Management, Wiley, New York.
• Make clear limitations and implications of risks.
• Discuss near misses and incidents with peers and take
action on findings. FURTHER READING
Commence communication verbally and look for a Safety US (2003) http://www.quality.org/config/cm-guide.html
sound, simple solution. Many issues would have been Safety Groups in US.
100: Safety Analysis Methods
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
established by the use of a suitable safety standard that
1 Some Terms and Definitions Commonly Used is appropriate for the project.
in Safety Design and Evaluation 651 Some key safety standards are
2 Standards and Other Information on Safety 651
• DEFSTAN 00-56
3 Overview of Safety Assessment Methods 652 • IEC 61508
4 Work Safety Assessment (WSA) 653 • CENELEC EN50126
5 Failure Mode and Effects Analysis (FMEA) 653
All of the above are useful, but can be a little difficult
6 Failure Mode and Effects and Criticality to come to grips with. For overall guidance, each has its
Analysis (FMECA) 654 own kind of explanations. The defense safety standards
7 Fault Tree Analysis (FTA) 655 are written in a directive tone with numerous updates and
8 Hazard and Operability Analysis (HAZOP) 655 revisions to be taken into account as they are used; these are
Related Articles 657 not always easy to read and maybe difficult to appreciate.
References 657 Several texts on safety are available – see Brauer (1994),
Railtrack (2000), Roland and Moriarty (1990) and Rough-
ton and Mercurio (2002).
Information given in texts should not be applied to the
1 SOME TERMS AND DEFINITIONS letter where definite standards are clearly defined to be
COMMONLY USED IN SAFETY DESIGN used in a development. The texts should be viewed as
AND EVALUATION useful reference works and are essential for building up
one’s personal safety attitude and design experience. The
Many terms will be encountered in the safety aspects of difficulty is that whereas they might maintain adherence to
designing and operating a system. No single definitive set one standard, they, more often than not, pick up terms and
is used; several exist. Where a safety standard is used for a ideas from several sources and can also use the author’s
project, the terms used there should be adhered to. Table 1 own terminology from time to time.
provides a composite set for use where they are not defined One useful, Web accessed, general introduction and guide
elsewhere for a project. that was compiled recently for use by a very wide range
of civilian users – as are found in a large railway system,
is Railtrack (2003) – known as the ‘Yellow Book’. Its fea-
2 STANDARDS AND OTHER tures are as follows:
INFORMATION ON SAFETY • Prepared only as a guide to concepts and methods.
• Developed for use within the breadth of activities of the
It is essential that terms, methods, and reporting be done railway sector.
consistently throughout a project. Consistency can be • Principles are applicable to all sectors.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
652 Measuring System Design
Table 1. Composite set of terms and definitions used in safety.
Term or acronym Meaning
Accident Unintended event that results in harm
Accident trigger Event that causes a hazard leading to an accident
ALARP As Low As Reasonably Practicable
BSM Board of Safety Management
Causal factor State of system that is systematic, for which a cause is clearly defined – once
established
CCB Change Control Board
CENELEC EN50126 Electrical Industry Standard on safety
CMC Change Management Committee
DEFSTAN 00–56 UK Ministry of Defence. Standard on safety
ECM Engineering Change Proposal
ESM Engineering Safety Management
ETA Event Tree Analysis
Failure Deviation from intended performance, usually in a catastrophic manner
FTA Fault Tree Analysis
FMEA Failure Mode and Effects Analysis
FMECA Failure Mode and Effects and Criticality Analysis
GAMAB Globalement Au Moins Aussi Bon
H&SW Health and Safety at Work Act of UK
Hazard log Record of hazards and potential accidents identified from analysis of safety
HAZOP Hazard and Operability Analysis
HIA Hazard Identification and Analysis
HS&E Health, Safety, and Environment
IEC 61508 International Electrotechnical Commission safety standard
Incident Event that could have safety significance
Individual risk Risk experienced by the individual. Usually expressed as probability of fatality
per unit time
Intolerable risk Risk that is totally unacceptable
MEM Minimum Endogenous Mortality
MIL-STD-882C/D US Defense Department Standard on safety
PSP Preliminary Safety Plan
Risk analysis Analysis of system risk, by various methods
Safety Freedom from unacceptable risk
Safety Approval Process through which the Safety Board grants approval to make changes
Safety Audit Activity to verify that project is following the Safety Plan and maybe the
veracity of that plan
Safety Case Formal presentation aiming to provide assurance system will be safe and fit for
purpose
Safety certificate Formal statement authorizing the system for use
Safety integrity level SIL. Discrete level, usually from 5, for specifying safety integrity zero being
the lowest level where safety is not relied upon
Safety Plan Documented plan that details how the system will be made sufficiently safe
Safety Requirements Requirements needed to make a system safe under the definitions that apply
TLV Threshold Level Value
Value of Preventing a Fatality VPF. Defined monetary figure used to indicate reasonable expenditure for
preventing a single fatality
WSA Work Safety Assessment
• Developed by a committee of Railtrack employees and 3 OVERVIEW OF SAFETY ASSESSMENT
supported by the committee as versions with updates METHODS
provided on the Web site, Railtrack (2003).
Failure of a system can cause death, injury, and property
It is important to recognize that it is all too easy to spread damage to a product or system. Examples are loss to an
unsafe practices by sloppy use of terms due to the semantic individual and the more geared-up situations where, say,
nature of the necessarily subjective statements and varied a ship and its passengers are affected. It can also cause
use of terms. destruction of facilities and the environment.
Safety Analysis Methods 653
Safety standards are essential as they contribute to safety methodology. Whichever method is adopted, it must
management by providing generic means to design safe be used under advice to prevent shortcomings arising
systems: they do not specify design solutions. from differences in use of terms, process steps, and
Methods deployed must be acceptable to all involved documentation.
and cannot be imposed without user ownership across There is a particularly strong need for a tightly specified,
the system. hardware and software codesign: the simplest of changes
To prevent or minimize impact of human error, make use to software can negate prior safety analyses so regression
of key mechanisms like for safety assessment can be very expensive to carry out.
Several methods of safety analysis are now introduced
• adoption of fail safe, or not fail at all, operations, to set the scene. For application, these short overviews will
activities, and events; need expansion using the appropriate standards in force for
• redundant system units; a project and the texts referred to above.
• fault-tolerant (error tolerant) design of hardware and
computing;
• preventing unsafe actions by the way the system is 4 WORK SAFETY ASSESSMENT (WSA)
designed and operated;
• providing assistance with making key decisions on a Work Safety Analysis (WSA) is a simple means for system-
safe basis; atically identifying hazards and potential corrective mea-
• using comprehensive integration and testing procedures. sures, Sage and Rouse (1999). It is used mainly for physical
operational work by people, but its relatively basic method-
Some key characteristics of failures are ology has a place in safety assessments where deep detail
is not required.
• failure of individual small components and subassem-
This is a hierarchical task analysis in which the hazards
blies are relatively easily controlled with low risk – that
for each step are identified; once these are established, their
is, high confidence, by allocation of more resource;
likely causal effects are identified.
• many disastrous failures have been the result of a series
It is predominantly based on the judgment of an expert(s),
of apparently unimportant events that line up to create
plus considerable relevant data from past studies. It is often
a pathway through the whole system, which adds up to
found that a lay person, one unfamiliar with the system, can
major failure – like it did in the Challenger disaster and
often ‘see’ additional hazards that experts do not!
some Mars explorer missions;
Likely issues are best handled using a checklist to ensure
• Reason’s ‘Swiss cheese’ model points out that the ‘holes
things already known from prior histories are not over-
in the cheese need to line up for both good and bad
looked. Checklists for use by safety assessors, auditors, and
actions to result’.
designers are provided in Railtrack (2000, 2003).
Sources of safety requirements can be from a Regulatory Once established, each hazard is classified in terms of its
Authority or be derived from assessment and analysis. relative likelihood of occurrence on a 0 (hazard eliminated)
Types of safety requirement are many such as to 5 (very serious) scale and its severity of consequence.
Relative risk is then calculated as the product of proba-
• failure probability target – random or systematic; bility and consequence.
• hazard removal requirement; Once the risk table has been reduced by calculation it
• functional safety requirement; will become clear where corrective measures can be taken
• process requirement. to reduce the salient hazards to acceptable levels.
WSA is extended in different ways, as follows.
Safety requirements must be
• auditable
• testable
5 FAILURE MODE AND EFFECTS
• reasonable ANALYSIS (FMEA)
• unambiguous
• uniquely identified. Failure Mode and Effects Analysis (FMEA) is another
commonly used method. It has the same basic steps as
Safety can be regarded as fitting into two groups of WSA but is made more objective by use of the methods
interest – safe technical behavior and safe human operation. of reliability analysis. It too is used to establish potential
Several well-developed and documented methods exist systems failures that are then analyzed to establish the
for establishing safety risk. All are similar in their safety regime and make improvements.
654 Measuring System Design
This method was developed from hardware reliability • Define the blocks using a clear numbering system such
engineering and thus there exist many similarities between as that commonly used in functional flow diagrams
safety and reliability analysis; this is to be expected for both (FFD) used in system functional decomposition.
are for investigating the risk of failure. • Locate the system requirements documents.
It tabulates the percentages of failures by the mode of • Define and record failure modes and mechanisms using
failure and then categorizes these as critical or noncritical. a checklist that defines them along with the relevant
It leads to event trees. environmental parameters to be applied.
As with most safety assessment methods, it is most • Work on all potential failures found suggesting cor-
effective if used early in a development to flush out likely rective actions that avoid them or sufficiently attenu-
failures at the time when they can be avoided at lesser cost ate them
by design or operational correction. • List any special conditions of use that are essential for
An FMEA study is also of value to setting up the logistics safe operation.
of through life maintenance as it determines what needs to
Traceability is essential for all of the above process doc-
be, or not to be, done when servicing the system.
umentation for any open link may leave issues unaddressed
Carrying out an FMEA study can be done on either
or see rekeyed starts using different semantic appreciations.
hardware, or a functional basis. The hardware approach is
only practical once sufficient design drawings and analyses
have become available – that is, in the later development 6 FAILURE MODE AND EFFECTS AND
life cycle stages. It can be carried out with various levels
of system subassembly. CRITICALITY ANALYSIS (FMECA)
In the functional approach, the system is studied using
the system functional decomposition breakdown levels. It Failure Mode, Effects, and Criticality Analysis (FMECA) is
starts studies of likely failure at the top level, working a similar process to the FMEA in that it is for identifying
down to the level of detail needed. It is used to precede and investigating the safety and failure spots of potential
a hardware study. systems designs.
It is suggested, extending the list given in Sage and The impact of unsafe situations will vary from low to
extreme. FMECA extends FMEA to take in the critical-
Rouse (1990), that an FMEA be carried out with the
ity factors.
following steps:
It is mainly applicable in the concept and preliminary
design stages of the development life cycle for preventing
• Use the functional or hardware approach according to unsafe events but it can also be used to study the safety of
the stages of development available. existing designs.
Define system requirements
Accomplish functional analysis
Accomplish requirements allocation
Identify failure modes
Determine causes of failure
Determine effects of failure
Identify failure detection means
Rate failure mode severity
Rate failure mode frequency
Rate failure mode detection probability
Analyze failure mode critically
Feedback and correction
Figure 1. Steps of a typical FMECA analysis. Reproduced by permission of Pearson.
Safety Analysis Methods 655
Raw material Equipment Environment
Operating temperature
Lube and cleaning
Dust particles
Cleaning Material composition Machine vibration
Set up
Raw material
Tooling Magnetic/electric field
preparation
Product
Tolerances Set up procedure
Handling
requirements
Skill level required
Raw material Procedures
Figure 2. Fish bone diagram example: identifying faults in a manufacturing process. Reproduced by permission of Pearson.
The steps of a typical FMECA study are shown in Undeveloped Logic AND
Top level
Figure 1. More detail of each step is available applied to event gate
event
an example in Blanchard and Fabryky (1998).
An essential difference between FMECA and FMEA
studies is that the former allocates tighter understanding Intermediate
of the failure data using statistical methods. This allows a fault event Logic OR
Input event gate
more discerning study of the criticalness to be produced
where many parameters are involved.
Lowest failure
event
7 FAULT TREE ANALYSIS (FTA)
Figure 3. Some symbols used in fault tree construction.
Fault Tree Analysis (FTA) studies failure, and thus safety,
using deductive use of a graphical enumeration and anal- the Blanchard and Fabryky (1998) example is given in
yses of the different ways in which a fault can arise. Its Figure 2.
methodology can be applied in the above methods to tease Fault trees are constructed using commonly agreed stan-
out potential safety problems. It is a method for charting dard symbols – see Figure 3 for the basic ones used. A
the hazards in a hierarchical manner. more complete list, with explanations of each, is avail-
Its process is similar to the widely applied decision tree able on http://www.relexsoftware.com/reliability/brief/fta
method in that it is a top down procedure that identifies brief2.asp.
how issues can be broken down. Here it is used to locate An example Fault tree for events that can lead to failure
causal reasons for failure. in an elevator, drawn with the Reflex Fault Tree tool, is
As causes of potential failure are determined they branch given in Figure 4.
out with probabilities of each arising being allocated at each
branch point. Once the full tree has been identified, each
of the many paths is calculated back to the start to identify 8 HAZARD AND OPERABILITY
each risk, thus revealing the most unsafe sets of circum-
stances. High-risk paths are rectified by redesign to obtain ANALYSIS (HAZOP)
a more suitable risk distribution across the whole project.
In analysis of the safety-regime application of the deci- Hazard and Operability Analysis (HAZOP) was developed
sion tree concept, use is often made of the Ishikawa cause in the 1960s for the UK chemical industry. It was quoted
and effect or the fish bone diagram. The one used in as being a
656 Measuring System Design
Passenger injury occurs
in elevator
Top event
Q:0.0690691
Box free falls Door open error
Box fall Door open
Q:0.0679412 Q:0.00121007
Cable off pulley Holding brake is failed Broken cable Door close failure Box not at level
Event1 Holding brake Cable Gate5 Gate6
Q:0.00995017 Q:0.0438203 Q:0.0154297 Q:0.00989076 Q:0.122344
No holding brake Motor tums free Latch failure Controller failure Cable slips Box early or late stop
Gate7 Gate8 Event10 Event11 Event13 Event12
Q:0.157178 Q:0.278793 Q:0.00249688 Q:0.00741239 Q:0.0054849 Q:0.117503
Wom friction material Stuck brake solenoid
Event3 Event4
Q:0.108634 Q:0.0544609
Figure 4. Example fault tree created with the Reflex Fault Tree tool.
‘Qualitative technique whose purpose is to identify all • PART OF: Some qualitative decrease compared to what
deviations from the design’s expected operation and all was intended.
hazards associated with these deviations’
• REVERSE: Logical opposite of intention.
• OTHER THAN: Something completely different.
It is a systematic, creative examination of a design that is
best performed by a multidisciplinary team who can bring • EARLY: Something happens before it is expected, rel-
a wide variety of viewpoints to the study. ative to clock time.
It has been widely adapted for use in other sectors. • LATE: Something happens after it is expected, relative
In use, each system component is inspected in turn keep- to clock time.
ing in mind the design intention – what is this component • BEFORE: Something happens before it is expected,
supposed to do? relating to order or sequence.
Standardized words are used to minimize the semantic • AFTER: Something happens before it is expected, relat-
nature of normal language in this use. A selection from the ing to order or sequence.
thesaurus is as follows:
• NO/NOT: No part of the intention is achieved. The first step, identifying the system elements, for a
• MORE: Some quantitative increase over what was simple example HAZOP study of a two component fuel
intended. mixer (Mugridge, 2003) is given in Figure 5.
• LESS: Some quantitative decrease compared to what From each such study is derived plausible (that is, may
was intended. well be true but not necessarily so until investigated)
• AS WELL AS: Some qualitative increase over what deviations from the design intention. Avoid prejudgment
was intended. of any issues before they are analyzed; include those that
Safety Analysis Methods 657
Tank with fuel 1 Valve 1 Pump 1
A
Overflow
AB
Fuel mixing system C
Output mixture
B Combiner
Tank with fuel 2 Valve 2 Pump 2
Figure 5. First step in an HAZOP study – the system block diagram.
Guide Word Deviation Cause Effect
NO No flow of A Pump failure Explosion
MORE Excess flow Valve failure Poor quality
LESS Reduced flow Pump/valve failure Explosion
Contamination
AS WELL Line rupture Fire
with air
REVERSE Pump reverse flow Maintenance error Explosion
OTHER THAN A solidifies Low temp Explosion
Figure 6. HAZOP table for fuel mixer system.
seem wildly unlikely. Work on the basis that all elements REFERENCES
are unsafe until proven to the contrary.
Figure 6 shows part of an analysis table formed from Blanchard, B.S. and Fabryky, W.J (1998) Systems Engineering
study of the system given in Figure 5. and Analysis, Prentice Hall, Englewood Cliffs, NJ.
It can now be seen that while each method has the Brauer, R.L. (1994) Safety and Health for Engineers, John Wiley,
same basic features, they variously incorporate features New York.
of their own. Each of the above analyses methods finds Mugridge, C. (2003) Teaching Notes for Safety Course, Witch-
use according to the Standard requirements in force and wood Engineering, UK.
the experience of the analyst. It is essential to use only Railtrack (2000) Yellow Book – 3, Railtrack PLC, Euston Square,
one methodology for all related parts of a system; mixing London, distributed by Praxis Critical Systems Ltd, 20 Manvers
methods can lead to confusion, the enemy of reliable safety Street, Bath BA1 1PX, UK.
studies. Safety is built up over years and can be destroyed Railtrack (2003) Engineering Safety Management Guide, http://
in moments by simple design or operational changes. www.yellowbook-rail.org.uk/site/the yellow book/the yellow
book.html.
Roland, H.E. and Moriarty, B. (1990) System Safety Engineering
RELATED ARTICLES and Management, Wiley-Interscience, New York.
Roughton, J.E. and Mercurio, M.J. (2002) Developing an Effec-
Article 63, Outline of Systems Thinking, Volume 2; tive Safety Culture: A Leadership Approach, Butterworth-
Article 64, Executing A Measuring System Design, Heinemann.
Volume 2; Article 99, Safety Organization, Volume 2; Sage, A.G. and Rouse, W.B. (eds) (1999) Handbook of Systems
Article 101, Assessing and Demonstrating Safety, Engineering and Management, Wiley, New York.
Volume 2.
101: Assessing and Demonstrating Safety
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
can have its integrity reduced by subtle changes in the chain
1 Safety as a Whole of Life Cycle Factor 658 of circumstances in application, by new requirements of
2 The Seven Stage Process 658 regulators, and because of the occurrence of a previously
unseen event that attracts attention. All of these can require
3 Stage 1 – Hazard Identification 659
extensive system redesign.
4 Stages 2 and 3 – Causal and Consequences The detailing designer should work within existing direc-
Analyses 659 tives aiming to second-guess failure mechanisms that are
5 Stage 4 – Loss Analysis 661 then mitigated by the use of appropriate design strategies.
6 Stage 5 – Options Analysis 661
7 Stage 6 – Impact Analysis 663
2 THE SEVEN STAGE PROCESS
8 Stage 7 – Demonstration of ALARP 663
9 Safety Process Reviews 664 The ‘Yellow Book’ of Railtrack (2000) suggests that a
Related Articles 665 seven stage process be used to demonstrate that the sys-
References 665 tem risks are as low as reasonably practicable (ALARP).
(While ALARP is used here, this basis of safety is not
the only one in use – see Article 99, Safety Organization,
Volume 2.) Notes of a short course on safety (Mugridge,
1 SAFETY AS A WHOLE OF LIFE CYCLE 2003) are summarized here to introduce the methodology
FACTOR for assessing the risk level of a system.
The steps of the Seven Stage process are shown in
Safety must be treated as a whole of system life cycle issue. Figure 1. The sequence is as follows:
Each design activity, when adding its own contribution, Stage 1. Identify the hazards.
must carry out safety assessments as well as is possible at Stage 2. Carry out an analysis of the causes of each
the time. hazard.
Overarching the individual design contributions, to Stage 3. Analyze the consequences of the hazards. Stages
ensure that they integrate and do not degrade the overall 2 and 3 can be done in either order.
system safety, will be the duty of the Safety Board with its Stage 4. Assess the loss in terms of harm to people.
safety policies and practices. Stage 5. Consider the options for risk reduction.
Applying a well executed, top-down, System Engineer- Stage 6. Assess the likely cost benefits of each hazard
ing, SE planning and project management, and working reduction measure.
within a design change management process will help in
Then, after systems adjustment to reduce the risks:
reducing rework at a later stage.
The difficulty, however, is that safety is an ongoing Stage 7. Demonstrate that the system risks are as low as
maturing process. A system that has been ‘safe’ for years reasonably practical.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Assessing and Demonstrating Safety 659
Stage 2:
Causal
analysis Implementation
Total loss Option Total loss Option Benefit Cost
cost
without with
mitigation mitigation
Stage 1: Options
Hazard Stage 7:
ID Stage 4: Stage 5: Stage 6: Demonstration
Loss analysis Options analysis Impact analysis of ALARP
Stage 3:
Consequence
analysis
Figure 1. Seven stage process for demonstrating ALARP.
Carrying out this process needs speed, but with adequate • Hazard and operability (HAZOP) studies – see
care. If the process takes too long, the system may change Article 100, Safety Analysis Methods, Volume 2.
significantly, making the analyses following the data used • Empirical approaches, while not well ordered can reveal
to perform the task no longer useful. unexpected issues.
Each step is now covered in turn. • Checklists from other projects, and the organization’s
safety documentation.
• Structured Walk-throughs of each likely, plausible
3 STAGE 1 – HAZARD IDENTIFICATION situation.
• (FMEA) Failure Mode and Effects Analysis – see
The purpose of this stage is to identify any potential hazards Article 100, Safety Analysis Methods, Volume 2.
associated with the system and the way in which it is to • Task Analysis.
be used.
For best results, the task should not be guided by intuition More details of these methods are available in Railtrack
to come up with likely hazards, but should be approached in (2000, 2003). These methods take time to do and need care
a systematic and comprehensive (that is, systemic) manner. to identify issues. Bias and preconceived ideas of improve-
Ideas and inspiration can be found from a variety of sources ments should not be allowed to prevail as potential risks
such as are uncovered. This is not the stage to decide whether their
level of importance is such that they are not worth includ-
• expert knowledge, ing. The chain effect must be remembered for even small,
• layperson experiences, apparently innocent, events that might lead to unaccept-
• previous experience for the same kind of circumstances, able risk.
and from the prior history of the project (if recorded –
all too often this has not been done!)
• new analysis of the current situation. 4 STAGES 2 AND 3 – CAUSAL AND
CONSEQUENCES ANALYSES
Some of the various techniques that can be deployed to
make use of the above sources are as follows:
The purpose of causal analysis is to determine credible
• Creative thinking by individuals including a wide cross combinations or sequences of causal factors that can lead
section of users. to hazards.
• Brainstorming that teases out ideas as members of the It is carried out using a diagrammatic, functional, hier-
team spark ideas from each other’s suggestions. They archical representation as its main schema. A fault tree
should make use of charts to record ideas trees and other is generated from a study of drawings, component lists,
connections between the different ideas. and operational instructions – see Fault Tree Analysis
660 Measuring System Design
(FTA) and Failure Modes and Effects Analysis (FMEA) Some rules of fault trees are that
in Article 100, Safety Analysis Methods, Volume 2. The
• a top event is ALWAYS fed by a gate,
probability of a condition arising is best stated as the quan-
• no gate-to-gate states can occur,
titative probability estimate.
• no event-to-event links can exist.
FTA starts with an undesired top event being identified.
Logic gates are used to form a logic network showing As well as either train causing the unwanted event, it
how the various fault-inducing conditions are related. An is also feasible that both trains can cause the unwanted
example is given in Article 100, Safety Analysis Methods, event together. In this case, one train AND the other
Volume 2. is a condition to be investigated. Figure 3 shows this
Working downwards, one identifies the next immediate condition.
causes, eventually stopping when basic events are reached. Stage 3 is a Consequence Analysis. The purpose here
Probabilities are then inserted for the basic events from is to determine what are the intermediate conditions and
an FMEA – see Article 100, Safety Analysis Methods, the final consequences resulting from the occurrence of the
Volume 2. identified hazards.
FTA is usually carried out using a software package. The It is carried out using cause consequence diagrams (CCD)
calculations can be done manually, but they are tedious. and Event Trees. An example of a CCD is given in Figure 4.
Tools must be used with great care in safety-critical situa- A fire starts with ignition. If the fire protection system is
tions, for a tool may have an inbuilt fault error of its own operating as planned, the sprinkler systems operate and the
in the way it computes the calculations or in the assump- fire is put out. If it does not operate, but the fire alarm
tions it uses. Super safe systems usually need to have their sounds, then the severity of damage caused by the fire is
own safety tools developed under the control of the safety
process in order to keep control over all the factors of a
Fire put
safety assessment. out
Minor fire Major fire
Figure 2 shows how an undesired top event (e.g. a train
collides with another train) can cause a top event. In this
Yes No
case, an OR gate is appropriate – for one train OR the other
can cause the collision.
Alarm sounds
P (A + B)
Event • Undesired top event (e.g. a train collides with
another train) Yes No
OR Sprinkler
• ALWAYS fed by a gate
– No gate to gate operates
A B – No event to event
P(A) P(B)
Mathematical expression: Ignition
P(A + B) = P(A) + P(B) − P(A)P(B)
(assuming A and B are independent)
Figure 4. Simple example of a cause consequence diagram
Figure 2. OR gate condition for causing an unwanted event. (CCD).
• Undesired top event (e.g. a train collides with
P(A.B) another train)
Event • ALWAYS fed by a gate
– No gate to gate
And – No event to event
• Next immediate causes
needed-no jumping
A B
P(A) P(B) • NB: always check for common mode
events when ANDing
Mathematical expression:
P(A.B) = P(A)P(B)
(assuming A and B are independent)
Figure 3. AND gate condition for causing an unwanted event.
Assessing and Demonstrating Safety 661
lessened. Good design for least risk sets up situations that
have less risk due to alternative protective states being in
place.
System safety is aided by the existence of safety barriers.
These are design situations that will assist in preventing the Electronic
escalation of a safety risk – a barrier mechanism will have a Detector Evaluator
junction box
barrier strength associated with it. Assessment of this value
is needed in quantitative mathematical terms.
It is important to recognize that safety rectification tem-
porarily put into place can lead to reduced safety because
of its use. A study of such regression steps is needed as
much as the main lines of expected risk.
Figure 5. Schematic of a train axle detector.
5 STAGE 4 – LOSS ANALYSIS harm to a surviving person can cost more depending on
individual legal judgments.
The purpose of carrying out a Loss Analysis is to determine The VPF is, therefore, project specific. Advice is needed
the loss associated with each consequence identified during as to what values to use; it will come from the legal fra-
Consequence Analysis. This study is done to see what the ternity. Note that these figures are used only for supporting
current situation is before rectification is implemented. It is ALARP decisions – and they are not a guide to legal set-
not the loss of the improved system. tlements.
A loss here is harm to people, ranging from slight effects To illustrate some of these principles, consider the simple
to fatalities. The principles used can, however, be adapted to example of an axle detector set up on the side of a railway
take in other losses such as environmental and commercial track, see Figure 5, to warn a railway gang working on the
but that is more for studies of risk assessment in general. track of an approaching train.
A Loss Analysis needs to estimate the extent of harm, The CCD for this element of a railway system is devel-
damage, or disruption that could potentially arise from oped in Figure 6. The various situations are self-evident
the identified consequences. Harm is initially expressed in from inspection of the tree. The failure F for the base fault
qualitative terms for loss assessment and starts from guided is decided along with the probabilities for dividing risk at
estimates, rather than from exact measures. each branching point. The latter values are used to decide
For example, if a measuring system fails and causes the failure values at the end of each branch of the tree.
its aircraft to crash, it is not possible to be exact about With the CCD now set up, it is possible to carry out
the harm to life that might arise. Guidelines commonly the calculations needed to convert the data for the various
accepted are that 1 fatality = 10 minor injuries, and 1 major severities of collision into monetary values. This is done
injury = 20 minor injuries. These subjective conversions in Figure 7 where the total cost is shown for major, minor,
allow the various levels of harm to be reduced to the and zero collisions.
common unit of fatalities per annum. Those values can then be used – Figure 8 – to calculate
Once the subjective figures are decided, the next step is the total annualized cost of the systems as it is before any
to convert them into objective values to ensure that any improvements are carried out.
further work with them has minimal added subjectivity. More details for conducting a Loss Analysis are available
This results in tabulation of the nature of the loss against in Railtrack (2000, 2003).
the cost associated with each consequence.
The next step involved is arriving at the value of pre-
venting a fatality (VPF), a value put on a human life. This 6 STAGE 5 – OPTIONS ANALYSIS
varies considerably; in the rail sector, the VPF ranges over
£1 to 3 million for a life, or its equivalent. The purpose of the Options Analysis is to ensure that all
A higher figure is needed for multifatality accidents reasonable measures for reduction of losses have been con-
for the public is disproportionably averse to this! A sidered. There are two ways to consider this – reduce the
weighting coefficient is used – called the Factor of Gross rate of occurrence of the hazard, or limit its consequences.
Disproportion. For each system hazard, it is necessary to identify phys-
Other sectors are often less open about this value. A road ically possible and known mitigation strategies and then
accident death is far less than the above, but any long-term to calculate the cost of each option. Only direct costs are
662 Measuring System Design
No Minor Major
collision F = 1.0918/yr collision F = 4.9 × 10−3/yr collision F = 3.3 × 10−3/yr
YP = 0.6 NP = 0.4
Impact low
momentum
YP = 0.75 NP = 0.25
2nd train avoids
YP = 0.9 NP = 0.1
No 2nd train
YP = 0.9 NP = 0.1
Signalman detects
track occupied
F = 1.1/yr
System shows
unoccupied when
track occupied
Figure 6. Cause consequence diagram (CCD) for axle detector system.
Consequence Losses Cost £m
Major collision 30 fatalities 115.9
50 major injuries
200 minor injuries
= 36 equivalent fatalities
Minor collision 2 major injuries 0.3
6 minor injuries
= 0.23 equivalent fatalities
No collision 2 minor injuries 0.01
(emergency application of brakes) = 0.01 equivalent fatalities
Figure 7. Cost of the consequences of collision for the axle detector system.
included here – indirect benefits arising due to mitigation Suppose two risk reduction options have been identified:
are costed into the safety assessment at a later stage.
It is done as a risk reduction exercise using the same kind 1. Rework the evaluator component’s software
of discovery methods as are listed for Stage 1. Indeed, some
ideas may well already be on record as earlier stages were • Total cost £2 million
undertaken. • Expected life 20 years
Consider an Options Analysis for the train axle detector
given earlier. – Annualized cost: £0.1 million
Assessing and Demonstrating Safety 663
Consequence Calculation Cost £m Impact analysis is continued as iterative action loops until
all variables under control have settled to give the safest
Major collision 115.9 × 0.0033 0.38
situation for allowable factors.
From these outcomes, selection decisions need to be
made using background factors existing at the time – polit-
Minor collision 0.3 × 0.0049 0.01 ical, financial, public pressure, and so on.
No collision 0.01 × 1.0918 0.01
8 STAGE 7 – DEMONSTRATION OF
Total annualized loss 0.40 ALARP
ALARP is a well-documented legal principle used in some
Figure 8. Annualized loss of the consequences of collision for health and safety activities – to see more on this topic,
the axle detector system. search the Internet under ALARP. It is used here to show
the last stage of a system safety analysis. It is not, however,
2. Complete system redesign the only one required to be used, so always checkout
the appropriate safety standard relevant to the project –
• Total cost £5 million see Article 100, Safety Analysis Methods, Volume 2.
• Expected life 20 years Two things must be shown in order to justify that the
acceptance of identified risks are ALARP:
– Annualized cost: £0.5 million
• The overall risk, and each individual risk is below the
7 STAGE 6 – IMPACT ANALYSIS upper limit of the tolerability that has been decided for
the project and its circumstances.
The purpose of an Impact Analysis is to assess the • All practical and cost-effective risk reduction measures
effects of implementing each of the risk reduction options. have been implemented. This is largely a scientific
Its approach is to estimate the net cost or benefit of and engineering exercise but interpretations of what
each option. constituted such parameters as ‘all practical and so on’
For each option (separately), revisit the previous stages need to be done against a background of appropriate
of the analysis, inserting revised figures. The safety value of legal advice.
a change is calculated from the differences revealed. In this
Figure 9 shows the zones of the ALARP principle.
revision process, carry out each modification on the same
Figure 10 shows how to tabulate the cost/benefits result-
qualitative or quantitative basis that was used originally;
ing from the Loss and Impact Analyses.
to change this, will not allow the baseline to be properly
Setting limits of acceptability are usually within the
compared with the revision state.
prerogative of the appropriate Board of Safety, or are
Considering the two options identified for the axle
dictated by national or international law. The task of safety
counter and placing in some realistic data leads us to the
analysis is to produce a Safety Case that demonstrates
following:
adherence being as low as it is reasonably practical –
1. Rework evaluator software ALARP.
Bodies issuing the safety compliance certificate can
• Reduces failure rate by a factor of 3 inspect the report of the 7 Stage Safety Assessment process
• Reduces annualized loss from £0.4 million to to ascertain that a sound and thorough study has been made.
£0.13 million Legal litigation is heavily based on the principle of
– Reduction in risk = £0.27 million per annum reasonableness. This means that best practice needs to be
reached in safety design because the knowledge of what can
2. Complete system redesign be done and how it can be achieved is available, making it
reasonable to expect it to be used. This basis can, however,
• Reduces failure rate by a factor of 10
be slow to encourage new technology to be used, so some
• Reduces annualized loss from £0.4 million to
legal systems now include judgments that need a designer
£0.04 million
to apply current technology that would make the difference
– Reduction in risk = £0.36 million per annum as long as that technology is reasonably available; it is not
664 Measuring System Design
Unjustifiable
Unacceptable risk
region
Tolerable
region Control measures
Increasing individual required to drive
risks and societal residual risk down
concerns
Broadly Level of residual
acceptable risk regarded as
region insignificant
Figure 9. Zones of the ALARP principle.
Cost/benefit Loss per Annual Annual cost Option
calculations year (£m) benefit over of required?
current modification
design (£m)
Current 0.4 n/a n/a n/a
Loss design
analysis
Option 1 0.13 0.27 0.1 Yes
Impact (rework
analysis evaluator s/w)
Option 2 0.04 0.36 0.5 No
(complete
redesign)
Figure 10. Table demonstrating the costs and benefits of a change to the system.
possible under those rules to retain old systems that could • the use of unstated assumptions and unexpected
be made safer. events – we cannot totally, definitively, predict all
likely failure mechanisms! Their discovery is an ongo-
ing process.
9 SAFETY PROCESS REVIEWS
An independent professional review is essential to make
It is necessary to give time to internally review the safety sure internal bias does creep into assessments. The purpose
process put into place. of reviews is to
Ask these questions of that process:
• provide assurance that the work that has been car-
• Is the hazard identification up to date? ried out has been (or is being) done in an adequately
• Have all potential accidents been assessed? safe manner (i.e. in accordance with the agreed Safety
• Have all relevant probabilities been checked? Plan);
• Is probability assessment conservative? • form a judgment as to whether the risk associated
• Has risk tolerability been assessed? with the system has been (or will be) reduced to an
• Have the risks reduced, the system still being ALARP? acceptable level.
Beware of Reviews are carried out as a series of appropriately scaled
• common cause failures, and scheduled audits and assessments that review both
• excessive optimism regarding systematic failures, process and product.
Assessing and Demonstrating Safety 665
External audits and assessments are also needed. These will not be available. Prior checks will be carried out to
are carried out on the safety process by suitable third party uncover any obvious safety defect, but there they are more
authorities. a matter of requirements and configuration control.
Safety Audits focus on the Engineering Safety Manage- They will also be made immediately prior to safety
ment (ESM) processes. Are the processes adequate? Are case submission to ensure that all is in order right up to
they being adhered to adequately? the minute.
Safety Assessments focus on the project’s product; they The audit/assessment process starts off with the forma-
evaluate the risk. Is the system safe? Is it fit for the purpose? tion of a suitable plan. This includes
Does it meet its safety claims?
In practice, there is some overlap between these two • familiarization with the project,
kinds of activity. • review of previous audit or assessment,
Audits and assessments must be conducted in an inde- • requirements, activities, timescales,
pendent manner. The level of independence will depend • project dependencies,
upon the system involved. Several systems that are under • audit/assessment team,
the care of an organization may warrant different levels of • delivery of documentation.
independence.
These are driven by the system’s defined Safety Integrity The plan is then carried out and the results reported to
Level (SIL). Other factors, however, that may need to be the agency calling for the assessment or audit.
considered are Useful checklists of parameters to address in assessments
and audits are given in Railtrack (2000, 2003).
• degree of developer’s previous experience with a simi-
lar design,
• degree of complexity,
• degree of novelty, RELATED ARTICLES
• novelty of design,
• novelty of technology. Article 63, Outline of Systems Thinking, Volume 2;
Article 64, Executing A Measuring System Design,
The timing of audits will generally start some way Volume 2; Article 99, Safety Organization, Volume 2;
through a development when it is late enough for something Article 100, Safety Analysis Methods, Volume 2.
to be there to be audited, but early enough to allow
improvement to the processes while error correction costs
are low.
REFERENCES
Extra audits are carried out after transfer of safety respon-
sibilities to the user agency and during all of the operational
Mugridge, C. (2003) Teaching Notes for Safety Course, Witch-
life. The system may be very well maintained and be kept wood Engineering, UK.
within original specification, but the working environment,
Railtrack (2000) Yellow Book – 3, Railtrack PLC, Euston Square,
and political and public scene factors may have changed to London, distributed by Praxis Critical Systems Ltd, 20 Manvers
degrade the ALARP levels of understanding in the origi- Street, Bath BA1 1PX, UK.
nally certified system. Railtrack (2003) Engineering Safety Management Guide,
Assessments are done upon completion of the detailed http://www.yellowbook-rail.org.uk/site/the yellow book/
engineering design phase. If done earlier, sufficient details the yellow book.html.
102: Introduction to the Legal Process
Christopher Sweet
Minter Ellison, Adelaide, South Australia, Australia
The instrument or system designer will need to be aware
1 Introduction to Legal Issues 666 of the law that will impact on the design and test processes
2 Definitions 666 such as the preparation of contracts and patents and, then,
3 Legal Systems in Force 667 the legal context in which the system or product is to be
tested and subsequently used.
4 Application of the Principles and Process of
Designers and test engineers should carefully consider
the Law 667
how the law might impact on each step in the design,
5 Difficulties in Application of the Law 668 development, and test process, whether it be the contrac-
6 The Litigation Process 669 tual arrangements, protection of intellectual property, or
7 Commencing Litigation 669 liability, and seek appropriate advice to reduce the risk of
8 Evidence and Documentation 669 adverse legal consequences, whether these be losses due
9 Reducing Liability 670 to inadequate contracts or liability due to a failure in the
system. This article covers a broad introduction to some of
Further Reading 670
the legal concepts and issues that might arise. A case study
about legal liability is given in Article 103, Legal Liability
Issues for Designers – A Case Study, Volume 2.
1 INTRODUCTION TO LEGAL ISSUES A brief outline of some of the basic concepts follows.
The sorts of legal issues that arise in measurement systems
can be many and varied. This is because there are different
systems of law in place, and different legal issues can 2 DEFINITIONS
occur at different stages of the design, implementation,
and application times of the life cycle of a measuring In most countries, the main sources of law are legislation
system. For example, there might be specific legislative and case law.
requirements such as those imposed by quality standards
that have to be met in the design and production of a
particular component of a measurement tool. Then, during 2.1 Case law
the operational phase in which the measuring system is
used, there may be other standards such as environmental This is law made by the courts. The process of courts
standards that the designer of the measuring system needs to making law through judicial officers (e.g. magistrates or
have in mind in setting the parameters for the measurement judges) depends on a number of factors:
system. The sorts of legal issues that arise will depend on
the nature of the system or product being developed and on • Level of the decision-making court in the hierarchy of
the laws of the countries in which the instrument or system courts. Most countries have a hierarchy of courts, and it
is to be developed and used. usually follows that the higher the court in the hierarchy,
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Introduction to the Legal Process 667
the more binding or persuasive its decision will be on England. By common law, we refer to a system of case
the courts below. law developed by the courts in which a complex set of
• Nature of the cases or disputes coming before the precedents are gradually developed by judges considering
courts, that is, the fact situations that allow the courts cases coming before the courts.
to adjudicate and make law. Most countries that were previously in the British Com-
• Case law from other countries, which may influence monwealth have a system based on the common law. The
courts or be considered binding on courts in another common law is law made by judges when they adjudicate
country. disputes between parties. The authority to make common
• As cases come before the courts for adjudication, a law is derived from the authority exercised by English
body of law is developed by judges applying judicial judges over many centuries. A few examples of common-
decisions from other cases to new fact situations. Many law countries are Great Britain, Australia, many of the states
court proceedings are recorded in official court reports of America, and Canada. In these countries, judges have
and so become available to judges and the public for retained a significant direct role in creating law rather than
future reference. just interpreting law made by legislation.
Many European countries have a different system of
‘civil law’, which is based on systematic legislation extend-
2.2 Legislation
ing over more areas of law rather than case law being made
by the judiciary. The law is ‘codified’ in detailed legislation
Legislation is law made by a parliament. Legislation sets
and interpreted or applied by the courts.
out the law relating to its subject matter in one document.
These are just some of the systems of law in force
Amendments can then subsequently be made by further
throughout the world. It is important to interpret a situation
amending legislation.
according to the particular system in place in each country
Legislation is produced following a lengthy process of
in which a system is developed or used. This also means
debate in parliament and public consultation. The primary
where the system is to be used rather than where it is
document is termed an Act of Parliament, which is usually
designed and made.
required to be passed through parliament.
Some acts are termed Consolidating Acts because they
consolidate legislation on a particular subject. Acts that
amend legislation are usually called amending acts.
4 APPLICATION OF THE PRINCIPLES
Some Acts adopt international conventions into the law AND PROCESS OF THE LAW
of a particular country. An example is the Warsaw Conven-
tion, which concerns aviation liability and which has been In practice, the law is applied in various ways.
introduced into the laws of many countries through aviation During the initial contract phase, this can be protecting
liability legislation. intellectual property, compliance with standards, or dealing
Acts also allow for regulations to be made under the with a liability claim.
Act, which give effect to certain administrative matters. In the case of complying with standards, which is one
An example would be regulations that contain published of the most common applications of the law, that is, via
documentary and physical SI standards. Another might be legislation, those responsible for compliance will normally
a law to protect the environment, that is passed in an act of familiarize themselves with the standards relevant to their
parliament, such as ‘The Environmental Protection Act’. system design and incorporate the standards into the design
The specific standards for emission of certain industrial as part of the design process. This will involve researching
gases would then be set out in regulations made under or obtaining advice on the standards that apply to the partic-
that act. ular design. Most final contracts will dictate the standards
Both case law and legislation comprise the law in most to use.
countries, and both will need to be taken into consideration An initial nondisclosure agreement is usually signed by
by all designers and testing processes at various stages. the would-be contractor and the customer. Another prepara-
tory legal mechanism used at the first encounter stage is
the use of a Memorandum of Understanding (MOU), other
3 LEGAL SYSTEMS IN FORCE names used being ‘a letter of agreement’, ‘heads of discus-
sion’, and more.
Many countries have different systems of law, which gov- In the case of protecting the intellectual property of a
ern the way the law regulates processes. For instance, many system, applications can be made for patents or by dealing
countries have a system of common law such as that in with the rights of the parties in a contract. All these
668 Measuring System Design
processes involve the application of the law and the legal to another. For those providing professional services, the
rights of the parties. tort of negligence is probably the most frequently used
Another common application of the law is in the adju- common-law action.
dication of liability disputes between individuals or cor- Explaining the law of torts and the concept of negligence
porations. As mentioned earlier, this is the way in which easily needs a large text book to introduce its many facets.
case law is developed and applied by courts in factual In very simple terms, it involves the concept of owing
situations. a duty of care to another, the standard of that duty of
This is more difficult to research because case law is care, breaching the duty, and causing damage to the other
developing everyday on different subject matter and not party. Unlike a situation in which you may have a contract
all cases are officially reported. While the most significant with another party, under which you have certain remedies
decisions of the highest courts are reported, those decisions against that other contracting party for a breach of the
of lower courts may not be reported. Other sources of contract, the duty of care in the law of torts does not depend
information on court decisions are textbooks and journals on a contract existing between the parties. For instance, if an
and Web services. engineer contracts with an aircraft manufacturer to design
All jurisdictions have a hierarchy of courts. The criteria an aircraft and if the aircraft is faulty, the manufacturer
for each level of the hierarchy will determine the court in has contractual rights against the engineer. If the aircraft
which a particular dispute is adjudicated. In most jurisdic- crashes and injures passengers, the passengers may have a
tions, the level of the court is set by criteria such as the type claim against the engineer, under the tort of negligence.
of crime or the monetary value of the compensation being Laws also vary from country to country and, therefore,
sought. For instance, most jurisdictions have a supreme or so do standards. Those designing measurement systems for
high court at the top of the hierarchy, a district or county use in different countries need to be aware of the differing
court at the middle level, and a local or magistrate’s court standards between countries. A measurement system used
at the lower level. With all the various different levels in in one country may be subject to different thresholds of
the court system, cases on the same sort of subject mat- liability in another.
ter may be subject to adjudication in different courts for Fortunately, with the development of more international
the same case. You might find that the legal liability for standards and more uniformity of standards for products
a person injured in the same accident is being determined between countries, it is becoming easier for a designer and
in different courts because each person is seeking different manufacturer to produce a product, which will ultimately
amounts of compensation. comply with the standards in a wider variety of countries.
While complying is more straightforward with legislation,
it is not so with the common law.
5 DIFFICULTIES IN APPLICATION In the same way as mandatory products standards vary
OF THE LAW from country to country, the threshold for tortious liability
at common law will also vary even between those countries
The common law probably presents a more difficult concept with the same basic system of common law. For instance,
than legislation. Legislation can be obtained from most gov- the standard of care required by a design engineer in
ernment printers and is usually identified by subject matter. one common-law country such as the United Kingdom
Because legislation is constantly undergoing change by par- may be more or less stringent than in another common-
liament, subsequent amendments also need to be checked. law country such as Australia even where the second
Helpfully though, many government printers will produce country’s common-law system is largely based on that of
a consolidation from time to time, which brings together the first.
the principal act and subsequent amendments. There are Because the standard of care expected by any common-
also many annotated versions of legislation produced by law system will be subject to change at any time by the
academics and law firms. courts, there is a much greater imposition on any designer
The common law, on the other hand, requires constant or manufacturer, particularly one offering a professional
research for new decisions by the courts. Judicial consid- service in the nature of expert advice, to be conscious of the
eration of previous cases and facts can sometimes produce relevant standard in the jurisdiction in which they operate
only a minor variation to the law but one that can mean the and the likely standard in any other jurisdiction in which the
difference between liability and no liability for a particu- expert advice is likely to be relied upon. Fortunately, many
lar act or omission. One of the most relevant common-law common-law countries are now incorporating the concepts
concepts for professionals is the law of torts (or wrongs). of standard of care into legislation to try to provide more
This aspect concerns civil liability for damage caused certainty and consistency into these concepts.
Introduction to the Legal Process 669
Where legislation exists, it will generally override the The defending party is then usually required by the court
common law on the same subject matter. However, legisla- rules to file a responding document called a defense. If
tion is still subject to interpretation by the courts, so expert no defense is filed, the court rules usually provide for the
advice from those legally trained should be considered suing party to enter judgment by default. It is, therefore,
where there is doubt about which law or standard applies. extremely important that a party being sued responds to the
service of a letter of demand or writ. Usually, a lawyer is
consulted when either of these documents is received. The
6 THE LITIGATION PROCESS statement of claim and defense are called the pleadings.
They define the issues of fact and law in dispute between
The litigation process is complex. Litigation involves the the parties and are usually drafted by lawyers.
process by which a series of events (facts) become a dispute The type of claim and the monetary value of the claim
between two parties and ends up in court. The term often determine the forum in which the court action is com-
used is suing another party. menced. The process of commencing the claim is the same
An event that starts the litigation process is usually one for both individual and corporation.
where something goes wrong, which leads to injury to a The action then proceeds through the court list toward a
person, damage to property, or a loss to business caused trial. Certain steps need to be taken by the parties before
by failure. For instance, it may be a failure that leads to trial. These are
interruption of a business activity. There are any numbers
of failures that can cause such an event. • finalization of the pleadings;
The party (person or corporation) suffering the loss can • disclosure of all relevant documents to each other;
make a claim for compensation against the other. If a • formal pretrial conference to facilitate settlement of the
contract exists between the parties, the usual claim is for dispute.
breach of contract. If no contract exists, the claim can be
At any stage, the possibility for a negotiated settlement
in tort if such a right is recognized by law.
or other form of resolution such as mediation can be
The process of litigation often starts with a letter of
explored. In general, 90% of all litigated disputes settle
demand written between the parties to try to resolve the
problem or to negotiate a settlement for money. A lawyer before going to trial. This is because the outcome of a trial
may be consulted at this stage or the parties may chose is considered to be uncertain and many parties prefer to
to try to reach a resolution between themselves. They may reach a compromise between them.
have consulted a lawyer to find out their rights under the At all times during the process, parties (clients) are
law or the strength of their respective positions. In more expected to cooperate with the court and their lawyers
serious events, most corporations involve lawyers from the by providing frank and timely instructions and by making
time of the event. evidence available. Evidence can be in any form such as
If a letter of demand is not successful in providing products, components, data, information either written or
resolution, lawyers are often called in to advise the parties recorded in any other form, or oral testimony.
on their rights. These rights may include commencing
litigation in the courts. There are other alternatives such
as further negotiation, mediation, and arbitration. Each has 8 EVIDENCE AND DOCUMENTATION
pros and cons in terms of cost and effectiveness.
In any situation in which legal issues are concerned, but
especially in a litigated case, evidence is crucial to the way
7 COMMENCING LITIGATION the case is conducted.
Because a party must prove its case to a court, it is by
Litigation is usually commenced with a formal document the evidence called at a trial that this is achieved. Whether
called a writ or summons. This document is filed with the the evidence is the oral testimony of a witness, documents
court and served on the other party being sued. The formal or data, or expert evidence, the case will only be as good
court process is called an action. The receipt of a writ or as the evidence available to prove the case.
summons may be the first notice a party has of a claim or It is, therefore, important that good quality and complete
action against them. information of the various steps in the design and imple-
Once the litigation process has commenced, the suing mentation of measuring systems is recorded and retained.
party may then file a more detailed statement of its claim Claims may not arise for many years after the product or
against the other, called a statement of claim or particulars system is released into the public environment, and dis-
of claim. putes may not reach trial for many years. Memories as to
670 Measuring System Design
what took place will fade and the information recorded by in defending a claim if it does arise, an organization or
the individuals involved will often be the only reminder of individual should
what took place. There are legislated time limits imposed
on claims for damages and under the law of contract in • be familiar with the applicable law and standards that
most jurisdictions, but these can still be many years and apply and that will apply to the system, test environ-
extensions to these are possible in some cases. ment, and the final system or product;
Information and data that might be needed to trace a • maintain and preserve clear and comprehensive records
process for those involved should therefore always be stored throughout the entire design and test process, whether
in a secure environment for long preservation. Legal advice it is written notes, data, or hardware;
can be obtained before information is generated to ensure • seek appropriate expert advice to establish a sound
it is likely to be suitable. Legal advice is needed before and comprehensive contractual basis for any activity to
destruction as to the relevant time limits that might apply protect intellectual rights and to deal with possible con-
for claims. sequences such as by maintaining adequate insurance
Good preservation of sound and relevant information is to cover all potential liabilities;
one of the best risk-management procedures for litigation. • seek timely expert legal advice in responding to all
The important issue for each designer to keep in mind is major incidents and threats of legal action such as a
to inform themselves of the relevant legislative standards letter of demand or writ.
and the relevant common-law duties that they are required
to comply with. Following these basic principles will not, of course, guar-
In most common-law jurisdictions, for instance, the stan- antee a litigation-free project, but it should help reduce
dard of care likely to be used to judge a professional service, the numbers of adverse situations and assist in dealing
such as might be provided with expert advice in the design with such a situation, if it arises, in the best and most
of a measuring system, is that of a test of reasonableness. efficient way. By being prepared and being able to demon-
While it is often difficult to identify the precise standard as strate, a case can be well defended and will deter claims
to reasonableness, the best guide is usually what is consid- being made.
ered to be appropriate industry practice by one’s peers. A Some useful reference texts for further material on the
rider to this is that today what is reasonable to use because legal processes are given.
of technological change also needs to be considered. All
this becomes quite difficult when the technology is consid-
ered to be at the very cutting edge of industry practice and
FURTHER READING
standard or where, for instance, research into a completely
new system is being carried out.
Cooke, J.R. (2001) Architects, Engineers and the Law, 3rd edn,
The Federation Press, New South Wales.
Kutz, M. (1998) Mechanical Engineers’ Handbook, Wiley, New
9 REDUCING LIABILITY York.
In order to reduce the risk of legal liability arising during Many other more detailed law texts are available on specific legal
the design and development of a system, and to best assist concepts and topics.
Legal Liability Issues for Designers – A Case
103:
Study
Christopher Sweet
Minter Ellison, Adelaide, South Australia, Australia
sophisticated product, such as a commercial aircraft, can
1 Introduction 671 fail, leading to legal liability being attached to several
2 Facts and Background of Role Study 671 parties including, perhaps, the measuring system designer.
3 Legal Consequences in General 672 The liability might be for damage to person, property, or
for loss of business.
4 Liability for the Deaths of the Test Crew and
It is quite easy to see how an omission, however minor,
Damage to Property 672
during the test and evaluation phases of a part of the
5 Public Inquiries 673 design of an instrument or system can lead to catastrophic
6 Criminal Investigations 673 consequences for the testing body, the manufacturer, and
7 Product Liability and Recall 673 any third parties injured by the omission.
8 Legal Process 675 The legal concepts in the example can be readily applied
9 Legal Risk Mitigation 675 to systems designed for measuring or testing or those that
Reference 676 might be used during the operation and performance of a
system or product. Equally, the concepts can be applied to
the provision of advice as well as the provision of services
that are based on software or data.
1 INTRODUCTION Legal risk is not totally avoidable but it can be mitigated
by taking appropriate actions from the start of a project.
A general introduction to the legal aspects of engineering This is covered in Section 9.
applications is presented in Article 102, Introduction to
the Legal Process, Volume 2. It is shown in that article
that there are many legal situations to guard against and
2 FACTS AND BACKGROUND OF ROLE
that expert legal assistance is often essential throughout
a project. STUDY
This case study extends that discussion by illustrating
some of the legal liability issues that can arise in the The case study example involves a fictitious university
perspective of an application. It is intended to highlight research unit, which has tendered for a contract concerned
issues, not to constitute legal advice: it is important to note with testing the accuracy of a navigational instrument to
that legal advice must be taken on a case-by-case basis be installed in the cockpit panel of a commercial aircraft.
using experts who can chart the way through the law and The navigational instrument incorporates the use of a novel
its processes. method for processing the output signals from a conven-
In particular, this study demonstrates how a system tional directional gyroscope, which indicates the direction
designed for measurement during the design phase of a of flight to the flight crew.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
672 Measuring System Design
The university unit decides to tender as a joint venture These are each likely to be very expensive and time-
with a private technology company. In particular, the uni- consuming exercises. Their impact on the providing parties
versity brings to the contract the expertise of one of its and their affiliations can be very harsh, leading to loss of
senior researchers on the behavior of measuring instruments employment, criminal penalties, corporate collapse, pay-
under extreme temperatures. Temperature is a critical fea- ment of compensation, and loss of goodwill in the market
ture in the proper operation of the new instrument. place, to name a few.
A contract is drawn up between the university and private
company for each to provide certain aspects of the tender
services. The tender is successful and provides a significant 4 LIABILITY FOR THE DEATHS
grant to the research budget of the university department. OF THE TEST CREW AND DAMAGE
This university has never really been overly concerned
about its exposure for liability on these sorts of projects and, TO PROPERTY
therefore, decides that its normal insurance arrangements
are adequate for the joint venture. There is likely to be a claim by the dependants of the test
The contract goes ahead without event until there is a crew killed during the test flight. This claim may be brought
change in personnel within the university department. The under one or more of the following
senior researcher with the expertise in temperature effects
• aviation law
has a major disagreement with the head of the department
• at common law
and resigns on short notice. His area is quite specialized,
• under fatal accident legislation
and the university finds it hard to replace his expertise.
• under employee/employer liability.
The head of department assures the private company that
they can continue the contract with other researchers and In each case, there is likely to be a complicated legal
so it continues. analysis of the various competing claims.
The testing process is completed and the instrument is Any claim brought by the dependants is most likely to be
rated to perform satisfactorily at its declared operating against the manufacturer of the prototype, but it may also be
temperatures. However, due to a miscalculation by one contemplated against the contractors such as the university
of the researchers during testing, it is actually out of and private technology company or the flight licensing
specification and indicates incorrectly when subject to very authority. To make matters more complex, those influencing
low temperature. the legal forces at play can be constantly changing.
The instrument is fitted to a prototype aircraft by the Even if the claim, or claims, is brought against the
manufacturer for final operational test and evaluation. On manufacturer and not the university unit/private company,
its first test flight, the directional instrument fails to function the manufacturer or their insurer will almost certainly seek
accurately at high altitude (due to the low temperature there) to recover its losses from the university/private company.
and the prototype aircraft crashes, resulting in the loss of These losses may include compensation for the loss of the
the test crew. prototype hull and/or for loss of profits.
The contracts for the project may (or may not) have
included detailed clauses providing for indemnities between
3 LEGAL CONSEQUENCES IN GENERAL the parties in the event of certain types of claims. This
may further complicate (or possibly simplify) matters as to
A number of significant legal consequences can flow from which party is ultimately liable.
this scenario. The key issues are The same could be said in the case of claims by any
person injured on the ground or any property of such
• liability for the deaths
persons damaged by the falling wreckage.
• liability for damage to property, that is, loss of the
If the various companies involved have insurers, each is
aircraft and damage to any buildings it destroyed
likely to become involved with their own team of lawyers
while crashing
in addition to the lawyers involved for the parties, adding
• public inquiry to ascertain what went wrong
further complexity and cost to the resolution of any claims
• criminal investigation
arising from the crash.
• product recall if the component has already been placed
In all the legal scenarios above, the manufacturer, univer-
into service.
sity, or private company could be exposed to considerable
These are addressed below as separate issues but they loss. A potential lack of adequate insurance could also
usually will be interdependent in reality. expose the university to severe financial embarrassment,
Legal Liability Issues for Designers – A Case Study 673
depending on the level of insurance arrangements for the irresponsible acts that led to the loss of life. In many
university, under normal circumstances. countries, that situation has been changing over recent
years. Courts are now more prepared to look into the inner
workings of a company to find those persons whose key
5 PUBLIC INQUIRIES actions may have led directly or indirectly to the event.
In many countries, the crash of an aircraft will involve an So, in this example, an investigation might reveal that
official investigation by the local air safety bureau. Human the head of the university department misled the man-
fatalities in such circumstances will usually also result in ufacturer as to their ability to continue the project and
an investigation by the local coroner. consider that his or her actions led directly to the failure of
In each case, an investigation can take months or years the instrument.
to complete, will cause much anxiety for all those involved Also, one of the directors of the manufacturer may
(particularly those who will appear as witnesses), and will have known or suspected that the university did not have
potentially expose the organizations to adverse publicity the competence to satisfactorily complete the project but
during the investigations, not to mention the close scrutiny may have remained silent for the sake of the project
each organization will undergo during any legal process. being completed.
During the safety bureau investigation, documents and In all cases, prosecutions for these sorts of disasters can
design material will be taken and fully considered by a carry significant penalties for companies and the individu-
whole series of experts. Those personnel involved in all als. Companies can be fined; individuals can be imprisoned
stages of the design and test process including the university or fined. The more severe the disaster, the more severe the
team, both past and present, will be interviewed a number penalties for those found to be responsible.
of times for witness statements. For instance, it is possible
that each could be interviewed up to four to five times
by the crash investigators, coroner’s investigators, police, 7 PRODUCT LIABILITY AND RECALL
and lawyers in connection with each legal part of the case.
Each legal inquiry may result in significant consequences In addition to the above legal issues, another costly allied
for each of the parties involved. liability situation, which could occur if any component has
The coronial investigation may result in findings, which already been distributed in another product or in addition
lead to criminal investigations or recommendations, which to the above types of liability is product liability. For a
affect the licence of the manufacturer or the way in which component that has already found its way into the market
the aviation research industry is regulated. place in a product, the need to carry out a product recall may
The air safety bureau investigation and its findings may result. This is usually done to correct a defect in a design
also lead to further investigations by the police or to or its manufacture or to call back products with subsequent
recommendations for regulation of the industry. refund of purchase price and possibly compensation.
All of these actions will produce adverse publicity for all When these involve large numbers of a product item,
those involved but particularly for the commercial reputa- they are commonly seen advertised in newspapers for such
tion of the aircraft manufacturer. In this example, there are things as defects in automobiles, consumer goods, and
likely to be at least two unavoidable public hearings irre- foodstuffs. Where it is not possible to accurately identify
spective of the compensation claims. Prior to the hearings all customers, these notices are directed at the purchaser or
and for the hearing itself, staff time and traveling costs will user.
be needed to prepare maternal and to attend the inquiries In the notice issued for the defect, the serial numbers and
when called. product names of the defective lot are identified. The defect
is explained in lay terms and the recommended immediate
6 CRIMINAL INVESTIGATIONS action to be taken is stated. Persons who believe they have
a defective item are asked to make contact to arrange for
This is a third category of possible legal process that may withdrawal and refund, replacement, or rectification, free
affect this example. In disastrous events such as plane of charge.
crashes, train collisions, or building collapses where lives Other less publicly noticeable statements are sent to
are lost, there is an increasing trend for those in positions wholesalers, franchisers, distributors, and retailers.
of responsibility to be held publicly accountable for their In the case of a defect in a sophisticated aircraft and
actions or omissions. where there are a low number of early models, the manufac-
Historically, the corporate veil provided a shield for turer will be aware of all sales and can make rapid contact
individuals against criminal prosecution for reckless or with the buyers, and a public notice will not be needed.
674 Measuring System Design
Where it is not economic to rectify the item, the its data processor unit to an aircraft instrument systems’
maker will usually need to recover it, compensating the supplier. Also assume that no unit, apart from the one in
buyer. the crashed aircraft, had yet been sold for deployment.
In some cases, it will be economic and prudent to make For safety reasons, the maker needs to immediately notify
a modification to each item via a rework activity through the supplier to not sell or use any of the units and to retrieve
provision of parts to be replaced or added by the owner, or them for design alteration or destruction.
as a factory operation by the supplier. Many of the costs of a recall are for tangible items.
Recalls can be surprisingly expensive exercises. The Some items, however, are less easy to identify in clear-
overall cost might well not only take an individual project cut monetary terms, examples being the value of loss of
into a loss balance sheet but also place the organization goodwill and the impact of disturbing normal design and
into liquidation. manufacturing operations.
One way to demonstrate what is involved is by listing the Assume that the defect in this case study was shown to
factors involved for the example given, along with likely be quite simple, being that an electrical resistor involved
costs; refer to Table 1. in the calibration could marginally overheat, changing its
In this case study, assume that the maker had also resistance, which in turn could alter the gain of a critical
delivered the same incorrectly calibrated prototype gyro and signal-processing preamplifier.
Table 1. Costs associated with recall to repair or withdraw a defective low volume item.
Item Time Cost ($) No. of units Recall to fix Refund and
estimate option withdraw
(h) product option
Designing a change fix
Engineer labor (20 h @ $75/h−1 ) 20 75 All units 1500
Drawings, part lists, work instructions 20 80 All units 1600
Press release to lessen impact
Media labor costs 8 80 640 640
Newspapers/radios costs 5000 5000
Freight and handling
Freight of units back from customer 100 10 1000 1000
Tracking costs for 20 days 100 2 10 2000 2000
Outward dispatch labor 1 60 10 600
Freight of units to customer 200 10 2000
Compensation paid for inventory changes – not 2000 10 000
for any liabilities
Media relations costs 5000 5000
Execution of repair
Disassemble labor 10 80 10 8000
Repair assemble 10 80 10 8000
Test and recertification 40 80 10 32 000
Other costs
Loss of other production due to diverted staff – 1000 10 10 000
assessed as $1000 per recalled item repaired
Replacement loan gyros 500 10 5000
Legal advice on change made 5000
Total cost estimate of recall and repair of 89 340
defective instruments
Total cost to withdraw and refund for items 23 640
Minimum cost for option chosen 89 340 23 640
Sales income for 10 units 5000 50 000
Overall income (loss in this case) for option ($39 340) ($23 640)
chosen
Note: The assumption is that 10 instruments have been sold and all need to be rectified. They were sold for $5000 each, with a profit margin of 20%
of sales. The problem was soon found to be quite simple, being that an electrical resistor’s power-carrying capacity was inadequate and could overheat,
changing the gain of a critical signal-processing amplifier.
Legal Liability Issues for Designers – A Case Study 675
This problem could have arisen from such reasons as made may be the difference between accurately recalling
ones actions or not. That is why the concept of proper
• supply of the wrong part by errors of specification or
record keeping is reinforced constantly by the legal pro-
purchasing, or selection;
fession. When a person gives evidence to a court, their
• ventilation around the resistor had been reduced from
credibility to recall events is often determined by whether
the design value to save cost and size;
they have properly recorded their past actions.
• another component generating local heat was placed
next to this critical resistor;
• the whole unit was placed in an environment hotter or
colder beyond specification. 9 LEGAL RISK MITIGATION
From the details of Table 1, several points are clear: The above examples show how legal liability will affect
• The costs of the ‘recall to fix’ and the alternative ‘refund all concerned if a design causes harm in some way in
and withdraw’ options are both major – but one has to its normal or abused use. Designers need to be able to
be chosen. The cost of recall cannot be avoided. defend any design that has been practiced to ‘best practice’
• The ‘recall to fix’ option is marginally the dearest to standards.
implement, but it will possibly be seen to be necessary Some indicators on how to limit that risk as design
for business to continue and for more sales to be made. proceeds are now presented. It would be ideal to have
• The ‘refund and withdraw’ option is the least costly a lawyer in the design team to give advice on every
option, but it loses not only goodwill but also consid- decision made but that is often not practical. Designers must
erable income from sales of the product. exercise judgment and make use of regular legal reviews
• The cost of either option may well exceed the costs of of the project.
development, influencing the future of the project. The main line of defense will usually be the ability
to demonstrate ‘sufficient standard of care’ or that ‘due
These costs are apart from any legal liability arising
diligence has been practiced’. Where a situation cannot be
from the aircraft crash but would involve a commercial
formally described in complete reductionist terms – most of
decision taken after the legal consequences have been
those that give rise to litigation – the case will be argued on
considered. It is often the case that commercial decisions
the basis of ‘reasonableness’, a highly subjective cognitive
are taken in conjunction with advice from lawyers. Clearly,
concept that lawyers are experienced in arguing one way
design errors like this need to be controlled early to reduce
or the other.
legal risk.
Preparation for legal defense starts when the project
commences. Following practices of reasonable design and
8 LEGAL PROCESS keeping a suitable document trail are absolutely necessary.
Without these, it would be difficult to provide the evidence
In all of the legal consequences mentioned above, the needed to withstand scrutiny during any of the legal pro-
process will begin with the personnel being contacted for cesses discussed above.
interviews. Either the air safety investigators or the police If adequate records are not made at the time, it may well
or the coroner’s investigators will be the first to investigate. be impossible to later find the documented evidence needed
They will request documents and other materials such to support witnesses or the defense case. It is acknowl-
as data, software, and hardware for preservation and for edged that detailed documentation will impose significant
the purposes of their inquiries. Equipment and files may additional overhead on a project and record information that
be impounded. may never be called upon. Taking such precautions is, how-
One would also hope that the companies have contacted ever, much like taking out additional insurance; one hopes
their own lawyers and insurers at the earliest opportunity it is never needed but it is a comforting thing to have in
to assist with responding to the investigations. place when needed.
The quality of the documentation made during the var- The design team is not expected to provide for every
ious processes used in design and testing becomes critical possible circumstance in a design, as that would sup-
while responding to any allegations. All the documentation press creation and cost-effectiveness. The team should
and other materials have the potential to become evidence allow for that which is reasonable. Allowance has to be
during the investigation. Notes made by each of the per- made for all situations that can reasonably be expected to
sonnel during the project will become their aide memoires arise.
when called upon to give statements. Human memory is Some key points to assist in the better defense of a legal
notoriously inaccurate after time has elapsed and the notes liability suit are:
676 Measuring System Design
• description of, and adherence to, a formalized design • locality and extent of application (office, city, country,
process by project personnel; international)
• peer design review and approval of project design • level of perceived public acceptance (automobile safety
information and documentation; vs aircraft safety, natural environment, low or height-
• management of system safety by a formally constituted ened public interest, child or adult user)
Safety Board, with a sound and documented safety and • lifetime of system use (days to decades)
design change process; • business issues (profits made, risks to be accepted)
• reference and adherence to Standards and Regulatory • potential population affected by the system.
requirements for the appropriate aspects of the design;
• appropriate training for designers, with records of atten- Clearly, legal advice is crucial when considering proce-
dance and achievement in learning; dures and processes to minimize the risk of legal conse-
• findings of independent quality audits conducted during quences for any project. Lawyers can anticipate the con-
the execution of project activities that confirm that sequences and advise on how to protect the project from
adequate duty of care has been exercised; such risk.
• availability of the above evidence that is suitably framed A well-prepared set of documentation and a sound best
to suit legal requirements; practice process form a sound basis for fending off claims;
• maintenance of records of appropriate calibrations and either an adverse consequence will not occur or, if it
certifications; does, a claimant will find it harder to argue that standards
• availability of information to show that current practice have been breached, leading to liability. Sound processes
used in the project is in accordance with other similar will also make it much easier for a legal team to assess
designs; whether a claim will succeed and so assist them to properly
• well-run risk and safety assessment process that has manage the consequences and reduce public and commer-
been in place. cial damage.
Adequate security of records and equipments is also nec- A corollary is that a legally sound process being in place
essary to support the veracity of any case made. Relevant is highly likely to avoid litigation in the first place because
details must be secured and staff access controlled. the delivered system is less likely to incur difficulties
Just how far these various issues should be addressed because of its greater integrity.
will depend on factors such as Further information on this topic is available in Syden-
ham (2004). An extensive table of issues to address at each
• nature of the project (such as defense, government, sea, phase of the project life cycle is presented there.
air, land, utilities, commercial, personal, health, etc.)
• kind of system delivered (benign, hazardous, domes-
tic, factory) REFERENCE
• environmental impact, guidance being found in an envi-
ronmental impact statement (EIS) for the project, if Sydenham, P.H. (2004) Systems Approach to Engineering Design,
one exists Artech House, Norwood, MA.
104: Models of the Measurement Process
Luca P. Mari
Università Cattaneo, Castellanza (VA), Italy
• Although human beings are able to directly sense a fair
1 Measurement as a Comparison Model 681 amount of quantities and are well trained to express
2 The Output/Input Black Box Model 682 in linguistic form their perception (e.g. ‘it is rather
3 Set-theoretical Model 683 cold’, ‘this is heavier than that’), their statements are
affected by subjectivity, that is, they report information
4 Generalized Model 683
on both the sensed system and the perceiver state; to
Related Articles 684 avoid the influence of the latter, and thus to enhance the
References 684 objectivity of the operation, the measurand is transduced
Further Reading 684 by a sensing system whose output ideally depends
only on the measurand and is unaffected by influence
quantities and internal imperfections.
1 MEASUREMENT AS A COMPARISON • While related to the measurand, the quantity provided
by sensors still depends on their specific behavior; as a
MODEL consequence, distinct sensors, even if perfectly repeat-
able, produce different outputs from the same input;
Measurement is an operation of data acquisition and pre-
furthermore, in many cases the sensor output quantity,
sentation, aimed at expressing in symbolic form the infor-
appropriate for signal conditioning and for driving pre-
mation empirically obtained on a system about a quantity,
sentation devices, is not dimensionally homogeneous
the measurand (we accept the common ambiguities of call-
to the measurand. The sensor output must then be dealt
ing ‘measurand’ both the system under measurement and
with as an instrument reading, not as a measurand value.
the measured quantity, and the latter in both its general and
To make the information obtained by the measurement
specific forms, for example, length and length of a given
intersubjective, a common reference must be adopted so
object in a given time).
Peculiar to measurement is the requirement of being that measurand values are expressed in comparison to
objective and intersubjective, where objectivity implies that such a standard. Therefore, the possibility of tracing the
measurement results convey information only related to the readings to the agreed standard is critical, a condition
system under measurement and not its environment, and operatively ensured by instrument calibration.
intersubjectivity requires that measurement results convey
the same information to different subjects. As such, these The requirement of empirical comparison to traceable
properties appear an ideal target, justifying the efforts to standards is so fundamental that it can be assumed as
constantly enhance measurement devices and procedures. distinctive of measurement; generic scale-preserving evalu-
To achieve an acceptable degree of objectivity and ations can be formalized as homomorphisms from empirical
intersubjectivity, measuring systems are adopted, which to symbolic relational systems, as shown in Figure 1 (see
include selective and repeatable sensors and traceable stan- also Article 8, Formal Theory of Measurement, Vol-
dards. Indeed: ume 1). In the case of measurement, such mappings are
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
682 Modeling Measuring Systems
Homomorphic ‘Direct’ RS of measurand Instrument RS of derived
Empirical RS Symbolic RS
Evaluation comparison: states standard states
Figure 1. A generic scale-preserving evaluation.
‘Indirect’ RS of derived
Instrument
comparison: standard states
RS of symbols used
RS of measurand states RS of measurand Calibrated
as measurand values
states instrument
Empirical
comparison Standard definition Figure 4. The different usages of the measuring systems as
RS of standard states comparators.
Figure 2. Measurement as a scale-preserving evaluation obtained
by the comparison to a standard. measurement involves (at least) two interactions: standard-
instrument and measurand-instrument.
In its interaction with the measurand, the instrument
RS of symbols used
RS of measurand states
as measurand values generates an output; a general problem of measurement
can then be stated as follows: from the output of the
3. Empirical
comparison measuring instrument (‘the reading’) its input (the state of
RS of derived
standard states
1. Standard the system under measurement and its environment) must be
definition reconstructed, and from this state a measurand value must
2. Traceability
chain be inferred.
To cope with this input-from-output inference problem,
RS of primary
standard states two basic strategies can, in principle, be followed:
Figure 3. Measurement as a scale-preserving evaluation obtained • The analytical model of the measuring system behavior
by the comparison to a standard derived by a primary standard. is identified and the obtained characteristic function is
inverted, so that from the output readings the input
not direct but mediated by the comparison to standards, as signals are computed. Because of its complexity, this
shown in Figure 2. Finally, the situation in which primary approach is seldom adopted.
standards are not directly available can be schematized • The system is regarded as a black box and only its
as in Figure 3. In such a case, operations 1 and 2 are input–output behavior is taken into account: the instru-
usually carried out before measurement: nevertheless, mea- ment operates by interacting with a set of (known)
surement cannot be completed without them and therefore standard states and the corresponding output readings
such operations play an essential role in the definition of are recorded; by a suitable interpolation, this collec-
measurement itself. tion of couples becomes the so-called calibration curve,
As a consequence, measurement results must state a shown in Figure 5, which can be thought of as a map-
measurand value with reference to the adopted standard, ping from measurand values to instrument readings.
usually expressed in the form of a measurement unit (see This function is then inverted so that each instrument
also Article 43, Units, Volume 1). reading can be associated with a measurand value.
2 THE OUTPUT/INPUT BLACK BOX
Interpolated calibration curve
[Reading units]
MODEL
Instrument
readings
It is a well-known fact that different methods of mea-
surement exist, each of them corresponding to a specific Calibration points
technique to perform the comparison between the measur-
and and the standard (see Figure 4). While some methods
require the synchronous presence of the measurand and
the standard (e.g. following the paradigm of the two-
Measurand values [measurand units]
arm balance provided with a set of standard weights: a
direct comparison), many others are based on the usage of Figure 5. A diagram with the example of a curve generated by
devices acting as serializers of the comparison, so that a calibration.
Models of the Measurement Process 683
[Measurand • Sensitivity: ideally, x1 = x2 implies f (x1 , w, t) =
units] f (x2 , w, t), that is, distinct measurand values always
Measurand produce distinct outputs; the ratio y/x expresses the
values
aptitude of the sensor to reproduce measurand variations
to output values.
• Selectivity: ideally f (x, w1 , t) = f (x, w2 , t) even if
Measurand w1 = w 2 , that is, the sensor output is not affected by
value
the variations of influence quantities; the less is the
x
variability of y due to w, the better is the sensor
(therefore, selectivity corresponds to nonsensitivity to
x influence quantities: the relative contribution of the
Instrument readings [Reading units] measurand to the output can be formalized as a signal-
Instrument output to-noise ratio).
Figure 6. The example of an inverted calibration diagram, for
• Repeatability and stability: ideally f (x, w, t1 ) =
usage in measurement. f (x, w, t2 ) even if t1 = t2 , that is, the sensor output is
not affected by short-term (fluctuations) and long-term
The interactions standard-instrument and measurand- (aging) time effects; the less is the variability of y due
instrument have therefore a complementary function: while to t the better is the sensor (a stable sensor does not
the former is aimed at creating a calibration diagram, the require frequent recalibrations).
latter uses the inverted diagram, shown in Figure 6, to • Linearity: ideally y = ax + b (where a and b are given
find the measurand value that corresponds to the obtained coefficients, possibly with b = 0), that is, f is a straight
reading. line, the better the actual sensor behavior is approxi-
To enhance the user-friendliness of the measuring sys- mated by this equation, the better the sensor is usually
tems, it is customary to set up their presentation component considered to be (a linear, zero-crossing sensor is cal-
so that the data they display are expressed directly in mea- ibrated in a single operation, aimed at determining the
surand units, that is, the calibration diagram is embedded slope a).
into the systems. While measurement always requires cal-
ibration information, in these cases one can specifically In addition to these static parameters, the dynamic behav-
speak of calibrated instruments. ior of the sensor is synthesized by parameters such as
its frequency response (see also Article 29, Relationship
Between Signals in the Time and Frequency Domain,
3 SET-THEORETICAL MODEL Volume 1; Article 27, Signals in the Frequency Domain,
Volume 1; and Article 36, Systems in the Frequency
The sensor behavior, therefore critical for both calibration Domain, Volume 1).
and measurement, is usually expressed as a characteris- The technical specifications for sensors usually include
tic function formalizing the input–output conversion per- some quantitative evaluation for these parameters in the
formed by the sensor itself. nominal conditions of usage, expressed by the allowed
The sensor input, a couple (x, w) where x = ranges of measurand and influence quantities.
x(t) ∈ X is the measurand and w = w1 , . . . , wn =
w1 (t), . . . , wn (t) ∈ W is a collection of further quantities
influencing the sensor behavior, is transformed to its output
4 GENERALIZED MODEL
y ∈ Y . Therefore, the sensor characteristic function:
f :X×W ×T →Y (1) The inference process that leads to the evaluation and the
expression of a measurand value is always only plausible
takes the measurand x(t), the influence quantities w(t) in its results, and in general nothing can be inferred with
and the current time t, included to take into account certainty about the measurand value. The causes of this
possible time-dependent effects, and associate them with lack of certainty are various, and in particular the follow-
the output signal y(t) = f (x(t), w(t), t) to which both the ing:
measurand (‘the signal’) and the influence quantities (‘the
noise’) contribute. • The model of the measurement system has not identified
This simple formalization allows us to introduce some all of the relevant influence quantities, and any one of
basic parameters describing the static behavior of a sensor: them may have a significant variability, such that the
684 Modeling Measuring Systems
environmental conditions (including human operators) An even more general approach could be adopted by
change after the calibration. expressing the uncertainty estimation as a standard devi-
• The measuring system is less stable than expected ation, and therefore in a probabilistic framework, as rec-
when the calibration procedure was defined, that is, ommended by the ISO Guide to the Expression of Uncer-
the instrument would require a recalibration before tainty in Measurement (1993) (GUM). The Guide, based
its usage. on a recommendation by the International Committee for
• The interpolation shape of the calibration curve does Weights and Measures (CIPM, 1981), states that measure-
not adequately map the actual instrument behavior (e.g. ment uncertainty can be estimated on the basis of both
it is significantly nonlinear where a piecewise linear statistical and nonstatistical methods, and specifies a pro-
interpolation was chosen), so that for some instrument cedure to combine such components into a combined stan-
reading subsets the instrument is wrongly calibrated – dard uncertainty. The set-theoretical formalization can then
see Article 58, Description of Accuracy, Linearity, be regarded as a specialization of this framework: if the
and Drift, Volume 1. combined standard uncertainty is multiplied by a cov-
erage factor, then an expanded uncertainty is obtained,
which is thought of as the half-width of an uncertainty
All these cases can be formally characterized by rec- interval.
ognizing that the certainty implied in the choice of a The inherent presence of uncertainty justifies the funda-
single-valued association between instrument readings and mental assumption that the result of a measurement must
measurand values is not adequate. In the interaction with the state not only a (nominal) measurand value but also its
measuring system during calibration, each measurand value uncertainty estimation. Uncertainty in measurement is cov-
generates an instrument reading that should be considered ered in Article 54, Explanation of Key Error and Uncer-
a sample drawn from a whole set of possible readings. tainty Concepts and Terms, Volume 1.
Such variability can be formalized according to a set-
theoretical model, so that the information obtained in the
calibration is expressed by a calibration strip, in which an
interval of possible readings, whose center and width can RELATED ARTICLES
be considered as the nominal reading and an uncertainty
interval respectively, is associated with each measurand
Article 54, Explanation of Key Error and Uncertainty
value (see Figure 7). (Note the changes of the calibration
Concepts and Terms, Volume 1.
strip width along the measurand axis, taking into account
nonuniformities in the uncertainty evaluation.)
As in the previous (certain, and therefore ideal) case, this
diagram is used in its inverted form during measurement: REFERENCES
for any given instrument reading, an uncertainty interval
of possible measurand values is obtained together with a CIPM, BIPM (1981) Procès-Verbaux des Séances du Comité
nominal value (see also Article 54, Explanation of Key International des Poids et Mesures, 49, 8–9, 26, (in French);
Giacomo, P. (1982) News from the BIPM, Metrologia 18,
Error and Uncertainty Concepts and Terms, Volume 1).
41–44.
ISO (1993) Guide to the Expression of Uncertainty in Measure-
Nominal calibration curve
ment, International Organization for Standardization, Geneva.
[Reading units]
Instrument Calibration
readings Nominal reading strip
FURTHER READING
Uncertainty
Interval Krantz, D., Luce, R., Suppes, P. and Tversky, A. (Vol. 1: 1971,
Vol. 2: 1989, Vol. 3: 1990) Foundations of Measurement, Aca-
demic Press, New York.
x See also the volumes of the Proceedings on the International
Measurand values [Measurand units]
Measurand value Workshop Advanced Mathematical Tools in Metrology, in the
under calibration Series on Advances in Mathematics for Applied Sciences, World
Scientific, Singapore, New Jersey, London, Hong Kong (cur-
Figure 7. A diagram with the example of a strip generated by a rently published: Vol. 1: 1994, Vol. 2: 1996, Vol. 3: 1997,
calibration in which uncertainty has been taken into account. Vol. 4: 2000, Vol. 5: 2001).
105: Modeling with LabVIEW
Wiesław Tłaczała
Warsaw University of Technology, Warsaw, Poland
1.1 Virtual instruments
1 General Overview and Nature of the
Virtual instruments are computer-controlled instrumentation
LabVIEW Environment 685
systems. The physical instruments, which exist in the back-
2 LabVIEW G-programming Rapid ground of the computer (Figure 1), may be controlled using
Information 687 the computer keyboard, mouse, and display. Also, a wide
3 Data, Wires, and Terminal Types in variety of the computer-based applications for process mod-
LabVIEW 688 eling, instrument simulation or animation of the execution
4 Quick Start with LabVIEW – VI Example 688 of the measuring system are possible using virtual instru-
5 General Application to Different Classes of ments. In this sense, all LabVIEW programs are Virtual
Measuring Systems with VIs 689 Instruments, called VI s for short.
6 Relevance to Presented Applications 690
7 Communication and Data Transfer Inside 1.2 Graphical programming language G – VI
and Outside VI 691 components
8 VI Behavior Customizing 692
The graphical programming language G uses icons and
9 Modularity and Hierarchy 693
wires to create graphical source code. In graphical language,
10 Data Flow Principle Presentation 694 icons are used instead of text-lines, which are used in the
References 694 sequential text-based programming languages. Graphical
Further Reading 694 code is created on the Block Diagram behind the user
interface called the Front Panel (Figure 2). The SubVI
icon represents a program placed as a subprogram within
another VI. During program execution, data are transferred
1 GENERAL OVERVIEW AND NATURE between the Front Panel and the Block Diagram and carried
between icons on the block diagram. The VI connector pane
OF THE LabVIEW ENVIRONMENT gives the possibility of passing data to/from the VI. The
path of data flow on the block diagram is determined by
LabVIEW (Laboratory Virtual Instrument Engineering wire connections. LabVIEW is a system based on the data
Workbench) is the trademark of National Instrument Cor- flow principle.
poration (NI). It is a programming environment including
all of the tools necessary for instrument control, data stor- 1.3 Data sources
age, data analysis, and data presentation. It uses graphical
programming and a graphical language called G. The main Data used by the program can be supplied by controls on the
goal of LabVIEW is the creation of virtual instrumentation. front panel, loaded from files as off-line data, and received
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
686 Modeling Measuring Systems
Real-world instruments in the
background of the computer
Detector
2Θ
w
Sample
X-ray tube
Virtual instrument
Data
visualization
GONIOMETER
MODULAR DATA HANDLING SYSTEM
Control
Figure 1. Virtual instrument presentation.
Title bar
Menu bar
Tools bar
available in
running mode
Window with
front panel–
user interface
Data transfering between
front panel and block diagram
Tools bar
available in
editing mode
Window with
block diagram
source code
Figure 2. VI components.
Modeling with LabVIEW 687
through ports from physical instruments as on-line data and setting breakpoints, probes and shortcut menus, and also
also from the Internet. for operations usually used in a paint program. The Tools
After modification, data can be displayed by indicators palette is common to both front panel and block dia-
on the front panel, saved to files, or sent back to peripheral gram operations.
physical instruments.
Each control and indicator placed on the front panel is
represented by its corresponding terminal, which appears 2.2 The front panel
automatically on the block diagram when VI is built.
Data are passed to/from the object on the front panel The front panel window contains various types of controls
through the object terminal on the block diagram. and indicators and models a real-world instrument front
panel. Front panel design activity needs to place on it
objects used by a user to control VI and to display data
2 LabVIEW G-PROGRAMMING RAPID produced by the program. Figure 4 shows, as an example,
INFORMATION the front panel of the virtual X-rays diffractometer, which
gives a user possibility to calibrate the sample and detector
After launching the LabVIEW application, the start-up positions; select parameters for the scanning; acquire and
screen appears and a user can select a New VI. A new present data; and save data in the file.
VI named Untitled 1(2,3. . .) appears as a composition of The complete set of front panel objects is available from
two windows that comprise a VI, the front panel, and its the Controls palette and the All Controls subpalette.
associated block diagram. At the initial stages of building
a new VI, it is helpful to see both windows side by side to
determine what can be done from the menu bar. 2.3 The block diagram
The block diagram window contains the program code
2.1 Editing tools in a graphical form. It contains terminals of the front
panel objects, programming structures, functions, constants,
Editing tools used to build a VI are the Tool bar and Tools, subVIs, and wires.
Controls, and Functions palettes (Figure 3). Palettes can be The complete set of block diagram functions is available
visible after selecting them from the menu bar. from the Functions palette and All Functions subpalette.
The Tool bar contains buttons and indicators to build, Objects can be placed on the front panel/block diagram
edit, debug, and run VIs. by dragging and dropping, or copying. Front panel objects
The Tools palette is a graphical panel that contains var- and constants on the block diagram can be also created
ious tools used to create and operate VIs. A tool is a by selecting from the shortcut menus of the objects placed
special mouse cursor for performing specific operations earlier on the block diagram.
such as positioning, wiring, labeling, operating, scrolling,
Figure 4. The front panel of the X-rays diffractometer. The
application has been developed by M. Sarzyński, a PhD student
Figure 3. LabVIEW editing tools. at the faculty of physics in WUT, under the author’s direction.
688 Modeling Measuring Systems
2.4 Wires and terminals 3.4 Debugging techniques
Wires are used to carry data from one terminal to another. The broken Run button appears on the tool bar in the case of
Wires are analogous to variables in traditional program- a broken VI. A broken VI is a VI that cannot compile or run
ming languages. Terminals are objects or regions on nodes because of programming errors. LabVIEW has many built-
through which data pass. in debugging features like Execution Highlighting, Probe
and Breakpoint tools, Pause, and Single-Stepping through
a VI.
3 DATA, WIRES, AND TERMINAL TYPES With the Execution Highlighting, an animation of the
IN LabVIEW execution of the VI block diagram can be viewed. Using
Probes, the intermediate values created during VI execution
3.1 Data types can be checked. Single-Stepping through a VI and its
subVIs can be used to execute a block diagram node by
Data type is a format for information. Data can be the node. Breakpoints can be set at certain locations of the VI
instrument status information data, reference data repre- using the Breakpoint tool. The Pause button temporarily
senting node properties or status, data obtained as results of stops execution.
measurements, or data used for addressing. Data types used The programmer can also open the Error List window
in LabVIEW and acceptable for most VIs and functions are by clicking the broken Run button and locate the object
numeric, Boolean, string, waveform, array, cluster, refnum, reporting error.
enumeration, and path.
4 QUICK START WITH LabVIEW – VI
3.2 Wire types
EXAMPLE
Wires can be of different types, colors, line styles, and
thickness, depending on carried data types. Wire types are 4.1 Building VI
unique for each data type. For example, orange color is
reserved for floating point numbers, blue for integers, green As an example, consider the VI created for presenta-
for Booleans, and pink for strings. Basic wire styles used in tion of the 137 Cs spectrum data loaded from the file.
block diagram are as follows: thin, solid lines are used to For building a user interface as shown in Figure 5, there
carry scalars; a bit thicker solid lines for 1D arrays; double is need to create a waveform graph, a slide, and path
lines are reserved for 2D arrays of numbers; and thick lines objects. These are created as objects by selecting them
are used to carry 2D arrays of Booleans or strings. A broken on the Controls palette. The Positioning tool is used to
wire appears as a dashed black line. To find out why the rearrange or resize selected objects. Right-clicking the
wire is broken, the Wiring tool is moved over the broken object and selecting Properties from its shortcut menu
wire to display a tip strip with an error description. can open the dialog box to configure the object. The
Labeling and Operating tools change the defaults for the
object. Data paths between output terminal and input ter-
3.3 Terminal types minals on the block diagram are defined using the Wiring
tool.
Control terminals and node outputs terminals have thick
borders, while indicator terminals and node inputs terminal
borders are thin. Terminal color indicates what data type
will be supplied or accepted by this terminal. The program-
mer can open the Context Help window, move the cursor
onto the selected node, and get information about data
types accepted or supplied by the terminal. The Context
Help window also displays the function and VI parameters,
the object description, and the parameter type definitions.
It can be opened by selecting the same from the menu
bar. A broken wire can be removed by selecting and delet-
ing it. Figure 5. Simple VI example – its front panel and block diagram.
Modeling with LabVIEW 689
5 GENERAL APPLICATION TO
DIFFERENT CLASSES OF MEASURING
SYSTEMS WITH VIs
5.1 Where can LabVIEW be used?
Areas of the typical uses of LabVIEW are electrical
and mechanical engineering, physics, chemical engineer-
Figure 6. LabVIEW programming structures. ing, biology, biomedical engineering, civil engineering, and
many others. For example, scientists at NASA used Lab-
For VI execution control, the While Loop is created by VIEW to analyze and display Mars Pathfinder Sojourner
selecting the same on the Functions palette. The While rover engineering data (Bishop, 1999). At the Laboratory
Loop is a structure used to repeat the subdiagram inside of Dielectrics and Magnetics at Warsaw University, X-
it until a conditional terminal receives a particular Boolean ray measurements of the layer spacing in the SmA phases
value. The iteration terminal provides the current loop iter- were performed using the diffractometer (Figure 4) con-
ation count. trolled by the software developed in LabVIEW (Pociecha
et al., 2002).
4.2 Programming structures 5.2 LabVIEW in industry and research
LabVIEW applications are implemented worldwide in
Programming structures, shown in Figure 6, are program many industries and in research for test and measure-
control nodes, such as Flat/Stacked Sequence structures, ment applications, process monitoring and control, machine
Case structure, Event structure, For Loop, or While Loop. monitoring and control, and design process control. For
Structures can execute code repeatedly or conditionally. example, Construzioni Apparecchiature Elettroniche Nucle-
They are not created on the block diagram immediately ari (CAEN) S.p.A, offers a VME (Versa Module Eurocard)
after being selected. Instead, a small icon representing controller which can be operated from the parallel port
the structure appears in the block diagram, giving the of a standard PC using LabVIEW software developed in
programmer the opportunity to size and position it to CERN. Analog Devices (AD) and Crossbow Technology,
encompass all the objects that need to be placed inside Inc. used LabVIEW software for the demonstration dual
the structure and then execute within it. The structure- axis, 2 g accelerometer.
creating technique is shown on the left-hand side in
Figure 6.
5.3 LabVIEW in universities
LabVIEW programs used in university laboratories improve
4.3 Running VI the didactic process. These applications can be used in e-
learning and distance learning. With its Remote Panel, the
LabVIEW application can be turned into a remote labora-
A user can run a VI by clicking its Run button. If
tory. With their Web browser, users can open a link to a
a user runs the VI, LabVIEW compiles it and starts
remote application, and control it using the Remote Panel,
the VI execution. While the VI is executing, the Run
and observe how an experiment runs. Such a panel, as
button changes appearance to indicate that the VI is
an example, is shown in Figure 7. This is the statistical
running.
distribution analyzer for remote control of the experiment
A user can stop execution by clicking the Abort Exe-
devoted to the statistical character of the gamma decay
cution button or by using the Stop button placed on the
investigations (see also Article 106, Virtual Instrumen-
front panel. The program will stop immediately when the
tation in Physics, Volume 2).
Abort Execution button is selected, but data can be cor-
rupted this way. The VI will reach the end before stopping
when the Stop button is used. This is the recommended 5.4 Communicating with LabVIEW
stopping mechanism. The continuous Run button causes the
VI to run over and over until the Abort Execution button LabVIEW simplifies the tasks of communicating with
is pushed. GPIB, USB, serial, Ethernet, VXI, and PXI instruments. For
690 Modeling Measuring Systems
of experiments. One of them is used to control X-
rays diffractometer (Figure 4) and the other to perform
experiments devoted to the statistical character of gamma
decay investigations (Figure 7).
The presented graphical user interfaces are intuitive
in operation and simple to apply. Designing of similar
interfaces using a traditional computer language can take
many days. With LabVIEW, this can be done in an
extremely short time, even with very little experience in
G-programming.
LabVIEW terminology and the block diagram represen-
tation of VIs (Figure 8) are easy to interpret and natural
to use by scientists and engineers. Typical advanced oper-
ations usually needed for communication or analysis have
been encapsulated in VI libraries for quick application.
The productivity with LabVIEW is significantly higher,
and generally, a system’s development time is much
reduced compared to traditional programming. For exam-
Figure 7. The front panel of the poisson distribution analyzer as
a web page. ple, the LabVIEW flexibility allowed rapid modification of
the virtual diffractometer (Figure 4) to then make it appli-
example, the Metex M-3660D multimeter uses the built-in cable (Figure 7) to measurements devoted to the statistical
RS 232 port for serial communication, and with the Pt100 character of gamma decay investigations. The modifica-
(see also Article 197, Metallic Resistance Temperature tion can be made easily because both instruments use the
Detectors (RTDs), Volume 3) element, it can be used as a same subVIs needed to control gate time, collect data in the
digital thermometer controlled by a computer in thermody- time limited by the gate, present acquired data on the XY
namics experiments. The GPIB (General Purpose Interface graphs, and communicate through serial RS 232 interface
Bus) interface is optionally installed in the Hameg HM-507 with peripheral instruments.
oscilloscope and can be use to communicate between a PC LabVIEW also improves the teaching process and the
and the HM-507. It is relatively easy to build applications way by which students learn in universities. The appli-
using LabVIEW to control these real-world instruments. cations used during the laboratory work help students to
understand better how an experiment is running because
VIs can provide a lot of useful information that is quickly
5.5 Data analysis with LabVIEW available from the menu bar.
LabVIEW provides more than 400 built-in functions for
extracting useful information from acquired data and
for signal processing and analysis. Tools available from
advanced libraries can be used for solving systems of linear
algebraic and ordinary differential equations; for integration
and differentiation (with G Math Toolkit); for probability
and statistics analysis; for curve fitting; and for signal gen-
eration and signal processing with DFT (Discrete Fourier
Transform) and FFT (Fast Fourier Transform) analysis (see
also Article 143, DFT and FFTs, Volume 3), windowing
and filtering.
6 RELEVANCE TO PRESENTED
APPLICATIONS
In Figures 4 and 7 are shown two front panels of
the exemplary VIs built to control the different types Figure 8. The block diagram of the X-rays diffractometer.
Modeling with LabVIEW 691
7 COMMUNICATION AND DATA selected item placed on the global front panel. Figure 9(b)
TRANSFER INSIDE AND OUTSIDE VI shows how data passes between the VI B and the VI C
using global variables.
7.1 Local variables
7.3 Serial communication
A local variable can be used to represent the front panel
object in several locations on the block diagram of a The program developed for communication between serial
single VI. It can also be used to pass data between instruments using Virtual Instrument Software Architecture
nodes that are not connected by a wire. It is necessary (VISA) is shown in Figure 9(c). The VISA Configure Serial
to use local variables to control several While Loops Port initializes the selected port to the specified settings.
by using one switch. The local variable can be created VISA Write writes a value in a string to the port. VISA
from the shortcut menu of the object or its terminal. Read reads the number of characters specified by the
Figure 9(a) shows how data passes in the VI A using local requested byte count from the port after a timeout period
variables. A local variable always contains the value of its controlled by the Property Node. VISA Close is used to
associated terminal. terminate communication.
The sequence of the operations in this program is
determined by wiring the output terminals to the input
7.2 Global variables
terminals used for addressing in the line of nodes. GPIB
(see also Article 170, IEEE-488 Instrumentation Bus,
Global variables can pass data between several VIs. A
Volume 3) communication can also be performed using
global variable can be used to represent the front panel
the VISA.
object in several locations on the block diagrams of the
several VIs. The global variable can be created by dragging
the Glob from the All Structures panel and then dropping 7.4 In/out port communication
it on the block diagram. From the shortcut menu of the
global variable, the front panel of the special kind of VI Most of the peripheral devices in the PC are controlled
has to be opened first and the controls that define the data using the I/O ports. Figure 9(d) shows the VI built for
types accepted by this global variable have to be placed data loading to counter card using Out Port functions and
on it. A global variable always contains the value of the Sequence structure. The first frame of the Sequence is used
(a) (c)
(b) (d)
Figure 9. Communication and data transfer inside and outside VI.
692 Modeling Measuring Systems
for the Control Byte writing to the card through the Out Port checked/unchecked by setting the respective attributes. The
before the initial count is written. Then, the initial count in built menu can be viewed in the Preview area. A tag
two bytes is loaded to the card, LSB byte followed by the identifies the custom item in the block diagram from the
MSB byte. output of the Get Menu Selection function (Figure 10b).
The current VI’s Menu bar returns the menu bar refnum of
the current VI. The Item Tags are strings and can be used,
8 VI BEHAVIOR CUSTOMIZING for example, to select the subVIs or functions placed in the
After the VI building, a user can customize the VI Menu bar different frames of the Case structure.
by its editing and the VI behavior using the VI Properties The VI documentation, user’s manual, and so on, can be
dialog boxes. selected from the custom menu bar, for example, to instruct
a user on how the instruments can be executed and how the
measurements can be performed.
8.1 Customizing the menu bar
The custom menu template can be built at the time 8.2 Customizing the VI properties
of editing after opening the Menu Editor window by
selecting the Run-Time Menu from the default menu bar. By popping up the VI icon in the top right of the front panel
In Figure 10(a), for example, the Menu Editor window corner and selecting the VI Properties, a user can open the
for the virtual spectrum analyzer (VSA) is shown (see dialog boxes and turn on/off all the properties available for
also Article 106, Virtual Instrumentation in Physics, the current VI.
Volume 2). On the left-hand side of this window, the menu From the Window Size category, for example, the width
hierarchy is displayed. The tools needed for the menu and height of the front panel window can be determined.
hierarchy editing are available from the Menu and Tools From the Execution category, the dialog box, used to turn
bars. After the highlighting of the menu item in the menu on/off the execution options, and the Window Appearance
hierarchy, the item properties appear on the right-hand side category (Figure 10c), used to turn on/off the window
of the window. A user can select the item type, write the appearance options available for the current VI, can be
item name and tag identifier, and define for it the shortcut accessed. The window options apply to VI when in the
key combination. A user item can be enabled/disabled and execution mode.
(a) (b)
(c) (d)
Figure 10. VI customizing windows.
Modeling with LabVIEW 693
The customizing of VI also needs the icon editing, which larger task. The concept of the modular programming, pro-
can be done in the Icon Editor window (Figure 10d) by moted by LabVIEW, gives a user possibility to execute each
accessing the pop up on the VI icon and then selecting the subprogram by itself, making debugging relatively easy.
Edit Icon. Furthermore, many low-level subprograms often perform
tasks common to different applications and can be used
independently by every individual application.
9 MODULARITY AND HIERARCHY The hierarchy nature is also the significant characteristic
of LabVIEW and VIs. From the VI Hierarchy window
VIs are modular in design, so any VI can be used as the (Figure 11a), a user can simply find the file path for every
top-level program and can run by itself or as a subpro- subVI, open it selectively by double clicking, and customize
gram within other programs. A programmer can divide an individually. There is an unlimited number of layers in
application into a set of simple subtasks and then build the hierarchy structure of every VI. Each subVI can be
VIs to accomplish all the subtasks and finally combine also opened from the VI Browse available in the menu bar
those VIs on the top-level block diagram to complete the (Figure 11b).
(b)
(a)
Figure 11. The VI hierarchy window.
Data flow direction
Figure 12. The data flow principle presentation.
694 Modeling Measuring Systems
10 DATA FLOW PRINCIPLE and appears when the frame containing it is selected. The
PRESENTATION button names can be edited. By default, they are Yes, No,
or Cancel. The programmer can replace them by wiring the
string constants with the needed names to the corresponding
10.1 The Data Flow principle
dialog function inputs.
The One/Two Button Dialog functions return a value
LabVIEW is a programming environment based on the Data
of True or False depending on which button is pushed.
Flow principle. In practice, it means that the nodes execute
This output value can be used to select the following VI
only when they receive all required input data and produce
behavior.
the output data immediately when they execute. Because the
The Three Button Dialog .vi is a more advanced function
path of the data flow on the block diagram is determined by
and can be edited in a more advanced form. A programmer
the wire connection, the wiring can be used to control the VI
can use one, two or three buttons for the application–user
execution order. Such a solution is applied in the program
communication, justify message, place the Window Title on
shown in Figure 12, presenting different types of dialog
the box, and select the keyboard shortcuts and the window
functions responsible for the messages that appear on the
close button, which allow a user to close window without
VI front panel in the order determined by wire connections
selection. The Yes, No, or Cancel buttons can be edited. The
between corresponding structures.
enumerated output of the Three Button Dialog .vi indicates
which button was selected.
10.2 Application–user communication
REFERENCES
For the application–user communication, LabVIEW offers
the One/Two/Three Button Dialog functions (Figure 12), Bishop, R.H. (1999) Learning with LabVIEW, Addison Wesley
which automatically display the messages in the situation Longman, Inc., Menlo Park, CA.
where a user needs to be instructed or has to take a Pociecha, D., Szydłowska, J., Ga̧sowska, J., Kardaś, D., Miecz-
decision about what to do next. They are available by kowski, J. and Tłaczała, W. (2002) New Mesogenic Com-
selecting the same from the Time & Dialog panel on the pounds having Fork-like or Cyclic Amide Terminal Groups.
Liquid Crystals, 29(5), 663–667.
Functions palette.
The Button Dialog function displays a dialog box that
contains a message addressed to the user and a set of the FURTHER READING
buttons, one, two, or three, on the VI front panel. The
dialog box has a standard form view and stops VI execution Johnson, G.W. (1997) LabVIEW Graphical Programming: Prac-
until the moment a user pushes selected button. A message tical Applications in Instrumentation and Control, 2nd edn,
should be prepared earlier as a string constant on the block McGraw-Hill, New York.
diagram and wired to the corresponding input of the dialog National Instruments (2004) The Measurement and Automation
function. Usually, it is placed inside the Case structure Catalog 2004, Austin.
106: Virtual Instrumentation in Physics
Wiesław Tłaczała
Warsaw University of Technology, Warsaw, Poland
1.1 Determination of the energy of γ-rays
1 Nuclear Physics Laboratory for a Student
Training with VIs 695 The exercise aims at instructing students about how the
2 Gamma-rays Intensity Analyzer – VI scintillation spectrometer works and what typical spectrums
Example 696 of γ -rays look like. Two sources, 137 Cs and 60 Co, are used
to calibrate the spectrometer. The third one, 22 Na, is then
3 Data Analysis and Presentation with
examined with respect to the energy of γ -photons emitted
LabVIEW 696
by it. The front panel of the spectrum analyzer, used in this
4 Multiuser Measuring Systems with VIs 697
real experiment to control data receiving, its analysis and
5 VIs Application to a Teaching Process 698 presentation, is shown in Figure 1.
6 Virtual Spectrum Analyzer (VSA) – VI
Example 699
7 Virtual Nuclear Physics Experiment for
e-learning 699 1.2 Nuclear decay statistical character
Further Reading 701
In this exercise, students compare the experimental distri-
butions of emitted particles with theoretical distributions
known to be true for a small number of events, which
can be described mathematically using the Poisson dis-
1 NUCLEAR PHYSICS LABORATORY tribution theory. The front panel of the Poisson distri-
bution analyzer, used in this experiment to control data
FOR A STUDENT TRAINING WITH VIs acquisition, its analysis and presentation, are shown in
Figure 2.
The Central Physics Laboratory (CPL) is where students
of the Warsaw University of Technology study the fun-
damentals of physics in practice. The Nuclear Physics 1.3 Absorption of γ-rays passing through matter
Laboratory (NPL) is a part of the CPL. The basic equip-
ments used in the NPL are six scintillation spectrometers, This exercise helps students to obtain elementary knowl-
for performing basic nuclear physics experiments. Three edge about the different types of nuclear radiation inter-
of them are presented here as examples of the use of actions with matter. The front panel of the Gamma-rays
LabVIEW . An introduction to the use of LabVIEW is Intensity Analyzer (GIA) used in this experiment to control
given in Article 105, Modeling with LabVIEW , Vol- data acquiring, its analysis and presentation, is shown in
ume 2. Figure 3.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
696 Modeling Measuring Systems
passing through matter interacts with it. Students measure
the γ -rays with respect to the thickness of Pb, Cu, and Al
by using a GIA (Figure 3). The acquired data are then used
to examine the attenuation law and calculate the absorption
coefficients.
Using the analyzer, a user can obtain the radiation
source spectrum, measure the apparatus background or the
radiation intensity by selecting the corresponding acquisi-
tion mode.
The front panel of the analyzer contains graphical objects
used for mode and parameter settings, measurement con-
trol, measurement data presentation, and registration data
display. The measurement parameters can be set using
numeric controls. With the Measurement start button, a user
Figure 1. Front panel of the VI spectrum analyzer.
can start/stop measurement. The analyzer takes data from
the counter card connected to the scintillation spectrome-
ter. The current Counts number and the average apparatus
Background value are displayed in corresponding numeric
indicators.
The current gamma intensity value is indicated by the
Color Ramp. The set of collected data is visible in the table
and presented on the XY graph. The Exponential Fitted
Curves are plotted on the same graph. Currently, after every
measurement, the absorption coefficients and their errors
are computed and presented in the string indicator as the
Absorption curve parameters. From customized pull-down
menus, the user can open programs, which help perform
measurements.
Figure 2. Front panel of the Poisson distribution analyzer. 3 DATA ANALYSIS AND PRESENTATION
WITH LabVIEW
A beam of photons passed through matter is attenuated in
intensity. The fraction of photons surviving a distance, d, is
exponential with respect to the thickness. The logarithmic
form of the attenuation law is shown in (1)
ln N (d) = ln N0 − µd (1)
where N0 –the incident intensity; d –thickness of the absor-
ber; µ–absorption coefficient.
Measurement of γ -rays’ intensity with respect to the
thickness of Pb, Cu, and Al are performed with the GIA.
Obtained data are used to examine the attenuation law and
calculate the absorption coefficients for different types of
Figure 3. Front panel of the γ -rays intensity analyzer. absorber using the Least Squares Method (LSM).
The program shown in Figure 4(a) considers the opera-
tions used for the curve fitting performed by the GIA.
2 GAMMA-RAYS INTENSITY If the data set is collected as a 2D array, like those
ANALYZER – VI EXAMPLE presented in Figure 4(b), it is first necessary to transpose
it using the Transpose 2D Array function, and then to
The absorption coefficient is a quantity, which is char- reshape it into a 1D array with the Reshape Array. Next
acteristic of the absorbing material. A beam of photons is the need to extract elements, which represent the data
Virtual Instrumentation in Physics 697
important when multiple measuring systems are needed in
the laboratory for training students. Scintillation spectrom-
(a)
eters are the base parts of the instrumentation used for data
taking in the NPL.
The analog signal from the detector is shaped by the
amplifier and then sent to the pulse height analyzer devices,
which are of different types in different experiments.
4.1 Multiuser measuring systems
(b) (c)
The multiuser measuring systems used in NPL are presented
in Figure 5. In two measuring systems, threshold discrim-
inators are used for pulse height selection. The measuring
systems used for the statistical distribution, Figure 5(a), and
gamma absorption investigations, Figure 5(c), need data
only in the logical form. The discriminators deliver a stan-
dard logic signal for every analog signal with its amplitude
higher than a given threshold. In each of the presented
Figure 4. Data analysis and presentation with LabVIEW.
multiuser system, the discriminator output signals are dis-
tributed with fan-out circuits to four computers.
set for a particular curve. For this operation, the Array In the experiments shown in Figure 5(a), the 16-bit
Subset functions and the Array Min&Max function are used counters, controlled by the microcontrollers, (see also
twice. In this way, two arrays are extracted, [di ] and [lnNi ], Article 158, Microcontroller Systems, Volume 3) are
which represent an 8-point data set for γ -rays attenuation used for data collection. The counter cards with the
for Pb. Obtained this way, the data set is used for the microcontrollers are serial instruments and communicate
curve fitting with the Exponential Fit. vi. Before fitting with with the VI through RS 232 (see also Article 168,
Linear Fit. vi, the data has to be computed with the Natural RS 232 and EIA/TIA 232 Serial Interface, Volume 3)
Logarithm function. After fitting curves to the data, they port. In this measuring system, one spectrometer delivers
can be presented on the graphs. signals continuously to four computers, but users perform
Two methods are used for this purpose. One, the classic measurements independently.
method, uses the Bundle functions to prepare two sets of In the experiments shown in Figure 5(c), counter cards
clusters (di , lnNi ), placed on the XY Graph L after building are used for data collection. Digital I/O ports control the
an array with them. counter cards by using the VI developed for gamma absorp-
The second method, Express VI method, uses the Build tion investigations. In this measuring system, one spectrom-
XY Graph subVI, which needs only to convert arrays eter supplies signals continuously to four computers, but
with the Dynamic Data Converters before wiring them to users are asked to perform measurements synchronously.
the corresponding inputs. This method was used for data
preparation before placing them on the XY Graph E. The
(a)
Linear Fit. vi also finds the slope and intercept coefficients,
(c)
and the Exponential Fit. vi finds the damping coefficient.
The slope gives the gamma absorption coefficient. The
Array Min&Max and X/YScale. Maximum property nodes
are used for X/Y Axis scaling. In Figure 4(c), two XY (b)
graphs are shown with data points and fitted curves, and
numeric indicators that display the curves coefficients.
4 MULTIUSER MEASURING SYSTEMS
WITH VIs
With virtual instruments it is easy to build relatively
low cost multiuser measuring systems. This is particularly Figure 5. Multiuser measuring systems with VIs.
698 Modeling Measuring Systems
The local area network (LAN) (see also Article 171, Local the law of black body radiation, the Planck’s Law, as shown
Area (LANs) and Wide Area Networks (WANs), Vol- in (2)
ume 3) is used for message sending that allows synchro-
nized measurements to be performed with computers work- b 1
M(λ, T ) = , (2)
ing in parallel. In these experiments, users can also obtain λ5 ehc/λkT − 1
a radiation source’s spectra. In this case, the threshold dis-
criminator is replaced with a single-channel analyzer. where b = 107 in relative units; λ–wavelength; h–Planck
Figure 5(b) shows the multiuser measuring system using constant; k –Boltzmann constant; c –speed of light in vac-
a LAN for data distribution. In this measuring system, only uum; T –temperature in Kelvin degrees.
one computer is equipped with the multichannel analyzer The block diagram, built to place on the graph the radia-
(MCA) used to take spectra. Data acquired by the MCA tion emission curves for objects at different temperatures, is
are distributed by the server-client software combination, shown in Figure 6(b). The relative radiation intensity as a
which is a communication option available in LabVIEW. function of the wavelength and temperature is plotted by the
Remote computers receive online data as soon as the implementation of (2), using the Formula Node structure.
MCA sends it to the MCA server, which then distributes The Formula Node is similar to other structures avail-
it to the MCA clients. Data acquisition is synchronized by able in LabVIEW (see also Article 105, Modeling with
the MCA, but remote users can still analyze the spectra LabVIEW , Volume 2), but instead of containing a subdi-
independently. agram, it contains one or more formula statements delimited
by a semicolon. Using the Formula Node, a user can enter
algebraic formulas directly into the block diagram. The
5 VIs APPLICATION TO A TEACHING input and output terminals of the Formula Node can be
PROCESS created by selecting it from the shortcut menu. After cre-
ating terminals, a user has to enter variable names into the
input/output boxes.
LabVIEW can be used not only for VI building of real-
The data for one point is calculated in one iteration of
world measurements but also for building applications that
For Loop 1 using the Formula Node. The data set for one
simulate real-world instrument operation, useful in the
curve is collected in the tunnel with an array symbol, during
teaching process, for example, teaching the fundamentals
calculations performed in 10 000 iterations. After For Loop
of physics or to demonstrate experiments in physics.
1 completes all iterations, the data collected in the tunnel
are provided to the XY graph using the Bundle function,
which also determines the scaling factors, such an offset –
5.1 Planck’s Law and Wien’s Displacement Law λo and a multiplier – λ. Using For Loop 2, a user can
observe one by one the set of 31 curves at the temperatures
Figure 6 shows how LabVIEW application can be used in the range 200 to 6200 K. Also, it is possible to observe
for demonstration of the Planck’s Law and the Wien’s how the peak of the black body spectral distribution shifts to
Displacement Law. When the temperature of the object is the shorter wavelengths as the temperature increases. This
known, the distribution of the radiating energy is given by relationship is described by the Wien’s Displacement Law,
(a) (b)
Figure 6. A model of Planck’s Law and Wien’s Displacement Law.
Virtual Instrumentation in Physics 699
as shown in (3) VSA menu bar. A user can also open animation programs
developed for demonstrating how scintillation spectrome-
λp T = 2897, 9, (3)
ter and electronic devices, like a scintillation detector, an
amplifier, and the MCA (Figure 7), operate.
where λp –wavelength of peak energy in microns;
With the VSA a user can virtually measure spectrums
T –temperature in Kelvin degrees.
of the γ -rays source, calibrate the spectrometer and finally,
A user can select a static or dynamic presentation by
determine energy of the photons emitted by 22 Na. The data
switching on/off the Wien’s Displacement Law button.
collected in the virtual experiment are similar to the data
Tunnels are terminals on a structure used to get data
acquired in the real physics laboratory.
into the structure and to pass data out of the structure.
Tunnels created on the For Loop or While Loop border
can be used with the Enabled Indexing for arrays or the 6.2 Virtual physics experiments
Disabled Indexing for scalars.
Several virtual physics experiments were built, using Lab-
VIEW, in the Warsaw University of Technology (WUT)
6 VIRTUAL SPECTRUM ANALYZER to demonstrate physics phenomena or the methodology of
(VSA) – VI EXAMPLE the measurements. There are nuclear physics experiments,
thermodynamics experiments based on the Ångström Law,
6.1 The virtual spectrum analyzer (VSA) experiments for the Hall effect and Curie temperature
investigations, and electric resonance measurements. These
The Virtual Spectrum Analyzer (VSA) front panel is visible experiments can be run through quickly during a lecture or
in the background of the window, with the signal converting in the ‘real’ time mode in a virtual physics laboratory for
simulation opened from the VSA menu bar (Figure 7). The training students. One of these experiments is devoted to
buttons, switches, selectors, graphs, and so on are used to determination of γ -rays energy. This experiment, likewise
select the type of spectrum to acquire, save and load data, other experiments, has its prototype in the real experiment
calibrate the spectrometer, start and stop data acquisition, existing in the CPL (see Figure 1) and simulates it.
present the spectrums, and finally present the calibration
curve. The complete VSA front panel is presented in 6.3 Determination of γ-rays energy – a virtual
Figure 9. nuclear physics experiment
The VSA not only helps a user to conduct an experiment
but also gives the full information concerning it, for exam- The experiment aims at instructing students as to how the
ple, elementary knowledge about the γ -rays interactions basic nuclear apparatus works, how the typical spectrum of
with matter, how to turn on electronics, set correct param- γ -rays looks, and how to perform measurements to deter-
eters and run experiments, how to collect and analyze data mine energy of the unknown γ -photons. Two sources, 137 Cs
then verify obtained result, and also how to prepare a lab- and 60 Co, are used to obtain calibration curve of the scintil-
oratory report. This information can be selected from the lation spectrometer. The third source, 22 Na, is then exam-
ined with respect to the energy of γ -photons emitted by it.
7 VIRTUAL NUCLEAR PHYSICS
EXPERIMENT FOR E-LEARNING
After switching on the apparatus, the parameters setting
(Figure 8(a)), and the source selection (Figure 8(b)), a user
can start to perform the virtual experiment. The experiment
for energy determination needs at first the spectrometer to
be calibrated. It can be done to measure the energy spectra
of the sources with emissions of the known energies.
7.1 Spectrum acquisition and presentation
Figure 7. LabVIEW application for the spectrum building The calibrating sources used in the virtual experiment are
137
animation. Cs and 60 Co. To perform the measurement for obtaining
700 Modeling Measuring Systems
(a) (b)
Figure 8. LabVIEW applications for the experimental setup arrangement and the source selection.
(a) (b)
(c) (d)
22
Figure 9. Spectra for the VSA calibration and the Na gamma energy determination.
the 137 Cs spectrum, a user has to select the Data acquisition identify the peaks with the aid of the isotope tables, or
mode, the Normal or Fast mode on the VSA front panel, by selecting the needed values in the Ring control labeled
then select the Data acquisition window, and finally push Energy, which is visible on the VSA front panel when the
the Cs-137 button. After some time, depending on the windows with spectrums of 137 Cs and 60 Co are selected.
acquisition mode to be selected the 37 Cs spectrum will The spectra obtained during the experiment are collected in
be obtained. Figure 9(a–c).
After selecting the channel number of the characteristic
peak on the calibrating source spectrum with the help of the
7.2 VSA calibrating and data analysis graph cursor, and selecting the corresponding energy value
in the Ring control, placed above the graph, the one point
After finishing data acquisition, performed with two cali- can be plotted on the calibration curve graph, using the
brating sources, a user can calibrate the spectrometer by Add point to plot/Fit button. A correspondence between
determining the calibration line. For this, a user has to the MCA channel number and energy can be made this
Virtual Instrumentation in Physics 701
way. With the 137 Cs and 60 Co spectrums, three such points FURTHER READING
are plotted on the graph. The VSA automatically draws
the calibration curve by taking the best fitting straight line Bishop, R.H. (1999) Learning with LabVIEW, Addison Wesley
through the points. The energy scale is indicated on the Longman, Inc., Menlo Park, CA.
left-hand side, but the MCA channel numbers are indicated Johnson, G.W. (1997) LabVIEW Graphical Programming: Prac-
on the X-Axis of the graph shown after opening the Data tical Applications in Instrumentation and Control, 2nd edn,
McGraw-Hill, New York.
analysis window.
After taking the 22 Na spectrum (Figure 9(c)) and select- National Instruments (2004) The Measurement and Automation
Catalog 2004 , Austin.
ing a corresponding peak, the energy value of the γ -rays
emitted by 22 Na can be read by using the calibration curve Tłaczała, W. and Zagórski, A. (2003) A virtual nuclear physics
laboratory for e-learning, Advances in Technology-Based Edu-
available in the Data analysis window. cation: Toward a Knowledge-Based Society, Proceedings of
The data collected in the virtual experiment are not 2 nd International Conference on Multimedia and Information
simulated. They are similar to data obtained in a real & Communication Technologies in Education, Badajoz, Spain,
physics laboratory because they are based on the real December 3–6th .
measurement results, which have been incorporated in the
software when the application was built.
107: Principles of Sensor Science
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
concerned with the design, analysis and utilization of
1 The Nature and Scope of Sensor Science 707 these machines.
2 The Central Role of Systems Engineering 708 A transducer is a device that transforms the input signal
3 Recursion and Reticulation Principles 709 of one energy form into an output of another energy form.
4 Importance of Classification in Sensor It does this in such a way that a prescribed relationship
Science 709 is maintained between the input and output signals. An
5 Sensor Systems as Information Machines for example is the mercury-in-glass thermometer in which heat
Measurement 710 energy is converted into mechanical energy, forcing the
mercury to move along the glass as it expands or contracts.
6 Holistic Design 711
To provide a wide view of sensing and transduction, it
7 Maintenance Engineering 711 is important to examine what constitutes the science of
8 Principles of Sensor Modeling 712 sensors. Sensor Science is closely akin to Measurement Sci-
9 Parametric and Nonparametric Models 713 ence (Bentley, 2004; Finkelstein and Grattan, 1994; Syden-
10 Materials Science for Sensors and Sensing 714 ham, 1982, 1983; Sydenham and Thorn, 1992; Sydenham,
Related Articles 714 Hancock and Thorn, 1992).
This article focuses upon sensors, sensing and sensing
References 714
systems. The specific goal of the article is to analyze the
elements of sensor science. An analytical approach leads
to proposals that specify the constituent parts defining the
1 THE NATURE AND SCOPE OF SENSOR ordered study of sensors.
SCIENCE The extensive literature on sensors is the principal tool
of this analysis. The article shows that Sensor Science has
Sensors and transducers are manufactured systems (Mc- a sufficiently distinct and noteworthy constitution, which
Ghee, Henderson and Sydenham, 1999), which extend the demands both greater emphasis by measurement scientists
human abilities to sense the two highly relevant problems and more recognition in the wider scientific community.
of measurement described in Article 5, Humans in the Calling the discipline under consideration Sensor Sci-
Real World, Volume 1. Hence, since they are the front-end ence may be controversial. However, there has been notable
elements in Instrumentation and Measurement Technology, success in developing specific types of sensors and sensor
or I&MT, they are of basic importance in all fields of systems as exemplified by the authoritative 8-volume trea-
science, industry, and commerce. I&MT, which may also tise edited by Göpel, Hesse and Zemel (1989), with its
be referred to as Information Technology, or IT, is that subsequent update (Baltes, Göpel and Hesse, 1996), and
field of scientific activity where humans use machines to the handbook of Fraden (1996). In addition, the excel-
extend their ability to perform the information handling lent texts by Hauptmann (1993), Sinclair (2000) and Usher
operations associated with measurement, calculation, and Keating (1996), have emphasized the importance of
communication, and control (Article 14, Hierarchical organizing the basic material of sensors. This need is also
Aspects of Measurement Systems, Volume 1). It is also strongly supported by other books grouped according to
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
708 Elements: A – Sensors
the COMETMAN energy form classification due to Stein, has been made possible by the cataloging and classification
which is described in some detail in Article 109, Structure of physical material effects (Ballentyne and Lovett, 1980;
and Energy in Sensor Systems, Volume 2. Good exam- Schubert, 1984), by sensor modeling (Finkelstein and
ples of books according to this grouping are as follows: Watts, 1983; Neubert, 1973), sensor design (Finkelstein
and Finkelstein, 1983; Cook, 1993a,b; Sydenham, 1994a,b;
• For Chemical sensors (Göpel et al., 1989; Janata, 1989; Finkelstein et al., 1993), system reliability (Cluley, 1974;
Taylor, 1996), including gas (Maskell, 1996; Moseley Hoyland, 1994; O’Connor, 2002; Ramakumar, 1993; Lyn,
and Tofield, 1987), immunological (Ngo, 1987), phar- 1996) and fault diagnosis (Patton, Frank and Clark, 1989),
maceutical (Cosofret and Buck, 1992) and food industry all of which represent a selection of these contributions.
units (Kress-Rogers, 1993; Karube, 1994); In spite of this progress, the constituent body of coherent
• For Optical sensors (Wagner, Dendliker and Spenner, principles of the discipline lack the organization necessary
1994; Agousti, 1996; Wolfbeis, 1991; Krohn, 1992; before that group of principles may be referred to as
Dadswell and Dean, 1995); a science.
• For Mechanical sensors (Bau, de Rooij and Kloeck, The purpose of this article is to address this question.
1994) such as in manufacturing (Soloman, 1994), rob- By doing so, proposals are made as to what constitutes the
otics (Everett, 1995; Ruocco, 1987), robotics and medi- science of sensors and giving an approach to its systematic
cine (Webster, 1988) as well as in mechatronics (Brad- organization is achieved.
ley et al., 1991); As shown above, there is substantial literature dealing
• For Electrical sensors, Chapter 15 in Fraden (1996); with all aspects of sensors, sensing, and sensor systems.
• For Thermal sensors (McGee, 1988; Michalski et al., Included in this area are those given specific reference
2001; Nicholas and White, 2001; Scholz and Ricolfi, above and at different points in this article.
1990);
• For Magnetic sensors (Boll and Overshot, 1990; Mae-
naka, 1994); 2 THE CENTRAL ROLE OF SYSTEMS
• For Acoustic sensors (Crocker, 1998; Lerch, 1994;
Lynnworth, 1994); ENGINEERING
• For Nuclear sensors, Chapter 14 in Fraden (1996).
Many different methods and theories are used in Sensor
In addition to the classification of sensors by the energy Science. The bubble and block diagram of Figure 1 is a
form, alternative schemes are grouped by the sensor man- good way to visualize some of these Sensor Science dis-
ufacturing technology. Examples using this approach are ciplines. In the search for these contributory disciplines an
solid state and semiconductor sensors (Gardner, 1994; Mid- important key is found in the definition that sensors are
delhoek and Audet, 1989; Sze, 1994) and their on-chip systems, which refine and extend the human faculties of
compatibility (Wolffenbuttel, 1996) as well as thin film type sensing and perception. Because sensors are systems, the
sensors (Ciureanu and Middelhoek, 1992). science of sensors must be holistic by always using the
Mention also needs to be made of smart sensors (Bar- whole-life-whole-system approach characterizing the meth-
ney, 1988; Chapman, 1995; Frank, 1996) as well as the ods of Systems Engineering described in Article 63, Out-
associated conditioning and support circuits (Carr, 1993; line of Systems Thinking, Volume 2; Article 64, Execut-
De Sa, 1990; Pallas-Areny and Webster, 1991; Regtien, ing A Measuring System Design, Volume 2; Article 66,
1992; Van Putten, 1996) – see Article 160, Smart Sensor Phases of System Life Cycle, Volume 2; Article 65, Life
System Features, Volume 3. Cycle Concept, Volume 2; Article 67, Principle of Con-
Complementing the multitude of published books about cept of Operations (ConOps), Volume 2; Article 68, Set-
sensors, there are also a large number of technical arti- ting the System Boundaries, Volume 2; and Article 69,
cles that have contributed to the overall understanding of Requirements Allocation, Volume 2 (Sandquist, 1985;
sensors. Journals where these articles appear, are Mea- Weinberg, 1975).
surement; Sensors and Actuator – A Physical, Sensors and Thus, sensors and sensor systems should be regarded as
Actuators: B Chemical; IEEE Trans on Instrumentation; information machines. This class of machine performs a
Instrument Science and Technology; and Measurement and diversity of sensing functions involving the holistic rela-
Sensor Review to name only five. tions among specific sensor structures within different
Achievements in the area of classification of sensors, energy domains for the acquisition, capture, communica-
which can be traced back to the work of Kurt Lion (1969), tion, or distribution of information.
have been built upon by Peter Stein (1969) and more The information, which is acquired about the states
recently by many other researchers. Significant progress of equilibrium and motion of solids, liquids, gases, and
Principles of Sensor Science 709
Signal
engineering Identification Design concept Spec
generation writing
Communication Calculation
Holistic Fault
design diagnosis
Measurement
Control
Information
machines Systems Maintenance
engineering and
Information logistics
theory
Sensor
Information Science
Mathematical
technology Reliability
modeling
engineering
Software
engineering
Artificial Materials
intelligence science
Figure 1. A bubble and block diagram illustrating the constituent disciplines of Sensor Science.
their constituent systems, may be carried or captured 4 IMPORTANCE OF CLASSIFICATION IN
in a variety of signal forms. This statement is similar SENSOR SCIENCE
to the definition given in Article 2, Organization of
Instrument Science, Volume 1. As both function, or ‘what A science is a systematically organized body of knowledge.
they do’, and structure, or ‘how they do’, are specified Hence, since Sensor Science possesses a corporate body of
in these definitions, the importance of considering sensors knowledge with an organized structure, it is as necessary
as embedded within sensor systems is reinforced. In other to organize it as it is to organize other scientific fields. The
words, sensors are human-made elements embedded within nature and scope of the constitution of Sensor Science may
human–machine systems, which help humans to acquire be objectively examined using the methods of taxonomy.
information, by the process of sensing, and to handle data, When applied in Sensor Science, taxonomy requires the
by performing information handling operations. identification of that series that depends upon the increase
Sensor systems possess physical, also called organic, or decrease of some variable in the scheme of ordering.
structure. The systems approach to their specification and A taxonomy of Sensor Science will be erroneous if it is
design is basic to Sensor Science. based upon its ends, as this will only lead to a cataloging
of sensors. Hence, the only legitimate scheme of sensor
classification will ensure that all of the divisions of Sensor
3 RECURSION AND RETICULATION Science are always determined by one common principle.
PRINCIPLES Indeed, only by organizing the constitution of the topic on
the basis of contributory disciplines can Sensor Science be
Important tools for use in Sensor Science, taken from arranged according to its basic nature and inherent charac-
the Systems Engineering approach, are the techniques of teristics. This will ensure that its arrangement is only based
reticulation and recursion. Reticulation is that approach upon its basic nature and inherent characteristics. It will also
to the analysis of systems, which allows them to be make sure that its organization is not based upon either its
successively subdivided down to the simplest element. ends or upon anything lying outside itself. In other words,
Recursion is that property of a procedure that allows applying classification in Sensor Systems is not about the
operations to be performed in a nested manner using the sum of their ends but about coordinating the science of
same basic method of analysis. By this means, all of the sensors so as to give it an organized or systematized struc-
subdivided portions or elements of the sensor system may ture. The systemic nature of sensors leads to the holistic
themselves be individually analyzed using the approach of approach in their classification as shown above.
Systems Engineering. It is the systems aspect of sensors, By considering the nature and scope of the disciplines
with its holistic implications, which provides the key to the constituting the taxonomy, analysis, design, and utilization
nature and scope of their science. of sensors and sensor systems, this article shows that Sensor
710 Elements: A – Sensors
Science is important in its own right. Hence, it should objectives of sensor classification are
be given the attention it deserves by all of the scientific 1. the concrete discrimination between different sensors,
community, despite the close relationship it shares with 2. the consensus regarding standards for the principles of
other aspects of Measurement Science within IT. description,
The Four Orders or Problems of Classical Taxonomy 3. the bringing of order or systematization to sen-
are considered in Article 2, Organization of Instru- sor systems.
ment Science, Volume 1. An understanding of Sensor
Science requires the application of the principles of classi- Similarly, the three functions of classification should
cal taxonomy. allow
Previous work has well established the application of 1. the organization of the means of communication and
these principles within the context of I&MT. Sensor sys- retrieval of the descriptions used for sensors,
tems, as the front-end elements in information machines, 2. the acquisition of new information in the extension of
form one of the two divisions of the machine kingdom. sensor descriptions,
These require an identifiable body of scientific principles, 3. the highlighting of unifying factors between sensors
which depend upon the holistic relations among specific but without diminishing the importance of any existing
diverse sensors and sensor systems. Classical taxonomy has differences.
always been used in the life sciences to bring order into the The materials of taxonomy in Sensor Science are the
questions of likeness between different living entities (Daly diverse types of sensors and their operating principles.
and Linsley, 1970). It is also important in the earth sciences Assembling the various sensor types is the main activ-
(Von Engelhardt and Zimmermann; 1988) for the classifica- ity of classification in Sensor Science, because it allows
tion of minerals, in technology transfer, and in Knowledge the possibility for further study. Assembling sensors into
Engineering or KE. progressively larger groups to obtain a hierarchical order-
Bringing order to Sensor Science recognizes the univer- ing by function, structure, and energy form constitutes the
sality of the four principal orders or problems of classifica- final ingredients of discrimination and ordering in Sen-
tion Article 2, Organization of Instrument Science, Vol- sor Science.
ume 1. The classification task in Sensor Science depends on
discriminational criteria using reticulation, which requires 5 SENSOR SYSTEMS AS INFORMATION
the whole system to be ‘broken down’ by a process of MACHINES FOR MEASUREMENT
subdivision.
These criteria are based upon two main approaches. In Sensors are important elements belonging to the class of
the life sciences, ordering may be built around evolutionary, Information Machines, which is made up of machines for
or phyletic, analysis. measurement, calculation, communication, and control –
Phenetic analysis, based upon physical appearance or see Article 14, Hierarchical Aspects of Measurement
structure, is also widely used in bioscience classification. Systems, Volume 1.
In the theory for classifying sensors and sensor systems, Consequently, the underlying theory of information han-
the same names may be adopted. Hence, the phyletic dling is also relevant to Sensor Science. This theory indi-
and phenetic approaches are also appropriate for machine cates that an economical, possibly minimum, number of
kingdom grouping. alphabet symbols should be coded as messages for carrying
Phyletic analysis in the life sciences is based upon genetic or capturing information.
considerations. Functional analysis, which is similar to This distinction between information carrying and infor-
evolutionary/propagative analysis, concentrates upon the mation capturing is important. While information carrying
functions performed by physical systems in the machine (i.e. communication) machines perform the primary func-
kingdom. Hence, analyzing the function of a machine tion of high fidelity information transmission, information
may be called phyletic analysis. For the same reason, capturing (i.e. measuring) machines perform the primary
structural analysis, performed using methods analogous to function of high precision information capture. As is stated
the phenetic criteria of bioscience taxonomy, may also in Article 5, Humans in the Real World, Volume 1, pre-
be called phenetic in machine kingdom classification. cision measurement may be concerned with either direct or
These represent some of the necessary modifications to inferential measurement.
the theory of bioscience taxonomy before it can be applied Both of these problems of measurement require the use
to instrumentation. of sensors. In the sensing role, the flow of information and
A taxonomy has three objectives and three functions. energy, characterizing the states of equilibrium and motion
These are also relevant to Sensor Science. Thus, the three of solids, liquids, and gases, are captured. Identification,
Principles of Sensor Science 711
essentially a problem in precision measurement, captures • signal processing (Bellanger, 1989; Ludeman, 1987).
information about those features of a system’s ability
to store, transform, transmit, or dissipate energy and/or When equipment is referred to as faulty there is an
information by a process of interrogation. implication that some abnormality in the operating con-
This discussion indicates the important contribution made ditions exists (McGhee, Henderson and Baird, 1997).
by signal design (Godfrey, 1993; Henderson, McGhee and A more general definition of faults is that there is a
El Fandi, 1997) and signal processing (Bellanger, 1989; substantial degradation in system performance. This may
Ludeman, 1987) in Sensor Science. The interrelationships be due to gradual (graceful degradation) or abrupt changes
between these contributors to Sensor Science are shown in (catastrophic failure) in the parameters of some system or
Figure 1. process parameter or malfunction of equipment causing
uncertainties in measured values.
The most common way of implementing such a philos-
6 HOLISTIC DESIGN ophy is by using some scheme of maintenance or plan for
maintenance, which will replace ‘worn out’ parts before
Sensing in real environments can only be performed with they malfunction.
real structures. These sensor forms and the systems within A second way of preventing malfunction of a process is
which they are embedded need to be purposefully designed. to provide some monitoring scheme.
It could be argued that, like Measurement Science, Sensor Thus, it appears that fault detection, fault location, fault
Science should be considered within the context of design diagnosis and fault correction are the principal functions
rather than solely analysis. constituting a monitoring scheme.
This valid point of view deserves widespread support. In short, such a monitoring scheme may be called Fault
However, another point of view concerning design needs Detection and Isolation, or (FDI). For fault diagnosis
to be expressed. It is highly unlikely that all measurement purposes, the system has three subsystems. These consist
technologists will find themselves involved in the actual of the actuators, the main structure of the plant or process
design of sensors. It is much more likely that they will and the instruments or information machines making up the
be involved in including existing sensors into new designs technical instrument subsystem. For this reason, the three
of sensing systems. Consequently, design would seem to FDI schemes are
be better addressed within the context of measurement
system design. 1. Actuator Fault Detection, or AFD, scheme,
Effective design methodologies in Sensor Science require 2. the Process Fault Detection, or PFD, scheme,
a combination of design processes, which involve a number 3. Instrument Fault Detection, or IFD, scheme.
of phases (Finkelstein and Finkelstein, 1983). These include
(i) a feasibility assessment, (ii) a specification formulation,
Sensor Fault Detection, or SFD, constitutes an important
which may use KE in the same way as (iii) design
part of IFD. In the case of SFD, the field of both sensor
concept generation. Further knowledge on how systems
validation (Henry and Clarke, 1993) and in situ testing
engineering impacts engineering detail design is found in
are the focus of some interest (Kerlin, Hashemian and
Sydenham (2003).
Petersen, 1982; Michalski et al., 1993; McGlone, McGhee
and Henderson, 1998).
Although many diagnostic systems are in use, they only
7 MAINTENANCE ENGINEERING
provide testing for most of the sensing conditioners up
to the point where the sensor signal is available. At the
Maintenance engineering, an important branch of all engi-
neering activity, is also essential for the efficient day-to-day moment, it is unusual to find sensor system diagnostics
activities in Sensor Science where it is now receiving the where the sensor is one of the system elements to have
attention it deserves. Three important aspects of this tech- both its static and dynamic behavior tested. In the search
nology are for more reliable systems, the importance of whole-system-
each-element-in-situ-testing will almost certainly lead to
• hardware (Cluley, 1974; Hoyland, 1994; Lyn, 1996; more sophisticated fault diagnosis and prediction. Detection
O’Connor, 2002; Ramakumar, 1993) and software reli- and prediction of faults uses similar processes as found in
ability engineering (Henderson et al., 1991; Lyn, 1996); system safety assessment – see Article 99, Safety Organi-
• fault diagnosis and prediction (Henry and Clarke, 1993; zation, Volume 2; Article 100, Safety Analysis Methods,
McGhee, Henderson and Baird, 1997; Patton, Frank and Volume 2; and Article 101, Assessing and Demonstrating
Clark, 1989); Safety, Volume 2.
712 Elements: A – Sensors
8 PRINCIPLES OF SENSOR MODELING In order to simplify the process of modeling, it is
necessary to apply methods of abstraction, which have
Since sensing essentially ascribes numbers to physical vari-
five discriminable levels depending upon the purpose of
ables in such a way as to describe both them and the
the abstraction. The three main processes of abstraction
relations between them, it is necessary to develop quan-
allow increasing generality, objective approximation, or
titative methods to describe these numerical assignations.
reduction, and perspectivization, or partial viewing.
Although sensors must be considered within the wider con-
In most cases of sensor modeling, it is necessary to
text of Information Technology, there are aspects of their
conduct some kind of decomposition so that one large
analysis and design, which require a specialized, unique
problem to be solved is broken down into a number of
treatment. An important example illustrating this assertion
smaller more manageable and presumably more easily
is the contribution made by the theory of modeling (Abdul-
solved problems.
lah et al., 1994).
Formalization is another important aspect of the model-
Sensor models also allow a deeper understanding of
ing process where formal models take the form of linguistic
sensor behavior to be gained. Sensor models were orig-
groups, pictorial representations, functional representations,
inally visualized as extensions of those widely used for
or mathematical models. Since all models require valida-
most other dynamic systems (Finkelstein and Watts, 1983;
tion, this is an important step in all modeling activity.
Neubert, 1973). Although this approach was very useful
The modeling process, which may also be summarized
for the visualization of the state and rate variables, it has
as shown in Figure 2, covers the three main problems
some shortcomings in the representation of the states of the
associated with the
local mass element in a material and how they relate to
the environmental states. The only approach that accepts • system, plant or process under measurement,
this possibility uses mathematical models to describe the • goal or aim of the measurement,
flow of energy and information (Kwaaital, 1993; Van Duyn, • result of the measurement.
1993; Van Duyn and Middelhoek, 1990; Ylilammi, 1989).
Developments in this area have shown that sensor model- The central focus is placed upon the sensor whose
ing should be based upon the three laws of thermodynamics dynamic properties are to be modeled and determined. This
(Van Duyn, 1993). element, which is characterized by its input and output, may
This indicates that as the acquisition and communication be visualized by humans who represent a valuable source
of information are impossible without the flow of energy of prior knowledge about the sensor.
or matter, considerations of energy form and material type This knowledge allows the proposal of appropriate math-
are essential. ematical models as well as the design of appropriate model
The diversity of energy forms in which systems are man- validation experiments.
ifest are described in Article 109, Structure and Energy Deriving the model of most dynamic sensor systems is
in Sensor Systems, Volume 2. These may be summarized intimately connected with the principles of conservation of
using the COMETMAN acronym. Consequently, the under- mass, energy, and momentum in time and space.
lying nature of all physical effects in diverse materials, and Consequently, modeling starts with the equations describ-
in whatever energy form, must play an essential role in all ing this balance. These relations are typically nonlinear,
aspects of the modeling of sensors and their systems. The partial differential equations. Since these equations are dif-
concepts of intensive and extensive state variables may be ficult to solve, they are not really of much practical use.
used to describe the equilibrium relations of the local mass Most practical systems are so complex that it is difficult
element and the manner in which these are coupled to to find an accurate model to describe the behavior of the
the environmental conditions respectively. In most cases, process by physical modeling.
this representation is useful for implementing computer- To give models that are practically useful and still
based solutions for the analysis of sensor design candidates. realistic, the experimenter needs to make some well-
However, it lacks the transparency necessary for deep and informed assumptions for the purposes of simplification.
simplified qualitative understanding. Nevertheless, the com- The first of these steps is the routine of linearization.
plexity of the model should be of such a level as to be of By assuming typical average operating conditions, all of
practical use. the parameters are represented by nominal values at the
The process of modeling consists of six main steps, operating points of the process. Parameters may be material
which may form a recursive process. When the important properties, environmental conditions, and other physical
objectives of the model are fixed, it is then necessary to influences. In most industrial systems, it is possible to
gather whatever knowledge is appropriate for the model- recognize some geometrical property of its structure. Such
ing purpose. a property can be employed usefully by using the ideas
Principles of Sensor Science 713
A priori
knowledge
Design process
A
Experiment
design
System,
plant or
process
Data measurement
Data acquisition
Data processing
Balance of mass Market forces
and energy in Data reduction Research
Measurement process
time and space Data correction Development
Modeling process
Design goals
Measurement knowledge
Structural knowledge
Simplifications
(a posteriori)
-linearization
Time Frequency
-shape
(a priori)
domain domain
-model reduction
analysis analysis
Parametric
or explicit
Coefficient of Coefficient of
a differential or a rational
Model structure
difference equation transfer function
Nonparametric
or implicit
Response to Bode and Nyquist
(a) impulse diagrams
(b) step Nicholls chart
Repeat from A Model verification
Figure 2. Steps in the modeling problem.
of shape simplification. Examples of this occur when found in many places – see Article 71, Modeling Metho-
there may be cylindrical symmetry in the real physical dology, Volume 2; Article 114, Models of the Sensor
object. Interface, Volume 2; and Article 105, Modeling with
Having linearized the model and introduced geometrical LabVIEW , Volume 2.
features in model simplification, the resulting mathematical
model may still be difficult to understand and use.
In such a case, the experimenter may be forced to 9 PARAMETRIC AND NONPARAMETRIC
apply a process of model reduction. Model reduction MODELS
usually replaces linear partial differential equations
with constant coefficient ordinary differential equations. Mathematical models obtained by the process described
McGhee, Henderson and Sydenham (1999) apply these above have parameters that depend on physical materials
techniques to the modeling of a resistance thermometer in used in the structure of the system. Hence, they are called
its sheath. Modeling in measurement systems pursuits is parametric models or explicit models.
714 Elements: A – Sensors
It often happens that a model based on physical behavior manufacture or processing and so on (Cahn, Haasen and
contains a number of unknown parameters even if the Kramer, 1993; de Podesta, 1996; Gardner, 1994) – see
structure is derived from physical laws. Experimentation, Article 117, Materials in Measuring Systems, Volume 2.
which can help by inferring the values of these parameters Because of the vast amount of published information
from measurements, is an essential part of the model on materials, it is essential to discriminate between those
validation process. general aspects of materials science and the more specific
The two most common parametric models refer to either requirements of sensor science. This may be achieved using
the time domain description of the system, normally in a scheme of classification based upon the COMETMAN
the form of systems of linear difference equations, or the classification of measurand energy form.
frequency domain description as rational transfer functions. It has been noted above that Knowledge Engineering,
Other types of models make a minimum number of or KE (Charniak and McDermott, 1985; Finkelstein, 1994)
assumptions appropriate for the purpose of the model. is an area of growing importance in Sensor Science. It is
These models are generally referred to as nonparametric likely that this has much promise for the resolution of many
models or implicit models. As with parametric models, cognitive issues in Sensor Science even though it is an
nonparametric models may characterize the behavior of the area where there are a number of outstanding problems.
system in either the time domain or the frequency domain. Some proposals concerned with these issues have been
The impulse response of a linear physical system, which made (Cropley, 1994, 1997; Cropley and Cropley, 1994;
is a statement of the time domain output of the system Ferris and Sydenham, 1994) – see also Article 160, Smart
when the input is an impulse, is of basic importance. Sensor System Features, Volume 3.
If the impulse response is known, either explicitly, from
a parametric model, or implicitly, from a nonparametric
model, then the response of the system to any other type RELATED ARTICLES
of input signal may be predicted. The frequency domain
Article 2, Organization of Instrument Science, Vol-
equivalent of the impulse response is called the frequency
ume 1; Article 6, Substructure of Human–Machine Sys-
response function sometimes simply called the frequency
tems, Volume 1; Article 16, Reduction of Influence Fac-
response. This model is a representation of the way in
tors, Volume 1; Article 108, Transducer Fundamentals,
which a system responds to sinusoidal signals of different
Volume 2; Article 111, Classification of Sensors, Vol-
frequencies. Consequently, when the system is linear, it
ume 2; Article 112, Systematic Description of Sensors,
can be used to predict the response of the system to any
Volume 2; Article 116, Selection of Sensors, Volume 2.
signal that can be modeled by a number of composite
frequency components.
When the process of modeling has reached an appropriate REFERENCES
stage, the next stage involves the problems of identification.
Identification is concerned with the synthesis and analysis Abdullah, F., Finkelstein, L., Khan, S.H. and Hill, W.J. (1994)
of signals used in inferential measuring instruments. Infer- Modelling in Measurement and Instrumentation – An Over-
ential measurement, also called identification, is defined as view. Measurement, 14(1), 41–45.
the second problem of measurement in Article 5, Humans Agousti, A.T. (1996) Introduction to Fiber Optic Sensors, Chap-
in the Real World, Volume 1 and also by McGhee et al. man & Hall.
(2001). DSP is widely used in identification. Ballentyne, D.W.G. and Lovett, D.R. (1980) A Dictionary of
Named Effects and Laws in Chemistry, Physics and Mathemat-
ics, Chapman & Hall, London.
10 MATERIALS SCIENCE FOR SENSORS Baltes, H. Göpel, W. and Hesse, J. (eds) (1996) Sensors Update
AND SENSING (in 2 Volumes), VCH Publishers, Weinheim.
Barney, G.C. (1988) Intelligent Instrumentation, Prentice Hall,
From thermodynamic modeling, it has become even more UK.
apparent that Materials Science plays a major role in Bau, H.H., de Rooij, N.F. and Kloeck, B. (1994) Mechanical
Sensor Science, quite apart from the pivotal role it plays Sensors, Vol 7 of Göpel, W. et al. (1989).
in the general science of machines. This role, although Bellanger, R. (1989) Digital Processing of Signals, John Wiley &
predominantly concerned with those physical effects that Sons, Chichester.
may be used to realize the sensing or transduction function, Bentley, J.P. (2004), Principles of Measurement Systems,
is also important from other points of view. Examples Prentice-Hall.
are mechanical rigidity or flexibility, chemical inertness Boll, R. and Overshot, K.J. (1990) Magnetic Sensors, Vol 5 of
or sensitivity in adverse environments, relative ease of Göpel, W. et al. (1989).
Principles of Sensor Science 715
Bradley, D.A., Dawson, D., Burd, N.C. and Loader, A.J. (1991) Finkelstein, L., Ginger, R., El-Hami, M. and Mirza, M.K. (1993)
Mechatronics, Chapman & Hall, New York. Design Concept Generation for Instrument Systems – A
Cahn, R.W., Haasen, P. and Kramer, E.J. (eds) (1993) Materials Knowledge Based Approach. Measurement, 11(1), 45.
Science and Technology: A Comprehensive Treatment (in 18 Fraden, J. (1996) Handbook of Modern Sensors: Physics, Designs,
Volumes), VCH Publishers (UK) Ltd., Cambridge, MA. and Applications, American Institute of Physics, Woodburg,
Carr, J.J. (1993) Sensors, Transducers and Measurement, Prentice New York.
Hall, Englewood Cliffs, NJ. Frank, R. (1996) Understanding Smart Sensors, Artech House,
Chapman, P. (1995) Smart Sensors, ISA, Research Triangle Park, Norwood, MA.
NC. Gardner, J.W. (1994) Microsensors: Principles and Applications,
John Wiley & Sons, Chichester.
Charniak, E. and McDermott, D. (1985) Introduction to Artificial
Intelligence, Addison-Wesley, Reading, MD. Godfrey, K.R. (ed.) (1993) Perturbation Signals for System Iden-
tification, Prentice Hall, Hemel Hempstead.
Ciureanu, P. and Middelhoek, S. (eds) (1992) Thin Film Resistive
Sensors, A. Hilger, Bristol, CT. Göpel, W., Hesse, J. and Zemel, J.N. (1989) Sensors: A Compre-
hensive Survey (in 8 Volumes), VCH Publishers, Cambridge,
Cluley, J.C. (1974) Electronic Equipment Reliability, MacMillan,
MA.
London.
Göpel, W., Jones, T.A., Kleitz, M., Lundström, I. and Seiyama, T.
Cook, S.C. (1993a) A Methodology for Formulating Measuring
(eds) (1989) Chemical and Biochemical Sensors, Vols 2 and 3
Instrument Specifications. Measurement, 11(1), 1.
of Göpel, W. Hesse, J. and Zemel, J.N.
Cook, S.C. (1993b) A Knowledge-Based System for Computer- Hauptmann, P. (1993) Sensors: Principles and Applications,
Aided Production of Measuring Instrument Specifications. translated by T. Pownall, Prentice Hall, Hemel Hempstead.
Measurement, 11(3), 235.
Henderson, I.A., McGhee, J. and El Fandi, M. (1997) Data Mea-
Cosofret, V.V. and Buck, R.P. (1992) Pharmaceutical Applica- surement, ISBN 0 9531409 0 3, Industrial Control Centre,
tions of Membrane Sensors, CRC Press, Boca Raton, FL. University of Strathclyde, Glasgow.
Crocker, M.J. (ed.) (1998) Handbook of Acoustics, John Wiley & Henderson, I.A., McGhee, J., Szaniawski, W. and Domaradzki, P.
Sons, Chichester. (1991) Incorporating High Reliability into the Design of
Cropley, D.H. (1994) Qualitative Analysis and Optimisation of Microprocessor Based Instrumentation. IEE Proceedings A,
Measurement Systems, ACTA IMEKO XIII, 2, 849. 138(2), 105.
Cropley, D.H. (1997) Information and Semiotics in Measurement, Henry, M.P. and Clarke, D.W. (1993) The Self-Validating Sen-
PhD thesis, University of South Australia. sor: Rationale, Definitions and Examples. Control Engineering
Cropley, D.H. and Cropley, A.J. (1994) Cognition in Electronic Practice, 1(4), 585–610.
Sensing Systems from the Viewpoint of Psychology, ACTA Hoyland, A. (1994) Reliability Theory: Models and Statistical
IMEKO XIII, 2, 843. Methods, John Wiley & Sons, New York.
Dadswell, N.G. and Dean, R. (1995) Optical Sensors for Power Janata, J. (1989) Principles of Chemical Sensors, Plenum Press,
Industry Applications, ERA Technology, Leatherhead. New York.
Daly, H.V. and Linsley, E.G. (1970) Taxonomy, in Encyclopaedia Karube, I. (1994) On-line Sensors for Food Processing, Gordon
of the Biological Sciences, 2nd edn (ed. P. Gray), Van Nostrand and Breach, Switzerland.
Reinhold, New York, 920. Kerlin, T.W., Hashemian, H.M. and Petersen, K.M. (1982)
de Podesta, M. (1996) Understanding the Properties of Matter, Response Characteristics of Temperature Sensors Installed in
UCL Press Limited, London. Processes, ACTA IMEKO.
De Sa, A. (1990) Principles of Electronic Instrumentation, Edward Kress-Rogers, E. (ed.) (1993) Instrumentation and Sensors for the
Arnold, London. Food Industry, Butterworth-Heinemann, Oxford.
Everett, H.R. (1995) Sensors for Mobile Robots, A.K. Peters. Krohn, D.A. (1992) Fiber Optic Sensors: Fundamentals and
Applications, 2nd edn, ISA, Research Triangle Park, NC.
Ferris, T.L.J. and Sydenham, P.H. (1994) Epistimological Aspects
of Measurement, ACTA IMEKO XIII, 2, 865. Kwaaital, Th. (1993) The Fundamentals of Sensors. Sensrs and
Actrs, 39, 103–110.
Finkelstein, L. (1994) Intelligent and Knowledge Based Instru-
mentation – An Examination of Basic Concepts. Measurement, Lerch, R. (1994) Sensors for Measuring Sound, Chapter 16 of Bau
14(1), 23. et al. (1994).
Finkelstein, L. and Finkelstein, A.C.W. (1983) Review of Design Lion, K.S. (1969) Transducers: Problems and Prospects. IEEE
Methodology. Proceedings of IEE Pt A, 130(4), 213. Transactions, IECI-16, 2–5.
Finkelstein, L. and Grattan, K.T.V. (1994) Concise Encyclopae- Ludeman, L.C. (1987) Fundamentals of Digital Signal Processing,
dia of Measurement and Instrumentation, Pergamon, Oxford. John Wiley & Sons.
Finkelstein, L. and Watts, R.D. (1983) Fundamentals of Trans- Lyn, M.R. (ed.) (1996) Handbook of Reliability Engineering,
ducers – Description by Mathematical Models, in Handbook of McGraw-Hill, New York.
Measurement Science, Vol. 2 (ed. P.H. Sydenham), John Wiley Lynnworth, L.C. (1994) Ultrasonic Nonresonant Sensors, Chap-
& Sons, Chichester (pp. 747–793). ter 8 of Bau et al. (1994).
716 Elements: A – Sensors
Maenaka, K. (1994) Hall Sensors, Chapter 6 of Bau et al. (1994). Schubert, J. (1984) Physikalische effekte, Physika-Verlag, Wein-
Maskell, W.C. (1996) Gas Sensors, Chapman & Hall. heim, BDR.
McGee, T.D. (1988) Principles and Methods of Temperature Sinclair, I.R. (2000) Sensors and Transducers, 3rd edn, Butter-
Measurement, John Wiley & Sons, New York. worth-Heinemann, Oxford.
McGhee, J., Henderson, I.A. and Baird, A. (1997) Neural Net- Soloman, S. (1994) Sensors and Control Systems in Manufactur-
works Applied for the Identification and Fault Diagnosis of ing, McGraw-Hill, New York.
Process Valves and Actuators. Measurement, 20(4), 267–275. Stein, P.K. (1969) The Engineering of Measurement Systems.
McGhee, J., Henderson, I.A. and Sydenham, P.H. (1999) Sensor Journal Metals, 231, 40.
Science – Essentials for Instrumentation and Measurement Sydenham, P.H. (ed.) (1982) Handbook of Measurement Sci-
Technology. Measurement, 25, 89–113. ence, Vol 1 Theoretical Fundamentals, John Wiley & Sons,
Chichester.
McGhee, J., Henderson, I.A., Kulesza, W. and Korczynski, M.J.
(2001) Measurement Data Handling: Theoretical Technique, Sydenham, P.H. (ed.) (1983) Handbook of Measurement Science,
ISBN 83-7283-007-X, The Technical University of Lodz, Lodz. Vol 2 Practice Fundamentals, John Wiley & Sons, Chichester.
McGlone, P., McGhee, J. and Henderson, I.A. (1998) Perfor- Sydenham, P.H. (1994a) Disciplined Design of Sensor Systems –
mance Evaluation of a Differential Pressure Cell, IEEE Trans- Part 1. Types of Design. Measurement, 14(1), 73–80.
actions on Instrumentation and Measurement, 47(5) Sydenham, P.H. (1994b) Disciplined Design of Sensor Systems –
1271–1276. Part 2. Knowledge Based Systems in Sensor and Sensor Sys-
Michalski, L., Eckersdorf, K., Kucharski, J. and McGhee, J. tems Design. Measurement, 14(1), 81–88.
(2001) Temperature Measurement, 2nd edn, John Wiley & Son, Sydenham, P.H. (2003) Systems Approach to Engineering Design,
Chichester. Artech House.
Michalski, L., Eckersdorf, K., Kucharski, J., Sankowski, D. and Sydenham, P.H. and Thorn, R. (eds) (1992) Handbook of Mea-
Urbanek, P. (1993) Experimental Computerised In-situ Identi- surement Science, Vol 3 Elements of Change, John Wiley &
fication of Thermocouple Sensors, in Proceedings of TEMP- Sons, Chichester.
MEKO 93, 5th International Symposium on Temperature and Sydenham, P.H., Hancock, N.H. and Thorn, R. (1992) Introduc-
Thermal Measurement in Industry and Science, Prague (pp. tion to Measurement Science and Engineering, Wiley.
107–111).
Sze, S.M. (ed.) (1994) Semiconductor Sensors, John Wiley &
Middelhoek, S. and Audet, S. (1989) Silicon Sensors, Academic Sons, New York.
Press, London.
Taylor, R.F. (1996) Handbook of Chemical and Biological Sen-
Moseley, P.T. and Tofield, B.C. (eds) (1987) Solid-State Gas sors, Adam Hilger, Bristol, CT.
Sensors, Adam Hilger, Bristol, CT.
Usher, M.J. and Keating, D. (1996) Sensors and Transduc-
Neubert, H.K.P. (1973) Instrument Transducers, 2nd edn, Claren- ers: Characteristics, Applications, Instrumentation, Macmillan,
don Press, Oxford. Basingstoke.
Ngo, T.T. (ed.) (1987) Electrochemical Sensors in Immunological Van Duyn, D.C. (1993) Multi-Signal Domain Modelling of Solid-
Analysis, Plenum Press, New York. State Transducers, Delft University Press, Delft.
Nicholas, J.V. and White, D.R. (2001) Traceable Temperatures, Van Duyn, D.C. and Middelhoek, S. (1990) Information Trans-
John Wiley & Sons, Chichester. duction in Solid-State Transducers: A General Thermo-Dyna-
O’Connor, P.D.T. (2002) Practical Reliability Engineering, 4th mic Systems Approach. Sensrs and Actrs, A-21-A23, 25.
edn, John Wiley & Sons, Chichester. Van Putten, A.F.P (1996) Electronic Measurement Systems: The-
Pallas-Areny, R. and Webster, J.G. (1991) Sensors and Signal ory and Practice, Institute of Physics, London.
Conditioning, John Wiley & Sons, New York. Von Engelhardt, W. and Zimmermann, J. (1988) Theory of Earth
Patton, R.J., Frank, R.N. and Clark, R.N. (1989) Fault Diagnosis Science, Cambridge University Press, Cambridge, MA.
in Dynamic Systems: Theory and Applications, Prentice Hall, Wagner, E., Dendliker, R. and Spenner, K. (eds) (1994) Optical
Englewood Cliffs, NJ. Sensors, Vol 6 of Göpel, W. et al. (1989).
Ramakumar, R. (1993) Engineering Reliability: Fundamentals Webster, J.G. (ed.) (1988) Tactile Sensors for Robotics and
and Applications, Prentice Hall, Englewood Cliffs, NJ. Medicine, John Wiley & Sons, New York.
Regtien, P.P.L. (1992) Instrumentation Electronics, Prentice Hall, Weinberg, G.M. (1975) An Introduction to General Systems Think-
Hemel Hemstead. ing, John Wiley & Sons, New York.
Ruocco, S.R. (1987) Robot Sensors and Transducers, Open Uni- Wolfbeis, O.S. (ed.) (1991) Fiber Optic Chemical Sensors and
versity Press, Milton Keynes, MA. Biosensors, CRC Press, Boca Raton, FL.
Sandquist, G.M. (1985) Introduction to Systems Science, Prentice Wolffenbuttel, R.F. (ed.) (1996) Silicon Sensors and Circuits: On-
Hall, Englewood Cliffs, NJ. chip Compatibility, Chapman & Hall, London.
Scholz, J. and Ricolfi, T. (1990) Thermal Sensors, Vol 4 of Ylilammi, M. (1989) Thermodynamics of Sensors. Sensrs and
Göpel, W. et al. (1989). Actrs, 18, 167–178.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
108: Transducer Fundamentals
Paul P.L. Regtien
University of Twente, Enschede, The Netherlands
so as to control that quantity. So, a more explicit defini-
1 Definition of a Transducer 717 tion of a transducer is an electrical device that converts one
2 Categorizing Sensors 718 form of energy into another, with the intention of preserving
information.
3 Terminology of Sensors 720
According to common terminology, these transducers are
4 Sensor Performance 721 also called sensor and actuator respectively (see Figure 1).
5 Role and Use of Actuators 723 So, a sensor is an input transducer and an actuator is
References 724 an output transducer. It should be noted, however, that
this terminology is not standardized. In literature, other
definitions are found. In particular, some authors make
a clear differentiation between a sensor and a (input)
1 DEFINITION OF A TRANSDUCER transducer, stressing a distinction between the element that
performs the physical conversion and the complete device,
A transducer is an essential part of any information process- for instance, a strain gauge (transducer) and a load cell
ing system that operates in more than one physical domain. with one or more strain gauges (sensor), or even vice versa.
These domains are characterized by the type of quantity Modern sensors not only contain the converting element but
that provides the carrier of the relevant information. Exam- also contain part of the signal processing (analog processing
ples are the optical, electrical, mechanical, and magnetic such as amplification and filtering, Analogue-to-Digital-
domains. A transducer is the part of the instrument that conversion and even some digital electronics). Many such
converts information about a measurand from one domain sensors have the electronics integrated with the transducing
into another, ideally without information loss. So, a trans- part onto a single chip. Present day sensors may have a
ducer can formally be defined as a device that converts one bus compatible output, implying full signal conditioning
form of energy into another, with the intention of preserving on board. There is a trend to also include transmission
information. electronics within the device, for instance, in biomedical
A transducer has at least one input and one output. In applications.
measuring instruments, where information processing is Signal conditioning may be included
performed by electrical signals, either the output or the
input is of electrical nature (voltage, current, resistance, • to protect the sensor from being loaded or to reduce
capacitance, etc.), whereas the other is a nonelectrical signal loading errors;
(displacement, temperature, elasticity, etc.). A transducer • to fit the sensor output to the input range of the
with a nonelectrical input is an input transducer, intended Analogue-to-Digital Converter (ADC);
to convert a nonelectrical quantity into an electrical signal • to enhance the S/N (signal-to-noise ratio) prior to further
so as to measure that quantity. A transducer with a non- signal processing;
electrical output is called an output transducer, intended • to generate a digital, bus compatible electrical output;
to convert an electrical signal into a nonelectrical quantity • to transmit measurement data for wireless applications.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
718 Elements: A – Sensors
sensor output energy mainly comes from this auxiliary
Nonelectrical in Electrical out
source, and just a fraction of energy is withdrawn from the
Sensor
measurement object. In this respect, modulating sensors do
not significantly load the measurement object and hence are
more accurate than direct sensors. Most sensors are of the
modulating type, for instance, all resistive and capacitive
Electrical in Nonelectrical out
sensors and many inductive sensors.
Actuator
There are many sensors on the market, over a million
types worldwide. A categorization of sensors would help
make the proper choices and also make it easier to under-
Figure 1. Input transducer (sensor) and output transducer
stand, but a useful basis for categorization is difficult to
(actuator).
define. There are various possibilities such as
In conclusion, the boundaries between sensor and trans- • according to the measurand,
ducer, as proclaimed in many sensor textbooks, are disap- • according to the conversion principle,
pearing or losing their usefulness: the user buys and applies • according to the domain of the measurand,
the sensor system as a single device, with a nonelectri- • according to application fields.
cal input and an electrical (analog, digital, bus compati-
ble) output. All of them have their limitations. For instance, the number
of measurands is rather large, making the first option not
very practical. Figure 3 gives an overview of the most
2 CATEGORIZING SENSORS common physical quantities for which sensors are available,
after Middelhoek and Noorlag (1981). However, the list is
A sensor (or input transducer) performs the conversion of not complete. Moreover, many quantities can be measured
information from the physical domain of the measurand to by a variety of sensor types. For example, position can
the electrical domain. Many authors have tried, more or be measured using resistive, capacitive, inductive, acoustic,
less successfully, to build up a consistent categorization of and optical methods.
sensors (see also Article 112, Systematic Description of The second option, according to the conversion principle,
Sensors, Volume 2). It is not easy to create a consistent is often used for the reason that the sensor performance
systematic description encompassing all sensor principles. is mainly determined by the physics of the underlying
There is at least consensus on a division into two groups of principle of operation. On the other hand, a particular
sensors: direct and modulating sensor types (see Figure 2). type of sensor might be suitable for a variety of physical
The distinguishing property is the need for auxiliary energy. quantities and in many different applications. For instance,
Direct sensors do not require additional energy for con- a magnetic sensor of a particular type could be applied as a
version. As information transport cannot exist without displacement sensor, a velocity sensor, a tactile sensor, and
energy transport, a direct sensor withdraws the output so on. For all these applications, the performance is limited
energy directly from the measurement object. As a con- by the physics of this magnetic sensor, but the limitations
sequence, loss of information about the original state of manifest in completely different ways.
the object may occur. There might also be energy loss, for A closer look at the various conversion effects may lead
instance, heat. An important advantage of a direct sensor is to the observation that the electrical output of a sensor
its freedom from offset: at zero input, the output is essen- depends either on a material property, or the geometry,
tially zero. Examples of direct sensors are the thermocouple or a movement. Figure 4 tabulates these three phenomena,
and the piezoelectric force and acceleration sensor. for various types of sensors. The figure gives the mate-
Indirect sensors or modulating sensors use an additional rial parameter, the geometric parameter, and the velocity
energy source that is modulated by the measurand; the induced parameter, together with associated sensors.
Signal in Signal out Signal in Signal out
Loss Energy in
(a) (b)
Figure 2. (a) Direct sensor and (b) modulating sensor.
Transducer Fundamentals 719
1. Mechanical, solids vibration 5. Acoustic pH
acceleration weight sound frequency polymerization
angle sound intensity degree
angular velocity 2. Mechanical, sound polarization reaction rate
area fluids sound pressure redox potential
diameter density sound velocity thermal
distance flow direction time of flight conductivity
elasticity flow velocity water content
expansion level 6. Nuclear
filling level pressure radiation 8. Magnetic and
force rate of flow ionization degree electrical
gradient viscosity mass absorption capacity
hardness volume radiation dose charge
height radiation energy current
length 3. Thermal radiation flux dielectric constant
mass enthalpy radiation type electric field
moment entropy strength
movement temperature electric power
7. Chemical
orientation thermal capacity electric resistance
cloudiness
pitch thermal conduction frequency
composition
position thermal expansion inductivity
concentration
pressure thermal radiation magnetic field
electrical
proximity strength
conductivity
rotation phase
4. Optical humidity
roughness pulse duration
color impurity
shape signal distortion
light polariaztion ionization degree
tension
light wavelength moisture
torque
luminance molar weight 9. Time
torsion
luminous intensity particle form time
velocity
reflection particle size
refractive index
Figure 3. Parameters for which sensors are on the market (after Middelhoek and Noorlag (1981)).
Type Material property Geometry Relative movement
Resistive Resistivity Relative length
(piezoresistor,LDR) (potentiometer; metal strain
gauge)
Capacitive Permittivity Relative electrode distance
(fluid level sensor) capacitive displacement
(LVDC)
Magnetic Permeability Distance source-detector Induction
(magnetoresistor) (magnetic displacement sensor) (magnetic velocity sensors)
Inductive Inductance, Induction
self-inductance, (inductive velocity sensors)
mutual inductance
reluctance
(inductive displacement sensors,
LVDT, resolver)
Optical Index of refraction; Distance transmitter−receiver Doppler frequency
absorptivity (intensity modulation sensors; (Doppler velocimeter)
(fibre optic sensors) interferometer; TOF sensor)
transmissivity, reflectivity
(optical encoder, tachometer)
Acoustic Acoustic impedance Distance transmitter−receiver Doppler frequency
(TOF displacement sensors) (Doppler velocimeter)
Piezoelectric Polarization Deformation
(piezoelectric sensors) (piezoelectric sensors)
Figure 4. Three groups of parameters, with examples of sensors.
720 Elements: A – Sensors
A categorization based on the domain of the measurand Sensors for the measurement of force and related quan-
is too coarse, and domain definitions are not unambiguous. tities are as follows:
Finally, an application field provides no restricted set
of sensors, since in each field (biomedical, automotive, Pressure sensor: measures pressure difference,
agriculture), almost all types of sensors could be applied. relative to either vacuum
(absolute pressure), a reference
pressure or ambient pressure;
Force sensor: measures the (normal and/or
3 TERMINOLOGY OF SENSORS shear) force exerted on the
active point of the transducer;
Torque sensor: measures torque (moment);
We have defined a sensor as a device that performs the Force–torque sensor: measures both forces and torques
conversion of information from the physical domain of the (up to six components);
measurand to the electrical domain. For instance, a position Load cell : force or pressure sensor, for
sensor converts position information into an electrical signal measuring weight;
(a voltage, current, etc.). Obviously, a position sensor Strain gauge: measures linear relative
measures position. However, there are different names for elongation (positive or
different applications and situations, for instance, negative) of an object, caused
by compressive or tensile
Distance sensor: measures the length of the stress.
straight line between two
defined points; Many sensors have been given names according to
Position sensor: measures the coordinates of a their operating principle or construction. Examples are as
specified point of an object in follows:
a specified reference system;
Displacement sensor: measures the change of position LVDT : or linear variable displacement
relative to a reference point; transformer, a device that is basically a
Proximity sensor: (i) determines the sign voltage transformer, with a linearly
(positive or negative) of the movable core;
linear distance between an Gyroscope: a device for measuring the angular
object point and a fixed velocity, based on the gyroscopic effect
reference point; occurring in rotating or vibrating
(ii) a contact free displacement structures;
or distance sensor for short Hall sensor: measures magnetic field on the basis of
distances (down to zero); the Hall effect, after the American
Level sensor: measures the distance of the top physicist Edwin Hall (1855–1938).
level of a liquid or a granular
substance in a container with Some sensors use a concatenation of conversion steps.
respect to a specified A displacement sensor combined with a spring can act
horizontal reference plane; as a force sensor. The measurand (force) is converted
Angular sensor: measures the angle of rotation to a displacement, which in turn is converted into an
relative to a reference electrical signal. In combination with a calibrated mass,
position; a displacement sensor can serve as an accelerometer: the
Encoder: displacement sensor (linear or measurand (acceleration) is converted into a force (by
angular) with digital output: the inertial effect), the force into a displacement (by the
Tilt sensor: measures the angle relative to spring), and the displacement into an electrical signal.
the earth’s normal; The performance of such transducers not only depends on
Tachometer: measures the rotational speed; the original sensor but also on the added components: in
Vibration sensor: measures the motion of a the case of the accelerometer, it depends on the spring
vibrating object in terms of compliance and the seismic mass respectively.
displacement, velocity or Information about a particular quantity can also be
acceleration; obtained by calculation or by additional electronic signal
Accelerometer: measures acceleration. processing using relations between quantities. The accuracy
Transducer Fundamentals 721
of the result not only depends on the errors in the quan- First, we define some common specifications that apply
tities that are measured directly but also on the accuracy to sensors:
of the parameters in the model that describes the relation
• sensitivity;
between the quantities involved. For instance, in an acous-
• nonlinearity and hysteresis;
tic distance measurement, the distance is calculated from
• resolution;
the measured time-of-flight (with associated errors) and the
• accuracy;
sound velocity. An accurate measurement result requires
• offset and zero drift;
knowledge of the acoustic velocity of the medium at the
• noise;
prevailing temperature.
• response time;
Speed and acceleration can be measured using a displace-
• frequency response.
ment sensor, by differentiating its output signal once or
twice respectively, and vice versa: by integrating the out-
put signal of an accelerometer, a velocity signal is obtained, 4.1 Sensitivity
and by a second integration, a position signal is obtained.
Obviously, the performance of the final result depends on The sensitivity of a sensor is defined as the ratio between
the quality of the signal processing. The main problem with a change in the output value and the change in the input
differentiation is the increased noise level (in particular in value that causes that output change. Mathematically, the
the higher frequency range), and integration may result in sensitivity is expressed as S = ∂y/∂x, where x is the
large drift due to the integration of offset. input signal (measurand) and y is the output (an electrical
signal). Usually, a sensor is also sensitive to changes in
quantities other than the intended input quantity, such
4 SENSOR PERFORMANCE as the ambient temperature or the supply voltage. These
unwelcome sensitivities should be specified as well, for a
Imperfections of a sensor are usually listed in the data sheets proper interpretation of the measurement result. To have a
provided by the manufacturer. These sensor specifications better insight in the effect of such unwanted sensitivities,
inform the user about deviations from the ideal behavior. it is often related to the sensitivity of the measurement
The user has to accept technical imperfections, as long as quantity itself.
they do not exceed the specified values.
Example 1 The sensitivity of a particular displacement
Any measuring instrument, and hence any sensor, has to
sensor with voltage output is specified as 10 mV mm−1 .
be fully specified with respect to its performance. Unfortu-
Its specified temperature sensitivity is 0.1 mV K−1 . Since
nately, many data sheets show lack of clarity and com-
0.1 mV corresponds to a displacement of 10 mm, the tem-
pleteness. Gradually, international agreements about for-
perature sensitivity can also be expressed as 10 mm K−1 . A
mal error descriptions are being established. An extensive
temperature rise of 5 ◦ C results in an apparent displacement
description of measurement errors and error terminology
of 50 mm.
can be found in IOS (1995). Further, there is an interna-
tional standard on transducer nomenclature and terminol- Example 2 The sensitivity of a particular type of temper-
ogy ISA (1975). Finally, various international committees ature sensor is 100 mV K−1 , including the signal condition-
are working toward a uniform framework for specifying ing unit. The signal conditioning part itself is also sensitive
sensors. to (ambient) temperature and appears to create an extra
The characteristics that describe sensor performance can output voltage of 0.5 mV for each ◦ C rise in ambient tem-
be classified into four groups: perature (not necessarily the sensor temperature). So, the
unwanted temperature sensitivity is 0.5 mV K−1 or 0.5/100
• static characteristics, describing the performance with = 5 mK K−1 . A change in ambient temperature of ±10 ◦ C
respect to very slow changes; gives an apparent change in sensor temperature equal to
• dynamic characteristics, specifying the sensor response ±50 mK.
to variations in time and in the measurand (the quantity
that has to be measured);
• environmental characteristics, relating the sensor per- 4.2 Linearity and hysteresis
formance after or during exposure to specified external
conditions (pressure, temperature, vibration, radiation); If the output y is a linear function of the input x, the
• reliability characteristics, describing the sensor life sensitivity S does not depend on x. In the case of a
expectancy. nonlinear transfer function y = f (x), S does depend on
722 Elements: A – Sensors
the input or output value. Often, a linear response is as 10−4 ; assuming this is relative to the full-scale value,
preferred to reduce computational burden in, for instance, it means that the output changes discontinuously in steps
multisensor control systems. In that case, the sensitivity equivalent to input displacements of 10 mm.
can be expressed with a single parameter. The transfer of a
sensor with a slight nonlinearity may be approximated by Example 4 A particular type of optical encoder that has a
a straight line, to specify its sensitivity by just one number. resolution of 14 bit. The smallest change in angle that can
The user should be informed about the deviation from the be detected by this encoder is 2π/214 ≈ 1.9 × 10−4 rad or
actual transfer; this is specified by the nonlinearity. 0.022◦ .
The linearity error of a system is the maximum devia-
tion of the actual transfer characteristic from a prescribed
straight line. Manufacturers specify linearity in various 4.4 Accuracy
ways, for instance, as the deviation in input or output
units: xmax or ymax , or as a fraction of FS (full scale):
Formally, the accuracy reflects the closeness of the agree-
xmax /xmax . Nonlinearity should always be given together
ment between the actual measurement result and a true
with a specification of the straight line. The following def-
value of the measurand. The accuracy specification should
initions are in use:
include relevant conditions and other quantities. Many sen-
• Terminal nonlinearity: based on the terminal line–a sor manufacturers specify the sensor performance in terms
straight line between 0 and 100% theoretical full- of accuracy. This specification should be viewed with sus-
scale points. picion, because it may or may not include particular imper-
• End-point nonlinearity: based on the end-point line–the fections of the sensor (nonlinearity, hysteresis, drift), and
straight line between the calibrated end points of the may be only valid under strict conditions.
range; coincides with the terminal (theoretical) line after
calibration of zero and scale.
• Independent nonlinearity: referring to the best-fit strai- 4.5 Offset and zero drift
ght line, according to a specified error criterion, for
instance, the line midway between two parallel lines
Most sensors are designed such that the output is zero at
enclosing all calibration points; if the least-square error
zero input. If the transfer characteristic does not intersect
criterion for the best-fit straight line is used, this linear-
the origin (x,y = 0,0) the system is said to have offset.
ity error is as follows.
The offset is expressed in terms of the input or the output
• Least-square nonlinearity: based on the least square
quantity. Specifying the input offset is preferred to facilitate
line, the line for which the summed squares of the
a comparison with the measurand.
residuals is minimized.
Hysteresis is the maximum difference in output signal Example 5 The sensitivity of a particular type of force
when the measurand first increases over a specified range, sensing system is 0.1 V N−1 . At zero force, the out-
and next returns to the starting value. The traveled range put appears to be 3 mV. The (input) offset of this sys-
should be specified because hysteresis strongly depends tem is the output offset divided by the sensitivity, so
on it. 0.03 N.
A nonzero offset arises mainly from component toler-
4.3 Resolution ances. Offset compensation can be performed in the inter-
face electronics or the signal processing unit. Once adjusted
The resolution indicates the smallest detectable increment to zero, the offset may nevertheless change, owing to tem-
of the input quantity. When the measurand varies contin- perature variations, changes in the supply voltage or aging
uously, the sensor output might show discontinuous steps. effects. This relatively slow change in the offset is called
The value of the corresponding smallest detectable change zero drift. In particular, the temperature induced drift (the
in the input variable is the resolution: xmin . Sometimes temperature coefficient or t.c. of the offset) is an important
this parameter is related to the maximum value xmax that item in the specification list.
can be processed (full-scale value), resulting in the resolu- Sometimes a system is deliberately designed with offset.
tion expressed as xmin /xmax . Many industrial transducers have a current output ranging
from 4 to 20 mA. This facilitates the detection of cable
Example 3 The resolution of a particular type of wire- fractures or a short circuit, producing a zero output clearly
wound linear potentiometer with range 10 cm is specified distinguishable from a zero input.
Transducer Fundamentals 723
4.6 Noise Various possibilities of error reduced designs are described
in Article 16, Reduction of Influence Factors, Volume 1.
Electrical noise is also specified in terms of the input quan-
tity, to show its effect relative to that of the measurand.
White noise (noise with constant power over a wide fre- 5 ROLE AND USE OF ACTUATORS
quency range) is usually expressed in terms of spectral
√
noise power (W/Hz), spectral noise voltage (V/ Hz) or Similar to the definition of a sensor (or input transducer)
√
spectral noise current (A/ Hz). Thermal noise is an exam- an actuator (or output transducer) can be defined as a
ple of ‘white noise’. device that converts information from the electrical domain
Another important type of noise is 1/f noise (one-over- to another physical domain, with preservation of infor-
f noise), a collection of noise phenomena with a spectral mation. Actuators and sensors can be considered as each
noise power that is proportional to f −n , with n = 1 to 2. other’s opposites. Actually, many transduction principles
Quantization noise is the result of quantizing an analog are reversible (notably the direct conversion types), which
signal. The rounding off results in a (continuous) deviation means that the same device can act as both – a sensor as
from the original signal. This error can be considered well as an actuator (for instance, a piezoelectric sensor and
as a ‘signal’ with zero mean, and a standard deviation actuator). However, an actuator should, in general, be able
determined by the resolution of the AD converter. to deliver output energy, whereas a sensor operates best
when the amount of energy taken from the measurement
object is close to zero. Reversibility does not imply that
4.7 Response time both options can be combined in a single device.
Since an actuator has to produce output energy, its con-
The response time is associated with the speed of change struction and dimensions differ largely from that of sensors.
in the output upon a stepwise change of the measurand. There are far less physical conversion principles suitable
The specification of the response time always needs to for actuation than for sensing. However, the variability
be accompanied by an indication of the input step (for in types is larger, according to the application and the
instance, FS – full scale) and the output range for which required energy.
the response time is defined, for instance, 10 to 90%. Creep Some actuator types, grouped according to the physical
and oscillations may make the specification of the response domain of the output are as follows:
time meaningless or at least misleading.
• Mechanical output: induction motor; permanent mag-
net electromotor; stepper motor, piezoelectric actuators
(linear and rotating movements), relay;
4.8 Frequency response
• Optical output: light emitting diode (LED), solid state
laser diode (SSLD); incandescent lamp;
The sensitivity of a system depends on the frequency or
• Thermal output: electric heater; (thermo)electric cooler;
rate of change of the measurand. A measure for the useful
• Acoustic output: piezoelectric ultrasound transmitter;
frequency range is the frequency band. The upper and
electrostatic ultrasound transmitter.
lower limit of the frequency band are defined as those
frequencies for which the output signal has dropped to half As with sensors, actuators too can operate either in a
the nominal value, at constant input power. For voltage or direct way or in a modulating mode. Most actuators are of
√
current quantities, the criterion is 1/2 2 of the nominal the direct type: no auxiliary energy is required. Examples
value. The lower limit of the frequency band may be zero; of modulating actuators are an oscilloscope (the electrical
the upper limit has always a finite value. The extent of signal modulates the deflection of an electron beam and
the frequency band is called the bandwidth of the system, hence the position of the light spot on the screen) and an
expressed in Hz. LCD (the electrical signal modulates the transmission of an
All specification items only apply within the operating auxiliary light source by influencing the transmittance of
range of the system, which should also be specified cor- the crystal material).
rectly. It is given by the measurement range, the required The performance of a sensor system can sometimes be
supply voltage, the environmental conditions and possibly improved by including the sensor in a feedback loop. The
other parameters. electric output of the sensor is amplified and supplied to an
Despite the specified limitations of sensors, a sensing actuator, having the inverse conversion characteristic. So
system can be configured in such a way that the effect of its output is of the same nature as that of the measurand,
some of these limitations are eliminated or at least reduced. or at least is closely related.
724 Elements: A – Sensors
Measurand [M] →0 In the past decades, sensors tended to get smaller and
Sensor Output actuators larger. Development in microtechnology, how-
+ _ [E] Gain ever, allows the creation of microactuators as well see, for
=0 instance, Tabib-Azar (1997). Consequently, such microac-
[M] Actuator [E]
tuators can produce little energy, and are therefore only
useful as part of a microsystem, including (micro)sensors,
Figure 5. Feedback sensor configuration; [M] denotes the do- signal processing, and actuation.
main of the measurand, [E] the electrical domain.
The role of an actuator in a measurement system can be
In the steady state, and assuming stability and a high • indication of the measurement result (optical display),
loop gain, the input of the sensor is compensated (made • registration of the measurement result (magnetic or
zero) by the feedback action, see Figure 5. The compensat- optical head; driving a plotter pen),
ing signal is delivered by the actuator, and hence its input • control of the measurement system (scanning in 1, 2, or
equals the output of the measurement system. Since the 3 dimensions),
input signal of the sensor is zero, its static characteristics • control of a process to perform measurements (for
(sensitivity, nonlinearity, limited range) are irrelevant to the instance, in wear detection, fatigue tests)
measurement result. The only requirement is zero offset. • control of the sensing part of the measurement system
The performance of the measurement completely relies on (feedback sensors).
the characteristics of the actuator. Further details of feed-
back sensing systems are discussed in more detail in the
article Article 113, Force-feedback Sensors, Volume 2. REFERENCES
Like with sensors, also some actuators operate on the
basis of a concatenation of conversions. Examples are Middelhoek, S. and Noorlag, D.J. (1981) Three-Dimensional
as follows: Representation of Input and Output Transducers. Sensors and
Actuators, 2, 29–41.
• Thermal microactuator: a particular element of the Tabib-Azar, M. (1997) Microactuators – Electrical, Magnetic,
microstructure that is heated by an electric current upon Thermal, Optical, Mechanical, Chemical and Smart Structures,
which it expands, producing a displacement of the tip. Kluwer Academic Publishers, Boston, MA.
• Reed switch: a coil around the switch that is activated IOS (1995) Guide to the Expression of Uncertainty in Measure-
by a current, producing a magnetic field, which closes ment, International Organization for Standardization, Geneva.
the contacts of the switch. ISA (1975) S37.1: Electrical Transducer Nomenclature and
• Incandescent lamp: electric current heats up a filament Terminology.
that emits visible light above a particular temperature.
109: Structure and Energy in Sensor Systems
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
measurement system described in Article 68, Setting the
1 Boundary Perspective of Sensor Systems 725 System Boundaries, Volume 2. The existence of intrinsic
2 Sensors, Transducers, and Modifiers – and extrinsic boundaries allows the discrimination between
Terminology and Definitions 725 internally and externally arisen effects respectively. In this
3 Representing an Instrument Stage by way, internally caused errors, which give rise to ‘eigen’
Signal/Energy Notation 728 errors, or characteristic errors (Solopchenko, 1994), are
noted to be qualitatively different from externally arising
4 Classification of Transducers by Information
effects. From this boundary view, it can also be seen
and Energy 728
that inputs and outputs, which can be subdivided by
5 The Classification of Energy 729
energy form, can also be classified as within the resource,
6 Spatial Modeling of Transducers and Sensors 731 information, or contamination groups of boundary impacts.
7 The Transducer Space 732 Interfering influences – which arise from inside the sen-
8 The Sensor Effect Tetrahedron for sor, are due to either parameter sensitivity or internal funda-
Modulating Sensors 733 mental fluctuations. Parameter sensitivity is caused by the
9 The Sensor Effect Tetrahedron for Self- inexorable aging of all of the components that make up the
supporting Cross-converting Sensors 734 sensor structure. Fundamental fluctuations occur as a result
Related Articles 735 of equipartition of energy between the thermodynamic state
References 735 and the states of the other energy forms. External interfer-
ence is caused by disturbance power flows or by other forms
of direct contamination. Highlighting the various sources of
contaminating inputs provides a clear view of the limits on
1 BOUNDARY PERSPECTIVE OF SENSOR the ultimate resolution of a sensor in executing its primary
SYSTEMS function. In the case of sensor systems, the resource output
is also the information output. It is the influence effects that
make good design difficult to achieve.
When analyzing sensor systems, the Systems Engineering
approach described in Article 5, Humans in the Real
World, Volume 1; and Article 14, Hierarchical Aspects
of Measurement Systems, Volume 1, can fruitfully be 2 SENSORS, TRANSDUCERS, AND
applied by combining the ideas of the boundary view of MODIFIERS – TERMINOLOGY AND
a measurement system and the information and energy
tetrahedron as described in Article 6, Substructure of DEFINITIONS
Human–Machine Systems, Volume 1 to obtain Figure 1.
This approach is consistent with that used in the application Through the study of the various blocks used in routine
of the Systems Engineering boundary view to the practice, it has become clear that two prime kinds of energy
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
726 Elements: A – Sensors
Contamination/interference Output [E o /I o]
[E c /I c ]
Energy/information energy/information
Nuclear
Acoustic
N
Magnetic
A
Thermal
M
T Electrical
E Mechanical
M Optical
Measurand/input [E /I ]
O Chemical energy/information
m m
C C O M E T M A N
C
O COMETMAN
M
E
[E c /I c ]
T C C
M O O
A M M
E E
N [E m /I m] [E o/I o]
T T
M M
A A
N N
Resource/support [ E s /I s ]
Energy/information [E s /I s]
COM ET M AN
Figure 1. Boundary perspective of a sensor system and its influencing effects.
COMETMAN
I – information form; E – energy form
E c /Ic(contamination/influence) Suffixes:
m – measurand/input
c – contamination/influence
C o – output C
E m /Im Self-sustaining E o /Io
O s – support/resource O
M Cross-converter M
E Input transducers Output E
Modifier
T transducer T
M E m /Im E o /Io M
A Modulator A
N N
E s, Support energy COMETMAN
E c/Ic(Contamination/influence)
COMETMAN
Formal analog DM components Formal analog DM components
Formal A-to-D DM components
Formal digital DM components
Figure 2. Functional block diagram of a basic instrumentation system illustrating modifier elements and input and output transducers.
Structure and Energy in Sensor Systems 727
handling operational structures are evident in instrumenta- contributions to sensor modeling have been taken into
tion systems. Figure 2 illustrates the most basic structure account. Stein (1969), Middelhoek and Noorlag (1981)
of an instrumentation system made up of input and out- and Middelhoek and Hoogerwerf (1986) use the term self-
put transducers and modifiers (Middelhoek and Noorlag, generators. In other places, this group is also referred to as
1981; Middelhoek and Hoogerwerf, 1986). The relation- passive (Fraden, 1997) or self-sustaining (Areny and Web-
ship between this structure and Formal Data Measurement ster, 1991).
Components, or FDMCs, outlined in Article 14, Hierar- All of these titles share common shortcomings associated
chical Aspects of Measurement Systems, Volume 1, is with ambiguity and imprecision. To propose a replacement
also given in this figure. title, which conveys the important aspects previously stated,
A transducer is a device that transforms the input signal requires a clear view of the principles of sensing and the
of one energy form into an output of another energy form, effects used in the sensing process.
doing so in such a way that a prescribed relationship is Operating principles, including sensor effects, which
maintained between the input and the output signal. have been described by Ylilammi (1989), Van Duyn and
A sensor is an input transducer involving the process Middelhoek (1990) and Kwaaital (1993), represent a useful
of measuring those variables that characterize the flow of way of classifying sensors. From these principles, it is
information and energy. clear that self-supporting cross-converting sensors convert
An example is the mercury-in-glass thermometer in energy from one form at the input of the sensor to
which heat energy is converted into mechanical energy, another different form at its output without the use of
forcing the mercury to move along the glass as it expands a supporting energy source. For this reason, they may
or contracts. be called cross-converters. However, since some types of
Another example is the bellows pressure sensor, where an modulating sensors also act as cross-converters, the use of
increase in pressure causes the bellows to change its length. the term cross-converter to describe a self-generator will
The most commonly used sensors and transducers, which lead to ambiguity. For this reason, McGhee et al. (1999)
are especially suitable for automation and control pur- proposed that self-generators should be renamed as self-
poses, use information carrying signals in the electrical sustaining cross-converting sensors, or as self-sustaining
energy form. cross-converters for short. It is likely that this description is
It is also important to distinguish between those sensors probably the most concise but still an adequately complete
that do not require the supply of energy from any source description is possible.
other than the system under measurement, and those that Examples of self-sustaining cross-converters, mislead-
do. The former are now commonly referred to as self- ingly referred to as passive in various places, are
generators while the latter are referred to as modulators.
They are also misleadingly referred to as passive and active • the thermocouple, which generates a voltage propor-
respectively. tional to input temperature;
Self-supporting cross-converting transducers are energy • the photovoltaic cell, which gives an electrical output
converters, which do not require the supply of energy for when exposed to light;
support or excitation purposes. • the mercury-in-glass thermometer, whose mercury col-
Referring to this species of sensor as a self-generator umn expands when exposed to increasing temperature.
does not really express the completeness of the functional
energy transformation, which they perform. Hence, in some The other kind of sensor structure, which needs an
respects, calling this group of sensors self-generators could excitation or auxiliary supply of energy for its operation,
be viewed as incorrect. Strictly speaking, they should be is called a modulating type. During their operation, the
regarded as converters of an input energy form to a dif- energy flow is modulated by the input measurand. They
ferent output energy form. For this reason, it would be of are also misleadingly classified as active. Examples of the
benefit if they were referred to by a name with a more modulating group of sensors are
incisive description than self-generators. Such a new name,
which should preferably be short, should also clearly and • all resistance-based sensors, where electrical energy
unambiguously convey the idea that no auxiliary energy must be applied to allow the resistance to be measured;
form is required to support the flow of information in • the photodiode, where a bias electrical current is pro-
the energy transduction process involved. McGhee et al. vided by an electrical support energy source;
(1999) have proposed such a renaming, which attempts • the ultrasonic Doppler flow sensor where an acoustical
to express a complex phenomenon by a concise linguistic interrogating energy form is modulated by the flow
synonym. In making this renaming proposal, previous velocity.
728 Elements: A – Sensors
It is recommended that the misleading reference to where the term [OSs /Oes ] indicates that there is no sup-
modulating sensors as active and the naming of self- port signal/energy form for self-sustaining cross-converter
supporting cross-converting sensors as passive should be sensors.
avoided in sensor nomenclature. This position is adopted
to remove any ambiguity that the terms introduce.
In contrast to other kinds of instrument subsystems, 4 CLASSIFICATION OF TRANSDUCERS
modifiers alter, or modify, the input energy form to BY INFORMATION AND ENERGY
give an output in the same energy form. They do not
convert the energy to another form. An electronic filter, Classification in sensing and transduction follows the
for instance, processes an electrical, or more commonly
same general route outlined in Article 2, Organization of
electronic, energy signal by modifying the distribution of
Instrument Science, Volume 1. This scientific approach to
the frequency content of the signal. It still provides an
grouping uses the four main orders or problems of classifi-
electrical energy output. Other examples of modifiers are
cation to assist in the ordering.
electrical transformers, mechanical levers, and gearboxes,
In the First Order, known as Generality or Resemblance,
hydraulic rams, and heat exchangers.
the question of those relations describing likeness between
Since this article is concerned with sensors, the reader is
sensors is considered.
referred elsewhere for descriptions of modifiers and output
Sensors and transducers must also be grouped on the
transducers (Middelhoek and Noorlag, 1981; Middelhoek
basis of the Second Problem or Order of taxonomy, which
and Hoogerwerf, 1986).
is concerned with their Composition or Collectivity.
While this order is concerned with the relationship of
3 REPRESENTING AN INSTRUMENT the parts or elements of a sensor system to the whole
STAGE BY SIGNAL/ENERGY system and vice versa, the Third Order of classification,
NOTATION called Hierarchy, when applied to sensors and transducers,
takes account of the relation of rank between the heads or
Figure 3(a, b) illustrates representative block diagrams for central members of the constituent groups of the elements
self-supporting cross-converting and modulating types of comprising sensor systems.
sensors with their input and output notation. These should then be related in the order of composition.
It is possible to write down the functional relationship In this process, the place occupied by each concrete
between the signal/energy forms of the output, [So /eo ], sensor is assessed in every order relative to other sensors
in Figure 3 in terms of the respective forms for the making up both the same order and different orders.
measurand, [Sm /em ], the support/resource, [Ss /es ], and the Thus, the uniqueness of each sensor in the ordering is
contamination/interference, [Sc /ec ], inputs. Hence, in the distinguished by its similarity with and difference from
case of a modulating sensor, the functional relationship every other sensor.
between the output Signal/Energy form and the three inputs Perhaps the most important Fourth Order in the classifi-
as shown in Figure 3 can be written as cation of sensors hinges upon the notions of kinship through
the relations describing ascent, descent, and collaterality.
So Sm S S
=f , s , c (1) As with the orders of Composition and Hierarchy, the
eo em es ec
last problem of classification just described, which is also
In the case of a self-sustaining cross-converter, there is no known as Genealogy and Evolution in the biosciences, is
support energy. Under these conditions, (1) can be rewritten also concerned with the actual sensors to be classified.
in the form These latter orders, when applied for sensor systems,
emphasize the importance of a functional and structural
So Sm OSs S
=f , , c (2) hierarchy.
eo em O es ec Taxonomy in Sensor Science may thus be regarded as
possessing the six main features of its objectives and func-
S i, e i S o, e o S i, e i S o, e o tions, its materials and activities, together with the methods
of discrimination used and hierarchical ordering (Article 2,
S s, e s Organization of Instrument Science, Volume 1).
(a) (b) Taxonomy is useful for proving the newness claimed for
Figure 3. Block diagram and notation for the two types of new sensors. It places a sensing stage into a set position on
sensors: (a) self-supporting cross-converting sensors and (b) mod- the basis of its properties. If sensors were so classified, it
ulating sensors. would show up those that are actually of the same type –
Structure and Energy in Sensor Systems 729
yet appear as a new kind. Too many sensors are constantly may be considered on either a macroscopic or microscopic
being reinvented for lack of knowledge that one already has level. There are a number of different, equally valid ways
been reported of the same kind. of classifying energy.
To understand the overall classification of energy, it is
important to realize that there is a general tendency in
5 THE CLASSIFICATION OF ENERGY physics to represent the largest number of phenomena by
the smallest number of models (McGhee et al., 1999, 2001).
Consistent with the previously described orders of classi-
fication, it can be seen that the appropriate grouping of Physical considerations lead to the classification of
sensors must be on the basis of function, structure, informa- gravitation, atomic, magnetic, and electric as possibly being
tion/energy form, and signal output. A good way of group- the most representative, irreducible groups.
ing transducers uses the transduction principle involved, The energy acronym GAME, shown in Table 2, gives
and the form of the measurand, as shown in Table 1. a convenient shorthand for these four basic energy forms.
Kurt Lion (1969) produced such a classification similar to Because they lie at the top of the hierarchical grouping of
that in Table 1. In this classification, it is possible to relate energy domains, they are not very informative from the
the physical effect in the energy domain of the measurand, point of view of either physics or measurement systems.
which allows a transduction into another energy form. The GAME energy domains can also be written as the
Thus, a resistance transducer for measuring displacement expanded subdomains also given in Table 2. Each of these
is classified as a resistance displacement transducer. Other subdomains is associated with at least one characteriz-
classification examples are a pressure bellows or a force ing physical phenomenon. The subdomains of the gravi-
diaphragm. Notice that all of the diagonal entries in this tation and atomic groups of energy forms can be seen to
table correspond to modifiers as defined above. be similar.
Questions about energy and information, which are Differentiation between both groups is made on the basis
highly relevant in measurement systems, are of fundamental of distinguishing potential energy characteristics and kinetic
importance and of immediate interest in this article. Energy energy manifestations.
Table 1. Examples of transducers and their logical grouping for different input and output signals.
IN
Radiation Mechanical Thermal Electrical Magnetic Chemical
OUT
Radiation Geometric-optical Interferometer; Thermal radiation Kerr cell; Faraday cell Transducer based
system filter Photoelastic transducer Electro- on emission &
transducer luminescent absorption
transducer, spectroscopy
LED
Mechanical Crooke’s Gravity balance; Thermometers; Electrometer; Magnetometer; Hydrometer; Elec-
radiometer; Ballistic Bimetallic strip; Electrostrictive Magnetostriction trodeposition
Radiation pendulum; Dilatation transducer; transducer cell
pressure system Bellows; Gear thermometer Piezoelectric
box transducer
Thermal Bolometer Adiabatic Mixing Thermal Eddy-current Combustion
system; calorimeter; converter transducer calorimeter
Friction Heat exchanger
calorimeter
Electrical Photoelectric, Resistive; Thermistor; Transformers; Magnetoresistance; Transducers
Photoresistive & Inductive; Thermoelement; Langmuir Hall-effect based on
Photogalvanic Capacitive; Seebeck effect probe; Charge transducer potentiometry,
cells; Piezo resistor collectors; conductimetry,
Photodiode Transistor polarography;
pH-meter
Magnetic Curie-effect Magnetoelastic Paramagnetic Electromagnetic Magnetic recorder Magnetic
radiation meter transducer transducer transducer; resonance
device Coil system
Chemical Photographic Ultrasonic Thermal dye Electrolytic Paramagnetic Chemical system
emulsion chemical indicators integrator; oxygen analyzer modifiers
transducer Electrical
plating
730 Elements: A – Sensors
Table 2. A conventional classification scheme for energy forms.
Energy forms
Physics Expanded physical Sensor measurement Characteristic physical phenomenon
(GAME acronym) forms forms
Gravitation Gravitational Attraction between masses
Mechanical Mechanical Mass in motion has equivalent energy forces,
displacements
Atomic Nuclear Chemical Binding energy within nuclei
Atomic Binding energy of nuclei and electrons
Molecular Bond energy of atoms in molecules
Thermal Thermal Kinetic energy of molecules
Magnetic Magnetic Magnetic Magnetic fields
Electric Electric Electrical Electric fields, potentials and currents
Electromagnetic Radiation Radio waves, through UV to X-, γ -and cosmic rays
radiation
The gravitational subgroup is connected with natural The separation of thermal from the other atomic forms of
forces of attraction between masses. energy is due to its kinetic energy aspects.
In a similar way, the bonding energy within nuclei, The magnetic form is self explanatory due to its particular
within atoms and within molecules is also connected with manifestation.
natural forces of attraction but on a microscopic scale. Electric circuit variables are distinguished from the entire
Atomic forces can also be repulsive. This contrasts with radiation group.
the macroscopic scale of gravitational attraction. Although the grouping of energy by the forms in Table 2
Mechanical forces, which are associated with physical is perfectly acceptable, the individual groups are not easily
motion, cause kinetic energy effects. remembered. For this reason, the method used in this
The classification of energy using the GAME acronym article recalls the organization of energy classification
and its expanded groups may be regarded as the most given by Peter Stein (1969) in his presentation of the
precise, since the grouping results from the application of transducer space.
the formal principles of taxonomy. It provides an entirely His scheme of energy classification grew from his wide
consistent physical view. experience as a teacher of measurement and instrumenta-
Van Dijck (1964) describes another way of grouping tion. He perceived that the sensor measurement grouping
energy forms by extracting six from the extended group of energy forms given in Table 2, although very close to
of nine of the GAME acronym. This abbreviated group the central heads of the GAME grouping are not easily
of six, which can be identified as being of particular remembered. Thus, arose his scheme for classifying energy
relevance to measurement systems (McGhee et al., 1996, forms using the memory-aiding (energy acronym COMET-
1999, 2001), are called sensor and measurement energy MAN, which is summarized in Table 3 (McGhee et al.,
forms, in Table 2. These six have been used previously 1999, 2001).
in classifying instrument systems. The second column of Table 3 provides the cross-link
This group of six energy forms, used by Kurt Lion (1969) between it and Table 2. Whereas the previous methods of
and Middelhoek and coworkers (1981, 1986), are classification are based upon rigidly linking the energy by
similarity, COMETMAN is specifically formulated as a
1. radiant energy,
memory aid.
2. mechanical energy,
The acronym arises from classifying by Chemical (i.e.
3. thermal energy,
molecular), Optical (i.e. radiation), Mechanical (i.e. includ-
4. electrical energy,
5. magnetic energy, and ing gravitation and mass), Electrical (i.e. both electric
6. chemical energy. circuit forms and electromagnetic radiation), Thermal, Mag-
netic (i.e. including electrical and atomic forms), Acoustic
Gravitational energy, which appears to be unimportant (i.e. correctly speaking within the mechanical group) and
here, is grouped within the mechanical energy form since Nuclear (i.e. strictly speaking within the atomic group)
mechanical variables are more important in measurement. energy forms.
All of the microscopic potential energy forms are Once again, the important point to emphasize here is
grouped together within the domain of chemical energy. not whether one form of classification is more correct than
Structure and Energy in Sensor Systems 731
Table 3. A classification scheme for energy forms based upon the COMETMAN acronym.
Energy forms
Modified acronym Expanded physical forms Physical variables Human senses
COMETMAN (GAME acronym)
Chemical Molecular Concentration, reaction rate, redox potential, biological Smell, Taste
(Atomic) properties, clinical analysis
Optical Electromagnetic radiation Intensity, phase, wavelength, amplitude, transmittance, Sight
(Electric) polarization
Mechanical Mechanical Position, velocity, acceleration, mass, density, force, Touch
(Gravitation) stress, pressure, moment, torque, shape, roughness,
orientation, viscosity
Electrical Electric Current, voltage, resistance, inductance, capacitance, Pain sensors or
(Electric) permittivity nociceptors
Thermal Thermal Heat flux, temperature, thermal conductivity, thermal Touch and sight
(Atomic) capacitance, specific heat, expansion
Magnetic Magnetic Field strength, magnetic moment, permeability, flux None
(Magnetic) density
Acoustic Mechanical Sound pressure Hearing
(Gravitation)
Nuclear Nuclear, atomic Nociceptors
(Atomic)
another. It is the convenience of the most easily remem- fulfilling the primary aim of measurement as stated
bered grouping that tends to be used, for convenience by Solopchenko (1994). This primary aim is ‘. . . tak-
is an important ingredient in any scheme of classifica- ing account of errors, their sources and their quantifi-
tion (Thomson, 1911). cation . . .’. This article describes a spatial tetrahedron
In this classification of energy form, the optical form method for modeling and for predicting the possible
is extracted from the radiation group, the acoustic from energy sources of interfering effects. Errors are covered
the mechanical group, and the nuclear from the atomic/ in depth in Article 48, Common Sources of Errors in
chemical group, to allow the formation of the acronym. Measurement Systems, Volume 1; Article 49, General
The inputs of a sensor system then become the energy Characterization of Systematic and Stochastic Errors,
and information form of the input, the resource energy form, Volume 1; Article 50, Errors in Signal Systems, Vol-
supporting the flow of information, and the energy and ume 1; Article 51, Errors in Digital Signal Systems, Vol-
information form of the contamination. Thus, the resource ume 1; Article 52, Error Models, Error Budgets and
and contamination classes have a possible total of eight their Calculation, Volume 1; Article 53, Calculation and
energy form inputs each. Treatment of Errors, Volume 1; Article 54, Explanation
of Key Error and Uncertainty Concepts and Terms,
Volume 1; and Article 55, Uncertainty Determination,
6 SPATIAL MODELING OF Volume 1.
TRANSDUCERS AND SENSORS A complete spatial representation for the boundary
inputs and outputs of a generalized sensor system
Sensor systems have been clearly placed within human– requires a multidimensional space. As the boundary
machine supersystems (Article 5, Humans in the Real view of human–machine supersystems reveals (Article 6,
World, Volume 1). This allows sensor systems to be intro- Substructure of Human–Machine Systems, Volume 1;
duced as information machines for measurement (Arti- McGhee et al., 1996; Sandquist, 1985), there is a
cle 107, Principles of Sensor Science, Volume 2) where total of three groups of inputs and three groups of
the importance of sensors in the overall view of mea- outputs. Hence, the representation becomes too complex
surement was also emphasized. The importance of the for graphical illustration on a two-dimensional paper
boundary view of systems, and consequently of sensor surface. Previously, Kurt Lion (1969), Peter Stein (1969)
systems, including the concepts of extrinsic and intrinsic and Middelhoek and coworkers 1986, 1981) and McGhee
sources of interference is highlighted in Article 68, Set- et al. (1999, 2001) realized this requirement.
ting the System Boundaries, Volume 2. These sources Lion formulated the most basic spatial model by
of interference and their discrimination are essential in concentrating on the input/output relations already outlined
732 Elements: A – Sensors
in Table 1. A highly complex space is immediately reduced 7 THE TRANSDUCER SPACE
to a two-dimensional problem.
Although this model does not include either the support It is possible to use the COMETMAN classification of
energy inputs or the interference inputs, it is still a energy to visualize the wide range of possible input and
convenient starting point for focusing upon transduction as output transducers. To develop this approach, it is necessary
an energy transforming operation. Each one of the input to realize that any modulating sensor can be represented by
and output groups may be specified as information-bearing the energy/signal notation in the [x, y, z] space, where the
signals or energy flowing forms. measurand energy form, x, modulates the support energy
Stein (1969) extended the two-dimensional representa- form, z, to produce a variation in the output energy form,
tion of Lion to three dimensions using what he called the y, bearing some relationship to the measurand energy form,
transducer space. This uses a three-space to illustrate sens- x. This model is represented mathematically by
ing and transduction. Adding a third axis to represent the
range of possible support energy forms resulted in a three- y = f (x, y) (3)
dimensional Cartesian space. Each of the axes in this space
uses the COMETMAN energy form classification. Some The representation described by (3) can be drawn in a
time later, Middelhoek and coworkers used the sensor mea- pseudoisometric view on a two-dimensional page as shown
surement forms of energy classification of Table 2 in the in Figure 4.
three-space sensor effect cube. A detailed description of Each of the axes in this space is labeled along its
spatial transducer modeling is given later in this article. length with the COMETMAN acronym to give a three-
The 3-D transducer space has one significant disadvan- dimensional, or three-space, energy, and/or signal form of
tage. In this diagram, no account is taken of contaminating the inputs and outputs. In this way, the space can now
inputs. Consequently, the influence of interference from provide an aid in understanding the interacting relationship
whatever source is not transparent. A spatial representation between the support energy form and the input and output
for the boundary inputs and outputs of a sensor system, in information carrying signals in their range of possible
its most general form, requires a multidimensional space. energy forms. This is now the three-dimensional transducer
Any spatial representation, which does not take all inputs space, as it was called by Peter Stein, representing the input
and outputs into consideration, has serious shortcomings. signal in its energy form along the conventional x-axis, the
As there is a total of three groups of inputs and three support energy form along the conventional z-axis and the
groups of outputs, the representation becomes very complex output energy along the conventional y-axis as shown in
for graphical illustration on paper surfaces. In spite of Figure 3. Two examples are used to illustrate the use of the
this, McGhee et al. (1999) extended the transducer space transducer space for modeling energy/information flow in
using the concept of the sensor effect tetrahedron, which is sensors and transducers.
also described in some detail later in this article. This model Two examples illustrate the application of Figure 4. The
is an adaptation of the information and energy tetrahedron first example uses self-supporting cross-converters, called
used in system modeling (Article 6, Substructure of self-generating transducers by many other sensor scientists,
Human–Machine Systems, Volume 1; McGhee et al., and modifiers, which operate without an auxiliary energy
1996, 2001). source. They are represented in the vertical x-y plane.
Other contributors in the field of measurement have also Consider, for example, the thermocouple. Thermal infor-
realized the importance of spatial modeling for simplifying mation flows from the thermal energy input on the x-axis
the complex problems of classification in sensors such to the information output on the y-axis transduced to the
as in the case of optical sensors. Initial attempts which electrical energy form. The appropriate block diagram also
give good basic insight into optical sensor classification, appears in Figure 4.
utilized the three-dimensional structure of the transducer Modulating types of sensors, as the second example,
space and the sensor effect cube but named the axes can be shown in the [x, y, z] Cartesian transducer space
of the representation differently. Subsequently, a more in Figure 4 depending upon which support energy source
rigorous analysis gave an expanded classification called the is used.
Tri-Angular Pyramid Sensor Representation, or TAPSER, For example, a photoconductance sensor would have
scheme (El-Hami et al., 1993). its input information at x = Optical, its output at y =
The sensor effect tetrahedron, which is described below, Electrical and its support energy at z = Electrical. The
is an appropriate combination of the transducer space/sensor radiating optical input energy incident upon the sensor
effect cube, but using the COMETMAN energy classifica- modulates the flow of electrical energy from the support
tion, and the multifaced objects from the TAPSER space. energy port to the output energy port. For this reason,
Structure and Energy in Sensor Systems 733
y Output energy/signal form
Si S o′ e o S i′ e i S o′ e o
Nuclear
Acoustic
S s′ e s
Magnetic
Thermal
Photoconductive cell Thermocouple
Electrical
Mechanical
Optical
to give electrical Chemical
output C O M E T M A N
C Input energy/signal form x
O
Optical measurand
M
E
M
od
T El
ul
ec
at
M tri
e
y/
ca
s
rm erg
A ls
up
l fo en
N po
na or t
rt
sig pp
Su
z
Figure 4. The transducer space – a three-dimensional (three)-space for transducers by energy/information.
modulating sensors of this type bring the advantage of energy symbol, eS , in the diagram models the effect of
much higher power gain to measurement systems than do modulation of eS by Sm .
self-supporting cross-converters.
The energy/information space diagram, which may also
be interpreted as a signal/energy space, allows the predic- 8 THE SENSOR EFFECT TETRAHEDRON
tion or visualization of transducer possibilities. It is clear FOR MODULATING SENSORS
that there are numerous different possibilities as feasible
candidates, when proposing a transduction principle to meet Starting from the boundary limit viewpoint of a sensor sys-
a sensing need. While all known possible combinations of tem given in Figures 1, 2 and 3, it is possible to represent
input, output, and support energy forms can be located in a sensor and a transducer effect space, which is more com-
this 3-D representation, there are still many positions where plex than that proposed by either Lion (1969), Stein (1969)
practical devices have not been proposed so far. or Middelhoek and his coworkers (1981, 1986), but is still
In accordance with Figure 3, it can be seen from the useful. In a sensor system, the main output is an information
block diagrams at the top of Figure 4 that any instrument output, which also corresponds to the resource output.
system building block, has three possibilities on inputs and Contamination is also produced at the output. This
outputs. The self-supporting cross-converting sensor has an contamination is the information output when there is cross-
input signal, Sm , with its associated input self-supporting coupling into some other information-bearing channel.
or excitation energy, em . At the same time its output signal It is clear that the output of a sensor system may be
is So , and output energy eo . The difference between the regarded as either the resource output, the information
self-supporting and modulating types of cross-converters is output, or even a contaminating source.
a matter of how em is supplied to the system as shown in Although it is essential to design any system so that the
Figure 4. contamination it produces is negligible, this question will
In self-supporting cross-converting sensors, the sys- not be considered further for the present purposes. Hence,
tem under measurement supports the information flow by the output energy/information form for a sensor system is
supplying energy to the sensor. Modulating sensors, which given more emphasis in this figure than the contamination
extract negligible energy from the system under measure- produced by the sensor system.
ment, modulate the support energy input. The energy sym- The proposed representation consists of one axis nor-
bol, em , is removed to indicate that negligible energy is mal to a base plane as shown in Figure 5. Along the base
drawn from the system under measurement. Inserting the plane of this spatial representation introduce a topological
734 Elements: A – Sensors
(S o, e o) Output energy/signal form)
Co gna
Nuclear
si
nt l fo
am rm
Acoustic
(S c
ina
tio
N
, ec
Magnetic
n
)
A
en
Thermal
er
M
gy
Electrical
/
T
E Mechanical
M Optical
O Chemical
C C O M E T M A N
C Input energy/signal form (S i, e i)
O
rm
M
l fo
E na
ig
y /s
T
rg
ne
M
te
or
A
pp
Su
N
s)
e
s,
S
(
Figure 5. The information and energy tetrahedron for a modulating sensor.
grouping consisting of three axes set mutually at 120◦ one base plane surface and three inclined surfaces. This
spacing. Each of these axes, taken as representing one represents a tetrahedron or four-faced object. This way
of the three input groups with their COMETMAN infor- of representing the boundary impacts on sensor systems
mation or energy classification, clarifies, and justifies the may be called the information/energy tetrahedron or the
way in which the acronym aids memory. The organization sensor/transducer effect tetrahedron.
of the three axes on this base plane is a collapsed, two- Illustration of the manner in which the contamination
dimensional representation of the input three-space consist- inputs influence the output may be extended to cover con-
ing of the information, resource, and contamination inputs. tamination from other energy forms. The dot-dot-dash lines
This organization of inputs may be regarded further in Figure 5 show the case when the contaminating energy
as the input triangle. Such a name is appropriate as form is due to the influence of temperature on the wire of
illustrated by the input triangle given by the dotted lines the strain gauge. The deep perception allowed by the sen-
in Figure 5. Here, the support energy form is electrical, the sor/transducer effect tetrahedron given in Figure 5, which is
contamination energy/information form is chemical, and the fairly straightforward, is still realistic and comprehensive.
measurand energy form is mechanical using a sensor like a It is quite important to note that the sensor effect
resistance strain gauge. tetrahedron extends the transducer space by including the
A strain gauge, which measures a mechanical energy possibility for predicting the energy forms of the various
domain variable, has been chosen as a typical, but not contaminating sources that may be present.
exclusive, example. It is straightforward to represent resis- McGhee et al. (1999) illustrate the application of the
tance transducers for variables in other measurand energy sensor effect tetrahedron to a modulating photodiode,
domains. The chemical contamination could be responsible a modulating ultrasonic Doppler flow sensor and self-
for parasitic electrochemical potentials at the contacts in the supporting cross-converting thermocouples.
electrical circuit.
Each point on the input triangle axes may be joined to
an energy/information form on the output axis. When the
9 THE SENSOR EFFECT TETRAHEDRON
output is in the form of an information-bearing electrical FOR SELF-SUPPORTING
energy form, the dash-dot line combinations shown in CROSS-CONVERTING SENSORS
Figure 5 allow the link between all three input forms and
the output form to be clearly seen. The geometrical shape, So far the conditions in Figure 5 correspond to the
which is outlined by the dotted and the dash-dot lines, has case of modulating types of sensors and transducers.
Structure and Energy in Sensor Systems 735
(So, e o) Output energy/signal form)
Co gna
si
Nuclear
nt l fo
am rm
(S c
Acoustic
ina
,ec
N (S i , e i ) (So, e o)
tio
Magnetic
n
A
en
Magnetic field M Thermal
er
gy
T Electrical
/
+ = E Mechanical Magnetic field + Vibration
Vibration M Optical
O Chemical
C C O M E T M A N
Collapsing axis
Input energy/signal form (S i, e i)
Nonexistent support energy/information form (O s, os)
Vibration in a magnetic field induces a nonthermal emf in a thermocouple
Figure 6. The sensor effect tetrahedron for self-supporting cross-converting sensors using a thermocouple as an example.
Self-supporting cross-converting sensor possibilities may to a plane, or two-space, from a three-space or 3-D space,
also be represented by the sensor/transducer effect tetrahe- in the case of self-supporting cross-converting transducers,
dron as shown in Figure 6 using a thermocouple for illustra- the sensor effect tetrahedron collapses to a three-space for
tive purposes. In self-supporting cross-converting sensors, the same kind of sensor. The different results of the col-
there is no support energy form. Consequently, the support lapsing are due to the fact that the sensor effect tetrahedron
energy axis of the input triangle collapses to the origin of represents the input information/energy groupings using the
the system of axes in the tetrahedron. This origin, usually idea of the input triangle illustrated in Figures 5 and 6 with
representing the number zero in a numerical space, may be the added transparency resulting from the inclusion of the
regarded as the point defining the nonexistence of a support interfering energy source forms.
information/energy form. Under these conditions, (2) is the
mathematical representation for the sensor.
The case for the sensor effect tetrahedron representation RELATED ARTICLES
of a thermocouple is illustrated in Figure 6 for two types
of interference/contamination sources. The chemical inter- Article 41, Terminology and Classification of Measure-
ference shown may give rise to a series mode parasitic ment Systems, Volume 1; Article 108, Transducer Fun-
electrochemical emf in the electrical circuit of the ther- damentals, Volume 2; Article 110, Signal/Energy Matrix
mocouple as in the previous example for the modulating Modeling, Volume 2; Article 111, Classification of Sen-
sensor. A situation illustrating the interaction of different sors, Volume 2; Article 112, Systematic Description of
energy forms to produce a propagated contaminating signal Sensors, Volume 2.
is also shown in Figure 6.
If the thermocouple sensor assembly, including its lead
wires, are mounted on a structure that is vibrating in an REFERENCES
electromagnetic field, the sensor effect tetrahedron allows
the possibility of an induced series mode interfering elec- Areny, R.P. and Webster, J.G. (1991) Sensors and signal condi-
tioning, John Wiley & Sons, New York.
trical signal to be predicted and visualized. The powerful
El-Hami, M., Finkelstein, L., Grattan, K.T.V. and Palmer, A.W.
predictive property of the sensor effect tetrahedron is well
(1993) Expert System Application Using Optical Fibre Sensor
illustrated by this example. Classification Data. Sensors and Actuators, A-39, 181.
The representation of self-supporting cross-converting
Fraden, J. (1997) Handbook of Modern Sensors, American Insti-
sensors and transducers shown in Figure 6 may be called tute of Physics, New York.
the self-supporting sensor tetrahedron. Kwaaital, T. (1993) The Fundamentals of Sensors. Sensors and
This model of a collapsing four-space may also be Actuators, 39, 103–110.
viewed as forming a three-space model, when the input and Lion, K.S. (1969) Transducers: Problems and Prospects. IEEE
contaminating axes are drawn mutually perpendicular. This Transactions, IECI-16, 2–5.
is analogous to the effect, which occurs in Stein’s trans- McGhee, J., Henderson, I.A., Korczynski, M.J. and Kulesza, W.
ducer space for self-supporting cross-converting sensors (1999) The Sensor Effect Tetrahedron: An Extended Transducer
given in Figure 4. Whereas this transducer space collapses Space. Measurement, 24, 217–236.
736 Elements: A – Sensors
McGhee, J., Henderson, I.A., Kulesza, W. and Korczynski, M.J. Solopchenko, G.N. (1994) Formal Metrological Components of
(2001) Measurement Data Handling: Theoretical Technique, Measuring Systems. Measurement, 13, 1–12.
ISBN 83-7283-007-X, The Technical University of Lodz, Lodz.
Stein, P.K. (1969) The Engineering of Measurement Systems.
McGhee, J., Korczynski, M.J., Kulesza, W. and Henderson, I.A. Journal of Metals, 21, 40.
(1996) Scientific Metrology, ISBN 83 904299 9 3, A.C.G.M.
Lodart, Lodz. Thomson, A.J. (1911) Introduction to Science, Williams & Nor-
gate Ltd, London.
Middelhoek, S. and Hoogerwerf, A.C. (1986) Classifying Solid-
State Sensors: The ‘Sensor Effect Cube’. Sensors and Actua- Van Dijck, J.G.R. (1964) The Physical Basis of Electronics,
tors, 10, 1. Centrex, Eindhoven/Macmillan, London.
Middelhoek, S. and Noorlag, D.J.W. (1981) Three Dimensional Van Duyn, D.C. and Middelhoek, S. (1990) Information Trans-
Representation of Input and Output Transducers. Sensors and duction in Solid-State Transducers: A General Thermodynamic
Actuators, 2, 29. Systems Approach. Sensors and Actuators, A-21–A23, 25.
Sandquist, G.M. (1985) Introduction to Systems Science, Prentice Ylilammi, M. (1989) Thermodynamics of Sensors. Sensors and
Hall, Englewood Cliffs, NJ. Actuators, 18, 167–178.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
110: Signal/Energy Matrix Modeling
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
Consider first when the sensor is a mercury-in-glass ther-
1 Principles of Signal/Energy Matrix Modeling 737 mometer. The input transduction process transforms the
2 Case Studies in Signal/Matrix Modeling of thermal energy into mechanical energy, due to the cubic
Sensors – Shaft Torque Measurement 738 expansion of the mercury in the bulb of the thermome-
3 Case Studies in Signal/Energy Modeling – ter. This expansion is propagated through the thermometric
Level Measurement 739 fluid, the mercury, by modifying the original mechanical
cubic expansion into a change in the length of the mer-
Related Articles 740
cury column. This column length belongs to the group of
References 740 mechanical variables.
Further Reading 740 Radiation, which is incident upon the column, is modu-
lated by the reflective, refractive, and absorptive properties
of the mercury column so that the human eye perceives
the change in length. This perception is by a noncontact-
1 PRINCIPLES OF SIGNAL/ENERGY ing form of sensing through the 108 videceptors in the
MATRIX MODELING human sense of seeing, which are mentioned with the other
human senses in Figure 2 of Article 5, Humans in the
The basic ideas of classification in sensors and transducers Real World, Volume 1.
allow a visualization of the diverse possibilities for realizing If the thermometer is a digital electronic unit, the thermal
measuring functions. energy is transduced to the electrical domain using a suit-
Another very important aid in visualizing how the energy able electrical temperature sensor, which could be any one
and signal are transduced uses the concept of a sig- within the contacting group in the classification of tem-
nal/energy flow matrix (Middelhoek and Noorlag, 1981). perature sensors. Modification of the electrical energy at
Figure 1 illustrates this for two different kinds of tem- the electrical output port of resistive sensors could use a
perature sensors. One of these is a mercury-in-glass ther- Wheatstone bridge circuit.
mometer and another is a modern digital thermometer. Further modification could be performed with an elec-
The fundamental block diagram representation of input tronic instrumentation amplifier. Subsequently, the analog
and output sensing/transducing elements and modifying signal at the output of the instrumentation amplifier is con-
elements, shown in Figure 2 of Article 109, Structure and verted to a digital form using an analog-to-digital converter,
Energy in Sensor Systems, Volume 2, is used to construct or ADC. The digital output of the ADC is then encoded
the matrix shown in Figure 1. Above this block diagram in and applied to a light emitting diode, or LED. The LED is
Figure 1, is a matrix of rows and columns with the left-hand a photon generator since it transduces its digital electrical
column indicating the COMETMAN classification scheme input signal to visible radiation at its output.
for input energy/signal form. The energy domain of the It is also worthwhile illustrating the application of the
temperature input signal is, of course, the thermal domain. signal/energy matrix model of sensors using two other
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
738 Elements: A – Sensors
Chemical
method, a bonded strain gauge is to be used as the sensor,
while in the second an optical method is recommended.
Optical
Optical energy The block diagrams of two torque-measuring instruments
Mechanical are given in Figures 2 and 3. In Figure 2, strain gauges
Electrical are used to sense changes in applied mechanical force by
Thermal transducing mechanical force to electrical resistance.
Magnetic To measure axial torque, a rosette of strain gauges
(McGhee et al., 1996) should be bonded to the shaft under
Acoustic
measurement. Mechanical torque is converted into strain on
Nuclear
the shaft surface. The strain gauges bonded to the shaft at
angles of 45◦ to the perpendicular, are under strain. Two
Input Output
transducer Modifier transducer of them are under tension and two are under compression.
The strain gauges both change their dimension depending
Figure 1. Generalized but simplified energy regime for a mer- upon changes in the forces on the shaft surface.
cury-in-glass thermometer and a modern digital thermometer. Dimensional changes are transduced to changes in the
resistance of the strain gauges. These resistance variations
examples, which will be presented as case studies. The first are then measured, after suitable conditioning, by applying
example illustrates two methods of measuring mechanical a Wheatstone bridge. The output of the bridge is amplified,
torque, while the second demonstrates two different trans- converted into a digital signal in an ADC, whose output is
duction principles for level measurement. fed to a driver for an LED or an LCD display element. The
LED operates on the principle of optical photon generation.
In the case of the LCD (liquid crystal display), the elec-
2 CASE STUDIES IN SIGNAL/MATRIX trical input modulates the structure of the liquid crystal
MODELING OF SENSORS – SHAFT whose orientation acts as a modulator of ambient photons.
TORQUE MEASUREMENT Recording on computer disk or magnetic tape is a transduc-
tion from the electrical energy form to the magnetic energy
The instantaneous value of axial torque on a shaft is to form. The block diagram of the whole instrument is also
be measured and displayed using an appropriate method. shown in Figure 2. As an exercise, the reader should write
Simultaneously, storage of the measured torque signal is a short description of the energy transductions and signal
required on a computer diskette or magnetic tape for further flows that are shown in Figure 2.
analysis and processing. Only two methods of measuring In the optical method of torque measurement represented
torque will be considered in the proposed design. In the first in Figure 3, the principle of optical modulation is proposed
Chemical
Optical
Mechanical
Electrical
Thermal
Magnetic
Acoustic
Nuclear
Chemical
Force Torque Displacement Resistance Voltage Light
Rosette LCD
Shaft Bonding Bridge circuit/ Display
gauges DC amp driver LED
Modulating
Modifier Mag
input sensor R/W flux
heads
surface
Output transducers
Figure 2. The signal/energy domain matrix and block diagram of a proposed method for axial torque measurement using strain gauges.
Signal/Energy Matrix Modeling 739
Chemical
Optical
Mechanical
Electrical
Thermal
Magnetic
Acoustic
Nuclear
(a)
Chemical
Force Torque Displacement Resistance Voltage Light
Rosette Bridge circuit/ Display LCD
Shaft Bonding
gauges DC amp driver LED
Modulating
Modifier R/W Mag
input sensor
heads flux
surface
(b) Output transducers
Figure 3. (a) The signal/energy domain matrix and (b) block diagram of a proposed method for axial torque measurement using torque
to modulate a light beam.
as a suitable solution. Two rotating circular vanes, each hav- This misalignment causes a reduction in the light that is
ing a regular array of small holes at a fixed radius from their incident on the photodetectors. The applied shaft torque, a
centers, are attached to a shaft with a separation of l mm rotary mechanical energy form, modulates the amount of
between them. Light sources are attached adjacent to the light that is incident upon the photodetectors.
outside face of one vane. An array of stationary photode-
tectors is mounted alongside the outside of the other vane.
The light from the sources passes through the holes 3 CASE STUDIES IN SIGNAL/ENERGY
in the first vane. If the second vane is aligned properly, MODELING – LEVEL MEASUREMENT
there will be a maximum amount of light incident on
the photodetectors. When a torque is applied to the shaft, It is required to measure the level of fluid in a container,
misalignment of the arrays of holes in the vanes will occur. to display the measured value, and to record it upon either
Chemical
Optical
Mechanical
Electrical
Thermal
Magnetic
Acoustic
Nuclear
(a)
Angular
Level Resistance Current Voltage Chemical
displacement
Light
Resistance/ Current/ Display LCD
Float Resistor current voltage driver LED
R/W Mag
Converters flux
heads
(b) surface
Figure 4. (a) The signal/energy domain matrix and (b) block diagram of a proposed method for level measurement using a float and
a rotary potentiometer.
740 Elements: A – Sensors
Chemical
Optical
Mechanical
Electrical
Thermal
Magnetic
Acoustic
Nuclear
(a)
Level Displacement Flux Current Voltage Chemical
Light
Flux/ Current/ LCD
Display
Float LVDT current voltage driver LED
R/W Mag
Converters flux
heads
surface
(b)
Figure 5. (a) The signal/energy domain matrix and (b) block diagram of a proposed method for level measurement using a float and
a linear variable differential transformer, or LVDT.
a magnetic diskette or a magnetic tape. The signal/energy REFERENCES
domain flow diagrams and block diagrams are shown for
each respective system in Figures 4 and 5. Since it is essen- McGhee, J., Henderson, I.A., Kulesza, W. and Korczynski, M.J.
tial to have some primary sensing element to transduce the (1996) Scientific Metrology, ISBN 83-904299-9-3, printed by
level, both the proposed methods of level measurement use A.C.G.M. LODART, Lodz.
a float as a primary sensing element. The first also uses Middelhoek, S. and Noorlag, D.J.W. (1981) Three Dimensional
a rotary potentiometer and the second uses a linear vari- Representation of Input and Output Transducers. Sensors and
Actuators, 2, 29.
able differential transformer (LVDT). As the schematics are
quite explanatory, it is left to the reader to complete a writ- Sydenham, P.H., Hancock, N. and Thorn, R. (1992) Introduction
to Measurement Science and Engineering, Wiley, Chichester.
ten description of both – the matrix and the block diagram.
Further explanations of the method are given in Syden-
ham, Hancock and Thorn (1992). FURTHER READING
Middelhoek, S. and Hoogerwerf, A.C. (1986) Classifying Solid-
RELATED ARTICLES State Sensors: The ‘Sensor Effect Cube’. Sensors and Actua-
tors, 10, 1.
Article 109, Structure and Energy in Sensor Systems,
Volume 2; Article 111, Classification of Sensors, Vol-
ume 2; Article 112, Systematic Description of Sensors,
Volume 2.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
111: Classification of Sensors
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
For example, a simple electrical strain gauge mounted
1 Need for Classification of Sensors 741 on a thin circular diaphragm can be used to measure many
variables. Table 1 lists many of the methods used.
2 Classification Rules in General 741
What is needed is a universal way to classify sensors
3 Classification by Energy 742 as this assists retrieval of information regarding their exis-
4 Applying the Energy Classification 743 tence, reveals new combinations, and links potential users
5 Three-space Modeling of Transducers and to deeper knowledge about them.
Sensors 743 In the life sciences, classification and taxonomy has been
6 The Sensor Effect Tetrahedron 745 a key branch of scholarship for centuries. It was needed
7 Signal/Energy Matrix Modeling of Sensors to assist understanding of the myriad of living objects.
and Transducers 747 With respect to sensors, however, there has not been much
8 General Principles of Temperature Sensor attention paid to setting up a serious universal classification
Classification 748 schema. This article outlines the main approaches that can
9 Temperature Measuring Ranges of be taken – the energy classification that has a universal
Temperature Sensors 749 application but does not cover the form of sensors, and how
temperature sensors have been classified as an example of
Related Articles 750
a species taxonomy.
Reference 750
2 CLASSIFICATION RULES IN GENERAL
1 NEED FOR CLASSIFICATION OF
Classification in sensing and transduction follows the sci-
SENSORS entific approach for grouping objects using the four main
orders of classification to assist in the ordering.
New ideas for sensors are constantly being invented – and In the First Order, known as Generality or Resemblance,
reinvented. This is not efficient for in many cases the the question of those relations describing likeness between
knowledge on a particular sensor form is already available sensors is considered.
and is redeveloped instead of being advanced with each Sensors and transducers must also be grouped on the
new application and experience. basis of the Second Problem or Order of taxonomy, which
With the existence of hundreds of potential transduction is concerned with their Composition or Collectivity.
principles, and numerous ways in which they can be imple- While this order is concerned with the relationship of
mented, it is possible to form thousands of sensor forms that the parts or elements of a sensor system to the whole
are often little different to others yet to the new inventor, system and vice versa, the Third Order of classification,
seem novel and a contribution. called hierarchy, when applied to sensors and transducers,
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
742 Elements: A – Sensors
Table 1. Use of simple instrumented diaphragm to measure dif- Table 2. A scheme of classification for energy forms.
ferent measurands.
Energy form Characteristic physical phenomenon
Sensor Method of using the diaphragm
Electromagnetic Radio waves, through to cosmic rays
Pressure gauge Strain gauge on round diaphragm in sealed radiation
chamber under pressure Gravitational Attraction between masses
Small deflection Diaphragm used with pin resting on it to Mechanical Lifting, stretching, and movement
meter deflect diaphragm Thermal Kinetic energy of molecules
Sound pressure Sensitive form of pressure gauge above with Electrical Electric and magnetic fields, currents
meter large cone connected to the diaphragm Molecular Bond energy of atoms in molecules
Volume sensor Chamber with pressure diaphragm at Atomic Binding energy between nuclei and
bottom measuring force exerted by height electrons
of fluid or solid material Nuclear Binding energy within nuclei
Accelerometer Flexibility of diaphragm with added mass Mass Relativistic energy
forms seismic system
Level sensor Chamber with pressure diaphragm at
bottom measuring force exerted by height Energy and signal are of immediate interest. Energy
of fluid or solid material may be considered on either a macroscopic or microscopic
level.
takes account of the relation of rank between the heads or Serious consideration of the fundamentals of physics
central members of the constituent groups of the elements shows that the most basic starting point might be at its four
comprising sensor systems. These should then be related in prime forces – electric, magnetic, gravitational, and atomic.
the order of composition. These, however, are too much removed from practice to be
In this process, the place occupied by each concrete sen- useful to designers of sensing systems.
sor is assessed in every order relative to other sensors There are a number of different, equally valid ways of
making up both the same order and different orders. Thus, classifying the energy found in the physical world. For
the uniqueness of each sensor in the ordering is distin- example, a comprehensive but not exclusive grouping of
guished by its similarity with, and difference from, every nine various types of existing energy forms is given in
other sensor. Table 2; it stems from experience.
Perhaps the most important Fourth Order in the classifi- From this group of nine, an abbreviated group of six can
cation of sensors hinges upon the notions of kinship through be identified as being of general relevance to instrument
the relations describing ascent, descent, and collaterality. systems. These six, used by Kurt S. Lion and later by Simon
Middelhoek and coworkers, are
As with the orders of Composition and Hierarchy, the
last problem of classification just described, which is also
1. radiant energy
known as Genealogy and Evolution, is also concerned with
2. mechanical energy
the actual sensors to be classified. These latter orders, when
3. thermal energy
applied for sensor systems, emphasize the importance of a
4. electrical energy
functional and structural hierarchy. Taxonomy in Sensor
5. magnetic energy
Science may thus be regarded as possessing the six main
6. chemical energy.
features of its objectives and functions, its materials and
activities, together with the methods of discrimination used Transducers can also be classified according to the trans-
and hierarchical ordering. duction principle involved and the form of the measurand.
Thus, a resistance transducer for measuring displacement
is classified as a resistance displacement transducer. Other
3 CLASSIFICATION BY ENERGY classification examples are pressure bellows or a force
diaphragm. In the classification, grouping on the basis of
Consistent with these orders of classification, it can be seen transduction principle has been developed by a number of
that the appropriate universal grouping of sensors must be people. Kurt S. Lion once produced such a classification
on the basis of: similar to that shown in Table 2. In this classification, it is
possible to relate the physical effect in the energy domain
• function of the measurand that allows a transduction into another
• structure energy form. The topic is also discussed under the modeling
• energy form of sensors in Article 112, Systematic Description of Sen-
• signal output. sors, Volume 2.
Classification of Sensors 743
4 APPLYING THE ENERGY another. Rather, it is from the convenience of that which is
CLASSIFICATION most easily remembered.
Although these methods of grouping by energy form are 5 THREE-SPACE MODELING OF
perfectly acceptable, they are not easily remembered in
a consistent manner. For this reason, another possibility TRANSDUCERS AND SENSORS
is organization by energy classification given by Peter K.
Stein in his presentation of the transducer space. It is possible to use this COMETMAN classification of
In this scheme, the energy forms are classified with energy to visualize the wide range of possible input and
the memory-aiding acronym COMETMAN. Whereas the output transducers from two approaches.
previous methods of classification are based upon a more The first approach uses a three-dimensional or three-
rigid linking of the energy by similarity, COMETMAN is space energy and/or signal form of the inputs and outputs to
specifically designed as a memory aid. The acronym comes provide an aid to understanding the interacting relationship
from classifying by between the form of support energy used and the input
and output information carrying signals in their range of
• Chemical (i.e. molecular) possible energy forms.
• Optical (i.e. radiation) This three-dimensional transducer space represents the
• Mechanical (i.e. including gravitation and mass) input signal in its energy form along the conventional x-
• Electrical axis, the support energy form along the conventional z-axis,
• Thermal and the output energy along the conventional y-axis as
• Magnetic (i.e. including electrical and atomic forms) shown in Figure 1.
• Acoustic (i.e. correctly speaking within the mechanical Self-generating transducers and modifiers, which operate
group) without an auxiliary energy source, lie in the vertical x–y
• Nuclear energy forms. plane. Consider, for example, the thermocouple in which
thermal information flows from the thermal energy input on
Once again, the important point to emphasize here is the x-axis to information output on the y-axis transduced
not whether one form of classification is more correct than to the electrical energy form.
y Output energy/signal form
Nuclear
Acoustic
Magnetic
Thermal
Photoconductive cell Thermocouple
Electrical
Mechanical
Optical
To give electrical output Chemical
C O M E T M A N
C Input energy/signal form x
O
M Optical measurand
E Mo
du
El
T lat
es
ec
al
tri
ign
M
ca
m y/s
ls
A
for erg
up
po
n
N
te
rt
or
pp
Su
z
Figure 1. The three-dimensional (three) space for transducers by energy/information.
744 Elements: A – Sensors
Table 3. Examples of transducers and their logical grouping for different input and output signals.
IN
Radiation Mechanical Thermal Electrical Magnetic Chemical
OUT
Radiation Geometric-optical Interferometer; Thermal radiation Kerr cell; Faraday cell Transducer based
system filter Photoelastic transducer Electro- on emission &
transducer luminescent absorption
transducer, spectroscopy
LED
Mechanical Crooke’s Gravity balance; Thermometers; Electrometer; Magnetometer; Hydrometer; Elec-
radiometer; Ballistic Bimetal strip Electrostrictive Magnetostriction trodeposition
Radiation pendulum; transducer; transducer cell
pressure system Bellows; Gear Piezoelectric
box transducer
Thermal Bolometer Adiabatic Mixing Thermal Eddy-current Combustion
system; calorimeter; converter transducer calorimeter
Friction Heat exchanger
calorimeter
Electrical Photoelectric, Resistive; Thermistor; Transformers; Magnetoresistance; Transducers
Photoresistive & Inductive; Thermoelement; Langmuir Hall-effect based on
Photogalvanic Capacitive; Seebeck effect probe; Charge transducer potentiometry,
cells; Piezo resistor collectors; conductimetry,
Photodiode Transistor polarography;
pH-meter
Magnetic Curie-effect Magnetoelastic Paramagnetic Electromagnetic Magnetic recorder Magnetic
radiation meter transducer transducer transducer; resonance
device Coil system
Chemical Photographic Ultrasonic Thermal dye Electrolytic Paramagnetic Chemical system
emulsion chemical indicators integrator; oxygen analyzer modifiers
transducer Electrical
plating
Modulating types of sensors can be shown in the [x, output signal is So and output energy eo . The difference
y, z] ‘transducer space’ in Figure 1 according to which between the self-generating and modulating types is a
support energy source is used. A photo-conductance sensor matter of how ei is supplied to the system as shown in
would have its input information at x = optical, its output at Figure 2.
y = electrical, and its support energy at z = electrical. The In self-generating sensors, the system under measure-
radiating optical input energy incident upon the sensor mod- ment supports the information flow by supplying energy
ulates the flow of electrical energy from the support energy to the sensor.
port to the output energy port. For this reason, modulat- Modulating sensors, which extract negligible energy
ing sensors of this type bring the advantage of much higher from the system under measurement, modulate the support
power gain to measurement systems than do self-generators. energy input. The energy symbol ei is removed to indicate
The energy/information space diagram allows the pre- that negligible energy is drawn from the system under mea-
diction or visualization of transducer possibilities. It is surement. The energy symbol es is inserted in the diagram
clear that there are numerous different possibilities as fea- to model the effect of modulation of es by Si .
sible candidates when proposing a transduction principle It is clear that any instrument stage can be represented
to meet a sensing need. While all known possible combi- by the energy/signal notation in the [x, y, z] transducer
nations of input, output, and support energy forms can be
located in this three-dimensional representation, there are
Si, ei So, eo Si So, eo
still many positions where practical devices have not been
proposed. Table 3 lists many forms in terms of the energy
classification.
Ss, es
An instrument system building block, therefore, has three
(a) (b)
possibilities for inputs and outputs. The self-generating
sensor has an input signal Si , with its associated input self- Figure 2. Block diagram and notation for (a) self-generating and
generating or excitation energy ei . At the same time, its (b) modulating sensors.
Classification of Sensors 745
Extrinsic boundary
Controlling Parameter Internal
input fluctuations fundamental
fluctuations
Information/ Information/
energy input Intrinsic Boundary energy output
Sensor
system
Support
energy/signal form
Direct input (resource) Disturbances and Direct output
contamination fluctuations in contamination
power flows
Figure 3. Boundary perspective of a sensor system and its influencing effects.
space, in which the measurand energy form, x, modulates S - Signal form;
the support energy form, z, to produce a variation in Input transducers e - Energy form
S c /e c Contamination Suffixes:-
the output energy form, y, bearing some relationship to
i - Input; C - Contamination
the measurand energy form, x. These possibilities are all Si /ei So /eo o - Output
included in Figure 2. The diagonals in the blocks represent Self-generator
the general transduction from one energy form to another.
The relationships on Figure 1 and Figure 2 may be written Support energy Output
Contamination eS Modifier
as a mathematical relation with the form S c /e c transducer
Si /ei So /eo
y = f (x, y) (1) Modulator
6 THE SENSOR EFFECT TETRAHEDRON Figure 4. Functional block diagram of a basic instrumenta-
tion system illustrating modifier elements and input and output
The three-dimensional transducer space has one signifi- transducers.
cant disadvantage. In this diagram, no account is taken of
contaminating inputs. Consequently, the influence of inter-
ference from whatever source is influencing the system is Hence, the output energy/signal form for a sensor system
not transparent. A spatial representation for the boundary is given more emphasis in this figure than the contamination
inputs and outputs of a sensor system, in its most gen- produced by the sensor system at its output.
eral form, requires a multidimensional space. As there is The proposed representation consists of one axis nor-
a total of three groups of inputs and three groups of out- mal to a base plane as shown in Figure 5. Along the base
puts, the representation becomes too complex for graphical plane of this spatial representation, introduce a topological
illustration. grouping consisting of three axes set mutually at 60◦ spac-
This situation can be understood by starting from the ing. Each of these axes, taken as representing one of the
various perspectives at the boundaries of a sensor system, as three input groups with their COMETMAN information or
are given in Figure 2, Figure 3, and Figure 4. It is possible energy classification, clarifies and justifies the way in which
to represent a sensor and transducer effect space that is the acronym aids memory. The organization of the three
more complex than the schemes proposed previously. axes on this base plane is a collapsed, two-dimensional
In spite of the increased complexity, this sensor effect representation of the input three-space consisting of the
space is still useful, since it also includes all of the possi- information, resource, and contamination inputs. This orga-
bilities for contamination inputs. nization of the inputs may be regarded further as the input
In a sensor system, the main output is an information triangle. Such a name is appropriate as illustrated by the
output, which also corresponds to the resource output. input triangle given by the dotted lines in Figure 5. Here,
Although contamination is also produced at the output, it the support energy form is electrical, the contamination
is not necessary to consider it for the present purposes. energy/information form is chemical, and the measurand
746 Elements: A – Sensors
Co
(So, eo) Output energy/signal form
nt
am
ina
Nuclear
tio
n
Acoustic
en
er
(S c
gy
N Magnetic
,e
/si
c
)
gn
A
al
Thermal
fo
M
rm
Electrical
T
E Mechanical
M Optical
O Chemical
C C O M E T M A N
C
Input energy/signal form (S i, e i)
O
M
m
E
r
fo
al
ign
T
y/s
M
rg
ne
A
te
or
pp
N
Su
s)
e
s,
S
(
Figure 5. The information and energy tetrahedron for a modulating sensor.
energy form is mechanical, using a sensor such as a resis- Extending (1) to include the effects of contamination
tance strain gauge. is easily understood. Hence, the functional relationship
The chemical contamination could be responsible for between the output energy/signal forms can be written as
parasitic electrochemical potentials at the contacts in the
electrical circuit. Each point on the input triangle axes may So Si S S
be joined to an energy/information form on the output axis. =f , s , c (2)
eo ei es ec
When the output is in the form of an information bearing
electrical signal, the dash-dot line combinations shown in So far, the conditions in Figure 5 correspond to the case
Figure 5 allow the link between all three input forms and of modulating types of sensors and transducers, which are
the output form to be clearly seen. The geometrical shape, represented in Figure 2 and Figure 4.
which is outlined by the dotted and the dash-dot lines, Self-generating sensor possibilities may also be repre-
has one base plane surface and three inclined surfaces. sented by the sensor/transducer effect tetrahedron as shown
This represents a tetrahedron or four-faced object. This in Figure 6 using a thermocouple for illustrative purposes.
way of representing the boundary impacts on systems In self-generating sensors, there is no support energy
may be called the information/energy tetrahedron or the
form. Consequently, the support energy axis of the input
sensor/transducer effect tetrahedron.
triangle collapses to the origin of the system of axes in the
Illustration of the manner in which the contamination
tetrahedron. This origin, usually representing the number
inputs influence the output may be extended to cover
zero in a numerical space, may be regarded as the point
contamination from other energy forms. The dot-dot-dash
defining the nonexistence of a support information/energy
lines in Figure 5 show the case when the contaminating
form. Under these conditions, (2) can be rewritten in the
energy form is due to the influence of temperature on the
form
wire of the strain gauge. The deep perception allowed by the
information/energy tetrahedron given in Figure 5, which is So Si Os S
=f , , c (3)
fairly straightforward, is still realistic and comprehensive. eo ei os ec
Classification of Sensors 747
Co
nt
am
(So, eo) Output energy/signal form
ina
t
Nuclear
ion
en
Acoustic
er
(S c
gy
N
, ec
/si
Magnetic
gn
)
A
al
Thermal
for
Magnetic field M
m
T Electrical
+ E Mechanical
Vibration M Optical
O
Chemical
C C O M E T M A N
Collapsing axis Input energy/signal form (Si, ei)
Nonexistent support energy/information form (Os, os)
Vibration in a magnetic field induces a nonthermal emf in a thermocouple
Figure 6. The sensor effect tetrahedron illustrating self-generating sensors and transducers using a thermocouple as an example.
where the term [Os /os ] indicates that there is no support 7 SIGNAL/ENERGY MATRIX MODELING
signal/energy form for self-generating sensors.
OF SENSORS AND TRANSDUCERS
In the case of the thermocouple, illustrated in Figure 4,
the measurand energy form is thermal. The chemical inter-
ference shown once again may give rise to a series mode The basic ideas of classification in sensors and transducers
parasitic electrochemical emf in the electrical circuit of allow a visualization of the diverse possibilities for real-
the thermocouple. A situation, illustrating the interaction of izing measuring functions. Another very important aid in
different energy forms to produce a propagated contaminat- visualizing how the energy and signal are transduced uses
the concept of a signal/energy flow matrix. Figure 7 illus-
ing signal, is also shown in Figure 6. If the thermocouple
trates this for two different kinds of temperature sensor. One
sensor assemblies, including its lead wires, are mounted
of these is a mercury-in-glass thermometer and another is
on a structure, which is vibrating in an electromagnetic
a modern digital thermometer. The fundamental block dia-
field, the sensor effect tetrahedron allows the possibil-
gram representation of input and output sensing/transducing
ity of an induced series mode interfering electrical signal
to be predicted and visualized. The powerful predictive
property of the sensor effect tetrahedron is well illustrated Chemical
by this example.
Optical
The representation of self-generating sensors and trans-
ducers shown in Figure 6 may be called the self-generator Mechanical
tetrahedron. This model of a collapsing four-space may also
Electrical
be viewed as forming a three-space model, when the input
and contaminating axes are drawn mutually perpendicular. Thermal
This is analogous to the effect that occurs in Stein’s trans- Magnetic
ducer space for self-generating sensors given in Figure 1.
Whereas this transducer space collapses to a plane, or two- Acoustic
space, from a three-space or three-dimensional space, in Nuclear
the case of self-generating transducers, the sensor effect
tetrahedron collapses to a three-space for the same kind of
Input Output
sensor. The different results of the collapsing is due to the transducer
Modifier
transducer
fact that the sensor effect tetrahedron represents the input
information/energy groupings using the idea of the input Figure 7. Generalized but simplified energy regime for a mer-
triangle illustrated in Figure 5 and Figure 6. cury-in-glass thermometer and a modern digital thermometer.
748 Elements: A – Sensors
elements and modifying elements, shown in Figure 4, is photon generator, transduces its digital electrical input sig-
used to construct the matrix shown in Figure 7. Above nal to visible radiation at its output.
this block diagram in Figure 7 is a matrix of rows and
columns, with the left-hand column indicating the COMET-
MAN classification scheme for input energy/signal forms. 8 GENERAL PRINCIPLES OF
The energy domain of the temperature input signal is of TEMPERATURE SENSOR
course the thermal domain. CLASSIFICATION
Consider first when the sensor is a mercury-in-glass
thermometer. The input transduction process transforms While little systemic classification work has been con-
the thermal energy into mechanical energy, because of tributed for the sensor family as a whole, a significant
the cubic expansion of the mercury in the bulb of the foundation showing the general rules of classification has
thermometer. This expansion is propagated through the been developed for temperature. It provides an example that
thermometric fluid, the mercury, by modifying the original illustrates the characteristics and scope for other categories.
mechanical cubic expansion into a change in the length of Elements for measuring temperature extend the human
the mercury column. This column length belongs to the faculties to sense hotness relations between bodies or enti-
group of mechanical variables. Radiation, which is incident ties in the real world. This functional extension of the
upon the column, is modulated by the reflective, refractive, human faculties is held in common with other instruments
and absorptive properties of the mercury column so that the for measurement, calculation, communication, and control.
human eye perceives the change in length. This perception From the classification perspective, the tree classification
is by a noncontacting form of sensing through the 108 diagram of Figure 8 illustrates benefits that the science of
videceptors in the human sense of seeing. taxonomy can bring to the study of temperature sensors.
If the thermometer is a digital electronic unit, the ther- It is obvious from Figure 8 that there are diverse methods
mal energy is transduced to the electrical domain using a for temperature measurement. Only contacting sensors with
suitable electrical temperature sensor, which could be any heat transfer by convection or conduction will be consid-
one of the contacting group in the classification of tempera- ered in detail. None of the wide variety of noncontacting
ture sensors given in Figure 8, explained later in Section 8. sensors are considered here.
Modification of the electrical energy at the electrical out- To provide breadth of perspective, it is essential to
put port of resistive sensors could use a Wheatstone bridge. introduce some kind of ordering so that similarities between
Further modification could be performed with an electronic each kind may be identified without in any way diminishing
instrumentation amplifier. Subsequently, the analog signal their important differences.
at the output of the instrumentation amplifier is converted to The main function of taxonomy, or classification, is
a digital form using an analog-to-digital converter (ADC). to discover the relationship between different entities that
The digital output of the ADC is then encoded and applied unites them. In the first instance, at the metaphysical, the-
to an LED (light emitting diode). The LED, which is a oretical, or fictitious level, it is necessary to deal with the
Families of the ordering genus Sensing
Temperature Other
Levels of the key
species Contacting Noncontacting
Subspecies by structure
and/or energy form Nonelectrical Electrical
Self-generators
Modulators Self-generators
and modulators
Thermal Thermal Quartz Fibre optic Thermocouples Noise
expansion indicators thermometry thermometry
Conductors Semiconductors
Solid Liquid Gas Pyrometric Refractive
cones Two-terminal Three-terminal Other
index
Bimetallic Liquid variations No One- Two- Chips
Dilatation crystals Black-body junction junction junction
Paints cavities
Capillary Manometric Fluorescence Diodes Transistors
Resistors Thermistor
Liquid- Liquid- Gas-filled
in-glass filled Wired Thin film Silistors
Vapour pressure
Figure 8. Classification of temperature measuring instruments/sensors by structure and/or energy form.
Classification of Sensors 749
generality or resemblance of specific methods of tempera- may belong to the groups of sensor known as either self-
ture measurement. This is concerned with establishing the generators or modulators.
relationship of likeness. The generality and resemblance level in temperature
The second important problem in the classification of sensor ordering is best approached by comparison with the
temperature sensors should analyze two relationships. The functions of the human senses. Humans can detect hotness
first is that between those similar subspecies of sensing relations by looking at an object, by approaching it, or by
methods. The second is the relationships between one part touching it.
of a temperature sensor with other parts, which together Neither looking at nor approaching an object requires
collectively contribute to the composition of the sensor. It physical contact to sense its hotness. Touching an object to
is also necessary to construct a hierarchical ordering based sense its hotness requires physical contact. Thus, the senses
upon the relationship between the heads or central members of touch, by contact, and sight or proximity, with no contact,
of groups of sensors. are those used by humans to sense hotness relations.
Composition and collectivity play a role in this ordering, It is apparent that temperature sensors may be based upon
as they relate the places occupied in each order, relative to heat transfer by contacting or noncontacting methods. This
other sensors of the same order. Another last, but not least, grouping at the species level of the hierarchy of temperature
ingredient in ordering temperature sensors is concerned sensors, classifies them on the basis of these heat transfer
with the kinship of one sensor type with another type. mechanisms. This approach to temperature measurement
is recognized as the direct form of measurement in the
The relationships of ascent, descent, and collaterality are
thermal field.
the essential factors in this aspect of temperature sensor
Other possibilities for the noncontacting species use
ordering. Ascending and descending relationships place
inferential methods. The principle of this technique applies
the sensor by structure and energy form relative to the
an external energy as an interrogating medium in the
head or central members of principal groups of sensors.
measuring scheme for the purposes of information capture
Collaterality is that relationship which one type of sensor
about the abilities of the body under measurement to store,
has with another at the same hierarchical level in the
dissipate, transmit, or transform thermal energy.
classification.
Applying the above principles, it can be seen that tem-
perature sensors are a specific genus of sensor within the 9 TEMPERATURE MEASURING RANGES
much wider family of sensors. This is emphasized by the OF TEMPERATURE SENSORS
branch that shows that other genus of sensor exist. Also,
in accordance with the general classification of sensors by It is also possible to classify temperature sensors on the
their energy relationships given earlier, temperature sensors basis of the temperature range of application. Such a
thermometers
Total radiation pyrometers
Radiation
Photoelectric pyrometers 3000°C
Disappearing filament thermometers
Two- colour (ratio) pyrometers 3500°C
Resistance thermometers (Conductors: wire/foil/film)
Thermo-
resistors
Modulators
thermometers
Electrical
Thermistor thermometers
conductor
Semi-
Contacting thermometers
Silicon resistors
Diodes and Transistors
Thermocouples Self-generator
Vapour pressure thermometers
Typical Extreme
Liquid filled thermometers
thermometers
Nonelectrical
range range
Mercury-in-glass thermometers
Organic liquids-in-glass thermometers
Dilatation thermometers
Bimetallic thermometers
Scale of 1990
Temperature
International
Radiation thermometer
Platinum resistance thermometer
0 500 1000 1500 2000 °C
Figure 9. Classification of temperature measuring instruments/sensors by measuring range.
750 Elements: A – Sensors
classification is given in Figure 9. An abbreviated form Volume 2; Article 110, Signal/Energy Matrix Model-
for the temperature ranges of the standard thermometers ing, Volume 2; Article 112, Systematic Description of
of ITS-90 is included in this diagram for the purposes Sensors, Volume 2; Article 114, Models of the Sensor
of comparison. Interface, Volume 2; Article 115, Designing the Sensor
A simple classification tree on the various types of Interface, Volume 2.
elastic flexure used in instruments has been published,
Sydenham (1984).
REFERENCE
RELATED ARTICLES
Sydenham, P.H. (1984) Elastic Design of Fine Mechanism in
Article 108, Transducer Fundamentals, Volume 2; Instruments. Journal of Physics E: Scientific Instruments, 17,
Article 109, Structure and Energy in Sensor Systems, 922–30.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
This article is based in part on a chapter by J. McGhee et al originally published in Measurement Data Handling: Hardware Techniques
Vol. 2, J. McGhee et al , Technical University of Lodz, 2001.
112: Systematic Description of Sensors
Paul P.L. Regtien
University of Twente, Enschede, The Netherlands
occur in physics. The description is, in particular, useful
1 Principles of Systematic Description of in the field of material research and optimization of sen-
Sensors 751 sor materials. Derived from the thermodynamic approach
2 Systematic Description Based on Energy is the Bond graph notation (Blundell, 1982) with a division
Domains 752 of variables into two groups called effort and flow variables.
3 Systematic Description of Physical Quantities 753 This method is not only useful for the description of sen-
4 Systematic Description Based on sors but has great significance in the design of all kinds of
Thermodynamic Laws 754 technical systems, irrespective of the domain type.
A further distinguishing property for a subdivision of sen-
5 A Unified Approach to Transducer Modeling 756
sors into just two categories is the need for auxiliary energy
References 757
(see Article 108, Transducer Fundamentals, Volume 2).
Sensors that need no auxiliary energy for their operation
are called direct sensors or self-generating sensors. Sen-
1 PRINCIPLES OF SYSTEMATIC sors that use an additional energy source for their operation
DESCRIPTION OF SENSORS are called indirect sensors, modulating sensors, or inter-
rogating sensors. The names direct and indirect are rather
A sensor performs the exchange of information (thus obscure; the name self-generating is misleading: neither the
energy) from one domain to another and therefore it oper- energy nor the information is generated by the sensor for it
ates on the interface between different physical domains. comes from the measurand.
Several frameworks have been developed for a systematic The terms modulating and interrogating refer to the fact
description of sensors. Basically, they are based on either that the measurand affects a specific material property,
of two different approaches. The first method follows a which in turn is interrogated by an auxiliary quantity. Most
categorization on the basis of the various energy domains. sensors belong to this group: all resistive, capacitive, and
However, there is no clear definition of an energy domain, inductive sensors are based on a parameter change (resis-
leaving room for different opinions on this description. The tance, capacitance, inductance) caused by the measurand.
other approach is based on a categorization of physical Likewise, most displacement sensors are of the modulating
quantities. Although more fundamental, this method also type: displacement of an object modulates optical or acous-
leaves scope for discussions. Obviously, the two approaches tic properties (transmission, reflection, interference), where
are connected to each other, since quantities describe par- light or sound is the interrogating quantity. The distinction
ticular phenomena in which energy or conversion of energy between these two groups can be useful when considering
is involved. offset and load errors of sensors (Article 108, Transducer
A common categorization according to physical effects Fundamentals, Volume 2). In the last section of this arti-
is based on thermodynamic principles. It is the most fun- cle, a universal model of a sensor is presented that avoids
damental, and covers most of the conversion effects that any misinterpretation due to badly chosen names.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
752 Elements: A – Sensors
The operation of sensors is governed by This classification is rather unpractical for the description
of sensors. Stein has proposed a system of eight distinct
• physical effects,
energy forms:
• material properties, and
• system layout. • Chemical
Physical effects may either act within one-energy domain • Optical
or cross domain boundaries. The latter describe the funda- • Mechanical
mental operation of sensors. Material properties determine • Electrical
the quantitative characteristics of the conversion process, • Thermal
and hence form a guide to the proper material choice for a • Magnetic
particular sensor. Finally, with an appropriate sensor layout, • Acoustic/fluid
the performance of the sensor can further be optimized. • Nuclear
forming the acronym COMETMAN (Stein, 1963).
Lion reduced this to only six domains, and adopted
2 SYSTEMATIC DESCRIPTION BASED
the term signal domain (Lion, 1969). These six domains
ON ENERGY DOMAINS are
A systematic representation of sensor effects based on • radiant
energy domains involves a number of aspects. First, the • mechanical
energy domains have to be defined. Next, the energy • thermal
domains should be allocated to both the sensor input and • magnetic
output. Finally, since many sensors are of the modulating • chemical and
type (Article 108, Transducer Fundamentals, Volume 2) • electrical.
also, the domain of the auxiliary quantity should be con-
The number of domains is actually a rather arbitrary
sidered.
choice, so for practical reasons we continue with the
From a physical point of view, the nine energy forms that
system of six domains and remain calling them energy
can be distinguished are (Middelhoek and Audet, 1989)
domains.
• electromagnetic radiant energy Information contained in each of the six domains can
• gravitational energy be converted to any other domain. These conversions can
• mechanical energy be represented in a 6 × 6 matrix. Figure 1 shows that
• thermal energy matrix, including some of the conversion effects. An input
• electrostatic and electromagnetic energy transducer or sensor performs the conversion from a non-
• molecular energy electrical to the electrical domain (the shaded column), and
• atomic energy an output transducer or actuator performs the conversion
• nuclear energy from the electrical to another domain (the shaded row in
• mass energy. Figure 1).
In↓ Out→ Radiant Thermal Electrical Magnetic Mechanical Chemical
Luminescense Radiation heating Photoconductivity Photomagnetism Radiation Photochemical
Radiant pressure process
Thermal Incandescense Thermal Seebeck effect; Curie−Weiss law Thermal Endothermal
conductivity Pyroelectricity expansion reaction
Electrical Injection Peltier effect PN junction effect; Ampere’s law Converse Electrolysis
luminescense Ohm’s law piezoelectricity
Magnetic Faraday effect Ettinghausen Hall effect Magnetic induction Converse
effect magnetostriction
Mechanical Photoelastic effect Friction heat Piezoelectricity Magnetostriction Gear Pressure
induced reaction
Exothermal Volta effect Explosive Chemical
Chemical Chemoluminescense reaction
reaction reaction
Figure 1. Conversion effects. (Reproduced from ‘Physics of Silicon Sensors’, Middelhoek et al, Copyrigtht (1989), with permission
from Elsevier.).
Systematic Description of Sensors 753
z ra 3 SYSTEMATIC DESCRIPTION OF
th PHYSICAL QUANTITIES
el
ma Several attempts have been made to set up a consis-
tent framework for quantities and material properties.
me
Most of these descriptions are based on energy consider-
ch ations, or more precisely, on the first and second law of
0 y thermodynamics.
ch ch me ma el th ra
me For each energy domain, two variables can be defined
ma in such a way that their product equals the energy (or a
el derived energy quantity) of the domain type, the conjugate
th
ra quantities. For instance:
x
Mechanical: Tension T (N m−2 ) and deformation S [–]
Figure 2. Sensor cube with four transducer types; thermocouple:
[th, el, 0], Hall sensor: [ma, el, el], LED: [el, ra. 0] and LCD: [ra,
Shear tension τ (N m−2 ) and shear angle
ra, el]. χ [–]
Electrical: Field strength E (V m−1 ) and dielectric
displacement D (C m−2 )
This two-dimensional representation can be extended to Field strength E (V m−1 ) and electric
three dimensions, when the interrogating energy domain polarization P (C m−2 )
is included. This gives 216 energy-triplets. To get a clear Magnetic: Magnetic induction B (Wb m−2 ) and
overview of all these possible combinations, they can be magnetic field H (A m−1 )
represented in a 3D Cartesian space, the sensor cube shown Thermal: Temperature (K) and entropy σ (J K−1
in Figure 2. The three axes refer to the input energy or J Km−3 ).
domain, the output energy domain, and the domain of the
interrogating input. The variables E, D, B, H, T, and S are vector vari-
Each of the 216 elements of the 6 × 6 × 6 matrix repre- ables, whereas σ and are scalars (so often denoted as
sents one or more conversion effects. When restricting to σ and ). Note that the dimension of the product of
electrical transducers, we have 5 direct input transducers, 5 each domain pair is J m−3 (energy per unit volume) in
direct output transducers, 25 modulating input transducers, all aforementioned cases (note a possible confusion with
and 25 modulating output transducers. To facilitate nota- respect to the thermal domain).
tion, the transducers can be indicated by indices, like in The variables in this list show a remarkable analogy:
crystallography, the so-called Miller indices: [x, y, z]. The they belong to either of two classes: through-variables
x-index is the input domain, the y-index, the output domain and across-variables. To explain this classification, we first
and the z-index, the domain of the interrogating quan- introduce the term lumped element. A lumped element
tity. With these three indices, a transducer can be typified symbolizes a particular property of a physical component;
according to the energy domains involved. Some exam- that property is thought to be concentrated in that element
ples are between its two end points or nodes. Exchange of energy
or information only occurs through these terminals.
A through-variable is a physical quantity that is the same
• direct input transducer: thermocouple: [th, el, 0]
for both terminals of the lumped element.
• modulating input transducer: Hall sensor; [magn, el, el]
An across-variable describes the difference with respect
• direct output transducer: LED: [el, rad, 0]
to a physical quantity between the terminals. They are
• modulating output transducer: LCD: [rad, rad, el].
also called intensive variables and extensive variables
respectively.
These transducers are also visualized in Figure 1. The In an electronic system, current is a through-variable,
value of such a representation is rather limited. It may voltage (or potential difference) is an across-variable.
serve as the basis of a categorization for overviews, or as Therefore, through-variables are called generalized I-vari-
a guide in the process of sensor selection. This also shows ables and across-variables are called generalized V-vari-
why there are so many alternatives for sensing a given ables. However, this is just a matter of viewpoint. It
variable, and why the uncontrolled generation of so-called is perfectly justified to call them generalized forces and
new sensors is hard to keep in clear perspective. displacements.
754 Elements: A – Sensors
In the above groups of variables, T, E, and are across- The work dW is the sum of the different energy forms
variables. On the other hand, S, D, and σ are through- involved, for instance, electrical and mechanical energy:
variables. They are related to each other through physical V dQ and F dx. Since dQ = dσ (J) (temperature and
material properties or system layout. entropy), we can generalize the energy change as
For each domain, we can define a pair of conjugate
variables. In the preceding examples, the product was
dU = Ai dBi (J) (5)
energy density (J m−3 ). It is also possible to find other
i
pairs of variables whose product is, for instance, energy or
power. A few examples are
where Ai is an extensive variable and Bi the conjugated
−2 3 intensive variable. So if we consider only the thermal,
mechanical: pressure P (N m ) and volume V (m );
mechanical: force F (N) and distance x (m); electrical, and mechanical domains, the energy change is
electrical: voltage V (V) and charge Q (C); governed by the equation
all with product energy (J); these pairs are energy conjugate
dU = dσ + V dQ + F dx (J) (6)
variables.
Other examples are
When energy per unit of volume is considered, this equation
mechanical: force F (N) and velocity v (m s−1 ); reads as
electrical: voltage V (V) and current I (A = C s−1 );
both with product power (W); these pairs are power dU = T dS + E dD + dσ (J m−3 ) (7)
conjugate.
For other domains, such pairs can be defined as well, but (Note that for entropy we have used the same symbol
with sometimes rather unusual quantities. but the dimension is now J Km−3 .) Apparently, in this
From the last examples, we observe regularity in the equation only through-variables affect the system. If, on
various domains. Within one domain, we can distinguish the other hand, only across-variables affect the energy state
state variables and rate variables. They are related as of the system, the equation for the energy change per unit
volume is
d
Xrate = Xstate = Ẋstate or Xstate = Xrate dt (1)
dt
dG = −S dT − D dE − σ d (J m−3 ) (8)
The most obvious cases are
d where G is the Gibbs potential that can be found from
I= Q or Q = I dt (2)
dt the free energy U by a Legendre transformation. We con-
d tinue the discussion with the last expression because the
v = x or x = v dt (3) resulting parameters are more in agreement with experi-
dt
mental conditions. After all, it is easier to have the across-
variables as inputs or independent quantities (temperature;
electric field, or voltage; force) and to measure the resulting
4 SYSTEMATIC DESCRIPTION BASED
through-variables or dependent quantities (strain; dielectric
ON THERMODYNAMIC LAWS displacement or current).
The energetic state of a system can be described by
On the basis of laws on energy conservation, several frame- pairs of conjugate variables, the pairs being determined by
works for a systematic description of material properties the energy forms that have to be regarded for a specific
and physical effects and hence for the description of sensors situation. However, the system configuration or the material
can be built up. couples the conjugate variables of each pair. Examples
According to the first law of thermodynamics, the energy within one domain are
content of an infinitely small volume of an elastic dielectric
material changes by adding or extracting heat dQ (J) and
T = c × S (Hooke’s law, with c, the mechanical elasticity)
by work dW (J) exerted upon it is given by
P = ε0 χe E (with χe , the electrical susceptibility)
dU = dQ + dW (J) (4) J = µ0 χm H (with χm , the magnetic susceptibility)
Systematic Description of Sensors 755
In general, the extensive variable, Ai and the intensive ∂σ ∂σ ∂σ
dσ = dT + dE +
variable, Bi within one domain are connected according to ∂T E, ∂E ∂ E
Ai = αi Bi or Bi = βi Ai (9) Combination of (13) and (14) results in
2
where αi is a (generalized) modulus and βi is a (gen- ∂ 2G ∂ 2G ∂ G
eralized) compliancy. These are the state equations of S= T + E+
,E ∂T 2∂T ∂E ∂T ∂ E
the system. 2 2 2
As an example, in the mechanical domain, Hooke’s law ∂ G ∂ G ∂ G
D= T + 2
E+
can also be written as ∂E∂T ∂E ,T ∂E∂ T
(15)
2 2 2
T = c × S or S = s × T (10) ∂ G ∂ G ∂ G
σ = T + E+
∂∂T E ∂∂E T ∂2 E,T
where c is the modulus of elasticity (or Young’s modulus)
and s, the compliance. Now we have a set of equations connecting the (depen-
Likewise, variables in one domain are linked to variables dent) through-variables S, D, and σ with the (independent)
in another domain, expressing the cross-effects. So across-variables T , E, and . The second-order derivatives
represent material properties; they have been given special
Ai = γij Bj or Bj = γj i Ai (11) symbols.
The constant variables are put as superscripts, to make
where γij and γj i represent material properties linking
place for the subscripts denoting orientation (applied in
domains i and j.
nonisotropic materials).
The aforementioned equations form the basis for a formal
description of sensors, since it is now possible to describe
S = s E, T + d E + α E
physical effects crossing domain boundaries. We illustrate
the concept with a system in which only the mechanical, D = d T + ε,T E + p T (16)
the electrical, and the thermal domains are involved. So the ρ
σ = α E T + p T E + cE,T
starting point is the equation: T
dG = −S dT − D dE − σ d(J m−3 ) (12) These superscripts denote constancy with respect to the
indicated parameters, for instance, s E, is the compliance
The through-variables S, D and σ in this equation can be at zero electric field E and constant temperature .
written as partial derivatives of the Gibbs potential: The nine associated effects are displayed in Table 1.
Table 2 shows the corresponding properties. The param-
∂G eters for just a single domain are ε, cp , and s. The other
S(T , E, ) =
∂T ,E parameters (p, α, and d) denote cross-effects.
Note that direct piezoelectricity and converse piezoelec-
∂G
D(T , E, ) = (13) tricity have the same symbol (d) because the dimensions
∂E T , are equal (m V−1 and C N−1 ). The same holds for the
∂G pair pyroelectricity and converse pyroelectricity, as well as
σ (T , E, ) =
∂ T ,E for thermal expansion and piezocaloric effect. Further, the
second order derivatives in (15) correspond with the param-
From these equations, we can derive the various material eters γij of (11). Since the order of differentiation can be
and sensor parameters. To that end, the variables S, D, reversed, the parameters γij and γj i are equal.
and σ are approximated by linear functions, that is, we
take only the first term of the Taylor series expansion
Table 1. Nine physical effects corresponding to the parameters
in the points T = 0, E = 0 and = 0 of the functions in (16).
S(T , E, ), D(T , E, ), and σ (T , E, ):
Elasticity Converse Thermal expansion
∂S ∂S ∂S piezoelectricity
dS = dT + dE +
∂T E, ∂E ∂ E
Direct Permittivity Pyroelectricity
piezoelectricity
∂D ∂D ∂D Piezocaloric effect Electrocaloric Heat capacity
dD = dT + dE + (14) effect
∂T E, ∂E ∂ E
756 Elements: A – Sensors
Table 2. Symbols, property names, and units of the effects in Fx Fy
Table 1.
Ex System Ey
Symbol Property Unit
s Compliance, elasticity m2 N−1 (a)
d Piezoelectric constant mV−1 = CN−1 Fy
Fx
α Thermal expansion coefficient K−1
p Pyroelectric constant Cm−2 K−1 Ex System Ey
ε Permittivity, dielectric constant Fm−1
cE,T (Specific) heat capacity Jkg−1 K−1
Fz
Equations (7) and (8) can be extended just by adding (b) Ez
other couples of conjugate quantities, for instance, from
the chemical or the magnetic domain. Obviously, this intro- Figure 3. Port models of a sensor: (a) two-port (direct sensor)
and (b) three-port (modulating sensor).
duces many other material parameters. With three couples
we have nine parameters, as listed in Table 1. With 4
couples of intensive and extensive quantities we get 16 where the measurand affects specific material or geometric
parameters, so 7 more (for instance, the magnetocaloric parameters. To extract the information from such a sensor,
effect, expressed as the partial derivative of entropy to mag- it has to be interrogated using an auxiliary signal. The infor-
netic field strength). mation stored in the sensor is available latently, in the latent
information parameters or LIP (Stein, 1992). These param-
eters are modulated by the input signal, and interrogated by
5 A UNIFIED APPROACH TO
the auxiliary or interrogating input.
TRANSDUCER MODELING At zero input, the LIPs of a modulating sensor have
initial values, set by the material and the construction.
As a consequence of the energetic basis for the sen- Generally, the input has only a small effect on these
sor description in the preceding sections, the information
parameters, resulting in relatively small deviations from the
exchange between a sensor (or measuring system) and
initial values.
its environment (input devices, output devices, auxiliary
Note that direct sensors also have LIPs set by materials
sources, interfering sources) always implies two quanti-
and construction. They determine the sensitivity and other
ties: an across-variable and an associated through-variable.
transfer properties of the sensor. So, the input port of all
Energy (and thus information) enters or leaves the system
sensors can be denoted as LIP-input port. As a consequence,
through a pair of terminals making up a port. We distin-
any sensor can be described with the three-port model of
guish input ports and output ports. A direct sensor can be
Figure 3(b). Only the functions of the ports may differ,
described by a two-port model or a four-terminal model
notably the LIP-input port and the interrogating-input port.
(Figure 3a). The input port is connected to the measurand
According to the unified transducer model as introduced
and the output port corresponds with the electrical connec-
by Stein, an input port can be controlled either by design (it
tions of the sensor. Likewise, a modulating sensor can be
conceived as a system with three ports: an input port, an has a fixed value) or by the environment (the measurand or
output port, and a port through which the auxiliary energy some unwanted input variable). So we have four different
is supplied (Figure 3b). cases (Figure 4).
In these models, the across-variables are denoted by E
(effort) and the through-variables by F (flow) respectively.
For each domain, the two variables E and F are defined INT.
Environment
LIP input Design controlled
in such a way that their product equals the energy of the controlled
input
domain type or a time or space derivative of energy. The
subscripts x, y, and z are chosen in accordance with the Design controlled Source
Direct
sensor cube from Figure 1. sensor
Direct sensors provide the information about the measur-
and as an output signal, an energetic quantity. Modulating Environment Modulating Multiplying
controlled sensor devices
sensors contain the information as the value of material
property, or a geometric quantity, not an energetic signal.
The information enters the system through the input port, Figure 4. Unified transducers.
Systematic Description of Sensors 757
The characteristics of these four cases are briefly 4. Environment-controlled LIP-input and environment-
reviewed. controlled interrogating input: These are multiplying
transducers: the output depends on the quantities at
1. Design-controlled LIP-input and design-controlled
both inputs, often in a multiplicative relation. For
interrogating input. All inputs are fixed. This type
instance, a Hall sensor could act as such when
represents a signal or information source, for instance,
the interrogating input is not a fixed current (by
a standard, or a signal source with a constant or
design) but a current that is related to just another
predetermined output. The output is totally determined
measurand.
by the construction and the materials that have been
chosen. Any environmental effect on the output is It is important to note that any practical transducer shows
(ideally) excluded. all four types of responses to varying extents. A strain
2. Design-controlled LIP-input and environment-control- gauge (a modulating transducer) produces, when inter-
led interrogating input. Since the latent information rogated, an output voltage related to the strain-induced
parameters are fixed by design, the output only depends change in resistance. But the circuit can also generate
on what is connected to the interrogating input. When spurious voltages caused by capacitively or magnetically
this is the measurand, the transducer behaves as a induced signals. A thermocouple (a direct transducer)
direct sensor. produces an output voltage proportional to the measur-
Examples are as follows: and at the interrogating input. If, however, the material
parameters change due to (for instance) strain or nuclear
• Thermocouple sensor: the Seebeck coefficient is radiation (inputs at the LIP port) the measurement is
fixed by the choice of the materials. corrupted.
• Piezoelectric accelerometer: the sensitivity is fixed Since just one response is desired, other responses should
by the seismic mass and the piezoelectric properties be minimized by an appropriate design. This universal
of the crystal. approach helps identify such interfering sensitivities and
their relative importance.
3. Environment-controlled LIP-input and design-control-
led interrogating input. The measurand affects
particular material properties or geometric parameters.
REFERENCES
These changes are interrogated by a fixed or well
defined signal at the interrogating input. The transducer Blundell, A.J. (1982) Bond Graphs for Modelling Engineering
behaves as a modulated sensor. Systems, Ellis Horwood Publishers, Chichester.
Examples are as follows: Lion, K. (1969) Transducers: Problems and Prospects. IEEE
Transaction on Industrial Electronics and Control Instruments,
• Strain gauge bridge: strain alters the resistance of 16, 2–5.
the strain gauge(s); a bridge voltage converts this
Middelhoek, S. and Audet, S. (1989) Silicon Sensors, Academic
resistance change into an output voltage. Press, London.
• LVDT (Linear Variable Differential Transformer): a
Stein, P.K. (1963) Classification System for Transducers and
displacement of an object connected to the moving Measuring Systems, in Symposium on Environmental Measure-
core of the LVDT will change the transfer ratio of ments: Valid Data and Logical Interpretation, 4–6 Septem-
the differential transformer. An AC signal on the ber; US Department of Health Education and Welfare, 1964
primary coil acts as an interrogating quantity. (pp. 65–84).
• Hall sensor: the measurand is a magnetic induc- Stein, P.K. (1992) The Unified Approach to the Engineering
tion field that acts on moving charges imposed by of Measurement Systems – Basic Concepts, Stein Engineering
a fixed (or known) current applied to the interro- Services, Phoenix, AZ.
gating input.
113: Force-feedback Sensors
Barry E. Jones
Brunel University, Uxbridge, UK
force). The resultant error is usually transduced into elec-
1 Feedback-measuring System 758 trical form and amplified to give the output indication.
2 General Properties of Feedback Systems 758 Normally, a sensor and associated circuit has a nonelectrical
3 Stability Considerations 759 input and an electrical output, for example, a thermistor,
strain gauge, and photodiode, whereas a so-called inverse
4 Examples of Functional Architecture 760
transducer or precision actuator has an electrical input and
5 Force and Torque Balance 761 a low-power nonelectrical output: for example, a piezoelec-
6 Block Diagram and Damping 761 tric crystal, translational, and angular moving-coil elements
7 Examples of Force-feedback Sensors 762 can be used as inverse transducers. The sensor, inverse
8 Other Feedback-measuring Systems 763 transducer, and usually the amplifier must be close to the
Related Articles 763 point of measurement, whereas the indicator may be some
distance away. The feedback loop must have sufficient
References 763
negative gain and the system must be stable. The system
is driving fairly low-power devices at its output, and the
inverse transducer essentially determines the characteristics
1 FEEDBACK-MEASURING SYSTEM of the system, although noise connected with the sensor
and amplifier input stage may well be important. In prac-
There are only a few structural schemes employed in the tice, the measurand may not be directly connected to the
construction of instruments and instrument systems, and null or balance point, and there may well be one or more
one such scheme is the use of a feedback. A feedback primary sensing elements inserted between points X and Y
system can be thought of as a system that tends to auto- as in Figure 1.
matically maintain a prescribed relationship of one system
variable to another by comparing functions of these vari-
ables and using the difference as a means of control. The 2 GENERAL PROPERTIES OF
main characteristic of a feedback system is its closed- FEEDBACK SYSTEMS
loop structure (see also Article 18, Nature and Scope of
Closed-loop Systems, Volume 1). A measuring system in A measuring system may have many components or ele-
which feedback is the basic structural arrangement is called ments and it is convenient to consider each such element
a feedback-measuring system, and a systematic analysis, as a block with its own input–output relationship that is its
description, and review of such systems has been under- own transfer function. The blocks representing the various
taken by Jones (1979). elements of a system are connected to use their functional
A general block diagram of such a system is shown relationship within the system, thus producing a block dia-
in Figure 1. Here, the output signal (usually electrical) is gram for the system.
converted to a form (usually nonelectrical, e.g. force) suit- A generalized block diagram of a simple feedback system
able for comparison with the quantity to be measured (e.g. is shown in Figure 2. The diagram illustrates negative
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Force-feedback Sensors 759
Feedforward The main properties of this general system can be
Balance
point Error network summarized as well-defined gain (giving accuracy), wide
point (amplifier)
bandwidth (giving fast response), low or high output
X Y+ N
r
_
c impedance (allowing remote indication), reduction in noise
Quantity to Indication of
be measured Transducer
effects and extraneous disturbances, and the property of
measured
quantity inversion. Negative feedback can be employed to produce
Inverse
Nonelectrical transducer Electrical instrument high input impedance, and unity-gain feed-
quantities quantities back often allows noncontact measurement. Measurement
devices providing a frequency output usually employ feed-
Figure 1. Diagram of a feedback-measuring system: - - - - , back, and feedback can be used to enhance the character-
electrical units or signals; , mechanical units or signals.
istics of frequency-selective networks.
(Reproduced from Feedback in Instruments and its Applications,
Jones, 1979, with permission of Institute of Physics Publishing
Ltd.)
3 STABILITY CONSIDERATIONS
n1 n2
It is important to investigate the output response of a
+ N + + system, particularly a feedback-measuring system, when the
r a A c
− + + input changes rapidly. This transient response characterizes
the stability of the system. A system is defined as stable
b
if its impulse response approaches zero as time approaches
infinity. A system is defined as being unstable, if, with zero
Figure 2. Generalized block diagram of a simple feedback sys-
tem. (Reproduced from Feedback in Instruments and its Appli- input, the output increases indefinitely. If the output of a
cations, Jones, 1979, with permission of Institute of Physics system has continuous oscillation of constant peak-to-peak
Publishing Ltd.) amplitude, the system is considered to be neutrally stable.
Consideration of the degree of stability of a system often
feedback, in that the gain around the loop has negative provides valuable information about its behavior. Clearly,
polarity. In this case, the measurand r and the output it is desirable that feedback-measuring systems are stable
indication c are in phase, but if in the practical system and respond satisfactorily to rapid changes in the quantity
they are in antiphase, then the amplification A will have to be measured.
a negative sign associated with it and both inputs to the It should be noted that in general for a feedback system,
comparator will have positive signs. designs for accuracy and stability tend to be incompati-
In Figure 2 a fraction α of the quantity to be measured ble. From (1) accuracy required |Aβ| to be high, while
(r) is fed in and a fraction β of the output is fed back; α, for second- and third-order systems, high |Aβ| produces
β and A are transfer functions, while n1 and n2 are noise less damped and unstable systems respectively. In practice,
sources at the input and output of the amplifier respectively. when it is necessary, the open-loop frequency characteris-
Assuming n1 = n2 = 0, the following equation expresses tics of a feedback system are tailored such that at quasi-
the situation of Figure 2: static frequencies, the gain is very high to give the required
accuracy at these frequencies, and at frequencies near to
c unity gain, the loop phase shift is less than −180◦ to give
αr − βc =
A the required damping characteristics. Feedback-measuring
or systems often use a combination of mechanical and elec-
c α Aβ trical means of damping. There are a number of useful
= (1) texts available on the design of systems with control loops,
r β 1 + Aβ
for example, by Jones (1979), and by DiStefano, Stub-
The ratio c/r is called the closed-loop gain, while Aβ is the berud and Williams (1990); see also Article 19, Dynamic
open-loop gain and both are transfer functions. Provided Behavior of Closed-loop Systems, Volume 1.
that Aβ 1, c/r ≈ α/β, and c/r is largely independent Systems incorporating feedback involve comparison of
of A and is simply determined by transfer functions α and two physically similar variables and production of a min-
β. Thus, the sensitivity of such a system can be changed imum or null (at point N in Figure 1). In feedback-
by altering α or β. In some practical feedback-measuring measuring systems, pervariable (or ‘through’ variable) bal-
systems, the feedback also makes α and β well defined. ance occurs with force, torque, current, and heat flow, while
Actually, α and β may alter, but so long as their ratio in the case of transvariables (or ‘across’ variables), voltage,
remains fixed, c/r is well defined. temperature, and displacement balances are common.
760 Elements: A – Sensors
Torque
+ Torque to
W Beam position G
converter
−
W
G
(a) Spring
Beam
Moving coil
+
element
Torque
−
Beam position
detector Torque to
W position
converter
Moving coil Position
element detector
and amplifier
Current Indication
meter
i
Amplifier
Current i
(b) Block diagrams
Real schemes
Figure 3. Measurement of small weights: (a) simple system and (b) more complex system. (Reproduced from Instrument Systems:
Functional Architectures from Concise Encyclopedia of Measurement and Instrumentation, Finkelstein and Grattan, 1994, with permission
of Elsevier.)
4 EXAMPLES OF FUNCTIONAL between G and W is very dependent on the linearity
ARCHITECTURE between G and the spring torque.
A torque balance that results in very little beam
movement and, therefore, greater accuracy is shown in
A simple system for measuring small weights is shown Figure 3(b). Here, the clockwise torque is produced by the
in Figure 3(a). The weights W are placed on a plate current i in a moving-coil element fixed about the pivot
fixed to the end of a pivoted beam such as to produce a point. Without weights the beam remains stationary and i is
torque in an anticlockwise direction. A spiral spring fixed zero. When weights are applied, the beam tends to move in
about the pivot produces a torque in a clockwise direction an anticlockwise direction, and this movement is detected
proportional to the beam angular deflection G. At some by sensors, for example, photodiodes, which produce an
point the torques will balance, and G is an indication of electrical signal; this is amplified to give current i, which, in
W . The block diagram for this system shows that there turn, produces the counter torque to almost maintain beam
is inherent feedback, in that the mechanical indication G balance. As there is little angular movement of the moving-
produces a torque via the spring constant (feedback path), coil element, the current-torque relationship is linear and
which subtracts from the torque produced by the weights the electrical output i is proportional to W . The output
on the beam (feedin path) to give an error torque. Linearity meter can be some distance from the point of measurement.
Force-feedback Sensors 761
Because there is little mechanical movement using this the time constant (s is the Laplace complex variable). The
method of measurement, it can be used to measure more mechanical moving element has a second-order transfer
rapid changes in W than the fully mechanical scheme of function, where m is the mass (or moment of inertia), k is
Figure 3(a). the spring stiffness, and d is the viscous damping (inherent
or artificial). Kd is the sensitivity of the displacement sensor
and Ka is the gain of an amplifier, while the term 1 + Ta s
5 FORCE AND TORQUE BALANCE is the transfer function of a (ideal) phase-advance network,
providing rate compensation and feedback damping.
Force and torque-balance systems are very common and In fact this network is likely to be essential to stabilize
are employed, for example, in precision weighing, for the the system, and this is illustrated by the typical Bode magni-
measurement of acceleration, pressure, flow, level, elec- tude plots in Figure 5(b). Curve A is the system open-loop
trical power, and high voltage. These quantities can be gain measured in the region of unity open-loop gain, with
converted to a force or torque; for example, acceleration Ta = 0 and Ka = 1. The curve has a slope of slightly more
of a fixed seismic mass produces a force on the mass pro- than −12 dB/octave at the unity-gain frequency of 10 Hz,
portional to acceleration, and a dynamometer movement and the system is unstable in the closed loop. A suitable
produces torque on a shaft proportional to electrical power phase-advance characteristic is shown as curve P (in prac-
in the movement. The relationship between the quantity to tice it is limited as shown by the dotted line), and with this
be measured and force may be nonlinear. phase-advance in the loop, the system open-loop gain is
The main methods of force and torque balance are shown given by curve B. It is clear that the slope is now less than
in Figure 4, where F is the force to be measured and Ff is −12 dB/octave at unity gain and the closed-loop system
the balancing or feedback force generated in the particular is stable. A resultant closed-loop step response is shown
instrument. For the beam and shaft, F and Ff are converted in Figure 5(c), indicating that the damping is still fairly
to torques T and Tf respectively. In each case, the element small. The steady state gain of the instrument or static char-
being used (seismic mass, diaphragm, beam or shaft) is acteristic c/r = K1 /Kf , provided that the open-loop static
displaced by a small amount in the direction in which F acts gain Kd Ka Kf /k 1; the accuracy of such instruments is
before the equilibrium balance F = Ff is achieved. This usually better than 1% of full scale. This static gain is
displacement is the information required by the instrument
chosen to give just sufficient accuracy, because as its value
to generate Ff .
is increased stabilizing the instrument becomes more dif-
ficult. Most feedback-measuring instruments incorporate a
stability network of some form. Having demonstrated the
6 BLOCK DIAGRAM AND DAMPING
need and use of such networks, we will not show them
The block diagram of a basic force (or torque) balance in subsequent diagrams. The detailed stability analysis of
instrument is given in Figure 5(a). The input creates a force automatic weight balances has been given by Smirnova
(or torque) via, say, a constant K1 , while the output creates
the balancing force (or torque) via a precision actuator; F (or T )
if this is of a moving-coil type, it will have the transfer Output
Input + 1
function of a single-order system as shown, where T1 is r K1 Kd Ka(1 + Tas) c
− ms2 + ds + k
Diaphragm Ff(or Tf)
K f /1 + T1s
F Ff (a)
F
(b) F
40 P
Magnitude (dB)
Ff B
M F Shaft 20 A
01
Output c
10 100 ω(Hz)
(a) Ff (c) (d) Ff −20
Beam
(b) −40 (c) Time
Figure 4. Force balance with: (a) a seismic mass M, (b) a
diaphragm; torque balance with (c) a beam, and (d) a shaft. Figure 5. Basic force (or torque) balance instrument: (a) block
Broken line, position before application of F , when F = Ff = 0. diagram, Kf /1 + T1 s (b) open-loop Bode magnitude plots, and
(Reproduced from Feedback in Instruments and its Applications, (c) step response (time scale 20 ms per division). (Reproduced
Jones, 1979, with permission of Institute of Physics Publishing from Feedback in Instruments and its Applications, Jones, 1979,
Ltd.) with permission of Institute of Physics Publishing Ltd.)
762 Elements: A – Sensors
(1960), and Gal’perin and Kolesov (1972). An example of distance. An input movement on a force spring, due to
an automatic chemical balance has been given by Smith a measurand, creates a torque on a beam with a flexure
and Stevens (1959). pivot. Any movement of the beam is detected by the posi-
tion detector and a current is created in the moving-coil
actuator (or force motor) to produce the balancing torque.
7 EXAMPLES OF FORCE-FEEDBACK The output DC is linearly related to the input movement.
SENSORS This two-wire transmitter has a ‘live zero’ current out-
put, and its sensitivity can be altered by changing the
Various force- and torque-balance instruments are shown spring stiffness.
schematically in Figure 6, and illustrate the methods of A pneumatic pressure transmitter is shown in Figure 6(b).
balance, some typical devices used, and the wide applica- The pressure unit is completely sealed and P2 could be
tion of this form of balance. The arrangement of Figure 6(a) the pressure at the bottom of a petrol tank, enabling the
can be used to transmit force information over a long level of petrol in the tank to be monitored. The output of
Force Circuit
motor block Power
Flexure Line
supply
pivot (up to 30 km)
Force Recorder
spring Position detector
Beam Dashpot Indicator
Transmitter
(a) Input
Flapper-nozzle
position detector
Output Moving
pressure thread
Beam LED
Tf
Pneu- Phototransistor
matic
amplifier
L Bellows Tf T
Diaphragm Flexure
q A
pivot
Restriction Shaft
T P2 > P1 Tension plate
DC motor Output
Air
P1 P2 supply
(b) Pressures (c)
Acceleration
A
F
Capacitance
transducer
Seismic mass
Ff Moving-coil
actuator
S
N
(d) Housing Output
Figure 6. Various force- and torque-balance instruments: (a) two-wire transmitter, (b) pneumatic pressure transmitter, Flexure pivot,
T1 ⇒ Tf (c) thread-tension monitor, and (d) seismic-mass accelerometer. (Reproduced from Feedback in Instruments and its Applications,
Jones, 1979, with permission of Institute of Physics Publishing Ltd.)
Force-feedback Sensors 763
this transmitter is air pressure, and the sensitivity of the Besides force and torque balance, other physically similar
transmitter is inversely proportional to length L. Various variables can be balanced in sensory systems: electrical cur-
forms of feedback transmitter are used in industry. rent (and magnetic flux), heat flow, voltage, pressure, tem-
The monitor of Figure 6(c) employed in the textile indus- perature, displacement, phase, and radiation, are described
try uses the torque of a DC motor to balance torque by Jones (1979).
created by tension in the thread; an optical transducer Measuring devices where the output is in frequency form
is used to determine the angular position of the motor usually employ feedback to maintain oscillation. Feedback
shaft and the monitor is described by Auckland and Hawke is used in instruments for control purposes, and the oper-
(1978). ation of the feedback is similar in both the control and
In the accelerometer of Figure 6(d), the inertial force measuring applications; the design for accuracy and stabil-
derived from the acceleration is balanced by the feedback ity is basically the same.
force created by current in the moving-coil actuator; this
has been described by Macdonald (1961).
It should be noted that the moving-coil actuators (both RELATED ARTICLES
translational and angular) make use of the electromagnetic
force created when a current-carrying conductor is at right Article 18, Nature and Scope of Closed-loop Systems,
angles to a magnetic field. The force is at right angles to Volume 1; Article 19, Dynamic Behavior of Closed-loop
both the conductor and the field, and is proportional to both Systems, Volume 1; Article 109, Structure and Energy
the current if and field flux density; as a consequence, force- in Sensor Systems, Volume 2.
balance instruments employing such actuators usually have
linear characteristics with a current as the output. The ratio
Tf /if may vary from about 0.1 to 50 mN mA−1 , while the REFERENCES
ratio Ff /if may vary from about 0.1 to 20 N A−1 .
Auckland, D.W. and Hawke, J.R. (1978) A Thread-Tension
Transducer Using Torque-Balance About the Axis of a Motor,
8 OTHER FEEDBACK-MEASURING Proceedings of Transducer ’78 Conference, Trident Interna-
tional Exhibitions, Tavistock (pp. 1–4).
SYSTEMS
DiStefano, J.J., Stubberud, A.R. and Williams, I.J. (1990) Feed-
back and Control Systems, Schaum’s Outline Series, 2nd edn,
Feedback is widely used in instruments, and may well be McGraw-Hill, New York.
the main structural element of a measuring system, linking
Gal’perin, L.N. and Kolesov, Yu.R. (1972) Theory of Automatic
the output indication with the measurand. Using feedback, it Balances with an Electrical Compensator. Measurement Tech-
is possible to improve accuracy and speed of measurement, niques, 15, 546–549.
reduce the effect of interfering and modifying inputs, and Jones, B.E. (1979) Feedback in Instruments and its Applications.
allow remote indication and noncontact measurement. The Journal of Physics E: Scientific Instruments, 12, 145–158.
property of inversion assists instrument design, and can Macdonald, W.R. (1961) in Flight Test Instrumentation (ed.
assist in provision of digital indication. In general, the M.A. Perry), Pergamon Press, Oxford (pp. 15–23).
main disadvantages are increased complexity, size and Smirnova, N.A. (1960) The Theory of Automatic Beam Scales
cost, but the advent of small cheap integrated circuits has with Electrical Balancing. Measurement Techniques, 10,
reduced these. The development of transducers and accurate 844–850.
precision actuators seems certain to extend the development Smith, B.O. and Stevens, J.W. (1959) Quick-Acting Chemical
and use of feedback-measuring systems. Balance Journal of Scientific Instruments, 36, 206–209.
114: Models of the Sensor Interface
Qing Ping Yang
Brunel University, Uxbridge, UK
According to the level of abstraction, models may be used
1 Types of Models and their Characteristics 767 to represent power flow, signal flow, or information flow.
2 Information and Knowledge Flow 769 Models may be functional or object oriented, for exam-
ple, computer-based models implemented in functional
3 Energy and Mass Flow Interaction 769
or object-oriented programming languages. According to
4 Semiotic Issues 770 whether the model is based upon the physical attributes
5 Set Theoretical Expression of Interface and structures or the observed data, functional models may
Characteristics 770 be further classified as data models or physical models.
6 Allowance for Influence Effects 771 This section will mainly compare functional and object-
7 Practical Sensor Interface Modeling 771 oriented models. Other types of models are discussed
References 772 elsewhere – Article 13, Signals, Information and Knowl-
edge, and Meaning, Volume 1; Article 71, Modeling
Methodology, Volume 2; Article 104, Models of the Mea-
surement Process, Volume 2; and Article 105, Modeling
1 TYPES OF MODELS AND THEIR with LabVIEW , Volume 2.
CHARACTERISTICS
1.1 Types of models 1.2 Model characteristics
1.2.1 Functional models
There are several different types of models for a measure-
ment system interface. The simplest model (Figure 1) is Traditionally, measurement systems are usually modeled
based upon the traditional concept of measurement, which with a functional representation. This kind of model typ-
is defined as the comparison of the measurand with some ically consists of three basic stages, that is, input stage
fixed standard unit. including a sensor or transducer, signal processing stage and
According to the representation scheme, there are phys- output stage, each of them represented as an input–output
ical, linguistic, pictorial, and mathematical models. Mathe- functional block, as shown in Figure 2.
matical models are by far the most important because of the Functional representations are well established in mea-
nature of measurement. In general, a measuring instrument surement and other branches of science. This kind of model
may be described as is often used as the basis for classification of measuring
instruments. They are very useful for determining system
Z = f (X, Y, W ) (1) static and dynamic characteristics by means of the transfer
characteristic and transfer function of each element – see
where Z – output; X – input measurand(s); Y – interfer- Article 57, Static Considerations of General Instrumen-
ence or influence quantities; and W – latent parameters. tation, Volume 1.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
768 Elements: A – Sensors
Measurand
the real world also provides a unifying platform for all
Output
Measurement
the related activities, for example, analysis, design, and
Reference implementation.
standard Essential characteristics of the object-oriented approach
Figure 1. The simplest comparison model. include information hiding or encapsulation, inheritance
and piecewise refinement. Objects encapsulate both their
data structures and operations and they thus work at a higher
Sensor/ Signal Data level of abstraction. This makes object-oriented models
transducer processing presentation more stable than conventional functional models because
Input Output
changes in operations or functions are decoupled and are
Figure 2. Functional model. localized within objects.
Inheritance comes from the class hierarchy structure, in
Common to the use of modeling in various problem which a class can have its own data structure and opera-
domains, the intrinsic limitations of a functional model tions, and can also inherit them from its parent class(es).
arise from its low level of abstraction. Its reusability is Although the concept originated in artificial intelligence,
very poor as it is intended for specific systems use. Despite its inherent existence in class hierarchy structure is unique
wide use of structural or modular design concepts, it is to object-oriented approach. The reusability of similar
rather inflexible and often difficult to modify. The size of objects is one of the most important benefits of object-
this kind of model, especially a computer-based one, often oriented methods.
increases exponentially with complexity. The feature of piecewise refinement results from object-
As has originated mainly from control engineering, func- orientation and, in particular, its separated external and
tional models of measurement systems have often failed internal views of an object. Together with encapsulation, it
to address some issues that are of theoretical and prac- provides a powerful mechanism to achieve high productiv-
tical importance to measurement, for example, operating ity, good maintainability, system integrity, and reliability.
space of measurement discussed in Stein (1992, 1996). As Yang and Butler (1998) have demonstrated the applica-
a result, incomplete descriptions of measuring systems are tion of object-oriented modeling for a generic measurement
widely used, and the distinctiveness of measurement sci- system, based upon a generalized measurement concept. A
ence has been seriously obscured. This is probably one of new framework was proposed in Yang and Butler (1997).
the important reasons measurement science has attracted Depicted in Figure 3 is the architecture of the object-
little theoretical interest when compared with control engi- oriented model for measurement systems. It captures impor-
neering. tant relations in a measurement system and embraces five
objects or subsystems, that is, measured object, measur-
1.2.2 Object-oriented models ing instrument, human observer, referencing standard, and
operating environment. The model represents a measure-
The object-oriented approach views a system as a collection
ment system at three system levels: internal (measuring
of discrete objects that contain both data structure and
instrument), operational, and environmental.
behavior. The data structures and behavior or operations
of an object are defined by a class. An object is an instance
of a class. Objects can interact with other objects.
The development of object-oriented methods has been
closely associated with computer programming. Use of Measured
object
object orientation first appeared in the development of a dis-
crete event simulation language (Simula) in Norway, Bjørn
(1989). As they began to mature in the late 1980s, object-
Measuring Human
oriented methods have spread to other areas, for example, operator
instrument
system analysis and design. Indeed, they have become
a general modeling approach in recent years, Rumbaugh
(1991). Referencing Operating
The power of the object-oriented approach lies largely in standard environment
its closeness to the natural view of the real world. By raising
the level of abstraction from the function to the object level,
it focuses on the real-world aspects of a system, providing
a better model of the problem space. Its consistence with Figure 3. The architecture of an object-oriented model.
Models of the Sensor Interface 769
2 INFORMATION AND KNOWLEDGE instrument and, indeed, a physical system can be modeled
FLOW on the basis of the energy and power interactions.
One methodology first proposed by Paynter (1961) has
A measuring instrument or system may usually be concep- been further developed by many researchers. The model
tually viewed as a measuring chain, which is defined as is typically presented in a graphic form called a bond
a series of elements constituting the path of measurement graph that represents a physical system in terms of power
signal from the input to the output. Sensor interface is the bonds connecting the system elements to so-called junction
key element of a measuring chain. structures that model different types of energy constraints.
The measuring chain can be modeled at different levels The system elements are generally classified into sev-
of abstraction. At the top level is information and knowl- eral standard idealized types, namely, storage elements
edge flow, as the purpose of any measurement is to acquire (C-element and I-element), resistor, sources, transformer,
information and knowledge about the measured system or and gyrator. There are two types of junction structures,
process. The measuring chain can be viewed as the infor- namely, 0- and 1-junctions corresponding to the two gener-
mation path, handling sensing, processing, transmission, alized Kirchhoff’s laws. They are generalized elements and
storage, and display of information. Although this chain is domain-independent, and can describe different systems
represented by different terms and blocks, they are invari- (e.g., mechanical, electrical, hydraulic, acoustical, thermo-
ably of this form. dynamic, material) in the same way, using analogies in
The information-flow model of a sensor interface repre- both, their governing mathematical equations and physi-
sents only the relevant information operation performed by cal concepts.
the sensor interface. It typically takes the form of flow or The bonds, normally labeled and directed, are the energy
block diagram. connections between the two power ports of two interacting
As an information carrier, signal flow always accompa- elements. Each port represents the physical interaction with
nies information flow in a measuring chain. Signal flow is at other elements or the system environment. The energy flow
the intermediate level of abstraction and is commonly used along each bond or port has a physical dimension of power
in the analysis, design, and evaluation of measuring instru- that can be described with a pair of terminals or variables,
ments. Signal-flow models represent a measuring chain or flowing in opposite directions.
process as transformed signals in various signal domains, Such a combination can be found in various physical
for example, time domain, frequency domain, and time- domains, for example, voltage and current in electrical
frequency domain. domain, force and velocity in mechanical domain. One of
Since a signal results from the change or modulation of them may be classified as effort variable (e.g., voltage,
the energy state, useful signals and unwanted ones (noises) force), the other as flow variable (e.g., current, velocity).
always exist together because of energy interactions. It is A bond graph has many properties of object-oriented
thus necessary to apply some form of signal conditioning modeling. It can also be easily converted to signal-flow
and signal processing techniques to reduce the effect of graph and block diagram. An example bond graph for an
noise and to increase the signal-to-noise ratio. Further pro- RLC circuit is shown in Figure 4.
cessing including compensation and correction is also often A power-flow model can be used to completely explain
required. The information is finally decoded or determined the behaviors and characteristics of a physical system
with reference to calibration and measurement scale. such as transducers because it encompasses all the signifi-
Computer tools, both hardware and software, for mod- cant energy flows related to the system. System equations
eling and processing signals are highly developed. There governing variables at various ports or elements can be
are many packages commercially available for handling systematically derived from a bond graph. The procedure
the entire process of signal flows, for example, MATLAB
(http://www.mathworks.com/products/matlab/), and Lab- R:R
VIEW (http://amp.ni.com/niwc/labview/lv.jsp) – see
Article 105, Modeling with LabVIEW , Volume 2. L
VR i
R
Vs VL
Se:vs 1 I:L
i i
3 ENERGY AND MASS FLOW Vs C VC i
INTERACTION C:C
(a) (b)
At the lowest level of abstraction of the measuring chain is
energy or power flow. The system behavior of a measuring Figure 4. (a) RLC circuit and (b) corresponding bond graph.
770 Elements: A – Sensors
may be algorithmized and performed by computer soft-
ware, for example, ENPORT , ARCHER , Camp-G , and q1 n1
so on.
The modeling of instruments based upon this method- q3
ology is given in Finkelstein and Watts (1982). A similar n3
approach was presented in Stein (1992, 1996) as a three- q2
n2
port six-terminal transducer model.
Q N
Figure 6. Mapping from observed space to number space.
4 SEMIOTIC ISSUES
Semiotics is the study of signs and symbols in various
fields, especially language. Its theory explains how the q1 n1
meaning arises during the perception and interpretation of
signs or sensory data. Since measurement can be regarded q3
n3
as a symbolic measurement process, there are similar semi-
q2
otic issues concerned with measurement. n2
Different models can be used to explain the relation-
ship between signs and their meaning. The most popular Q N
one is the triadic model or semiotic triangle developed
by Charles Sanders Peirce (1839–1914), as shown in
Figure 5. R3 P1
R1 P3
The semiotic triangle based upon the triadic model con-
sists of three elements:
R2 P2
1. the sign or symbol presented to an observer;
2. the meaning or the concept formed by the observer as R P
the result of the perception of the sign; and
3. the object or referent the sign is referring to. Figure 7. Mapping of individual observation value and the set
of empirical relations.
The meaning of a sign is not only dependent upon the
sign itself but also influenced by the observer’s experience Indeed, the relationship between the human perception
with the referent. The semiotic model is obviously appli- and machine perception as shown in the previous block
cable to an empirical measurement process in which the (Figures 6 and 7) is a triadic model, and the ‘measurement
measurand is the ‘object’ or ‘referent’, the measured signal triangle’ in Figure 10 (see Article 13, Signals, Informa-
is the ‘sign’, and the information about the measurand or tion and Knowledge, and Meaning, Volume 1) is also a
the measured object can be regarded as the ‘meaning’ of semiotic triangle.
the signal. Measurement is primarily concerned with how measure-
ment symbols (signals or data) are related to the observed
Meaning
object(s), whereas semiotics centers on how the symbols
(or concept) are interpreted. They are represented as two sides of the
triangle, but they share the common third side, that is, the
direct experience of the human observer.
5 SET THEORETICAL EXPRESSION OF
INTERFACE CHARACTERISTICS
Sign Object
(or symbol) (or referent)
The nature of a transducer and indeed a measurement
Figure 5. Semiotic triangle. process may also be described using the representational
Models of the Sensor Interface 771
theory of measurement based upon set theory. Measure- Measured Measuring Human
ment may be viewed as a mapping process from an object instrument observer
observed space (Q) to a number space (N), as shown in Coupling Interface
Figure 6.
The possible observation values of the measurand can be
represented as a set, for which a set of relations exist. The
measurand can thus be regarded as an empirical relational
Reference 1+ Operating
system, including the set of observation values and the set standard environment
of relations. Calibration Interference
According to the representation theory of measure-
ment, as discussed in Pfanzagl (1968), Krantz et al. (1971),
Narens (1985), and Finkelstein (1982), the nature of mea-
surement is to map the above empirical relational system
to a numerical relational system in which there are a set of Figure 8. Object-oriented model of measurement systems.
numbers and a set of relations defined on the set of num-
bers. There are two mappings involved, one from the set of developed an object-oriented model for a generic measure-
individual observational values (Q) onto the set of numbers ment system. For example, the interactions and influence
(N), and the other from the set of empirical relations (R) effects in a measurement system may be modeled as shown
onto the set of numerical relations (P). Together, these sets in Figure 8.
and mapping define a scale of measurement, Finkelstein It should be mentioned that the bond-graph model is
(1982). The two mapping processes are also shown in essentially object oriented, and as such it can also con-
Figure 7. veniently model the various influence effects.
Given a scale of measurement in addition to the mea-
surement result as numbers, the information or knowledge 7 PRACTICAL SENSOR INTERFACE
about the measurands and their relations can be deter-
mined. MODELING
The above discussions are based upon numerical rep-
A number of practical techniques are available for modeling
resentation, which can be easily extended to a symbolic
a sensor interface. They broadly fall into two categories.
representation.
The first one is to model the system functions based
upon observed or experimental data. It represents the func-
tional relationship between the inputs and outputs of the
6 ALLOWANCE FOR INFLUENCE sensor interface without explicit correspondence between
EFFECTS the model parameters and physical attributes.
Traditional techniques of doing the above include regres-
As the sensor interface interacts with other objects includ- sion models, time series models, and system identifica-
ing a measured object, reference object, environment, and tion. More recently, artificial neural networks are also
even possibly a human observer, the energy connections used for modeling sensor interfaces. It has been proved
prevailing will undoubtedly influence the behavior and that a feed-forward artificial neural network with one
characteristics of the sensor interface. Most models, based hidden layer can approximate arbitrary continuous map-
upon traditional functional representation, represent well ping, Cybenko (1989) and Funahashi (1989). An example
the internal functional aspects of a sensor interface, but of three-layered feed-forward artificial neural network is
are weak in their modeling of the operational and, to some shown in Figure 9.
extent, environmental interactions. These traditional techniques and artificial neural net-
In the modeling of measurement systems, functional works are normally applied using computer software pack-
models tend to be oversimplified, neglecting some impor- ages, for example, MATLAB and its relevant toolboxes.
tant aspects of a measurement system, for example, the The second category of modeling techniques is based
coupling interactions between the instrument and the mea- upon physical modeling, which relates model parameters
sured object, although some improvement has been made to physical attributes and structures.
by Finkelstein (1994) and Sydenham et al. (1994). Simple models may be developed manually, but comput-
A better approach is object-oriented modeling as it can ers have to be employed for more complicated modeling.
easily accommodate these external influences. In addition The models can be developed using general program-
to the conceptual framework, Yang and Butler (1998) have ming languages or special software packages. LabVIEW
772 Elements: A – Sensors
Conference held in Krakow, Poland, June, 2002. Key papers
are ‘in production’ with the Measurement Jnl at the time
Output layer of writing.
REFERENCES
Hidden layer
Bjørn, K. (1989) Object-Oriented Programming with Simula,
Addison-Wesley, Wokingham.
Cybenko, G. (1989) Approximation by Superposition of a Sig-
moidal Function. Mathematical Control Signal Systems, 2,
303–314.
Input layer Finkelstein, L. (1982) Theory and Philosophy of Measurement,
in Handbook of Measurement Science, Volume 1 Theoretical
Fundamentals (ed. P.H. Sydenham), John Wiley & Sons,
Figure 9. A three-layered feed-forward neural network. Chichester (pp. 1–30).
Finkelstein, L. (1994) Measurement and Instrumentation Sci-
ence – An Analytical Review. Measurement, 14, 3–14.
from National Instruments and HP-VEE (Hewlett-Packard
Finkelstein, L. and Watts, R.D. (1982) Mathematical Models of
Visual Engineering Environment) are good examples for Instruments – Fundamental Principles, in Instrument Science
instrument modeling – see Article 105, Modeling with and Technology, Vol. 1 (ed. B.E. Jones), Adam Hilger Ltd,
LabVIEW , Volume 2. They are essentially a graphical Bristol, CT (pp. 9–27).
programming language specially developed for instrumen- Funahashi, K. (1989) On the Approximation Realization of Con-
tation and are intended for virtual instrumentation applica- tinuous Mappings by Neural Networks. Neural Networks, 2(3),
tions. 183–192.
Another well-established technique is finite element Krantz, D.R., Luce, R.D., Suppes, P. and Tversky, A. (1971)
method (FEM) as a mathematical simulation of the actual Foundations of Measurement, Academic Press, New York.
physical structure. The model is divided into a number Narens, L. (1985) Abstract Measurement Theory, MIT Press,
of regions known as elements, each of them defined by Cambridge, MA.
a pattern of nodes. The method defines an approximation Paynter, H.M. (1961) Analysis and Design of Engineering Sys-
within each element, with appropriate continuity conditions tems, MIT, Cambridge, MA.
imposed on the interelement boundaries. FEM can Pfanzagl, J. (1968) Theory of Measurement, Physica-Verlag,
be applied to linear and nonlinear structural, thermal, Würzburg, Vienna.
dynamic, electromagnetic, and flow analysis. A number of Rumbaugh, J. (1991) Object-Oriented Modelling and Design,
software packages are commercially available, for example, Prentice Hall, Englewood Cliffs, NJ.
ANSYS , EMFlex , and so on. These make the task Stein, P.K. (1992) The Unified Approach to the Engineering
relatively easy, for the mathematics is hidden, allowing of Measurement Systems-Basic Concepts, Stein Engineering
Services Inc., Phoenix, AZ.
practitioners to apply the practical inputs needed.
As described in Section 3 here, bond graphs have been Stein, P.K. (1996) The Unified Approach to the Engineering
of Measurement Systems for Test & Evaluation – A Brief
successfully applied to build power-flow models. A bond-
Survey, in Proceedings IEEE Instrumentation and Measurement
graph model can be both power and information oriented. Technical Conference, Brussels (pp. 1–28).
There are several software packages for general bond-graph
Sydenham, P.H. et al. (1994) Unsolved Problems of Measure-
modeling, for example, ENPORT , ARCHER , Camp-G , ment – An International Study, in Proceedings XIII IMEKO
BondLab , SYMBOLS , Modelica , and so on. One par- World Congress, Turino (pp. 827–831).
ticular package, developed specifically at City University, Yang, Q. and Butler, C. (1997) On the Framework of Measure-
London for transducer modeling, was MEDIEM (Multi- ment Science, in Proceedings XIV IMEKO World Congress,
Energy Domain Interactive Element Modeling). Vol. 5, Tampere (pp. 101–106).
Further information on the difficulties and shortcomings Yang, Q. and Butler, C. (1998) Object-Oriented Model of Mea-
of modeling of the measurement systems interface is found surement Systems. IEEE Transactions on Instrumentation and
in the keynote and in other papers of the IMEKO TC-7 Measurement, 47(1), 104–107.
115: Designing the Sensor Interface
Qing Ping Yang
Brunel University, Uxbridge, UK
force, load, torque, and so on;
1 Deciding the Input and Output Forms of pressure, sound, flow, and so on;
Energy 773 Thermal : temperature, heat, and so on;
2 Transformation Representation Methodology 774 Radiant: intensity, phase, frequency (wavelength), po-
3 Specifying the System Parameters 775 larization, power, and so on;
Magnetic: field intensity, and so on;
4 Progressive Development Using the
Chemical : concentration, crystal structure, aggregation
Transducer Conversion Logic 776
condition, and so on.
5 Use of Tools to Support Sensor Interface
Development 777
A transducer has energy connections with some other
References 777
elements and the environment. Each connection interface is
usually called power port. Intrinsically (without considering
environmental interactions), a transducer has two or three
1 DECIDING THE INPUT AND OUTPUT power ports, that is, input, output, and modulation (or
FORMS OF ENERGY sometimes called interrogation), although modulation is not
always necessary. Two-port and three-port transducers are
The information in a transducer is carried and conveyed by shown in Figures 1 and 2.
signals, which in turn reside in certain power or energy A two-port transducer is also called self-generating or
form. Signals can be regarded as the factors of energy a passive transducer, and a three-port transducer is called
or power. modulating or active transducer (Sydenham, Hancock and
Signals may be classified into various domains accord- Thorn, 1989; Middlehoek and Noorlag, 1983).
ing to their energy forms, that is, electrical, mechanical, For example, a piezoelectric transducer is a two-port
thermal, radiant, magnetic, and chemical. The above six transducer and a strain gauge is a three-port transducer.
signal domains, as proposed by Middlehoek and Noorlag Since each power port of a transducer may be in one of
(1981), are most commonly used, although it is possible to the six energy domains, a two-port transducer has 62 = 36
add additional domains. Some example signals in these six combinations, each as a transducing or conversion mech-
domains are given below. anism, for example, a piezoelectric transducer has direct
conversion from mechanical to electrical.
Electrical : resistance, capacitance, inductance, current, Similarly, a three-port transducer has 63 = 216 possible
voltage, and so on; conversion mechanisms. For example, a strain gauge con-
Mechanical : motion and dimensional: displacement, posi- nected to a bridge circuit has a conversion from mechanical
tion, strain; length, height, width, thickness, to electrical, which is modulated by electrical power. With
area, volume; velocity; acceleration, vibration, both two-port and three-port transducers combined, there
and so on; are 252 possible conversion mechanisms.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
774 Elements: A – Sensors
fi fo Transducer
element
ei Pi Two-port Po eo
transducer Electrical
Mechanical
Thermal
Figure 1. Two-port (self-generating or passive) transducer.
Radiant
Magnetic
fi fo
Chemical
ei Pi Three-port Po eo
transducer Figure 3. Matrix representation of a transducer. (Reprinted from
Sensors and Actuators, Middlehoek, S. and Noorlag, D.J.W.,
fm Pm Vol 2, Three-Dimensional Representation of Input and Putput
Transducers, page 27, Copyright (1981), with permission from
em Elsevier.)
Figure 2. Three-port (modulating or active) transducer.
(1981) and Sydenham, Hancock and Thorn (1989). A line
may be drawn to connect the input and output domains,
2 TRANSFORMATION representing the transformation from one energy to another.
REPRESENTATION METHODOLOGY All the elements of a measuring chain can be shown
as such columns and connected together as a matrix. The
The input, output energy, and, if necessary, the modulation conversion lines may be joined together in the transfor-
energy, together define a conversion mechanism or trans- mation matrix of a measuring chain, (Middlehoek and
duction principle of a transducer. Noorlag, 1981).
For example, assume that a pressure transducer has a
mechanical diaphragm to convert the measured pressure to
2.1 3D coordinates and matrix representation displacement, which is measured by an intensity-modulated
transducer. The light intensity is first converted to electrical
A useful way of classifying transducers and physical effects current by a photodiode and then to a voltage with an ampli-
is to represent the conversion mechanism in 3D space, with fier. The corresponding matrix representation is shown in
the energy form of input, output, and modulation as the Figure 4.
three axes (e.g. x, y, and z axes, respectively), and each It is quite straightforward to use this kind of matrix
conversion principle or physical effect can thus be shown representation to describe conversion processes and energy
as a ‘sensor effect cube’ (i.e. a cube region in 3D coordinate flows. However, it does not include the modulation input
system), as presented in Middlehoek and Noorlag (1981). directly, and the representations of energy flow and signal
Consequently, each element of a measuring chain may be flow are not integrated.
expressed as 3D coordinates, for example, [x, y, z], where Both the coordinates and matrix approaches proposed are
x, y, and z are one of the six energy forms, (Middlehoek mainly graphic presentations, and, as such, it is neither
and Noorlag, 1981). easy nor efficient to handle these representations using
On the basis of the 3D classification, Middlehoek and computers. See also Article 109, Structure and Energy
Noorlag (1981) have further proposed a methodology for in Sensor Systems, Volume 2.
describing a measuring instrument by joining the 3D coor-
dinates of each stage, with additional lines drawn to show Optical
Diaphragm diplacement Photodiode Amplifier
the energy source/flow as well as the signal flow. transducer
Although it is a novel and interesting idea, one has to
Electrical
constantly check to see which element and which coordi-
nate the lines are from, and to which element and which Mechanical
coordinate they go to. With all the coordinates quite simi- Thermal
lar or the same because only the energy form is given, it is
Radiant
quite difficult to interpret a conversion process.
A simpler approach shows each transducer as a trans- Magnetic
formation column (Figure 3), with the left and right ver- Chemical
tical edges corresponding to the input and output energy
form respectively, as presented in Middlehoek and Noorlag Figure 4. A pressure transducer using matrix representation.
Designing the Sensor Interface 775
2.2 Transducer conversion logic representation complicated process in which the modulation itself is con-
verted from another source, that is,
A better representation is presented here, which combines
energy and signal flows. It is compact, highly efficient, and x|z → y (5)
easy to use. It also has a natural link to signal flow diagram
and block diagram. This approach may be generally referred where z = x1 → z1 or z = x1 |y1 → z1 .
to as transducer conversion logic or language. The combined conversion can be expressed as
According to this approach, a two-port (or self-
generating) and three-port (or modulating) transducer or x|(z ← x1 ) → y (6)
element can be represented as
or
x→y (1) x|(z ← x1 |z1 ) → y (7)
and Take the above strain gauge connected to a bridge circuit
x|z → y (2) for example; assume the voltage supply is converted from
an optical light source, which in turn is controlled by an
respectively, where x, y, and z are input, output, and electric current source, then the process can be expanded as
modulation of a transducer respectively. They take a form
Ds, where s is the signal concerned, D is one of six energy Me.ε → E.R|(E.Sv ← R.I |E.Si ) → E.v (8)
forms, which may be identified with the initials, that is,
where R.I is the light intensity in radiant (R) domain, E.Sv
E = Electrical is the voltage source, and E.Si is the current source.
Me = Mechanical
T = Thermal 3 SPECIFYING THE SYSTEM
R = Radiant PARAMETERS
Ma = Magnetic The best way to specify the system parameters for a
C = Chemical measurement system is to follow a systematic approach,
given by Bosman (1978) and the framework proposed by
For example, a piezoelectric transducer has a direct Yang and Butler (1997, 1998). The latter is based upon
conversion process systems approach and object-oriented modeling.
Figure 5 shows the general model architecture. It cap-
Me.F → E.q (3) tures several important aspects of a measurement system.
It encompasses five subsystems: measuring instrument,
which converts mechanical force to electrical charge. A
strain gauge connected to a bridge circuit can be repre-
sented as
Me.ε → E.R|E.Sv → E.v (4) Measured
object
where the strain is converted to electrical resistance,
which is then converted to a voltage output with voltage
Measuring Human
modulation. operator
instrument
Of course, longer names can also be used for signals, if
necessary. But as the signal domain is specified, it is quite
easy to understand standard symbols for variables or signals Referencing Operating
that are commonly used. In fact, the same symbol, but in standard environment
different domains, can be easily resolved. This makes the
representation more powerful.
The above conversion expression for a three-port or
modulating transducer can be expanded to model a more Figure 5. Object-oriented model for system specification.
776 Elements: A – Sensors
Table 1. Quality matrix for system specification.
Internal Operational Environmental
Features Size Data format Power supply
Weight Coupling method Operating environment
Principle Calibration method
Material Human interface
Capability Measurand(s) Networking interface Temperature compensation
No. of channels
Intelligence
Performance Resolution Flexibility (object dependence) Cross sensitivity
Measuring range User friendliness EMC
Dynamic range Calibration frequency
Repeatability
Uncertainty
Frequency response
Drift
measured object (measurand), referencing standard, human converted to electrical current by a photodiode and finally
operator, and operating environment. The measurement converted to a voltage with a transimpedance amplifier. The
results can be affected by all the five subsystems, with measuring chain may be first represented as
the instrument as the most important one. The system
may be considered at three different system levels, that Me.p → Me.ε → R.I → E.i → E.v (9)
is, internal, operational, and environmental. At each level,
important attributes, capability, and performance can be where p is the measured pressure; ε is the diaphragm
specified. The specification may be presented as a qual- deformation; I is the output light intensity from the optical
ity matrix as shown in Table 1, which was first used by fiber sensor; i is the current output from the photodiode;
Yang (1995). and v is the amplifier voltage output.
More details about the light intensity modulation can be
added as follows:
4 PROGRESSIVE DEVELOPMENT USING
THE TRANSDUCER CONVERSION Me.ε|R.SI → R.I (10)
LOGIC where SI is the light source, and
The transducer conversion logic or language presented in Me.ε|(R.SI ← E.Si ) → R.I (11)
Section 2 has the key features of object-orientation, such
as hierarchy and inheritance, piecewise refinement, and where Si is the current source.
information hiding. Thus, it is essentially an object-oriented The latter expression shows that the mechanical defor-
approach to description and modeling of a measuring chain mation is converted to light intensity, with the light
and conversion processes. source powered by an electric current source. Note that
Since the transducer conversion logic has a very simple more details can be simply added without affecting the
syntax and consists of only texts, it can be easily handled structure.
and processed by computers. For many transducers, typical Assume the photo detector is voltage biased, then we
conversion mechanisms can be used as default or options, have
which makes it much easier to automate analysis and design
of transducers or instruments. R.I |(E.v = const) → E.i (12)
In addition, the representation can be easily converted to
a signal flow or block diagram. The use of the transducer The expanded chain can thus be expressed as
conversion logic can be further demonstrated, taking the
above pressure transducer, for example. Me.p → Me.ε|(R.SI ← E.Si ) → R.I |E.Sv → E.i → E.v
The measured pressure is applied to a mechanical dia- (13)
phragm, whose deformation is sensed by an intensity- In fact, this expression may be easily converted to a block
modulated optical fiber sensor. The light intensity is then diagram as shown in Figure 6.
Designing the Sensor Interface 777
Optical package developed specifically for transducer modeling is
p e i
Diaphragm displacement I Photodiode Amplifier v the MEDIEM (multi-energy domain interactive element
transducer modeling, Sydenham, Hancock and Thorn, 1989).
Once the transducer models are created, simulation and
Light source Voltage source
optimization can be further performed. The criteria for
optimization need to be clearly identified and defined,
Figure 6. Block diagram of the pressure transducer. and optimization generally involves analysis of a num-
ber of different design configurations and/or parameters.
The process can be quite computing intensive and time
5 USE OF TOOLS TO SUPPORT SENSOR consuming.
INTERFACE DEVELOPMENT Alternatively, it may be easier to follow the design of
experiment methodology (DOE) to experimentally model
The complete process of sensor interface development start- and optimize the design. One popular methodology is the
ing from requirement and specification encompasses several robust design developed by Taguchi (1986), consisting of
important stages, namely, design (including modeling, sim- system design, parameter design, and tolerance design.
ulation, optimization), construction, test, and evaluation. There are also software tools available for the use of design
Various tools and methods, especially computer-aided tools, of experiment. The design of an experiment may be used
are needed during the development process. together with the above theoretical modeling.
Computer-aided design (CAD) is widely used in engi-
neering design, and there are many commercial packages,
for example, AutoCAD , Pro/ENGINEER , SolidWorks . REFERENCES
There are also software packages for both design and man-
Bosman, D. (1978) Systematic Design of Instrumentation
ufacturing, that is, CAD/CAM (computer aided manufac- Systems. Journal of Physics E: Scientific Instruments, 11,
turing) tools. In addition, CAD packages are also available 97–105.
for electronic or optical design. Middlehoek, S. and Noorlag, D.J.W. (1981) Three-Dimensional
Modeling and simulation have become increasingly im- Representation of Input and Output Transducers. Sensors and
portant in product development. As discussed in the pre- Actuators, 2, 29–41.
vious article, mathematical modeling may be carried out Middlehoek, S. and Noorlag, D.J.W. (1983) Silicon Micro-Trans-
using some software packages, for example, MATLAB , ducers, in Instrument Science and Technology, Vol. 2 (ed.
LabView , and so on. These models may be implemented B.E. Jones), Adam Hilger Ltd, Bristol, CT (pp. 33–44).
as traditional procedural algorithms or as the more recent Sydenham, P.H., Hancock, N.H. and Thorn, R. (1989) Introduc-
object-oriented programs – see also Article 105, Modeling tion to Measurement Science and Engineering, John Wiley &
with LabVIEW , Volume 2 and Article 71, Modeling Sons, Chichester.
Methodology, Volume 2. Taguchi, G. (1986) Introduction to Quality Engineering – Design-
The finite element method (FEM) is a very powerful ing Quality into Products & Processes, Asian Productivity
Organization, Tokyo.
tool and can be applied to linear and nonlinear structural,
thermal, dynamic, electromagnetic, and flow analysis. A Yang, Q. (1995) Basics and Theory of Measurement, in Man-
ufacturing Measurement, Part 1 (ed C. Butler and Q. Yang),
number of software packages are commercially available, Chapter 2, Brunel University, Uxbridge (pp. 1–16).
for example, ANSYS , EMFlex , and so on.
Yang, Q. and Butler, C. (1997) On the Framework of Measure-
The modeling can also be based upon the use ment Science, in Proceedings XIV IMEKO World Congress,
of bond graphs. A number of software packages Vol. 5, Tampere (pp. 101–106).
have been developed for this kind of modeling, Yang, Q. and Butler, C. (1998) Object-Oriented Model of Mea-
namely, ENPORT , ARCHER , Camp-G , BondLab , surement Systems. IEEE Transactions on Instrumentation and
SYMBOLS , Modelica , and so on. One special Measurement, 47(1), 104–107.
116: Selection of Sensors
Paul P.L. Regtien
University of Twente, Enschede, The Netherlands
know-how contained in the minds of people and acquired
1 The Design Process 778 through experience.
2 The Requirements 778
3 Selecting the Measurement Principle 779 2 THE REQUIREMENTS
4 Selecting the Sensing Method 779
5 Sensor Selection 779 Sensor selection means meeting requirements. Unfortu-
nately, these requirements are often not known precisely
References 780
or in detail, in particular when the designer and the user
are different persons. The first task of the designer is to get
as much information as possible about the future applica-
tions of the measurement instrument, all possible conditions
1 THE DESIGN PROCESS
of operation, the environmental factors, and the specifi-
cations with respect to quality, physical dimensions, and
The design of a technical system involves making choices costs.
on the basis of criteria (from a list of requirements), The list of demands should be exhaustive. Even when
availability of parts and materials, financial resources, and not all items are relevant, they must be indicated as
time. These aspects play a significant role when designing such. This will leave more room to the designer, and
a measurement system. will minimize the risk of being forced to start all over
Blanchard and Fabrycky (1998) distinguish six major again at a later date. Rework is an expensive process and
phases of the design process: (a) conceptual design; should be avoided where possible by reducing errors as
(b) preliminary design; (c) detail design and development; early as possible in the systems engineering life cycle
(d) production/construction; (e) operational use/mainten- process. The requirements list should be made in such
ance; (f) retirement – see also (Sydenham, 2004). Thus, a way as to enable unambiguous comparison with the
sensor selection is a crucial activity in the systems design final specifications of the designed instrument. Once the
process, as it will make a great impact on the production designer has a complete idea about the future use of
of the measurement instrument and the performance during the instrument, the phase of the conceptual design can
its entire lifetime and may even have consequences related start.
to disposal. Before thinking about sensors, the measurement princi-
Design methods have evolved over time, from purely ple has to be considered first. For the instrumentation of
intuitive (as in art) to formal (managerial). The process each measurement principle, the designer has a multitude
of sensor selection is somewhere in between: it is an of sensing methods at his or her disposal. For realization
act of engineering, in which the design is supported by of a particular sensor method, the designer has to choose
advanced tools for simulating system behavior based on the optimal sensor type out of a vast collection of sensors
scientific knowledge. The basic attitude is (still) the use of offered by numerous sensor manufacturers.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Selection of Sensors 779
3 SELECTING THE MEASUREMENT For instance, where the fluid must remain in the tank
PRINCIPLE during measurement, principles based on volume or mass
flow are excluded. If the tank contains a chemically aggres-
The design process is illustrated by using an example of sive fluid, a noncontact measurement principle is preferred,
a measurement for a single, static quantity: the amount of placing principles B, C, and D lower on the list, and so on.
fluid in a container. The first question to be answered in Also, method A can possibly be eliminated because of
this case is in what units the amount should be expressed – its high costs for large containers.
volume or mass. It is important to have a sound under- In this way, the conceptual design ends up with a set of
standing of the physics involved and the circumstances of principles having related pros and cons, ranked according
the situation. These may influence the final selection of to the prospects of success.
the sensor.
Figure 1 shows the various measurement principles that
could be used in this case: 4 SELECTING THE SENSING METHOD
A: the tank placed on a balance, to measure its total
weight; After having specified a list of candidate principles, the
B: a pressure gauge on the bottom of the tank; next step is to find a suitable sensing method for each of
C: a gauging-rule from top to bottom with electronic them. In the example in Figure 1, we will further investigate
readout; principle E, a level detector placed at the top of the tank.
D: level detector on the bottom, measuring the col- Again, a list of the various possible sensor methods is made.
umn height; This may be
E: level detector from the top of the tank, measuring the
1. a float, connected to an electronic readout system;
height of the empty part.
2. an optical time-of-flight measurement;
Obviously, many more principles can be used to measure 3. an optical range measurement;
a quantity that is related to the amount of fluid in the tank. 4. an electromagnetic distance measurement (radar);
In the conceptive phase of the design, as many principles 5. an acoustic time-of-flight (ToF) measurement; and
as possible should be considered, even unusual or unortho- so on.
dox ones. On the basis of the list of demands and not as
As in the conceptual phase, these methods are eval-
a ‘hunch’, it should be possible to find a suitable candi-
uated using the list of demands, so not only the char-
date principle from this list, or at least delete many of the
acteristics of the sensing method but also the proper-
principles, on the basis of arguments.
ties of the measurement object (liquid level) and the
environment should be taken into account. For the tank
system, the acoustic ToF method could have an excel-
lent chance because of being contact-free; or just the
contrary, for instance, because of possible high temper-
C
E atures. In this phase, it is also important to consider
methods to reduce such environmental factors (see Arti-
cle 16, Reduction of Influence Factors, Volume 1). Any-
how, this phase ends up with a list of candidate sensing
methods with merits and demerits with respect to the
requirements.
B 5 SENSOR SELECTION
D
The final step is the selection of the components that
make up the sensing system. Here, a decision has to be
made between a commercially available system and the
A
development of a dedicated system. The major criteria are
Figure 1. Measurement principles for the amount of fluid in costs and time: both are often underestimated when a new
a reservoir. development is considered.
780 Elements: A – Sensors
In this phase of the selection process, sensor specifica- More detail is given on the systems engineering
tions become important. Sensor providers publish speci- process in Article 63, Outline of Systems Thinking,
fications in data sheets or on the Internet. However, the Volume 2; Article 64, Executing A Measuring System
accessibility of such data is still poor, making this phase Design, Volume 2; Article 66, Phases of System Life
of the selection process critical and time consuming, in Cycle, Volume 2; Article 67, Principle of Concept of
particular for nonspecialists in the sensor field. Operations (ConOps), Volume 2; Article 68, Setting the
Computer-aided sensor selection programs are under System Boundaries, Volume 2; Article 69, Requirements
development and are partly realized, but up to now their Allocation, Volume 2; Article 65, Life Cycle Concept,
use has been limited. A prerequisite for a general, suc- Volume 2; and Article 70, Measuring System Design
cessful tool facilitating sensor selection is the continuous Methodologies, Volume 2.
availability of sensor data on the Internet, in a more
standard format, and regularly updated. A good start is
www.globalspec.com. REFERENCES
Obviously, the example of the level sensor is greatly
simplified here. Usually, the selection process is not that Blanchard, B.S. and Fabrycky, W.J. (1998) Systems Engineering
and Analysis, Prentice Hall, Upper Saddle River, NJ.
straightforward. Since the sensor is often just one ele-
ment in the design of a complex technical system, close Sydenham, P.H. (2004) Systems Approach to Engineering Design,
Artech House.
and frequent interaction with other design disciplines is
necessary.
117: Materials in Measuring Systems
Peter H. Sydenham
GSEC Pty Ltd, Adelaide, South Australia, Australia
In their passive role, materials provide the capability to
1 Role of Materials in Sensor Construction 781 support, connect, control, contain, or display the sensing
phenomenon provided by the active components of a con-
2 Metals 782
structed object.
3 Ceramics 782
All passive materials are active to some degree in that
4 Plastics 784 they respond to one or more forms of external energy. They
5 Influence Effect on Materials 785 are not, however, regarded as transducer materials unless
6 Selecting a Material 786 the active transduction action is deliberately enhanced and
7 A Rationale for Selecting Passive Materials 787 controlled. For example, normal passive structural steel has
8 Altering the Active and Passive Behavior 787 mechanical hysteresis that varies far more than it does in the
9 An Example in Material Selection 788 active instrumentation alloys processed to make precision
springs and load cells.
10 Sources of Knowledge on Materials 790
Another characteristic often exhibited in a sensor mate-
References 790 rial is reversibility of the transducer action. For example,
Further Reading 790 piezoelectric material will produce electric charge when its
shape is deformed; conversely, when an electric charge
is supplied to it, the shape changes. Not all materials
exhibit this.
1 ROLE OF MATERIALS IN SENSOR Sensor materials ideally provide a linear, or at least repro-
CONSTRUCTION ducible, energy transformation. Efficiency of the energy
transformation is usually less important than its all-impor-
A report of the Fellowship of Engineering, RFE (1983), tant fidelity.
classified materials as having two main roles. The concern of technology for producing sensor materials
In their active role, they are ‘hosts of potentially useful is to enhance the efficiency of the information transfor-
scientific phenomena, and always bring about a conversion mation action. A requirement is also that the transduction
of energy (or information) from one form to another’. They action remains constant regardless of time and the in-service
can provide this conversion using a gas (such as in a filled level of influencing variables and operational signals.
thermometer), fluid (liquid crystals in a display), or a solid Some sensors rely on loss of some of their mass to
(piezo-electric material in an accelerometer). produce signals. Examples include many chemical sensors
In the sensors of measuring systems, energy conversions where intimately related mass/energy transformations take
are concerned with the provision of information about the place; and special, electrically resistive, tape used to mon-
physically existing world; modulation of an energy link itor the abrasive level of magnetic recording tapes, whose
carries the required information from the system being resistance rises as the tape is worn away. In these cases,
studied to the observer. the design aim is to prolong the useful sensing life, usually
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
782 Elements: A – Sensors
by increasing the sensing sensitivity, thereby requiring less on species protection, and lack of uniformity of properties
mass consumption per unit time. of the materials.
By the traditional design methodology, a sensor is created
from a selected mixture of different active and passive
materials, their integration chosen to enhance the sensing 2 METALS
effect while attenuating undesirable internal and external
influence effects. This group comprises the common, relatively inexpensive,
In contrast, the modern integrated microelectronic form base metals and alloys such as steel, brass, aluminum,
of sensor makes use of a common base material that is mod- and copper. Also used are special metal alloys such as
ified to enhance the sensing parameters while adequately Invar , Ni-Span C , Nichrome , and others sold under
controlling the nonsensing ones. In this microelectronic proprietary names.
sensor form, highly pure silicon is doped, with a range Literature on the common metals is widely available,
of materials, at minute levels of impurity to make it sen- especially on those used for structural purposes. Informa-
sitive to one of a wide range of measurands in certain tion on properties of the specialist metals, however, must
positions in its bulk. By use of masks and other process- usually be obtained directly from the alloy makers.
ing techniques – see Article 162, Principles of MEMS, The common metals have found time-honored acceptance
Volume 3; Article 163, Uses and Benefits of MEMS, as passive engineering constructional materials due to their
Volume 3; Article 164, Principles of MEMS Actuators, ease of working, wide temperature operating range, general
Volume 3 – the doping and etching can be made to pro- availability, well-known history of performance, and range
duce both active and passive areas in the two-dimensional of useful structural properties.
and three-dimensional geometry of a single chip of the host The engineering use of metals is covered extensively
material. Here, the technology is required to also control the in terms of analysis of their upper working load limits.
passive properties. Their use in measuring systems, however, is often more
concerned with allowable deflection and stiffness under a
Another class of integrated sensor is the fiberoptic sensor,
working load.
wherein a basic element – the optical fiber – is modified by
Table 1 is a representative list of the metals used in
a range of methods to form many different sensors.
instrument construction. Typical applications are given to
Many transduction principles in use today were discov-
indicate, not prescribe, which to use.
ered as physical and chemical effects in the nineteenth
century (Sydenham, 1979). The twentieth century has been
a period of enhancement of the passive and active uses of 3 CERAMICS
materials; their use in sensing systems requires extensive
interdisciplinary understanding. The ceramic group of materials (glasses and semiconduct-
Three main groupings of material types that find appli- ing materials are included here but are not strictly defined
cation in measuring systems are as such) provide for such functions as thermal insulation or
conduction, high-temperature mountings, and can provide
• metals very useful transduction processes for use as both sensors
• ceramics, including semiconductors here for conve- and actuators. While the use of ceramics is less common in
nience instruments, they can be vital because they augment metals
• plastics. and plastics by providing properties the latter two cannot
(Ichinose, 1987). Table 2 lists representative ceramics used
Depending on composition, processing, or application in instrument construction along with some typical appli-
arrangement, these variously form either active or pas- cations.
sive components. Silicon semiconductor material is the key to the advance
Optical materials are covered in Article 210, Optical of many miniature sensors and actuators, see Article 162,
Transducers for Chemical Measurements, Volume 3; Principles of MEMS, Volume 3; Article 163, Uses and
Article 89, Optical Materials, Volume 2; and Article 91, Benefits of MEMS, Volume 3; and Article 164, Princi-
Light Sources and Detectors, Volume 2. ples of MEMS Actuators, Volume 3.
Brief mention is also needed of the many natural More information is provided on the ultrasonic use
materials that were once commonly used, such as wood, of ceramics in Article 118, Ultrasonic Sensors, Vol-
leather, bone, ivory, mother of pearl, tortoiseshell, and the ume 2; Article 119, Ultrasonic Instrumentation Princi-
like. In general, these are rarely used now because of their ples, Volume 2; and Article 120, Ultrasonic Instrumen-
relatively high cost, problems of supply, legal restrictions tation Design, Volume 2.
Materials in Measuring Systems 783
Table 1. Metals commonly used in instrument construction.
Metal Application
Commonly available passive metals (often alloyed)
Aluminum Panels, mountings, leads, fittings
Brasses (copper–zinc alloys) Diaphragms, mounts, contacts, springs
Bright steel (often hardened) Frame, panels, shafts, gears, machined parts, pins, clips,
springs, cams, knife-edges, pulleys, chains, flexure strips
Bronzes Cast frames, cast mechanisms, bearings, gears, cams
Copper Leads, printed circuit connections, connectors contacts, heat
sinks, heat conductors, shielding
Iron Castings, temperature compensation
Nickel Corrosion resistant needs, temperature match with glasses
Platinum, Silver, and Gold Contacts, electrodes, corrosion resistant layers
Solders (silver, tin–lead) Connections, contact materials
Stainless steel Springs, support frames, panels, shafts, flexure strips
Special passive alloy groups
Beryllium–copper and phosphor–bronze (copper–tin) Springs, sealing diaphragms
Copper–nickel, such as Monel Corrosion resistance
Copper–nickel–chromium, such as Mu-Metal Magnetic shielding
Nickel–iron, such as invar, and Ni Span-C Temperature stable parts
Minimal mechanical hysteresis
Nickel–chromium–iron, such as Hastellow C Diaphragms, springs
Nickel–chromium, such as Nichrome Electrical heaters
Silicon–iron Transformer laminations
Active metals for sensing applications
Bimetal strips Temperature actuators and detectors
Electrical resistance alloys Stable resistors, strain gauge foils
Load cell materials Diaphragms, load cell structures
Magnetic alloys Thermomagnetic effects, magnetic detectors
Nickel alloys Magnetostriction
Thermally sensitive electrical resistance alloys and thermocouple alloys Temperature sensors
Table 2. Some ceramic materials used in instrument construction A useful introductory text on fine ceramic materials is
(From Ichinose, 1987). Ichinose (1987) from which much of this explanation is
Ceramic type Application extracted. More detailed data must be sought from the
Passive constructional roles
makers because ceramics, like plastics, are materials with
Alumina • Heat resistance support very complex formulae; numerous controlled production
Beryllia • Heat conducting support factors decide the final performance.
Fused silicas • Low TEMPCO parts Some instrument ceramics are made with reasonably
Graphite • Heat resistant support, colloidal
straightforward, specialized processes that convert natural
lubricant
Hydroxyl apatite • Artificial bones and tooth root materials such as sand and clays into the final form. Exam-
replacement ples here include the manufacture of fused silica (needing
Manganese oxide • Insulators, IC substrates electric arc refining of special sands) and insulating porce-
Silicon carbide • Cutting tools, low-wear surfaces lains (liquid pug is cast in molds, followed by drying, and
Zirconia • High-temperature mechanisms.
then high-temperature firing).
Active sensing roles
Alumina • Sodium batteries The advanced materials, so-called hi-tech forms, are
Barium titanate (a PZT) • Piezoelectric sensing and much more complex to manufacture. These fine ceramics
actuation require to be manufactured using highly refined raw materi-
Ferrites • Recording heads, RF transformer als, rigorously controlled composition, and strictly regulated
cores
forming and sintering.
Silicon dioxide • Optical fibers
Stable zirconia • Oxygen detection Production of ceramic powder, the starting material of
Tin oxide • Gas sensor many fine ceramics, is made with several processes, each
Zinc oxide–beryllium oxide • Voltage-dependent resistor. being complex and needing expensive plant investment.
Source: Reproduced from Introduction to Fine Ceramics: Applications in The powder is formed into the required shape by one of
Engineering, Ichinoise (1987) John Wiley & Sons Limted. five main sintering processes such as
784 Elements: A – Sensors
• die casting chemical resistant or biodegradable, and materials can be
• rubber mold pressing blended to achieve any desired property, or combination
• extrusion molding of properties. The maximum temperature of operation of
• slip casting selected plastics is constantly being raised, being now well
• injection molding over the boiling point of water.
However, success in their use is very much affected by
The item is then sintered to bring about the minimum the part’s design and processing. The designer’s knowledge
system energy through densification of the powder. These of all of these variables can profoundly affect the ultimate
processes require high temperatures (1400–1800 ◦ C) to be success or failure of a consumer or industrial product.
present while the pressing action is in place. Final mechan- Plastics fall into two major groupings
ical properties are decided by some twenty major parame-
ters, each needing strict control of several parameters. • thermoplastics, which can be melted and reused;
To illustrate the complex productions involved, consider • thermosets, which are chemically changed by the man-
making silicon ‘chip’ sensors. This is a long, many-stage ufacturing process and cannot be remelted
process that requires a plant that is so expensive that many
countries cannot afford to have such facilities. The process Typically, thermoplastics are shaped by melting, forcing
starts by pulling a pure silicon crystal, followed by zone the (often highly viscous) liquid into the desired shape,
refining it by sweeping it with heat. This is vital to obtain and allowing it to freeze. Injection molding (for bulk
the highest purity material. material) and thermoforming (sheet material) are processes
The quite large crystal is then sliced and polished to form of this type.
thin wafers of pure silicon. These are subjected to several Thermosets are compressed into a mold and heated,
similar cycles of masking, doping, and developing to form the heating process changing the chemical structure and
the miniature electronic and mechanical components within forming the molding shape. However, these simplistic
the silicon. Vacuum deposition is then used to add the metal divisions are blurred, and injection-moldable thermosets are
connections. available. The heating required to set off the thermoset is
The silicon die is then tested and mounted on a support also often aided by catalysts (as in dough molding) and the
package, electrical leads being bonded from the die to the chemical change can be brought about by catalysis alone
connection pins. with no external heat as in composite lay-up (for example,
Finally the product is tested, packaged, and labeled. fiberglass) construction.
Most silicon microelectronic manufacturing plants are Crystalline thermoplastics are generally stronger, more
built to make electronic circuits and, as such, are not chemically resistant, and often self-lubricating. Examples
always capable of providing the processes needed for cer- are polyethylene, nylon, polypropylene, and acetal. Amor-
tain sensor manufacturing needs – micromachining needed phous thermoplastics such as polystyrene vinyl (PVC),
in MEMs forms, for example, requires special addi- acrylic, and polycarbonate are more easily formed, deco-
tional processes. rated, and fastened and can be transparent.
Once a ceramic or semiconductor item is made, it cannot Many plastic materials can be formed as foams with
be easily remanufactured, so these components must be closed or open cell structure. Polystyrene foam is famil-
ready formed and processed for specific applications. The iar as a rigid, yet light and strong, packaging material.
high costs of manufacturing plant must be supported by Polyurethane foams can be rigid though soft for applications
large-volume production; the designer of small-runs of like cushioning material in instrument carry cases. Materi-
sensors has to make use of stock proprietary forms. als can also be made to foam during molding, forming a
solid external skin with a foam core. This provides struc-
tural benefits (high strength to weight) and material savings
4 PLASTICS as well as molding benefits from a reduced cycle time.
Individual plastic materials are supplied with a wide
Plastics have steadily displaced metals in many instrument range of properties affected by the molecular weight of the
application areas, such as in frames, in cases, and in bulk material, additives for modification of production per-
many precision mechanisms. Caren (1988) summarizes formance and final product properties, and coloring agents.
their place well. Suppliers have ranges of product to suit different processes
Plastics permit a greater amount of structural design free- and provide specialty blends for specific requirements.
dom than any other material. Plastic parts can be large or An important subset of thermoplastics is the elastomer
small, simple or complex, rigid or flexible, solid or hollow, group. These are plastics with rubber-like elastic properties.
tough or brittle, transparent, opaque or virtually any color, SBR (styrene butadiene rubber) is used for protective boots,
Materials in Measuring Systems 785
Table 3. Plastics used in instrument construction. (Compiled from industry catalogs – a guide only – manufacturers must be consulted
in making a detailed choice).
Common plastic name Application
Passive constructional roles
ABS • Telephone handsets, high-quality cabinets, PCB plugs, transparent covers – is platable
Acetals • Gears, pawls, links, cams, cranks – is platable
Acrylics • Optical lenses and covers
Cellulosics • Blister packaging, poor durability
Epoxys • Circuit boards, hard surfaces
Fluoroplastics (includes PTFE) • Electrical insulation, bearings, valves, linings, implants
Nylons • Gears, snap-on parts, hinges, guide rolls, cams, plug connectors, brush holders, print
wheels, hot-water valve bodies, switches, coil formers press buttons, bearings
Polycarbonates • Tough, transparent structures, lenses, safety enclosures, cases
Phonelics • Low-cost molded parts
Polyimids • High temperature, low-creep, instrument parts
Polyesters • Resin for glass, carbon and aramid fiber-based composites, containers, films
Polyethylenes • Nozzles, containers, cable clamps, cases, cable insulation
Polystyrenes • Lamp shades, diffusers, instrument frames, panels, video cassette parts, foam thermal
insulation and packaging
Polypropylenes • Appliance housings, cable sleeves, sterilizable parts
Polyurethane • Cushion soft to glass hard needs, fascias, tubing, cords
Poly (vinyl chloride) • Cable ducts, pipes insulation on cables switch covers, medical parts
Silicones • Flexible seals, protective covers, molded seals Active sensing roles
Polyvinylidene fluoride (PVDF) • Piezo electric
Polyacetylene (and others) • Electrical conduction
shock-absorbing mounts, and grommets. Polyurethane (a Fillers are mainly inorganic materials such as talc, kaoli-
thermoset) can be made in hardness ranging from glass hard nite, feldspar, and glass microspheres. They can improve
to soft and rubbery. It is prepared by mixing two liquids processability, reduce shrinkage, increase stiffness, and
and finds applications in cast-in-place elastic elements. reduce cost.
Silicone rubbers provide rubbery consistency with excellent Reinforcers are fibrous materials including various glass
chemical resistance, although if used in contact with fibers, aramid, carbon, and more exotic materials. Their
sensitive items (such as foil strain gauges), a grade should major use is increasing strength and stiffness, often
be chosen which does not emit acetic acid on setting. anisotropically, providing finished products, which, on a
Some combinations of polymers can also be mixed strength-to-weight basis, are often comparable with high-
together to form copolymers (where the chains of the differ- strength steels.
ent materials cross-link) or alloys in which the ingredients All additives can affect surface finish and subsequent
remain discrete. ABS is a typical terpolymer (a copolymer decorative processes, those added for physical property
of three ingredients) while PC/PVC is an alloy. improvement frequently adversely affecting the finish.
While the range of plastic materials available presents a The selection of plastics is a highly skilled task. Plas-
large spectrum of opportunity for the designer, the possi- tics manufacturers, however, provide extensive data and
design services including electronic forms of adviser,
bilities added by additives, fillers, and reinforcers increase
Endura (2004) and Rapra (2004) being examples of guides.
this enormously. These can modify finished part strength,
Kroschwitz (1990), Rubin (1988) provide greater and more
impact resistance, density, and color. The performance of
generic detail.
the material in production can also be affected. Materials
Table 3 is provided as a first guide to the type of
can be added to plastics in bulk, during molding or by hand
plastics commonly used for the construction of various
lay-up.
instrument parts.
In general, the measuring instrument and sensor designer
need only be aware of the possibilities arising from the use
of additives, leaving detailed recipes to the plastics supplier 5 INFLUENCE EFFECT ON MATERIALS
and manufacturer. Such additives as biocides, fungicides,
heat and light stabilizers, and antistatic agents can be quite As explained elsewhere – see Article 16, Reduction of
vital to the success of a product, especially those that relate Influence Factors, Volume 1 – all materials respond, to
to dimensional stability and to life. some degree, to a range of external influence effects.
786 Elements: A – Sensors
Table 4. Physical effects give rise to unexpected activity in materials.
Term describing effect of Input variable on Output variable
(most are reversible)
Gas law effect (at constant volume) Pressure Temperature
Electrochemical effect Electric current Material decomposition
Electromagnetic effect Electric current Force
Electrostatic effect Electric voltage Force
Faraday effect Magnetic field Polarization angle
Hall effect Magnetic field Current/voltage
Magnetoresistive effect Magnetic field Electrical resistance
Magnetostrictive effect Magnetic field Length
Photoelastic effect Strain Optical polarization
Piezoelectric effect Force Charge
Piezoresistive effect Force Electrical resistance
Poisson effect Longitudinal load Transverse length
Pyroelectric effect Heat Charge
Rectifying effect Electric current Flows only one way
Resistance–effect Strain Electrical resistance
Self-heating effect Electrical power Temperature
Shear modulus effect Shear load Length
Temperature velocity effect Temperature Velocity of sound
Thermoelastic effect Temperature Elasticity
Thermoelectric effect Temperature Voltage
Thermoresistive effect Temperature Electrical resistance
Young’s modulus effect Tensile load Length
These can change shapes of passive material support inhibited by their manufacturing difficulties existing for
structures and sensitivities of sensing active materials and small volume production.
as such can considerably impact sensor and instrument It might be thought that as materials have been develop-
performance. ing since the time of early man, the designer always has
Table 4 is provided to assist in identifying likely effects. at hand materials with the right combination of properties
It lists often-met transduction effects that may give to suit any task. Unfortunately, this is far from the case
unwanted activity in a material normally regarded as pas- because the number of well-defined and controlled material
sive or that may alter the active performance. Due attention properties required is great and each material application
to such, often unexpected, factors is needed to reach a satis- will usually involve the need for several particularly strin-
factory design outcome. Many of the effects are reversible. gent properties at the same time. This will be seen in the
Generally, only one or two would be expected to be signif- illustrative case study given in Section 9.
icant in a chosen material but as the system sensitivity is Measuring instruments are made in production volumes
increased more effects become significant. ranging from one-off to many thousands but rarely in
the numbers seen in consumer electronic goods. Plastic
materials, and the production processes required to convert
6 SELECTING A MATERIAL
them into products, are generally more suited to high-
Metal and ceramic materials tend to be used in instruments volume production. For example, injection molding is
to provide design solutions where plastics are unsuited, for one of the most widely used processes but the simplest
instance, where very high temperature and high mechanical molding dies can cost around $10 000. It is an elegant and
stability are needed. versatile process but production runs of less than several
Another factor that dominates the choice of material is thousand inevitably carry a high tool amortization cost
the volume of the production run. For small volume runs, per part.
tooling costs associated with the often preferred use of Die-cast metals, once very popular, also can be used
plastics are usually very high compared with the cost of but these are falling from favor except where plastics
computer-controlled machining of metal parts. Ceramics, cannot cope because of lack of a key property such as the
usually requiring very elaborate manufacturing processes need to have metal shielding. Even there, however, special
to obtain the forms needed, are often used in stock electrically resistive, sprayed on and plated coatings can be
shapes with minimal machining. Their use is somewhat used to shield the contents of plastic cases.
Materials in Measuring Systems 787
7 A RATIONALE FOR SELECTING are formed to allow the needed freedoms and constrain the
PASSIVE MATERIALS others. An example is a ball bearing where the rollers and
tracks are made to exhibit as little deformation as possible
The first step toward material selection for the passive part while allowing one-dimensional slide action.
of a sensor is to set up the system architecture that will The alternative is to purposefully provide certain parts
satisfy the operational requirement for the measurement with elasticity so that movements can occur at intended
system. This means deciding the functions needed and places. This is known as elastic mechanical design,
how they are to be provided as the various needed see Article 85, Elastic Regime of Design – Design
organic modules. Principles, Volume 2; Article 86, Elastic Regime of
The module connecting the observed process to the Design – Spring Systems, Volume 2; and Article 87,
subsequent signal processing and the output actuating Elastic Regime of Design – Plates and Bimorphs,
device both need an active sensing material, whereas the Volume 2. An example is a flexure strip hinge used to allow
mountings and following signal processing assembly are limited rotation.
made from sufficiently passive materials. In general, kinematic design has the widest applicability
These building blocks, in turn, are made from a combina- but when relatively small movements are needed, its options
tion of parts. This reticulation process eventually identifies often cannot compete with the cost-effectiveness and small
the basic components needed, calling for readymade or new excursion fidelity of the elastic method.
parts to be made from appropriate materials. Once the overall constraint of freedoms strategy is
Crucial in this design process is the simultaneous iden- decided, a suitable physical framework is evolved to
tification of which salient physical effects, the influence support the various parts needed to provide the functions.
parameters, need to be allowed for. A collection of basic parts set up to provide a particular
The passive role of materials in instruments can be function is termed a mechanism. This, in turn, is formed of
found by considering the generalized mechanical restraining members, linkages, joiners, bearings, and parts that provide
purpose of the passive support providing components. transfer motion. The classic overview of the detailed design
Reuleaux, in 1875, defined a machine as a combination of such parts for fine instruments is given in Trylinski
of resistant forces so arranged that by their means the (1971), with a shorter account given in Trylinski (1983).
mechanical forces of nature can be compelled to do work Parts and roles identified by this top-down design process
accompanied by certain determined motions. As instrument then allow the choice of material. As can be seen from the
and sensor design is highly interdisciplinary, there are abridged list of material properties given in Table 4, the
usually many such resistant forces and determined motions choice of a suitable material can be complex and often
to be considered. impossible to totally satisfy.
In mechanical design, the main system requirement aims
to allow movements, in a three-dimensional space, of as
many of the six possible degrees of freedom as are needed. 8 ALTERING THE ACTIVE AND PASSIVE
These freedoms, for an extended object in space, can BEHAVIOR
occur as three translations and three rotations. A support
framework has to hold all six stable so that parts mounted An often-met difficulty is that the material that must be
on it are constrained within the required dimensional limits. used is not passive enough to some external influences.
A slideway allows one degree of translation; a pivot one Conversely, for sensing use, it may be desirable to increase
degree of rotation and so forth. the activity level. Several methods can be used.
Two approaches can be used to allow the neces- First, unwanted material effects can often be reduced by
sary freedoms in a mechanical system. The first is placing the active part, or the whole unit, in a suitably
to make use of materials, selected in shape and type, controlled environment.
to form a structure that is effectively nonelastic. This Second, some form of compensation can be incorporated.
method is known as kinematic mechanical design, see For example, to render a mechanical length formed from
Article 80, Principles of Fine Mechanics – Kinematic a metal bar less variant with temperature, the bar can
and Elastic Designs, Volume 2; Article 81, Princi- be mounted end to end with another bar of material that
ples of Fine Mechanics – Systems Considerations, expands in the opposite direction. By use of materials with
Volume 2; Article 82, Kinematical Regime – Members different TEMPCOs and different lengths, it is possible to
and Linkages, Volume 2; Article 83, Kinematical set up a system that has closely stable length. In practice,
Regime – Fasteners, Bearings, Volume 2; and Article 84, this concept usually needs to apply an opposing parallel
Kinematical Regime – Rotary Motion, Volume 2. These action because few materials have negative TEMPCOs.
788 Elements: A – Sensors
Another compensation method is to measure the influenc- These methods add the effects. A mechanical, hydraulic,
ing variable and from knowledge of the systematic nature pneumatic, or optical lever system can be used to increase
of the activity, change a value in the system at some appro- a small deflection by simple multiplication.
priate place.
The principle of these two compensation methods can be
applied to systems in any energy regime. Often, transducers
9 AN EXAMPLE IN MATERIAL
are used to convert energy forms in order to make the
comparison and compensatory actuation. An example of SELECTION
this type is the use of a second resistance strain gauge
placed in one arm of a Wheatstone bridge interrogation As an example of the necessary interplay between the
circuit such that its temperature-induced changes cancel out active and passive roles of materials forming a component,
those of the sensing gauge. This second gauge is mounted consider the materials needed to form a spring and its
such that it is not subjected to any strain but is subjected support framework for a precise and stable weighing
to the same temperature as the sensing gauge. Feedback in balance that is little influenced by its ambient temperature.
instruments is discussed in Article 113, Force-feedback The measurement purpose of the spring element (the
Sensors, Volume 2. clearly defined active component) is to produce a repro-
There is sometimes a need to increase the effective ducible and repeatable elongation proportional to the mass
activity of a material. A commonly used method is to place hung on its end.
more than one element in an arrangement that adds or If the material from which the spring is made changes
multiplies the effect. An example is the use of a set of length with temperature, it will exhibit a displacement
series-connected single element thermocouples to form a error for a fixed load as its ambient temperature changes.
thermopile. Another example is where the cold junction This suggests, on first considerations (but see later), that
of a Peltier cooling cell is mounted on a second unit, the the spring should be made from a material that has a
temperature difference then being more than one unit alone zero thermal coefficient of expansion. Figure 1 illustrates
can provide. how this coefficient varies in typical instrument materials,
Steels, woods, Cold worked
Monel-K Invar 46 Cold worked
Invar 36
+50 Brasses, phenolics
Nylon, acrylic, ABS,
polycarbonate Aluminum
plastics Quartz glass,
are 4 to 10 x fused silica
ULE coefficient of steel
Borosilicate Annealed
Relative length change, parts in 106
glass Invar 36
Glass-ceramic
Zerodur
20 °C
Nonexistent ideal zero expansion material
0 +
ULE
Titanium silicate
Zerodur
Carbon fiber composite
Quartz glass (longitudinal)
−30
−200 0 +200
Temperature °C
Figure 1. Thermal expansion of selected metals, plastics, and ceramic materials.
Materials in Measuring Systems 789
showing that some virtually zero coefficient materials are, mass being weighed is removed. This is its mechanical
indeed, available. hysteresis. Steels are quite poor in this regard as they obtain
However, measurement of spring displacement is made some offset with load cycling. Special alloys, such as Ni-
by comparing the length of the spring with a reference Span C , have been developed that exhibit insignificant
support framework that should at first sight be made of mechanical hysteresis, but, unfortunately their TEMPCO
a passive material – but again see later. and TEC are not at the preferred values.
What is actually required, however, is that the framework To make material selection still more complex, the
and the spring have the same thermal expansion coefficient properties mentioned above all may change with history
so that their changes in length track each other as tem- of use, with temperature of operation, and sometimes with
perature changes. As it is expedient to make a frame from level of the present and past history of stress. Additionally,
steel, it does appear feasible to use a steel spring, despite its they are usually time-variant, often in nonsystematic ways.
being quite active to temperature effects, because the frame- The same considerations apply for the support frame
work will change length to the same degree as the spring. design as that provides the positional reference framework.
However, there is another aspect still to be covered. They are, however, usually of less significance.
A vital physical property needed of the measuring spring In practice, precision springs can be made to only certain
material is that its modulus of elasticity remains adequately performance limits. Use of some form of active or passive
constant with variation of operating temperature – this is compensation can then be used to squeeze a little more
expressed as its thermoelastic coefficient or (TEC). sensing performance from the spring. These material limits
Study of the properties of the steel family shows that
were gradually met in the evolving design of spring-mass
its TEC is relatively large. Thus, a steel spring is prone to
clock movements, finally being replaced by better defined
temperature error in that its weighing sensitivity increases
elastic properties of the piezoelectric quartz crystal now
as the temperature rises. For this reason, precision weighing
commonly used in electronic timekeepers.
springs made of steel are not suitable unless a temperature
The design and operation of highly precise and stable
correction is applied.
springs is a matter of setting up a system that recognizes
The material needed ideally has a thermal coefficient of
the sources of likely error arising from material activity
expansion similar to structural materials, yet also has a low
and somehow operates with all of the material restraints in
TEC. No entirely suitable material has yet been developed –
a common situation to be faced in instrument design. a manner that optimizes the operation. In the case of the
The nickel–iron–chromium alloys, however, provide some spring, this may mean tightly controlling the temperature
scope. Figure 2 shows how the TEC varies with alloy to keep the spring at its optimum operational condition.
composition for this family. Note how the alloying ratios An example of the lengths that are needed is seen in
are so critical. the practical design of a top-performance spring-mass geo-
Even if the above needs could be met, there is still physical gravity meter, which is, in effect, an ultrasensitive
another important parameter to be considered – the spring spring balance that can be carried in the field over rough
should return to the same displacement position when a terrain! To retain the best operational performance, it is
necessary to precisely temperature control the spring at all
times. The temperature is selected as that where the com-
600
plex combination of material properties is optimized.
Thermoelastic coefficient,
Thus, it is seen that the design of a sensor is not
just the procedure of selecting an active material and
parts in 106/°C
Cr = 0 setting it up within a system of passive supporting and
containing materials. To get the best performance sensor
design invariably needs complex interaction of material
0
properties with both the active and the passive parts being
Cr = 12% carefully designed.
The problem of interaction between material proper-
−400 ties will be found in any design regime. For example,
20 40 60 80 in optical lens design, obtaining a certain magnification
Nickel% in Ni: Fe alloy power for a given spectral bandwidth will be a matter
Figure 2. Thermoelastic coefficient variation versus alloying
of juggling between availability of a few materials having
composition for the Ni-Fe-Cr alloys. (Copyright Inco Europe limited ranges of refractive index and varying wavelength
Ltd.) dependent transmission losses, see Article 89, Optical
790 Elements: A – Sensors
Materials, Volume 2; and Article 91, Light Sources and Measurement IMEKO, Elsevier; Lausanne.
Detectors, Volume 2. Sensors and Actuators, A – Physical, B–Chemical, Elsevier;
Lausanne.
10 SOURCES OF KNOWLEDGE ON
MATERIALS REFERENCES
10.1 Books Caren, S. (1988) Product design, basic parameters, basic require-
ments, in Handbook of Plastic Materials and Technology, (ed.
Budinski, K.G. and Budinski, M.K. (2001) Engineering I.I. Rubin), Wiley, New York.
Materials: Properties and Selection, Prentice Hall, Engle- Endura (2004) http://www.endura.com/material1.htm.
wood Cliffs, NJ. Ichinose, N. (1987) Introduction to Fine Ceramics, Wiley, Chich-
Crane, F.A.A. and Charles, J.A. (1996) Selection and Use ester.
of Engineering Materials, Butterworths, London. Kroschwitz, J.I. (1990) Concise Encyclopedia of Polymer Science
Crawford, R.J. (1998) Plastics Engineering, Butterworth- and Engineering, Wiley, Chichester.
Heinemann. Rapra (2004) http://www.rapra.net/technicalservices/plastics
Farag, M. (1989) Selection of Materials and Manufacturing rubberselection.htm.
Processes for Engineering Design, Prentice Hall, Engle- RFE (1983) Modern Materials in Manufacturing Industry, Fel-
wood Cliffs, NJ. lowship of Engineering, UK.
Flinn, R.A. and Trojan, P.K. (1986) Engineering Materials Rubin, II. (1988) Handbook of Plastic Materials and Technology,
and Their Applications, Houghton Mifflin, Boston, MA, Wiley, New York.
International Student Edition, (1992), Wiley, New York. Sydenham, P.H. (1979) Measuring Instruments: Tools of Knowl-
Harper, C.A. (2000) Modern Plastics Handbook, McGraw- edge and Control, Peter Peregrinus, Stevenage.
Hill Professional, New York. Trylinski, W. (1971) Fine Mechanism and Precision Instruments;
Harwood, F. (1987) Fine Ceramics: Technology and Appli- Principles of Design, Pergamon Press, Oxford.
cations, Norton Krieger Publishing Company. Trylinski, W. (1983) Mechanical Regime of Measuring Instru-
Jones, D.R.H. and Ashby, M. (1996) Engineering Materi- ments, in Handbook of Measurement Science, Vol. 2, (ed
als, Vol. 1, Butterworth-Heinemann, New York. P.H., Sydenham), Wiley, Chichester.
Kalpakjian, S. and Schmid, S. (2002) Manufacturing Pro-
cesses for Engineering Materials, Prentice Hall, Engle-
wood Cliffs, NJ. FURTHER READING
Phule, P.P. and Askeland, D.R. (2002) The Science and
Engineering of Materials(with CD-ROM), Brooks Cole. Caren, S. (1990) Product design, basic parameters, basic require-
ments, in Handbook of Plastic Materials and Technology, (ed.
I.I., Rubin), Wiley, New York.
10.2 Journals Rubin, I.I. (1990) Handbook of Plastic Materials and Technology,
Wiley, New York.
Measurement Science and Technology (formerly Jour-
nal of Physics E: Scientific Instruments) Institute of
Physics. UK.
This article is based in part on a chapter by W.E. Duckworth, D.D. Harris and P.H. Sydenham originally published in Handbook of
Measurement Science Vol 3, P.H. Sydenham and R. Thorn, Wiley, 1992.
118: Ultrasonic Sensors
Peter J. Lesniewski
University of South Australia, Mawson Lakes, South Australia, Australia
• doppler shift (representing velocity of individual
1 Overview of Ultrasonic Techniques in objects);
Sensing 791 • resonant frequency (affected by load e.g. deposited mass
2 Theory of Ultrasound 793 or fluid level in single frequency and interdigital (IDT)
surface acoustic wave sensors (SAW));
3 Key Ultrasound Principles Used in Sensing 793
• statistical/averaging (e.g. cross-correlation flow mea-
4 Overview of Sensors Using Ultrasonic surement, steady state monitoring in alarm/proximity
Technique 795 sensors, absorption estimated from reverberation time
5 Use of Ultrasound Principles in Sensing in or scattering, and Doppler or spectral characterization
Solids 796 of scatterers such as fish or blood particles).
6 Use of Ultrasound Principles in Sensing in
Liquids 797 Influences by other variables are minimized through
7 Use of Ultrasound Principles in Sensing in compensation, numerical correction, noise/drift reduction,
Gases 798 and evaluation of methodology. An overview is given
8 Limitations of Ultrasonic Methods of Sensing 799 by Hauptmann et al. (1998).
Piezoelectric transducers are typically highly resonant.
References 800
Increasing their bandwidth (BW) reduces sensitivity (in-
creasing losses) and is difficult. The upper frequency range
is limited by the molecular granularity of physical medium
(low-loss ZnO films use GHz range), but common appli-
1 OVERVIEW OF ULTRASONIC
cations use frequencies from 40 kHz to a few MHz. At
TECHNIQUES IN SENSING high frequency f , the wavelength λ = c/f is short and
the particle velocity high, allowing small transducer size,
Although there are a few different sensing principles for microelectronic implementation, and use of wave filters
ultrasonic waves, the focus here is on the leading, linear and lens for focusing. For typical indicative values of the
piezoelectric transducers. Individual measurands are rep- wave velocity c for several materials, refer to Table 1 in
resented by the following categories of the transmitted Section 3.
wave parameters: The acousto-optic effect and laser/microwave excitation
of piezoelectrics and the glow-discharge effect addressed
• propagation or time delay (representing medium flow, further, extend measurement principles.
wave velocity, range/distance); Conversion between the magnetic field and mechanical
• attenuation or complex frequency response (charac- vibrations is available through the magnetostrictive effect.
terizing materials using spectrometry and transmis- Unlike Cr-Ni-Co-Fe-based alloys and nickel, it is the new
sion tomography); giant magnetostriction of Terfenol (Tb–Fe alloy), rare
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
792
Table 1. Comparison of piezoelectric data for various materials.
Piezoelectric material ZnO CdS AlN SiO2 p SPN LiNbO3 BaTiO3 PVDF PZT PZT4 PZT5 LiTaO3 LiSO4 KNbO3
→ parameter (SI units) ↓ p p p Quartz f f p
Piezoel. strain coefficient −58–78 23 110 −123 −171
[×10−12 ]
d31 3
d32 /laterally clamped 10.6 10.3 5–8 2–3 127 40 149, 190 −33 289 374, 593 15, 16
d33 −22
dt
Elements: A – Sensors
Piezoel. stress −5.5–5 216 10
constant [×10−3 ]
g31 19
g32 /laterally clamped 135 122 66–105 57 29 116 14.1–13 −339 25 25, 20 156, 175
g33 −207
gt
Coupling coefficient
k31 0.194–0.21 0.12/1 kHz 0.3
k33 0.28 0.15 0.2 0.1 0.46 0.49 0.48 0.19/1 kHz
kt 0.38 0.29 0.5 0.49, 0.5 0.35, 0.33
Acoustic impedance ρc
[×106 ]
All directional 30 3.9 30
1directional 2.7
Wave velocity c
Longitudinal 6400, 6330 4500, 4460 10 400, 10 700 5700, 5740 6940 7400 5470 2200 (1500) 4600 4350, 4560 5460
(transverse)
Density [103 ] (i.e. vs
water)
ρ 5.68 4.82 3.26 2.65 4.46 4, 6.64 5.55, 5.7 1.78 7.5 7.5 7.75, 7.5 2.06
Relative permittivity
T
ε33 8.84 9.53 8.6 4.58 39 1700 12 1200 1300 1700, 3400 10.3
S
ε33 1260 635 830
Curie temperature/range, 575 <0, >115 >145 328 365, 193 450 410
◦
C
Breakdown voltage field 50 50 200 50 50 80, 100
[V/µm]
Max oper. voltage field 30
[V/µm]
Elasticity const [×10−12 ]
CK33 9.5 15.5 18.8
CD33 7.1 7.9 9.46
CK11 9.1 12.3 16.4
Mechanical tanδ/kHz 0.1
Q factors: QE /QM 100/300 250, 500 50/75
Electrical tanδ [×10−3 ] 15/10 Hz
20/10 kHz
Note: These values should serve as orientation and are not intended for design purposes. Design parameters should be obtained from suppliers of transducers or samples.
Ultrasonic Sensors 793
earth metals, and some compound semiconductors that 3 KEY ULTRASOUND PRINCIPLES USED
offer competitive parameters. Magnetostrictive transducers IN SENSING
exhibit significantly larger conversion than traditional
piezoceramics, are more robust, heat insensitive, and do
not require high voltage, high impedance electric port. The widely used piezoelectric effect offers linear con-
Like piezoelectric, the magnetostrictive transducers are version between the stress tensor representing ultrasonic
reversible (can be used as sensors or actuators). Their ultra- vibration and electric field K received as the output volt-
sonic applications remain currently in the power area, such age. The mathematical formulation, similar to the one
as ultrasonic cleaning. Ultrasonic magnetostrictive sen- used for electrostatic transducers, represents the class of
sors are likely to emerge soon (http://www.public.iastate. reciprocal transducers (whereby electromagnetic transduc-
edu/∼terfenol/). More detail on ultrasonic methods is given ers are antireciprocal). The equivalent circuit uses the
in Article 119, Ultrasonic Instrumentation Principles, dynamic (impedance) type port, (the antireciprocal class
Volume 2 and Article 120, Ultrasonic Instrumentation uses the mobility type) or else a cumbersome gyrator
Design, Volume 2. is required.
The transduction process, best described by single mode
relations, can be extended to the tensor form. Detailed,
valid-today description of both was given by Mason in the
2 THEORY OF ULTRASOUND 1940s. For a review of equivalent circuits, refer to Kino
(1987). Formulation for magnetostrictive transduction is the
same, except for the electric field being replaced by the
Ultrasonic wave is defined as the vibration of medium magnetic field strength H.
particles exceeding the audio frequency (>18 kHz). Field Figure 1 presents a single mode transduction (specific
variables (common with the acoustic wave) are particle orientation of electric and mechanical field) in a thin piezo-
velocity v (vector) and acoustic pressure p in fluids (a electric slice governed by (1a, b).
scalar) or stress tensor [T ], including shear forces in solids The voltage E, related to the electric field strength (K =
or viscous fluids. The ratio p/|v| or Tij /vi represents the E/x), produces strain (S = x/x) if no external force is
medium impedance for the relevant vibration mode and applied:
the products represent complex power density per unit area
x
(defined for a plane wave). x = Ed or = Kd (1a)
Ultrasonics in liquids exhibit specific effects, for x
example, molecular absorption, manifested by a sharp The stress in the material (T = F /A) corresponds to free
increase of losses in wave propagation at a frequency charge Q appearing between the electrodes, manifested
corresponding to the molecular relaxation time of a fluid as the polarization (D = Q/A), if no additional external
(ranging from 30 kHz for carbon dioxide to 10 MHz for
hydrogen).
Nonlinear phenomena result from high particle velocities I E Electric port
or at high ultrasonic signal frequency or from high pressure
level in focused/high energy fields (where either the particle
velocity |v| cannot be regarded as being much smaller than
the wave velocity, and the Euler equation governing the
wave equation becomes nonlinear, or the flux continuity ∆x Variation of electrode distance
equation is nonlinear).
They include F Stress force
• streaming (medium flowing away from transducer’s
A Electrode area
face);
• cavitation (breaking the continuity in a vigorously
vibrating liquid and the development of bubbles);
• fog formation (breaking molecules of a liquid medium
x Electrode distance
away from its surface);
• audibility of modulated waves – used in focused beam Figure 1. Operation of a piezoelectric transducer: stress F /A
audio transmission. causes strain x/A, which produces voltage E (and vice versa).
794 Elements: A – Sensors
polarization occurs: The combined shear-longitudinal motion, in the presence
of a boundary, forms waves propagating even slower than
Q F ct – such as Lamb wave (in waveguides) and Rayleigh
Q = F d or =d (1b)
A A wave (on a surface) utilized in SAW transducers.
Properties of piezoelectric materials are described by a
where d denotes the piezoelectric strain coefficient. large number of parameters with significant redundancy (16
Using incremental notation, strain is expressed as the basic parameters in 4 pairs of constitutive equations). In the
ratio of the material’s infinitesimal displacement over the general tensor form, two essential pairs are
increment of the position along the chosen axis, that is,
for a one-dimensional case: S = ∂ux /∂x. Because the dis- [T ] = [cK ][S] − [e] [K]
(6)
placement itself is a vector, its variation has to be related [D] = [e][S] + [εS ][K]
to a three-dimensional position – thus, strain is a size 3 × 3
[S] = [s K ][T ] + [d][K]
tensor [S] relating two vectors: (7)
[D] = [d] [T ] + [εT ][K]
du = [S] dw where du = [dux , duy , duz ] and where,
dw = [dx, dy, dz] (2) [c] matrix of elastic constants,
[e] matrix of piezoelectric stress coefficients,
Similarly, stress is the variation of the force vector [ε] matrix of dielectric constants,
over the infinitesimal spatial grid. For one-dimensional [d] matrix of piezoelectric strain coefficients,
case: T = ∂Fx /A∂x, but for three dimensions, including [s] matrix of compliance constants,
shear forces, a symmetric 3 × 3 stress tensor [T ] with six
independent variables (linearity condition), is used: while symbol [..]’ denotes the transpose of matrix [..].
Superscript within a matrix denotes that the parameter
F is measured with invariant voltage, that is, K = 0 meaning
d = [T ] dw (3) shorted electric terminals. Similarly, T denotes nulling the
A
variation of internal stress, T = 0, whilst superscript S
The relation between stress and the resulting strain is, in denotes zero displacement condition S = 0, that is, rigidly
fluids, determined by the elasticity or compliance, but in clamped transducer.
solids, the elasticity constant [c] is a 6 × 6 matrix defined Other parameters in use include: piezoelectric stress con-
by the general Hooke’s law: [c] = ∂[T ]/∂[S]. stant g = e−1 , transmitting constant h = e/εS , transforma-
For isotropic solids, [c] contains only three different tion ratio α = d/Cm (where mechanical compliance Cm =
components, all related to two Lamé constants (µ, λ). dx/ dF = s.x/A), and coupling coefficient k = d(εA)−1/2
These relations and the equations of motion lead to the that are important in assessing piezoelectric materials since
wave equation for isotropic materials (4) that can be split it determines the fraction of the input transmitted between
into two simpler potential wave equations (5a, b), one for the mechanical and the electric port in an unloaded trans-
longitudinal (index l) and one for transversal (index t) ducer. Values of k closer to unity are preferred except
wave, each propagating along z with a different velocity for such cases where the Q-factor is exceptionally high
√
cl ≥ ct 2. (Quartz). Note that k33 = kt , since kt is defined for trans-
versely clamped operation.
∂u2 λ+µ µ Many transducers have one dominant mode of oper-
= ∇(∇ · u) + ∇ 2 u (4)
∂t 2 ρ ρ ation (either longitudinal/thickness ‘33’ or shear/bender
1 ∂ 2 vt ∂ 2 vt ‘31’, ‘32’ mode) but full optimization requires following
2 ∂t 2
= (5a) strict relations in mathematical modeling for secondary
ct ∂z2
effects. Table 1 lists the typical parameters of several
1 ∂ 2 vl ∂ 2 vl piezoelectrics indicating discrepancies between independent
2 ∂t 2
= (5b)
cl ∂z2 measurements.
The piezoelectric stiffening phenomenon where piezo-
These two waves, transversal and longitudinal, in ceramics become ‘harder’ after being poled results from
piezoelectric materials are interrelated (disturbing one (6, 7). To prove this, calculate [c] or [s] with and without
affects the other). To avoid dispersion and linear distortion, piezoelectric effect, that is, nulling d or e and assum-
only one mode of propagation is facilitated in ultrasonic ing D = 0 (open electric circuit). Consequently, the lon-
sensors. gitudinal wave velocity in poled materials is increased.
Ultrasonic Sensors 795
Ultrasonic fundamentals are covered well by Ristic (1983) ZnO on silicon) for hair or SAW microsensors are obtained
and Kino (1987). by laser or microwave vacuum sputtering, but also heat
Improved modeling should avoid using inaccurate mate- only (200–500 ◦ C) bonding of piezosensors on silicon is
rial constants – when conditions imposed on a sample’s an inexpensive option since the 1980s.
surface cannot be maintained throughout it. Piezoelectric The fundamental series resonant frequency (Figure 2)
transducer equivalent circuits based on Mason’s descrip- is well defined by the electromechanical parameters of
tion have been related to the general linear transducer the sensor and its load (housing and the medium). The
theory for the general case of distributed parameter devices parallel resonance, which involves electric capacitance of
by Lesniewski and Thorn (1995) leading to a more univer- the material and connections, is less stable. Minute varia-
sal transmission-line circuit. tions of series resonant frequency relate well to the load
SPICE simulations by Püttmer et al. (1997) using the impedance (representing fluid density, phase transition,
Mason-based Redwood model, for impedance interferome- temperature, or deposits on the vibrating surface) and are
ter 1-MHz piezoceramic transducers showed that incorpo- measurable with reasonable accuracy. A self-resonating
rating losses, helps reduce simulation error to 10% and in PZT sensor with frequency resolution of 0.1 Hz in the
some cases below 0.1%. MHz range allows reliable measurement of 10–1000 Hz
The polar pattern of linear transducers in fluids (both variation caused by mechanical loading such as deposi-
sensors and actuators) is determined by the radiation theory tion/growth of biochemical material (refer to Guibault and
of Kirchoff but in practical cases the working integral Luong, 1994).
formula of Rayleigh (8) is used, Narrowband sensors are also suitable in single frequency
systems where phase delay of a fixed frequency continuous
1 e−jkR
(x,y,z) = − vn(x,y,0) dS (8) (CW) or semi continuous signal represents the measurand
2 S R
(e.g. in noncontact surface profiling).
where for each frequency component ω (with k = ω/c The Q-factor of a sensor (damping), which may be deter-
being the wave number) of the velocity distribution vn(x,y,0) mined at its electric port, is useful in characterization of
normal to the transducer’s surface S, the resulting acoustic molecular absorption of solutions. In such measurements,
potential is observed in the far field at a distance R from many resonant frequencies (overtones) specific to the sen-
the source. sor’s geometry are available, allowing frequency profiling
This formula is valid for flat baffled surfaces but it
is often used for slightly curved ones in small enclo-
sures. Resulting inaccuracies are of limited concern because I
nonuniformity of vibrations in transducers cannot be real-
Mech. compliance
istically determined. A better modeling, limited to analysis E
only, is offered by the finite element method.
Vibrating mass Electric
capacitances
Losses
4 OVERVIEW OF SENSORS USING (a)
ULTRASONIC TECHNIQUE Plane transducer, area A, thickness
represented by a transmission line:
Z = Ap cl , cl = (C D/p)0.5,
An ultrasonic wave can be generated in solid dielectrics
containing electric dipoles through application of varying v1 v2
electric field (piezoelectricity in piezoceramics and piezo- −Ce I
composites). While crystals and ferroelectrics have natural F1 F2
permanent polarization created by their dipole or domain
(eA /l ):1
lattice, dipoles in electrostrictive materials are aligned per- Ce + C wire E
manently by applying high electric field (several kV), for (b)
example, by corona or contact poling, preferably cooling Ce = esA /l
them down from the Curie point. Piezoceramics man-
ufactured from precalcinated fine powders mixed with Figure 2. Equivalent circuits of single mode solid-state ultra-
sonic transducers, (a) simplified lumped network or electric
binders, dry pressed, baked, and fired, are finally poled impedance and (b) the exact transmission-line equivalent circuit
and have electrodes deposited. Thin films (0.05–5 µm of developed by Mason in 1948 and verified by Redwood in 1956
Ba2 Si2 TiO8 , TeO2 , Si3 N4 , PZT (lead zirconate titanate), or and 1961.
796 Elements: A – Sensors
Multilayered piezoceramics tungsten are still used for backing piezoceramics but reduc-
ing reverberation and cross talk in arrays is better served
by lossy resin composites, possibly loaded with tungsten.
Epoxy resin Concentrator/lens Composite ceramics offer lower acoustic impedance and
losses with good piezoelectric parameters. Refer to Patel
and Nicholson (1991) and Richter, Reibold and Molken-
Impedance
struck (1991) for performance comparison of conventional
Housing/shielding and composite transducers and to Silk (1984) for back-
matching coupler
ing methods.
The most versatile surface acoustic wave, SAW sensors,
Inert backing developed in the 1970s utilize bulk acoustic wave (BAW)
Electrodes
or thickness shear mode (TSM). They are usually microde-
vices with thin film piezoelectrics, deposited on silicon
substrate, operating at frequencies from a few MHz to a few
hundred MHz, either band pass or resonating at frequencies
Terminals relating to IDT structure. Being extremely sensitive to any
acoustic loading on the surface, responding to deposited
Figure 3. Cross section of a simple bidirectional transducer for
NDT or medical imaging, utilizing a concentrator lens for beam mass, ionic and molar concentrations, temperature, pres-
focusing and offering an increased efficiency/sensitivity through sure, and humidity, they are suitable as probes for conven-
using multilayer piezoelectric structure with damped housing. tional structural tests (http://www.sofratest.com/site/Sales/
Catalog/Systems/SAW Probes/kcu023a.pdf) as well as
probes for microsensing (http://www.techtransfer.anl.gov/
of absorption (relaxation spectroscopy). For a detailed out- techtour/sensors.html).
line, refer to Edmonds (1981, Ch. 2, 3, 4, 8). Since they can be excited remotely by microwaves,
Wider BW (see Figure 3) sensors are needed in pulse- mixed technology devices are possible. Because of their
echo techniques, where a short transient burst represents sensitivity, they require careful compensation against
either time delay (in ranging systems) and/or signal magni- irrelevant variables. For a versatile overview, refer
tude (in defectoscopy). The BW representing the difference to Drafts (2001).
between the maximum and minimum 3-dB angular cutoff Piezoelectrics such as SiC and piezopolymers can also
frequency (ω = ωmax − ωmin , typically ∼50% of the cen- be excited by laser beams.
ter frequency) determines the duration (tmin ∼
= 1/ω) of the A quality ultrasonic sensor is also a small size
sensor’s electrical response to the shortest received tran- capacitance microphone (such as the Bruel&Kjaer 4138
sient indicating time delay error (if correlation techniques type with BW exceeding 100 kHz) or inexpensive 1/8 inch
in signal detection are not used). electret capsules usually offering BW of just over
The group of broadband (BB) sensors use transient or 50 kHz.
wideband signals (chirps, modulated, or coded signals) in
imaging systems, tomography, holography, and sonar; and
requires flat frequency response with minimum distortion. It
5 USE OF ULTRASOUND PRINCIPLES IN
is possible to compensate/equalize for time-invariant linear
distortion using optimized Wiener inverse filter if spectral SENSING IN SOLIDS
components of interest have sufficient energy (above the
noise level). For an inexpensive way of broadening BW Ultrasonic wall thickness/flaw sizing in metals uses the
of resonant transducers and references to the practical pulse-echo mode at a few MHz with single transmit–
use of the Wiener filter, refer to Lesniewski and Thorn receive piezoceramic transducers (backed by dense lossy
(2001). resin) and pads of oil, wax, or resin; or else fully immersed
Transducers for imaging applications (sonar, medical) in liquid to facilitate coupling. For timber, concrete, and
appear as planar or circular arrays with electronic beam- tires, the frequency is reduced by a decade due to higher
forming, but in single low power devices, for example, losses. The actuator is separate from the sensor, although
in nondestructive testing (NDT), lenses/concentrators (of both may remain in one housing. Flaws are assessed accord-
aluminum, epoxy resin, magnesium) offer fixed focus. ing to the delay time of observed reflected pulse and its
For details of a 4.5-MHz PZT array for medical imag- relative amplitude (refer to Krautkrämer and Krautkrämer,
ing, refer to Turnbull and Foster (1992). Bronze, brass, or 1990) (http://www.krautkramer.com/).
Ultrasonic Sensors 797
For monitoring a machining tool (wear, cooling), Nayfeh, for sonar and medical imaging (refer e.g. to Schlaberg and
Eyada and Duke (1995) employed the pulse-echo method- Duffy, 1994).
ology and precise measurements of a reflected 10-MHz Sonar transducers are either small arrays of piezoce-
burst using an ultrasonic transducer mounted within the tool ramic elements or single transducers with focusing con-
holder. Similarly, Morton, Degertekin and Khuri-Yakub cave face and/or lenses embedded in epoxy resin. They
(1999) monitored baked photoresist by relating its tem- are used for bottom profiling or fish location ({echo-
perature and thickness to the reflection coefficient in sounding}) using pulse-echo principle and narrowband sig-
pulse reflection measurement using a GHz BW, ZnO-on- nals (http://www.simrad.com/). A multibeam commercial
saphire transducer. unit Simrad EM3000 offers depth resolution of 1 cm within
Single transducers for medical imaging tend to employ 200-m range at 300-kHz operating frequency (for minimum
PZT in epoxy housing, but a copolymer PVDF-TrFE with attenuation in seawater).
lossy polymeric backing offers lower acoustic impedance, Other applications include velocimetry, where a measur-
reducing the mismatch while maintaining a reasonably flat and that uniquely influences ultrasonic wave velocity in a
120 MHz BW. given liquid is found through time of flight (ToF) of an
Research in constructing improved piezoelectric trans- ultrasonic transient traveling over known distance. Mea-
ducers for medical imaging has led to specific profiling surement of fluid composition, its temperature, density, or
of the surface, subdividing the front electrode, and/or concentration (water–air, water–alcohol, oils/fats, alcohols,
graded poling (for state-of-the-art microfabricated trans- molecules in a solution) using the above approaches has
ducers, refer to http://www.sensant.com/pro advDiag.html). been reviewed by Povey (1997). The variation of wave
This has coincided with new solutions of the acoustic velocity tends to be as high as 10% (e.g. for temperature
harmonic wave equation from 1987 (following a simi- variation of 30 ◦ C), but effects on wave attenuation by a
lar, earlier development in the electromagnetic wave the- liquid medium, although under 1% and strongly influenced
ory), representing a new class of limited diffraction and by frequency, are also measured (refer e.g. to examples
localized waves (nondiffracting waves, X-waves, Bessel by Harker and Temple, 1991). The use of ultrasonic sensors
Beams, transient Bessel Bullets), which form very nar- in velocity measurements is straightforward (using edge
row beams, although they decay faster than conventional triggered timer with transient signal amplitude detection)
wave types. as long as the sensor’s low linear distortion allows it to
An experimental pulse-echo 2-D imaging system with maintain undistorted transient and good definition of the
small limited diffraction beam transducer arrays (48 or arrival time. If distortion is significant, then a reference
64 elements) has been developed by Lu (1997, 1998) transient propagating in a known reference liquid can be
(http://bioe.eng.utoledo.edu/Laboratories/UltrasoundLab/ used and two distorted transients correlated to establish the
jian yu lu lab.html). time delay.
It offered improved images of tissue phantoms up to a Imperfect sensors can still be used in velocimetry based
depth of 200 mm utilizing a new Fourier transform–based on interferometry utilizing the superposition of continu-
algorithm allowing high frame rate processing. Theoretical ous, sinusoidal signal with its delayed version (through
analysis of the new waves and an outline of other develop- a known length of transmission in an unknown fluid).
ments is offered by Salo, Fagerholm and Friberg (2000), This interference is manifested by resonances affecting
a numerical analysis of Bessel Bullets by Stepanishen the resulting frequency response of such systems where
(1999), and an investigation of finite planar transducers the wave velocity c can be found from the location of
by Chatzipetros et al. (1998). resonant frequencies: fn = n(c/ l), or notch frequencies
(antiresonances): fn = (2n + 1)(c/ l), n = 1, 2, 3, 4, 5 and
so on.
6 USE OF ULTRASOUND PRINCIPLES IN With a single transmit–receive transducer (where T
SENSING IN LIQUIDS and S1 would become the same transducer in Figure 4),
the reflected wave travels twice the length of the fluid-
Steady coupling and reduced acoustic impedance mismatch filled tube terminated by a reflecting/rigid end. Examples
between the medium and the immersed transducer offers of acoustic interferometers are given by Trusler (1991)
good interface; thus, solid-state ultrasonic transducers in and review of measurement approaches is given by Mc-
liquids are readily used. Resonances are damped and the Clements (1995). Errors in this method are caused by
sensor’s BW is larger than the BW in gases. The mismatch imprecise location of the resonance frequencies, which
may be efficiently reduced by using novel composite mate- results from multipath transmission in the fluid tested.
rials (elastomer based piezorubbers or copolymers) suited The transducer cannot radiate a perfect plane wave with
798 Elements: A – Sensors
Distance c by offering two such equations (Figure 5). A commer-
cial five-path gas flowmeter has an error less than 0.5%
T
S1 Fluid S2 within the range of 0.4 to 20 ms−1 – Fluenta, FMU700
(http://www.fmckongsbergsubsea.com/).
Att.
The delay time used for ranging/positioning (in robotics)
Sinewave ∑ Absorbing and level control allows sensing distance to reflecting
generator termination objects (locally flat, of size larger than signal wavelength
Display
(amplitude vs with an abrupt change of acoustic impedance). Often 40-
frequency) kHz center frequency, narrowband piezoceramic sensors
with aluminum horn/scatterer are used to reduce impedance
Figure 4. Principle of an ultrasonic interferometer with two mismatch else polyvinylidene fluoride (PVDF) sensors are
independent sensing points (Note: (i) The sensor S1 may be used – for an overview refer to Manthey, Kroemer and
omitted and the signal taken directly from the transmitter T.
(ii) Only one transmit–receive transducer is used and the tube Magori (1992). The impulse response is of irregular ring-
termination is fully reflecting in a single transducer system). ing form having autocorrelation function with many lobes.
Therefore, the limiting error reaches several microns and
yet the wave velocity has to be determined. In simpler sys-
a constant phase over its body – therefore, the wave
tems (Vargas et al., 1997), the sensed amplitude envelope
reaching the sensor may be interpreted as a sum of
of the signal leads to a pulse duration of some 100 µs with
beams, some reflected at the tube walls and having dif-
relative error reaching 3%. Another robotic system devel-
ferent phases, and yet reaching the sensor with differ-
oped by Munro et al. (1990), with conventional 50-kHz
ent phases.
transducers offered 10-m sensing range.
Acousto-optic effects, such as diffraction of light,
For the state-of-the-art selection, refer to: http://www.
variation of refractive index or scattering, and frequency
hexamite.com/.
shifting by nonuniform medium density (and by surface
Reduction of errors limiting ultrasonic ranging and/or
waves in solids) created by a perpendicular ultrasonic
imaging can be obtained through integrating results from
standing wave are used for nonacoustic sensing of high-
a system producing uncorrelated errors.
frequency ultrasonic waves. The deformed light beams
In a robotic collision avoidance system, Chou and Wykes
are sensed by photodetectors or interferometers (the
(1999) combined four-element 100 kHz cf−1 ultrasonic lin-
most popular commercial sensors being the Bragg Cells)
ear array (giving poor angular, but good range resolution,
(http://www.goochandhousego.com/page/home.html). Re-
reaching 1.7 mm at a distance of some 140 mm) with a
view of relevant sensing techniques is offered by Scruby
charge-coupled device (CCD) optical imager having differ-
and Drain (1990). Chiang, Chan and Gardner (1990)
ent characteristics/limitations.
have outlined an improved sensor that can be used
Abreu et al. (1999) combined electromagnetic an ultra-
for detection of ultrasonic wave up to the MHz range
sonic positioning 3-D system with one transmitter (pulse
(sufficient for medical applications) based on birefringence
source: an electric spark) and three remote receivers (con-
in optical fiber (ultrasonic field varying the refractive
ventional resonant 40-kHz transducers) obtaining a resolu-
index and causing dispersion of differently polarized light
tion of 1 mm within a range of 5 m.
beams).
Combining a conventional 40-kHz ultrasonic CW Dopp-
ler system (utilizing 5-Hz frequency resolution) with a
microwave motion detector unit has improved the elimina-
7 USE OF ULTRASOUND PRINCIPLES IN tion of false alarms in an intruder alarm system described
SENSING IN GASES by Ruser and Magori (1998).
An efficient transducer, developed for operation in air
Because of the low wave velocity in gases, ultrasonic by Kielczynski, Pajewski and Szajewski (1990), comprised
sensors find application in flow velocity (v) measure- one or more axially vibrating PZT rings embedded in
ment, by sensing the time delay, td = distance/(c ± v), low impedance resin, transmitting vibrations to the air.
between two transmission points along the gas flow. A The resulting device had a narrow beamwidth (<4◦ ) and
two-path system with simultaneous transmission along and bandwidth of 100–200 kHz (larger than the rings alone),
against the flow allows elimination of the wave velocity although containing several resonances.
Ultrasonic Sensors 799
Sensor
Medium flow a
Velocity vm
C +V Direct
t2
t1 transmission
C −V into
solid wall
V = Vm × COS a
Fast wave in solid
t2/t1 = (1 + v/c)/(1 − v/c) Causes interference/error
c = acoustic wave velocity Sensor
Figure 5. The principle of operation of a two-path ultrasonic flow velocity measurement (Note: the dotted path identifies the cross-talk
transmission, interfering with the measured transient and causing errors).
An option that may prove successful is the use of sensors tissue in medical imaging is also a source of signal attenu-
where glow-discharge is modulated by acoustic pressure – ation that cannot be avoided due to impedance discontinu-
refer to Kocis and Figura (1996). ity/instability.
8 LIMITATIONS OF ULTRASONIC 8.2 Delay time
METHODS OF SENSING
The usual assumption in signal transmission is that signal
General limitations are gradually lifted with the steady travels through a homogeneous and time-invariant medium.
progress in technology (frequency range, automation) and In fact, a flowing medium is unsteady and nonuniform,
material processing (material purity, profiling accuracy) but with bubbles or multiple phase, yet there are also reflected
the following fundamental limitations remain. signals from instrument housing – leading to significant
uncontrollable distortion of the received signal. This limits
the accuracy of the delay measurement and therefore ultra-
8.1 Impedance mismatch sonic transducer needs only the BW sufficient to generate
a test signal whose autocorrelation has a width comparable
Affects any method involving transmission in gases, where with the above error. Narrowing beamwidth may be jus-
the huge impedance ratio (7000:1 or worse) between a tified in tomographic reconstruction where many sensors
solid-state transducer and the medium introduces signal are used.
loss. In actuators, the peak signal level may be increased In signal reflection, narrow beam and wide, BW of
to some 100 dB at 1 m by reaching the maximum operating transducers is most important, while limitations come from
voltage of the piezoelectric material (electric breakdown interfering reflections and imprecise determination of the
or linearity limit), but in sensors impedance transforma- wave velocity (a differential method with reference path
tion is necessary. Wave filters offer any transformation but is useful).
over a narrow frequency range, while horns increase the
level/sensitivity by some 30 dB at the expense of reduced
frequency range and a likely dispersive transmission (for 8.3 Polar and spatial response
horn theory refer to Ensminger, 1988 and for modeling,
to Lesniewski, 1995). In the case of small size sensors with A narrow response is useful in isolating interfering signals
reduced source capacitance (as low as 10 pF), even a short but it is costly and it negatively affects frequency response
connecting cable affects the response, thus, sensor-amplifier and sensitivity. If lenses or special techniques (digitally
hybrids increase sensitivity by more than 10 dB (Zheng equalized/filtered arrays) are not used, then the main lobe
et al., 1997). Contact between solid-state transducer and is frequency dependent and sidelobes are created. Finally,
800 Elements: A – Sensors
severely distorted interfering signals may be more difficult Lesniewski, P.J. and Thorn, R. (2001) Generating Broadband Test
to handle (e.g. spikes turning into longer ringing). Signals with Resonant Piezoelectric Transducers. Measurement
Science and Technology, 12, 723–735.
Lu, J. (1997) 2D and 3D High Frame Rate Imaging with Limited
REFERENCES Diffraction Beams. IEEE Trans. on Ultrason. Ferroel. and Freq.
Control, 44(4), 839–856.
Abreu, M.J.M., Ceres, R., Calderon, L., Jimenez, M.A. and Lu, J., (1998) Experimental Study of High Frame Rate Imaging
Gonzalez-de-Santos, P. (1999) Measuring 3D-Position of a with Limited Diffraction Beams. IEEE Trans. on Ultrason.
Walking Vehicle Using Ultrasonic and Electromagnetic Waves. Ferroel. and Freq. Control, 45(1), 84–97.
Sensors and Actuators, 75, 131–138. McClements, D.J. (1995) Chapter 4, in Ultrasonic Characteriza-
Chatzipetros, A.A., Shaarawi, A.M., Besieris, I.M. and Abdel- tion of Foods, in: Characterization of Food (eds A.G. Gaonkar),
Rahman, M.A. (1998) Aperture Synthesis of Time-Limited X Elsevier, Amsterdam; New York.
Waves and Analysis of their Propagation Characteristics. Jour- Manthey, W., Kroemer, N. and Magori, V. (1992) Ultrasonic
nal of the Acoustical Society of America, 103(5), 2287–2295. Transducers and Transducer Arrays for Applications in Air.
Chiang, K.S., Chan, H.L. and Gardner, J.L. (1990) Detection of Measurement Science and Technology, 3, 249–261.
High-Frequency Ultrasound with a Polarization-Maintaining Morton, S.L., Degertekin, F.L. and Khuri-Yakub, B.T. (1999)
Fiber. Journal of Lightwave Technology, 8(8), 1221–1227. Ultrasonic for Photoresist Process Monitoring. IEEE Transac-
Chou, T.N. and Wykes, C. (1999) An Integrated Ultrasonic Sys- tions on Semiconductor Manufacturing, 12(3), 332–339.
tem for Detection, Recognition and Measurement. Measure- Munro, W.S.H., Pomeroy, M., Rafiq, M., Williams, R., Wybrow,
ment, 26, 179–190. M. and Wykes, C. (1990) Ultrasonic Vehicle Guidance Trans-
Drafts, B. (2001) Acoustic Wave Technology Sensors. IEEE ducer. Ultrasonics, 28, 250–354.
Transactions on Microwave Theory and Techniques, 49(4), Nayfeh, T.H., Eyada, O.K. and Duke, J.C. (1995) An Integrated
795–802. Ultrasonic Sensor for Monitoring Gradual Wear On-Line Dur-
Edmonds, P.D. (ed.) (1981) Ultrasonics, Methods of Experimen- ing Turning Operations. International Journal of Tools Manu-
tal Physics, Vol. 19, Academic Press, Orlando, FL. facturing, 35(10), 1385–1395.
Ensminger, D. (1988), Ultrasonics, Marcel Dekker, New York. Patel, N.D. and Nicholson, P.S. (1991) Influence of Effec-
tive Diameter and Piezoelectric and Dielectric Properties on
Guibault, G.G. and Luong, J.H.T. (1994) Piezoelectric Immu- the Performance of Ultrasonic Transducers. Ultrasonics, 29,
nosensors and their Applications in Food Analysis, Chapter 7, 389–394.
in Food Biosensor Analysis (eds G. Wagner and G.G. Guibault),
Marcel Dekker, New York (pp. 151–172). Povey, M.J.W. (1997) Ultrasonic Techniques for Fluids Charac-
terization, Academic Press.
Harker, A.H. and Temple J.A.G. (1991), Propagation and Attenu-
ation of Ultrasound in Suspensions, in Developments in Acous- Püttmer, A., Hauptmann, P., Krause, O. and Henning, B.
tics and Ultrasonics (eds M.J.W. Povey and D.J. McClements), (1997) SPICE Model for Lossy Piezoceramic Transducers.
IOP Publishing, Bristol, UK; 1992 (pp. 19–31); Proceedings of IEEE Trans. on Ultrason. Ferroel. and Freq. Control, 44(1),
the meeting organised by the IOP Physical Acoustics Group, 60–66.
Leeds, UK, 24–25 September 1991. Richter, K.P., Reibold, R. and Molkenstruck, W. (1991) Sound
Hauptmann, P., Lucklum R., Püttmer, A. and Henning, B. (1998) Field Characteristics of Ultrasonic Composite Pulse Transduc-
Ultrasonic Sensors for Process Monitoring and Chemical Anal- ers. Ultrasonics, 29, 76–80.
ysis: State-of-the Art and Trends. Sensors and Actuators A, 67, Ristic, V.M. (1983) Principles of Acoustic Devices, John Wiley &
32–48. Sons.
Kino, G.S. (1987) Acoustic Waves: Devices, Imaging and Analog Ruser, H. and Magori, V. (1998) Highly Sensitive Detection
Signal Processing, Prentice Hall, Englewood Cliffs, NJ. with Combined Microwave-Ultrasonic Sensor. Sensors and
Kielczynski, P., Pajewski, W. and Szajewski, M. (1990) Ring Actuators A, 67, 125–132.
Piezoelectric Transducers Radiating Ultrasonic Energy into Salo, J., Fagerholm, J. and Friberg, A.T. (2000) Unified Descrip-
the Air. IEEE Transactions on Ultrasonics Ferroelectrics and tion of Nondiffracting X and Y Waves. Physical Review E,
Frequency Control, 37(1), 38–43. 62(3), 4261–4275.
Kocis, S. and Figura, Z. (1996) Ultrasonic Measurements and Schlaberg, H.I. and Duffy, J.S. (1994) Piezopolymer Composite
Technologies, Chapman & Hall, London. Arrays for Ultrasonic Medical Applications. Sensors and Actu-
Krautkrämer, J. and Krautkrämer, H. (1990), Ultrasonic Testing ators A, 44, 111–117.
of Materials, Springer-Verlag, Berlin; New York. Scruby, C.B. and Drain, L.E. (1990) Laser Ultrasonics – Tech-
Lesniewski, P. (1995) Discrete Component Equivalent Circuit for niques and Applications, Adam Hilger.
Webster’s Horns. Applied Acoustics, 44(2), 117–123. Silk, M.G. (1984) Ultrasonic Transducers for Nondestructive
Lesniewski, P. and Thorn, R. (1995) A Z-Matrix Model of the Testing, Adam Hilger.
Thickness-Mode Piezoelectric Transducer. Acoustics Letters, Stepanishen, P.R. (1999) Acoustics Bessel Bullets. Journal of
19(3), 43–46. Sound and Vibration, 222(1), 115–143.
Ultrasonic Sensors 801
Trusler, J.P.M. (1991) Physical Acoustics and Metrology of Flu- Vargas, E., Ceres, R., Martin, J.M. and Calderon L. (1997) Ultra-
ids, Adam Hilger. sonic Sensor for Liquid-Level Inspection in Bottles. Sensors
Turnbull, D.H. and Foster, F.S. (1992) Fabrication and Character- and Actuators A, 61, 256–259.
ization of Transducer Elements in Two-Dimensional Arrays for Zheng, X.R., Lai, P.T., Liu, B.Y. and Cheng, Y.C. (1997) An
Medical Ultrasound Imaging. IEEE Trans. on Ultrason. Ferroel. integrated PVDF ultrasonic sensor with improved sensitivity
and Freq. Control, 39(4), 464–474. using polyimide. Sensors and Actuators A, 63, 147–152.
119: Ultrasonic Instrumentation Principles
Lawrence C. Lynnworth
Lynnworth Technical Services, Waltham, MA, USA; Formerly Chief Technologist at GE Panametrics
If more than one parameter is varying, say temperature
1 Active and Passive Transducers 802 and pressure, it may be necessary to observe more than
2 Effects of Measurand on Ultrasonic one characteristic of the generated or received wave to
Properties 802 resolve ambiguities. Generally speaking, one can say that
3 Transducer Effects Used in Ultrasonic each measurand (flow, temperature, pressure, etc.) influ-
Instrumentation 803 ences one or more of the following: generation, propaga-
4 Guiding Ultrasonic Waves to Sense a tion, and detection of ultrasound. For a fuller account of
Measurand Remotely; Buffering an Ultrasonic such influences and their measurement see Asher (1997),
Transducer 805 Auld (1990), Baker (2000), Brown and Lynnworth (2001),
Hunt, Stubbs and Lee (2003), Kleppe (1989), Lynnworth
5 Container Walls as Sensors or Windows for
(1989), Lynnworth and Mágori (1999), Youngdahl and
Noninvasive Interrogation 806
Ellingson (1982), and Royer and Dieulesaint (1999).
6 Limitations on Precision and Accuracy 807
References 807
Further Reading 808 2 EFFECTS OF MEASURAND ON
ULTRASONIC PROPERTIES
1 ACTIVE AND PASSIVE TRANSDUCERS Consider Figure 1. Without further study, it is not obvi-
ous as to which is the best way to proceed, where, say the
Transducers for launching or detecting ultrasonic waves object is to measure water temperature. Ultrasonic trans-
utilize mechanical, electromagnetic, chemical, pneumatic, ducers and instruments have been available commercially
fluid dynamic, or thermal phenomena. Transducers for for measuring signal characteristics related to temperature.
ultrasonic instruments are mainly in the electroacoustic Their operation is based on the temperature dependence
category. Here, electrical energy is converted to acoustic of: (i) the transducer frequency; (ii) the speed of sound in
energy, when transmitting; acoustic or mechanical wave water; (iii) the speed of sound in an intrusive sensor or
energy is converted to an electrical signal, when receiving. probe; or (iv) the speed of sound in the beaker material.
An important concern in the design of ultrasonic trans- In discussing the laboratory experiment depicted in
ducers and measuring instruments is the optimization of Figure 1, the influence of pressure has been disregarded
the system with respect to a reliable and accurate deter- because this effect is relatively minute unless pressure is
mination of the measurand – the parameter whose value very high. Actually, the speed of sound depends on pressure
is to be inferred or computed on the basis of ‘ultrasonic’ in all the media involved. These include the solids (glass or
observations. In an ultrasonic experiment, one may observe metal beaker; quartz crystal, which we shall imagine to be
transit time, amplitude of the received signal, oscillator fre- embedded within the housed transducer assembly), the liq-
quency, oscillator strength, or other terms related to these. uid (water), and the gas (air) that remain after all the water
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Ultrasonic Instrumentation Principles 803
liquid at an approximately or accurately known pressure
is illustrated for several liquids in Matson et al. (2002).
Air Figures 3 and 4 suggest methods for determining air pres-
Pulser or sure or density.
pulser/receiver
Transmitting Water Receiving
transducer Oscilloscope
transducer
3 TRANSDUCER EFFECTS USED IN
Beaker
ULTRASONIC INSTRUMENTATION
Hot plate
3.1 Piezoelectric, magnetostrictive, electrostatic,
Figure 1. Ultrasonic effects of heating water in a beaker on and electromagnetic mechanisms
which a diametrically-opposed pair of resonant piezoelectric (e.g.
quartz) transducers are mounted. (Reproduced by permission of
Panametrics, Inc.) The piezoelectric effect discovered by the Curie brothers
over a century ago, involves a coupling between elastic
has boiled away. One might expect pressure effects to be
negligibly small in an ‘incompressible’ liquid like water, Plastic pipe, 6"
schedule 80 PVC
and certainly negligible in a solid such as quartz. But this
is not the case. For example, in the particular cut of quartz
used by one manufacturer in their quartz pressure gauge,
a change of pressure of 69 Pa (0.01 psi) was resolvable as
a difference in its resonant frequency. In the case of air,
Air, no
pressure may be calculable if one can utilize the amplitude flow
of the received signal, or the attenuation coefficient, as a
measure of air pressure. Once pressure is approximately
known, the temperature of dry air (or air of known relative
humidity) can be calculated from the speed of sound.
The dependence of the speed of sound on temperature
and density in a fluid such as oxygen is given in Figure 2
(Straty and Younglove, 1973). Their data suggests that, over
the range studied, the density of fluid oxygen could be
computed from measurements of speed of sound and tem- 100
Plastic pipe, 6"
perature. The idea of determining the density of a known 90 schedule 80 PVC
Signal amplitude (mV)
80
70 Air, no
1200 flow
70 K 60
50
1000 100 K 40
Velocity of sound c (m s−1)
30
20
800 130 K 10
0
160 K 0 10 20 30 40 50 60 70 80
600 Saturation
190 K Air pressure (psig)
250 K 200 K
280 K
400 300 K
Figure 3. Example of air pressure measured at room temperature,
and at no flow with clamp-on ultrasonic transducers. This mea-
surement utilizes the principle that the amplitude of the received
200 signal is proportional to the acoustic impedance of air and there-
fore, to a good approximation, proportional to the air pressure.
0 0.2 0.4 0.6 0.8 1.0 1.2 Under turbulent flow, signals become jittery. At Mach num-
Density r (g cm−3) bers >0.1, beam drift further complicates the simple relationship
depicted in the graph. Changes in temperature or gas composition
Figure 2. Velocity of sound versus density for saturated liquid also influence the pressure/amplitude relationship. (Reproduced
and compressed fluid oxygen. (After Straty and Younglove, 1973.) by permission of Panametrics, Inc.)
804 Elements: A – Sensors
Transmitter/receiver
ultrasonic measuring
instrument
Reflector
o28.5 µs
detail
Sensors
Transducer Lead-in A B C D E...K
Receiver monitor trace
(a) (b)
1.3
Air density vs sound speed
0° C to 60° C, dry to 100% RH
1.2
Dry air
P = 760 mmHg
Density (gl−1)
10% RH
1.1
50% RH
Dry air sound speed and density data from Weast, R.C., Handbook of
1.0 Chemistry and Physics. Perturbations from dry air properties due to
water vapor are calculated from known relationships (see Bohn, D.A.,
Environmental Effects on the Speed of Sound, J. Audio Eng. Soc., Vol.
36, No. 4, April 1988) which were normalized to the values for dry air.
100% RH
0.9
330 340 350 360 370 380
Sound speed (ms−1)
(c)
Figure 4. Ultrasonic thermometry using the speed of sound as a multipath indicator of temperature: (a) in the medium itself, (b) in an
intrusive waveguide with one or more measuring zones, (c) extending concept (a) to density measurement based on the speed of sound
in air. (Reproduced by permission of Panametrics, Inc.)
and dielectric phenomena. Piezoelectricity is found only in quartz, and approximately 1200 ◦ C for LiNbO3 . The usable
materials that have no center of symmetry. Quartz, lithium frequency range extends down below the ultrasonic band
niobate, tourmaline, and other crystals exhibit the effect, into the audio and infrasonic range, and up to the giga-
but probably most industrial transducers, as used in testing hertz range.
and process measurements today, use poled ceramics as Magnetostrictive devices are usually limited to frequen-
the electroacoustic element. PZT (lead zirconate titanate) cies below a few hundred kilohertz by the electrical skin
and LM (lead metaniobate) are examples of such materi- effect, although operation into the megahertz range has been
als. As a transmitter, piezoelectric (ferroelectric) materials reported. Ni, Fe, Co, and their alloys are magnetostrictive,
expand (or contract), or exhibit shear, in response to a driv- that is, they twist or change their dimensions in response
ing voltage. As a receiver, an electrical signal is generated to an applied magnetic field. Conversely, in response to
in response to an incoming pressure or shear wave. These a torsional, tensile, or compressional load, they generate a
effects can be observed over a wide temperature range, from magnetic field that is typically sensed by a coil surrounding
nearly absolute zero up to the Curie point or the transfor- the magnetostrictor. As a compressional stress wave prop-
mation temperature of the element, for example, 573 ◦ C for agates along a magnetostrictive wire, one can imagine a
Ultrasonic Instrumentation Principles 805
little bar magnet moving through the wire at the speed of waveguiding is beyond the scope of this section.) The sens-
sound. When that magnet passes through an encircling coil, ing and buffering purposes of interest here include (i) using
a voltage is induced in that coil. propagation in the waveguiding structure to sense one or
Electrostatic forces have been used in microphones for a more characteristics of the waveguide, such as its elastic
long time. Micromachining has ‘modernized’ the options, moduli or its temperature; (ii) to sense a characteristic in
leading to CMUTs (capacitive micromachined transducer). the medium adjacent to the waveguide, such as the den-
Many of their performance and manufacturability character- sity or viscosity of the fluid in which the waveguide is
istics were summarized by Oralkan et al. (2002). Because immersed; or (iii) to buffer an ultrasonic transducer from
of the low mass of the radiating membrane, CMUTs have a hot fluid into which the buffer radiates in order to mea-
been used in air transducers and immersion transducers. sure some characteristic of that fluid, for example, speed of
Electromagnetic acoustic transducers (EMATs) utilize sound, attenuation coefficient, or flow velocity. Waveguides
Lorentz forces in conductors. The Lorentz force is relatively can convey ultrasound to and from a remote region where
small, but if the conductor is magnetostrictive, the combined the transducer, if exposed to the harsh environment of that
effect is large enough to be utilized with essentially stan- remote region, might fail because of thermal or chemi-
dard flaw detectors designed for ordinary piezoelectric cal attack, or excess nuclear exposure. Another reason for
transducers. Their chief advantage is the avoidance of using waveguides is because the specimen size or shape,
any couplant. This can be important if the object under for example, micrometer-diameter glass fiber, is such that
test has a rough surface, for example, a heavily oxidized it is awkward to introduce the interrogating energy directly
iron pipe. from the transducer to that specimen. In other cases, the
waveguide assists in mode-converting from extensional to
4 GUIDING ULTRASONIC WAVES TO torsional; from transverse shear to torsion; from extensional
to flexural, and so on. Sometimes, multiple purposes are
SENSE A MEASURAND REMOTELY; simultaneous goals.
BUFFERING AN ULTRASONIC Table 1 illustrates waveguides and buffers investigated
TRANSDUCER by various authors between 1934 and 2003 in monostatic
(pulse-echo) and bistatic (through-transmission, or pitch-
Ultrasonic waves can be guided along elastic structures such catch) arrangements.
as rods, tubes, and plates; and through fluids in pipes or Figure 5, after Nguyen, Smart and Lynnworth (2004),
tubes. (In the ocean, acoustic waves can be guided in a shows pulse-echo and through-transmission torsional-
sandwiched channel when the water above and below has wave interrogation of a noncircular portion of an
a sound speed higher than that in the channel. But such otherwise-circular cross-section waveguide, in order to
Table 1. Buffers for cryogenic, high temperatures, or for other nasty fluids.
For longitudinal waves: References For extensional waves: References
Plain thick rod F (1948) Thin rod B (1957)
Threaded rod Marginally dispersive rod L (1992)
A & M (1977);
– Uniform pitch Y & E (1978, 1982) L, L & Z (1998);
Bundle of thin rods L & L (1999 & 2002)
– Non uniform pitch or concentric tubes Fiberacoustic
N & M (1993)
bundle
– Knurl, undulating, other irregular surface Spiraled sheet H & K (2002)
Clad rod
For torsional waves: see thin rod, above B (1957);
– Cylindrical J & L (1998) A & T (1957)
For transverse shear waves:
L, J & C (1996);
–Tapered J & L (1998) Thin blade
L (2000)
Threaded rod - see diagram, above left, or
Hollow tube S & G (1934) A & M (1977)
Article 120, Ultrasonic Instrumentation
Design, Volume 1, Figure 6, cell for 1977
Clad rod-see diagrams, above left J & L (1998)
Reproduced by permission of Panametrics, Inc.
806 Elements: A – Sensors
parameters affecting propagation. If temperature is sensed
by a thermocouple, only the two remaining variables need
f6.
4m to be sorted out ultrasonically.
m
(b)
25 5 CONTAINER WALLS AS SENSORS OR
.
mm 4
(a)
WINDOWS FOR NONINVASIVE
100 INTERROGATION
mm
When categorizing ultrasonic instrumentation principles
and techniques, it is convenient to organize them according
Figure 5. (a) Through-transmission configuration with two pairs
to whether the sensor is the medium itself or an intrusive
of transverse shear transducers, one pair as transmitters and one sensor. Here we use ‘sensor’ to mean the medium, material,
pair as receivers, at each end of a waveguide. These generate and or structure in which the generation, reception, or prop-
detect torsional waves by a mode conversion process (Nguyen, agation of ultrasound is measured and in which the said
Smart and Lynnworth, 2004; Lynnworth, Cohen and Nguyen, generation, reception, or propagation is influenced by the
2004). The waveguide with the diamond cross section comprising
state, characteristics, or value(s) of measurands in the path.
its center section responds to liquid density (Kim and Bau, 1989;
Shepard et al., 1999; Lower, Brow and Kurkjian, 2004) and Referring again to Figure 1, if we think of water as its
(b) Pulse-echo configuration with one pair of transducers coupled own sensor (‘Propagation’ column in Table 2), we must
to the circular cross-section end of the waveguide. (Reproduced nevertheless acknowledge that the electroacoustic transduc-
by permission of Panametrics, Inc.) ers might be legitimately called sensors. By designating
water as the sensor, we can emphasize that the speed of
sense the density of fluid surrounding the noncircular sound in water is the basis of the way in which we obtain
portion. This technique has a resolution on the order of information on the water’s temperature. The electroacous-
1 mg cm−3 and, therefore, while sensitive enough for many tic piezoelectric quartz transducers shown in this particular
industrial liquid applications, it is not particularly useful for illustration could be replaced by an external laser genera-
sensing the density of gases. Air, for example, has a density tion and detection system, focused on opposite sides of a
of 1.29 mg cm−3 at ordinary conditions. On the other hand, black beaker, yet the speed of sound in water could still
when the sensor portion is immersed in a low-density gas, it serve as the basis for determining water temperature.
can sense the gas temperature, T , nearly independent of gas Finally, one must not overlook what might appear to be
density or viscosity, especially if the T -sensor is a round an intermediate category, that is, using the existing wall
rod or tube. used as a sensor. As one example, the acoustic ‘ringdown’
What happens if the sensor is immersed in a high- or echo decay pattern in the wall of a vertical standpipe is
viscosity liquid? The torsional wave slows down in response often proposed to tell if liquid is present at that level. In the
to density and viscosity, besides being responsive to tem- common example of a clamp-on flowmeter, the wall of the
perature. In this case, compensation is required to separate pipe provides an acoustically transparent window through
the variables of interest. Solutions may include using addi- which the interrogating beam gains access to the liquid in
tional cross-sections and modes such that three independent the pipe. The liquid is its own sensor. The pipe and suitable
ultrasonic measurements allow one to sort out the three transducer assemblies make it possible to utilize that sensor
Table 2. Examples of heating effects on the generation, propagation, and detection of ultrasound in water, in the laboratory test of
Figure 1.
Heating effects on:
◦
Temperature ( C) Generation Propagation Detection
Increase from 20 to 74 Frequency increases Speed of sound increases; absorption Signal arrives progressively
decreases earlier
75 to 99 Frequency increases Speed of sound decreases; absorption Signal arrives progressively
decreases later
100 Frequency increases Propagation is largely blocked by boiling Signal weakened by
and then by low-impedance air after all scattering due to boiling
the water boils away and may be undetectable
Ultrasonic Instrumentation Principles 807
A B
Thickness Transverse Radial
shear shear
(a)
C D
Figure 6. Liquid level obtained noninvasively from an up-
ward-looking pulse-echo transducer, together with four side-
Plastic Plastic Plastic
wall-mounted through-transmission transducers to compensate for
wear plate wedge lens
sound speed gradients as well as to have multiple discrete mea- (b)
suring heights (two in this example). Transducers B and D can be
deleted if transducers A and C operate in pulse-echo mode, like Figure 7. Examples of: (a) thickness mode, transverse shear
transducer E. (Reproduced by permission of Panametrics, Inc.) (often called thickness shear), and radial mode crystals and
(b) epoxy-sealed NDT transducers. (Reproduced by permission
without invading or physically penetrating the boundaries of Panametrics, Inc.)
of the fluid (see Figure 6).
in hot water near saturation conditions are often interrupted
by ‘flashing’ or boiling because of the severe mismatch in
6 LIMITATIONS ON PRECISION AND acoustic impedance between liquid and gas phases.
ACCURACY Another limitation in ultrasonic measuring systems is
spatial resolution. Apart from speckle and wavelength-
6.1 Limitations on coupling, propagation, spatial dependent limits, the propagation characteristics of the
resolution, and correlations among medium between the transducer and the region under study
propagation and measurands influence resolution in that region. As in optical systems,
diffraction limits the lateral resolution. Depth resolution is
In the field of ultrasonic nondestructive testing (NDT), usually limited by the time duration of the interrogating
a variety of couplants – liquids, oils, pastes, gels, and pulse, or its bandwidth. Refraction (Snell’s law) limits the
elastomers – are available for transmitting the interrogat- angle that a beam can be aimed relative to an interface.
ing wave from the transducer assembly to the part being This means noninvasive interrogations may be unable to
tested (see Figure 7). The couplant tends to match acoustic examine all areas inside a bounding pipe or vessel.
impedances as it replaces the low-impedance air gap that Yet another limitation applies to the attempts to general-
would otherwise block most of the transmission between ize the correlations between variables such as, say, strength
a solid transducer and a solid test object. At temperature and the speed of sound, molecular weight and the speed of
extremes, the choice of couplants is greatly diminished. sound, or Young’s modulus and the speed of sound. Such
Accordingly, in cryogenic or high-temperature situations, correlations may be quite useful when judiciously applied to
one may resort to pressure coupling (sometimes aided by a a limited class of materials, or to specimens of similar alloys
compliant film of metal) or to solid bonds of epoxy, solder, of like cross-section. One must be alert, however, to avoid
or ceramic cements. Coupling to liquid metals is sometimes errors introduced by unwanted variables, such as moisture,
facilitated by coating the probe so that it is wetted but not temperature, pressure, boundary dimensions comparable to
dissolved by the liquid metal. wavelength, and dispersion.
Temperature extremes may be associated with severe
sound speed gradients or other inhomogeneous conditions
that block propagation. On the one hand, ultrasonic mea- REFERENCES
surements have been made in the laboratory from nearly
absolute zero to plasma temperatures of 20 000 K. On the Asher, R.C. (1997) Ultrasonic Sensors for Chemical Process
other hand, practical measurements in cryofluids or merely Plant, Institute of Physics Publishing, London.
808 Elements: A – Sensors
Auld, B.A. (1990) Acoustic Fields and Waves in Solids, 2nd edn, FURTHER READING
Krieger Publishing, Malabar Florida.
Baker, R.C. (2000) Flow Measurement Handbook, Cambridge Andreatch, P. Jr. and Thurston, R.N. (1957) Disk-Loaded Tor-
University Press, Cambridge. sional-Wave Delay Line. Journal of the Acoustical Society of
Brown, A. and Lynnworth, L.C. (2001) Ultrasonic Flowme- America, 29(1), 16–19.
ters, Chapter 20, in Flow Measurement – Practical Guides Ao, X. (2003) Clamp-On Steam/Gas Flow Meter , U.S. Patent
for Measurement Control, 2nd edn (ed D.W. Spitzer), ISA 6,626,049.
(pp. 515–573).
Apfel, J.H. (1962) Acoustic Thermometry. Review of Scientific
Hunt, W.D., Stubbs, D.D. and Lee, S.-H. (2003) Time-Dependent Instruments, 33(4), 428–430.
Signatures of Acoustic Wave Biosensors. Proceedings of the
IEEE, 91(6), 890–901. Araki, H. and Matsunaga, Y. (1977) Ultrasonic Flow Meter , U.S.
Patent 4,014,21.
Kim, J.O. and Bau, H.H. (1989) On Line, Real-time Densimeter-
theory and Optimization. Journal of the Acoustical Society of Bell, J.F.W. (1957) The Velocity of Sound in Metals at High
America, 85(1), 432–439. Temperatures. Philosophical Magazine, 2, 1113–1120.
Kleppe, J.A. (1989) Engineering Applications of Acoustics, Artech Bohn, D.A. (1988) Environmental Effects on the Speed of Sound.
House, Boston, MA. Journal of the Audio Engineering Society, 36(4), 1–9.
Lower, N.P., Brow, R.K. and Kurkjian, C.R. (2004) Inert Failure Frederick, J.R. (1948) Ultrasonic Measurement of the Elastic
Strains of Sodium Aluminosilicate Glass Fibers. Journal of Properties of Polycrystalline Materials at High and Low Tem-
Non-crystalline Solids, 344, 17–21. peratures (Abstr.). Journal of the Acoustical Society of America,
Lynnworth, L.C. (1989) Ultrasonic Measurements for Process 20, 586.
Control, Theory, Techniques, Applications, Academic Press. Heijnsdijk, A.M. and Klooster, J.M.van P. (2002) Ultrasonic
Lynnworth, L.C., Cohen, R. and Nguyen, T.H. (2004) Clamp-On Waveguide, U.S. Patent 6,400 (p. 648).
Shear Transducers Simplify Torsional and Extensional Investi- Hill, J. and Weber, A. (2002) Qualification of Ultrasonic Flowme-
gations, Proceedings Ultrasonics Symposium IEEE, Montreal, ters of Natural Gas Using Atmospheric Air Calibration Facil-
August 2004. ities, in Proceedings of the 20th North Sea Flow Measurement
Lynnworth, L.C. and Mágori, V. (1999) Industrial Process Con- Workshop, St. Andrews.
trol Sensors and Systems, Chapter 4, in Ultrasonic Instru- Hurd, J.C., Nguyen, T.H. and Lynnworth, L.C. (2004) Shear
ments and Devices: Reference for Modern Instrumentation, Wave Polarization Follows Twist of Rectangular Steel Bar.
Techniques, and Technology (E.P. Papadakis Guest Editor), Materials Evaluation, 62(1), 37–42.
23 in the Series Physical Acoustics, Academic Press
(pp. 275–470). Jen, C.K. and Legoux, J.G. (1998) Clad Ultrasonic Waveguides
with Reduced Trailing Echoes, U.S. Patent 5,828,274.
Matson, J., Mariano, C.F., Khrakovsky, O. and Lynnworth, L.C.
(2002) Ultrasonic Mass Flowmeters Using Clamp-On or Wetted Khuri-Yakub, B.T., Kim, J.H., Chou, C.-H., Parent, P. and Kino,
Transducers, in Proceedings of the 5th International Symposium G.S. (1988) A New Design for Air Transducers, in 1988
on Fluid Flow Measurement . Ultrasonics Symposium Proceedings, IEEE (pp. 503–506).
Nguyen, T.H., Smart, C.D. and Lynnworth, L.C. (2004) Trans- Lemmon, E.W., McLinden, M.O. and Friend, D.G. (2000) Ther-
verse Shear to Torsion Mode Converter and Applications. mophysical Properties of Fluid Systems, in NIST Chemistry
Materials Evaluation, 62(6), 690–698. WebBook, NIST Standard Reference Database Number 69
(eds W.G. Mallard and P.J. Linstrom), National Institute of
Oralkan, Ö., Sanli Ergun, A., Johnson, J.A., Demirci, U., Kara-
Standards and Technology, Gaithersburg, MD 20899. URL:
man, M., Kaviani, K., Lee, T.H. and Khuri-Yakub, B.T. (2002)
http://webbook.nist.gov.
Capacitive Micromachined Ultrasonic Transducers: Next-
Generation Arrays for Acoustic Imaging? IEEE Transactions Liu, Y., Lynnworth, L.C. and Zimmerman, M.A. (1998) Buffer
on UFFC, 49(11), 1596–1610. Waveguides for Flow Measurement in Hot Fluids. Ultrasonics,
Royer, D. and Dieulesaint, E. (1999) Elastic Waves in Solids II, 36(1–5), 305–315.
Springer. Lynnworth, L.C. (1979) Ultrasonic Flowmeters, Chapter 5,
Shepard, C.L., Burghard, B.J., Friesel, M.A., Hildebrand, B.P., in Physical Acoustics – Principles and Methods, Vol. 14
Moua, X., Diaz, A.A. and Enderlin, C.W. (1999) Measure- (eds W.P. Mason and R.N. Thurston), Academic Press
ments of Density and Viscosity of One- and Two-phase Flu- (pp. 407–525).
ids with Torsional Waveguides. IEEE Trans UFFC, 46(3), Lynnworth, L.C. (1992) Marginally Dispersive Ultrasonic Waveg-
536–548. uides, U.S. Patent 5,159,838.
Straty, G.C. and Younglove, B.A. (1973) Velocity of Sound in Lynnworth, L.C. (2000) Ultrasonic Buffer/Waveguide, U.S. Patent
Saturated and Compressed Fluid Oxygen. Journal of Chemical 6,047,602.
Thermodynamics, 5(5), 305–312.
Lynnworth, L.C. and Liu, Y. (1999) Ultrasonic Path Bundle and
Youngdahl, C.A. and Ellingson, W.A. (1982) Acoustic System Systems, U.S. Patent 5,962,790; (2002) 6,343,511.
for Monitoring Pressure Boundary Wear, Presented at 1982
Symposium on Instrumentation and Control for Fossil Energy Lynnworth, L.C., Jossinet, G. and Chérifi, E. (1996) 300 ◦ C
Processes, Houston, TX. Clamp-On Ultrasonic Transducers for Measuring Water Flow
Ultrasonic Instrumentation Principles 809
and Level, in 1996 Ultrasonics Symposium Proceedings, IEEE Thurston, R.N. (1978) Elastic Waves in Rods and Clad Rods.
(pp. 407–412). Journal of the Acoustical Society of America, 64(1), 1–37.
Nygaard, O.G.H. and Mylvaganam, K.S. (1993) Ultrasonic Ting, V.C. and Ao, X. (2002) Evaluation of Clamp-On Ultra-
Time – Domain Reflectometry for Level Measurement in sonic Gas Flowmeters for Natural Gas Applications, in Pro-
Molten Metals. Technisches Messen 60(1), 4–14. ceedings of the 20th North Sea Flow Measurement Workshop,
St. Andrews.
Sherratt, G.G. and Griffiths, E. (1934) The Determination of the Weast, R.C. (1989) Handbook of Chemistry and Physics, CRC
Specific Heat of Gases at High Temperatures by the Sound Press (pp. E-46, F-8).
Velocity Method, I-Carbon Monoxide. Proceedings of the Royal
Society (London), Series A, 147, 292–308. Youngdahl, C.A. and Ellingson, W.A. (1978) Development of
Ultrasonic Techniques for Remote Monitoring of Erosive Wear
Szabo, T.L. (2004) Diagnostic Ultrasound Imaging – Inside Out, in Coal-Conversion Systems, in 1978 Ultrasonics Symposium
Academic Press. Proceedings, IEEE (pp. 305–310).
120: Ultrasonic Instrumentation Design
Lawrence C. Lynnworth
Lynnworth Technical Services, Waltham, MA, USA; Formerly Chief Technologist at GE Panametrics
likely to be a compromise that nevertheless benefits from
1 Selection and Influence of New the designer’s judicious choice of transducer material and
Electroacoustic Materials, Components, mode, methods of data acquisition, signal processing, dis-
and Techniques 810 play features, output functions, and so on. The total ultra-
2 Special Considerations for Temperature and sound design task includes many factors (Figure 2). Design
Pressure Extremes 812 principles underlying a diagram of this type are discussed in
3 Ultrasonic Transducers for Gases 812 books such as those by Asher (1997), Auld (1990), Baker
4 Ultrasonic Transducers for Liquids – Wetted, (2000), Lynnworth (1989), Royer and Dieulesaint (2000),
Clamp-on, Hybrid 812 Rogers (1984), or others. In this section, we deal only with
5 Ultrasonic Transducers for Nondestructive ‘air’ transducers, admittedly in an overview fashion.
Testing and Evaluation of Solid Materials 816 It will be understood that the relatively low-acoustic
References 819 impedance of air (compared to water or common engineer-
ing metals) has been a major barrier to ultrasonically testing
Further Reading 820
and measuring materials in air, or even testing air itself,
that is, measuring air flow or air temperature. Low-mass
membranes and low acoustic–impedance electroacoustic
1 SELECTION AND INFLUENCE OF NEW materials are appropriate for ultrasonic measurements of
ELECTROACOUSTIC MATERIALS, air, gases, or solids tested by air-coupled transducers. NDT
COMPONENTS, AND TECHNIQUES ‘immersion transducers’, developed many years ago for
testing solids immersed in water, are sometimes used for
The pie chart of Figure 1 categorizes ultrasonic measure- biomedical applications in body tissue, technical applica-
ment or instrumentation applications according to the four tions in liquids, and in some gases other than air, as well as
principal process control measurands (flow, temperature, air. In other words, sometimes one transducer can be used
pressure, and liquid level). Other slices express explicitly to test solids, liquids, or gases. Optimizing for a specific
or implicitly, combinations, NDT(nondestructive testing), application in one state of matter or another, however, may
and analyzer measurands (e.g. gas concentration, relative involve selecting or improving the electroacoustic material.
humidity, etc.). To keep this section of manageable size, let As far as impedance mismatch is concerned, bear in mind
us focus on a specific task. that the characteristic acoustic impedance of a gas, ZGAS ,
Consider the task of designing an ultrasonic system to equals the product of gas density times sound speed. This
measure without contact, that is, using air-coupled trans- means ZGAS increases roughly in proportion to pressure
ducers, several characteristics of a fast-moving solid. If the times the square root of gas MW (molecular weight).
measurement is important, there are likely to be competi- As a numerical example, the Z for argon at 10 bar is
tors, so the task requires a solution that is better, faster, approximately thirty times higher than Z for helium at
and cheaper than the competition’s. The best solution is atmospheric pressure.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Ultrasonic Instrumentation Design 811
Point Gas pressure
Line average Liquid pressure
Time-dependence Stress in a solid, e.g. in a bolt
Distribution in space Stress along axes of a crystal
Maximum or minimum Force, weight
Tension of musical string; of yarn;
of webbing, e.g. paper sheet
T P
Velocity V (m/s)
Volumetric Q (m3/s) Discrete
Mass flow rate (kg/s) Continuous
Liquids
Flow switches (V > Vsetpoint) Flow Level Solids
Flow rate per mode Foam
in multiphase flow Shape of interface
Flow transients (dV /dt )
Flow mapping [V (x, y, z )]
Swirl angle (ψ), circulation (Γ)
Cross-flow (Vx) Other
Combinations measurands
Energy flow (V & ∆T ) Composition, concentration
Flow velocity & level Thickness (x )
Flow and composition Range (distance)
Reynolds number (Re) Diameter (D )
where Re = rVD /h Viscosity (h)
Combine focussed ultrasound, Density (r)
acoustic streaming and range- Elastic moduli (E, G )
gated sampling of the streaming Poisson's ratio (s)
fluid
Other combinations derived
from sound speed and
attenuation, or from sensor
resonant frequency and
width of resonance peak;
combinations of acoustic
and nonacoustic sensors
Figure 1. Pie chart categorizes ultrasonic low-intensity sensing, measurement and control applications. (Reproduced by permission of
Panametrics, Inc.)
X proof (explosion proof) case
Damped for
broad bandwidth No dispersion (L or S waves)
Compensate for
buffer/transducer delays
Signal processing
Challenging Noise cancelation
medium, Cross-correl detection
Buffer harsh Buffer
Fixed or portable
Xdcr environment Xdcr
Transmitter Coupling Receiver Digital output
Measure sound
propagation,
1 or > 1 path, Display
Spike Visual
interpret c, a,
in terms of process IR (infrared) link
Tone bust
measurands Wireless to
Code remote site, if needed
(not available in 2002)
Continuous waves
Figure 2. Factors and choices to be considered when designing an ultrasonic measuring system. (Reproduced by permission of
Panametrics, Inc.)
812 Elements: A – Sensors
New electroacoustic materials, new manufacturing tech- Apart from low-mass membrane, ultrasonic transducers
niques [e.g. layered manufacturing, Safari et al. (2002)], and low acoustic–impedance transducers, which are appro-
piezo polymer composites, CMUTs (capacitive microma- priate for air, there are several housed piezoelectric disk
chined ultrasonic transducers), polyvinylidene fluoride and designs to be considered. These may be categorized as
copolymer piezoelectric sheet materials, and electrets com- unmatched and matched. Unmatched designs may encap-
prise areas that continue to be dynamic. sulate a resonator (thickness, radial, or flexural mode) in
an impervious housing. Matched designs, for which more
2 SPECIAL CONSIDERATIONS FOR details have been published, include a single λ/4 matcher
having an impedance as close as practical to the geomet-
TEMPERATURE AND PRESSURE ric mean of the impedances of the piezoelement and the
EXTREMES gas. Two-layer designs also exist (Figure 3). The simpler
one-layer design has proved adequate in over a thousand
Temperature extremes can cause failure of transducers
flare gas flowmeters installed in pipelines, refineries, and
in several ways: depolarization when the Curie point is
chemical plants between 1983 and 2003. This design has
exceeded; chemical reactions when hot; cracking or dis-
also been used in high-accuracy flowmeters for natural gas,
bonding due to differential thermal expansion; couplant
that is, Hill and Weber (2002).
boils away or becomes embrittled at hot or cold extremes;
Acoustic isolation of transmitter from receiver, to prevent
transducer material becomes electrically conductive when
cross talk, is often one of the most difficult design aspects of
hot; attenuation in the transducer backing or speed of sound
gas transducers used in pairs. Some approaches are illus-
in an adjacent wedge changes too much; soldered connec-
trated in Lynnworth (1989), or in Lynnworth and Mágori
tions melt. High-purity gas applications may require that
(1999) utilizing transducers mounted on membranes; inten-
during vacuum bakeout the transducer, if wetted, neither
tionally alternated impedance mismatches; small contact
outgases nor leaks. For use at high pressure, the trans-
areas; time delays or damping. If the same transducer is
ducer must neither deform nor leak. In a downhole tool
used as transmitter and receiver, ringdown often becomes
for exploring geothermal reserves, transducers may be sub-
the key problem as far as coherent noise is concerned. Sig-
jected simultaneously to high temperature and high pressure
nals indicative of the measurand cannot be detected until
(e.g. T = 100 to 350 ◦ C; P = 20 to 70 MPa). In testing red-
hot steel, buffer rods or momentary contact may be able coherent noise generated on transmission has subsided suf-
to isolate the transducer element from the high tempera- ficiently or unless it can be subtracted. Some examples of
ture of the product and the high pressure required for dry isolating or damping transducers, and damping cross talk
coupling. Lasers and emats (electromagnetic acoustic trans- or ringing, are shown in Figure 4. In manufacturing broad-
ducers) offer noncontact approaches that avoid couplants. band NDT transducers, the backing impedance matches that
Eventually, such methods may become routine in industry. of the piezoelement; it is intentionally attenuative; and may
have one or more antireflection surfaces (Papadakis and
Fowler, 1972).
3 ULTRASONIC TRANSDUCERS FOR
GASES
4 ULTRASONIC TRANSDUCERS FOR
In contrast to liquids or solids, gases are of low density
ρ, low speed of sound c, and consequently of low acous- LIQUIDS – WETTED, CLAMP-ON,
tic–impedance Z – see comparisons in Table 1. HYBRID
This means that if a high acoustic–impedance transducer
is used, it is much more difficult to transmit ultrasound into In this section, we introduce three categories of transducer
or out of a gas compared to liquids or solids. One generally designs for liquids, Figure 5. Principal characteristics are
seeks low-impedance transducers for gases. compared in Table 2.
Table 1. Comparison of density and two ultrasonic propagation characteristics of ultrasound for
air, water, and stainless steel (longitudinal waves).
Medium (20◦ C, Density (ρ) Speed of sound (c) Acoustic impedance (Z)
105 Pa) (kg/m3 ) (m/s) (kg/m2 s) (Mrayls)
Air 1.29 343 450 0.00045
Water 998 1482 1.48 × 106 1.48
Stainless steel 316 7833 5760 45.1 × 106 45.1
Ultrasonic Instrumentation Design 813
Table 2. Categories of transducer designs for liquids.
Type Advantages Disadvantages
Wetted Beam path is controllable; used in the most Requires penetration of pressure boundary;
accurate ultrasonic methods; usable over wide requires isolation valve or empty pipe for
frequency range e.g. 0.1 < f < 10 MHz. May installing and removing; cavities distort flow.
comprise part of replaceable module, e.g.
avionic flowmeter.
Nonwetted clamp-on, Noninvasive, removable unless bonded; easy to Oblique beam path depends on liquid; tendency
strap-on, snap-on, retrofit, no cavities. In many cases, it would for acoustic short-circuit noise when conduit
epoxy-on, braze-on, appear to be the safest way to measure flow in wall is thick; usually f < 2 MHz. Limited
weld-on the sense that the pipe integrity is not usually to liquids; sometimes usable with gases
jeopardized. at high pressure, or gases at atmospheric
pressure in special circumstances only.
Hybrid Combines advantages of wetted and nonwetted Requires initial penetration of pressure boundary
approaches. In some cases, the ‘window’ can be if window is installed as a retrofit; introduces
welded, molded, or cast into a valve or pipe cavities.
fitting to accommodate the occasional or
permanent attachment of the transducer.
Wetted designs, Figure 5(a), typically employ piezoele- independently, as a remedy to the occasional nuisance
ments resonant near 0.5, 1, or 2 MHz and may be narrow- and sometimes intolerable errors caused by uncertainty or
band or broadband depending on the backing impedance drifting of the obliquely refracted beam launched in the
and on the window thickness. Low-impedance backings liquid by a clamp-on angle beam transducer. A hybrid
and thin windows lead to narrowband waveforms. Back- transducer consists of a flat-faced plug that can permanently
ings that match the element’s impedance and thick metal maintain the pressure boundary while defining one terminus
windows lead to broadband waveforms. But ringing in a of a known fixed measuring path normal to the wetted
thick window can compromise the broadband response. face of the plug, and a removable nonwetted transducer
Tapered pipe–threaded housings can be rated up to 20 Mpa that repeatedly mates with the said plug. A hybrid flowcell
(3000 psi). Flanged designs, however, are usually pre- may be taken to mean a spool with transducers that do
ferred for high pressure, depending in part on the fluid and not comprise part of the pressure boundary. These worked
the industry. with liquids first and later with gases. A few designs are
Nonwetted external or clamp-on types, Figure 5(b), are collected in Figure 6.
designed to fit in standard yokes or tracks. Sometimes In a round conduit of area A, if the flow profile were
they are permanently attached or integral with the spool. known precisely, then measurements at a point or along a
Tracks are available with scales so that the axial separation chord could be readily converted to the area-averaged flow
between transducers, as required in contrapropagation or VAVG , from which the volumetric flow rate Q could be
tag flow measurements, can be set easily. For large pipes, calculated as AVAVG . For example, if the flow is steady and
magnets or wrap-around straps are commonly used to hold laminar, the profile is parabolic and a velocity measurement
the yokes, tracks, or transducers. For small pipes, below on the axis, VAXIS , would be exactly twice VAVG . Along
50-mm inside diameter, one or more transducers may be the midradius chord the chordal average equals VAVG . A
fixed temporarily or permanently within a given clamp- measurement along the tilted diameter, VDIAM , is higher
on or snap-on assembly. Such dual-transducer flowmeter than VAVG by 33.3% but is corrected by a meter factor
assemblies, for example, became available in the 1980s K of value 0.750 in this instance. In other words, Q =
from Transonic Systems for soft biomedical tubing having KAV DIAM = 0.750AV DIAM .
internal diameters down to 1 mm, and from Panametrics in For turbulent flow, if pipe relative roughness ∈R is
1990 for metal tubing from 10- to 50-mm inside diameter. known, models exist for estimating the flow profile as
The two principal liquid process control uses of external a function of the Reynolds number Re. At Re = 106 , if
nonwetted transducers have been to measure (i) liquid level, the pipe is smooth, the diameter path yields VDIAM about
and (ii) flow in pipes. The same or similar transducers can 5% higher than VAVG , which is routinely corrected using
often be used to measure thickness and integrity of pipes K = 0.95, in this instance. The correction generally is
and pressure vessels. slightly incorrect because of (i) uncertainty in ∈R and its
Hybrid flowmeter transducers, Figure 5(d, e) evolved influence, and (ii) uncertainty in the effective beam diam-
first as a convenience for removability and second, and eter. Multipath flowmeters typically include two or more
814 Elements: A – Sensors
3 4
2
1 5
Nonoptimum first
matching layer
9 mm
Air
0.25 mm
Best second
layer
Piezoetectric
Straight element
(b) (d)
90°
45°
(a) (c) (e)
(2) (3)
Air &
ath
Air & other
ue dard
ep
q3 other gases,
ve
gases
ob Stan
with
Water q3.AIR some
liq
q3 ~ 25° cross-flow
Steel pipe Long straight Steel pipe
(1) Stainless (4) steel pipe (5)
steel pipe
Clamp-on air
LOX transducer as
(liquid in (c).
oxygen)
q3 ~ 16°
(6)
(7)
(f)
Ultrasonic Instrumentation Design 815
1977 1984 1996
7 7
8 14 9 8
5 29
6 4
Y
3
1 21
14 20 12 22
10 X 3
2 25 SIGNAL
PROCESSOR
T T
OSC
13 23 17
0 19
FLOW AXIS
11
27 9 S/G FLOW
11
16 2
X 15 12
12 13 R R
N
1 15 7
5
10 21
S
Y 6 23
4 5
9 14
Lynnworth, L. C., Ultrasonic Measuring System Sheen, S-H. and Raptis, A. C., Acoustic
with Isolation Means, U.S. Patent 4,004,461 Rogers, S. C., Tunable Damper for An Acoustic Cross-Correlation Flowmeter for Solid Gas
(Jan. 25, 1977). Wave Guide, U.S. Patent 4,452,334 (1984). Flow, U.S. Patent 4,598,593 (July 8,1996)
1994 1996 2000
12
30 51 53 53 51 40
L 50
11
10
13
40 A 40 C
37
37 A 37 C
38 38 39
39
37 B 40
37 D
Baumoel, D. S. and Baumoel, J., Pipe Spool
40 B 40 D
Section Having Square or Rectangular Cross-
Lynnworth, L. C., Ultrasonic Transducer System with Itoh, I., Masanori, H. and Akio, Y., Ultrasonic Section for Clamp on Transducer and Method
Crosstalk Isolation, U.S. Patent 5,275,060 (Jan. 4, Fluid Vibrating Flowmeter, U.S. Patent 5,503,035 for Flow Measurement, U.S. Patent 6,026,693
1994). (Apr. 2, 1996). (Feb. 22, 2000).
Figure 4. Isolation and damping methods that have been used when the object is to measure a gas, liquid, or solid medium. (Reproduced
by permission of Panametrics, Inc.)
paths off the diameter so that a quadrature integration or as a function of distance from the wall. Then the cor-
other weighting procedure can obtain VAVG substantially responding annular areas are multiplied by their average
independent of the profile. For small pipes, the entire duct velocities and the products are integrated to yield Q =
area can be insonified, as practiced in methods termed AV AVG (Takeda et al., 2000). To enhance the interaction
100% area-averaging. In some Doppler flowmeters, range between the interrogating wave and the fluid, and thereby
gating is employed, whereby measurements are obtained improve performance of Doppler or other scatter-based
Figure 3. Transducers for gases: (a) metallurgically sealed single-layer impedance-matched designs for flare gases according to a patent
by Lynnworth, Fowler and Patch (1981); legend: 1 thin window, 2 quarter wave impedance matcher, 3 piezo element, 4 potting/backing
material, 5 leadwires, (b) two-layer unencapsulated design for air, after Khuri-Yakub et al. (1988), (c) clamp-on air transducer, after Ao
(2003); see Ting and Ao (2002), (d) gas flowmeter using quadrature integration of flow data obtained in three parallel planes, courtesy
of RMG; see Hill and Weber (2002), (e) gas transducer similar to straight one shown in (a) and usable in air at atmospheric pressure
and methane up to 100 bar, and (f) comparison of clamp-on paths. Path angles are calculated assuming a refracted angle θ2 of 60◦
for a shear wave in the steel pipe. Refracted angle depends on sound speed c3 in the fluid. Referring to items (1)–(3) commercial
contrapropagation clamp-on flowmeters available since the early 1990s include Panametrics’ PT868 and 6068 for measuring the flow
of liquids. In water, the vee path usually works, and θ3 is about 25◦ at room temperature. In air, θ3.AIR is only about 6◦ and unless
pressure is high, the transducers usually need to be placed on opposite sides of the pipe. (5) For liquid clamp-on, the vee path (shown
in (1)) tends to cancel cross-flow as well as double the sensitivity to flow compared to a single traverse. For gases, odd numbers of
traverses are preferred, to reduce cross talk. This means, if cross-flow is significant, crossed paths are recommended. The velocities
measured along the legs of the X should be averaged. Best solution: find a long straight run far from disturbances and joints. For gases
(diagrams (4) & (5)) the flowmeter instrument (6) is Panametrics’ GC868 (Ting and Ao (2002)). Diagram (7), drawn for LOX (liquid
oxygen) or LN2 (liquid nitrogen), shows the refracted angle θ3 ≈ 16◦ , nearly midway between water (25◦ ) and air (6◦ ). (Reproduced
by permission of Panametrics, Inc.)
816 Elements: A – Sensors
(a) (b)
(c) (d)
(e)
Figure 5. Transducers for liquids: (a) wetted, (b) clamp-on, (c) snap-on, (d) hybrid for ordinary temperature, and (e) hybrid for
temperature extremes, using a buffer consisting of a rigid bundle of thin rods within a sealed tube. (Reproduced by permission of
Panametrics, Inc.)
flowmeters in cases in which the fluid does not already con- 5 ULTRASONIC TRANSDUCERS FOR
tain sufficient scatterers, contrast agents are added. Besides NONDESTRUCTIVE TESTING AND
physical contrast agents, investigators sometimes use cav-
itation bubbles (Takeda et al., 2000). In Lynnworth, Carey
EVALUATION OF SOLID MATERIALS
and Pedersen (1974), it was suggested that a circumferential
Although the markets and the manufacturers may differ
external ultrasonic cylindrical transducer, when energized,
according to one’s interpretation of ultrasonic instrumen-
could generate a hot tag near the axis. tation (i.e. ‘analytic’ instrumentation, ‘process’ instrumen-
Another way to improve flowmeter performance is to tation, ‘NDE’(nondestructive evaluation)/NDT instrumen-
condition the flow to eliminate uncertainty in profile. Flow tation), one would expect that the physics underlying
conditioners (Miller, 1996, Chap. 5; Gallagher, 2002) usu- the wave/measurand interaction would largely determine
ally reduce cross-flow and swirl, but in principle they could whether the design of transducers for such applications
accomplish their purpose of reducing profile uncertainty must differ or could be identical. In practice, many hand-
if they generated a predetermined pattern that includes a held transducers for manual inspection of manufactured
known asymmetry and swirl (Hill, 2001). In small ducts, parts reflect ergometric design considerations; size, weight,
flow conditioners have taken the form of static mixers, tube surface contours and finish. Other design considerations
bundles, concentric tubes comprising a Clausius-Mosotti include part accessibility, inspectability, transducer wear
dielectric – constant–based fuel densitometer, or blades resistance and longevity. Because of the wide range of
(Figure 6). materials, geometries, dimensions, microstructures, bonds,
Ultrasonic Instrumentation Design 817
1962 1973
8 4 1
5 Sealed diametrically-
opposed sight port
Liquid to see if bubbles are
oxygen present
10
9
12
6
Adapted from Pedersen, N. E., and Lynnworth, L. C., Nonintrusive Dynamic
Flowmeter, pp. 178–181 in: 1973 Ultrasonics Symposium Proceedings, IEEE
Lake, D.B., Flowmeters, U.S. Patent 3,050,997 (Aug. 28, 1962) (1973)
16
1974 1987 2000
14 15
DIFF. PRESS
TRANSDUCER Example for small pipe ... square meter body 10 42
12 20
40
CLAMP-ON 20 46
52 50
13 18
TRANSDUCER 42 51 43
19 51
Flow PIPE SQUARE PIPE 42 47
21 CAP TUBE CAP
37
4 49 42
9 811 19
6 36
5 7 17 ZIGZAG PATH
39
3 42
42
2 1 SIGNAL
22 33
PROCESSING
CIRCUITS 42 30
OUTPUT
SIGNAL
Source: Matson, J. E., Nguyen, T. H. and Baumoel; D. S. Baumoel. J., Pipe Spool Section
Lynnworth, L.C., Ultrasonic Measurement of Having Square or Rectangular Cross-Section for
Turtle, Q., Electroacoustical Flow Metering Liquid Flow Using Clamp-On Rayleigh Wave Clamp on Transducer and Method for Flow
Apparatus, U.S. Patent 3,788,140 (Jan. 29, Transducers, 197-206, IEEE Trans I & M Measurement, U.S. Patent 6,026,693 (Feb. 22,
1974) Conference Proceedings (April 27–29, 1987) 2000)
1977 1996 2000
Ultrasonic
transducer SS (stainless steel)
14 square-shaped tube
16b 16b Ultrasonic
16 16 24 transducer (1" square)
16a 16a Flow
Flow
16b
Lynnworth, L. C., Jossinet, G., and Chérifi, E.,
16b
22 16a 16a 12 300°C Clamp-On Ultrasonic Transducers for
1" NPTF
16 16 3000# SS cap Measuring Water Flow and Level, 1996 Ultrasonics
Symposium Proceedings, pp. 407–412, IEEE
Araki, H., and Yoshihioro, M., Ultrasonic Flow (1996); Lynnworth, L. C., Ultrasonic Buffer/
Meter, U.S. Patent 4,014,211 (March 29, 1977) © 1996 Panametrics, Inc. Waveguide, U.S. Patent 6,047,602 (April 11, 2000)
1975 1982
R Absorber 16
14
T
Orthogonal 32A 12 32B
beams
intersect
Inlet Flow Outlet
56
straighteners
Other flow conditioners
Pedersen, N. E., Lynnworth, L. C., and Bradshaw, J. E., Zacharias, E. M., Flow Straighteners in Axial Flowmeters, U.S. Patent
USAAMRDL-TR-75-8, page 178 (June 1975) 4,365,518 (Dec. 28, 1982)
1985 2001 2002
1
30 26 120
12 20
214 110
10 42 24 40 44 10
34 112 5
16 35 34 38 22
2 16
Fig. 2 34A P 3 4
Fig. 1 28 32
115 115 2
Zacharias, E. M., Sonic Flow Meter Gallagher, J. E., Method for Determining
Having Improved Flow Straighteners, Hill, J. A., Disturbance Simulating Flow Plate, Flow Velocity in a Channel,
U.S. Patent 4,523,478 (June 18, 1985) U.S. Patent 6,186,179 (Feb. 13, 2001) U.S. Patent 6,494,105 (Dec. 17, 2002)
Figure 6. Flowcells and interrogation methods for liquids, some of which were later applied or adapted to measuring gas flow.
Transducers are wetted in some cases and external (not wetted) in others. Flow conditioners were known at least since Turtle (1974);
see also, Miller (1996), Chapter 5. A Kenics static mixer, not shown, was used as a flow conditioner in 1974 in an R&D (research and
development) program and reported by Pedersen, Lynnworth and Bradshaw (1975). (Reproduced by permission of Panametrics, Inc.)
818 Elements: A – Sensors
quality levels and a market that has matured over the past in the 1 to 10-MHz decade. Others are available down to
half-century, a far greater variety of transducers exists for 20 kHz and up to 100 MHz using bondable electroacous-
NDT applications than for any other single process measur- tic elements, and up to the gigahertz range using deposited
and such as flow. Most ultrasonic NDT transducers operate piezoelectrics for acoustic microscopy, see Figure 7.
(a) (b) (c)
(d) (e) (f)
(g) (h)
Figure 7. NDT transducers: (a) contact, flat, (b) wetted, focused, (c) angled, (d) dual element for corrosion testing; emat electro-
magnetic acoustic transducer, (e) slidable magnetostrictive transducer for generating extensional and/or torsional waves, (f) shielded
magnetostrictive transducer for generating extensional or torsional waves, (g) emat (electromagnetic acoustic transducer), and (h) shear
wave piezo couple used to generate and detect torsional waves by a mode conversion process. The wide range of sizes and form factors
for ‘fixtures’ that position NDT transducers is exemplified by comparing (i) a miniature ‘clamp-on’ transducer that briefly clamps
to one’s finger; (j) a hand-held transducer in a penlike case versus (k) gantry for robotic ultrasonic scanning system including water
squirters for coupling and means to orient transducers normal to an airplane fuselage’s contour. Another aspect of form factor is the
shape or diameter of the specimen. If the specimen is a thin elastic rod, e.g. diameter in the several µm to mm range (l) the methods
of Bell (1957) or Fowler (in Lynnworth, Papadakis and Fowler, 1977) may be considered. Post-2000 versions of (l) adapted to moduli
measurements and related (derived) characteristics in thin glass fibers are found in Krohn et al. (2003). ( 2003 Panametrics, Inc. and
reproduced courtesy of GE Panametrics.)
Ultrasonic Instrumentation Design 819
(i) (j)
(k)
N S
f 76 - f 25 -
f 760 µm 25 µm 2.5 µm
Modulus
transducer
Magnetostrictive lead-in f 0.6 to 1 m 6.4 to
f 0.58 mm (580 µm) Quasi-conical drawn glass 254 mm
(l) by ~1 m long impedance transformer specimen
Figure 7. (Continued ).
REFERENCES Gallagher, J.E. (2002) Method for Determining Flow Velocity in a
Channel, U.S. Patent 6,494,105.
Hill, J.A. (2001) Disturbance Simulating Flow Plate, U.S. Patent
Ao, X. (2003) Clamp-On Steam/Gas Flow Meter, U.S. Patent
6,186,179.
6,626,049.
Hill, J. and Weber, A. (2002) Qualification of Ultrasonic Flowme-
Asher, R.C. (1997) Ultrasonic Sensors for Chemical Process
ters of Natural Gas Using Atmospheric Air Calibration Facil-
Plant, Institute of Physics Publishing, London.
ities, Proceedings of the 20 th North Sea Flow Measurement
Auld, B.A. (1990) Acoustic Fields and Waves in Solids, 2nd edn, Workshop, St. Andrews.
Krieger Publishing, Malabar Florida.
Khuri-Yakub, B.T., Kim, J.H., Chou, C.-H., Parent, P. and Kino,
Baker, R.C. (2000) Flow Measurement Handbook, Cambridge G.S. (1988) A New Design for Air Transducers, Proceedings
University Press. of the Ultrasonics Symposium, IEEE (pp. 503–506).
Bell, J.F.W. (1957) The Velocity of Sound in Metals at High Lynnworth, L.C. (1989) Ultrasonic Measurements for Process
Temperatures. Philosophical Magazine, 2, 1113–1120. Control, Theory, Techniques, Applications, Academic Press.
820 Elements: A – Sensors
Lynnworth, L.C. and Mágori, V. (1999) Industrial Process Con- Dabirikhah, H. and Turner, C.W. (1994) Leaky Plate Wave
trol Sensors and Systems, Chapter 4, in Ultrasonic Instruments Airborne Ultrasonic Transducer. Electronic Letters, 30(18),
and Devices: Reference for Modern Instrumentation, Tech- 1549–1550.
niques, and Technology (E.P. Papadakis Guest Editor), 23 in Estrada, H. (2001) Theory of Ultrasonic Flow Measurement –
the Series Physical Acoustics, Academic Press (pp. 275–470). Gases and Liquids, Caldon Technical Paper TP58, Presented
Lynnworth, L.C., Carey, C.A. and Pedersen, N. (1974) Non- at ISHM 2001 (International School of Hydrocarbon Measure-
intrusive (Noninterfering) Ultrasonic Techniques to Measure ment), Class 3175.
Gas Mass Flow Rates, AEDC-TR-74-77, Arnold Engineering Guizot, J.-L. (2003) Ultrasonic Liquid Flow Measurement, Caldon
Development Center, Tennessee (p. 19). Technical Paper TP68, Presented at ISHM 2003 (International
Lynnworth, L.C., Fowler, K.A. and Patch, D.R. (1981) Sealed, School of Hydrocarbon Measurement), Class 2405.
Matched Piezoelectric Transducer, U.S. Patent 4,297,607; Lyn-
Hurd, J.C., Nguyen, T.H. and Lynnworth, L.C. (2004) Shear
nworth, L.C., Patch, D.R. and Mellish, W.C. (1984) Impedance-
Wave Polarization Follows Twist of Rectangular Steel Bar.
Matched Metallurgically-Sealed Transducers. IEEE Trans Son-
Materials Evaluation, 62(1), 37–42.
ics and Ultras, SU-31(2), 101–104.
Itoh, I., Masanori, H. and Akio, Y. (1996) Ultrasonic Fluid
Lynnworth, L.C., Papadakis, E.P. and Fowler, K.A. (1977) Ultra-
Vibrating Flowmeter, U.S. Patent 5,503,035.
sound Propagation Measurements and Applications, in Inter-
national Advances in Nondestructive Testing, Vol. 5 (ed. Kleppe, J.A. (1989) Engineering Applications of Acoustics, Artech
W.J. McGonnagle), Gordon & Breach (pp. 77–115). House, Boston, MA.
Miller, R.W. (1996) Flow Measurement Engineering Handbook, Krohn, M.H., Hellman, J.R., Pantano, C.G., Lower, N. and
3rd edn, McGraw-Hill. Brow, R.K. (2003) Effects of Tin on the Physical Properties and
Papadakis, E.P. and Fowler, K.A. (1972) Broadband Transducers: Crack Growth in Soda-Lime Silica Float Glass. Proceedings
Radiation Field and Selected Applications. Journal of the of the 8th International Symposium on Fracture Mechanics of
Acoustical Society of America, 50,(3, Part 1), 729–745. Ceramics, Plenum Press, New York [Houston, Texas February
2003].
Pedersen, N.E., Lynnworth, L.C. and Bradshaw, J.E. (1975) US-
AAMRDL-TR-75-8 (p. 178). Lake, D.B. (1962) Flowmeters, U.S. Patent 3,050,997.
Rogers, S.C. (1984) Tunable Damper for an Acoustic Wave Guide, Lipták, B.G. (1995) Process Control, Instrument Engineers’
U.S. Patent 4,452,334. Handbook, 3rd edn, Vol. 2, ISA.
Royer, D. and Dieulesaint, E. (2000) Elastic Waves in Solids I, Lynnworth, L.C. (1977) Ultrasonic Measuring System with Isola-
Free and Guided Propagation, Springer. tion Means, U.S. Patent 4,004,461.
Safari, A., Ebrahimi, M., Toreu, S., Hall, A., Brenan, R. and Lynnworth, L.C. (1992) Ultrasonics in Instrumentation, Chap-
Hagh, N.M. (2002) Layered Manufacturing for Prototyping of ter 38, in Handbook of Measurement Science (eds P.H. Syden-
Novel Transducers, Proceedings of the International Ultrasonic ham and R. Thorn), Wiley (pp. 1655–1689).
Symposium, IEEE (pp. 1060–1068). Lynnworth, L.C. (1994) Ultrasonic Transducer System with Cross-
Takeda, Y., Furuichi, N., Mori, M., Aritomi, M. and Kikura, H. talk Isolation, U.S. Patent 5,275,060.
(2000) Development of a New Flow Metering System Using Lynnworth, L.C. (2000) Ultrasonic Buffer/Waveguide, U.S. Patent
UVP, Preliminary Performance Assessments Using NIST Flow 6,047,602.
Standards, Proceedings of the ASME FEDSM 2000 .
Lynnworth, L.C., Jossinet, G. and Chérifi, E. (1996) 300◦ C
Ting, V.C. and Ao, X. (2002) Evaluation of Clamp-On Ultrasonic Clamp-On Ultrasonic Transducers for Measuring Water Flow
Gas Flowmeters for Natural Gas Applications, Proceedings of and Level, 1996 Ultrasonics Symposium Proceedings, IEEE
the 20 th North Sea Flow Measurement Workshop, St. Andrews. (pp. 407–412).
Turtle, Q. (1974) Electroacoustical Flow Metering Apparatus, Matson, J.E., Nguyen, T.H. and Lynnworth, L.C. (1987) Ultra-
U.S. Patent 3,788,140. sonic Measurement of Liquid Flow Using clamp-On Rayleigh
Wave Transducers, IEEE Transactions I & M Conference Pro-
ceedings (pp. 197–206).
FURTHER READING Nguyen, T.H., Smart, C.D. and Lynnworth, L.C. (2004) Trans-
verse Shear to Torsion Mode Converter and Applications.
Baumoel, D.S. and Baumoel, J. (2000) Pipe Spool Section Having Materials Evaluation 62(6), 690–698.
Square or Rectangular Cross-Section for Clamp On Transducer
Oldenziel, D. and Greissmann, M. (1996) Clamp-On Ultrasonic
and Method for Flow Measurement, U.S. Patent 6,026,693.
Volumetric Flowmeter, U.S. Patent 5,533,408.
Brown, A. and Lynnworth, L.C. (2001) Ultrasonic Flowmeters,
Chapter 20, in Flow Measurement – Practical Guides for Pedersen, N.E. and Lynnworth, L.C. (1973) Nonintrusive Dyna-
Measurement Control, 2nd edn (ed. D.W. Spitzer), ISA mic Flowmeter, 1973 Ultrasonics Symposium Proceedings,
(pp. 515–573). IEEE (pp. 178–181).
Cousins, T. and Augenstein, D. (2002) Proving of Multi-Path Shah, K.C. (1986) Ultrasonic Temperature Sensor, U.S. Patent
Liquid Ultrasonic Flowmeters, Proceedings of the North Sea 4,610,551.
Flow Measurement Workshop, St. Andrews, 22–25 October Szabo, T.L. (2004) Diagnostic Ultrasound Imaging – Inside Out,
2002. Academic Press.
Ultrasonic Instrumentation Design 821
Wada, S., Kikura, H., Aritomi, M., Mori, M. and Takeda, Y. Zacharias Jr., E.M. (1982) Flow Straighteners in Axial Flowme-
(2004) Development of Pulse Ultrasonic Doppler Method for ters, U.S. Patent 4,365,518.
Flow Rate Measurement in Power Plant – Multilines Flow Rate
Measurement on Metal Pipe. Journal of Nuclear Science and Zacharias Jr., E.M. (1985) Sonic Flow Meter Having Improved
Technology 41(3), 339–346. Flow Straighteners, U.S. Patent 4,523,478.
121: Signals in the Presence of Noise
Richard Burdett
Signal Recovery, Wokingham, UK
It will be found that the techniques, originally conceived
1 Signal Extraction When Noise is Present 827 for analog methods of signal processing, are often now
2 Basic Methods for Reducing Noise 827 implemented in digital form, and thus, in many cases
3 Nature of Noise 828 provide improved performance at comparable or lower cost.
Parts of the systems are, however, necessarily analog and
4 White Noise 829
always will be so, since the natural world is not all digital
5 Flicker Noise 829 in its structure or form.
6 Importance of Bandwidth 829
7 Equivalent Noise Bandwidth 829
Related Articles 830 2 BASIC METHODS FOR REDUCING
References 830 NOISE
Recovering or enhancing a signal, or improving a signal-
to-noise ratio (SNR) simply means reducing the noise
1 SIGNAL EXTRACTION WHEN NOISE
accompanying a signal.
IS PRESENT There are two basic ways of doing this:
A measurement system begins with sensing stages that 1. Bandwidth reduction, where the noise is reduced
couple to relevant measurands of the system under study. by reducing the system noise bandwidth (Bn ). This
The power level of the information-bearing signals formed approach works well if the frequency spectra of the
by the sensors is often very low and may be swamped by the noise and signal do not overlap significantly, so that
unwanted noise signals that are present. Careful attention to reducing the noise bandwidth does not affect the
sensor and circuit design and assembly, plus use of certain signal. With random white noise, the output noise
signal processing methods, makes it possible to greatly √
is proportional to Bn . With nonwhite noise, other
enhance the original signal-to-noise ratio to usable levels. relationships will apply.
This group of articles discusses the various strategies that 2. Averaging or integrating techniques, where successive
are available and provides a basis for their use. This is samples of the signal are synchronized and added
not a widely published topic, the best information proba- together. The signal will grow as the number (n) of
bly being available in the literature of companies marketing added samples; with random white noise, the noise
such products. It is surprising to find that, after well over √
will grow as n. This is only the case, if the signal
half a century of signal processing progress, there exist characteristics are stationary for the duration of the
so few full length texts that deal generally with signal extraction process.
recovery and enhancement in general instrumentation appli-
cations, see Kester (2002), and the classic work Vainshtein Sometimes it is useful to combine both techniques. In
(1985). many applications there is significant overlap between the
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
828 Elements: B – Signal Conditioning
signal and noise spectra and improving a signal-to-noise the instrumentation techniques used to reduce the remaining
ratio must be done at the expense of the response time noise content. Finally, special considerations involved in
or measurement time (T ); with random white noise inter- recovering pulse signals from photon (light), ion, or elec-
ference, the output signal-to-noise ratio is proportional to tron beams are covered in this section Part 12, Ele-
√
T . The bandwidth reduction technique is best looked ments: F – Signals and Noise, Volume 3.
at from a frequency-domain point of view; signal aver-
aging and correlation techniques lend themselves to time-
domain analysis. 3 NATURE OF NOISE
In this set of articles, – Part 12, Elements: F – Signals
and Noise, Volume 3 – mathematics, and theoretical con- Noise is an undesired signal. All systems have it present
siderations are kept to a minimum and the models presented to some degree. It usually becomes of interest when it
are basic. Practical best practice requires considerably more obscures a desired signal. Figure 1 shows the power spec-
processing to precondition signals than is shown here. For tral density (power/unit bandwidth) of the most commonly
further simplicity, it is assumed that all noise processes are encountered types of noise.
stationary and that both signal and noise are ergodic, analog Deterministic noise can range from simple discrete-
variables; we will not concern ourselves here with digi- frequency components such as power-line hum at har-
tal signals or discrete-time (sampled) signals except where monics of 50 or 60 Hz, to radio frequency interference,
such signals are involved in the enhancement techniques. RFI, caused by narrow, high-energy pulses from power-
They are essential in modern application methods but it is line switching spikes, pulsed lasers, radar transmitters, and
the basic ideas that drive the digital methods. the like.
In addition, only signal recovery techniques will be con- Stochastic or random noise is found in most systems
sidered. Further processing, such as least-squares polyno- both as white noise, where the power spectral density is
mial smoothing of a waveform or Fourier transformation to independent of frequency, and also as 1/f or flicker noise,
obtain a frequency spectrum, are not considered here. where the power spectral density decreases as frequency
Discussion is started by reviewing some basic concepts, increases. Power spectral density is usually measured in
then moving on to ways of avoiding adding noise (e.g. mean-squared-volts/Hz or mean-squared-amperes/Hz; for
hum pickup and preamplifier noise) and finally covering noise, such specifications are usually referred to as spot
Year−1
Power line
Change of classes,
work shifts, etc 50/60 Hz
Lifts, 150/180 Hz
106 Day−1
elevators Switched mode
Power/unit bandwidth (Arbitrary units)
PSUs PC monitors
100/120 Hz
Hour−1
AM
Temperature radio Analog
104 Min−1 TV
Typical RFI
102 frequency
envelope
1/f Noise
1
White Noise
10−8 10−6 10−4 10−2 1 102 104 106 108
Frequency (Hz)
Figure 1. Environmental noise. (Reproduced by permission of SIGNAL RECOVERY, a part of AMETEK Advanced Measurement
Technology.)
Signals in the Presence of Noise 829
noise data and usually are a function of frequency. Notice What this model represents is that it is increasingly harder
that for an rms voltage of v (volts) and a frequency range to remove noise as the frequency is lowered below 1 Hz or
of f (Hz), the power spectral density, S, is given by so. This makes practical sense when it is considered that
2 there is a link between noise generation and temperature
v2 v cycling of mechanical systems. It becomes harder to inte-
S= = √ (1)
f (f ) grate temperature cycling noise as the cycle time lengthens.
√
The quantity v/ (f ) is usually referred to as voltage
√
spectral density and is measured in rms volts/ Hz (volts 6 IMPORTANCE OF BANDWIDTH
per root hertz). Similarly, we can refer to current spectral
√
density specifications in units of rms amperes/ Hz. What do we mean by bandwidth? In the simple low-pass
filter circuit shown in Figure 2, for example, we usu-
ally and somewhat arbitrarily define the signal bandwidth
4 WHITE NOISE (Figure 3) to be the cutoff frequency, fc , where vo /vi =
70.7% (−3 dB) or vo2 /vi2 = 50% (the half-power point).
White noise is usually found in one of two forms: Johnson
noise and shot noise. Johnson, or thermal, noise is caused by
random motion of thermally agitated electrons in resistive
7 EQUIVALENT NOISE BANDWIDTH
materials, and the mean-square noise voltage is given by
Notice that frequencies above fc will obviously pass
vn2 = 4kTRf (2)
(although attenuated) through the filter, and therefore are
not really cut off. For noise, it is convenient to think in
where k is Boltzmann’s constant (1.381 × 10−23 JK−1 ),
terms of an equivalent noise bandwidth, Bn , defined by the
T is the absolute temperature (kelvin) and relationship
R is the resistance (ohm). ∞
1
Alternatively, from Ohm’s law, the mean-square noise Bn = 2 |H (jω)|2 df (5)
G 0
current is given by
v 2 4kT f R
in2 = n
= (3)
R R
ni C no
Shot noise is caused by the random arrival of elec-
trons – see Article 184, Photon Counting, Volume 3 –
at, for example, the electrodes of electron tubes or tran-
sistor junctions. A DC current, I , will have a noise-current Figure 2. Low-pass filter circuit. (Reproduced by permission of
component, in , given by SIGNAL RECOVERY, a part of AMETEK Advanced Measure-
ment Technology.)
in2 = 2AeI f (4a)
Noise bandwidth, B n
where e is the charge of one electron (≈1.6 × 10−19 C), A
Signal bandwidth, fc
is the mean gain experienced by each electron and I is in
0 Slope = −6 dB/octave
amperes. In many cases, A = 1, so that −3 (−20 dB/decade)
in2 = 2eI f (4b) G (f )
(dB)
5 FLICKER NOISE
log(f )
fc Bn
Flicker noise has many different origins and is not clearly
understood but exhibits a 1/f n power spectrum with n Figure 3. Low-pass filter transfer characteristic. (Reproduced
usually in the range of 0.9 to 1.35. Note that DC drift is a by permission of SIGNAL RECOVERY, a part of AMETEK
very low frequency form of flicker noise. Advanced Measurement Technology.)
830 Elements: B – Signal Conditioning
where H (jω) is the frequency response function of the sys- RELATED ARTICLES
tem and G is a gain parameter suitably chosen to be a
measure of the response of the system to some parameter Article 125, Outline of Purpose of Analog Data
of the signal: for low-pass systems (e.g. Figures 2 and 3), Filters, Volume 2; Article 176, Signals and Signal-to-
G is usually taken to be the zero-frequency or DC gain. noise Ratio, Volume 3; Article 178, Noise Matching and
For band-pass responses, G is usually made equal to the Preamplifier Selection, Volume 3.
maximum gain.
Using the above definition, and taking G to be the zero-
frequency gain (i.e. unity), we can readily calculate that for REFERENCES
the simple RC filter shown in Figure 2 that
Fellgett, P.B. and Usher, M.J. (1980) Fluctuation Phenomena
1 in Instrument Science, Journal of Physics E: Scientific and
Bn = RC Hz (6) Instrumentation, 13, 104–106.
4
Kester, W. (2002) Mixed-signal and DSP Design Techniques,
Noise, of the stochastic form, has been reviewed in Engineering Staff of Analog Devices Inc., Newnes, London.
relation to instrument systems in a classic paper, Fellgett Vainshtein, L.A. (1985) Extraction of Signals from Noise, reprinted
and Usher (1980). from Dover Publications, Wokingham, UK, 1970.
This article is based in part on a chapter by D.M. Munro originally published in Handbook of Measurement Science Vol 1,
P.H. Sydenham, Wiley, 1982.
122: Operational Amplifiers
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
integrated circuit operational amplifier, abbreviated to IC
1 Introduction to Operational Amplifiers op-amp or simply op-amp.
(Op-amps) 831 It should also be noted that when the expression op-amp
2 Classification of Op-amps 832 is now used, it usually refers to the IC unit as purchased
3 Types of Semiconductor Amplifier Units from a manufacturer. Whenever the word amplifier is used,
used in Op-amps 832 it implies a circuit that employs an op-amp and some other
4 Open Loop Voltage Transfer Function 832 components to constitute the overall amplifier. Amplifiers
and the op-amps included in them, which are the basic
5 Op-amp Model 833
building blocks of almost all data handling measurement
6 Inverting Circuits 833 systems, have now reached a very high degree of sophisti-
7 Noninverting Circuits 834 cation. The detailed internal design of op-amps will not be
8 Frequency Domain Factors and Behavior 836 considered here; it is very specialized.
9 Offset Voltage, Bias, and Noise in Op-amps 837 Op-amps are directly coupled amplifiers having, for rea-
10 Equivalent Circuit for Real Op-amps 838 sons that will be discussed later, a high gain of the order of
Related Articles 839 >104 . In their own right, they are highly reliable electronic
systems. In addition, they bring the benefits of compact-
Further Reading 839
ness, versatility, and ease of use, which make them essential
electronic building blocks. This means they can make a
substantial contribution to increasing measuring systems
1 INTRODUCTION TO OPERATIONAL sensitivity while exhibiting versatility in the design of the
AMPLIFIERS (OP-AMPS) frequency characteristics of the systems in which they
are employed.
The voltage levels available from passive sensor condi- In many such industrial systems, the measuring points
tioners are usually too small to be practically useful; it is are fairly remote from the data processing location. Op-
necessary to amplify the signal. Since the development of amps make a considerable contribution to the effective
the triode valve by Lee de Forest early in the twentieth cen- transmission of these remotely acquired signals. They con-
tury, many other significant developments in amplification sume comparatively little power. Consequently, it can be
have been achieved. Notable milestones were the appear- concluded that the versatility brought to measuring systems
ance of the point contact transistor, the bipolar junction by electronic means is largely due to the impact of the
transistor (BJT), the field-effect transistor (FET) or junction op-amp.
field-effect transistor (JFET) and the metal oxide semicon- The continuing steadily improving performance of op-
ductor (MOSFET). amps since their first introduction around 1960 has been
From these developments in semiconductor technology, accompanied by significant reduction in their cost. Conse-
rapid evolution of solid-state amplifiers has resulted in the quently, op-amps now give measuring system designers a
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
832 Elements: B – Signal Conditioning
versatile building block component for inclusion in numer- Field-effect transistor, or FET, differential op-amp:
ous applications to realize many diverse and complex func- When it is important to minimize input bias current, it
tions. The versatility of op-amps lies in their application is usual to design an op-amp using a complementary pair
using associated feedback networks. Thus, op-amps can be of input FETs. This type of input also increases the input
configured to act as summers, subtracters, and difference impedance of the op-amp. Unfortunately, because the bias
amplifiers. They can also perform the functions of differ- current essentially depends on the reverse saturation current
entiation, integration, and of other types of filters, or as of the gate-to-source junction, its bias current drift is much
comparators. This concept has eventually resulted in the larger than for a bipolar input unit. Their potential for
universal application of the op-amp for a variety of uses, high slew rate and fast settling time is accompanied by
as is obvious from its universality in contemporary measur- capacitive pickup of noise signals due to the high input
ing systems. As there is a wide diversity of available types impedance. Some cases are also known where mechanical
of IC, the task of choosing a suitable op-amp for a particu- vibration causes problems if they introduce movement into
lar application is not difficult but requires some knowledge the wiring. They find use in general-purpose applications,
of their nature, scope, characteristics, and formal modeling. in differential amplifiers, low noise amplifiers from high
impedance sources and sample-and-hold amplifiers and
integrators, because of their low bias current.
2 CLASSIFICATION OF OP-AMPS FET op-amps may use either junction FETs or MOS-
FETs. They require a bias current in the range of 0.01 to
No single op-amp type meets the needs of every application. 100 pA at 25 ◦ C, which doubles with every 10 ◦ C increase in
All practically manufactured forms constitute a compromise temperature. Their drift voltage is around 2 to 100 µV/ ◦ C.
among the various aspects and features required. Hence, Open loop gain is 104 to 5 × 105 . Because their differential
the design of op-amps is an optimizing process. To give input resistance is very high, at around 1011 to 1012 , they
a good overall view, the three main op-amp types will be are particularly suitable for creating noninverting ampli-
considered by describing their principal features. fiers. MOSFET types require special input circuitry to
Originally, op-amps were designed for the application protect them against damage by electrostatic charges.
of external voltage feedback. This keeps the inverting
input current small by maintaining high input impedance Other measuring system amplifiers: Many amplifiers
as well as by keeping the input difference voltage small.
can be constructed by combining the basic op-amp unit
Contemporary designs make the inverting input respond
with other components. Those considered here are the
to current using current feedback in such a way that the
instrumentation amplifiers in Article 123, Instrumenta-
output voltage is proportional to the current. Major benefits
tion Amplifiers, Volume 2 and the chopper, chopper
of current feedback are the much larger slew rate and the
stabilized/compensated and auto-zero amplifiers, isolation
much decreased influence of stray input capacitance on op-
amplifiers, and the charge amplifier discussed in Arti-
amp bandwidth.
cle 124, Specialized Amplifiers for Measurement Sys-
tems, Volume 2.
3 TYPES OF SEMICONDUCTOR
AMPLIFIER UNITS USED IN OP-AMPS 4 OPEN LOOP VOLTAGE TRANSFER
Bipolar transistor differential op-amp: Op-amps using FUNCTION
bipolar technology are probably the most widely applied
general-purpose units. They are used as differential ampli- As shown in Figure 1, the voltage transfer relation of an
fiers, which are described in Article 123, Instrumentation op-amp is essentially nonlinear especially when vo is close
Amplifiers, Volume 2, in low noise applications with low to either of the supply rail voltages. In normal operation,
impedance sources and in low performance, and in noncrit- the op-amp is restricted to the linear range of the voltage
ical computational applications. transfer relation. Note that when vIg = vNg then vo = 0.
They are popular because of their low cost, modularity, This corresponds to the case of infinite common mode
packaging, and wide temperature range. Input bias current, rejection, or CMR. A figure of merit for an op-amp, which
which can be in the range 1 nA to 1 µA at 25 ◦ C, has a specifies its ability to reject common mode signals, is called
typical drift in excess of 0.2 nA with offset voltage drift the common mode rejection ratio, or CMRR.
in the range 0.25 to 100 µV/ ◦ C. Open loop gain is in the An op-amp will have an infinite CMRR if there is
range from 104 to 3 × 106 . (Note that these levels of gain symmetry of all of its parameters. It is never possible
are rarely used in open loop operation – see later.) to obtain infinite CMRR because production mismatches
Operational Amplifiers 833
vo 6 INVERTING CIRCUITS
Linear operating region
Basic inverting amplifier with finite gain op-amp: In
most practical situations, op-amps are connected with
(v Ig = v Ng) = vd
feedback to form a closed loop configuration. One of the
two basic closed loop circuits is the single input inverter
shown in its general form in Figure 3(a) and for a closed
<< 1 mV
loop gain of 1 in Figure 3(b). A stage gain of (−Rf /R1 )
is expected if the op-amp gain approaches infinity. The 5-
k resistor in Figure 3(b) is to maintain symmetry for bias
purposes. More will be said about this in Section 8.
Figure 1. Voltage transfer relationship for op-amp. Assume that the op-amp is ideal in the sense defined
before so that vo = −A0 vIN . Define the frequency indepen-
always produce asymmetries. This means that the open loop dent feedback factor as βI = R1 /R1 + Rf . Also, since the
gains of the op-amp to inverting and noninverting inputs input current i = 0 the closed loop gain, ACL , can be shown
are always different. The open loop gain of an op-amp to be
is typically >2 × 105 , which is sufficiently high for the
Rf β1 A0
majority of practical applications. vo = ACL vin = − × v
R1 1 + β1 A0 in
1
5 OP-AMP MODEL A
1 + G0 0
= −G0 vin (1)
1
Op-amps are usually modeled for their signal handling 1+ A0
properties. The simple block diagram and circuit model of 1 + G0
an ideal op-amp, which are shown in Figure 2, visualize
In the case when βI A0 1, it can be seen that the closed
the op-amp as a voltage controlled voltage source.
loop gain has the value ACL ≈ −Rf /R1 = −G0 , where G0
Although op-amps require the supply of electrical energy,
is the low frequency stage gain assuming the op-amp has
their connections to the power supply are rarely shown in
an infinite open loop gain.
circuit diagrams.
In other words, the closed loop gain at low frequencies
Various degrees of model complexity exist. Input termi-
is determined purely by the ratio of two resistors. This
nals, referred to as the inverting, I, and noninverting, N,
is an important finding for it means imperfections in
inputs, drive the amplifier as a difference signal to give an
manufacture, and the gain actually used, have little effect
output voltage, which is directly proportional to this differ-
on the closed loop gain in this feedback configuration. It
ence signal. The ideal amplifier has infinite input or output
also shows why basic op-amp units are made with such
impedance, has an infinite bandwidth of operation without
high values of gain.
gain change and a gain that is very high, but not quite infi-
Since it is assumed that the op-amp input resistance
nite. This is different from the definition given in many
is infinite, the stage has an equivalent input resistance
other books where the gain is assumed to be infinite.
Complete symmetry also characterizes the amplifier so given by
there is no input circuit mismatch or gain asymmetry in the Rf
Rin = R1 + (2)
signal propagation from I and N to the output. 1 + A0
−
v lg A
+ vo I vo
v Ng −Avd = −A(v lg − v Ng)
vd
N
g g g g
Figure 2. Block diagram and circuit model of an ideal IC op-amp.
834 Elements: B – Signal Conditioning
i in R if Rf R = 10 kΩ R f = 10 kΩ
i I
v in − vo v in − vo
+ +
N
5 kΩ
(a) (b)
Figure 3. Inverting amplifier.
The term, Rf /(1 + A0 ), in (2) is the effect of Rf across the so that
terminals IN of the op-amp in Figure 3 when the Miller
Rf Rf Rf
effect of Rf is taken into account. vo = − v1 − v2 − v3 (6)
It is obvious that there will be an error in the calculation R1 R2 R3
of the gain if it is assumed that the open loop gain of
Hence, also, the channels are noninteracting if the open
the op-amp is infinite. To express this error, (1) may be
loop gain is infinite.
rewritten as
Effect of op-amp input resistance: The value of the
vo = ACL vin = −G0 (1 − εG )vin
differential open loop input resistance, Rindd , of an op-amp
−1
1 is typically >300 k and is usually neglected.
= −G0 1 − 1 + A vin (3)
1 + G0 0
where εG = (1 + 1/(1 + G0 )A0 )−1 is the relative error
7 NONINVERTING CIRCUITS
introduced by assuming that the closed loop gain is G0 =
Basic noninverting amplifier: Circuits for noninverting
Rf /R1 . Usually, but not always so, the variation in gain not
amplifiers are shown in Figure 5. In these circuits the input
being infinite is insignificant.
signal is applied to the noninverting input terminal, N,
Summing amplifier: Several inputs may be connected of the op-amp, which is assumed to have a frequency
to the summing point as shown in Figure 4. independent gain of −A0 . Taking a voltage sum around the
Assume an infinite gain op-amp so that the inverting external nodes of the circuit for the noninverting amplifier
input node, I, can be considered as a virtual earth corre- in Figure 5(a) gives
sponding to vi ≈ 0. Assume infinite op-amp input resistance
vIg = vIN + vin = vIN + vNg (7)
so that iIN may be taken as zero to give
Assume infinite op-amp input resistance so that iIN may
i1 + i2 + i3 = −if (4)
be taken as zero. Convert the loop with vo , Rf , and R1
then into an equivalent voltage source looking left from the
terminal I. Also use the frequency independent feedback
v1 v2 v3 −vo
+ + = (5) factor, defined above for (1), to obtain
R1 R2 R3 Rf
R1
vIg = βI vo = v = vIG + vNg
R1 i1
R1 + Rf o
vo A0
v1
R2 =− + vin or vo = v (8)
i2
if Rf A0 1 + βI A0 in
v2 R3 i3 i IN In the case when βI A0 1, (5) can be simplified to give
− vo
vi I
v3 R
+ vo = 1 + f vin (9)
N R1
In the same way as for the inverter, there will be an
Figure 4. Summing amplifier. error in the calculation of the noninverting gain when the
Operational Amplifiers 835
Rf R f = 10 kΩ
I vo vo vo
i IN + + +
A0 A0 R i = 10 kΩ A0
− − −
N
v in v in v in
R
g
(a) (b) (c)
Figure 5. Noninverting amplifiers (a) with gain (b) as a follower and (c) with bias compensation.
open loop gain of the op-amp is assumed to be infinite. To Basic voltage follower: The basic voltage follower,
calculate the error, (8) can be rewritten in a similar way as shown in two versions in Figure 5(b) and (c), is a spe-
(3) to obtain the expression for the relative error. cial case of the noninverting amplifier in which Rf = 0.
This circuit is used as a buffer amplifier, allowing a trans-
Example 1. Deduce an expression for the effective input formation from a high source impedance to a low source
resistance of the noninverting amplifier whose op-amp has a impedance. Specific questions of interest are the value of
low frequency open loop gain of A0 and a differential input the input resistance and the value of the output resistance.
resistance of Rindd = 300 k. If A0 = 106 , Rf = 100 k For finite op-amp gain at low frequencies, it is easily shown
and R1 = 10 k, calculate the stage input resistance. that
Ao
vo = v (11)
Solution: Represent the loop with the output voltage, vo , 1 + Ao in
and the resistors Rf and R1 as an equivalent voltage source.
Figure 5(a) can then be redrawn as in Figure 6. The values of R1 and Rf are not critical. They may be
The effective stage input resistance can be deduced chosen to suit the biasing conditions.
immediately as
Example 2. An op-amp has an open loop gain of A0 =
v R R1 105 . Evaluate the error in the closed loop gain, ACL , of an
Rin = in = Rindd βI A0 + 1 and Rindd βI A0
iIN Rf Rf inverting stage if the feedback resistance is Rf = 100 k
and the lead-in resistance is R1 = 10 k. Also, calculate
A0 R
≈ Rindd with G0 = f (10) the input resistance of the stage.
1 + G0 R1
Comment upon the error in calculating the gain in the
Completing the calculations gives case of a noninverting amplifier with the same parameter
values
A0 300 × 103 × 106
Rin = Rindd = = 27.3 G Solution: Substituting into (3) gives the gain error as
1 + G0 1 + 10
−1 −1
1 1
Hence, noninverting amplifiers appear as near open circuits εG = 1 + A = 1+ × 10 5
to signals. 1 + G0 0 11
11
≈ ≈ 110 parts in 1 million or 0.011%
105
i IN I v IN = i INR indd
+
R indd A0 The input resistance of the stage is calculated from (4) as
v in
− vo = A 0v IN
N
Rf 100
R1//R f
b I = (1 + G 0)−1 Rin = R1 + = 10 +
1 + A0 1 + 105
G 0 = R1/R f
b Ivo = b IA 0v IN = (10 + 0.001) k ≈ R1
g
The error in calculating the gain of the noninverting
Figure 6. Circuit for evaluating the input resistance of a nonin- stage with the same values is equal to the error for the
verting amplifier. inverting case.
836 Elements: B – Signal Conditioning
Slew rate and full power frequency response: The slew
rate of an operational amplifier is the maximum rate of
−
Rs A0 change of output voltage at the rated output.
+ The slew rate is normally specified for the unity gain
vI
a follower configuration, as this is often the worst case.
C
+
Cs b If frequency compensation is optimized, higher closed
vs loop gain often yields higher slew rates. The slew rates for
vo positive and negative swing of the output need not always
R
be the same. For any desired output swing, the slew rate
can set the limit for the frequency response obtainable for
that particular swing.
Figure 7. Potential equalization.
An important frequency domain parameter is the full
power response frequency, fp . This is the maximum
Potential equalization: A useful application of the volt-
frequency at which rated output can be obtained without
age follower is for two nodes of a network to be at the same
significant distortion. It can be expressed in terms of the
potential without any physical connection. This can occur
slew rate of the amplifier as
on a printed circuit card where it is difficult to isolate items
for testing purposes. The circuit points a and b in Figure 7
are at the same potential so there is no current flowing in C. Sr
fp = (14)
Hence, the source is an open circuit. However, the amplifier 2πEo
drives a current through R. Hence, the source may exhibit
an open circuit frequency response. where Sr is the slew rate and Eo is the rated output voltage.
This relationship is obtained by equating the slew rate to
the maximum slope of a sinusoidal signal of peak value, Eo ,
8 FREQUENCY DOMAIN FACTORS AND
and frequency, fp . It indicates that the effective frequency
BEHAVIOR response of an amplifier for a specific output swing is often
more limited by the slew rate of the amplifier than by the
Frequency domain parameters of op-amps: Typical val-
small signal frequency response predicted by the open loop
ues of these parameters can vary widely depending upon
gain characteristics.
op-amp price and application. For example, very low-cost
units will have
Unity gain bandwidth = 1 MHz,
Inverter with frequency dependent gain: Consider the
Low frequency open loop gain = 2 × 105 ,
inverting amplifier in Figure 3 with generalized frequency
Roll-off rate = −6 dB/octave,
dependent impedances Zf and Z1 , replacing Rf and R1
Corner frequency = 5 Hz.
respectively, and an op-amp with an open loop gain
The small signal bandwidth for any closed loop gain ACL ≥ AOL = −A0 /[1 + j(ω/ωc )]. Assume that the op-amp is
10 is given by ideal in every other respect. Define the frequency dependent
feedback factor, βI = Z1 /Z1 + Zf . Conduct an analysis
106
Bandwidth ≈ Hz (12) similar to that for the basic inverter. To take account of
ACL
the frequency dependence of all of the circuit parameters,
For ultralow noise and ultralow distortion types these this is done in the frequency domain to obtain the voltage
parameters are frequency response function as
Unity gain bandwidth = 100 MHz,
Low frequency open loop gain = 20 × 106 , Vo (jω) Z βI × AOL
= ACL = − f (15)
Roll-off rate = −6 dB/octave, Vin (jω) Z1 1 + βI × AOL
Corner frequency = 5 Hz.
Equation (15) provides a general result for frequency
The small signal bandwidth for any closed loop gain ACL ≥
10 is given by domain behavior. Integrators, differentiators, and active fil-
ters, which can be analyzed using (13), are described in
107 Section 8, of Article 124, Specialized Amplifiers for Mea-
Bandwidth ≈ Hz (13)
ACL surement Systems, Volume 2.
Operational Amplifiers 837
9 OFFSET VOLTAGE, BIAS, AND NOISE typically about 1 to 6 µV/ ◦ C. Using this coefficient, the
IN OP-AMPS output offset voltage changes can be calculated in specific
cases. This approach is also used to calculate temporal drift
Asymmetrical fabrication mismatches in the input of op- and power supply changes.
amps give rise to two undesirable effects. The first is input
voltage offset and the second is asymmetry of bias currents. The input bias current: This has a typical value of
Even when the input voltage of an op-amp is zero, the 0.01 pA to 1.5 µA, with an effective offset current of 80 nA.
output is always nonzero. The bias current is the average of the individual bias
Input offset voltage is the value of an equivalent input currents, IB1 and IB2 , shown in Figure 8(a). The offset
voltage, which would give this output voltage. is half the difference. The presence of the bias current
The input bias currents are those currents necessary to introduces an additional offset error, depending on the input
bias the input transistors of the op-amp for operation. and feedback resistors used. Referring to Figure 8(b) the
These effects are modeled as shown in Figure 8(a). The output error, vo , resulting from the effect of bias current is
offset voltage, vos , and offset current, ios , consist of a
IB1 Rf Rf
nominal component plus components due to its sensitivity vo = − IB2 Ri2 1 + (16)
to changes in temperature, time, and power supply voltage. Ri1 Ri1
It is possible to design the circuit so that the nominal and The total output offset error will include that due to the
the temperature components are completely removed. input offset voltage. For an amplifier having a typical offset
voltage error, Vos , and bias current IB , there is a critical
Voltage offset: Even with the highly sophisticated tech-
resistance value of Ri = Vos /IB . If a value of Ri greater
niques used in modern IC fabrication, it is virtually impos-
than this is used, the offset error due to bias current will be
sible to fabricate op-amps without input stage asymmetries.
greater than that due to offset voltage.
Values of the input offset voltage, vos , can be as high as
2 to 7 mV in low-cost units to as low as 50 to 100 µV in
Input offset current: The input offset current, ios , is
ultralow distortion and ultralow noise units. The polarity of
the difference between the bias current IB1 and IB2 . In
the input offset voltage may differ between units. In prac-
many op-amps, the bias currents track each other well with
tice, the output offset, which is determined by the product
temperature and are very nearly equal in magnitude. Hence,
of gain and input offset, can be corrected by appropriate
it is possible to minimize the errors due to bias currents by
offset correction circuits.
proper choice of input resistors. Note that offset current can
The output offset can be calculated in a specific case
provided both the input offset voltage, vos , its sensitivity to exhibit sensitivity to the same circuit variables as the offset
variations in temperature, power supply voltage and time, voltage. From (16), if
as well as the gain of the stage, are known. Ri1 Rf
Analysis proceeds by treating the input offset voltage as Ri2 = (17)
Ri1 + Rf
an input at the noninverting terminal of the stage, which
is looked upon as a noninverting amplifier like that in then
Figure 5(a). vo = (IB1 − IB2 )Rf = ios Rf (18)
Average temperature coefficient of voltage offset: The If Ri2 cannot be selected as in (7), the value of bias
average temperature coefficient of offset, vos /T , is current will determine the DC output errors.
Rf
R i1
I − −
vos
A A
N + vo + vo
R i2
I B2 I B1 I B2 I B1
(a) (b)
Figure 8. Individual input bias currents in (a) require compensation for resulting offset using (b).
838 Elements: B – Signal Conditioning
Example 3. Estimate the output offset voltage and its current in as simple a way as for the offset components.
variation from 25 ◦ C to 45 ◦ C for an op-amp in the Since the mechanisms that lead to both voltage and current
inverting configuration, as shown in Figure 3(a), having noise are statistical, then, statistical methods, based upon
a gain of −10 and a lead-in resistance R1 of (a) 10 k correlation analysis, must be used to calculate effective
and (b) 100 k. Assume that op-amp offset parameters are levels. If the signal is low level, corresponding to small
vos = ±2 mV with a temperature coefficient of ±6 µV/ ◦ C vd , thus requiring high amplification, usually >100, noise
and ios = 80 nA with a temperature sensitivity of 1%/ ◦ C. sources will largely determine the fundamental resolution
of the amplifier.
Solution: Consider first (a) R1 = 10 k at 25 ◦ C. For a gain
Since, op-amps are made from either bipolar or field-
of −10, Rf = 100 k.
effect transistors; the op-amp noise will consist of the
The general expression for the output voltage combining
same types as occur in these devices. Fabrication technol-
the errors due to offset voltage and bias current is
ogy uses silicon nitride methods, which is a combination
Rf of special diffusion techniques and modern surface pas-
vo = ±vos 1 + + ios Rf (19) sivation, to give good surface stability. Bipolar noise is
Ri
predominantly low frequency voltage noise due to emitter
Therefore, taking into account the direction of current region dislocations, whereas the field-effect MOS device
flow, is mostly due to trapping levels within the oxide. Chan-
nel carriers jump into these levels and are later expelled
[vo ]25 ◦ C = ±2 × 10−3 (1 + 10) + 80 × 10−9 × 100 × 103
back into the channel. Ion-implanted field-effect technology
= +30 to −14 mV is used in FET input op-amps. These effects are modeled
using similar voltage and current sources as for the offset
Knowing the temperature sensitivities, the offset voltage effects.
and the bias current, the output offset can be calculated to
give
[vo ]45 ◦ C = +29.3 to −17.3 mV 10 EQUIVALENT CIRCUIT FOR REAL
OP-AMPS
If the offset voltage correction is arranged at 25 ◦ C, the
change in output will be Figure 9 illustrates a consolidated op-amp model for assess-
vo = (±1.32 − 2) mV = −0.68 to −3.32 mV ing the circuit behavior of an op-amp. Assessment is
important because the signal propagation properties must be
Repeating the above calculations for case (b) R1 = calculable. This can only be done with an effective signal
100 k gives Rf = 100 M, [vo ]25 ◦ C = +102 to +58 mV transmission model. Other important aspects are the offset
and [vo ]45 ◦ C = +83.3 to 36.7 mV. Hence, and noise behavior of the op-amp in the actual configuration
used.
vo = −18.68 to −21.32 mV In the model, the parameters are grouped in a logical
manner by offset sources, noise sources, and impedances.
Note that the errors due to bias current increase as R1
Although all of these parameters will generally be asym-
is increased.
metrical as is shown in the model, simplification is possible.
Voltage and current noise: Unfortunately, it is not The consolidated equations for offset voltage and current
possible to compensate for the equivalent noise voltage and are respectively
vos vos vos
eos = Eos + T + Vs + t
T Vs t
temperature power supply (20)
normal sensitivity + sensitivity temporal drift
= + +
in µV at 25 ◦ C µV/day
µV/ ◦ C µV/%
ios ios ios
ios = Ios + T + Vs + t
T Vs t
temperature power supply (21)
normal sensitivity temporal drift
= + + sensitivity +
in nA at 25 ◦ C nA/day
nA/ ◦ C nA/%
Operational Amplifiers 839
Noise Offset
sources sources
I
H
i n1 I B1 Z CH
Zo
ZD voCM vod
vo
i n2 I B2 Z CB
vn vos B vo = vod + voCM
N
vod = G (v Ig − v Ng) = Gvd
(v Ig + v Ng) v CM
voCM = G =G
CMRR CMRR
Figure 9. Consolidated signal, offset and noise model for an op-amp.
It can be seen from (20) and (21) that the offset voltage Systems, Volume 2; Article 133, Amplifiers and Filters
and current can be calculated for any given changes in for DAS, Volume 3.
temperature, power supply voltage and time period from
the drift sensitivity coefficients.
It is usually possible to adjust the initial offset voltage FURTHER READING
and current to zero using a suitable biasing network. The
standard practice is always to quote offsets and drift referred Clayton, G.B. (1975) Linear Integrated Circuits, Macmillan Press,
London.
to the input in op-amp specifications. This allows the
method to be used of multiplying the input referred effects Coughlin, R.F. and Driscoll, F.F. (2000) Operational Amplifiers
and Linear Integrated Circuits, Prentice Hall, NJ.
by the noninverting gain to obtain the effect of offsets on
the output. The methods used to calculate output offset Franco, S. (2001) Design with Operational Amplifiers and Analog
Integrated Circuits, McGraw-Hill, New York.
effects in instrumentation amplifiers, which are different
from those used in op-amps are described in Section 6 of Huijsing, J.H. (2000) Operational Amplifiers – Theory and
Design, Kluwer, Amsterdam.
Article 123, Instrumentation Amplifiers, Volume 2.
Because the noise values are normally considered as Jacob, J. (2000) Analog Integrated Circuits Applications, Prentice
Hall, NJ.
energy densities per unit bandwidth, it is more appropriate
to perform any noise budgeting analysis in the frequency Lang, T.T. (1987) Electronics for Measuring Systems, Wiley,
Chichester.
domain.
Mancini, R. (2003) Op Amps for Everyone, Newne, London.
Rangan, C.S., Sarma, G.R. and Mani, V.S.V. (1983) Instrumen-
tation: Devices and Systems, Tata McGraw Hill, New Delhi.
RELATED ARTICLES Rhodes, J.D. (1976) Theory of Electrical Filters, Wiley, Chi-
chester.
Article 123, Instrumentation Amplifiers, Volume 2; Stanley, W.D. (2001) Operational Amplifiers with Linear Inte-
Article 124, Specialized Amplifiers for Measurement grated Circuits, Prentice Hall, NJ.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
This article is based in part on a chapter by J. McGhee et al originally published in Measurement Data Handling: Hardware Techniques
Vol. 2, J. McGhee et al , Technical University of Lodz, 2001.
123: Instrumentation Amplifiers
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
developed as a single-ended output with respect to ground.
1 The Electronic Instrumentation Amplifier 840 Note that an IA differs fundamentally from an op-amp.
2 Differential Input to Single-ended Output An op-amp is an uncommitted open-loop device, whose
Stage 840 closed-loop performance can be designed depending upon
the external networks used to close the feedback loop.
3 Differential-input to Differential-output Stage 842
To extend the common-mode rejection ratio (CMRR) of
4 Two op-amp Configuration 843 the configurations, IAs using discrete realizations of two
5 Three op-amp Configuration 843 and three op-amp configurations were widely employed.
6 Specifying Instrumentation Amplifiers 844 In contemporary systems where the trend is toward 12-bit,
7 Improving the CMRR – Shielding and 14-bit, or even higher resolution, the discrete approach is
Grounding 845 no longer satisfactory. This trend for better resolution also
8 Improving the CMRR – Guard Techniques 847 means that tighter specifications for nonlinearity, offset and
Related Articles 849 gain drift, and noise are usually required for instrumentation
amplifiers.
Further Reading 849
In this article, the foundational design principles of flex-
ible high performance IAs are described. These properties
are presently obtained by combining additional specially
1 THE ELECTRONIC designed circuits with high performance op-amps on the
same monolithic chip. Hence, the section commences with a
INSTRUMENTATION AMPLIFIER
description of the basic difference stage and the differential-
input to differential-output stage, which are the two con-
An instrumentation amplifier, or IA, is a closed-loop, ded-
stituent stages of the most common realization of IAs.
icated circuit differential input gain block. These units can
This is followed by details of the two op-amp and three
justifiably be considered as one of the most important ele-
op-amp forms of IA realization. The importance of deal-
ments in contemporary measuring systems.
ing with common-mode signals is considered at different
The primary function of an IA is precision amplification
places.
of the differential signal applied to its inputs, while rejecting
any existing common-mode signals. Ideally, an IA only
responds to the difference in voltage between the two
signals applied to its input terminals. This difference is 2 DIFFERENTIAL INPUT TO
often called the differential input voltage. SINGLE-ENDED OUTPUT STAGE
IAs also exhibit extremely high impedances between the
I and N input terminals, and from each of these terminals to The circuit for the general case of an asymmetrical differ-
ground. The output voltage, which is equal to the product of ential stage, sometimes also called a difference amplifier,
the amplifier gain and the differential input voltage, is also is given in Figure 1. In differential amplifiers, asymmetries
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Instrumentation Amplifiers 841
R1 R2 In a similar way, let v1g = v2g = vCM to obtain the
common mode–output voltage, voCM , as
v1g I vo
− AI
R3 R2 R β A
N
+ voCM = − − 4 N N vd = −AvCM vCM (7)
AN R1 R3 βI AI
v2g R4
where the common mode–voltage gain, AvCM , is
R2 R β A
Figure 1. An asymmetrical difference amplifier. AvCM = − 4 N N (8)
R1 R3 βI AI
in all of the parameters of the circuit must be considered.
Using superposition, the total output voltage, which
Asymmetries will occur because of the inherent asymmetry
includes a difference-mode component and a differential
of the op-amp used. In addition to the offset asymmetries,
mode component, can be written as
there will be gain asymmetry in the signal paths between the
inverting input and the output and between the noninverting
1
input and the output. vo = −Avd vd − AvCM vCM = −Avd vd + v
Offset effects may be treated separately on the basis CMRR CM
(9)
that stage linearity allows the application of superposi-
where the common-mode rejection ratio, or CMRR, is given
tion. Although the resistors in the stage are high-tolerance
by
components, it is still necessary to take these tolerances
into account.
R2 R β A
Analyzing the stage gives its output voltage as + 4× N× N
Avd 1 R1 R3 βI AI
CMRR = = (10)
vo = −AI vIg + AN vNg AvCM 2 R2 R β A
(1) − 4 N× N
R1 R3 βI AI
Define
This rejection ratio is an important figure of merit for a
R1 R3 differential amplifier, since it allows a comparison of the
βI = and βN = (2)
R1 + R2 R3 + R4 ability of one arrangement to reject common-mode signals
with the same ability of some other arrangement.
Hence, If it is momentarily assumed that the op-amp is ideal
and the circuit is symmetrical, then βI AI = βN AN . This
R2 R4 shows that the matching condition for infinite rejection of
vIg = β v + βI v o and vNg = β v (3)
R1 I 1 g R3 N 2g common-mode inputs corresponds to the denominator of
(5) being zero or (R2 /R1 ) = (R4 /R3 ).
Substitute (2) into (3) and the result into (1) and simplify However, all of the circuit resistors have manufactur-
to obtain ing tolerances and the op-amp has asymmetrical gain.
R2 βI AI R β A These inadequacies ensure that there is a practical limit
vo = − v + 4 N N v (4) to the common-mode rejection property of the stage.
R1 1 + βI AI 1g R3 1 + βI AI 2 g
To evaluate the effects, it is possible to express the
CMRR as
Assume that βI AI > 1. Now let v1g = −v2g = 1/2vd to
obtain the differential mode output voltage, vod , as
1 + G0Q
CMRR = (11)
1 R2 R β A ±δ1 ± δ4 ± δ2 ± δ3 ± (1 + G0Q )CMRR −1
vod =− + 4 N N vd = −Avd vd (5)
2 R1 R3 βI AI
where the nominal closed-loop gain to inverting inputs
where the difference mode–voltage gain, Avd , is is G0Q = R2Q /R1Q and the factor cmrr is the common
mode–rejection ratio of the op-amp itself.
1 R2 R β A Equation (11) is easily understood. It demonstrates that
Avd =− + 4 N N (6)
2 R1 R3 βI AI the CMRR of a practical differential amplifier depends upon
842 Elements: B – Signal Conditioning
both the matching of the circuit resistors and the cmrr of 3 DIFFERENTIAL-INPUT TO
the op-amp. DIFFERENTIAL-OUTPUT STAGE
It is instructive to consider typical numerical values.
Let the stage have G0Q = 1 and also that cmrr dB = 100 ≡ Another important disadvantage of the basic difference
20 log10 (Avd /AvCM ) so that cmrr = (Avd /AvCM ) = 105 , amplifier considered before in Section 2 is its compara-
which is a typical value. This gives the last term in the tively low input resistance. This can be improved using
denominator of (5) a value of 2 × 10−5 . a suitable form of the type of noninverting amplifier previ-
Compared to typical resistor tolerances of ±1%, or 0.01, ously considered in Section 6 of Article 122, Operational
this figure of 2 × 10−5 can be neglected. It is only when Amplifiers, Volume 2.
(1 + G0Q ) cmrr −1 ≥ 0.004 that the term due to op-amp It will be recalled from (10), in Article 122, Operational
cmrr needs to be taken into account. This will corre- Amplifiers, Volume 2, that the input resistance of the
spond to G0Q ≈ 4 × 10−3 × cmrr = 400 for cmrr equal to noninverting stage and the follower stage are very much
105 . Hence, for good common-mode rejection, use tightly higher than that of the inverting stage. Also, noninverting
matched resistors, which can track each other well with stages essentially buffer a high-impedance source to a
temperature, select a low value for G0Q and a high value low-impedance source. The need for high input resistance
for cmrr. Even for values of resistor tolerance of ±1% and and high CMRR in amplifiers for low-level signals are
G0Q = 1, the value of the stage CMRR is still only ≈50, satisfied by combining differential-input to differential-
which is not very impressive. Clearly, steps need to be output amplifiers with the differential-input to single-ended
taken to design a differential amplifier with a much higher stage in Section 1.
CMRR. Of the candidate circuits shown in Figure 2, consider the
In the above analysis, it has been assumed that the first Figure 2(a) using only follower stages. This combina-
common mode–input resistance/impedance of the op-amp tion stage provides a differential output from a differential
has been neglected. To give a more complete interpretation, input with a very high input resistance. Provided the op-
the effects of asymmetry in these resistances need to be amps are well matched, it also has a low common-mode
taken into account. gain. Its only real disadvantage is its unity differential gain.
It is easily understood that the common-mode resistance Consider the second circuit in Figure 2(b), consisting
between both I and N op-amp inputs and ground act as of two cross-connected noninverting stages. The voltages
voltage dividers when combined with their external circuit around the circuit are
components. The problem is especially evident if the lead-
in resistors are increased in value in an attempt to increase vw = v1 and vz = v2 (12)
the common mode–input resistance of the stage. This can
The current, i0 , flowing in the resistor, R0 , is seen to have
be counterproductive, since it will lead to increased off-
the value (vw − vz )/R0 so that the output voltages, vo1 and
set and drift. In the case of FET type op-amps, which can
vo2 , are given by
tolerate large values of lead-in resistance because of their
significantly lower bias current, such an increase might be v1 − v2 v1 − v2
counterproductive by introducing an increase in noise and a vo1 = v1 + R2 and vo2 = v2 − R2
R0 R0
decrease in bandwidth when combined with the stray circuit (13)
capacitance. When there is no differential mode input, then v1 = v2 so
Resistor tolerance has been shown to be the critical factor that vo1 = vo2 = vCM giving a common-mode voltage gain
in determining the CMRR of the stage, since the tolerance
of this component exerts a much greater influence than the −
v1 − v1
cmrr of the op-amp used. Resistance asymmetry for what- A0 vo1 A0 vo1
+ +
ever reason or from whatever cause, effectively converts R1
common-mode interference into series mode interference. vw
These series mode sources are indistinguishable from the R0
true difference-mode input. vz
Although trimming the resistors can help improve the R1
− −
CMRR, this may be of no benefit if the trimmed resistors A0 vo2 A0 vo2
v2 + v2 +
do not track each other with changes in temperature. Other
(a) (b)
sources of drift are due to sensitivity to changes with time,
such as temperature drift and aging effects, and power Figure 2. Candidate circuits for differential-input to differen-
supply voltage. tial-output stages.
Instrumentation Amplifiers 843
AvCM of unity. For difference-mode inputs corresponding 5 THREE OP-AMP CONFIGURATION
to v1 = 1/2vd and v2 = −1/2vd , the stage has a difference
mode–voltage gain, Avd , given by Three op-amp IA configuration: This configuration,
illustrated in Figure 4, has a high accuracy and excellent
vo1 − vo2 R common-mode rejection. Provided that the amplifiers, A1
Avd = = 1+2 2 (14)
vd R0 and A2 , are well-matched, low-noise, low-offset, and high-
gain units then precision difference amplification with high
Because AvCM = 1, the CMRR of the stage is thus common-mode rejection may be obtained.
numerically equal to the difference-mode gain, which can The input stages, A1 and A2 , constitute a differential-
be chosen to be very high simply by reducing the value input to differential-output stage so that vo1 − vo2 has an
of R0 . Since the stage consists of noninverting amplifiers, amplified component of vd and an unamplified compo-
the output resistance at each of the output nodes will be nent of vCM . The stage A3 is a standard differential-
very low. input to single-ended output stage with unity gain and a
Common-mode signals, which are transmitted with high CMRR.
unity gain by both op-amps, can be converted to series The equations around the circuit are
mode–interfering components if the op-amps used are not
exactly matched. With such asymmetry, any common-mode 1
2R1 2
signal, vCM , will give rise to a differential signal given by vo1 = 1 + vd + vCM
Ro
1 1 1
vCM − (15) 2R1 2
CMRR 1 CMRR 2 vo2 = − 1 + vd + vCM (20)
Ro
Taking into account the finite CMRR 3 in A3 and using (17)
4 TWO OP-AMP CONFIGURATION gives the output voltage as
Basic two amplifier configuration: A circuit diagram for R 1
vo = − 1 + 2 1 vd + v
a basic two amplifier IA realization is shown in Figure 3. R0 CMRR 3 CM
The voltages around the circuit are 1
= −Gvd + v
CMRR 3 CM
v1 R2
vA = v1 [1 + (1 + R1 /R2 )] = with B1I = 1
B1I R1 + R2 = −G vd + v
(16) G × CMRR 3 CM
R4 v2 R3 1
vo = − vA +
with B2I = (17) = −G vd + vCM (21)
R3 B2I R3 + R4 CMRR o
R4 R1 + R2 R3 + R4 where CMRR o is the overall CMRR of the stage.
vo = − v1 + v2 (18)
R3 R2 R3 Note that CMRR o = G × CMRR 3 shows that the CMRR
is increased by the difference gain G = (1 + 2R1 /R0 ) of
With R4 = R2 and R1 = R3 , this becomes the differential-input to differential-output stage. This is
R1 + R2 R2
vo = (v1 − v2 ) = 1 + v (19) v1 −
R1 R1 d A1
v01
R2
+ R1 R2
v1 − vA R3 R4 −
vo Differential symmetry
R0 A3
+ + v0
R1
− R2
R1 R2
v2 + −
A2
R2 v2 + v02
Figure 3. Basic instrumentation amplifier configured using two
op-amps. Figure 4. Three op-amp configured instrumentation amplifier.
844 Elements: B – Signal Conditioning
possible because the common-mode gain of this stage Note that for programmable gain IAs, commonly used in
equals unity. The common-mode rejection ratio of A3 , computerized data-acquisition systems, the gain control is
CMRR 3 , is completely determined with negligible error by accomplished through a number of digital inputs.
the tolerance of the resistors, R2 , in accordance with (11).
Since the common-mode rejection ratio of A3 , CMRR 3 = Offset voltage and input bias currents: Voltage offset
1/2δ2 , (21) can be written finally as and input bias current are often considered as the key figures
of merit for IAs.
2δ
vo = −G vd + 2 vCM (22) Offset voltage is defined as the voltage required at
G the input to drive the output to zero in exactly the
same way as for op-amps. While initial offset can be
Example 1. A three op-amp IA, which uses resistors
adjusted to zero, shifts in offset voltage in temperature
with a tolerance of ±1%, has a closed-loop gain of 200.
Calculate the common-mode rejection ratio of the stage. and time introduce errors. They are by far the most
important source of error in precision IAs. The temperature
Solution: Since CMRR o = G × CMRR 3 , and CMRR 3 = coefficients of these parameters, which are always specified
(1 + G3 )/4δ, the overall CMRR is calculated as CMRR o = by IA manufacturers, have typical values of the order
200 × (1 + 1)/(4 × 0.01) = 10 000, which is equivalent to of 1.0 µV/ ◦ C in the temperature range −25 ◦ C ≤ TA ≤
CMRR odB = 20 log10 10 000 = 20 × 4 = 80 dB. +85 ◦ C.
The offset at the output of an IA consists of two terms.
The first is output offset, Vos , sometimes referred to as unity
6 SPECIFYING INSTRUMENTATION gain output offset, while the second is +GVos , sometimes
AMPLIFIERS referred to as inverting gain offset.
The input bias currents, which are currents flowing into
Closed-loop gain and nonlinearity: The idealized trans- or out of the two inputs of the amplifier, correspond to
fer function is the base currents IB1 for BJT and FET leakage currents,
depending upon the specific type of op-amp input stage.
vo = −G(v1 − v2 ) (23) In the same way as for other op-amp circuits, the bias
currents flowing through the source resistance will generate
The amplifier gain, G, is usually set by the user to
a voltage offset. Although initial bias currents are often
give typical values of 1,200,500, and 1000 with a single
adjustable to zero, their drift with temperature, especially
external resistor. Temperature coefficient of the gain and
for FET type input stages, is very troublesome. Recall
gain nonlinearity, shown in Figure 5, can be found on
from Section 1 of Article 122, Operational Amplifiers,
manufacturer’s data sheets.
Volume 2 that the JFET leakage current can double every
Gain nonlinearity, which is defined as the peak deviation
10 ◦ C.
from a best-fit straight line, expressed as a percentage
There are differences between the methods of calculating
of peak-to-peak full-scale output, is possibly of more
offset and drift effects in op-amps and in IAs.
importance than the gain accuracy, since the value of the
In the case of op-amps, as pointed out in Section 8
gain can be adjusted to compensate for simple gain errors.
of Article 122, Operational Amplifiers, Volume 2, offset
parameters are always quoted as values referred to as
their inputs.
Full-scale In IAs, extra care needs to be taken because of the effect
output voltage Vmax
of the gain-determining nature of the external component.
For fixed gain IAs, it does not matter whether the offset
Emax parameters are quoted as referred to the input or as referred
to the output, because one can be calculated knowing the
v1 − v2 other, provided the gain is also known.
With adjustable gain IAs, it is essential to remember that
Best-fit straight line each of the offset parameters will give rise to different
calculated effects depending upon the gain. In many cases,
when the drift sensitivities are quoted at two values of gain,
typically 1 and 1000, it is easy to calculate the effective drift
at other values of gain. This is achieved by assuming that
Figure 5. Illustration of nonlinearity error. there is a linear relationship between the drift sensitivity
Instrumentation Amplifiers 845
and the gain. Subsequently, the calculation in specific cases at the amplifier’s input terminal is defined as
of temperature and power supply ranges and time periods
v1 + v2
proceeds in the same way as for the op-amp in Section 8 vCM = (24)
of Article 122, Operational Amplifiers, Volume 2. 2
Frequency response: Specifications such as bandwidth, 7 IMPROVING THE CMRR – SHIELDING
slew rate, and settling time are defined and specified as for AND GROUNDING
op-amps.
A major source of error in instrumentation amplifier–based
Output impedance: IA output impedance, which is ∼10 systems is the coupling of undesired interfering currents
to 20 , allows very easy and direct interfacing with and voltages into systems with low-level signals. This kind
other following modules. IAs have much lower output of coupled interference, which usually takes the form of
impedances compared with the open-loop output resistance common-mode voltages and currents, can occur in many
of op-amps. ways through
• differences in potential due to ground loop current flow;
Consolidated signal, offset, and noise model: This • electrostatic coupling through mutual capacitance bet-
model, shown in Figure 6 may be summarized as ween adjacent conductors, such as power supply lines
possessing the properties of high-input impedance, low and other high frequency sources;
offset and drift, low nonlinearity, stable gain, high common- • electromagnetic coupling through mutual inductive cou-
mode rejection, and low effective-output impedance. pling in closed conductive loops;
Hence, it is widely used for applications requiring these • asymmetrical effects such as loading of the signal leads.
advantages. Examples include amplification for various Handling common-mode signals is crucial in precision
types of transducers such as strain-gauge deflection bridges, low-level measurements, where it is especially essential to
load cells, resistance and thermistor thermometer networks, improve the CMRR. Figures 7 and 8 illustrate two examples
thermocouples, current shunts, and many more, as well of common-mode interference.
as where small differential signals superimposed on high In Figure 7, the emf, e, that develops across a thermocou-
common-mode voltages require preamplification. ple, which is bonded to a metal plate, has to be measured.
This metal plate is at a potential vCM .
Common mode–rejection ratio: An ideal IA must res- In Figure 8, mechanical stress is to be measured with a
pond only to the difference between the input voltages. The strain-gauge deflection bridge. When the applied stress is
output of the ideal IA will be zero if the input voltages are zero, the bridge is at balance and the emf, e, is equal to zero.
equal so that v1 = v2 = vCM , the common-mode voltage. However, the common-mode voltage, which is applied at
However, the output voltage of practical IAs has one the amplifier’s input terminal, is equal to
component proportional to the differential input voltage
VA + VB E
and a second component proportional to the common-mode VCM = ≈ (25)
input voltage. The common-mode voltage, which appears 2 2
Thermocouple
vd/2 H
H
e
ZCH IA
Z0
Zd voCM vod B v0
vo
ZCL +
vd/2 B vo = vod + voCM M
vod = G(vHg − vBg) = Gvd vCM
vCM (vHg + vBg) GvCM
voCM = G =
CMRR CMRR
Figure 6. A simplified consolidated model of an instrumentation Figure 7. Thermocouple bonded to a metal plate which is at a
amplifier. potential vCM .
846 Elements: B – Signal Conditioning
Considering only the common-mode voltage, vCM , this
B
circuit may be drawn as shown in Figure 9(b). It is clear
that the undesirable voltage, vP , produced by vCM is zero
Strain
H only when the bridge is perfectly balanced.
e IA
gauge When specifying the error caused by vCM , it is usual
B vo
to consider the worst-case off-balance. Assume that the IA
used in this measurement system has infinite CMRR. The IA
M measures the potential difference, vP which exists across the
A
H and B input terminals. The CMRR of the measurement
+ system, expressed in dB, is
E
vCM
CMRR dB = 20 log10 (26)
Figure 8. Circuit for measurement of mechanical strain with a vP
strain gauge.
The worst-case off-balance specified by measuring instru-
ment manufacturers is that R1 is zero, while R2 is some
Common-mode interference can exert a critical, some- finite resistance such as, say, 1 k. The equivalent circuit
times even determining, influence in such systems as is now shown in Figure 9(c) so that the
industrial-process control systems whose remote transduc-
ers are often spread over large distances from their mea- CMRR dB is
surement instruments. Analog data must be transmitted over
ZCB + R2 Zd
these long distances. Since the input lead resistances are CMRR dB = 20 log10 (27)
no longer negligible, the undesirable voltages, which can R2 Zd
be developed, are added both to any common-mode signal
Normally, Zd R2 and ZCB R2 , so (27) simplifies to
and potential difference between two ground connections,
which may influence many transducer circuits. ZCB
In the equivalent circuit of a measuring scheme shown in CMRR dB = 20 log10 (28)
R2
Figure 9(a), R1 and R2 represent the input lead resistances,
Zd the IA input impedance, and ZCH and ZCB the leakage Example 2. In a certain measuring system, it is known
impedances from H to ground and B to ground respectively. that there is a direct common-mode voltage, VCM . The
ZCH
R1
H
vd Zd
R2 vo
+ B
vCM ZCB
(a)
vP
R1 R2
+ vP + R2 Zd
vCM A B vCM ZCH
Zd
ZCH ZCB ZCB
(b) (c)
Figure 9. Measuring system in (a) has the parasitic voltage, vP , in (b) produced by common-mode voltage, vCM , may be analyzed by
the equivalent circuit in the worst-case unbalance (R1 = 0) in (c).
Instrumentation Amplifiers 847
leakage impedance, ZCB , between the B terminal of the common-mode source, vCM , as shown in Figure 10(a) and
IA and ground is purely resistive with a value of RCB = its equivalent circuit of Figure 10(b).
100 M and the off-balance resistance is R2 = 1000 . Note that RG must be as low as possible and ZL
Calculate the CMRR dB . represents the leakage impedance from the guard shield
If the common-mode voltage is now alternating, and the to ground.
leakage impedance is the parallel combination of RCB = Referring to Figure 10 allows the unwanted voltage, vP ,
100 M and a capacitance of 100 pF, calculate the CMRR dB to be written as
if the frequency of vCM is 50 Hz.
RG R R R
vP = vCM × 2 ≈ vCM G × 2 (31)
Solution: Immediately applying (28) gives RG + ZL ZCB ZL ZCB
108 Hence, using (26), the CMRR dB can be expressed as
CMRR dB = 20 log10 = 100 dB (29)
103 ZCB Z
CMRR dB = 20 log10 × L (32)
Now calculate to get R2 RG
RCB 108 Example 3. An IA mounted inside a guard shield has
|ZCL | =
=
1 + jωRCB C 1 + j2π50 × 10 × 10
8 −7 ZCB = 1000 M. If the guard shield leakage impedance
is ZL = 1000 M, R2 = 103 , and RG = 100 calculate
108
≈ ≈ 320 × 103 = 320 k the CMRR dB if vCM is direct.
2π50 × 108 × 10−7 Comment upon the CMRR dB if vCM is alternating.
Substituting in (28) gives
Solution: Using (32) gives CMRR dB = 20 log10 106 +
320 × 103 20 log10 107 = 260 dB.
CMRR dB = 20 log10 ≈ 50 dB (30) When vCM is alternating, the effects of stray capacitance
103
must be taken into account.
Consider the connections illustrated in Figure 11. The
8 IMPROVING THE CMRR – GUARD instrumentation amplifier often has a very high com-
TECHNIQUES mon mode–rejection ratio. This ensures that the common
mode–voltage gain of the amplifier is very low compared
Guard techniques allow effective shunting of the common- to its gain for differential mode signals. In cases in which a
mode bridge and at the same time they provide high cable with a length of 1 m or more runs from the transducer
common-mode impedances, ZCH and ZCB . to the amplifier, the common-mode signals presented to the
To improve the CMRR, the input circuits of the IA amplifier will not be the same as those at the transducer.
are mounted inside a guard shield, which must be well This arises because of mismatches in the resistance and
isolated from the ground and connected directly to the capacitance of the cable. For example, in Figure 11(a)
R1 H R1 R2
ZCH vP
A B
ZL RG
Zd + Zd
vd ZCB
vCM ZCH
R2 ZCB
B
RG ZL
G
+
vCM
(a) (b)
Figure 10. IA mounted inside a guard shield in (a) has the equivalent circuit in (b).
848 Elements: B – Signal Conditioning
RG1
− +
vd/2
C1
Instrumentation
amplifer
+ RG2
vCM + v /2 −
d
C2
(a)
RG1
− +
vd/2
Instrumentation
amplifer
+
RG2
+ −
vCM vd/2
(b)
v1
− vo1
A R2
RG1 +
− + R1 R2
vd/2 R
a
−
Ro A
RG2 + vo
+ − b R2
vd/2
+ R R1 R2
−
vCM A
+ vo1
v2
(c)
Figure 11. Input guarding showing the effect of cable capacitance in (a), driving the shield from one of the inputs to enhance CMRR
in (b) and deriving the common-mode signal for driving the shielded cable in (c).
suppose that RG1 C1 is not equal to RG2 C2 , where RG1 If the common-mode voltage, vCM , is not large com-
and RG2 are the source resistances and C1 and C2 are pared to the difference-mode signal, vd , it is neces-
the cable capacitances. In that case, the common-mode sary to derive the common-mode voltage as the aver-
voltage, which is converted to a series mode interfering age of the two input signals. In the schematic shown
signal, is then amplified by the differential mode gain. in Figure 11(c), this is achieved by the resistive sum-
Consequently, the common mode–rejection ratio of the ming of the outputs of amplifiers, A1 and A2 , to get
system deteriorates significantly. This problem can be vCM = 1/2 (v1 + v2 ) where v01 = vCM + Gvd and v02 =
tackled by connecting the cables preferably to the common- vCM − Gvd , where G is the difference-mode gain of the
mode voltage, instead of to the ground, so that the voltage first stage.
across the cable capacitance then reduces to the difference- In situations in which large shield capacitances are
mode signal. In this manner, the common-mode voltage at to be driven at high frequencies, the divider resis-
the input of the amplifier is now unattenuated by the cable tances, R, have to be low. If this causes excessive
capacitance. output loading on the differential-input to differential-
When the common-mode voltage, vCM , is high, the cables output stage, a unity gain buffer amplifier should be
can be driven by one of the input signals itself, as shown added between the summing-point output and the shield
in Figure 11(b). connection.
Instrumentation Amplifiers 849
RELATED ARTICLES Franco, S. (2001) Design with Operational Amplifiers and Analog
Integrated Circuits, McGraw-Hill, New York.
Article 122, Operational Amplifiers, Volume 2; Arti- Huijsing, J.H. (2000) Operational Amplifiers – Theory and
cle 124, Specialized Amplifiers for Measurement Sys- Design, Kluwer, Amsterdam.
tems, Volume 2; Article 133, Amplifiers and Filters for Lang, T.T. (1987) Electronics for Measuring Systems, Wiley,
DAS, Volume 3; Article 176, Signals and Signal-to-noise Chichester.
Ratio, Volume 3. Stanley, W.D. (2001) Operational Amplifiers with Linear Inte-
grated Circuits, Prentice Hall, NJ.
FURTHER READING
Coughlin, R.F. and Driscoll, F.F. (2000) Operational Amplifiers
and Linear Integrated Circuits, Prentice Hall, NJ.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
This article is based in part on a chapter by J. McGhee et al originally published in Measurement Data Handling: Hardware Techniques
Vol. 2, J. McGhee et al , Technical University of Lodz, 2001.
124: Specialized Amplifiers for Measurement Systems
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
current offset, drift effects, and flicker noise effects. The
1 Amplifiers for Low-level Signals 850 three chopper type amplifiers to be described in this article
2 Chopper Amplifier 850 are the chopper amplifier, the chopper stabilized amplifier,
and the auto-zero amplifier.
3 Chopper Stabilized Amplifier 852
In the case in which isolation is required, applying a
4 Auto-zero Amplifier 852 generic type of amplifier known as an isolation ampli-
5 Magnetic and Optical Isolation Amplifiers 853 fier is usually appropriate. The magnetic isolation barrier
6 Necessity for the Charge Amplifier 856 amplifier and the optical isolation barrier amplifier are also
7 Principles and Practicalities of a Charge described.
Amplifier 856
8 Integrators 857
9 Differentiators 858 2 CHOPPER AMPLIFIER
Related Articles 859
Further Reading 859 A chopper amplifier, which operates in the manner outlined
in Figure 1, uses an important method to reduce to a great
extent the effects of low-frequency drift and noise in low-
level amplification. In measuring systems, the frequency
1 AMPLIFIERS FOR LOW-LEVEL ranges of most of the interference noise and inherent noise
SIGNALS sources at least overlap the bandwidths of the sensor signals
if they do not totally coincide with them. Interference-type
Instrumentation amplifiers (IA) covered in Article 123, noise is picked up from outside the circuit under consider-
Instrumentation Amplifiers, Volume 2 will solve many ation. Inherent noise, which arises inside the circuit under
measuring system amplification and conditioning prob- consideration, can be due to the equipartition of energy, the
lems but not all. There exist many situations in which randomness in flowing current or in the characteristics, and
low offset and low noise preamplification are required. fabrication of the materials.
Additionally, in many other applications, it is essential Interference noise, which may be periodic, irregularly
to ensure that the sensing input sections of the measur- occurring, or essentially random, can be minimized, if not
ing system are galvanically isolated from the rest of the very substantially reduced or even eliminated, using suit-
system. able precautions to avoid electrostatic and electromagnetic
Chopper amplifiers and others based upon the principles pickup from nearby radiators of electrical energy.
of signal chopping, or more precisely modulation, are per- Inherent noise is usually totally random in nature
haps the most common form of low-level signal amplifiers although a nonrandom inherent interference such as chopper
based upon the op-amp. All of them can help reduce unde- noise in chopper amplifiers can occur. Random inherent
sirable variations at low frequencies such as voltage and noise usually takes the form of:
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Specialized Amplifiers for Measurement Systems 851
v T (t ) vm(t ) vo (t )
Modulator Low-pass
AC amp De-mod
(buffered S&H) filter
n (t ) vc (t )
Oscillator
(a)
TIME DOMAIN FREQUENCY DOMAIN
Carrier signal
aT = Pulse duration
(t) Kc(w) Tends to a delta comb as a → 0
Time, t Frequency, f
Ts T = 1/fs fs ≥ fT
Sampling period s Sampling frequency
(b) (c)
Inherent noise signal
Drift components ∝ (1/f 2)
(t) Time domain shape of inherent Kn(w) Flicker components ∝ (1/f )
noise signal is essentially random
and cannot be accurately drawn t Frequency, f
(d) (e)
Noise transmitted by the AC amplifier
Noise floor – other transistor noise
Modulating signal
(t) KT(w)
Time, t Frequency, f
fT
(f) (g)
PAM signal Frequency response of AC amplifier
Km (w)
Time, t Frequency, f
fT
(h) (i)
Figure 1. Block diagram and operational waveforms of a chopper amplifier.
• Johnson noise, which occurs in resistors; Consider Figure 1. An oscillator, whose output signal,
• Schottky noise, which results when current carriers vc (t), has the time domain waveform with the frequency,
flow through a region depleted of carriers, such as the fs , in Figure 1(b), also has the frequency domain spectrum,
junctions of semiconductors; Kc (ω), shown in Figure 1(c).
• flicker noise, or (1/f ) noise, which is mainly caused This carrier signal is pulse amplitude modulated by
by surface effects and crystalline dislocation, which are the transducer signal, vT (t), shown in Figure 1(f), whose
particularly troublesome deficiencies in semiconductor typical spectrum, KT (ω), is given in Figure 1(g). For the
devices; purposes of comparison, the internal drift and noise in the
• popcorn noise, which is caused by erratic jittering of AC amplifier is shown in Figure 1(e). Note that the drift
the forward current gain of bipolar transistors. gives rise to the (1/f 2 ) asymptote, flicker noise gives the
(1/f ) asymptote, and other amplifier noise gives rise to the
The only really effective way to minimize the effects base, or floor, level noise. The resulting pulse amplitude
of random inherent noise is to use modulation to shift the modulated signal, vm (t), shown in Figure 1(h), which has
spectrum of the sensor signal into a band in which flicker the spectrum, Km (ω), in Figure 1(i), is applied to the AC
effect is much less. Hence, these op-amp-based structures band-pass amplifier whose frequency response is shown in
include some form of modulation and demodulation. In Figure 1(i).
effect, they allow the use of very narrow bandwidth filtering If the AC amplifier has internal drifts and noise, their
in a stable manner. spectrum being as shown in Figure 1(e). Comparison of the
852 Elements: B – Signal Conditioning
region where this interference spectrum is concentrated with Analyzing in the frequency domain, it can be seen that
Figure 1(i) shows that only the noise inside the bandwidth the relations in the circuit are
of the band-pass amplifier, with the comparatively low-level
noise floor spectrum, will be transmitted. If the bandwidth Vo = −Ad × Kd GVIg + Ad × G × Ac VIg (1)
adequately covers that of the transducer spectrum around
fs , then the net mean square value of the transmitted noise Assuming very large input impedance at I gives
appearing at the amplifier output will be very low. The
R1 Rf
chopper has removed most of the amplifier interference. VIg = V + V = βI Vo + Gex × βI Vin
Rf + R1 o Rf + R1 in
(2)
where the feedback factor βI is defined as βI =
3 CHOPPER STABILIZED AMPLIFIER R1 /Rf + R1 = 1/1 + Gex and the external ideal gain Gex ,
which is defined as Gex = Rf /R1 , is the gain, assuming an
Chopper amplifiers can be employed with other high- ideal op-amp.
performance op-amps to compensate for the already small Inserting (2) into (1) and simplifying gives
inadequacies of these high-quality op-amps. The resulting
Rf βI Ad × G(1 − Ac )
combination, shown in Figure 2, gives a circuit called a Vo = − V
chopper stabilized amplifier, which was first patented more R1 1 + βI Ad × G(1 − Ac ) in
than 50 years ago. Its designed aim was to combat the drift Ad
of directly coupled amplifiers by converting the DC voltage + Kd
1 + βI Ad × G(1 − Ac )
to an AC voltage.
The most important disadvantage of this inverting chop- = Desired output signal
per stabilized stage is the relatively low input resistance, + Disturbance caused output signal (3)
which is the same order as the lead-in resistance. A non-
inverting, high input resistance version of the chopper Assuming that βI Ad × G(1 − Ac ) 1 allows (3) to be
stabilized type first appeared about 30 years ago. By caus- written as
ing interference from digital switching noise, the bandwidth
Rf 1
is limited. Intermodulation distortion (IMD) between the Vo = − V + K
clock and the input signal, which is similar to aliasing, R1 in βI G(1 − Ac ) d
produces errors at the sum and difference frequencies. Rf 1 + (Rf /R1 )
Although both of these types have now been ostensibly =− Vin + Kd (4)
R1 G(1 − Ac )
replaced by the auto-zero amplifier, which is described in
Section 4, they are still worth examining. When Ac ≡ 0, which corresponds to a conventional
The chopper amplifier in Figure 2 measures the voltage inverting amplifier circuit, (4) becomes
of the inverting input of a conventional high-performance
op-amp and then applies its amplified output to the non- Rf 1 + (Rf /R1 )
Vo = − V + Kd (5)
inverting input of the same op-amp. In this way, any R1 in G
disturbances occurring inside the feedback loop are reduced
in accordance with the well-known theory of negative feed- Hence, with chopper stabilization, the disturbance input,
back. To illustrate this reduction, consider that a disturbance Kd , is reduced by the factor (1 − Ac )−1 . Note that the
Kd , shown in Figure 2, occurs inside the feedback loop. chopper amplifier gain Ac should be inverting at low
frequencies, corresponding to Ac (ω → 0) = −Ac0 .
Vin Vo
Rin Rf 4 AUTO-ZERO AMPLIFIER
I
− Kd Operating principle: Auto-zero chopper stabilized amp-
−G lifiers, with the block diagram in Figure 1, are now the
Ad
N G most widely available form of chopper stabilized ampli-
Ac +
Chopper fiers. Their operation consists of two phases. During Phase
amplifier High-gain op-amp 1, as in Figure 3(a), the input is applied only to the main
amplifier, A1 , whose nulling input voltage, VN1 , is supplied
Figure 2. Chopper stabilized amplifier. by the voltage on capacitor, CM1 .
Specialized Amplifiers for Measurement Systems 853
vos1 External feedback vos1 External feedback
vN + vo vN + vo
N N
A1 A1
I − I −
vI VN1 vI VN1
S2a S2a
vos2 vos2
+ +
S1a A2 S1a A2
S2b S2b
− S1b − S1b
VN2 VN2
CM1 CM2 CM1 CM2
(a) (b)
Figure 3. Phase 1 of an auto-zero amplifier in (a) is followed in operation by Phase 2 in (b).
Meantime, the I and N inputs of amplifier, A2 , are which is very similar to aliasing in sampled data theory,
connected by S2a to allow this amplifier to measure its auto- can be reduced by a careful layout design.
zero voltage that is stored on capacitor, CM1 . At the start of
Phase 2, the make-before-break switches, S1 and S2 , each
change over. This allows the previously measured auto- 5 MAGNETIC AND OPTICAL ISOLATION
zero values for A2 to be applied at VN2 . When Phase 2 is AMPLIFIERS
complete, the whole sequence of operations is repeated.
The isolation amplifier is particularly appropriate in data
Applications and characteristics of auto-zero amplifiers: acquisition systems where galvanic isolation (i.e. no DC
Auto-zero amplifiers possess a number of highly desirable electrical connection path) between the input and the output
characteristics. They are eminently suitable where very circuits is important such as in the following examples:
high input impedance buffering is necessary. For this
• amplification of low-level measurement signals in high
reason, they are mostly applied in integrators, electrode
common-mode environments;
buffers, ionization chamber buffers, and electrometer-type
• breaking of ground loops;
applications for femto-ampere current measurement.
• isolation protection of personnel and equipment due to
Although the clock frequency is in the range from 100 Hz
hazardous common-mode situations such as in biologi-
to a few kHz, the gain bandwidth product can be about 1 to
cal monitoring.
3 MHz or even more. The open-loop, low-frequency gain
is >107 while at the same time offering very low offset They also allow effective utilization of gain by inserting the
voltages in the region of 1 µV. Offset drift, which is a gain at the signal source where the bandwidth is likely to be
negligible 50 nV per month, gives the auto-zero amplifier lowest. The isolation barrier may be a screened transformer
the lowest drift of all contemporary types of amplifiers. or an optocoupler.
The same comment can be made about offset temperature Isolation amplifiers, which are represented in Figure 4(b),
sensitivity, which is ≈5 to 10 nV/ ◦ C. The CMRR and the appear inside the isolation units of the general system
power supply rejection ratio, or PSRR, which are also architecture shown in Figure 4(a). A generalized data
strongly affected by these low levels of offset, can have acquisition system in an industrial process environment,
values in the order of 140 dB or more. They are also requiring multiple signal isolation solutions, is given in this
characterized by low voltage and low current noise. system block diagram.
The high-speed chopper action means that the low- A diversity of signals is acquired from a range of resis-
frequency noise is nearly uniform down to zero frequency. tance type transducers, voltage sources, current sources, and
This should be compared with the flicker noise problems perhaps other types of transducers such as thermocouples.
of other noncompensated amplifiers, whose flicker noise Each of these signal sources is associated with an indi-
corner frequency can be in the range from ≈5 to 10 Hz. vidual conditioning card, whose function is to perform any
Auto-zero amplifiers use MOS input devices, which ensures necessary energizing power supply, screening, guarding,
that the bias currents required are in the range 10 to 20 pA and grounding, and perhaps preamplification. Power sup-
at 25 ◦ C while exhibiting drifts of about 0.5 pA/ ◦ C. This is ply and control lines connect the conditioning card to the
a considerable advantage if large value source resistances isolation units.
are likely. Chopper radiation can sometimes cause problems Isolation amplifiers consist of three parts. The first
due to intermodulation distortion. This type of distortion, part is a power supply, which is necessary for correct
854 Elements: B – Signal Conditioning
Unisolated
Conditioning Isolation units power supply
cards
Analogs
Resistance sensors Controls
(e.g. strain gauges,
RTD etc.) Isolated
power
(Microprocessor controlled)
Voltage
Microprocessor
Multiplexer
Current
Thermocouple
(a) Microprocessor clock
H
+ RISO
vd
CISO vo
B
+
vCM
+ i ISO
vISO
(b) Signal ground Instrument ground
Figure 4. A data acquisition system using isolation units containing isolation amplifiers in (a) has a model for isolation amplifiers
given in (b).
operation. It is arranged using some kind of isolating Where bandwidth and response speed are important, optical
and chopping DC-to-DC converter. An input stage with isolation amplifiers are preferable to the magnetic type,
floating input connections and an output stage with out- which are best suited for gain accuracy and linearity.
put connections are the other two subunits. All of the
subunits are electrically isolated from each other. Resid- Magnetic barrier isolation amplifier: The most com-
ual and stray impedances will always be associated with monly encountered form of isolation amplifier, which has
practical isolation amplifiers. As in Figure 4(b), isola- a block diagram similar to that in Figure 5, uses magnetic
tion resistance and isolation capacitance, which are the isolation barriers in the form of miniature toroidal trans-
most important parameters influencing the effectiveness formers.
of the isolation between the signal input terminals and A high-power oscillator is magnetically coupled through
the output terminals, have respective values of 1010 to a DC-to-DC converter to provide the power supply for
1012 and 10 to 20 pF, so ensuring isolation poten- the input stages. It is necessary to provide two stages
tials up to about 5 kV and leakage currents iISO ≈ 0.5 to of isolating modulation and demodulation units. One of
2 µA. these units allows transmission of the signal from the
It can be seen from Figure 4(b) that the effect of vISO is input across the magnetic isolation barrier to the output.
very similar to vCM . For this reason, it can be taken into A second modulator and demodulator allows transfer of
account in numerical evaluations by using a figure of merit the output signal back into the isolation side for feedback
called the isolation mode rejection ratio, or IMRR, which purposes.
is defined to give Magnetic barrier isolation amplifiers can cope with
v v typical continuous DC isolation voltages of 2 kV or a
vo = G vd + CM + ISO (6) value of 1.4 kV rms for alternating isolation voltages. The
CMRR IMRR
Specialized Amplifiers for Measurement Systems 855
Suitable screening,
Shielding, and guarding
Isolation
barriers
De-mod Modulator
Chopper stabilized Postamplifier
amplifier
vo
De-mod Modulator
DC-to-DC converter
Rectifier Chopper
Figure 5. Block diagram of a magnetic barrier isolation amplifier.
RB1 RB2
Rin Rf
i C1
vin
R2 R2 i C2
vo
g
(a) V+
R ZL
(b)
Figure 6. Optical barrier isolation amplifier.
IMRR is about 160 dB for direct voltages and 130 dB for The current in the input LED, which requires a standing
alternating forms. Leakage current can be about 0.5 µA. bias current, IB , to ensure full range input swing, is
controlled by the input op-amp. The standing bias is IB =
Optical barrier isolation amplifier: Figure 6 illustrates V + /RB . Hence, the collector current of the transistor in the
two versions of optical isolation amplifiers for two differ- input optoisolator is
ent possibilities. When optoisolation of analog voltages is
required, Figure 6(a) shows how feedback in the preampli- vin V+
iC1 = + (7)
fier can be used to linearize the response of the optoisolator. Rin RB1
856 Elements: B – Signal Conditioning
If the two optocouplers are closely matched, the currents, C, depending on the cable length, appearing in parallel with
iC1 and iC2 , in the collectors of the two transistors will be the amplifier input resistance RIN .
equal. It can be seen that the output op-amp is configured Referring to Figure 7, it can be seen that
as a current to voltage converter so that its output voltage is
sRIN
+ + Vc (s) = Q(s)
Rf V V R 1 + sRIN (Cc + C)
vo = v + − = f vin for RB1 = RB2
Rin in RB1 RB2 Rin sRIN (Cc + C) Q(s)
(8) =
In the case when the measured variable has been con- 1 + sRIN (Cc + C) Cc + C
verted to a current, a circuit like that in Figure 6(b) can be (s/ωci ) Q(s)
= (9)
used. The input current, iin , which consists of a standing 1 + (s/ωci ) Cc + C
bias current, IB , and the signal current, is , means that the
collector current of the input transistor iC1 = iin . The out- The stage acts like an active high-pass filter, whose Bode
put current io = vo /ZL , which is fed back to constrain the frequency response in Figure 7 has a cutoff frequency, fci ,
potential at the inverting input, I, of the op-amp to consist given by
of a component due to the standing bias current at the input ωci 1
and another component proportional to the signal current. fci = = (10)
2π 2πRIN (Cc + C)
The component due to the standing bias current is backed
off with the potential at the noninverting input, N. The out- As the frequency response at low frequencies is restricted,
put current is therefore directly proportional to the input slowly varying phenomena cannot be measured. A charge
signal current, iin . amplifier is needed to interface the PZT so that the high-
Optoisolators have DC isolation voltages up to about pass corner frequency can be reduced.
2 kV, which becomes about 5 kV for a pulse test. Their Since the output voltage of the transducer depends on C
rejection ratio is usually better than 140 dB for DC con- and thus on the cable length, a new calibration has to be
ditions or 120 dB at 50 to 60 Hz. Their leakage current is carried out every time the cable length is changed. This is
typically 0.25 µA and isolation leakage capacitance ≈2 pF. an unfortunate drawback of this measurement system.
6 NECESSITY FOR THE CHARGE 7 PRINCIPLES AND PRACTICALITIES
AMPLIFIER OF A CHARGE AMPLIFIER
Quartz and barium titanite are asymmetrical crystalline
In the simplified charge amplifier of Figure 8, assume
materials that produce an electrical charge when weights are
that the amplifier, A, is ideal as described previously in
placed on them. Piezoelectric transducers (PZT), which are
Section 1 of Article 122, Operational Amplifiers, Vol-
used for force measurement, particularly in accelerometers,
ume 2.
utilize this property where the produced electrical charge,
Q, is proportional to the applied force F . As a consequence,
PZT can be modeled as a current source i = dQ/dt in Cable C1
parallel with a capacitor, Cc . −
Transducers are often remote from measurement ampli- A
+ v0
fiers in practical measurement systems. The connection
between the transducer and an amplifier, as in Figure 7,
is realized with coaxial cables. These present a capacitance Figure 8. Functional diagram of a charge amplifier.
Coaxial cable Vc
Q(Cc + C) dB
i = dQ vc
dt RIN Frequency,f
Cc C vo
(a) (b) fci
Figure 7. A piezoelectric transducer in (a) connected to an amplifier with a coaxial cable with a capacitance depending on its length
has the Bode frequency response in (b).
Specialized Amplifiers for Measurement Systems 857
The charge, Q, produced by the transducer will be stored Hence, care must be taken to choose an adequate op-
in the capacitor, C1 . Therefore the amplifier output voltage amp in the design of charge amplifiers. It is possible to
will be equal to avoid this integration effect by placing a large resistance,
Q R1 , with a value of the order of 1 to 10 M in parallel with
vo = (11) C1 . Unfortunately, this also causes an output offset voltage
C1
with a value of
As this is no longer dependent on C and R, recalibration
vo = −R1 ib (15)
is not necessary when the cable length is changed.
Charge amplifier with a real op-amp: Suppose the
amplifier is nonideal with finite values of gain, A0 , and 8 INTEGRATORS
input resistance, RIN . Using the relations in Section 7
of Article 122, Operational Amplifiers, Volume 2 with The need for integrators and differentiators often arises
appropriate modifications, the amplifier output signal can in measuring systems. Integrators are employed when the
be expressed as acceleration measured has to be converted to velocity or
when velocity has to be converted to displacement. Con-
A0 sRIN versely, to obtain velocity from displacement, differentia-
vo (s) = −Q(s) (12)
1 + sRIN [Cc + C + C1 (1 + A0 )] tors are required.
The basic integrator circuit is shown in Figure 9(a). It has
Owing to the Miller effect transformation, the feedback
limited practical use because the nonzero input offset errors
capacitance appears as a high-value capacitor C1 (1 + A0 ) in
produce high output offset errors when they are integrated
parallel with the cable and the transducer capacitance. Since
at very low frequencies.
C1 (1 + A0 ) (Cc + C), the output signal is effectively
These errors can be minimized by reducing the gain
independent of the cable length.
at very low frequencies by the low-frequency feedback
By contrast with (10), the cutoff, or corner frequency, is
in the 15-M resistor in Figure 9(b). Though this results
now
in nonideal integrator characteristics, the final circuit is
1
fc = fci (13) useful for many applications over a limited frequency range.
2πRIN C1 (1 + A0 ) For the component values indicated, the phase lag of this
circuit is ±270◦ ±0.5◦ compared with +270◦ for an ideal
The high input resistance, RIN , implies the need for an op-
inverting integrator working over a frequency range of
amp with a low input bias current. This can be met by using
10 Hz to 10 kHz. At very low frequencies, the gain is
either op-amps with FET input stages or BJT input stages,
limited to −1000, resulting in output drift of less then a
specifically designed with low input bias currents. Although
few millivolts.
FET input op-amps have well under 1-nA input bias current
The gain, which is approximately −10 at 10 Hz, reduces
at 25 ◦ C, this current rises with increasing temperature to
to about 0.01 at 10 kHz. To ensure that the output at
values above that of bipolar input types. This small bias
10 kHz does not become comparable with the drift vin at
current, ib , causes an output voltage that increases with
this frequency, this should be of the order of a few volts.
time in the manner
The phase error at the high-frequency end is the direct
ib result of the limited unity gain bandwidth of the op-amp
vo = − t (14)
C1 used. Low-frequency phase errors are related to the basic
15 kΩ 0.1 µF 0.01 µF
C 15 MΩ 15 MΩ 750 kΩ 750 kΩ
vin +V
1 MΩ − 100 µF
R A 15 kΩ
−
A + vo −
+ vo −V 1 kΩ A
vin + vo
(a) (b) (c)
Figure 9. Integrator with the basic configuration of (a) is practically realized using (b) whose performance with frequency is improved
with (c).
858 Elements: B – Signal Conditioning
low-pass characteristics of the configuration used with a as
−3 dB cutoff, or corner, frequency of approximately 0.1 Hz,
relative to DC gain. The useful operating range of this (τA + RC)2 1
ζ = (17)
integrator can be shifted to a frequency range of 1 Hz to RCτA A0
1 kHz by increasing the feedback capacitance to 1 µF.
Further reduction of DC errors can be obtained by
where A0 is the low-frequency open-loop gain and τA is
providing higher feedback for DC and making this feedback
the open-loop time constant of the op-amp. If A0 = 105
ineffective for AC use in the desired frequency range.
and τA = 0.016 s, then ζ = 1/1000 for RC = 0.16 s. This
The arrangement shown in Figure 9(c) has a DC gain of
100, reducing the output DC errors by a factor of 10, as corresponds to a unity gain bandwidth product of 1 MHz.
compared to the performance of the integrator shown in The peaking of the frequency response means that the
Figure 9(b). differentiator, however, gives a very noisy output since
it amplifies its own internal noise in the 1-kHz range
by as much as 10 times. If the amplifier has a response
rolling off at more than 6 dB/octave, the circuit may even
9 DIFFERENTIATORS
oscillate.
The modified differentiator shown in Figure 10(c) can be
The basic differentiator shown in Figure 10(a) has limita-
used in practical situations since the closed-loop response
tions. Owing to the 6 dB/octave rolloff of the op-amp gain
is designed to be well damped. With the values shown
at high frequencies and the differentiating characteristics of
the configuration at the operating frequency, the closed-loop in Figure 10(c), the response is typically as shown in
gain shows a peak in its frequency response, similar to that Figure 10(d). The resistance in series with the input circuit
in Figure 10(b). Using a low cost op-amp with the param- capacitance flattens the response at about 30 kHz. At
eters in Section 1 in Article 122, Operational Amplifiers, 100 kHz, the response starts to roll off because of the
Volume 2 and a value of RC = 0.16 s, the gain is unity at influence of the open-loop gain of the op-amp. For this
1 Hz with the peak at a frequency of approximately 1 kHz. to happen at 100 kHz, the unity gain bandwidth of the op-
The damping factor, ζ , is obtained from the equation amp should be ≈10 MHz. This can be achieved readily with
many op-amps using an external compensating capacitor
Vo −RCs with a typical value of 3 pF. The phase shift is 90◦ lead, in
= (16)
Vi 1 RCτA 2 addition to the ‘inversion’ of the amplifier, with less than 2◦
1+ (τ + RC)s + s
A0 A A0 error for all frequencies less than 1 kHz. At frequencies of
Open-loop gain
100
160 kΩ
80
Closed-loop gain
Gain (dB)
1µF R
60
−
Vi C A 40
+ Vo
20
0
0 10 102 103 104 105 106
(a) (b) Frequency (Hz, log scale)
100
53 kΩ 80
Gain (dB)
0.1µF 60 Open-loop gain
− 40
Vi A
+ Vo
20
Closed-loop gain
3 pF
0
1 10 102 103 104 105 106 107
−20
(d) (c) Frequency (Hz, log scale)
Figure 10. A basic differentiator in (a) has the frequency response in (b), while the modified differentiator of (c) has the frequency
response of (d).
Specialized Amplifiers for Measurement Systems 859
0.1 Hz and below, the reduced gain for the input signal can Article 133, Amplifiers and Filters for DAS, Volume 3;
affect the performance. A differentiator for the frequency Article 181, Amplitude Modulated Signals: The Lock-in
range of 0.01 to 100 Hz is achieved by using the op-amp Amplifier, Volume 3.
with 30 pF compensation and by changing the input circuit
capacitance to 1 µF.
FURTHER READING
RELATED ARTICLES Huijsing, J.H. (2000) Operational Amplifiers – Theory and Design,
Kluwer, Amsterdam.
Article 122, Operational Amplifiers, Volume 2; Stanley, W.D. (2001) Operational Amplifiers with Linear Inte-
Article 123, Instrumentation Amplifiers, Volume 2; grated Circuits, Prentice Hall, NJ.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
This article is based in part on a chapter by J. McGhee et al originally published in Measurement Data Handling: Hardware Techniques
Vol. 2, J. McGhee et al , Technical University of Lodz, 2001.
125: Outline of Purpose of Analog Data Filters
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
passive R-L-C elements, frequently referred to as pas-
1 Role of a Signal Filter 860 sive filters, are used in the range from about 100 Hz to
2 Features That Need Selecting or Filtering >100 MHz. Active filters, which are widely applied in
Out 861 the frequency range from <0.01 Hz to about 100 kHz, are
normally realized using individual resistors, capacitors, and
3 Classification of Active Filter Types 862
op-amps. Microwave filters, also called cavity resonators,
4 Application of Filters 862
as their name implies, operate in the microwave frequency
5 Comparison of Analog and Digital Filters 862 range above about 10 to 50 MHz. Crystal filters using piezo-
6 Overview of Active Analog Filter Design electric techniques cover the range from about 100 Hz to the
Methodology 863 low microwave region. Monolithic filters, which use large-
7 Proprietary Filter Modules 863 scale integration for their realization, cover the same basic
Related Articles 863 range as active filters but have the possibility for higher
References 863 frequency operation because of their smaller general physi-
cal dimensions. This is because the R-C elements used are
normally included on a chip.
Analog filters are classified as passive if they use only
1 ROLE OF A SIGNAL FILTER passive R-L-C elements. More recently, active filters using
combinations of passive R-C elements and active op-amps
It is always the case that the signal acquired from a sensor have been widely developed. Digital filters, which process
or transducer requires conditioning of some kind or another. discrete interval sampled data signals, are realized using
Even after passive conditioning with constant voltage, con- dedicated digital hardware structures or software programs.
stant current or bridge circuits, or active conditioning with The digital hardware used takes the form of shift registers,
various kinds of amplifiers, there will still be the need to multipliers, and accumulators.
introduce some form of filtering. Filters, which usually are The concepts of filtering provide the measurement sys-
regarded as modifiers in the terminology of Article 109, tem designer with additional tools to tackle the problems
Structure and Energy in Sensor Systems, Volume 2, are of minimizing error, signal-to-noise ratio, response time,
specially designed to modify the range of their transmitted and bandwidth. At the same time, analog filters allow mod-
frequencies. Filtering operations may treat analog or digi- ification of the frequency content of the acquired signal
tal signals. Although the filters used in measurement and so that the problems of foldover associated with digi-
instrumentation use electric circuit realizations, filters exist tal signal processing are at least minimized by the use
in other energy forms. Electric filters are also called electric of antialiasing precautions. Before the era of ICs, filters
wave filters. were exclusively realized using passive R-L-C elements.
The frequency ranges of filters can be used as a means Presently, the op-amp has exerted a profound influence
of defining their different types. For example, filters using upon the design of data filters for analog signals. This
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Outline of Purpose of Analog Data Filters 861
Table 1. Advantages and disadvantages of active filters.
Advantages Disadvantages
1. Small in size and weight because there are no 1. Limitations on voltage/current levels makes them unsuitable
inductors in electrical power systems
2. Fairly easily designed 2. Inadequacies due to imperfections in the static and dynamic
characteristics of the active elements
3. Easily ‘tuned’ to give the desired response 3. Parasitic oscillations may occur when feedback is used
4. Isolation between cascaded stages 4. High sensitivity to tolerances of passive components
5. Large available gain so amplification is possible 5. A power supply is required
means that for frequencies up to about 100 kHz there is the measurement signal. Simultaneous preservation of rela-
no need for inductors. tive amplitude and relative phase difference are conflicting
The design techniques for active analog filters are largely requirements. In these cases, it is necessary to design some
the same as for passive filters. However, as active networks filter that compromises between the two requirements.
contain op-amps employing feedback, the stability of the Interference, more widely known as noise, occurs in all
filters needs to be considered. Also, as inductors are bulky, measurement systems. It is shown in Article 68, Setting
it is preferred that only R and C elements are used, even the System Boundaries, Volume 2 that noise may pen-
though inductors may be realized using gyrators. Gyrators etrate a measurement system from the outside, in which
will not be considered further. A general treatment is case it may be called interference or interfering noise.
quite difficult as there is such a large variety of possible Noise may also arise inside a measurement system owing
methods. Table 1 gives advantages and disadvantages of to imperfections in system design or of the materials used
active filters. in its construction. All of these noise sources contribute
to the overall error. Errors should be minimized to real-
ize the overall aim of measurement, which Solopchenko
2 FEATURES THAT NEED SELECTING (1994) asserts as being the discovery of errors and their
OR FILTERING OUT minimization.
One of the most commonly occurring problems in mea-
Filtering is essentially a signal-processing operation to surement and instrumentation systems is due to interfering
obtain optimal detection of some specified parameter or pickup, or ‘hum’, from the electric supply systems used in
parameters of a signal immersed in interference or noise. the electrical elements of the overall system. Thus, unde-
Problems of filter design are concerned with optimizing sirable inductive and capacitive coupling into the more
the preservation of the important signal parameters and sensitive parts of the system from nearby electrical sources
removing or at least minimizing the effect of the noise. needs to be guarded against. Removal of this type of inter-
The first item of interest is the signal and the manner in ference may also be accomplished using a band-reject filter,
which it carries the measurement information. Information sometimes also called a notch filter. Other kinds of inter-
can be carried in the amplitude, the phase, or the frequency ference may be due to cosmic effects.
of an analog measurement signal, or perhaps in the shape in Naturally occurring fundamental fluctuations are endemic
the case of a pulse modulated signal. When the amplitude throughout nature. In mechanical systems, they are man-
carries the information, the design of the filter concentrates ifested in Brownian motion, while in electrical systems,
upon uniform amplification of the frequencies within the they are referred to as thermal, Johnson, or Nyquist noise.
band of interest so that the relative magnitudes of each These fluctuations, which are due to the equipartitioning of
frequency component are maintained after the filtering oper- energy, have a uniform power spectral density up to about
ation. Ideally, all signal components outside the band of 10 GHz. Quantum corrections must be applied beyond this
interest should be attenuated by as large an amount as possi- frequency. Since most measurement systems use electri-
ble. Using the same reasoning, filter design concentrates on cal/electronic building blocks, other types of noise such as
optimizing the phase characteristic of the filter if the infor- flicker, or low frequency noise, shot noise, and other types
mation is carried in the phase angles of the signal compo- of noise characteristic of semiconductor components require
nents. In this case, optimal retention of the relative phases of precautions to be taken to minimize their effects.
the frequency components within the band of interest after In some cases, the measurement information is associ-
filtering is the aim. This is especially important if the shape ated with a narrow range of frequencies. To ensure that
of the waveform needs to be maintained. In some cases, unwanted components outside this range are not transmitted
information is carried in both the amplitude and the phase of in the measuring system, band-pass filters are used.
862 Elements: B – Signal Conditioning
Gain Gain in repeaters, and for harmonic suppression in some low-
frequency applications.
A major application area of filters is in measurement and
instrumentation systems. When measuring voltages with
digital voltmeters, it is important to reject interference
from noise sources and from power line harmonics. Similar
requirements for noise rejection occur in the study of vibra-
tion and shock. Many physiological signals such as those
w w
occuring in heart monitoring by electrocardiogram, or ECG,
(a) (b)
or in brain wave monitoring by electroencephalogram,
or EEG, have the information content of their spectrum
Gain Gain in a three-decade band below about 20 Hz. Special fil-
ters to remove ultralow-frequency interference below about
0.002 Hz, and noise penetration above 20 Hz are required.
Electric wave filters, and especially active filters, are
also widely used in control systems, such as servo-systems
and motor drive controllers, as well as in process control
equipment, consumer and entertainment systems such as
w w stereo amplifiers, videocassette recorders, or VCRs, digital
(c) (d) versatile disc, or DVD, systems and in all types of public
broadcast receivers.
Figure 1. Classes of filters: (a) low pass, (b) high pass, (c) band Applications also occur in sonar systems, telemetry sys-
pass, and (d) band stop/notch. tems, prefilters for analog-to-digital converters, spectrum
analyzer systems and many others.
3 CLASSIFICATION OF ACTIVE FILTER
TYPES
5 COMPARISON OF ANALOG AND
There are four main types of frequency responses real- DIGITAL FILTERS
ized using active filters as in Figure 1. Low-pass sections,
which widely occur in instrumentation systems, are used Some advantages and disadvantages of analog filters have
to overcome some limitations of a sensor signal by remov- been given above. Digital filters take the form of recursive
ing unwanted higher frequency noise and providing some filters, also called infinite impulse response filters, or IIR
antialiasing protection. The other three are the high-pass filters, and nonrecursive filters, sometimes referred to as
section, the band-pass section and the band-reject or notch- finite impulse response filters, or FIR filters.
filter section. All-pass sections, which are not shown in Digital filters have some significant advantages over ana-
Figure 1, are used as phase equalizers. They possess a flat log filters. In the first place, digital types possess guaranteed
amplitude response with a phase response, which can be accuracy, which is only limited by the number of bits,
designed to a given value of phase delay or advance. and guaranteed reproducibility since there are no varia-
tions due to component tolerances or aging or temperature
drift. In contrast, the accuracy and reproducibility of ana-
log filters cannot be guaranteed because they can depend
4 APPLICATION OF FILTERS heavily upon the tolerance of the R and C elements,
upon the characteristics of the op-amps used, and upon the
The rapid development of large-scale communication sys- aging and temperature drift effects, which they are all sub-
tems provided a major impetus in the need for and devel- jected to. Thus, digital filters have superior performance
opment of both passive and active filters. In this field, over analog types. The flexibility of digital filters arises
an very important application is the pulse-code modula- from their programmable nature, which allows a variety of
tion coder/decoder, or PCM-CODEC, in digital telephony. functions to be performed using a single hardware struc-
In this application, the filters require very tight specifica- ture. It is impossible to provide this flexibility with analog
tions of both their cutoff sharpness and pass band ripple. filters.
Other communication system examples of filtering occur Nevertheless, digital filters possess three main disadvan-
in signal preemphasis, equalization, impedance matching tages. Designing DSP can be time consuming unless the
Outline of Purpose of Analog Data Filters 863
development resources and knowledge are available. Real- low-pass-to-notch transformations (Ludeman, 1987; Ran-
izing DSP can be quite expensive, especially with wide- gan, Sarama and Mani, 1983) features.
band signals. Such signals need fast, and thus expensive,
analog-to-digital and digital-to-analog converters. Because
real-time applications are implemented using only a limited 7 PROPRIETARY FILTER MODULES
number of bits, there is a real possibility for errors due to
There are many IC manufacturers such as Burr-Brown,
quantization effects.
now a division of Texas Instruments, Maxim (http://www.
miximic.com), Linear Technology (http://www.linear.com)
6 OVERVIEW OF ACTIVE ANALOG Mixed Signal Integration, or MSI (http://www.mix-sig.
com), and many others.
FILTER DESIGN METHODOLOGY
Only the design of active analog filters will be described. RELATED ARTICLES
Many authors have described the methodology (Bowron
and Stephenson, 1979; Chen, 1986; Horowitz and Hill, Article 75, Electronic System Building Blocks, Vol-
1980; Huelsman, 1993; Lang, 1987; Ludeman, 1987; ume 2; Article 121, Signals in the Presence of Noise,
McGhee et al., 2001; Rangan, Sarama and Mani, 1983; Volume 2; Article 133, Amplifiers and Filters for DAS,
Schaumann and Van Valkenburg, 2000; Temes, 1985; Volume 3.
Winder, 2002).
In all cases, the design commences with a specification of
the filter frequency characteristics based upon the informa- REFERENCES
tion bearing parameters of the signal, its frequency range,
Bowron, P. and Stephenson, F.W. (1979) Active Filters for Com-
and the likelihood and types of any noise and their sources. munications and Instrumentation, McGraw-Hill, London.
This specification will most likely lead to an ideal filter, also
Chen, W.K. (1986) Passive and Active Filters: Theory and Imple-
called a brick-wall filter. If the specification points in the mentations, John Wiley & Sons, New York.
direction of high-pass, band-pass or notch sections, they are Horowitz, P. and Hill, W. (1980) The Art of Electronics, Cam-
transposed to an equivalent low-pass region for a low-pass bridge University Press, Cambridge, MA.
design. This design is then retransposed using suitable fre- Huelsman, L.P. (1993) Active and Passive Analog Filter Design:
quency domain transformations to gain the desired design. An Introduction, McGraw-Hill Education.
Since ideal filters cannot be realized in practice, a suitable Lang, T.T. (1987) Electronics of Measuring Systems, John Wiley
approximation must be designed on the basis of the four & Sons, Chichester.
main possibilities. These are Ludeman, L.C. (1987) Fundamentals of Digital Signal Processing,
John Wiley & Sons.
• the maximally flat amplitude filters, or Butterworth
McGhee, J., Henderson, I.A., Kulesza, W. and Korczynski, J.
filters; (2001) Measurement Data Handling: Hardware Technique,
• the maximally flat delay filters, also called Thomson ISBN 83 7283 008 8, A.C.G.M. Lodart S. A., Lodz.
filters or Bessel filters; Rangan, C.S., Sarama, G.R. and Mani, V.S. (1983) Instrumenta-
• the equiripple pass-band filter, or Chebyshev type 1 tion Devices and Systems, Tata McGraw-Hill, New Delhi.
filter; Schaumann, R. and Van Valkenburg, M. (2000) Design of Analog
• the elliptic filter, or Chebyshev type 2 filter. Filters, Oxford University Press.
Solopchenko, G.N. (1994) Formal Metrological Components of
Both kinds of Chebyshev filters are considered in the Measuring Systems. Measurement, 13, 1–12.
references given at the end of this article. Temes, G.C. (ed.) (1985) Selected Papers on Integrated Analog
In this way, the three main frequency regions of a low- Filters: Advances in Circuits and Systems, ISBN: 0879422157,
pass filter may be described. If a band-pass, high-pass, or IEEE Publications.
band-notch filter is required, these may be achieved using Winder, S. (2002) Analog and Digital Filter Design, Butterworth-
suitable low-pass-to-band-pass, low-pass-to-high-pass, and Heinemann.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
This article is based in part on a chapter by J. McGhee et al originally published in Measurement Data Handling: Hardware Techniques
Vol. 2, J. McGhee et al , Technical University of Lodz, 2001.
126: Electrical Bridge Circuits – Basic Information
Zygmunt L. Warsza
Polish Metrological Society, Warsaw, Poland
Similarly, after disconnecting two such linked branches
1 Balance in Measurement Circuits – General from the node, no changes will appear because the currents
Remarks 867 are opposite to each other – see Figure 1(b).
2 Basic Bridge Definitions 868 The equations describing balance conditions are usually
3 Wheatstone Bridge and History of Bridge much simpler than those describing the circuit in general
Developments 869 conditions, and, therefore, are often used for measurements.
4 Bridges in Measurement of Nonelectrical An unknown parameter, or combination of a few of them,
Quantities 871 is measurable if it is possible to calculate them from balance
equations as a function of other known values of parameters
5 Bridge Circuits Classification 871
obtained after balancing the circuit.
6 Z Matrix Equations of Two Port 872 Opportunities given by balanced states are used both
7 Impedance Bridge as Two Port and its Z in a single bridge, and also in the more sophisticated
Matrix 873 measurement circuits including a bridge as a part of
8 Basic Equations of Unbalanced Bridges 875 them.
Related Articles 876 In the circuits with time-variable signals, the variety of
References 877 other possible balance states are also applied for measure-
ments, for example, the equality of
Further Reading 877
• node potentials or branch currents as above, but only
in some instants of time, for example, as in successive
1 BALANCE IN MEASUREMENT approximations balance circuits, and A/D converters;
• integrals of two signals: in the different types of integral
CIRCUITS – GENERAL REMARKS A/D converters;
• both, module and phase of two pure sinusoidal signals
In the measurement process, special circuit states, such as
of the same frequency, or of harmonic components of
equality of two nodes’ potentials or of two branch currents
the differently disturbed periodic signals;
connected, for example, in the common node as shown
• only amplitudes, or only phases called the quasi-
in Figure 1(a), are of special significance. Such states are
balanced state;
known as balanced states. Hence, existing relations between
• maximum, mean, or RMS values of two signals of
circuit parameters are called balance conditions.
different shapes.
A circuit could have more than one state of balance.
If the potentials of two nodes are equal, it is possible to Other possible balance states also exist.
connect them together directly, or by some impedance, External and internal voltage and current sources, of
without changing the current and voltage distribution in stationary magnitude or that are adjustable, are used in the
all the circuit branches. measuring circuit.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
868 Elements: B – Signal Conditioning
C C
UCD = 0
D D
I MN = I NK
I MK
N′′
N N′
(a) (b)
Figure 1. (a) Examples of the voltage and current circuit balance states and (b) possible modifications of the circuit.
Adjustable parameters of circuit elements may be C
changed or modulated, either manually or automatically, + −
inside or outside the main circuit, such as in a bridge.
R1 R2
If, in the measurement process, the relationships between
circuit parameters in the balance state are only taken for I0
obtaining measuring results then such a method of mea-
surement is called a balanced or null-balance method. UCD
R0
If changes of parameters result in a different state from
A B
the null-balance one, such a method is a differential one
and the circuit is called deflection, off-balance, or unbal- R4 R3
anced one. IAB
Both methods offer high accuracy and are widely used
D
not only in electrical measurements.
E
Rw
2 BASIC BRIDGE DEFINITIONS
UCD = 0
An electrical measurement circuit including at least one when
quadrilateral topology network having all possible branches R1R 3 = R 2R 4
present is called a bridge circuit, or in short, the bridge.
Usually, such a circuit is excited (driven, powered sup- Figure 2. Wheatstone bridge and its balance condition R1 , . . .
plied) across one of its diagonals and has the output on the R4 – resistance of bridge arms; R0 – resistance of galvanometer
(bridge load resistance RL ); Rw – source internal resistance (series
other one (see Figure 2 as an example of the bridge). resistance to the primary bridge circuit RG ).
Bridges could work as balanced or unbalanced measure-
ment circuits.
Discovery of the various bridge circuits resulted origi- find application. They not only included measurements of
nally from the measurement needs of electricity and mag- parameters of the two terminal components but also their
netism science for an explanation of its historical emer- values in circuits with different structures of lumped and
gence – see also Sydenham (1979). distributed elements.
Later, their development became necessary for the Presented here is the basic unchanging foundational
research and practice of a rapidly growing electrical knowledge useful for setting up and understanding practical
technology, telecommunication, and other disciplines. bridge application.
More accurate measurements of the resistance R and The term bridge is used with two different meanings:
conductance G in the direct current (DC) circuits were ini- for a circuit and for a measuring device utilizing this
tially needed. Later, alternating current circuit parameters circuit as its main part. During recent years, the term
such as resistance R, inductance L, capacitance C, mutual ‘bridge’ has also been used sometimes by some instrument
inductance M, components of impedance Z or admittance producers, for a device measuring the same parameters as,
Y , frequency, and other auxiliary parameters were mea- for example, RLC bridges, but working on a completely
sured in the continuously rising range of frequencies that different measuring principle.
Electrical Bridge Circuits – Basic Information 869
Basic properties of different bridge-like circuits and the • value of applied voltage E;
background of their applications in measurements are pre- • sensitivity of the galvanometer now more usually called
sented in detail in the literature given in the end and in the detector – for details refer to Dally, Riley and
Article 127, Unbalanced DC Bridges, Volume 2. McConnell (1993) and Karandeev (1966).
Detailed technical data of bridge instruments, converters,
The influence of the thermoelectric voltages formed in
and computer measurement systems working with them that
these circuits should be avoided when two measurements
are offered on the market are given in the manufacturer’s
are made for opposite directions of supply. The correct
publications, including the Internet.
result is then found as a mean value of the two readings
that average out the effects.
3 WHEATSTONE BRIDGE AND HISTORY The structure of the Wheatstone bridge was later applied
to the design of several bridge versions for making R, L, C,
OF BRIDGE DEVELOPMENTS
and M measurements.
The history of bridges and their development is some The Wheatstone bridge is also commonly used in off-
170 years old now. The DC bridge principle was first balance measurements. In such a case, the supplying source
introduced by Christy in 1833, but it was unnoticed during E and resistances connected to both diagonals should be set
the next ten years. In 1843, Charles Wheatstone applied to appropriate values and kept stable during the measure-
the bridge arrangement for the comparison of resistors in ments. Then no additional adjustments are needed to obtain
his ‘differential analyzer’ gaining it a place as a basic the result.
electrical circuit. It is, thus, well known since as the The output signal level depends on changes of all arm
Wheatstone bridge. resistances from their values in the balance state. It can
Together with its balance condition for UCD = 0 this be measured by a direct reading instrument or conditioned,
bridge is shown in Figure 2. Resistance R1 is the one to converted, and transmitted to remote display devices, or to
be measured, R2 is the adjustable resistance and auxiliary the computer system.
resistances R3 and R4 are equal and used to set the potential Measurement of the ratio of the unbalanced bridge output
of point D. A battery is connected across A B diagonal. signal and of its supply current or voltage is also used.
A galvanometer (this was originally an electromechanical Not long ago, a quadrilateral structure was also applied in
detector but is now usually an electronic amplifier) ‘bridges’ active bridges with computer controlled voltage sources, for
the C D points and is used to detect the balance of their example, in the Cutkosky bridge.
potentials. No current will flow through the galvanometer The Christy–Wheatstone bridge principle opened the
R0 if R2 is adjusted to be equal to R1 . At that state, the 170-year-long history of bridge circuit developments and
value of R1 follows directly from the obtained value of R2 . their numerous applications in measurement, control, test-
By analogy to the process of mass weighing, resistances ing, monitoring, and diagnostics. The first 15 years of this
R3 and R4 were called by Wheatstone as the arms of the history are covered already in Section 3.
balance. This term, shortened to bridge arms has been used In 1862, Sir Thompson Kelvin developed a ‘double’ –
up to now for all the bridge side branches. six-arm DC bridge for measurements of small values of
Five years later, in 1848, Werner von Siemens developed resistances, lower than 1 . This bridge provided the means
a simple method of changing the range of R1 measurements. for avoiding the influence of the resistance of connecting
He applied unequal arms R3 , R4 with any desired ratio. lead wires and of their changes during measurements. With
The balance condition of the Wheatstone bridge given in this arrangement, small value, high precision four-terminal
Figure 2 was obtained from application of Kirchoff and standard resistances could be easily measured and compared
Ohm’s laws for linear resistances. It is independent of the with high accuracy.
voltage value of source E and of resistances connected The next invention was the application of the bridge
in both diagonals. Siemen’s idea made it possible to use circuit to measure and to compare passive parameters such
a slide-wire or decade adjustable resistor R2 for different as inductance L, mutual inductance M and capacitance C.
ranges of R1 . The bridge ratio R4 /R3 then is changed in The steady and transient states of the bridge circuit were
set sequences, for example, 10n . used together. Firstly, the bridge is balanced in the steady
The range, accuracy, and sensitivity of bridge measure- state by adjusting one of the resistances for the DC supply.
ments are limited by It is then also balanced in the unsteady state by L or C
adjustments of its arms to avoid pulse (AC) movements
• accuracy, values, and permissible power (as balance of the galvanometer working in the ballistic regime. The
currents still flow in all resistors!) loads of R2 , R3 , R4 ; pulses are generated just after repeated commutations of the
• resolution of adjustable resistor R2 ; supplying voltage or current DC (they can be switched off,
870 Elements: B – Signal Conditioning
switched on, or reversed) and are imposed. Such ‘ballistic their properties can be found in the references at the end,
bridges’ were developed, one after another, by de Sauty revised by T.R. Foord, edition of Hague and Foord (1971)
in 1871 – for C measurements; by Maxwell in 1873 – for famous book.
the L, M, LM measurements; by Carry-Foster in 1887 – Parallel to this, classical DC bridges have also been
for the C, M measurements; and also used and improved continuously improved. In the beginning of the second part
by Perry, Ayrton, and Brulion. After the invention of of the twentieth century, analog immittance measurements,
the telephone by Bell in 1875, Lord Rayleigh, Hevisade, provided by DC and AC bridges and by DC potentiometers,
Hughes, Kohlrausch, and others tried to use it in the above obtained levels of accuracy, sensitivity, and measurement
bridges supplied by an induction coil generator, but its ranges not very far from the present day ones. However,
voltage was of unknown shape and not of one frequency. they were not as ‘user friendly’ as the modern instruments,
The principle of the first modern AC bridge was intro- because they needed many manual lengthy adjustments to
duced by Max Wien in 1891 – for C measurements. He obtain balance and much skilled experimental experience.
first used an alternating current supply of defined steady During the last fifty years, new revolutionary techno-
frequency and a ‘vibrating’ galvanometer mechanically logical possibilities have been coming up for use in mea-
adjusted to resonance with this frequency. Its very light surements. This has been made possible by the avalanche
moving part was equipped with a small mirror, and a light of inventions and developments in physics and electron-
beam pointer. ics, such as transistors; new almost perfect ferromagnetic
Wien’s AC bridge, together with such a galvanometer, materials; Zener diodes; solid-state operational amplifiers
was subsequently used for several decades, mainly at fre- (OPAMP)s; analog to digital A/D and digital to analog D/A
quencies up to a few kilo hertz. For higher frequencies, signal converters; integrated digital and analog solid-state
high sensitivity electronic tube (valve) resonance amplifiers circuits; microprocessors; application specific integrated
were developed. circuits (ASIC); huge software possibilities; programmable
Presently, for all frequency ranges, integrated, solid state, digital signal processors (DSP); and so on.
very high sensitivity, adjustable resonance amplifiers, and An important stage in high accuracy AC bridges was
also phase detectors and shifters with operational ampli- that connected with the development of high precision,
fiers are commonly applied in AC bridges. Use of the adjustable induction dividers. They were made on cores
electromechanical indicator has all but become a museum of new ferromagnetic materials of very high permeability,
piece but its performance will still sometimes be useful in very narrow hysteresis loop, and very low losses. These
standard work. offered highly accurate, up to seven decades, AC voltage
The balance state: UCD (t) = 0, in the four-arm AC bridge division dependent nearly exactly on the ratio of the coils’
demands that for every frequency of this voltage there winding turns.
is equality of products of the complex impedances of The accuracy of earlier developed precision resistance
the opposite arms. This results in two real equations for decade dividers is limited, because it depends on the stabil-
equality of the polar or rectangular components of those ity of the resistor material and of their specialized technol-
two products. Here it is not always possible to obtain ogy process. The principle of the transformer bridge was
the solution. shown quite early by Walsh in 1930, but in practice it was
Furthermore, only a few AC bridges have their balance only first applied in 1949. Developments for high precision
conditions independent of the signal frequency. In the measurements have been due to Hill, Miller, and Decon in
balancing process, two, but not any two, circuit parameters the 1962–1968 period.
should be adjusted for a successively smaller and smaller A new idea in DC bridges was introduced by Mil-
readout from the detector. ianic and others in 1962, and by Kuster and Moore in
Thus, in all AC bridges, the process of balancing is much 1964–1966. They made use of the high precision adjustable
more complicated than for DC bridges. induction current comparator. In this device, the zero of
In the early days of development, dozens of AC bridge the DC magnetic flux in a ferromagnetic core is precisely
variation were developed for measurement of AC passive detected due to additional AC coils, as in transductors. They
parameters under different conditions. Between them in gave the possibility of accurate comparing DC ampere turns
1920, Schering introduced a high voltage bridge to measure of a few core windings.
the capacity C of a condenser and its dielectric losses In the book edited by Schnell and Sydenham (1993),
(expressed by tg δ) measurements, used up to now. In 1924, Osvad presented the details of the development of the
Robinson presented a bridge for frequency measurements. transformer bridges and also of active circuits based on
More about the history of the first period of developments operational amplifiers, both dedicated to immittance mea-
of AC bridges and deep studies of many circuits and surements.
Electrical Bridge Circuits – Basic Information 871
Some basic information about bridges are in instrumen- (CallTech) and by AC Rugy (MIT) in 1937. Very soon
tation handbooks, for example, in one edited by Nolting Baldwin Southwork Corp. started the production of metal
(1995) and revised by Boyes et al. (2003), and also in the strain gauges glued on paper background, ready to be stuck
latest one by Dyer (2001). Some valuable information about to a measured stressed object with a special adhesive. For
bridges can also be found in several academic textbooks their applications in static and in dynamic measurements,
and engineering books, including those given below in the special AC resistance bridges were developed.
References and Further Reading at the end of this article. Some other immittance sensors, such as photoresistors
During recent years, research on AC impedance mea- and magnetoresistors, can work in the DC or AC bridges.
surement has focused on the development of virtual mea- Self- and mutual-inductance sensors and capacitance sen-
surement circuits. High accuracy, high speed, and precise sors have to use AC bridges.
DSP processors are used for that; they make it possible to In all industrial and laboratory intelligent transmitters,
build virtual bridges. These circuits could include only two and in measuring cards of computer systems working with
hardware arms: the measured and the reference impedance. immittance sensors, unbalanced bridge circuits of both
Results obtained in the laboratory are comparable with the passive and active forms, are now commonly used. They
best technical data of hardware automatic digital bridges. are the input part of the signal conditioning circuits and still
New algorithms for such measurements have also been offer many properties not given by other circuits.
found, but it seems that the stability of the best R or C Bridge circuits and their theoretical backgrounds are
and M reference elements is still better than that of the still very useful in measurements, mainly for the follow-
A/D converters used as DSP inputs. ing reasons:
• Most physical quantities can be measured using analog
4 BRIDGES IN MEASUREMENT OF immittance sensors connected in the bridge circuit.
NONELECTRICAL QUANTITIES • Influence on sensor parameters of additional, not mea-
sured quantities, can be easily eliminated by using a
differential sensor connected to the half-bridge or two
Applications of the bridges in the measurement of non-
such sensors in the full bridge circuit.
electrical quantities by indirect means based on electrical
• If the measured changes of a sensors parameters are
detection began after immittance sensors of various quan-
very small (e.g. lower than 0.5%, as it is in metallic
tities were developed.
strain gauges), the bridge circuits give accurate out-
Siemens, in 1871, tried to design platinum resistance
put signals.
sensors for temperature measurement, but without success.
• Bridges can also easily operate with multiparameter sen-
Callendar, in 1891, described such a sensor and also for-
sors for simultaneous measurements of a few variables.
mulated the relation of its resistance to temperature.
• The equivalent circuit of four or more terminal sensors
Finally, in 1925, van Dusen developed corrections to this
that is integrated within a bridge circuit is accessible
curve, which has now been standardized. According to these
only from external terminals.
contributions, it became possible to build specialized, labo-
• Bridge circuits are also used in the measurement of
ratory and portable temperature bridges. In recent decades,
tensor components of material properties and of the dis-
electronic analog and digital intelligent industrial transmit-
tribution of multiparameter fields; and in nondestructive
ters have been made with bridges at their inputs.
testing, impedance tomography, and diagnostics of cir-
For recording temperature, Leeds, and Northrop Com-
cuit parameters.
pany, about 1940, designed the first automatic bridge
recorder. A robust electromechanical galvanometer was
Knowledge of bridge circuits and their properties are very
used with periodically operated special scissors closing on
important not only for designers but also for users of instru-
its deflected pointer. They actuated the clutch of a preci-
mentation. Thus, a unified description of bridge circuits
sion slide-wire resistor balancing the bridge step by step.
is needed.
Such an instrument had enough power to move the ink pen
on a slowly moved circular, or strip recording paper. This
instrument was still being used in dangerous areas even up
to 1975. Their new versions with electronic amplifier and 5 BRIDGE CIRCUITS CLASSIFICATION
servomechanism were applied later and it is still possible
to meet them in industry or in laboratories. Bridge circuits are commonly called by the names of
Thompson discovered the piezoresistive effect in 1856, their inventors; bridge instruments – by the names given
but it was applied firstly in measurements by Simmons by their manufacturers; or according to the field of their
872 Elements: B – Signal Conditioning
applications, or parameters to be measured. A large diver- for AC of one frequency; Fourier transforms for few fre-
sity of bridge circuits, especially AC ones, demands that quencies; Laplace transforms for any shape of signals in
there is a clear criteria of their classifications. Some of time; and partial differential equations for the distributed
them are presented below. Bridges could be classified parameter circuits.
according to For a single supply and single output set up, any bridge
could be transformed to a two port (two-terminal-pair)
• static characteristic of the parameters (validity of the use
circuit – see Figure 3.
of the superposition law in the circuit): linear, nonlinear
A set of equations of any linear circuit working as a two-
(for at least one parameter), being quasi-linear around
port device could be reduced to a set of only two linear
the working for relatively small excursions;
relations between currents and voltages at their ports.
• type or number of supplied sources:
It is thus possible to write these relations in six different
– DC and AC of different frequency ranges,
ways; and for simplicity in matrix form. Then the one of
– current or voltage sources,
the following matrices expresses the two-port parameters:
– low or high voltage,
Z, Y, A, B, H, H . If it is needed, anyone of them could
– one, or several sources,
– stationary and controlled sources; be easily transformed to another one.
• type of the bridge circuit: For a particular circuit, some of the above matrices will
– passive, active; not exist at all or their parameters are infinite in some states,
• number of branches: for example, parameters of the transmission matrix A of
– four, five, or six arms, and so on. the balanced bridge – see Hague and Foord (1971). More
• number of external terminals: valuable information about the two-port circuit theory can
– four, six terminal, and so on. be found in academic textbooks on electrical circuits, for
• number of output: example, such as the classic one by Carter and Richard-
• number of measured parameters; son (1972).
• number of variable arms; Impedances are used commonly to describe the bridge
• variability of the circuit structure: arms. Therefore, the Z matrix will be taken for further
– invariable, switched; consideration as the main one. Also, it is very important
• type and range of measured parameters: that this matrix exists in the balanced states of the bridge,
R, RC, RLC bridge, low impedance, high impedance, but matrix A does not.
and so on; The linear two-port device with all typical symbols is
• type of application: high precision, laboratory, portable; given in Figure 3. An additional assumption is taken that
• accuracy of measurement of different parameters for there are no stationary sources inside a two-port circuit. If
different ranges in percents; it is not so, the influence of such sources is eliminated from
• field of application: measurement considerations by short-circuiting the voltage
– temperature bridges, strain gauge bridges, conduc- ideal sources and disconnecting the current ones. They can
tometric bridges, and so on. also be taken into account in the next steps of the bridge
circuit analysis, for example, by the superposition method.
AC bridges can also be classified according to the phase The Z type matrix equation for a two-port device is the
relations of circuit arms. following:
U = ZI (1)
6 Z MATRIX EQUATIONS OF TWO PORT
In the general case, a bridge circuit can be consid-
Z1 I1 I2
ered as multinode, multiloop, and multiterminal [n-T] net-
work in which, for obtaining measurement result, its one, Z in Z out
J Z1 E U1 Z U2 Z2
or several branch parameters are changed or modulated
externally.
The two Kirchoff Laws are valid in any case, at any time
and any part of the circuit having negligible electromag- Current Voltage Input Output Load
or
source source port port
netic propagation. The circuit is linear if the superposition
theorem is valid for it. It can then be described by a set Figure 3. Two port network: standard conventions of currents,
of linear equations: algebraic ones for DC; complex form voltages, and impedances.
Electrical Bridge Circuits – Basic Information 873
or in expanded form Zout to the equation of voltage U2 we obtain
U z z12 I1 Z1 Z2 z21
1 = 11 (1a) U2 = J (1c)
U z z22 I2 (Z1 + z11 )(Z2 + z22 ) − z12 z21
2 21
In the algebraic form it is: Then, if zij elements were known from calculations or
terminal measurements for a particular bridge working as
U1 = z11 I1 + z12 I2 the two-port, equations of Table 1 would be a useful tool
to analyze it as a passive or active electrical linear circuit
U2 = z21 I1 + z22 I2 (1b)
with variable parameters.
In the general case, all impedance coefficients zij are dif-
ferent. A linear two-port circuit could include different, but
only linear, two- and more terminal passive elements, for
7 IMPEDANCE BRIDGE AS TWO PORT
example, R L C single or coupled impedances, transform- AND ITS Z MATRIX
ers, and so on, and only sources controlled proportionally
by the circuit’s voltages or currents. As it was mentioned earlier, four-terminal bridges including
If the two-port is nonreciprocal, then z12 = z21 holds. only linear elements, could operate as linear two-ports.
For only one source supplying input or output port, the If the bridge is a part of the larger circuit, it can be
two-port has two different balance states on its terminals: considered as a two-port set up, only when it has two pairs
z21 = 0 and z12 = 0. of terminals with currents opposite to each other in each
If z12 = z21 , then the Z matrix has only three independent pair. It means that currents in connections around the bridge
impedance elements. that is, between two terminals, each of a different pair, are
If external circuits connected to both ports are also linear, negligible, or their influence additionally balanced. If it is
then according to the Thevenin (after Carter and Richard- not so, the bridge should be treated as a four-terminal (4-T)
son (1972) this theorem first developed by H. Hemholtz 30 or generally an n-terminal (n-T) circuit.
years earlier) or Norton theorems, they have linear equiv- Any four arms being used in a linear impedance bridge
alent circuits. Such a one is shown for the primary side of is the X-type two-port. It is shown in Figure 4. Designa-
the two-port of Figure 3. tions of the two-port in Figure 3 are here changed to the
Input and output equivalent impedances, terminal cur- ones used in bridges, similar to the Wheatstone bridge in
rents, and voltages under different load Z2 and source Z1 Figure 2.
impedances are given in Table 1. The impedance matrix Z of the linear passive bridge is
These relations are valid for all linear two-ports including given below
bridge circuits. If elements of circuits connected to any side
(Z1 + Z2 )(Z3 + Z4 ) Z1 Z3 − Z2 Z4
of the two-port are of known nonlinearities, it is possible
also to find the terminal currents and voltages of this two- Zi Zi
Z= (2)
port by using its equations for each value of supply voltage Z Z − Z Z (Z + Z )(Z + Z )
1 3 2 4 1
4 2 3
E or current J .
Zi Zi
The practical meaning of the Z matrix elements is appar-
ent from the equations of Table 1. It is seen that element
z11 is the input impedance Zin of the two-port if the output where Zi ≡ Z1 + Z2 + Z3 + Z4 .
circuit is open, and z22 is the output impedance Zout when
Z1 → ∞. ZG IAB A Z4 D
Element z12 and z21 of the two-port matrix Z are current
I DC
to voltage transmittances, respectively in the reverse and Z1 Z3
J ZG
forward direction of transmission when the output port E UAB U DC ZL
is open. Z2
It is possible to find all elements of the Z matrix
B C
from two-port terminal parameters measured with an open
or short-circuited output port that is, Z2 = ∞, Z2 = 0 or Current
or
Voltage Bridge
source source
Z1 → ∞, Z1 = 0.
Equations of Table 1 could also be presented in different Figure 4. Notations of four-arm impedance bridge as X type
forms. For example, after inserting equations for Zin and two port.
874 Elements: B – Signal Conditioning
Table 1. Definitions and basic equations of terminal impedances, currents and voltages of linear two port under work of linear load
and supply source.
Parameters and General case Particular cases
basic relations ∞ > Z1 > 0, ∞ > Z2 > 0 Z1 = ∞ Z1 = 0 Z2 = ∞ Z2 = 0
U1 z12 z21 det Z + Z2 z11 ∞ Zout + Z2
0
∞ det Z
Zin ≡ z11 − = = Zin = z11 0
=
Input
Port impedance
∞ Zin Zin
I1 z22 + Z2 z22 + Z2 Zout + Z2 z22
∞ Zin + Z1 det Z + Z1 z22
0
Output
U2 z12 z21 ∞ det Z z12 z21
Zout ≡ z22 − = Zout ∞ Zout = z22 0
Zout = z22 − =
I2 z11 + Z1 Zin + Z1 z11 z11 + Z1 z11 + Z1
E Z2 E z21 Z2 E z21
Supply source E or J in primary port
z21 0 0
Zin + Z1 Zout + Z2 0
Zin Zout + Z2 z11 + Z1
U2
Z1 Z2 z21 Z2 Z1 z21
Output variables
J z21 J 0 J 0
Zin + Z1 Zout + Z2 z22 + Z2 z11 + Z1
E z21 E z21 E z21
0 0
Zin + Z1 Zout + Z2 Zin Zout + Z2
0 0
Zin + Z1 Zout
−I2
Z1 z21 z21 Z1 z21
J J 0 0 J
Zin + Z1 Zout + Z2 z22 + Z2 0
Zin + Z1 Zout
Supply source E or J in primary port
EZin z11 det Z
0 E E E
Zin + Z1 z11 + Z1 det Z + Z1 z22
U1
Zin Z1 z11 Z1 Z1 det Z
Input variables
J J Zin 0 J J
Zin + Z1 z11 + Z1 det Z + Z1 z22
E E E E z22
0
Zin + Z1 Zin Z1 + z11 det Z + Z1 z22
I1
Zin Z1 Z1 Z1 z22
J J 0 J J
Zin + Z1 z11 + Z1 det Z + Z1 z22
Note: Where det Z = z11 z22 − z12 z21 .
It is a symmetric matrix, that is, z12 = z21 , because the Balance condition could be also expressed in ratio forms
circuit is reciprocal. This means that open circuit output for example,
voltages UDC and UAB are equal when the bridge is supplied
by the same value current IAB = IDC at primary AB or Z1 Z Z Z
= 4 or 1 = 2 (3b) or (3c)
secondary DC port. Then the balance conditions for UDC = Z2 Z3 Z4 Z3
0 or UAB = 0 for both ways of supply are equal and the
following applies: Input and output bridge diagonal impedances are, in the
general case, different. They depend mainly on the bridge
Z1 Z3 − Z2 Z4 = 0 (3) arm impedances and to a smaller degree also on the
Or for arm admittances Yi = 1/Zi impedance connected to the opposite diagonal. Only in the
balanced state they are independent of the last ones, as it
Y1 Y3 − Y2 Y4 = 0 (3a) results from Figure 1(a).
Electrical Bridge Circuits – Basic Information 875
If the supply source E or J , as well as ZL and ZG are Z(ε) – above matrix expressed by relative increments of
given, the output voltage UDC of the loaded bridge as two- impedances and by the initial sensitivity t0 and unbalance
port (see equation (1c)) is function f (εi ) = L(εi )/M(εi ) of the output voltage given
below.
ZG ZL z21 All bridge parameters corresponding to the balance state
UDC = I (4)
(ZG + z11 )(ZL + z22 ) − z21
2
have now additional subscript 0. Then the arm impedances
are
Inserting in (4) the values of bridge matrix elements
given in (2) we obtain: Zi ≡ Zi0 + Zi = Zi0 (1 + εi ) (9)
Z Z (Z Z − Z2 Z4 ) Zi
UDC = J G L 1 3 (5) Zi ; εi ≡ – absolute and relative increments of
ZG ZL Zi + ZL (Z1 + Z2 )(Z3 + Z4 ) Zi0
+ZG 1 + Z4 )(Z2 + Z3 ) + A
(Z impedances Zi , from their values Zi0 in the
balance state;
Where: Z10 Z30
t0 ≡ –initial voltage sensitivity per one unit
Zi0
A= Zi Zj Zk = Z1 Z2 (Z3 + Z4 ) of current IAB of the open output of the
bridge;
+ Z3 Z4 (Z1 + Z2 ) = · · ·
L(εi )
f (εi ) = – voltage unbalance relative function
Moving the product of ZG ZL to the denominator we get: M(εi )
of bridge impedance increments and
Z1 Z3 − Z2 Z4 L(εi ) = ε1 − ε2 + ε3 − ε4 + ε1 ε3 − ε2 ε4 – numerator and
UDC = J (6)
1 Zi εZ
Zi + (Z + Z2 )(Z3 + Z4 ) M(εi ) = 1 + = 1 + i i0 – denominator of
ZG 1 Zi0 Zi0
1 A the above function.
+ (Z1 + Z4 )(Z2 + Z3 ) +
ZL ZG ZL
The following additional relations of arm initial impedances
From the equations of lines 1 and 2 in Table 1 it is also to the value of Z10 are defined as
possible to find Zin and Zout impedances of the analyzed
bridge for different values of source impedance ZG and
Z20 ≡ mZ10 ; Z40 ≡ nZ10 ; Z30 ≡ mnZ10 (10)
of the load one ZL – see Article 127, Unbalanced DC
Bridges, Volume 2.
Two-port matrix Z parameters of the bridge, in the above
new notation, are presented in Table 2.
8 BASIC EQUATIONS OF UNBALANCED The output voltage of the loaded bridge, given by (6)
also could be written now as
BRIDGES
L(εi )
In the analysis of bridges in off-balance conditions and UDC ≡ T0 (11)
ML (εi )
properties of unbalanced (deflection) bridges, it is conve-
nient to express matrix Z as a sum of two components T0 ≡ J Z10 Z30 /MU J 0 – initial voltage sensitivity of
loaded bridge;
Z = Z0 + Z = Z0 + Z(ε) (7) MU J ≡ MU J 0 + MU I (Zi ) – denominator in (6)
of UDC .
or in the expanded forms: L(εi ) = ε1 − ε2 + ε3 − ε4 + ε1 ε3 − ε2 ε4 –
numerator as before,
Z 0 ZAB z21
Z = AB0 + ML (εi ) ≡ 1 + MU J (Zi )/MU J 0 – new denominator of
0 ZCD0 z21 ZCD the voltage relative unbalance function of the
Z 0 εAB ZAB0 t0 f (εi ) loaded bridge with a nonideal supply source.
= AB0 + (8)
0 ZCD0 t0 f (εi ) εCD ZCD0
Increment MU J (Zi ) of the denominator is a com-
Where Z0 − Z matrix in the bridge balanced state; Z – plicated function of the balance impedances and of their
matrix of absolute increments zij from this state; increments for general cases of value ZG and ZL , because:
876 Elements: B – Signal Conditioning
Table 2. Parameters of matrices Z, Z0 , Z (εi ), Z of the four-arm linear impedance bridge as two port.
Matrix
Bridge
z11 z12 = z21 z22
No
state
In balance
Z40 (Z10 + Z20 ) n(1 + m) Z10 Z30 − Z20 Z40 Z10 + Z40 m(1 + n)
1 Z0 ZAB0 = Z10 =0 ZCD0 Z20 = Z10
Z10 + Z40 1+n Zi0 Z10 + Z20 1+m
(Z1 + Z2 )(Z3 + Z4 ) Z1 Z3 − Z2 Z4 ZCD (Z1 + Z4 )(Z2 + Z3 )
2
ZAB Zi Zi (ZG = ∞) Zi
(ZL = ∞)
Z or or
3 ZAB0 [1 + εAB ] ZCD0 [1 + εCD ]
or
εAB 1 nε12 ε43 εCD 1 mε14 ε23
+ + ε12 ε43 + + ε14 ε23
(ZL = ∞) M(εi ) 1+n 1+n (ZG = ∞) M(εi ) 1+m 1+m
Out of balance
ε1 + mε2 L(εi ) ε1 + nε4
4 ε ε12 ε12 = t0 ε14 ε14 =
1+m M(εi ) 1+n
ε4 + mε3 ε2 + nε3
ε43 ε43 = ε23 ε23 =
1+m 1+n
ZAB0 ZCD0
Zi0 + Zi Zi0 + Zi
1 1
ZAB n(Zi + Z2 ) + (Z3 + Z4 ) ZCD m(Z1 + Z4 ) + (Z2 + Z3 )
5 Z
(ZL = ∞) ×
n (ZG = ∞) × m
(Z1 + Z2 )(Z3 + Z4 ) (Z1 + Z4 )(Z2 + Z3 )
+ +
ZAB0 ZAB0
T0 Z20 Z40 Z10 mn
Note: Where initial bridge output voltage to current sensitivity – t0 ≡ = = ,
J Zi0 (1 + m)(1 + n)
numerator of unbalance function – L(εi ) = ε1 − ε
2 + ε3 − ε4 + ε1 ε3 − ε2 ε4 ,
Zi ε12 + nε43 ε14 + mε23 ε1 + mε2 + n(ε4 + mε3 )
denominator of unbalance function – M(εi ) = 1 + =1+ =1+ =1+ .
Zi0 1+n 1+m (1 + m)(1 + n)
∂M ∂ 2M It is also the case if some initial bridge arm impedances
MU J (Zi ) = UJ
Zi + UJ
Zi Zj are equal, or not changing and if pairs of the increments,
∂Zi i =j
∂Zi ∂Z j
or all of them, are of the same absolute value but of the
∂ 3 MU J opposite signs in the neighboring arms. Such cases are very
+ Zi Zj Zk (12) common in practice. Detailed discussion of these cases are
i =j =k
∂Zi ∂Zj ∂Zk
i =k in books by Bentley (1995), Dally, Riley and McConnell
(1993), Holman (1994), Tattamangalam (2000), Schnell and
There are 14 elements to add together (4 + 6 + 4). If Sydenham (1993), and also in Article 127, Unbalanced
modules of related resistance increments |εi | are not too DC Bridges, Volume 2.
high, the last sum in the above equation, dependant on the
products of the three of them, can be neglected; and also
for εi 1, the second sum is negligible. RELATED ARTICLES
Further simplification of (6) and (11) is obtained assum-
ing ideal J or E supply sources |ZG | |ZAB | or |ZG | Article 56, Transfer Characteristics of Instrument
|ZAB | respectively, for ideal voltage output |ZL | |ZCD | Stages, Volume 1; Article 112, Systematic Description
or ideal current output |ZL | |ZCD |. of Sensors, Volume 2; Article 127, Unbalanced DC
Electrical Bridge Circuits – Basic Information 877
Bridges, Volume 2; Article 202, Current Measurement, Sydenham, P.H. (1979) Measuring Instruments; Tools of Knowl-
Volume 3; Article 204, Capacitance and Inductance edge and Control, Peter Perigrinus.
Measurement, Volume 3 Tattamangalam, R.P. (2000) Industrial Instrumentation Principles
and Design, Springer-Verlag, London (pp. 29–51).
REFERENCES
This list extends the cited references given in the text above.
Bridges were of strong interest to publishers until the 1950s FURTHER READING
after which, despite, their still continued use, titles on the topic
twindled. Some of the best literature in bridges is to be found in Gregory, B.A. (1973) An Introduction to Electrical Instruments,
texts now considered as too old to be relevant by many! Macmilan Press Ltd, London (pp. 88–116).
Hai Hung, C. (1984) Electrical and Electronic Instrumentation,
Bentley, J.P (1995) Principles of Measurement Systems, 3rd edn, John Wiley & Sons, New York (pp. 48–92).
Longman Scientific & Technical, Longman Groupe UK Ltd,
Karandeyev, K.B. (1966) Bridge and potentiometer methods in
Harlow, Essex (pp. 178–187).
electrical measurements, Energia Moscow (1966) and Peace
Boyes, W., Parr, M. and Noltingk, B.E. (eds) (2003) Instrumenta- Publisher MIR (1967) (Eng. transl. from Russian edition
tion Reference Book, 3rd edn, Butterworth-Heinemann, Elsevier 1960).
Science, Boston (pp. 463–479).
McGhee, J., Korczyński, M.J., Henderson, I.A. and Kulesza W.
Carter, G.W. and Richardson, A. (1972) Technique of Circuit (1998) Scientific Metrology, 2nd edn Lodard S.A., Lodz
Analysis, Cambridge University Press, London (pp. 17–22, (pp. 205–217, 289–297).
248–281).
Morris, A.S. (1993) Principles of Measurement and Instrumenta-
Dally, J.W., Riley, W.F. and McConnell, K.G. (1993) Instrumen- tion, 2nd edn, Prentice Hall, New York (pp. 106–127).
tation for Engineering Measurements, 2nd edn, John Wiley &
Noltingk, B.E. (ed) (1995) Instrumentation Reference Book,
Sons, NewYork (pp. 170–176, 220–236).
Part 3, Butterworth-Heinemann, London (pp. 40–57).
Dyer, S.A. (ed) (2001) Survey of Instrumentation and Measure-
Oliver, B.M. and Cage, J.M. (1971) Electronic Measurements and
ment, Wiley-Interscience, John Wiley & Sons, New York (Nico-
Instrumentation, McGraw-Hill, New York (Chapter 9 by H.P.
las, J., Section: Bridge Instruments, pp. 309–326; Kumar, V.J.,
Hall, pp. 264–319).
Murti, V.G.K. and Sankaran, P., Sections: Methods of Mea-
surement of Self-inductance and . . . of Mutual Inductance, Sydenham, P.H. (ed) (1983) Handbook of Measurement Science,
pp. 352–357; Eren, H., Section: Capacitance Measurements, Vol. 2, Wiley-Interscience Publications, John Wiley (Chap-
pp. 369–371). ter 20.6 by L. Schnell).
Hague, B. and Foord, T.R. (1971) Alternating Current Bridge Webster, J.G. (ed) (1996, 2000) The Measurement, Instru-
Methods, 6th edn, Pitman Publishing, London. mentation and Sensors Handbook, CRC Press London,
Springer, IEEE Press New York; Eren, H. and Goh J.,
Holman, J.P. (1994) Experimental Methods for Engineers, 6th edn,
Chapter 45: Capacitance and capacitance measurements,
McGraw-Hill International, New York (pp. 136–144).
pp. 45-23–45-27; Misra D.K., Chapter 46: Permitivity mea-
Schnell, L. and Sydenham, P.H. (eds) (1993) Technology of Elec- surement, pp. 46-6–46-7; Szyper M., Chapter 50: Inductance
trical Measurements, Series on Measurement Science, John measurement, pp. 50-8–50-13; Dreher A., Chapter 51: Immi-
Wiley & Sons (Chapter 4 by P. Osvath, pp. 162–218). tance Measurement, pp. 51-11–51-15.
127: Unbalanced DC Bridges
Zygmunt L. Warsza
Polish Metrological Society, Warsaw, Poland
and of Analog Devices Co.). They use various differential
1 Introduction 878 operational amplifiers, feedback loops, single and double
2 Four-terminal (4T) DC Circuits 878 stabilized supply sources, many methods of switching (e.g.
3 Basic Equations of the Four-arm Resistance integrating or successive approximation AD converters).
Bridge (4R) as Twoport 879 Ratio measurements are also applied when analog cir-
cuit and AD converter are supplied by a common source.
4 Simplified Bridge Equivalent Circuits 884
However, classic unbalanced DC bridges are still most com-
5 Relations Between Open Circuit Terminal monly used either on their own or in above mentioned
Parameters of the 4R Bridge 884 circuits for measurement of resistance and conductance for
6 Bridge Open Circuit Terminal Parameters’ their merits and simplicity. They work most often with
Linearization 885 nearly unloaded (open circuit) voltage output, sometimes
7 Terminal Parameters of the DC Bridge of with current output loaded with resistance or short circuited,
any Supply Source and Load 886 and they deliver measurement signals for amplifying and
Related Articles 889 converting them to the digital form.
References 889 Furthermore, measured objects as well as multiterminal
Further Reading 889 solid state and integrated semiconductor sensors can also
include inseparable bridge structures in their equivalent cir-
cuits. Parameters of this structure are measurable indirectly
only on terminals of the circuit or on electrodes used to
1 INTRODUCTION check material properties.
Below, properties of the four terminal (4T) unbalanced
Bridge circuits are still commonly used in the input of auto- resistance bridge networks are analyzed in detail. The rel-
nomic immitance measurement instruments and transmitters evant knowledge in design and understanding of measure-
of industrial systems. It is so for the following reasons: ment bridge circuits and in programming digital circuits
• They are either directly sensitive to changes of immi- connected with them is required. Some new developments
tance components or else they are indirectly sensitive have also been introduced. A short history of measuring
to other variables actuating the immitance of ana- bridges and their basic equations have already been given
log sensors. in Article 126, Electrical Bridge Circuits – Basic Infor-
• They perform some simple signal-conditioning func- mation, Volume 2.
tions, such as dependence of output signal on changes
of parameters of several sensors easily and accurately.
2 FOUR-TERMINAL (4T) DC CIRCUITS
During the last decades, many new input signal-condi-
tioning circuits have been developed (see – Anderson Four-terminal circuits (4T) could have arbitrary currents
(1997), Kester (1999), Zakrzewski (2000), printed catalogs of three terminals and the fourth one determined by Kir-
and Internet information of National Instruments (2004) choff’s Law. If there are two pairs of opposite terminal
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Unbalanced DC Bridges 879
C C C
0′
R C0′ R ′2
R1 R2 R ′1 R ′2 R A0′ 0′
R5 R C0
R6 A B
A B A R A0 R B0 B R 00 R B0
0 R D0
R4 R3
R D0
(a) D
D (b) (c) D
Figure 1. Examples of 4T circuit, every one equivalent to each other: (a) without nodes inside – four arms bridge with diagonals;
(b) with one node inside; (c) with two nodes inside.
currents, then the 4T circuit behaves as the twoport – see, Resistance or conductance 4T circuits could be applied
for example, Carter and Richardson (1971). Only passive in measurements in different ways. In addition to measure-
and reciprocal 4T circuits are taken below for analysis. ments of branch immitances, they could be used for indirect
Their basic equivalent circuit has six immitance branches, measurements of one and theoretically up to six variables
which are connected to the quadrilateral network given for differently influencing these immitances.
DC on Figure 1(a). In measurements, a bridge could be supplied from voltage
Terminals are also the only nodes in this network, and it or current sources, ideal ones, or of the particular inter-
has the structure of the four-arm bridge with two diagonals. nal resistance. The classic type supply is connected to one
There is a direct relation of the admittance of each branch of the bridge diagonals and the output is taken from the
and the value of the off-diagonal element of the circuit second one. Two sources could be connected in different
admittance matrix with the sign ‘plus’. Any of four-terminal ways, for example, voltage sources, one in the diagonal
circuits, which are more complex inside, can always be and the second one in the bridge arm or in the output
transformed to the above network or to some of its sim- diagonal. Both the ones in the bridge branches, also have
pler structures. been used. Some years ago, this author proposed to use
In Figure 1(b) and (c), there are shown two other net- unconventional power supply for the bridge by providing
works of six branches equal to the first one. They had orig- two equal and ideal current sources connected in paral-
inally been developed by the author for the reciprocal part lel to opposite bridge arms, or by only one such source
of the Hall devices’ equivalent 4T network to find zero cor- switched between these arms and averaging of two mea-
rection circuits that were stabile in a wide range of temper- surement results obtained. Such a bridge has two outputs
atures – see Warsza (1980). The network in Figure 1(b) has from both diagonals and offers original metrological possi-
one node inside. It was obtained from circuit in Figure 1(a) bilities. It could be used simultaneously for two and more
by circuit substituting two smaller resistances, each one variable measurements – see Warsza (2002), (2004) and the
of two smaller opposite resistances’ products, for exam- literature list in it.
ple, R1 and R2 by two parallel ones respectively R1 , R1
and R2 , R2 . So, that is R1 R3 = R2 R4 = R5 R6 and to the
network of these six resistances the Rosen’s theorem of the
3 BASIC EQUATIONS OF THE
four-arm star-delta inverse transformation is applicable. The FOUR-ARM RESISTANCE
last network, given in Figure 1(c) has two nodes inside. It BRIDGE (4R) AS TWOPORT
is obtained from the previous one by transformation of ele-
ments of the triangular ACO to the equivalent star. There In Figure 2, the four-arm resistance bridge circuit (4R)
are six different networks of two serial immitances con- classically powered from current or from voltage source is
nected to neighboring nodes of the inside bridge and they given. This bridge appears under the name of Wheatstone
are equivalent to each other. All bridges of networks of bridge. It is drawn here as the twoport.
Figure 1 are in balance together. In the general case, the bridge input current IAB depends
The four-arm bridge (4R) is a particular case of the on supply source parameters J and RG , or E and RG
above networks of six immitances. It has no other four and also on the equivalent input bridge resistance RAB .
branch circuits that are equivalent to it. Then, in the circuit The last one is a function of actual values of all bridge
theory it plays an analogous role to the prime number in resistances Ri and of its load RL , if any. Then, for a
mathematics. full description of the bridge from Figure 2 as a linear
880 Elements: B – Signal Conditioning
RG IAB A C
R1 R2 IDC
RG UAB + − U DC
E J RL
− +
R4 R3
B D
Figure 2. The four-arm bridge as twoport, together with the single supply source J or E of the internal resistance RG and the load
RL , both connected to its diagonals.
circuit of variable parameters, twoport equations could be Because of the equivalence of both directions’ bridge
used – see such bridge equations of the impedance type transmittances (or transfer resistances) r21 = r12 , only three
given in Article 126, Electrical Bridge Circuits – Basic elements of the matrix ZR are different, but they depend
Information, Volume 2 and of the particular case when on four bridge circuit resistances. Values of these elements
all parameters are real, and matrix Z = Re(Z) ≡ ZR . In are always finite, including balance state when r21 = 0 and
generalized form, it is
R1 R3 = R2 R4 (2a)
U = ZR I
Let us consider a general case of the bridge when vari-
and after expansion of matrices: able resistances are in all arms. As it was introduced in
∞ Article 126, Electrical Bridge Circuits – Basic Informa-
U r r12 IAB tion, Volume 2, analysis of the circuit of Figure 2 should
AB = 11 r12 IAB ≡ RAB (1)
U r ICD r21 ∞
ICD be more general and easier if resistances are referenced to
DC 21 r22 RCD
one of them, for example, to R10 , and their changes – to
where current ICD = −IDC , initial values of resistances in the balanced state. Then,
∞ ∞
r11 ≡ RAB , r22 ≡ RCD – input and output resistances of Ri ≡ Ri0 + Ri ≡ Ri0 (1 + εi ) ≡ ri0 R10 (1 + εi ) (3)
twoport when the opposite side port is opencircuited.
r12 , r21 – current to voltage both direction transmit- where Ri0 – initial (in balance) value of the Ri resistance;
tances (transfer resistances) when the opposite side port Ri , εi – are absolute and relative increments;
is open circuited (RG → ∞, or RL → ∞). (Superscript ri0 – relative values of arms’ resistances
∞ or 0 is used here for open or shorted circuit of the in balance conditions.
opposite side of the bridge and subscript 0 – for the ∞
The open circuit output voltage UDC is given by the
balance state).
equation
Formula (1) in algebraic form and with the opposite
∞ IAB (R1 R3 − R2 R4 )
direction output current IDC as shown in Figure 2, is UDC = IAB r12 = (4)
Ri
∞
UAB = RAB IAB − r12 IDC
∞ (1a) where Ri = R1 + R2 + R3 + R4
UDC = r21 IAB − RCD IDC
With the notations given by formula (3) and after trans-
Matrix ZR of the four-arms resistance bridge is: formation of (4), one gets
r20 r40
(R1 + R2 )(R3 + R4 ) R1 R3 − R2 R4 ∞
UDC = IAB R10
ri0
Ri Ri
ZR ≡
R1 R3 − R2 R4 (R1 + R4 )(R2 + R3 ) (ε1 − ε2 + ε3 − ε4 + ε1 ε3 − ε2 ε4 )
×
Ri Ri ri0 εi
(2) 1+
ri0
where Ri = R1 + R2 + R3 + R4 – sum of resistances of
bridge main loop A B C D; ≡ T0 × f (εi ) (5)
Unbalanced DC Bridges 881
where T0 ≡ IAB t0 – initial output voltage sensitivity to the where
unbalance function f (εi ). – input and output bridge resistances in the balance state:
∞
When the bridge is in balance, output voltage UDC =0
n(1 + m)
and balance condition (2a) could be now written as: RAB0 = R10 (6a)
1+n
R30 = r20 r40 R10 ; (5a)
and
Balance state of the bridge is the property of this par- m(1 + n)
RCD0 = R10 (7a)
ticular network. Output signal is equal to zero only if the 1+m
bridge is supplied from the single source connected to one – relative increments of RAB and of RCD from the balance
of its diagonals. state:
If, for more simplicity of the notation, we put down addi-
tionally: R20 ≡ mR10 , R40 ≡ nR10 , then it is R30 = mnR10 ∞ 1 nε12 + ε43
εAB = + ε12 ε43 (6b)
and initial open circuit sensitivity of transmittance r21 is 1 + εR εi 1+n
R10 R30 mn and
t0 ≡ = R10 (5b)
Ri0 (1 + m)(1 + n)
∞ 1 mε14 + ε23
εCD = + ε14 ε23 (7b)
1 + εR (εi ) 1+m
and unbalance function of r21 :
(ε − ε2 + ε3 − ε4 + ε1 ε3 − ε2 ε4 ) – relative increment of the sum of Ri + Rj resistances of
f εi = 1 i, j arms;
ε1 + mε2 + n mε3 − ε4
1+ Ri + Rj
(1 + m)(1 + n) εij ≡ (8)
Ri0 + Rj 0
Lεi
≡ (5c)
1 + εR εi – relative increment of the sum Ri of bridge resistances:
where L(εi ); εR . – increments of its numerator ε12 + nε43 ε + mε23
εR ≡ = 14 (9)
and denominator. 1+n 1+m
Ri0 – value of Ri of the balanced bridge.
Matrix ZR from (2) could now also be written in a
From (5) and (5c), it is clear how output open-circuited
∞ different form as
voltage UDC and transmittance r21 depend on signs and ∞
values of relative increments εi of resistances of particular R 1 + ε∞ t0 f εi
ZR = AB0 AB ∞
∞
(10)
bridge arms. These increments of opposite signs in neigh- t0 f εi RCD0 1 + εCD
boring arms, given in Figure 2, unbalance the bridge in the
same direction. If absolute values of these increments are In Tables 1(a) and (b), are given elements of the matrix
the same, the output voltage is proportional to the number ZR of the resistance bridge 4R operating as the twoport
of variable arms (e.g. multiplied by 2 or by 4). X. In subsequent lines, there are open circuit terminal
∞ ∞
Transmittance r21 and its unbalance function f εi the- parameters r21 , RAB , RDC and their components. The first
oretically could take values from the range (−∞, +∞). In column of Table 1(a) corresponds to the general case of
practice, there are some limitations due to existing or per- the bridge. In columns a to d of this table and in columns
mitted extreme values of εi , maximum dissipated powers of e to j of Table 1(b), there are few particular cases depend-
arms and maximum voltage of the current source or maxi- ing on
mum current of the voltage source. Transmittance r21 = 0, • the number of variable arms,
if f (εi ) = 0. It could happen for many different combina- • the relations between their increments, and
tions of εi . The basic balance state is defined in such a one • the relations between initial arm resistances (in bal-
as all εi = 0. ance).
Open circuit bridge terminal resistances are
On the basis of these tables, it is possible to find terminal
∞
∞
1 + ε12 1 + ε34 parameters of the four-arm bridge in any kind of its oper-
RAB = RAB0 1 + εAB = RAB0 (6) ation as twoport. Some particular cases of unbalanced 4R
1 + εR
bridge are discussed in detail by Dall (1992), Karandeev
∞
∞
1 + ε14 1 + ε23 (1966), Morris (1993), Padmanabhan (2000), Sanderson in
RCD = RCD0 1 + εDC = RCD0 (7)
1 + εR Bentley (1995) handbook, Warsza (2004) and by others.
Table 1a. Matrix ZR elements of the resistance bridge as the linear twoport X of four variable arms – general case and its four particular cases – only two arms or single
882
arm variable.
General case Particular cases
variable: C (a) variable: R1 , R2 (b) variable: R1 , R4 (c) variable: R1 , R3 (d) variable: R1
No Bridge matrix ZR elements R1 R2
R1 = R10 (1 + ε1 ) C
(terminal parameters) B
Symbol
R2 = mR20 (1 + ε2 ) A R1 R 20
R3 = mnR10 (1 + ε3 ) A B
R4 R3
R4 = nR10 (1 + ε4 ) D
R 40 R 30
D
Transmittance:
ε1 − ε2 + ε3 − ε4 + ε1 ε3 − ε2 ε4 ε1 − ε2 ε1 − ε4 ε1 + ε3 ε1
UDC R1 R3 − R2 R4
r21 = = t0 t0 ε1 + mε2 t0 ε1 + nε4 t0 ε1 + mnε3 t0 ε1
J ε1 + mε2 + n(ε4 + mε3 ) 1+
Ri 1+ 1+ 1+ 1+
r21 (1 + n)(1 + m) (1 + n)(1 + m) (1 + n)(1 + m) (1 + n)(1 + m) (1 + n)(1 + m)
L(εi )
Elements: B – Signal Conditioning
≡ t0 f (εi ) ≡ t0
1 1 + M(εi )
Initial sensitivity of r21 mn
t0 t0 = R10
when : f (εi ) → 0 (1 + m)(1 + n)
Increment of function ε1 + mε2 + n(ε4 + mε3 ) ε1 + mε2 ε1 + nε4 ε1 + mnε3 ε1
M M = M12 = M14 = M13 = M1 =
f (εi ) denominator (1 + n)(1 + m) (1 + n)(1 + m) (1 + n)(1 + m) (1 + n)(1 + m) (1 + n)(1 + m)
Input resistance:
(R1 + R2 )(R4 + R3 ) R10 (1 + m + ε1 + mε2 )n(1 + m + ε4 + mε3 ) ε1 + mε2 ε1 ε4 ε1 mε3 ε1
r11 = 1+ 1+ 1+ 1+ 1+ 1+
∞ (1 + m)(1 + n) + ε1 + mε2 + n(ε4 + mε3 ) 1+m 1+m 1+m 1+m 1+m 1+m
Ri RAB RAB0 RAB0 RAB0 RAB0
(1 + ε12 )(1 + ε43 ) 1 + M12 1 + M14 1 + M13 1 + M1
∞ = RAB0 RL = ∞
≡ RAB0 1 + εAB 1 + M
n(1 + m)
2 In balance RAB0 RAB0 = R10
(1 + n)
1 nε12 + ε43 1+n 1+n
+ ε12 ε43 nε1 + ε4 + ε1 ε4 nε1 + mε3 + mε1 ε3
∞ 1 + M(εi ) 1+n n(ε1 + mε2 ) 1+m 1+m nε1
Its relative εAB
ε1 + mε2 ε4 + mε3 (1 + n)(1 + m) + ε1 + mε2 (1 + n)(1 + m) + ε1 + nε4 (1 + n)(1 + m) + ε1 + mnε3 (1 + n)(1 + m) + ε1
increment where ε12 ≡ ; ε43 ≡
1+m 1+m
Output resistance:
(R1 + R4 )(R2 + R3 ) R10 (1 + n + ε1 + nε4 )m(1 + n + ε2 + nε3 ) ε1 ε2 ε1 + nε4 ε1 nε3 ε1
r22 = 1+ 1+ 1+ 1+ 1+ 1+
∞ (1 + m)(1 + n) + ε1 + mε2 + n(ε4 + mε3 ) R 1+n 1+n 1+n 1+n 1+n 1+n
Ri RCD CD0 RCD0 RCD0 RCD0
(1 + ε14 )(1 + ε23 ) 1 + M12 1 + M14 1 + M13 1 + M1
∞ = RCD0 RG = ∞
≡ RCD0 (1 + εCD ) 1 + M
m(1 + n)
3 In balance RCD0 RCD0 = R10
(1 + m)
1 mε14 + ε23 1+m 1+m
+ ε14 ε23 mε1 + ε2 + ε1 ε2 mε1 + nε3 + nε1 ε3
∞ 1 + M(εi ) 1+m 1+n m(ε1 + nε4 ) 1+n mε1
Its relative εCD
ε1 + nε4 ε2 + nε3 (1 + n)(1 + m) + ε1 + mε2 (1 + n)(1 + m) + ε1 + nε4 (1 + n)(1 + m) + ε1 + mnε3 (1 + n)(1 + m) + ε1
increment whereε14 ≡ ; ε23 ≡
1+n 1+n
Table 1b. Matrix ZR elements of particular cases: if increments of two arm resistances R1 , R2 are opposite or small or if all arm initial resistances are equal.
Bridge of arbitrary resistance ratio m and n Initially double symmetric bridge M = N = 1
Variable: only two resistances: R1 , R2 Variable: all resistances: R1 , R2 , R3 , R4
(e) Large value opposite increments (f) Small increments1) (g) Large increments (h) Very small increments2) (i) |εi | = ε
Bridge open circuit
No terminal parameters
C
Symbol
(matrix ZR R1 = R10 (1 + ε); R1 = R10 (1 + ε1 ); R2 = R10 (1 + ε2 );
R2 1)
1 2 R1 |R1 + R2 | R10 Ri =
elements) A R2 = mR10 (1 − ε); R3 = R10 (1 + ε3 ); R4 = R10 (1 + ε4 )
U ′DC R10 (1 ± ε)
R3 = mnR10 ; equal to: ε1 = ε3 = ε
J
R4 R3 |ε1 + mε2 | (1 + m)(1 + n) 2) ε2 = ε4 = −ε
R4 = nR10 Ri 4Ri0 equal to: εi 4
D
B UDC = J r21
ε1 − ε2 2ε ε1 − ε2 + ε3 − ε4 + ε1 ε3 − ε2 ε4
Current to r21 t0 = t0 ≈ t0 (ε1 − ε2 ) t0 ε1 + ε2 + ε3 + ε4 ≈ t0 (ε1 − ε2 + ε3 − ε4 ) t0 4ε = εR10
1 + M12 ε(1 − m)
Voltage 1+ 1+
(1 + n)(1 + m) 4
transmittance
mn R10
1 Its initial t0 t0 = R10 t0 =
(1 + m)(1 + n) 4
sensitivity
ε(1 − m) ε1 + ε2 + ε3 + ε4
Its denominator M M12 = M12 ≈ 0 M = M ≈ 0 M = 0
(1 + n)(1 + m) 4
increment
Input resistance:
ε(1 − m) ε1 + ε2 ε3 + ε4
∞ 1+ 1+ 1+
RAB ≡ ∞ (1 + m) (1 + m)2 + ε(1 − m2 ) ε1 + mε2 2 2 ε1 + ε2 + ε3 + ε4
∞ RAB RAB0 =n R10 ≈ RAB0 1 + n R10 ≈ R10 1 + R10
RAB0 (1 + εAB ) 1 + M12 (1 + m)(1 + n) + ε(1 − m) (1 + n)(1 + m) 1 + 0, 25(ε1 + ε2 + ε3 + ε4 ) 4
2
n(1 + m)
In balance RAB0 RAB0 = R10 R10
(1 + n)
∞ nε(1 − m) n(ε1 + mε2 ) ε1 + ε2 + ε3 + ε4 + (ε1 + ε2 )(ε3 + ε4 ) ε1 + ε2 + ε3 + ε4
Its related εAB ≈ 0, 25 ≈ 0
(1 + n)(1 + m) + ε(1 − m) (1 + n)(1 + m) 1 + 0, 25(ε1 + ε2 + ε3 + ε4 ) 4
increment
Output resistance:
2
ε ε1 + ε4 ε2 + ε3
∞
1− 1+ 1+
RCD ≡ ∞ 1+n (1 + n)2 − ε2 ε1 + ε2 2 2 ε1 + ε2 + ε3 + ε4
∞ RCD RCD0 =m R10 ≈ RCD0 1 + R10 ≈ R10 1 + R10
RCD0 (1 + εCD ) 1 + M12 (1 + n)(1 + m) + ε(1 − m) 1+n 1 + 0, 25(ε1 + ε2 + ε3 + ε4 ) 4
3
m(1 + n)
In balance RCD0 RCD0 = R10 R10
1+m
1+m
−ε(1 − m) + ε2
∞ 1+n mε1 + ε2 ε1 + ε2 + ε3 + ε4 + (ε1 + ε4 )(ε2 + ε3 ) ε1 + ε2 + ε3 + ε4
Its related εCD ≈ 0, 25 ≈ 0
(1 + n)(1 + m) + ε(1 − m) (1 + n)(1 + m) 1 + 0, 25(ε1 + ε2 + ε3 + ε4 ) 4
increment
Unbalanced DC Bridges 883
884 Elements: B – Signal Conditioning
4 SIMPLIFIED BRIDGE EQUIVALENT A C
CIRCUITS R1 R2
+ −
In particular cases of the operation of the bridge as the UAB UDC
twoport type X, some of its equivalent circuits could − +
also be applied. Two of them are given in Figures 3(b) R4 R3
and 3(c). IDC
The circuit in Figure 3(b) is twoport of type T. It has only (a) B D
the same values of input–output terminal parameters as the
bridge (elements of its matrix ZR ) and could be used when A D
outside the bridge there is no other connection between IAB ∞ −r ∞
RAB 21 RCD − r21
circuits of both sides. In this twoport, two terminals, B
and C, are connected together but in the bridge they are UAB r21 UDC
not. The equivalent circuit type of twoport X also
exists and could be sometimes more proper, for example,
for describing the short circuit current of the conductance B C
(b)
bridge.
The other circuit, given in Figure 3(c), is the type 2T
A 0′ D
symmetric twoport. Two horizontally drawn resistances of
circuit in Figure 3(b) are divided here to two parts each. RA0′ RD0′
UAB UDC
They depend on bridge resistances, each one differently as
r21
follows ∞
RAB ∞
RCD
RB0″ RC0″
R4 (R1 + R2 )
RA0 = (11a)
Ri B 0″ C
(c)
R2 (R3 + R4 )
RB0 = (11b) Figure 3. Bridge 4R as twoport and its equivalent T and 2T
Ri twoport circuits: (a) bridge as differently drawn twoport X,
(b) twoport T circuit equivalent only for one sign of UDC and of
R4 (R2 + R3 ) input and of output parameters of the bridge (the same elements
RC0 = (11c) of matrix ZR ); (c) 2T symmetric twoport equivalent as circuit b)
Ri
and of all terminal voltages.
R2 (R1 + R4 )
RD0 = (11d)
Ri of T or of type structure. Reversing of output lines is
needed when output signal is crossing zero. Then the only
This circuit not only keeps the twoport matrix param- proper circuit for simulating in such a case is the twoport
eters equal but voltages of all terminals as well. Only X, that is, the four-arm bridge itself.
input resistances from the arms’ sides are here different
from ones of 4R bridge, for example, input resistance
from terminals AD. This equivalent circuit of the bridge 5 RELATIONS BETWEEN OPEN CIRCUIT
has not been found in literature yet but could be very
TERMINAL PARAMETERS OF THE 4R
useful in the analysis of the bridge and different cir-
cuits connected to it, including operational amplifiers, in BRIDGE
the case in which they are supplied from the common
source. From (6a) and (7a) of terminal resistances RAB0 , RCD0 , of
In both the equivalent circuits given above, the vertical the balanced 4R bridge, it resulted that their product was
resistance simulates the transmittance r21 . If r21 is negative, also equal to the product of opposite arms’ resistances as
this resistance should also be negative or wires of output in the balance condition (2a) or (5a), then
terminals should be reversed.
In conclusion, it is clear that in the general case it is RAB0 RCD0 = mn R10
2
= R10 R30 = R20 R40 (12)
impossible to describe the operation of any four-terminal
circuit, working as the twoport in full ranges of its internal The above expression could be called extended balance
resistance increment changes, only by three-terminal circuit condition of the four-arm bridge.
Unbalanced DC Bridges 885
Additionally, from (5b) results also the simple relation- 6 BRIDGE OPEN CIRCUIT TERMINAL
ship for initial parameters of the balanced 4R bridge PARAMETERS’ LINEARIZATION
RAB0 RCD0
= t0 (13) Bridge open circuit terminal parameters are related practi-
Ri0 cally linearly only for small values of independent resis-
tance increments εi . It is possible to obtain the linear
Then they are related to each other and an arbitrary value unbalance function f (εi ) for large values of εi too. In
cannot be assigned to all of them. Two above relationships this case, increments should simultaneously be at least in
could be applied to the syntheses of elements of the 4R two bridge arms and they are not independent. It happens
bridge equivalent to the unknown network’s inside struc- when increments of the sum of resistances of neighbor-
ture, based only on the results of measurements performed ing arms are equal to each other, that is, ε12 = ε34 , or
on its terminals. ε14 = ε23 . In these cases, from (6b) and (9) or from (7b)
∞
From the values of resistances RAB0 , RCD0 , ratios m and and (9) the outcomes respectively are εAB = ε12 = εR or
∞
n of balance state resistances of the four-arm bridge could εCD = ε14 = εR and according to the definition (8), gen-
be directly obtained: eralized bridge linearity conditions could be formulated as
ε1 + mε2 = ε4 + mε3 (15a)
rCD0 rAB0 − 1
m= (14a)
rCD0 − 1 or
ε1 + nε4 = ε2 + nε3 (15b)
and
With every one of them, formulas of bridge terminal
rAB0 rCD0 − 1
n= (14b) parameters become simpler. If condition (15a) is valid, then
rAB0 − 1
it is possible to find from (5) that
where: rAB0 ≡ RAB0 /R10 ; rCD0 ≡ RCD0 /R10 . r21 = R10
n
(ε − ε4 ) (16a)
When resistance R10 is given then single solutions 1+n 1
m > 0, n > 0 are obtain when rAB0 > 1 and rCD0 > 1 or and from (6a), that
rAB0 < 1 and rCD0 < 1. If rAB0 = 1 and rCD0 = 1 then there
ε1 + mε2
is infinite number of solutions n = m−1 . ∞
εAB = (16b)
These values of m and n and condition (13) should be 1+m
∞
taken also as some kind of criterion to choose the structure In this case, output resistance RCD depends nonlinearly
of the equivalent circuit of the two ports ‘black box’. Could on increments εi of arm resistances.
it be considered only as a four arms resistance bridge or If condition (15b) is satisfied then
if some additional resistances in series to bridge nodes m
are needed. r21 = R10 (ε − ε2 ) (17a)
∞
1+m 1
It follows from (6) and (7) that relative increments εAB
∞ ∞ ∞
and εCD of the bridge resistances RAB , RCD are related to and
∞ ε1 + nε4
each other through the common expression 1 + εR , as εDC = (17b)
1+n
follows
∞
and now open circuit input resistance RAB nonlinearly
∞ 1 + ε14 1 + ε23
1 + εCD depends on increments εi .
∞ =
1 + εAB 1 + ε12 1 + ε34 In the literature, only some particular cases of general lin-
earity conditions (15a) and (15b) can be found, for exam-
1 + n + ε1 + nε4 ple, for m = 1, n = 1. If each of them is to be separately
2
1+m × 1 + n + ε2 + nε3 satisfied, at most three increments can be independent, and
= (14c)
1+n 1 + m + ε1 + mε2 the fourth one results from others according to the above
× 1 + m + ε4 + mε3 formulas. It is difficult to implement such cases in practice.
Furthermore, in any pair of terminal parameter formu-
Actual arm resistances or their relative increments of the las (16a,b) or (17a,b) three increments εi are present and an
unbalanced bridge are not easy to reveal from measure- additional equation is needed to find all of them separately
ments of input–output terminal parameters. In the general from the terminal measurements. It is not so easy because
∞ ∞
case, the set of nonlinear equations should be solved. the third terminal parameter (RCD or RAB ) depends on εi
886 Elements: B – Signal Conditioning
nonlinearly. It is much easier to obtain two pairs of incre- parameters are different from the ones in matrix ZR of 4R
ments related to each other within each pair. It can be done bridge alone. Here they are called working terminal param-
by respectively forming the relations to common measured eters. Any four-terminal (4T) circuit operating as the X type
variables. For example, stationary resistance connected in twoport could be substituted by the equivalent 4R bridge of
series to the arm resistance (or parallel to arm conductance) arm resistances and its terminal parameters dependent now
proportionally decreases its increment. In force or pressure on diagonal resistances. Current to voltage transmittance
sensors, it is also possible to place two strain gauges in (or transfer resistance) r21 of unloaded bridge is replaced
points stressed differently but of known ratio. For condi- now by working transmittance T ≡ UDC /J and open cir-
∞ ∞
tion (15a), it could be cuit terminal resistances RAB , RCD – by working terminal
resistances RAB , RCD .
ε1 = mε3 , ε2 = mε4 (18a) From (6) of Article 126, Electrical Bridge Cir-
or cuits – Basic Information, Volume 2, if Zi = Ri , output
voltage UDC = RL IDC is
ε1 = −mε2 , ε4 = −mε3 (18b)
R1 R3 − R2 R4
It is obvious that the output signal is higher if increments UDC = J ≡ JT (20)
MUJ Ri , RG , RL
of neighboring arms are of the opposite signs, that is,
sign ε4 = −sign ε1 and sign ε2 = −sign ε1 . where
In (18a) conditions, positive values of increments εi are R1 R3 − R2 R4
limited only by the permissible dissipate power of arm T = (20a)
MUJ Ri , RG , RL
resistances or by maximum voltage of the current supply
∞ ∞ 1
source. Increment εAB of the input resistance RAB of the
MUJ = Ri + (R + R2 )(R3 + R4 )
unloaded bridge is now given by the formula RG 1
∞ ε1 + ε4 1 A
εAB = (19) + (R + R4 )(R2 + R3 ) +
1+m RL 1 RG RL
From formulas (16a), (18a), and (19) it is now possible A = Ri Rj Rk = R1 R2 (R3 + R4 )
to find increments of any arm.
+ R3 R4 (R1 + R2 ) = . . .
As an example of the (18b) case is the bridge built by
connecting two resistance potentiometers with output on Numerators of the sensitivity T in (20) and then the
their slides. Both ends of these resistances limit extreme balance condition are in all cases the same as before for the
increments and are as follows: |εi | ≤ 1 + m. In this bridge, bridge without diagonals. The denominator is the function
∞
the input resistance RAB of the open output is constant and of bridge resistances Ri and of diagonal resistances RG
the output resistance RCD nonlinearly depends on out-of- and RL . When RG → ∞, the second and the last addend
balance increments, but it should be measured instead. (component of the sum) in it become negligible. It is also
Linear relationships of the bridge transmittance r21 and the same with the third one and the last one of the addends
one of its terminal resistances could be used in one variable if RL → ∞. For J = constant, and if also together RG → ∞
measurement to correct the influence of another variable and RL → ∞ then output voltage UDC is the highest one
on the sensor set or to simultaneously obtain two variables ∞
and given in (5) as UCD .
measured indirectly on bridge terminals. Similarly it is possible to find equivalent working termi-
As an example of the first case, it is the use of the nal resistances RAB , RCD , of the bridge. After Karandeyev
strain gauge bridge input resistance changes to compensate (1966) they are
temperature influences on it.
RL (R1 + R2 )(R3 + R4 ) + A
RAB = (21)
RL Ri + (R1 + R4 )(R2 + R3 )
7 TERMINAL PARAMETERS OF THE DC
and
BRIDGE OF ANY SUPPLY SOURCE
RG (R1 + R4 )(R2 + R3 ) + A
AND LOAD RDC = (22)
RG Ri + (R1 + R2 )(R3 + R4 )
If in the circuit there are branches that can be considered Formulas (20) to (22) are quite complex. Then, in many
as internal or external diagonals connected to the 4R bridge cases it should be more useful in practice to apply equiv-
diagonals of finite equivalent resistances of values compara- alent terminal working parameters of the bridge only as
ble to terminal open circuit resistances, then bridge terminal functions of its matrix ZR elements.
Unbalanced DC Bridges 887
From (4) in Article 126, Electrical Bridge Cir- These resistances in balance state are equal to RAB0
cuits – Basic Information, Volume 2, formulas of the out- and to RCD0 . Out of the balance both of them decrease
put voltage of the bridge loaded by RL and supplied by the with the square of transmittance r21 , that is, indepen-
current or voltage source of equivalent internal resistance dently of its sign. Each of them also decreases on the
RG could be obtained. They are decreasing of the bridge opposite diagonal resistances RL
or RG .
RG RL r21
UDC = J ∞
∞
(23a) If RL = 0 or RG = 0, short circuit resistances obtained
RG + RAB RL + RCD − r21
2
0
are RAB 0
, RCD . After comparison of (25a) and (26a) for any
or after putting J = ERG−1 : value of r21 it is
R r 0 0 2
RCD RAB r21
UDC = E ∞
L 21 ∞ (23b) ∞ = ∞ = 1 − ∞ ∞ (27a)
RG + RAB RL + RCD − r21
2
RCD RAB RCD RAB
Formula (23a) simplified when RG → ∞ or/and RL → or
∞ and formula (23b) when RG = 0 or/and RL → ∞.
∞ ∞
The equivalent source resistance RG and load RL are gen-
0
RCD RAB = RAB
0
RCD (27b)
erally constant in measuring circuits. They could be variable
in equivalent circuits of four-terminal (4T) tested objects Then ratios of working open circuit and short circuit
and sensors, for example, Hall devices – see Chapters 6 terminal resistances of both ports are equal and decrease
and 7 by Warsza (1980) in the monograph by Kobus et al. 2
with r21 .
∞ ∞
After applying formulas of resistances RAB , RCD and When RL = 0, from (23a) or (24a), and (25a) two forms
putting RG ≡ gR10 ≡ kG RAB0 and RL ≡ lR10 ≡ kL RCD0 of current IDC0
of the bridge shortened output could
it is be found:
r21 ∞
UDC = J ∞ ∞
UDC RG r21
1 + εAB 1 + εCD r2 0
IDC = =J ∞
∞
1+ 1+ − 21 RCD RG + RAB RCD − r21
2
kG kL RG RL
(24a) J r21 1
= ∞
and RCD0 1 + εAB ∞
2
r21
1+ 1 + εCD − 2
E r21 kG kG mn R10
UDC = ∞ ∞
RG 1+ εAB 1 + εCD 2
r21 (28a,b)
1+ 1+ −
kG kL RG RL All the above formulas became simpler, in particular,
(24b) cases of ideal current or voltage supply sources and of the
Then the output voltage depends on resistances RG , RL , load resistance RL = 0 or RL = ∞. All such cases are given
and also on their ratios to terminal resistances of the in Table 2.
balanced bridge given by coefficients kG , kL . The output voltage of any case could be expressed
Working terminal resistances of the bridge within both ∞
similarly as UDC in formula (5), that is:
constant diagonal resistances are given by formulas:
∞
2
r21 UDC ≡ T0 fUJ (εi ) (29)
RAB = RAB − ∞
RCD + RL
where
2
∞ r12
= RAB0 1 + εAB − ∞
2
mn R10 1 + εCD + kL T0 ≡ J (R10 R30 )/MUJ0 = E/(RG )(R10 R30 )/MUJ0 – ini-
(25a,b) tial sensitivity of the working output voltage of the
and bridge supplied by source of internal resistance RG > 0
2
and loaded by any RL > 0,
∞ r21 fUJ (εi ) ≡ L(εi )/1 + εUJ (εi ) – unbalance function of
RCD = RCD − ∞
RAB + RG the working output voltage of the bridge.
2
r12 L(εi ) = ε1 + ε2 + ε3 − ε4 + ε1 ε3 − ε2 ε4 – numerator
∞
= RCD0 1 + εCD − ∞
of above function (as for 4R bridge alone),
2
mnR10 1 + εAB + kG
εUJ (εi ) ≡ MUJ (εi )/MUJ0 – relative increment of its
(26a,b) denominator,
888
Table 2. Bridge 4R output signals and terminal resistances of any mode of DC supply and load as functions of its ZR matrix elements.
General case Particular cases
∞
No Bridge output signals UAB = RAB IAB − r21 IDC (a) Current supply (b) Voltage supply (c) Source with resistance RG ,
∞
and terminal voltage UCD as output signal
Symbol
∞
resistances UDC = r21 IAB − RCD IDC RG → ∞, IAB = J RG → 0, UAB = E
RL → ∞, IDC = 0
Output voltage:
RG RL r21 RL r21 RL r21 RG r21
J ∞ ∞ 2 J E ∞ ∞ 2 J
∞ ∞
R1 R3 − R2 R4
(RG + RAB )(RL + RCD ) − r21 RL + RCD0 (1 + εCD ) RAB0 (1 + εAB )[RL + RCD0 (1 + εCD )] − r21 RG + RAB0 (1 + εAB )
UDC =J UDC J r21 r21 Er21 1 r21
MUJ = ∞ ∞
=J ∞ = ∞ 2
=J ∞
r2 1 + εCD RAB0 1 + εAB
Elements: B – Signal Conditioning
L(εi ) 1+ εAB 1+ εCD ∞ 1 + εCD r21
1+ 1+ − 21 1+ (1 + εAB ) 1+ − 1+
≡ T0 kG kL RG RL kL kL RAB0 RL kG
1 + ML (εi )
1 Initial sensitivity
R10 mn kG kL R10 mn kG
R10 R30 J R10 mn kL m kL T0 = J
T0 ≡ J (1 + m)(1 + n) (1 + kG )(1 + kL ) (1 + m)(1 + n) 1 + kG
MUJ0 T0 m 1 kL J E m 1
E R10 R30 =E (1 + m)(1 + n) 1 + kL (1 + m)2 (1 + kL ) ×E
= (1 + m)2 (1 + kG ) (1 + kL ) (1 + m)2 (1 + kG )
RG MUJ0
R10 [g(1 + n) + n(1 + m)]
n m m
n
Denominator of UDC MUJ0 R10 1+ n + (1+ m) 1+ m + (1+ n) R10 (1+ n) 1+ m + (1+ n) m R10 1 + n + (1 + m) (1 + m)
in balance g l l × 1 + m + (1 + n) g
l
Input resistance:
2 2
∞ r21 ∞ r21 ∞
2 In balance: RAB RAB 1− ∞ ∞ = RAB0 (1 + εAB )− ∞ RAB0 (1 + εAB )
n(1 + m) RAB (RCD + RL ) RCD0 (1 + εCD + kL )
RAB0 = R10
(1 + n)
2
Output resistance: ∞ r21 ∞
RCD 1− ∞ ∞
2 RCD0 (1 + εCD )
RCD (RAB + RG ) ∞ ∞ r2 ∞ r21 2
3 In balance: RCD 2 RCD0 (1 + εCD ) RCD − 21
∞ = RCD 1 − ∞ ∞
r21
∞ r21 RAB RCD RAB − ∞
m(1 + n) = RCD0 (1 + εCD )− ∞
RAB0 (1 + εAB + kG )
RCD0 = R10 RAB0 (1 + εAB + kG )
(1 + m)
0 J r21 1 r21 Er21 1 RG r21
Shorted output current: IDC ∞ 2
J ∞ 2
J ∞ ∞ 2
4 ∞ RCD0 1 + εAB r21 RCD0 (1 + εCD ) mnR10 r21 (RG + RAB )RCD − r21
0 U 1+ ∞
(1 + εCD )− (1 + ∞
εAB )(1 + ∞
εCD ) −
IDC = DC k R R 2
RCD G G CD0 mnR10
RG RL R10 mn ∞ ∞
Note: Some useful relations: E = J RG ; RG ≡ gR10 ; RL ≡ lR10 ; ≡ kG ; ≡ kL ; r12 = ; L(εi ) = ε1 − ε2 + ε3 − ε4 + ε1 ε3 − ε2 ε4 ; εAB , εCD – see Table 1a and b.
RAB0 RCD0 (1 + m)(1 + n)
Unbalanced DC Bridges 889
MUJ ≡ MUJ0 + MUI (εi ) – denominator of (20) ex- Sensors, Volume 2; Article 126, Electrical Bridge Cir-
pressed as a sum of its balance value and the cuits – Basic Information, Volume 2; Article 202, Cur-
resultant increment (sum of 14 components dependent rent Measurement, Volume 3; Article 204, Capacitance
differently on products of 6 resistances increments and Inductance Measurement, Volume 3.
of bridge with diagonals – 4T circuit) – see (8) and
(9) of Article 126, Electrical Bridge Circuits – Basic
Information, Volume 2.
REFERENCES
The working output voltage initial sensitivity T0 is Anderson, K.F. (1997) The New Current Loop, An Instrumen-
tation and Measurement Circuit Topology. IEEE Transac-
mn kG kL tions on Instrumentation and Measurement, October 1997,
T0 = J R10 (30a) http://www.vm-usa.com/links.html.
(1 + n)(1 + m) 1 + kG 1 + kL
Bentley, J.P. (1995) Principles of Measurement Systems, 3rd edn,
or Longman Scientific & Technical UK Ltd., Harlow, Essex, (pp.
178–185).
m kL
T0 = E (30b) Carter, G.W., Richardson, A. (1971) Technique of Circuit Anal-
(1 + m)2 (1 + kG )(1 + kL ) ysis, Cambridge University Press, London UK, (pp. 17–22,
248–281).
If supplied current J = constant, initial sensitivity T0 →
Dally, J.W. (ed.) (1993) Instrumentation for Engineering Mea-
T0 = J R10 (mn/(1 + m)(1 + n)) when: kG → ∞ and kL →
surements, 2nd edn, John Wiley & Sons, New York (pp.
∞. It is the highest one: T0 max = J R10 is when m → 170–176).
∞, n → ∞.
Karandeev, K.B. (1966) Bridge and Potentiometer Methods in
For E = constant T0 → T0 = E(m)/(1 + m)2 when Electrical Measurements, Energia, Moscow (1966) and Peace
kG → 0, kL → ∞ and is independent of n. Maximal Publisher MIR, Moscow (1967) (English translation from
sensitivity T0 max = 1/4E is now when m = 1. It is Russian edition 1960), Chapter 5.
independent of n. Particular cases are in Table 2. Kester W. (1999) Practical Design Techniques for Sensor Signal
After applying formula (5b) to (30a) is obtained Conditioning, Analog Devices Co., Norwood, MA.
Morris A.S. (1993) Principles of Measurement and Instrumenta-
kG kL tion, 2nd edn, Prentice Hall (pp. 106–118).
T0 = J t0 (31)
1 + kG 1 + kL Padmanabhan T.R. (2000) Industrial Instrumentation, Springer
Verlag, London (pp. 29–42).
where: t0 = (RAB0 RCD0 / Ri0 ) – initial sensitivity of
Warsza, Z.L. (1980) Chapters: 1, 6 i 7 in monography:
transmittance r21 . Kobus, A., Tuszynski J., Warsza Z.L., Technika Hallotronowa
This is the extension of formula (13) to the case of (Hall Devices Technology), WNT Warszawa, in Polish.
the resistance bridge with additional diagonal resistances. Warsza, Z.L. (2002) Bridges Supplied by Two Current Sources –
It relates the bridge initial working sensitivity T0 to the New Tool for Impedance Measurements and Signal Condi-
product of the current J , the initial open-circuit sensitivity tioning, in Proceedings of IMEKO-TC 7 Symposium, Cracow
t0 and coefficients of diagonal branches – on the right side. University of Technology, Cracow Poland, (pp. 231–236).
If changes of arm immitances are given as conductance Warsza Z.L. (2004) Resistance Bridges as Signal Condition-
increments, then it is possible to obtain linear relations ing Circuits of Few Variable Measurements, Prace Naukowe
of two bridge terminal parameters in dual circuit, that is, (Reasearch Works) ‘Elektryka’ 1(7), 2004, Pulaski Technical
University in Radom, Poland, pp. 78–157 (in Polish).
bridge supplied from the ideal voltage source E and the
output signal as the short circuited output current IDC0
. Zakrzewski, J. (2000) The Analogue Signal Processing in
Measuring Systems with a Single Passive Sensors, in Pro-
More information about unbalanced DC bridges can be
ceedings of XVI IMEKO World Congress, Vol. IX, Viena
found in the literature given below. Measurements of two to (pp. 185–190).
four variables provided by the resistance bridge, supplied
classically from one current source and unconventionally
by two of them connected in parallel to opposite bridge FURTHER READING
arms, are discussed in detail in Warsza (2002).
Boyes, W. (ed.) et al. (2003) Instrumentation Reference Book, 3rd
edn, Butterworth Heinemann, Elsevier Science, Boston, DC
RELATED ARTICLES bridge measurements, by M.L. Sanderson (pp. 463–468).
Dyer, S.A. (ed.) (2001) Survey of Instrumentation and Measure-
Article 56, Transfer Characteristics of Instrument ments, Wiley- Interscience, New York, Section Bridge Instru-
Stages, Volume 1; Article 112, Systematic Description of ments by Nicolas J. (pp. 309–317).
890 Elements: B – Signal Conditioning
Holman J.P. (1994) Experimental Methods for Engineers, 6th edn, Signal Conditioning and Data Acquisition Solutions Catalogue;
McGraw-Hill International Editions, New York, (pp. 141–144). Analog Devices Co., Norwood, MA, http://www.analog.
com/IOS.
McGhee, J., et al. (1998) Scientific Metrology, 2nd edn, Lodard
S.A., Lodz, (pp. 205–217, 289–297).
The Measurement and Automation Catalogue, Condensed Version
(2003) modules SCXI, SCC, National Instruments , Austin, TX
(pp. 22–24). [email protected].
128: Nature and Scope of AI Techniques
Ajith Abraham
Oklahoma State University, Stillwater, OK, USA
The Turing test, called the imitation game by Alan
1 Introduction to Computational Intelligence 893 Turing, measures the performance of a machine against that
2 Artificial Neural Networks 894 of a human being. The machine and a human (A) are placed
in two rooms. A third person, designated the interrogator,
3 Neural Network Architectures 895
is in a room apart from both the machine and the human.
4 Fuzzy Logic 895
The interrogator cannot see or speak directly to either A
5 If-then Rules and Fuzzy Inference Systems 896 or the machine, communicating with them solely through
6 Evolutionary Algorithms 897 some text messages or even a chat window. The task of the
7 Intelligent Paradigms 898 interrogator is to distinguish between the human and the
8 Hybrid Intelligent Systems 898 computer on the basis of questions he or she may put to
9 Models of Hybrid Systems 899 both of them over the terminals. If the interrogator cannot
10 Summary 900 distinguish the machine from the human, then, Turing
argues, the machine may be assumed to be intelligent. In
References 900
the 1960s, computers failed to pass the Turing test because
of the low processing speed of the computers.
The last few decades have seen a new era of arti-
1 INTRODUCTION TO COMPUTATIONAL ficial intelligence (AI) focusing on the principles, the-
INTELLIGENCE oretical aspects, and design methodology of algorithms
gleaned from nature. Examples are artificial neural net-
Machine intelligence dates back to 1936, when Alan Tur- works inspired by mammalian neural systems, evolutionary
ing proposed the idea of a universal mathematics machine, computation inspired by natural selection in biology, simu-
a theoretical concept in the mathematical theory of com- lated annealing inspired by thermodynamics principles, and
putability. Turing and Emil Post independently proved that swarm intelligence inspired by the collective behavior of
determining the decidability of mathematical propositions is insects or microorganisms, and so on, interacting locally
equivalent to asking what sort of sequences of a finite num- with their environment, therein causing coherent functional
ber of symbols can be recognized by an abstract machine global patterns to emerge. These techniques have found
with a finite set of instructions. their way into solving real-world problems in science, busi-
Such a mechanism is now known as a Turing ness, technology, commerce, and also to a great extent in
machine (Turing Machine, 2004). Turing’s research paper measuring systems.
addressed the question of machine intelligence, assessing Computational intelligence is a well-established para-
the arguments against the possibility of creating an digm, where new theories with a sound biological under-
intelligent computing machine and suggesting answers to standing have been evolving. The current experimental
those arguments; it proposed the Turing test as an empirical systems have many of the characteristics of biological com-
test of intelligence (Turing, 1950). puters (brains, in other words) and are beginning to be built
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
894 Elements: B – Signal Conditioning
to perform a variety of tasks that are difficult or impossible learning algorithm. Most applications of neural networks
to do with conventional computers. fall into the following categories:
To name a few, we have microwave ovens, washing
machines, and digital cameras that can figure out on their
own what settings to use to perform their tasks optimally; • Prediction: Use input values to predict some output
they have a reasoning capability, make intelligent decisions, • Classification: Use input values to determine the classi-
and learn from experience. fication
As usual, defining computational intelligence is not an • Data Association: Like classification, but it also recog-
easy task. In a nutshell, which becomes quite apparent nizes data that contains errors
in light of the current research pursuits, the area is • Data conceptualization: Analyze the inputs so that
heterogeneous with a combination of such technologies as grouping relationships can be inferred.
neural networks, fuzzy systems, evolutionary computation,
swarm intelligence, and probabilistic reasoning. A typical multilayered neural network and an artificial
The recent trend is to integrate different components to neuron are illustrated in Figure 1. Each neuron is char-
take advantage of complementary features and to develop a acterized by an activity level (representing the state of
synergistic system. Hybrid architectures like neuro-fuzzy polarization of a neuron), an output value (representing the
systems, evolutionary-fuzzy systems, evolutionary-neural firing rate of the neuron), a set of input connections (repre-
networks, evolutionary-neuro-fuzzy systems, and so on, senting synapses on the cell and its dendrite), a bias value
are widely applied for real-world problem solving. In the (representing an internal resting level of the neuron), and
following sections, the main functional components of
a set of output connections (representing a neuron’s axonal
computational intelligence are introduced along with their
projections). Each of these aspects of the unit is represented
key advantages and application domains.
mathematically by real numbers. Thus, each connection has
an associated weight (synaptic strength), which determines
the effect of the incoming input on the activation level of
2 ARTIFICIAL NEURAL NETWORKS the unit. The weights may be positive or negative. Refer-
ring to Figure 1, the signal flow from inputs x1 , . . . , xn is
Artificial neural networks (ANN) have been developed considered to be unidirectional, indicated by arrows, as is a
as generalizations of mathematical models of biological neuron’s output signal flow (O). The neuron output signal
nervous systems. O is given by the following relationship:
In a simplified mathematical model of the neuron, the
effects of the synapses are represented by connection
n
weights that modulate the effect of the associated input O = f (net) = f wj xj (1)
signals, and the nonlinear characteristic exhibited by neu- j =1
rons is represented by a transfer function, which is usually
the sigmoid, Gaussian, trigonometric function, and so on. where wj is the weight vector and the function f(net) i s
The neuron impulse is then computed as the weighted referred to as an activation (transfer) function. The variable
sum of the input signals, being transformed by the trans- net is defined as a scalar product of the weight and input
fer function. vectors
The learning capability of an artificial neuron is achieved
by adjusting the weights in accordance to the chosen net = wT x = w1 x1 + · · · · +wn xn (2)
Hidden layer
x1 Input layer
Output layer
x2 w1
w2 q f
x3 output (o )
w3
x4 w4
(a) Artificial neuron (b) Multilayered artificial neural network
Figure 1. Architecture of an artificial neuron and a multilayered neural network.
Nature and Scope of AI Techniques 895
where T is the transpose of a matrix and in the simplest In supervised learning, an input vector is presented at
case the output value O is computed as the inputs together with a set of desired responses, one
for each node, at the output layer. A forward pass is done
1 if wT x ≥ θ and the errors or discrepancies, between the desired and
O = f (net) = (3)
0 otherwise actual response for each node in the output layer, are found.
These are then used to determine weight changes in the net
where θ is called the threshold level , and this type of node according to the prevailing learning rule.
is called a linear threshold unit. The term ‘supervised’ originates from the fact that the
desired signals on individual output nodes are provided
3 NEURAL NETWORK ARCHITECTURES by an external teacher. The best-known examples of this
technique occur in the backpropagation algorithm, the delta
The behavior of the neural network depends largely on rule, and perceptron rule.
the interaction between the different neurons. The basic In unsupervised learning (or self-organization), a (out-
architecture consists of three types of neuron layers: put) unit is trained to respond to clusters of pattern within
the input. In this paradigm, the system is supposed to
1. Input discover statistically salient features of the input popu-
2. Hidden lation (Kohonen, 1988). Unlike the supervised learning
3. Output. paradigm, there is no a priori set of categories into which
In feed-forward networks, the signal flow is from input the patterns are to be classified; rather, the system must
to output units, strictly in a feed-forward direction. The data develop its own representation of the input stimuli.
processing can extend over multiple (layers of) units, but Reinforcement learning is learning what to do – how to
no feedback connections are present, that is, connections map situations to actions – so as to maximize a numerical
extending from outputs of units to inputs of units in the reward signal. The learner is not told which actions to take,
same layer or previous layers. as in most forms of machine learning, but instead must
Recurrent networks contain feedback connections. Con- discover which actions yield the most reward by trying
trary to feed-forward networks, the dynamical properties of them. In the most interesting and challenging cases, actions
the network are important. In some cases, the activation may affect not only the immediate reward but also the
values of the units undergo a relaxation process such that next situation and, through that, all subsequent rewards.
the network will evolve into a stable state in which these These two characteristics, trial-and-error search and delayed
activations do not change anymore. In other applications, reward, are the two most important distinguishing features
the changes of the activation values of the output neurons of reinforcement learning.
are significant, such that the dynamical behavior constitutes
the output of the network.
There are several other neural network architectures 4 FUZZY LOGIC
(Elman network, adaptive resonance theory maps, competi-
Zadeh (1965) introduced the concept of fuzzy logic to
tive networks, etc.) depending on the properties and require-
represent vagueness in linguistics and to further implement
ment of the application. The reader may refer to Bishop
and express human knowledge and inference capability in
(1995) for an extensive overview of the different neural
a natural way. Fuzzy logic starts with the concept of a
network architectures and learning algorithms.
fuzzy set.
A neural network has to be configured such that the
A fuzzy set is a set without a crisp, clearly defined
application of a set of inputs produces the desired set
boundary. It can contain elements with only a partial degree
of outputs. Various methods to set the strengths of the
of membership.
connections exist. One way is to set the weights explicitly,
A Membership Function (MF) is a curve that defines how
using a priori knowledge.
each point in the input space is mapped to a membership
Another way is to train the neural network by feeding it,
value (or degree of membership) between 0 and 1. The input
teaching patterns and letting it change its weights according
space is sometimes referred to as the universe of discourse.
to some learning rule.
Let X be the universe of discourse and x be a generic
The learning situations in neural networks may be
element of X. A classical set A is defined as a collection
classified into three distinct sorts of learning:
of elements or objects x ∈ X, such that each x can either
1. Supervised belong to or not belong to the set A.
2. Unsupervised By defining a characteristic function (or membership
3. Reinforcement. function) on each element x in X, a classical set A can
896 Elements: B – Signal Conditioning
be represented by a set of ordered pairs (x, 0) or (x, 1), It is possible to resolve the statement A AND B, where
where 1 indicates membership and 0, nonmembership. A and B are limited to the range (0, 1) by using the operator
Unlike the conventional set mentioned above, the fuzzy minimum (A, B). Using the same reasoning, we can replace
set expresses the degree to which an element belongs to the OR operation with the maximum operator, so that A
a set. Hence, the characteristic function of a fuzzy set is OR B becomes equivalent to maximum (A, B). Finally,
allowed to have a value between 0 and 1, denoting the the operation NOT A becomes equivalent to the operation
degree of membership of an element in a given set. If X 1 – A.
is a collection of objects denoted generically by x, then a In fuzzy logic terms, these are popularly known as
fuzzy set A in X is defined as a set of ordered pairs: fuzzy intersection or conjunction (AND), fuzzy union
or disjunction (OR), and fuzzy complement (NOT). The
A = {(x, µA (x))|x ∈ X} (4) intersection of two fuzzy sets A and B is specified, in
general, by a binary mapping T , which aggregates two
µA (x) is called the membership function of linguistic membership functions as follows.
variable x in A, which maps X to the membership space
M, M = [0,1], where M contains only two points 0 and 1, µA∩B (x) = T (µA (x), µB (x)) (5)
A is crisp and µA is identical to the characteristic function
of a crisp set. The fuzzy intersection operator is usually referred to as T-
Triangular and trapezoidal membership functions are norm (Triangular norm) operator. The fuzzy union operator
the simplest membership functions, formed using straight is specified in general by a binary mapping S.
lines. Some of the other shapes are Gaussian, generalized
bell, sigmoidal, and polynomial-based curves. Figure 2 µA∪B (x) = S(µA (x), µB (x)) (6)
illustrates the shapes of two commonly used MFs. The most
important thing to realize about fuzzy logical reasoning is This class of fuzzy union operators are often referred to as
the fact that it is a superset of standard Boolean logic. T-conorm (S-norm) operators.
4.1 Fuzzy logic operators 5 IF-THEN RULES AND FUZZY
It is interesting to note the correspondence between two- INFERENCE SYSTEMS
valued and multivalued logic operations for the AND, OR,
and NOT logical operators. The fuzzy rule base is characterized in the form of if-
then rules in which preconditions and consequents involve
linguistic variables. The collection of these fuzzy rules
1
forms the rule base for the fuzzy logic system. Owing to
0.75 their concise form, fuzzy if-then rules are often employed
to capture the imprecise modes of reasoning that play an
0.5 essential role in the human ability to make decisions in an
environment of uncertainty and imprecision. A single fuzzy
0.25 if-then rule assumes the form
0
if x is A then y is B
(a) 0 2 4 6 8 10
where A and B are linguistic values defined by fuzzy sets in
1
the ranges (universes of discourse) X and Y, respectively.
0.75 The if part of the rule ‘x is A’ is called the antecedent
(precondition) or premise, while the then part of the rule ‘y
0.5 is B’ is called the consequent or conclusion. Interpreting an
if-then rule involves evaluating the antecedent (fuzzification
0.25
of the input and applying any necessary fuzzy operators)
0
and then applying that result to the consequent (known
(b) 0
as implication). For rules with multiple antecedents, all
2 4 6 8 10
parts of the antecedent are calculated simultaneously and
Figure 2. Membership functions; (a) Gaussian and (b) trape- resolved to a single value using the logical operators. Simi-
zoidal. larly, all the consequents (rules with multiple consequents)
Nature and Scope of AI Techniques 897
Initialization Valuation Solution Yes
Start Stop
of population (fitness value) found?
No
Reproduction
Figure 3. Flowchart of genetic algorithm iteration.
are affected equally by the result of the antecedent. The 1977); evolutionary programming (Fogel, Owens and
consequent specifies a fuzzy set be assigned to the output. Walsh, 1967); genetic programming (Koza, 1992); and
The implication function then modifies that fuzzy set to the learning classifier systems. They all share a common
degree specified by the antecedent. For multiple rules, the conceptual base of simulating the evolution of individual
output of each rule is a fuzzy set. The output fuzzy sets structures via processes of selection, mutation, and
for each rule are then aggregated into a single output fuzzy reproduction. The processes depend on the perceived
set. Finally, the resulting set is defuzzified, or resolved, to performance of the individual structures as defined by the
a single number. environment (problem).
The defuzzification interface is a mapping from a space EA’s deal with parameters of finite length, which are
of fuzzy actions defined over an output universe of dis- coded using a finite alphabet, rather than directly manipu-
course into a space of nonfuzzy actions, because the output lating the parameters themselves. This means that the search
from the inference engine is usually a fuzzy set, while for is unconstrained neither by the continuity of the function
most practical applications, crisp values are often required. under investigation nor by the existence of a derivative
The three commonly applied defuzzification techniques function. Figure 3 depicts the functional block diagram of
are max-criterion, center-of-gravity, and the mean-of- a genetic algorithm, and the various aspects are discussed
maxima. The max-criterion is the simplest of these three below. It is assumed that a potential solution to a problem
to implement. It produces the point at which the possibility may be represented as a set of parameters. These parame-
distribution of the action reaches a maximum value. ters (known as genes) are joined together to form a string
The reader can refer to Nguyen and Walker (1999) for of values (known as a chromosome). A gene (also referred
more information related to fuzzy systems. It is typically to as a feature, character, or detector) refers to a specific
advantageous if the fuzzy rule base is adaptive to a certain attribute that is encoded in the chromosome. The particular
application. The fuzzy rule base is usually constructed values that the genes can take are called its alleles.
manually or by automatic adaptation by some learning Encoding issues deal with representing a solution in a
techniques using evolutionary algorithms and/or neural chromosome and, unfortunately, no one technique works
network learning methods (Abraham, 2001). best for all problems. A fitness function must be devised
for each problem to be solved.
Given a particular chromosome, the fitness function
6 EVOLUTIONARY ALGORITHMS returns a single numerical fitness or figure of merit,
which will determine the ability of the individual that the
Evolutionary algorithms (EA) are adaptive methods, which chromosome represents.
may be used to solve search and optimization problems, Reproduction is the second critical attribute of GAs,
based on the genetic processes of biological organisms. where two individuals selected from the population are
Over many generations, natural populations evolve accord- allowed to mate to produce an offspring, which will com-
ing to the principles of natural selection and ‘survival of prise the next generation. Having selected the parents, the
the fittest’, first clearly stated by Charles Darwin in his offsprings are generated, typically using the mechanisms of
work ‘The Origin of Species’. By mimicking this pro- crossover and mutation.
cess, evolutionary algorithms are able to ‘evolve’ solutions Selection is the survival of the fittest within GAs. It
to real-world problems, provided they have been suitably determines which individuals are to survive to the next
encoded (Fogel, 1999). generation. The selection phase consists of three parts.
Usually grouped under the term evolutionary algorithms The first part involves determination of the individual’s
or evolutionary computation, we find the domains fitness by the fitness function. A fitness function must be
of genetic algorithms (GA) (Holland, 1975), (Goldberg, devised for each problem; given a particular chromosome,
1989); evolution strategies (Rechenberg, 1973), (Schwefel, the fitness function returns a single numerical fitness value,
898 Elements: B – Signal Conditioning
which is proportional to the ability, or utility, of the 7.2 Swarm intelligence
individual represented by that chromosome.
The second part involves converting the fitness function Swarm intelligence is aimed at collective behavior of
into an expected value, followed by the last part where intelligent agents in decentralized systems. Although there
the expected value is then converted to a discrete number is typically no centralized control dictating the behav-
of offspring. ior of the agents, local interactions among the agents
Some of the commonly used selection techniques are often cause a global pattern to emerge (Eberhart, Shi and
roulette wheel and stochastic universal sampling. If the Kennedy, 2001).
GA has been correctly implemented, the population will Most of the basic ideas are derived from the real swarms
evolve over successive generations so that the fitness of the in nature, which includes ant colonies, bird flocking, honey-
best and the average individual in each generation increases bees, bacteria and microorganisms, and so on. Ant Colony
toward the global optimum. Optimization (ACO) has already been applied successfully
to solve several engineering optimization problems.
Swarm models are population based and the population
is initialized with a population of potential solutions. These
7 INTELLIGENT PARADIGMS individuals are then manipulated (optimized) over several
iterations using several heuristics inspired from the social
behavior of insects in an effort to find the optimal solution.
7.1 Probabilistic computing
Ant colony algorithms are inspired by the behavior
of natural ant colonies, in the sense that they solve
Probabilistic models are viewed as being similar to that of a their problems by multiagent cooperation using indirect
game; actions are based on expected outcomes. The center communication through modifications in the environment.
of interest moves from the deterministic to probabilistic Ants release a certain amount of pheromone (hormone)
models using statistical estimations and predictions. while walking, and each ant prefers (probabilistically) to
In the probabilistic modeling process, risk means uncer- follow a direction that is rich in pheromone. This simple
tainty for which the probability distribution is known. behavior explains why ants are able to adjust to changes
Therefore, risk assessment means a study to determine in the environment, such as optimizing the shortest path
the outcomes of decisions along with their probabilities. to a food source or a nest. In ACO, ants use information
Decision makers often face a severe lack of definite infor- collected during past simulations to direct their search
mation. Probability assessment quantifies the information and this information is available and modified through the
gap between what is known and what needs to be known environment. Recently, ACO algorithms have also been
for an optimal decision. Probabilistic models are used for used for clustering data sets.
protection against adverse uncertainty and exploitation of
propitious uncertainty (Pearl, 1997).
A good example is the probabilistic neural network 8 HYBRID INTELLIGENT SYSTEMS
(Bayesian learning) in which probability is used to represent
uncertainty about the relationship being learned. Before Several adaptive hybrid intelligent systems have, in recent
we have seen any data, our prior opinions about what years, been developed for model expertise, image, and
the true relationship might be can be expressed in a video segmentation techniques, process control, mechatron-
probability distribution over the network weights that define ics, robotics, complicated automation tasks and so on.
this relationship. Many of these approaches use the combination of dif-
After we look at the data, our revised opinions are ferent knowledge representation schemes, decision-making
captured by a posterior distribution over network weights. models, and learning strategies to solve a computational
Network weights that seemed plausible before, but which task. This integration aims at overcoming limitations of
do not match the data very well, will now be seen as individual techniques through hybridization or fusion of
being much less likely, while the probability for values of various techniques.
the weights that do fit the data well will have increased. These ideas have led to the emergence of several differ-
Typically, the purpose of training is to make predictions ent kinds of intelligent system architectures. Most of the
for future cases in which only the inputs to the network current Hybrid Intelligent Systems (HIS) consist of three
are known. The result of conventional network training essential paradigms: artificial neural networks, fuzzy infer-
is a single set of weights that can be used to make such ence systems, and global optimization algorithms (exam-
predictions. ple, evolutionary algorithms). Nevertheless, HIS is an
Nature and Scope of AI Techniques 899
Table 1. Hybrid intelligent system basic ingredients. 9 MODELS OF HYBRID SYSTEMS
Methodology Advantage
The various HIS architectures could be broadly classified
Artificial neural networks Adaptation, learning,
into four different categories based on the systems overall
and approximation architecture:
Fuzzy logic Approximate reasoning
Global optimization Derivative free
1. Stand alone
algorithms optimization 2. Transformational
3. Hierarchical hybrid
4. Integrated hybrid.
open, instead of conservative, concept, that is, it is evolv-
ing those relevant techniques together with the important
advances in other new computing methods. Table 1 lists 9.1 Stand-alone architecture
the three principal ingredients together with their advan-
tages (Abraham, 2002). Stand-alone models of HIS applications consist of inde-
pendent software components, which do not interact in any
Experience has shown that it is crucial for the design of
way. Developing stand-alone systems can have several pur-
HIS to primarily focus on the integration and interaction of
poses. First, they provide direct means of comparing the
different techniques rather than merge different methods to
problem-solving capabilities of different techniques with
create ever-new techniques. Techniques already well under-
reference to a certain application. Running different tech-
stood should be applied to solve specific domain problems
niques in a parallel environment permits a loose approx-
within the system. Their weakness must be addressed by
imation of integration. Stand-alone models are often used
combining them with complementary methods.
to develop a quick initial prototype while a more time-
Neural networks offer a highly structured architecture
consuming application is developed. Some of the benefits
with learning and generalization capabilities. The general-
are simplicity and ease of development using commercially
ization ability for new inputs is then based on the inherent
available software packages.
algebraic structure of the neural network. However, it is
very hard to incorporate human a priori knowledge into
a neural network. This is mainly due to the fact that the 9.2 Transformational hybrid architecture
connectionist paradigm gains most of its strength from a
distributed knowledge representation. In a transformational hybrid model, the system begins as
In contrast, fuzzy inference systems exhibit complemen- one type and ends up as the other. Determining which
tary characteristics, offering a very powerful framework technique is used for development and which is used
for approximate reasoning as it attempts to model the for delivery is based on the desirable features that the
human reasoning process at a cognitive level. Fuzzy sys- technique offers. Expert systems and neural networks have
tems acquire knowledge from domain experts and this is proven to be useful transformational models. Variously,
encoded within the algorithm in terms of the set of if-then either the expert system is incapable of adequately solving
rules. Fuzzy systems employ this rule-based approach and the problem, or it requires the speed, adaptability, and
interpolative reasoning to respond to new inputs. The incor- robustness of neural network. Knowledge from the expert
poration and interpretation of knowledge is straightforward, system is used to set the initial conditions and training
whereas learning and adaptation constitute major problems. set for neural network. Transformational hybrid models are
Global optimization is the task of finding the absolutely often quick to develop and ultimately require maintenance
best set of parameters to optimize an objective function. on only one system. Most of the developed models are just
In general, it may be possible to have solutions that are application oriented.
locally optimal but not globally optimal. Evolutionary com-
puting (EC) works by simulating evolution on a computer.
Such techniques could be easily used to optimize neural 9.3 Hierarchical hybrid architectures
networks, fuzzy inference systems, and other problems.
Owing to the complementary features and strengths The architecture is built in a hierarchical fashion, associ-
of different systems, the trend in the design of hybrid ating a different functionality with each layer. The overall
system is to merge different techniques into a more functioning of the model will depend on the correct func-
powerful integrated system to overcome their individ- tioning of all the layers. A possible error in one of the layers
ual weaknesses. will directly affect the desired output.
900 Elements: B – Signal Conditioning
9.4 Integrated hybrid architectures in Computer Science, Vol. 2084, (eds J. Mira and A. Prieto),
Springer Verlag, Germany, (pp. 269–276).
These models include systems, which combine different Abraham, A. (2002) Intelligent Systems: Architectures and Per-
techniques into one single computational model. They share spectives, Recent Advances in Intelligent Paradigms and Appli-
data structures and knowledge representations. Another cations, in Studies in Fuzziness and Soft Computing, Chapter 1,
(eds A. Abraham, L. Jain and J. Kacprzyk), Springer Verlag
approach is to put the various techniques on a side-by-side Germany, (pp. 1–35).
basis and focus on their interaction in the problem-solving
Bishop, C.M. (1995) Neural Networks for Pattern Recognition,
task. This method might allow integrating alternative tech- Oxford University Press, UK.
niques and exploiting their mutuality. The benefits of fused
Eberhart, R., Shi, Y. and Kennedy, J. (2001) Swarm Intelligence,
architecture include robustness, improved performance, and Morgan Kaufmann, San Francisco, CA.
increased problem-solving capabilities. Finally, fully inte-
Fogel, D.B. (1999) Evolutionary Computation: Toward a New
grated models can provide a full range of capabilities such Philosophy of Machine Intelligence, 2nd edn, IEEE Press,
as adaptation, generalization, noise tolerance, and justi- Piscataway, NJ.
fication. Fused systems have limitations caused by the Fogel, L.J., Owens, A.J. and Walsh, M.J. (1967) Artificial Intelli-
increased complexity of the intermodule interactions, and gence Through Simulated Evolution, John Wiley & Sons, New
specifying, designing, and building fully integrated models York.
is complex. Goldberg, D.E. (1989) Genetic Algorithms in Search, Optimiza-
tion, and Machine Learning, Addison-Wesley Publishing Cor-
poration, Inc, Reading, MA.
10 SUMMARY Holland, J. (1975) Adaptation in Natural and Artificial Systems,
University of Michigan Press, Ann Harbor, MI.
Artificial intelligence is the study of intelligent behavior.
Kohonen, T. (1988) Self-organization and Associative Memory,
Its ultimate goal is a theory of intelligence that accounts Springer-Verlag, New York.
for the behavior of naturally occurring intelligent entities,
Koza, J.R. (1992) Genetic Programming, MIT Press, Cambridge,
and this guides the creation of artificial entities capable MA.
of intelligent behavior. The stagnation of artificial intelli-
Nguyen, H.T. and Walker, E.A. (1999) A First Course in Fuzzy
gence during the 1970s and 1980s does not have much Logic, CRC Press, USA.
bearing on the likelihood of artificial intelligence to succeed
Pearl, J. (1997) Probabilistic Reasoning in Intelligent Systems:
in the future, since we know that the cause responsible for Networks of Plausible Inference, Morgan Kaufmann Publishers,
stagnation (mainly due to insufficient hardware resources) San Francisco, CA.
is no longer present. More detail of the various methods Rechenberg, I. (1973) Evolutionsstrategie: Optimierung technis-
introduced here is found in Article 129, Artificial Neu- cher Systeme nach Prinzipien der biologischen Evolution,
ral Networks, Volume 2; Article 130, Rule-based Expert Fromman-Holzboog, Stuttgart.
Systems, Volume 2; and Article 131, Evolutionary Com- Schwefel, H.P. (1977) Numerische Optimierung von Com-
putation, Volume 2. putermodellen Mittels der Evolutionsstrategie, Birkhaeuser,
Basel.
Turing, A.M. (1950) Computing Machinery and Intelligence
REFERENCES http://abelard.org/turpap/turpap.htm.
Turing Machine. (2004) http://www.turing.org.uk/turing/.
Abraham, A. (2001) in Neuro-Fuzzy Systems: State-of-the-
art Modeling Techniques, Connectionist Models of Neurons, Zadeh, L.A. (1965) Fuzzy Sets. Journal of Information and
Learning Processes, and Artificial Intelligence, Lecture Notes Control, 8, 338–353.
129: Artificial Neural Networks
Ajith Abraham
Oklahoma State University, Stillwater, OK, USA
human brain is a collection of more than 10 billion inter-
1 Introduction to Artificial Neural Networks 901 connected neurons. Each neuron is a cell (Figure 1) that
uses biochemical reactions to receive, process, and transmit
2 Neural Network Architectures 902
information.
3 Neural Network Learning 903 Treelike networks of nerve fibers called dendrites are
4 Backpropagation Learning 903 connected to the cell body or soma, where the cell nucleus is
5 Training and Testing Neural Networks 904 located. Extending from the cell body is a single long fiber
6 Higher Order Learning Algorithms 905 called the axon, which eventually branches into strands
7 Designing Artificial Neural Networks 905 and substrands, and are connected to other neurons through
8 Self-organizing Feature Map and Radial synaptic terminals or synapses.
Basis Function Network 906 The transmission of signals from one neuron to another
9 Recurrent Neural Networks and Adaptive at synapses is a complex chemical process in which specific
Resonance Theory 907 transmitter substances are released from the sending end of
10 Summary 908 the junction. The effect is to raise or lower the electrical
potential inside the body of the receiving cell. If the
References 908
potential reaches a threshold, a pulse is sent down the axon
and the cell is ‘fired’.
Artificial neural networks (ANN) have been developed
as generalizations of mathematical models of biological
1 INTRODUCTION TO ARTIFICIAL nervous systems. A first wave of interest in neural networks
NEURAL NETWORKS (also known as connectionist models or parallel distributed
processing) emerged after the introduction of simplified
A general introduction to artificial intelligence methods neurons by McCulloch and Pitts (1943).
of measuring signal processing is given in Article 128, The basic processing elements of neural networks are
Nature and Scope of AI Techniques, Volume 2. called artificial neurons, or simply neurons or nodes. In a
The human brain provides proof of the existence of mas- simplified mathematical model of the neuron, the effects
sive neural networks that can succeed at those cognitive, of the synapses are represented by connection weights that
perceptual, and control tasks in which humans are suc- modulate the effect of the associated input signals, and the
cessful. The brain is capable of computationally demanding nonlinear characteristic exhibited by neurons is represented
perceptual acts (e.g. recognition of faces, speech) and con- by a transfer function. The neuron impulse is then computed
trol activities (e.g. body movements and body functions). as the weighted sum of the input signals, transformed by
The advantage of the brain is its effective use of mas- the transfer function. The learning capability of an artificial
sive parallelism, the highly parallel computing structure, neuron is achieved by adjusting the weights in accordance
and the imprecise information-processing capability. The to the chosen learning algorithm.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
902 Elements: B – Signal Conditioning
2 NEURAL NETWORK ARCHITECTURES
Axon
The basic architecture consists of three types of neuron
Soma
Dendrites layers: input, hidden, and output layers. In feed-forward
networks, the signal flow is from input to output units,
strictly in a feed-forward direction. The data processing
Nucleus can extend over multiple (layers of) units, but no feed-
back connections are present. Recurrent networks contain
feedback connections. Contrary to feed-forward networks,
the dynamical properties of the network are important. In
Synaptic terminals
some cases, the activation values of the units undergo a
Figure 1. Mammalian neuron. relaxation process such that the network will evolve to a
stable state in which these activations do not change any-
more. In other applications, the changes of the activation
A typical artificial neuron and the modeling of a multi- values of the output neurons are significant, such that the
layered neural network are illustrated in Figure 2. Referring dynamical behavior constitutes the output of the network.
to Figure 2, the signal flow from inputs x1 , . . . , xn is con- There are several other neural network architectures (Elman
sidered to be unidirectional, which are indicated by arrows, network, adaptive resonance theory maps, competitive net-
as is a neuron’s output signal flow (O). The neuron output works, etc.), depending on the properties and requirement
signal O is given by the following relationship: of the application. The reader can refer to Bishop (1995)
for an extensive overview of the different neural network
n architectures and learning algorithms.
O = f (net) = f wj xj (1) A neural network has to be configured such that the
j =1 application of a set of inputs produces the desired set of
outputs. Various methods to set the strengths of the connec-
where wj is the weight vector, and the function f(net) is tions exist. One way is to set the weights explicitly, using
referred to as an activation (transfer) function. The variable a priori knowledge. Another way is to train the neural net-
net is defined as a scalar product of the weight and input work by feeding it teaching patterns and letting it change
vectors, its weights according to some learning rule. The learning
situations in neural networks may be classified into three
distinct sorts. These are supervised learning, unsupervised
net = wT x = w1 x1 + · · · · +wn xn (2)
learning, and reinforcement learning. In supervised learn-
ing, an input vector is presented at the inputs together with
where T is the transpose of a matrix, and, in the simplest a set of desired responses, one for each node, at the output
case, the output value O is computed as layer. A forward pass is done, and the errors or discrep-
ancies between the desired and actual response for each
1 if wT x θ node in the output layer are found. These are then used to
O = f (net) = (3)
0 otherwise determine weight changes in the net according to the pre-
vailing learning rule. The term supervised originates from
where θ is called the threshold level; and this type of node the fact that the desired signals on individual output nodes
is called a linear threshold unit. are provided by an external teacher.
Hidden layer
x1 Input layer
Output layer
x2 w1
w2 q f
x3 output (o )
w3
x4 w4
(a) Artificial neuron (b) Multilayered artificial neural network
Figure 2. Architecture of an artificial neuron and a multilayered neural network.
Artificial Neural Networks 903
The best-known examples of this technique occur in the where o is the desired output for
backpropagation algorithm, the delta rule, and the percep-
i = 1 to n(inputs).
tron rule. In unsupervised learning (or self-organization),
a (output) unit is trained to respond to clusters of pattern
Unfortunately, plain Hebbian learning continually streng-
within the input. In this paradigm, the system is supposed
thens its weights without bound (unless the input data is
to discover statistically salient features of the input pop-
properly normalized).
ulation. Unlike the supervised learning paradigm, there is
no a priori set of categories into which the patterns are to
be classified; rather, the system must develop its own rep- 3.2 Perceptron learning rule
resentation of the input stimuli. Reinforcement learning is
learning what to do – how to map situations to actions – so The perceptron is a single layer neural network whose
as to maximize a numerical reward signal. The learner is weights and biases could be trained to produce a correct
not told which actions to take, as in most forms of machine target vector when presented with the corresponding input
learning, but instead must discover which actions yield the vector. The training technique used is called the perceptron-
most reward by trying them. In the most interesting and learning rule. Perceptrons are especially suited for simple
challenging cases, actions may affect not only the imme- problems in pattern classification.
diate reward, but also the next situation and, through that, Suppose we have a set of learning samples consisting
all subsequent rewards. These two characteristics, trial-and- of an input vector x and a desired output d(k). For a
error search and delayed reward are the two most important classification task, the d(k) is usually +1 or −1. The
distinguishing features of reinforcement learning. perceptron-learning rule is very simple and can be stated
as follows:
3 NEURAL NETWORK LEARNING 1. Start with random weights for the connections.
2. Select an input vector x from the set of training
3.1 Hebbian learning samples.
3. If output yk = d(k) (the perceptron gives an incorrect
The learning paradigms discussed above result in an adjust- response), modify all connections wi according to:
ment of the weights of the connections between units, δwi = η(dk − yk )xi ; (η = learning rate).
according to some modification rule. Perhaps the most influ- 4. Go back to step 2.
ential work in connectionism’s history is the contribution Note that the procedure is very similar to the Hebb
of Hebb (1949), where he presented a theory of behav- rule; the only difference is that when the network responds
ior based, as much as possible, on the physiology of the correctly, no connection weights are modified.
nervous system.
The most important concept to emerge from Hebb’s
work was his formal statement (known as Hebb’s postu- 4 BACKPROPAGATION LEARNING
late) of how learning could occur. Learning was based on
the modification of synaptic connections between neurons. The simple perceptron is just able to handle linearly separa-
Specifically, when an axon of cell A is near enough to excite ble or linearly independent problems. By taking the partial
a cell B and repeatedly or persistently takes part in firing derivative of the error of the network with respect to each
it, some growth process or metabolic change takes place weight, we will learn a little about the direction the error
in one or both cells such that A’s efficiency, as one of the of the network is moving.
cells firing B, is increased. The principles underlying this In fact, if we take the negative of this derivative (i.e.
statement have become known as Hebbian Learning. Vir- the rate change of the error as the value of the weight
tually, most of the neural network learning techniques can increases) and then proceed to add it to the weight, the error
be considered as a variant of the Hebbian learning rule. The will decrease until it reaches a local minima. This makes
basic idea is that if two neurons are active simultaneously, sense because if the derivative is positive, this tells us that
their interconnection must be strengthened. If we consider the error is increasing when the weight is increasing. The
a single layer net, one of the interconnected neurons will obvious thing to do then is to add a negative value to the
be an input unit and one an output unit. If the data are rep- weight and vice versa if the derivative is negative. Because
resented in bipolar form, it is easy to express the desired the taking of these partial derivatives and then applying
weight update as them to each of the weights takes place, starting from the
output layer to hidden layer weights, then the hidden layer
wi (new) = wi (old) + xi o, to input layer weights (as it turns out, this is necessary since
904 Elements: B – Signal Conditioning
changing these set of weights requires that we know the data to get the network familiarized with noise and natural
partial derivatives calculated in the layer downstream), this variability in real data.
algorithm has been called the backpropagation algorithm. Poor training data inevitably leads to an unreliable and
A neural network can be trained in two different modes: unpredictable network. Usually, the network is trained for
online and batch modes. The number of weight updates of a prefixed number of epochs or when the output error
the two methods for the same number of data presentations decreases below a particular error threshold.
is very different. Special care is to be taken not to overtrain the network.
The online method weight updates are computed for By overtraining, the network may become too adapted in
each input data sample, and the weights are modified after learning the samples from the training set, and thus may
each sample. be unable to accurately classify samples outside of the
An alternative solution is to compute the weight update training set.
for each input sample, but store these values during one Figure 3 illustrates the classification results of an over-
pass through the training set which is called an epoch. trained network. The task is to correctly classify two pat-
At the end of the epoch, all the contributions are added, terns X and Y. Training patterns are shown by ‘ ’ and test
and only then the weights will be updated with the compos- patterns by ‘ ’. The test patterns were not shown during
ite value. This method adapts the weights with a cumulative the training phase.
weight update, so it will follow the gradient more closely. As shown in Figure 3 (left side), each class of test data
It is called the batch-training mode. has been classified correctly, even though they were not
Training basically involves feeding training samples as seen during training. The trained network is said to have
input vectors through a neural network, calculating the error good generalization performance. Figure 3 (right side) illus-
of the output layer, and then adjusting the weights of the trates some misclassification of the test data. The network
network to minimize the error. initially learns to detect the global features of the input
The average of all the squared errors (E) for the outputs and, as a consequence, generalizes very well. But after
is computed to make the derivative easier. Once the error prolonged training, the network starts to recognize indi-
is computed, the weights can be updated one by one. In the vidual input/output pairs rather than settling for weights
batched mode variant, the descent is based on the gradient that generally describe the mapping for the whole training
∇E for the total training set set (Fausett, 1994).
δE
wij (n) = −η∗ + α ∗ wij (n − 1) (4)
δwij 5.1 Choosing the number of neurons
where η and α are the learning rate and momentum respec-
tively. The number of hidden neurons affects how well the network
The momentum term determines the effect of past weight is able to separate the data. A large number of hidden
changes on the current direction of movement in the neurons will ensure correct learning, and the network is
weight space. A good choice of both η and α are required able to correctly predict the data it has been trained on,
for the training success and the speed of the neural- but its performance on new data, its ability to generalize,
network learning. is compromised. With too few hidden neurons, the network
It has been proven that backpropagation learning with may be unable to learn the relationships amongst the data
sufficient hidden layers can approximate any nonlinear and the error will fail to fall below an acceptable level.
function to arbitrary accuracy. This makes backpropaga- Thus, selection of the number of hidden neurons is a
tion learning neural network a good candidate for signal crucial decision.
prediction and system modeling.
Y Y
5 TRAINING AND TESTING NEURAL
NETWORKS
The best training procedure is to compile a wide range of
examples (for more complex problems, more examples are X X
required), which exhibit all the different characteristics of (a) Good generalization (b) Poor generalization
the problem. Training samples Test samples
To create a robust and reliable network, in some cases,
some noise or other randomness is added to the training Figure 3. Illustration of generalization performance.
Artificial Neural Networks 905
5.2 Choosing the initial weights The fourth method of Levenberg and Marquardt is specif-
ically adapted to the minimization of an error function that
The learning algorithm uses a steepest descent technique, arises from a squared error criterion of the form we are
which rolls straight downhill in weight space until the assuming. A common feature of these training algorithms
first valley is reached. This makes the choice of initial is the requirement of repeated efficient calculation of gradi-
starting point in the multidimensional weight space critical. ents. The reader can refer to Bishop (1995) for an extensive
However, there are no recommended rules for this selection coverage of higher-order learning algorithms.
except trying several different starting weight values to see Even though artificial neural networks are capable of per-
if the network results are improved. forming a wide variety of tasks, in practice, sometimes, they
deliver only marginal performance. Inappropriate topology
selection and learning algorithm are frequently blamed.
There is little reason to expect that one can find a uni-
5.3 Choosing the learning rate formly best algorithm for selecting the weights in a feed-
forward artificial neural network. This is in accordance
Learning rate effectively controls the size of the step that is with the no free lunch theorem, which explains that for
taken in multidimensional weight space when each weight any algorithm, any elevated performance over one class of
is modified. If the selected learning rate is too large, then the problems is exactly paid for in performance over another
local minimum may be overstepped constantly, resulting in class (Macready and Wolpert, 1997).
oscillations and slow convergence to the lower error state. The design of artificial neural networks using evolu-
If the learning rate is too low, the number of iterations tionary algorithms has been widely explored. Evolutionary
required may be too large, resulting in slow performance. algorithms are used to adapt the connection weights, net-
work architecture, and so on, according to the problem
environment.
6 HIGHER ORDER LEARNING A distinct feature of evolutionary neural networks is their
ALGORITHMS adaptability to a dynamic environment. In other words, such
neural networks can adapt to an environment as well as
Backpropagation (BP) often gets stuck at a local minimum changes in the environment. The two forms of adaptation,
mainly because of the random initialization of weights. evolution and learning in evolutionary artificial neural net-
For some initial weight settings, BP may not be able works, make their adaptation to a dynamic environment
to reach a global minimum of weight space, while for much more effective and efficient than the conventional
other initializations the same network is able to reach an learning approach. Refer to Abraham (2004) for more tech-
optimal minimum. nical information related to evolutionary design of neu-
A long recognized bane of analysis of the error sur- ral networks.
face and the performance of training algorithms is the
presence of multiple stationary points, including multiple 7 DESIGNING ARTIFICIAL NEURAL
minima.
Empirical experience with training algorithms show that NETWORKS
different initialization of weights yield different resulting
networks. Hence, multiple minima not only exist, but there To illustrate the design of artificial neural networks, the
may be huge numbers of them. Mackey-Glass chaotic time series (Box and Jenkins, 1970)
In practice, there are four types of optimization algo- benchmark is used. The performance of the designed neural
rithms that are used to optimize the weights. The first three network is evaluated for different architectures and activa-
methods, gradient descent, conjugate gradients, and quasi- tion functions. The Mackey-Glass differential equation is a
Newton, are general optimization methods whose operation chaotic time series for some values of the parameters x(0)
can be understood in the context of minimization of a and τ .
quadratic error function. dx(t) 0.2x(t − τ )
Although the error surface is surely not quadratic, for = − 0.1 x(t). (5)
dt 1 + x 10 (t − τ )
differentiable node functions, it will be so in a sufficiently
small neighborhood of a local minimum, and such an We used the value x(t − 18), x(t − 12), x(t − 6), x(t)
analysis provides information about the behavior of the to predict x(t + 6). Fourth order Runge-Kutta method was
training algorithm over the span of a few iterations and used to generate 1000 data series. The time step used in the
also as it approaches its goal. method is 0.1 and initial condition were x(0) = 1.2, τ =
906 Elements: B – Signal Conditioning
Table 1. Training and test performance for Mackey-Glass Series Table 2. Mackey-Glass time series: training and generalization
for different architectures. performance for different activation functions.
Hidden neurons Root mean-squared error Activation function Root mean-squared error
Training data Test data Training Test
14 0.0890 0.0880 TSAF 0.0439 0.0437
16 0.0824 0.0860 LSAF 0.0970 0.0950
18 0.0764 0.0750
20 0.0452 0.0442
24 0.0439 0.0437
24 1.06
Hidden neurons
20 0.89
17, x(t) = 0 for t < 0. The first 500 data sets were used
18 0.8
for training and remaining data for testing.
16 0.71
14 0.62
7.1 Network architecture 0.5 0.6 0.7 0.8 0.9 1 1.1
Billion flops
A feed-forward neural network with four input neurons, one Figure 5. Computational complexity for different architectures.
hidden layer and one output neuron is used. Weights were
randomly initialized and the learning rate and momentum two node transfer functions. The generalization looks better
are set at 0.05 and 0.1 respectively. The numbers of hidden with TSAF.
neurons are varied (14, 16, 18, 20, 24) and the general- Figure 5 illustrates the computational complexity in bil-
ization performance is reported in Table 1. All networks lion flops for different numbers of hidden neurons. At
were trained for an identical number of stochastic updates present, neural network design relies heavily on human
(2500 epochs). experts who have sufficient knowledge about the differ-
ent aspects of the network and the problem domain. As
the complexity of the problem domain increases, manual
design becomes more difficult.
7.2 Role of activation functions
The effect of two different node activation functions in 8 SELF-ORGANIZING FEATURE MAP
the hidden layer, log-sigmoidal activation function LSAF
AND RADIAL BASIS FUNCTION
and tanh-sigmoidal activation function TSAF), keeping
24 hidden neurons for the backpropagation learning algo- NETWORK
rithm, is illustrated in Figure 4. Table 2 summarizes the
empirical results for training and generalization for the 8.1 Self-organizing feature map
Self-organizing Feature Maps SOFM is a data visualization
0.8
technique proposed by Kohonen (1988), which reduces
0.7
the dimensions of data through the use of self-organizing
0.6
neural networks.
0.5
A SOFM learns the categorization, topology, and dis-
RMSE
0.4 tribution of input vectors. SOFM allocate more neurons
0.3 to recognize parts of the input space where many input
0.2 vectors occur and allocate fewer neurons to parts of the
0.1 input space where few input vectors occur. Neurons next
0 to each other in the network learn to respond to similar
25 150 500 1000 1500 2000 2500
vectors.
LSAF TSAF Epochs
SOFM can learn to detect regularities and correlations
Figure 4. Convergence of training for different node trans- in their input and adapt their future responses to that input
fer function. accordingly. An important feature of the SOFM learning
Artificial Neural Networks 907
algorithm is that it allows neurons that are neighbors to the 9 RECURRENT NEURAL NETWORKS
winning neuron to be output values. Thus, the transition of AND ADAPTIVE RESONANCE THEORY
output vectors is much smoother than that obtained with
competitive layers, where only one neuron has an output at
a time.
9.1 Recurrent neural networks
The problem that data visualization attempts to solve
is that humans simply cannot visualize high-dimensional Recurrent networks are the state of the art in nonlinear
data. The way SOFM goes about reducing dimensions is time series prediction, system identification, and temporal
by producing a map of usually 1 or 2 dimensions, which pattern classification. As the output of the network at time
plot the similarities of the data by grouping similar data t is used along with a new input to compute the output of
items together (data clustering). In this process, SOFM the network at time t + 1, the response of the network is
accomplish two things, they reduce dimensions and display dynamic (Mandic and Chambers, 2001).
similarities. Time Lag Recurrent Networks (TLRN) are multilayered
It is important to note that while a self-organizing map perceptrons extended with short-term memory structures
does not take long to organize itself so that neighboring that have local recurrent connections. The recurrent neural
neurons recognize similar inputs, it can take a long time for network is a very appropriate model for processing temporal
the map to finally arrange itself according to the distribution (time-varying) information.
of input vectors. Examples of temporal problems include time-series pre-
diction, system identification, and temporal pattern recog-
nition. A simple recurrent neural network could be con-
structed by a modification of the multilayered feed-forward
8.2 Radial basis function network network with the addition of a ‘context layer’. The context
layer is added to the structure, which retains information
between observations. At each time step, new inputs are
The Radial Basis Function (RBF) network is a three-layer
fed to the network. The previous contents of the hidden
feed-forward network that uses a linear transfer function for
layer are passed into the context layer. These then feed
the output units and a nonlinear transfer function (normally
back into the hidden layer in the next time step. Initially,
the Gaussian) for the hidden layer neurons (Chen, Cowan
the context layer contains nothing, so the output from the
and Grant, 1991). Radial basis networks may require more
hidden layer after the first input to the network will be the
neurons than standard feed-forward backpropagation net-
same as if there is no context layer. Weights are calculated
works, but often they can be designed with lesser time.
in the same way for the new connections from and to the
They perform well when many training data are avail-
context layer from the hidden layer.
able.
The training algorithm used in TLRN (backpropagation
Much of the inspiration for RBF networks has come from
through time) is more advanced than standard backprop-
traditional statistical pattern classification techniques. The
agation algorithm. Very often, TLRN requires a smaller
input layer is simply a fan-out layer and does no processing.
network to learn temporal problems when compared to
The second or hidden layer performs a nonlinear mapping
MLP that use extra inputs to represent the past samples.
from the input space into a (usually) higher dimensional
TLRN is biologically more plausible and computationally
space whose activation function is selected from a class of
more powerful than other adaptive models such as the hid-
functions called basis functions.
den Markov model.
The final layer performs a simple weighted sum with a
Some popular recurrent network architectures are the
linear output. Contrary to BP networks, the weights of the
Elman recurrent network in which the hidden unit activation
hidden layer basis units (input to hidden layer) are set using
values are fed back to an extra set of input units and the
some clustering techniques. The idea is that the patterns in
Jordan recurrent network in which output values are fed
the input space form clusters. If the centers of these clusters
back into hidden units.
are known, then the Euclidean distance from the cluster
center can be measured. As the input data moves away
from the connection weights, the activation value reduces.
This distance measure is made nonlinear in such a way that 9.2 Adaptive resonance theory
for input data close to a cluster center gets a value close to
1. Once the hidden layer weights are set, a second phase Adaptive Resonance Theory (ART) was initially introduced
of training (usually backpropagation) is used to adjust the by Grossberg (1976) as a theory of human information
output weights. processing. ART neural networks are extensively used for
908 Elements: B – Signal Conditioning
supervised and unsupervised classification tasks and func- Box, G.E.P. and Jenkins, G.M. (1970) Time Series Analy-
tion approximation. sis, Forecasting and Control, Holden Day, San Francisco,
There exist many different variations of ART networks CA.
today (Carpenter and Grossberg, 1998). For example, ART1 Carpenter, G. and Grossberg, S. (1998) in Adaptive Resonance
performs unsupervised learning for binary input patterns, Theory (ART), The Handbook of Brain Theory and Neural
Networks, (ed. M.A. Arbib), MIT Press, Cambridge, MA, (pp.
ART2 is modified to handle both analog and binary input 79–82).
patterns, and ART3 performs parallel searches of distributed
Chen, S., Cowan, C.F.N. and Grant, P.M. (1991) Orthogonal
recognition codes in a multilevel network hierarchy. Fuzzy Least Squares Learning Algorithm for Radial Basis Func-
ARTMAP represents a synthesis of elements from neural tion Networks. IEEE Transactions on Neural Networks, 2(2),
networks, expert systems, and fuzzy logic. 302–309.
Fausett, L. (1994) Fundamentals of Neural Networks, Prentice
Hall, USA.
10 SUMMARY
Grossberg, S. (1976) Adaptive Pattern Classification and Uni-
versal Recoding: Parallel Development and Coding of Neural
This section presented the biological motivation and fun-
Feature Detectors. Biological Cybernetics, 23, 121–134.
damental aspects of modeling artificial neural networks.
Hebb, D.O. (1949) The Organization of Behavior, John Wiley,
Performance of feed-forward artificial neural networks for
New York.
a function approximation problem is demonstrated. Advan-
Kohonen, T. (1988) Self-Organization and Associative Memory,
tages of some specific neural network architectures and
Springer-Verlag, New York.
learning algorithms are also discussed.
Macready, W.G. and Wolpert, D.H. (1997) The No Free Lunch
Theorems. IEEE Transactions on Evolutionary Computing,
REFERENCES 1(1), 67–82.
Mandic, D. and Chambers, J. (2001) Recurrent Neural Networks
Abraham, A. (2004) Meta-Learning Evolutionary Artificial Neu- for Prediction: Learning Algorithms, Architectures and Stabil-
ral Networks, Neurocomputing Journal, Vol. 56c, Elsevier Sci- ity, John Wiley & Sons, New York.
ence, Netherlands, (1–38). McCulloch, W.S. and Pitts, W.H. (1943) A Logical Calculus of
Bishop, C.M. (1995) Neural Networks for Pattern Recognition, the Ideas Immanent in Nervous Activity. Bulletin of Mathemat-
Oxford University Press, Oxford, UK. ical Biophysics, 5, 115–133.
130: Rule-based Expert Systems
Ajith Abraham
Oklahoma State University, Stillwater, OK, USA
thumb – could help find good heuristics easier. Humans use
1 Problem Solving Using Heuristics 909 heuristics a great deal in their problem solving. Of course,
2 What are Rule-based Systems? 910 if the heuristic does fail, it is necessary for the problem
solver to either pick another heuristic, or know that it is
3 Inference Engine in Rule-based Systems 911
appropriate to give up.
4 Expert System Development 911
Choosing random solutions, adopting greedy approaches,
5 Fuzzy Expert Systems 912 evolving the basic heuristics for finding better heuris-
6 Modeling Fuzzy Expert Systems 914 tics are just some of the popular approaches used
7 Illustration of Fuzzy Expert System Design 914 in heuristic problem solving (Michalewicz and Fogel,
8 Adaptation of Fuzzy Inference Systems 918 1999).
9 Summary 918 Heuristic problem solving involves finding a set of rules,
or a procedure, that finds satisfactory solutions to a specific
References 919
problem. A good example is finding one’s way through
a maze. To make the way toward the final goal, a step-
by-step movement is necessary. Very often false moves are
1 PROBLEM SOLVING USING made but in most cases we solve the problem without much
HEURISTICS difficulty. For the maze problem, a simple heuristic rule
could be ‘choose the direction that seems to make progress’.
A general introduction to artificial intelligence methods Another good example is the job shop scheduling prob-
of measurement signal processing is given in Article 128, lem wherein the task is to schedule Jn independent jobs,
Nature and Scope of AI Techniques, Volume 2. where n = {1, 2, . . . .N } on Rm heterogeneous resources
Problem solving is the process of finding a solution when and m = {1, 2, . . . ., M}, with an objective of minimizing
the path leading to that solution is uncertain. Even though the completion time of all the jobs and utilizing all the
we are familiar with several problem-solving techniques, in resources effectively.
the real world, sometimes many problems cannot be solved Each job Jn has processing requirement Pj cycles and
by a technique we are familiar with. resource Rm has speed of Si cycles/unit time. Any job Jn
Surprisingly, for some complicated problems, no straight- has to be processed in resource Rm , until completion. If Cj
forward solution technique is known at all. For these is the completion time and the last job j finishes processing,
problems, heuristic solution techniques may be the only then we define Cmax = max{Cj , j = 1, . . . , N }, the make-
alternative. A heuristic can be simplified as a strategy that span and Cj , as the flow-time.
is powerful and general, but not absolutely guaranteed to The task is to find an optimal schedule that optimizes
provide best solutions. the flow-time and make-span. Some simple heuristic rules
Heuristic methods are very problem specific. Previous to achieve this are by scheduling the Shortest Job on the
experience and some general rules – often called rules of Fastest Resource (SJFR), which would minimize Cj or by
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
910 Elements: B – Signal Conditioning
scheduling the Longest Job on the Fastest Resource (LJFR), the knowledge base, a different problem can be solved
which would minimize Cmax . using the same program without reprogramming efforts.
Minimizing Cj asks that the average job finishes Moreover, expert systems could explain the reasoning
quickly, at the expense of the largest job taking a long process and handle levels of confidence and uncertainty,
time, whereas minimizing Cmax , asks that no job takes too which conventional algorithms do not handle (Giarratano
long, at the expense of most jobs taking a long time. and Riley, 1989). Some of the important advantages of
In summary, minimization of Cmax will result in expert systems are as follows:
maximization of Cj , which makes the problem more
interesting. • ability to capture and preserve irreplaceable human
By contrast, algorithms are straightforward procedures experience;
that are guaranteed to work every time for they are fully • ability to develop a system more consistent than human
determinate and time invariant. For example, certain daily experts;
routine tasks could be formulated in a strict algorithm • minimize human expertise needed at a number of
format (example, starting up an automobile). However, for locations at the same time (especially in a hostile
a ‘problem solver’ to be more adaptive, novel elements or environment that is dangerous to human health);
new circumstances must be introduced. Many real-world • solutions can be developed faster than human experts.
problems cannot be reduced to algorithms, which leads us
to the quest to find more powerful techniques. The basic components of an expert system are illustrated
in Figure 1. The knowledge base stores all relevant infor-
mation, data, rules, cases, and relationships used by the
2 WHAT ARE RULE-BASED SYSTEMS? expert system. A knowledge base can combine the knowl-
edge of multiple human experts. A rule is a conditional
Conventional problem-solving computer programs make statement that links given conditions to actions or out-
use of well-structured algorithms, data structures, and crisp comes. A frame is another approach used to capture and
reasoning strategies to find solutions. For the difficult store knowledge in a knowledge base. It relates an object
problems with which expert systems are concerned, it may or item to various facts or values. A frame-based repre-
be more useful to employ heuristics: strategies that often sentation is ideally suited for object-oriented programming
lead to the correct solution, but that also sometimes fail. techniques. Expert systems making use of frames to store
Conventional rule-based expert systems, use human knowledge are also called frame-based expert systems.
expert knowledge to solve real-world problems that The purpose of the inference engine is to seek infor-
normally would require human intelligence. Expert mation and relationships from the knowledge base and to
knowledge is often represented in the form of rules or as provide answers, predictions, and suggestions in the way a
data within the computer. human expert would. The inference engine must find the
Depending upon the problem requirement, these rules and right facts, interpretations, and rules and assemble them
data can be recalled to solve problems. Rule-based expert correctly. Two types of inference methods are commonly
systems have played an important role in modern intelligent used – Backward chaining is the process of starting with
systems and their applications in strategic goal setting, conclusions and working backward to the supporting facts.
planning, design, scheduling, fault monitoring, diagnosis Forward chaining starts with the facts and works forward
and so on. to the conclusions.
With the technological advances made in the last decade,
today’s users can choose from dozens of commercial Expert knowledge Users
software packages having friendly graphic user interfaces
(Ignizio, 1991). Conventional computer programs perform
tasks using a decision-making logic containing very little Knowledge base
knowledge other than the basic algorithm for solving that Knowledge base User interface
acquisition facility
specific problem. The basic knowledge is often embedded
as part of the programming code, so that as the knowledge
changes, the program has to be rebuilt. Knowledge-based
expert systems collect the small fragments of human know- Explanation
Inference engine
how into a knowledge base, which is used to reason through facility
a problem, using the knowledge that is appropriate. An
important advantage here is that within the domain of Figure 1. Architecture of a simple expert system.
Rule-based Expert Systems 911
The explanation facility allows a user to understand how with the preconditions specifying as precisely as possible
the expert system arrived at certain results. The overall when different rules should fire.
purpose of the knowledge acquisition facility is to provide In the backward chaining method, processing starts with
a convenient and efficient means for capturing and storing the desired goal, and then attempts to find evidence for
all components of the knowledge base. proving the goal. Returning to the same example, the task
Very often specialized user interface software is used for to prove that D is true would be initiated by first finding a
designing, updating, and using expert systems. The purpose rule that proves D. Rule 4 does so, which also provides
of the user interface is to ease use of the expert system for a subgoal to prove that Z is true. Now Rule 2 comes
developers, users, and administrators. into play, and as it is already known that A is true, the
new subgoal is to show that X is true. Rule 3 provides
the next subgoal of proving that B is true. But that B is
3 INFERENCE ENGINE IN RULE-BASED true is one of the given assertions. Therefore, it could be
concluded that X is true, which implies that Z is true, which
SYSTEMS
in turn also implies that D is true. Backward chaining is
useful in situations where the quantity of data is potentially
A rule-based system consists of if-then rules, a bunch of very large and where some specific characteristic of the
facts, and an interpreter controlling the application of the system under consideration is of interest. If there is not
rules, given the facts. much knowledge what the conclusion might be, or there is
These if-then rule statements are used to formulate some specific hypothesis to test, forward chaining systems
the conditional statements that comprise the complete may be inefficient. In principle, we can use the same set
knowledge base. A single if-then rule assumes the form of rules for both forward and backward chaining. In the
‘if x is A then y is B’ and the if-part of the rule ‘x is A’ is case of backward chaining, since the main concern is with
called the antecedent or premise, while the then-part of the matching the conclusion of a rule against some goal that
rule ‘y is B’ is called the consequent or conclusion. There is to be proved, the ‘then’ (consequent) part of the rule is
are two broad kinds of inference engines used in rule-based usually not expressed as an action to take but merely as a
systems: forward chaining and backward chaining systems. state, which will be true if the antecedent part(s) are true
In a forward chaining system, the initial facts are pro- (Donald, 1986).
cessed first, and keep using the rules to draw new conclu-
sions given those facts. In a backward chaining system, the
hypothesis (or solution/goal) we are trying to reach is pro-
cessed first, and keep looking for rules that would allow 4 EXPERT SYSTEM DEVELOPMENT
to conclude that hypothesis. As the processing progresses,
new subgoals are also set for validation. Forward chaining Steps in the expert systems development process include
systems are primarily data-driven, while backward chain- determining the actual requirements, knowledge acquisi-
ing systems are goal-driven. Consider an example with the tion, constructing expert system components, implement-
following set of if-then rules ing results, and formulating a procedure for maintenance
and review.
Rule 1: If A and C then Y Knowledge acquisition is the most important element
Rule 2: If A and X then Z in the development of expert system (Niwa, Sasaki and
Rule 3: If B then X Ihara, 1988). Knowledge could be obtained by interviewing
Rule 4: If Z then D domain experts and/or learning by experience.
Very often people express knowledge as natural language
If the task is to prove that D is true, given A and B are true. (spoken language), or using letters or symbolic terms.
According to forward chaining, start with Rule 1 and go on There exist several methods to extract human knowledge.
downward till a rule that fires is found. Rule 3 is the only Cognitive Work Analysis (CWA) and the Cognitive Task
one that fires in the first iteration. After the first iteration, Analysis (CTA) provide frameworks to extract knowledge.
it can be concluded that A, B, and X are true. The second The CWA is a technique to analyze, design, and evaluate
iteration uses this valuable information. After the second the human computer interactive systems (Vicente, 1999).
iteration, Rule 2 fires adding Z is true, which in turn helps The CTA is a method to identify cognitive skill, mental
Rule 4 to fire, proving that D is true. Forward chaining demands, and needs to perform task proficiency (Militallo
strategy is especially appropriate in situations where data and Hutton, 1998). This focuses on describing the represen-
are expensive to collect, but few in quantity. However, tation of the cognitive elements that defines goal generation
special care is to be taken when these rules are constructed, and decision-making. It is a reliable method for extracting
912 Elements: B – Signal Conditioning
human knowledge because it is based on the observations membership grade (or membership value) between zero and
or an interview. one. Obviously (1) is a simple extension of the definition
Most expert systems are developed using specialized of a classical set in which the characteristic function is
software tools called shells. These shells come equipped permitted to have any values between zero and one.
with an inference mechanism (backward chaining, forward The intersection of two fuzzy sets A and B is specified
chaining, or both), and require knowledge to be entered in general by a function T : [0,1] × [0,1] → [0,1], which
according to a specified format. aggregates two membership grades as follows:
One of the most popular shells widely used throughout
the government, industry, and academia is the CLIPS µA∩B (x) = T (µA (x), µB (x)) = µA (x)¯∗µB (x) (2)
(CLIPS, 2004). CLIPS is an expert system tool that provides
where ∗¯ is a binary operator for the function T . This class
a complete environment for the construction of rule- and/or
of fuzzy intersection operators are usually referred to as
object-based expert systems. CLIPS provides a cohesive
T-norm operators (Jang, Sun and Mizutani, 1997). Four of
tool for handling a wide variety of knowledge with support
the most frequently used T-norm operators are
for three different programming paradigms: rule-based,
object-oriented, and procedural. CLIPS is written in C Minimum: Tmin (a, b) = min(a, b) = a ∧ b (3)
for portability and speed and has been installed on many
different operating systems without code changes. Algebraic product: Tap (a, b) = ab (4)
Bounded product: Tbp (a, b) = 0 ∨ (a + b − 1) (5)
5 FUZZY EXPERT SYSTEMS a, if b = 1
Drastic product: Tdp (a, b) = b, if a = 1 (6)
The world of information is surrounded by uncertainty 0, if a, b < 1
and imprecision. The human reasoning process can handle Like intersection, the fuzzy union operator is specified in
inexact, uncertain, and vague concepts in an appropriate general by a function S: [0,1] × [0,1] → [0,1], which
manner. Usually, the human thinking, reasoning, and per- aggregates two membership grades as follows:
ception process cannot be expressed precisely. These types
of experiences can rarely be expressed or measured using µA∪B (x) = S(µA (x), µB (x)) = µA (x) ∓ µB (x) (7)
statistical or probability theory. Fuzzy logic provides a
framework to model uncertainty, the human way of think- where ∓ is the binary operator for the function S. This class
ing, reasoning, and the perception process. Fuzzy systems of fuzzy union operators are often referred to as T-conorm
were first introduced by Zadeh (1965). (or S-norm) operators (Jang, Sun and Mizutani, 1997). Four
A fuzzy expert system is simply an expert system that of the most frequently used T-conorm operators are
uses a collection of fuzzy membership functions and rules,
Maximum: Smax (a, b) = max(a, b) = a ∨ b (8)
instead of Boolean logic, to reason about data (Schneider
et al., 1996). The rules in a fuzzy expert system are usually Algebraic sum: Sas (a, b) = a + b − ab (9)
of a form similar to the following:
Bounded sum: Sbs (a, b) = 1 ∧ (a + b) (10)
If A is low and B is high then X = medium
where A and B are input variables, X is an output variable. a, if b = 0
Here low, high, and medium are fuzzy sets defined on A, B, Drastic sum: Sds (a, b) = b, if a = 0 (11)
and X respectively. The antecedent (the rule’s premise) 1, if a, b > 0
describes to what degree the rule applies, while the rule’s
Both the intersection and union operators retain some
consequent assigns a membership function to each of one
properties of the classical set operation. In particular, they
or more output variables.
are associative and commutative.
Let X be a space of objects and x be a generic element
Figure 2 illustrates the basic architecture of a fuzzy
of X. A classical set A, A ⊆ X, is defined as a collection
expert system. The main components are a fuzzification
of elements or objects x ∈ X, such that x can either belong
interface, a fuzzy rule base (knowledge base), an inference
or not belong to the set A. A fuzzy set A in X is defined
engine (decision-making logic), and a defuzzification inter-
as a set of ordered pairs
face. The input variables are fuzzified whereby the member-
A = {(x, µA (x))|x ∈ X} (1) ship functions defined on the input variables are applied to
their actual values, to determine the degree of truth for each
where µA (x) is called the membership function (MF) for rule antecedent. Fuzzy if-then rules and fuzzy reasoning are
the fuzzy set A. The MF maps each element of X to a the backbone of fuzzy expert systems, which are the most
Rule-based Expert Systems 913
Fuzzification Inference Defuzzification
interface engine interface
Crisp Fuzzy Fuzzy Crisp
input input output output
Rules
Fuzzy rule
base
Figure 2. Basic architecture of a fuzzy expert system.
Input MF Output MF
min max
A1 B1 C1
c’1
m m m
C’
m
X Y Z1
A2 B2 C2
m m c’2 Z (COA)
m
X Y Z2
X Y Output Z
Input (x,y )
Figure 3. Mamdani fuzzy inference system using min and max for T-norm and T-conorm operators.
important modeling tools based on fuzzy set theory. The According to Mamdani, fuzzy inference system (Mam-
fuzzy rule base is characterized in the form of if-then rules dani and Assilian, 1975) – see Figure 3 – the rule ante-
in which the antecedents and consequents involve linguis- cedents and consequents are defined by fuzzy sets and has
tic variables. The collection of these fuzzy rules forms the the following structure:
rule base for the fuzzy logic system. Using suitable infer-
ence procedure, the truth value for the antecedent of each if x is A1 and y is B1 then z1 = C1 (12)
rule is computed, and applied to the consequent part of each
rule. This results in one fuzzy subset to be assigned to each There are several defuzzification techniques. The most
output variable for each rule. Again, by using suitable com- widely used defuzzification technique uses the centroid of
position procedure, all the fuzzy subsets assigned to each area method as follows
output variable are combined together to form a single fuzzy
subset for each output variable. Finally, defuzzification is µ (z) z dz
Centroid of area ZCOA = Z A (13)
applied to convert the fuzzy output set to a crisp output. Z µA (z) dz
The basic fuzzy inference system can take either fuzzy
inputs or crisp inputs, but the outputs it produces are always where µA (z) is the aggregated output MF.
fuzzy sets. The defuzzification task extracts the crisp output Takagi and Sugeno (1985) proposed an inference scheme
that best represents the fuzzy set. With crisp inputs and in which the conclusion of a fuzzy rule is constituted by a
outputs, a fuzzy inference system implements a nonlinear weighted linear combination of the crisp inputs rather than a
mapping from its input space to output space through a fuzzy set. A basic Takagi–Sugeno fuzzy inference system
number of fuzzy if-then rules. is illustrated in Figure 4 and the rule has the following
In what follows, the two most popular fuzzy inference structure
systems are introduced that have been widely deployed in
various applications. The differences between these two if x is A1 and y is B1 , then z1 = p1 x + q1 y + r1 (14)
fuzzy inference systems lie in the consequents of their
fuzzy rules, and thus their aggregation and defuzzification where p1 , q1 , and r1 are linear parameters. TSK Tak-
procedures differ accordingly. agi–Sugeno Kang fuzzy controller usually needs a smaller
914 Elements: B – Signal Conditioning
A1 B1
m m z1 =
w1
p1*x + q1*y + r1
X Y
m m z2 =
A2 B2
w2 p2*x + q2*y + r2
X Y w1*z1 + w2*z2
X Y Z = w1 + w2
Output Z
Input (x,y )
Figure 4. Takagi–Sugeno fuzzy inference system using a min or product as T-norm operator.
number of rules, because their output is already a linear
function of the inputs rather than a constant fuzzy set.
Large
R7 R8 R9
6 MODELING FUZZY EXPERT SYSTEMS
Medium
Fuzzy expert system modeling can be pursued using the
Input-2
R6 R5 R4
following steps.
• Select relevant input and output variables. Determine
the number of linguistic terms associated with each
Small
input/output variable. Also, choose the appropriate fam-
R1 R2 R3
ily of membership functions, fuzzy operators, reasoning
mechanism, and so on.
• Choose a specific type of fuzzy inference system
Small Medium Large
(for example, Mamdani, Takagi–Sugeno etc.). In most
cases, the inference of the fuzzy rules is carried Input-1
out using the ‘min’ and ‘max’ operators for fuzzy
intersection and union. Figure 5. Example showing how the two-dimensional spaces are
partitioned using three trapezoidal membership functions per input
• Design a collection of fuzzy if-then rules (knowledge dimension. A simple if-then rule will appear as If input-1 is
base). To formulate the initial rule base, the input space medium and input 2 is large, then rule R8 is fired .
is divided into multidimensional partitions and then
actions are assigned to each of the partitions.
The task is to develop a fuzzy expert system to forecast
In most applications, the partitioning is achieved using one- the reactive power (P ) at time t + 1 by knowing the load
dimensional membership functions using fuzzy if-then rules current (I ) and voltage (V ) at time t. The experiment
as illustrated in Figure 5. The consequent parts of the rule system consists of two stages: developing the fuzzy expert
represent the actions associated with each partition. It is system, and performance evaluation using the test data. The
evident that the MFs and the number of rules are tightly model has two in–out variables (V and I ) and one output
related to the partitioning. variable (P ). Training and testing data sets were extracted
randomly from the master dataset. Sixty percent of data
was used for training and the remaining 40% for testing
7 ILLUSTRATION OF FUZZY EXPERT (Abraham and Khan, 2003).
SYSTEM DESIGN
This section illustrates the development of a reactive power 7.1 Design and experiments: fuzzy expert systems
prediction model using Mamdani and Takagi–Sugeno fuzzy
inference expert systems. The MatLab fuzzy logic tool- First, the effects of (a) shape and quantity of mem-
box was used to simulate the various experiments (Fuzzy bership functions (b) T-norm and T-conorm operators
Logic Tool Box, 2004). (c) defuzzification methods and (d) inference method for
Rule-based Expert Systems 915
designing the fuzzy expert system is analyzed. Experiments each input variable and nine Gaussian MFs for the output
were carried out using four different settings using the same variable were used. The rule base consisted of nine if-
rule base. then rules. ‘min’ and ‘max’ as T-norm and T-conorm
Experiment 1 (To evaluate the effect on the number of operators, and the centroid method of defuzzification for
membership functions) The following settings were used Mamdani FIS and the weighted average defuzzification
for designing the expert system method for Takagi–Sugeno FIS were also used. The
developed fuzzy inference systems using Mamdani and
1. Two triangular membership functions (MFs) for each Takagi–Sugeno models are depicted in Figures 10 and 11.
input variable and four triangular MFs for the output Table 2 summarizes the training and testing RMSE values.
variable (power). Using the grid partitioning method
(Figure 5), four if-then rules were developed. Experiment 3 (To evaluate the effect of fuzzy operators) For
2. Three triangular MFs for each input variable and nine Mamdani FIS, three Gaussian MFs for each input variable
triangular MFs for the output variable (power). The and nine Gaussian MFs for the output variable were used.
rule base consisted of nine if-then rules. The rule base consisted of nine if-then rules. T-norm and T-
conorm operators were ‘product’ and ‘sum’ and the centroid
‘min’ and ‘max’ were used as T-norm and T-conorm oper- method of defuzzification for Mamdani FIS, and weighted
ators and the centroid method of defuzzification for Mam- average defuzzification method for Takagi–Sugeno FIS
dani inference method and weighted average defuzzification were used. Table 3 summarizes the training and testing
method for Takagi–Sugeno Fuzzy Inference System (FIS). RMSE values.
The developed fuzzy inference systems using Mamdani
Experiment 4 (To evaluate the effect of defuzzification
and Takagi–Sugeno models are depicted in Figures 6 to
operators) For the Mamdani FIS, three Gaussian MFs for
9. Table 1 summarizes the training and testing Root Mean
each input variable and nine Gaussian MFs for the output
Squared Error (RMSE) values.
variable were used. The rule base consisted of nine if-
Experiment 2 (To evaluate the effect of shape of membership then rules. T-norm and T-conorm operators were ‘product’
functions) For the Mamdani FIS, three Gaussian MFs for and ‘sum’ and the following defuzzification operators were
tested for Mamdani FIS.
Table 1. Empirical comparison of fuzzy inference systems and
quantity of Membership Functions (MFs). Table 2. Empirical comparison of fuzzy inference systems using
Gaussian MFs.
No. of Mamdani FIS Takagi – Sugeno FIS
MFs Mamdani FIS Takagi – Sugeno FIS
Root mean squared error
Root mean squared error
Training Test Training Test
Training Test Training Test
2 0.401 0.397 0.024 0.023
3 0.348 0.334 0.017 0.016 0.243 0.240 0.021 0.019
Voltage = 0.5 Current = 0.5
Power = 0.5
0 1 0 1
0 1
Figure 6. Mamdani fuzzy inference system using two triangular MFs for input variables.
916 Elements: B – Signal Conditioning
Voltage = 0.5 Current = 0.5
Power = 0.424
0 1 0 1
− 0.16 1.038
Figure 7. Takagi–Sugeno fuzzy inference system using two triangular MFs for input variables.
Voltage = 0.5 Current = 0.5 Power = 0.625
1
9
0 1 0 1
0 1
Figure 8. Mamdani fuzzy inference system using three triangular MFs for input variables.
Voltage = 0.5 Current = 0.527 Power = 0.429
1
9
0 1 0 1
−0.1291 1.058
Figure 9. Takagi–Sugeno fuzzy inference system using three triangular MFs for input variables.
Rule-based Expert Systems 917
Voltage = 0.5 Current = 0.5 Power = 0.584
1
9
0 1 0 1
0 1
Figure 10. Mamdani fuzzy inference system using three Gaussian MFs for input variables.
Voltage = 0.5 Current = 0.5 Power = 0.417
1
9
0 1 0 1
− −
−0.1364 1.095
Figure 11. Takagi–Sugeno fuzzy inference system using three Gaussian MFs for input variables.
Table 3. Empirical comparison of fuzzy inference systems for Table 4 summarizes the training and testing of RMSE
different fuzzy operators. values.
Mamdani FIS Takagi – Sugeno FIS
Discussions of Results and Problem Solution As depicted
Root mean squared error in Table 1, when the number of input MFs were increased
Training Test Training Test from two to three, the RMSE values reduced regardless
of the inference system used. However, when the shape
0.221 0.219 0.019 0.018
of the MF was changed to Gaussian, RMSE values for
Mamdani FIS decreased but the RMSE values for Tak-
• Centroid agi–Sugeno FIS increased (Table 2). Using Gaussian MFs,
• Bisector of Area (BOA) when the T-norm and T-conorm operators were changed
• Mean of Maximum (MOM) to ‘product’ and ‘sum’ (instead of ‘min’ and ‘max’) both
• Smallest of Maximum (SOM). the inference methods performed better (Table 3). Finally,
the selection of an ideal defuzzification operator also has
For the Takagi–Sugeno FIS, the weighted sum and a direct influence in the performance of FIS as shown in
weighted average defuzzification methods were used. Table 4.
918 Elements: B – Signal Conditioning
Table 4. Empirical comparison of fuzzy inference systems for different defuzzification operators.
Mamdani FIS Takagi–Sugeno FIS
Defuzzification RMSE Defuzzification RMSE
Training Test Training Test
Centroid 0.221 0.0219 Weighted sum 0.019 0.018
MOM 0.230 0.232 Weighted average 0.085 0.084
BOA 0.218 0.216
SOM 0.229 0.232
The design of the rule base (number of rules and how computation techniques has been widely explored. Evolu-
the inputs and outputs are related) is also very important tionary Computation (EC) is a population based adaptive
for the good performance of FIS. The role of weighting method, which may be used to solve optimization problems,
factors emphasizing the importance of certain rules also based on the genetic processes of biological organisms
bears a prominent role for the overall performance. When (Michalewicz and Fogel, 1999).
the input/output dimensions become larger, manual design Over many generations, natural populations evolve
becomes tedious and sometimes could even lead to poor according to the principles of natural selection and ‘sur-
design and implementation. vival of the fittest’, first clearly stated by Charles Darwin
in ‘On the Origin of Species’. By mimicking this pro-
cess, EC could ‘evolve’ solutions to real-world problems, if
8 ADAPTATION OF FUZZY INFERENCE they have been suitably encoded (problem representation is
SYSTEMS called chromosome). Automatic adaptation of membership
functions is popularly known as self tuning and the chromo-
Expert knowledge is often the main source to design some encodes parameters of trapezoidal, triangle, logistic,
the fuzzy expert systems. Figure 12 illustrates the various hyperbolic-tangent, Gaussian membership functions, and
parameters and components that need to be adapted for con- so on. Evolutionary search of fuzzy rules can be carried
trolling a process. According to the performance measure out using three approaches. In the first method (Michi-
of the problem environment, the membership functions, gan approach), the fuzzy knowledge base is adapted as a
rule bases, and the inference mechanism are to be adapted result of antagonistic roles of competition and cooperation
(Abraham, 2002). of fuzzy rules.
Neural network learning, self-organizing maps and clus- The second method (Pittsburgh approach), evolves a
tering methods could be used to generate rules. Gradi- population of knowledge bases rather than individual fuzzy
ent descent and its variants could be applied to fine- rules. Reproduction operators serve to provide a new
tune the parameters of parameterized input/output mem- combination of rules and new rules.
bership functions and fuzzy operators (Abraham, 2001). The third method (iterative rule learning approach),
Adaptation of fuzzy inference systems using evolutionary is very much similar to the first method with each
chromosome representing a single rule, but contrary to the
Michigan approach, only the best individual is considered to
Adaptation of fuzzy Performance form part of the solution, discarding the remaining chromo-
inference system measure somes of the population. The evolutionary learning process
builds up the complete rule base through an iterative learn-
ing process (Cordón et al., 2001).
Membership functions
+ if-then rules 9 SUMMARY
Process
−
Fuzzy operators
Rule-based expert systems have been applied in a vast
Knowledge base
number of application areas. An important advantage of
Fuzzy inference system
the fuzzy expert system is that the knowledge is expressed
as easy-to-understand linguistic rules. If we have data, the
Figure 12. Adaptation of fuzzy inference systems. fuzzy expert system can be taught using neural network
Rule-based Expert Systems 919
learning, EC, or other adaptation techniques. It is to Fuzzy Logic Toolbox, The MathWorks (2004) http://www.
be expected that the number of applications will grow mathworks.com/products/fuzzylogic/.
considerably in the future now that success is clearly proven Giarratano, J. and Riley, G. (1989) Expert Systems: Principles and
for these methods. Programming, PWS-Kent Publishing Co, Boston, MA.
Ignizio, J.P. (1991) Introduction to Expert Systems: The Devel-
opment and Implementation of Rule-Based Expert Systems,
McGraw-Hill, Inc, USA.
REFERENCES
Jang, J.S.R., Sun, C.T. and Mizutani, E. (1997) Neuro-Fuzzy and
Soft Computing: A Computational Approach to Learning and
Abraham, A. (2001) Neuro-Fuzzy Systems: State-of-the-Art Mod- Machine Intelligence, Prentice Hall Inc, USA.
eling Techniques, Connectionist Models of Neurons, Learning
Processes, and Artificial Intelligence, in Lecture Notes in Com- Mamdani, E.H. and Assilian, S. (1975) An Experiment in Lin-
puter Science, Vol. 2084, (eds. Mira., Jose and Prieto., Alberto) guistic Synthesis with a Fuzzy Logic Controller. International
Springer Verlag, Germany (pp. 269–276). Journal of Man-Machine Studies, 7(1), 1–13.
Michalewicz, Z. and Fogel, D.B. (1999) How to Solve It: Modern
Abraham, A. (2002) Intelligent Systems: Architectures and Per-
Heuristics, Springer Verlag, Germany.
spectives, Recent Advances in Intelligent Paradigms and Appli-
cations, in Studies in Fuzziness and Soft Computing, Chapter 1, Militallo, L.G., Hutton, R.J.B. (1998) Applied Cognitive Task
(eds A., Abraham, L., Jain and J., Kacprzyk), Springer Verlag, Analysis (ACTA): A Practitioner’s Toolkit for Understanding
Germany (pp. 1–35). Cognitive. Ergonomics, 41(11), 1618–1642.
Abraham, A. and Khan, M.R. (2003) Neuro-Fuzzy Paradigms Niwa, K., Sasaki, K. and Ihara, H. (1988) An Experimental Com-
for Intelligent Energy Management, Innovations in Intelligent parison of Knowledge Representation Schemes, in Principles of
Systems: Design, Management and Applications, in Studies in Expert Systems, (Eds A., Gupta and E.B., Prasad), IEEE Press,
Fuzziness and Soft Computing, Chapter 12, (eds A., Abraham, New York (pp. 133–140).
L., Jain and B., Jan van der Zwaag), Springer Verlag, Germany Schneider, M., Langholz, G., Kandel, A. and Chew, G. (1996)
(pp. 285–314). Fuzzy Expert System Tools, John Wiley & Sons, USA.
CLIPS (2004) Expert System Shell <http://www.ghg.net/clips/ Takagi, T. and Sugeno, M. (1985) Fuzzy identification of systems
CLIPS.html>. and its applications of modeling and control, IEEE Transactions
Cordón, O., Herrera, F., Hoffmann, F. and Magdalena, L. (2001) of Systems. Man and Cybernetics, USA (pp. 116–132).
Genetic Fuzzy Systems: Evolutionary Tuning and Learning of Vicente, K.J. (1999) Cognitive Work Analysis:∼ Towards Safe,
Fuzzy Knowledge Bases, World Scientific Publishing Company, Productive, and Healthy Computer-Based Work, Lawrence Erl-
Singapore. baum Associates, Inc. Press, USA.
Donald, W.A. (1986) A Guide to Expert Systems, Addison- Zadeh, L.A. (1965) Fuzzy Sets. Information and Control, 8,
Wesley, Boston, MA. 338–353.
131: Evolutionary Computation
Ajith Abraham
Oklahoma State University, Stillwater, OK, USA
This is the recombination operation, which is often referred
1 Introduction and Biological Motivation 920 to as crossover because of the way strands of chromosomes
cross over during the exchange. The diversity in the
2 Genetic Algorithms 921
population is achieved by mutation operation.
3 Schema Theorem 922
Evolutionary algorithms are ubiquitous nowadays, hav-
4 Selection and Reproduction 922 ing been successfully applied to numerous problems from
5 GA Demonstrations 924 different domains, including optimization, automatic pro-
6 Evolution Strategies 925 gramming, signal processing, bioinformatics, social sys-
7 Evolutionary Programming 928 tems, and so on. In many cases, the mathematical function,
8 Genetic Programming 928 which describes the problem, is not known, and the values
9 Genetic Programming Basics 929 at certain parameters are obtained from simulations. In con-
trast to many other optimization techniques, an important
10 Summary 931
advantage of evolutionary algorithms is they can cope with
References 931 multimodal functions.
Further Reading 931 Usually found grouped under the term evolutionary com-
putation or evolutionary algorithms (Bäck, 1996), are the
domains of genetic algorithms (GA) (Holland, 1975), evo-
lution strategies (Rechenberg, 1973; Schwefel, 1977), evo-
1 INTRODUCTION AND BIOLOGICAL lutionary programming (Fogel, Owens and Walsh, 1966),
MOTIVATION and genetic programming (Koza, 1992).
These all share a common conceptual base of simulat-
A general introduction to artificial intelligence methods ing the evolution of individual structures via processes
of measurement signal processing is given in Article 128, of selection, recombination, and mutation reproduction,
Nature and Scope of AI Techniques, Volume 2. thereby producing better solutions. The processes depend
In nature, evolution is mostly determined by natural on the perceived performance of the individual structures
selection or different individuals competing for resources as defined by the problem.
in the environment. Those individuals that are better A population of candidate solutions (for the optimization
are more likely to survive and propagate their genetic task to be solved) is initialized. New solutions are cre-
material. The encoding for genetic information (genome) ated by applying reproduction operators (crossover and/or
is done in a way that admits asexual reproduction, mutation). The fitness (how good the solutions are) of the
which results in offspring that are genetically identical resulting solutions is evaluated and suitable selection strat-
to the parent. Sexual reproduction allows some exchange egy is then applied to determine which solutions will be
and reordering of chromosomes, producing offspring that maintained into the next generation. The procedure is then
contain a combination of information from each parent. iterated, as illustrated in Figure 1.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Evolutionary Computation 921
Selection 3. Evolutionary algorithms can also be combined with
Parents more traditional optimization techniques. This may be
as simple as the use of a gradient minimization after
primary search with an evolutionary algorithm (e.g. fine
Population Reproduction tuning of weights of an evolutionary neural network)
or it may involve simultaneous application of other
algorithms (e.g. hybridizing with simulated annealing
Replacement or Tabu search to improve the efficiency of basic
Offspring evolutionary search).
4. The evaluation of each solution can be handled in
Figure 1. Flow chart of an evolutionary algorithm. parallel and only selection (which requires at least
pair-wise competition) requires some serial processing.
Implicit parallelism is not possible in many global
A primary advantage of evolutionary computation is that
optimization algorithms like simulated annealing and
it is conceptually simple.
Tabu search.
The procedure may be written as the difference equation:
5. Traditional methods of optimization are not robust to
the dynamic changes in the problem of the environment
x(t + 1) = s{v[x(t)]} (1)
and often require a complete restart in order to provide
a solution (e.g. dynamic programming). In contrast,
where x(t) is the population at time t under a representation
evolutionary algorithms can be used to adapt solutions
x, v is a random variation (reproduction) operator, and s is
to changing circumstance.
the selection operator (Fogel, 1999). 6. Perhaps, the greatest advantage of evolutionary algo-
rithms comes from the ability to address problems for
which there are no human experts. Although human
1.1 Advantages of evolutionary algorithms expertise should be used when it is available, it often
proves less than adequate for automating problem-
Following are some of the advantages of using evolutionary solving routines.
algorithms while compared to other global optimization
techniques (Fogel, 1999).
2 GENETIC ALGORITHMS
1. Evolutionary algorithm performance is representation
independent in contrast to other numerical techniques, A typical flowchart of a genetic algorithm is depicted in
which might be applicable for only continuous values Figure 2. One iteration of the algorithm is referred to as
or other constrained sets. a generation. The basic GA is very generic, and there are
2. Evolutionary algorithms offer a framework such that many aspects that can be implemented differently accord-
it is comparably easy to incorporate prior knowl- ing to the problem (e.g. representation of solution (chro-
edge about the problem. Incorporating such infor- mosomes), type of encoding, selection strategy, type of
mation focuses the evolutionary search, yielding a crossover and mutation operators, etc.). In practice, GAs are
more efficient exploration of the state space of pos- implemented by having arrays of bits or characters to rep-
sible solutions. resent the chromosomes. The individuals in the population
Yes
Start Evaluate objective Solution
(Initialize population) function found? End
No
Reproduction
(Crossover/mutation) Selection
Generate new population
Figure 2. Flow chart of basic genetic algorithm iteration.
922 Elements: B – Signal Conditioning
then go through a process of simulated evolution. Simple The chromosome represents the order in which the
bit manipulation operations allow the implementation of salesman will visit the cities. Special care is taken to ensure
crossover, mutation, and other operations. The number of that the strings represent real sequences after crossover and
bits for every gene (parameter) and the decimal range in mutation. Floating-point representation is very useful for
which they decode are usually the same, but nothing pre- numeric optimization (e.g. for encoding the weights of a
cludes the utilization of a different number of bits or range neural network).
for every gene. It should be noted that in many recent applications, more
When compared to other evolutionary algorithms, one of sophisticated genotypes are appearing (e.g. chromosome
the most important GA feature is its focus on fixed-length can be a tree of symbols or a combination of a string and
character strings, although variable-length strings and other a tree, some parts of the chromosome are not allowed to
structures have been used. evolve, etc.).
2.1 Encoding and decoding 3 SCHEMA THEOREM
In a typical application of GAs, the given problem is Theoretical foundations of evolutionary algorithms can be
transformed into a set of genetic characteristics (parameters partially explained by the schema theorem (Holland, 1975),
to be optimized) that will survive in the best possible which relies on the concept of schemata. Schemata are
manner in the environment. For example, optimizing a templates that partially specify a solution (more strictly, a
function solution in the genotype space). If genotypes are strings
built using symbols from an alphabet A, schemata are
min f (x1 , x2 ) = (x1 − 5)2 + (x2 − 2)2 for − 3 ≤ x1 3 ≤;
strings whose symbols belong to A ∪ (∗ ). This extra-symbol
− 8 ≤ x2 8 ≤ (2) (∗ ) must be interpreted as a wildcard, being loci occupied
by it, called undefined. A chromosome is said to match a
The parameters of the search are identified as x1 and x2 , schema if they agree in the defined positions.
which are called the phenotypes in evolutionary algorithms. For example, the string 10011010 matches the schemata
In genetic algorithms, the phenotypes (parameters) are usu- 1∗∗∗∗∗∗∗ and ∗∗ 011∗∗∗ among others but does not match
∗ ∗
ally converted to genotypes by using a coding procedure. 1 11∗∗∗ because they differ in the second gene (the first
Knowing the ranges of x1 and x2 , each variable is to defined gene in the schema).
be represented using a suitable binary string. This repre- A schema can be viewed as a hyperplane in a k-
sentation using binary coding makes the parametric space dimensional space, representing a set of solutions with
independent of the type of variables used. The genotype common properties. Obviously, the numbers of solutions
(chromosome) should in some way contain information that match a schema H depend on the number of defined
about solution, which is also known as encoding. GAs use positions in it. Another related concept is the defining-
a binary string encoding, as shown below. length of a schema, defined as the distance between the
first and the last defined positions in it.
Chromosome A: 110110111110100110110 The GA works by allocating strings to best schemata
Chromosome B: 110111101010100011110 exponentially through successive generations, this being the
selection mechanism mainly responsible for this behavior.
Each bit in the chromosome strings can represent some On the other hand, the crossover operator is responsible
characteristic of the solution. There are several types of for exploring new combinations of the present schemata
encoding (e.g. direct integer or real numbers encoding), in order to get the fittest individuals. Finally, the purpose
which directly depends on the problem. of the mutation operator is to introduce fresh genotypic
Permutation encoding can be used in ordering problems, material in the population.
such as the traveling salesman problem (TSP) or task-
ordering problem. In permutation encoding, every chromo-
some is a string of numbers, which represents numbers in a 4 SELECTION AND REPRODUCTION
sequence. A chromosome using permutation encoding for
a 9-city TSP problem will appear as follows: Individuals for producing offspring are chosen using a
selection strategy after evaluating the fitness value of
Chromosome A: 4 5 3 2 6 1 7 8 9
each individual in the selection pool. Each individual
Chromosome B: 8 5 6 7 2 3 1 4 9 in the selection pool receives a reproduction probability
Evolutionary Computation 923
depending on its own fitness value and the fitness value of size takes values ranging from two to the total number of
all other individuals in the selection pool. This fitness is individuals in the population.
used for the actual selection step afterwards. Some of the
popular selection schemes are discussed below.
4.4 Elitism
When creating a new population by crossover and mutation,
4.1 Roulette-wheel selection there is a big chance that we will lose the best chromosome.
Elitism is the name of the method that first copies the
The simplest selection scheme is the roulette-wheel selec- best chromosome (or a few best chromosomes) to the new
tion, also called stochastic sampling with replacement. This population. The rest is done in the classical way. Elitism
technique is analogous to a roulette wheel with each slice can very rapidly increase performance of GA because it
proportional in size to the fitness. The individuals are prevents losing the best-found solution.
mapped to contiguous segments of a line such that each
individual’s segment is equal in size to its fitness. A ran-
dom number is generated and the individual whose seg- 4.5 Genetic operators
ment spans the random number is selected. The process is
repeated until the desired number of individuals is obtained. Crossover and mutation are two basic operators of GA.
As illustrated in Figure 3, chromosome 1 has the highest Performance of GA depends very much on the genetic
probability for being selected since it has the highest fitness. operators. Type and implementation of operators depends
on encoding and also on the problem. There are many
ways of doing crossover and mutation. In this section, we
4.2 Rank-based fitness assignment will demonstrate some of the popular methods, with some
examples and suggestions as to how to do it for different
encoding schemes.
In rank-based fitness assignment, the population is sorted
according to the objective values. The fitness assigned to
each individual depends only on the position of the objec- 4.6 Crossover
tive values in the individual’s rank. Ranking introduces a
uniform scaling across the population. Crossover selects genes from parent chromosomes and
creates a new offspring. The simplest way to do this is
to choose randomly some crossover point and everything
4.3 Tournament selection before this point is copied from the first parent and then,
everything after a crossover point is copied from the second
In tournament selection, a number of individuals are chosen parent. A single point crossover is illustrated as follows (|
randomly from the population and the best individual is the crossover point):
from this group is selected as the parent. This process is
repeated as often until there are sufficient individuals to Chromosome A: 11011|00100110110
choose. These selected parents produce uniformly random Chromosome B: 11011|11000011110
offspring. The tournament size will often depend on the
Offspring A: 11011|11000011110
problem, population size, and so on. The parameter for
tournament selection is the tournament size. Tournament Offspring B: 11011|00100110110
Chromosome 5
Chromosome 4
Chromosome 3
Chromosome 1
Chromosome 2
Figure 3. Roulette-wheel selection.
924 Elements: B – Signal Conditioning
Parent 1 Parent 2 For many optimization problems, there may be multiple,
equal, or unequal optimal solutions. Sometimes, a simple
GA cannot maintain stable populations at different optima
of such functions. In the case of unequal optimal solutions,
Offspring 1 Offspring 2
the population invariably converges to the global optimum.
Niching helps to maintain subpopulations near global and
local optima. A niche is viewed as an organism’s envi-
Single point crossover ronment and a species as a collection of organisms with
similar features. Niching helps to maintain subpopulations
Parent 1 Parent 2 near global and local optima by introducing a controlled
competition among different solutions near every local opti-
mal region. Niching is achieved by a sharing function,
which creates subdivisions of the environment by degrading
Offspring 1 Offspring 2 an organism’s fitness proportional to the number of other
members in its neighborhood. The amount of sharing con-
tributed by individuals to their neighbor is determined by
Two point crossover their proximity in the decoded parameter space (phenotypic
sharing) based on a distance measure (Goldberg, 1989).
Parent 1 Parent 2
5 GA DEMONSTRATIONS
Offspring 1 Offspring 2 5.1 Rastrigin function
The Rastrigin function is a typical example of nonlinear
Uniform crossover multimodal function. It was first proposed by Rastrigin
(Törn and Zilinskas, 1989) as a two-dimensional function
Figure 4. Types of crossover operators. and has later been generalized. This function is a fairly
difficult problem due to its large search space and its large
As illustrated in Figure 4, there are several crossover number of local minima. Rastrigin’s function is defined as
techniques. In a uniform crossover, bits are randomly
copied from the first or the second parent. Specific
n
F (x) = 10n + xi2 − 10 cos(2πxi ), −5.12 ≤ xi ≤ 5.12
crossover made for a specific problem can improve the i=1
GA performance. (3)
The function has just one global minimum, which occurs
at the origin where the value of the function is 0. At any
4.7 Mutation local minimum other than [0, 0], the value of Rastrigin’s
function is greater than 0. The farther the local minimum
After crossover operation, mutation takes place. Mutation is from the origin, the larger the value of the function is
randomly changes the new offspring. For binary encoding, at that point. Figure 5 illustrates the surface of the function
mutation is performed by changing a few randomly chosen for two input variables.
bits from 1 to 0 or from 0 to 1. Mutation depends on the A real-value representation was used to encode the two
encoding as well as the crossover. For example, when we input variables. The following parameters were used for the
are encoding permutations, mutation could be exchanging GA experiments.
two genes. A simple mutation operation is illustrated as • Mutation: 0.05, crossover: 0.90
follows: • Population size: 20, number of iterations: 50, selection
Chromosome A: 1101111000011110 method: Roulette-wheel selection.
Chromosome B: 1101100100110110 Figure 6 illustrates how the best fitness values were
evolved during the 50 generations. As evident after 30
Offspring A: 1100111000011110
generations, the GA algorithm has succeeded in finding the
Offspring B: 1101101100110110 best optimal solution.
Evolutionary Computation 925
600
590
580
Objective value
570
560
550
540
530
520
5
4 5
2 3
0 1
Var 0
iab −1
le2
−3 −2
−4 1
−5 −5 ia ble
Var
Figure 5. Rastrigin’s function for two variables.
4
Objective function value
0
1 6 11 16 21 26 31 36 41 46
No. of generations
Figure 6. GA learning during the 50 generations.
5.2 Peaks function the given range of x and y values. Using a population
size of 30, the genetic algorithm was run for 25 itera-
Peaks function is a function of two variables, obtained by tions. Each input variable was represented using 8 bit.
translating and scaling Gaussian distributions (Jang, Sun Crossover and mutation rates were set as 0.9 and 0.1 respec-
and Mizutani, 1997). tively.
Figure 8(a), (b), and (c), illustrate the convergence of
F (x, y) = 3(1 − x)2 the solutions on a contour plot of the surface. After 10
x iterations, almost all the solutions were near the optimal
exp(−(x 2 ) − (y + 1)2 ) − 10
5 − x3 − y5 point.
exp(−x 2 − y 2 ) − 1
3
exp(−(x + 1)2 − y 2 )
for − 3 ≤ x ≤ 3 and −3≤y ≤3 (4) 6 EVOLUTION STRATEGIES
The Peak function surface is plotted in Figure 7, and Evolution strategy (ES) was developed by Rechenberg
the task is to find the optimum value (maximum) for (1973) and Schwefel (1977). ES tends to be used for
926 Elements: B – Signal Conditioning
−2
−4
3
−6 2
1
3
2 0
x
1 −1
0
−1 −2
y −2 −3
−3
Figure 7. Surface of Peaks function.
empirical experiments that are difficult to model mathe- 6.1 Mutation in evolution strategies
matically. The system to be optimized is actually con-
structed and ES is used to find the optimal parameter The mutation operator is defined as component-wise addi-
settings. tion of normal distributed random numbers. Both the objec-
Evolution strategies merely concentrate on translating tive parameters and the strategy parameters of the chromo-
the fundamental mechanisms of biological evolution for some are mutated. Objective parameter vector is calculated
technical optimization problems. The parameters to be as follows:
optimized are often represented by a vector of real numbers
(object parameters – op ). Another vector of real numbers op(mut) = op + N0 (sp ) (7)
defines the strategy parameters (sp ), which controls the
mutation of the objective parameters. Both object and where N0 (si ) is the Gaussian distribution of mean value 0
strategic parameters form the data structure for a single and standard deviation si .
individual. Usually, the strategy parameters mutation step size is
A population P of n individuals could be described as done by adapting the standard deviation si . This may be
follows: done (for example) as follows:
P = (c1 , c2 , . . . , cn−1 , cn ) (5) sp(mut) = (s1 ∗ A1 , s2 ∗ A2 , . . . , sn−1 ∗ An−1 , sn ∗ An ) (8)
where Ai is randomly chosen from α or 1/α, depending
where the ith chromosome ci is defined as: ci = on the value of equally distributed random variable E of
(op , sp ) [0,1]
Ai = α if E < 0.5
op = (o1 , o2 , . . . , on−1 , on ) and
1
Ai = if E ≥ 0.5 (9)
sp = (s1 , s2 , . . . , sn−1 , sn ) (6) α
Evolutionary Computation 927
3 3
2 2
1 1
0 0
y
y
−1 −1
−2 −2
−3 −3
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3
(a) x (b) x
0
y
−1
−2
−3
−3 −2 −1 0 1 2 3
(c) x
Figure 8. Convergence of solutions (a) generation 0; (b) after 5 generations; (c) after 20 generations (solution points are marked with *).
α is usually referred to as strategy parameters adaptation 6.3 Controlling the evolution
value.
Let P be the number of parents in generation 1 and let C be
6.2 Crossover (recombination) in evolution the number of children in generation i. There are basically
strategies four different types of evolution strategies: P ,C; P + C;
P /R, C; and P /R + C, as discussed below. They mainly
For two chromosomes c1 = (op(c1) , sp(c1) ) and c2 = differ in how the parents for the next generation are selected
(op(c2) , sp(c2) ), the crossover operator x is defined as and in the usage of crossover operators.
follows:
R(c1 , c2 ) = c = (op , sp ) 6.3.1 P,C strategy
with op(i) = (op(c1),i |op(c2),i ) The P parents produce C children, using mutation. Fitness
values are calculated for each of the C children and the
and sp(i) = (sp(c1),i |sp(c2),i ) (10)
best P children become next-generation parents. The best
By defining op(i) and sp(i) = (x|y), a value is randomly individuals of C children are sorted by their fitness value
assigned for either x or y (50% selection probability for x and the first P individuals are selected to be next-generation
and y). parents (C ≥ P ).
928 Elements: B – Signal Conditioning
6.3.2 P + C strategy 7.1 Evolutionary programming versus genetic
algorithms
The P parents produce C children, using mutation. Fitness
values are calculated for each of the C children and the 1. GA is implemented by having arrays of bits or charac-
best P individuals of both parents and children become ters to represent the chromosomes. In EP, there are no
next-generation parents. Children and parents are sorted by such restrictions for the representation. In most cases,
their fitness value and the first P individuals are selected the representation follows from the problem.
to be next-generation parents. 2. EP typically uses an adaptive mutation operator in
which the severity of mutations is often reduced as the
6.3.3 P/R, C strategy global optimum is approached, while GAs use a pre-
The P parents produce C children, using mutation and fixed mutation operator. Among the schemes to adapt
crossover. Fitness values are calculated for each of the C the mutation step size, the most widely studied is the
children and the best P children become next-generation ‘meta-evolutionary’ technique in which the variance
parents. The best individuals of C children are sorted by of the mutation distribution is subject to mutation by
their fitness value and the first P individuals are selected to a fixed variance mutation operator that evolves along
be next-generation parents (C ≥ P ). Except for the usage of with the solution.
crossover operator, this is exactly the same as P ,C strategy.
7.2 Evolutionary programming versus evolution
6.3.4 P/R + C strategy strategies
The P parents produce C children, using mutation and 1. When implemented to solve real-valued function opti-
recombination. Fitness values are calculated for each of mization problems, both typically operate on the real
the C children and the best P individuals of both parents values themselves and use adaptive reproduction oper-
and children become next-generation parents. Children and ators.
parents are sorted by their fitness value and the first 2. EP typically uses stochastic tournament selection,
P individuals are selected to be next-generation parents. while ES typically uses deterministic selection.
Except for the usage of crossover operator, this is exactly 3. EP does not use crossover operators, while ES (P /R, C
the same as P + C strategy. and P /R + C strategies) crossover. However, the
effectiveness of the crossover operators depends on the
problem at hand.
7 EVOLUTIONARY PROGRAMMING
The book Artificial Intelligence Through Simulated Evolu- 8 GENETIC PROGRAMMING
tion by Fogel, Owens and Walsh (1966) is the landmark
publication for evolutionary programming (EP). In this The genetic programming (GP) technique provides a frame-
book, finite state automata are evolved to predict symbol work for automatically creating a working computer pro-
strings generated from Markov processes and nonstation- gram from a high-level statement of the problem (Koza,
ary time series (AI–FAQ-Genetic). The basic evolutionary 1992).
programming method involves the following steps: Genetic programming achieves this goal of automatic
programming by genetically breeding a population of
1. Choose an initial population (possible solutions at computer programs, using the principles of Darwinian
random). The number of solutions in a population is natural selection and biologically inspired operations. The
highly relevant to the speed of optimization, but no operations include most of the techniques discussed in the
definite answers are available as to how many solutions previous sections. The main difference between GP and GA
are appropriate (other than > 1). is the representation of the solution. GP creates computer
2. New offspring are created by mutation. Each offspring programs in the LISP or scheme computer languages as the
solution is assessed by computing its fitness. Typically, solution. LISP is an acronym for LISt Processor and was
a stochastic tournament is held to determine the N solu- developed in the late 1950s (History of LISP, 2004). Unlike
tions to be retained for the population of solutions. It most languages, LISP is usually used as an interpreted
should be noted that, typically, evolutionary program- language. This means that, unlike compiled languages, an
ming method does not use any crossover as a genetic interpreter can process and respond directly to programs
operator. written in LISP.
Evolutionary Computation 929
The main reason for choosing LISP to implement GP is 9 GENETIC PROGRAMMING BASICS
because of the advantage that the programs and data have
the same structure, which could provide easy means for A parse tree is a structure that develops the interpretation of
manipulation and evaluation. a computer program. Functions are written down as nodes
In GP, the individual population members are not fixed- and their arguments as leaves. A subtree is the part of a
length character strings that encode possible solutions to the tree that is under an inner node of this tree, as illustrated in
problem at hand, they are programs that, when executed, Figure 10. If this tree is cut out from its parent, the inner
are the candidate solutions to the problem. These programs node becomes a root node and the subtree is a valid tree of
are expressed in genetic programming as parse trees rather
its own.
than as lines of code. For example, the simple program
There is a close relationship between these parse trees
‘a + b ∗ c’ would be represented as shown in Figure 9. The
and S-expression; in fact, these trees are just another way
terminal and function sets are also important components
of writing down expressions. While functions will be the
of genetic programming. The terminal and function sets are
nodes of the trees (or the operators in the S-expressions)
the alphabets of the programs to be made. The terminal set
and can have other functions as their arguments, the leaves
consists of the variables and constants of the programs (e.g.
will be formed by terminals, that is, symbols that may not
A, B, and C in Figure 9).
be further expanded. Terminals can be variables, constants,
The most common way of writing down a function
or specific actions that are to be performed. The process
with two arguments is the infix notation. That is, the two
of selecting the functions and terminals that are needed
arguments are connected with the operator symbol between
or are useful for finding a solution to a given problem is
them as follows.
one of the key steps in GP. Evaluation of these structures is
A+B straightforward. Beginning at the root node, the values of all
subexpressions (or subtrees) are computed, descending the
A different method is the prefix notation. Here, the tree down to the leaves. GP procedure could be summarized
operator symbol is written down first, followed by its as follows:
required arguments.
• generate an initial population of random compositions
+AB of the functions and terminals of the problem;
• compute the fitness values of each individual in the
While this may be a bit more difficult or just unusual for population;
human eyes, it opens some advantages for computational • using some selection strategy and suitable reproduction
uses. The computer language LISP uses symbolic expres-
operators, produce offsprings;
sions (or S-expressions) composed in prefix notation. Then,
• iterate the procedure until the required solution is
a simple S-expression could be
found or the termination conditions have been reached
(specified number of generations).
(operator, argument),
The creation of an offspring from the crossover operation
where operator is the name of a function and argument
is accomplished by deleting the crossover fragment of
can be either a constant or a variable or another symbolic
the first parent and then inserting the crossover fragment
expression, as shown below.
of the second parent. The second offspring is produced
(operator, argument(operator, argument) in a symmetric manner. A simple crossover operation is
illustrated in Figure 11. In GP, the crossover operation is
(operator, argument))
+ Subtree
+
5 −
A ∗
3 1
B C
(+5(−31))
Figure 9. A simple tree structure of GP. Figure 10. Illustration of a parse tree and a subtree.
930 Elements: B – Signal Conditioning
/ /
+ × − ×
√ c 6 a √ a + /
× + 4 a a c
4 a × ×
c a b −
Parent 1 Parent 2
a b
/ /
+ × − ×
√ c 6 a √ a
4 a
× +
+ / × ×
4 a a c c a b −
a b
Offspring 1 Offspring 2
Figure 11. Illustration of crossover operator.
/
+ ×
√ c 6 a /
+ ×
×
− c 6 b
4 a
Parent ×
Mutation of terminals
4 a
/
+ ×
√ c 6 a
× c
Mutation of subtree
a 9
Figure 12. Illustration of mutation operator in GP.
implemented by taking randomly selected subtrees in the second kind, an entire subtree can replace another subtree.
individuals and exchanging them. Figure 12 explains the concepts of mutation.
Mutation is another important feature of genetic pro- GP requires data structures that are easy to handle and
gramming. Two types of mutations are commonly used. evaluate and are robust to structural manipulations. These
The simplest type is to replace a function or a termi- are among the reasons why the class of S-expressions was
nal by another function or a terminal respectively. In the chosen to implement GP. The set of functions and terminals
Evolutionary Computation 931
that will be used in a specific problem has to be chosen Bäck, T. (1996) Evolutionary Algorithms in Theory and Practice:
carefully. If the set of functions is not powerful enough, Evolution Strategies, Evolutionary Programming, Genetic algo-
a solution may be very complex or may not be found at rithms, Oxford University Press, New York.
all. Like in any evolutionary computation technique, the Fogel, D.B. (1999) Evolutionary Computation: Toward a New
generation of the first population of individuals is important Philosophy of Machine Intelligence, 2nd edn, IEEE Press,
Piscataway, NJ.
for successful implementation of GP. Some of the other
factors that influence the performance of the algorithm are Fogel, L.J., Owens, A.J. and Walsh, M.J. (1966) Artificial Intelli-
gence Through Simulated Evolution, John Wiley & Sons, New
the size of the population, percentage of individuals that York.
participate in the crossover/mutation, maximum depth for
Goldberg, D.E. (1989) Genetic Algorithms in Search, Optimiza-
the initial individuals and the maximum allowed depth tion, and Machine Learning, Addison-Wesley Publishing Cor-
for the generated offspring, and so on. Some specific poration, Inc, Reading, MA.
advantages of genetic programming are that no analytical History of LISP. (2004) http://www-formal.stanford.edu/jmc/
knowledge is needed and still accurate results could be history/lisp/lisp.html.
obtained. GP approach does scale with the problem size. GP Holland, J. (1975) Adaptation in Natural and Artificial Systems,
does impose restrictions on how the structure of solutions University of Michican Press, Ann Harbor, MI.
should be formulated. Jang, J.S.R., Sun, C.T. and Mizutani, E. (1997) Neuro-Fuzzy and
Soft Computing: A Computational Approach to Learning and
Machine Intelligence, Prentice Hall, USA.
10 SUMMARY Koza, J.R. (1992) Genetic Programming, MIT Press, Cambridge,
MA.
This article presents the biological motivation and fun- Rechenberg, I. (1973) Evolutionsstrategie: Optimierung Tech-
damental aspects of evolutionary algorithms and its con- nischer Systeme nach Prinzipien der biologischen Evolution,
stituents, namely, genetic algorithm, evolution strategies, Fromman-Holzboog, Stuttgart.
evolutionary programming, and genetic programming. Per- Schwefel, H.P. (1977) Numerische Optimierung von Computer-
formance of genetic algorithms is demonstrated using two modellen Mittels der Evolutionsstrategie, Birkhaeuser, Basel.
function optimization problems. Important advantages of Törn, A. and Zilinskas, A. (1989) Global Optimization, Lec-
evolutionary computation as compared to classical opti- ture Notes in Computer Science, Vol. 350, Springer-Verlag,
mization techniques are also discussed. Berlin.
REFERENCES FURTHER READING
AI – FAQ-Genetic. http://www.faqs.org/faqs/ai-faq/genetic/, ac- Michalewicz, Z. (1992) Genetic Algorithms + Data Structures =
cessed on September 10, 2004. Evolution Programs, Springer-Verlag, Berlin.
132: Data Acquisition Systems (DAS) in General
Gerd Wöstenkühler
Hochschule Harz, Wernigerode, Germany
sampled data control systems. In addition, every laboratory
1 Introduction 937 digital multimeter or digital panel meter contains an A/D
2 Basic Data Acquisition System 937 converter.
3 Basic Data Distribution System 939
Further Reading 939
2 BASIC DATA ACQUISITION SYSTEM
Besides A/D and D/A converters, data acquisition and dis-
1 INTRODUCTION tribution systems may employ one or more of the following
circuit functions:
Data acquisition systems (DAS) interface between the real • transducers,
world of physical parameters, which are analog, and the • amplifiers,
artificial world of digital computation and control. With • filters,
current emphasis on digital systems, the interfacing function • nonlinear analog functions,
has become an important one; digital systems are used • analog multiplexers,
widely because complex circuits are low cost, accurate, and • sample-holds.
relatively simple to implement. In addition, there is rapid
growth in the use of microcomputers to perform difficult The interconnection of these components is shown in the
digital control and measurement functions. diagram of the data acquisition portion of a computerized
Computerized feedback control systems are used in many feedback control system in Figure 1.
different industries today in order to achieve greater pro- The input to the system is a physical parameter such
ductivity in our modern industrial societies. Industries that as temperature, pressure, flow, acceleration, and position,
presently employ such automatic systems include steel which are analog quantities. The parameter is first con-
making, food processing, paper production, oil refining, verted into an electrical signal by means of a transducer;
chemical manufacturing, textile production, cement man- once in electrical form, all further processing is done by
ufacturing, and others. electronic circuits.
The devices that perform the interfacing function be- Next, an amplifier boosts the amplitude of the trans-
tween analog and digital worlds are analog-to-digital (A/D) ducer output signal to a useful level for further processing
and digital-to-analog (D/A) converters, which together (see also Article 133, Amplifiers and Filters for DAS,
are known as data converters. Some of the specific Volume 3). Transducer outputs may be microvolt or mil-
applications in which data converters are used include data livolt level signals, which are then amplified to 1 to 10 V
telemetry systems, pulse code modulated communications, levels. Furthermore, the transducer output may be a high-
automatic test systems, computer display systems, video impedance signal, a differential signal with common-mode
signal processing systems, data logging systems, and noise, a current output, a signal superimposed on a high
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
938 Elements: C – Data Acquisition and Processing Systems
Physical Trans- Ampli- Active
parameter ducer fier filter Sample
hold
A/D
Computer data bus
Analog multiplexer
converter
Other
analog
channels
Programmer Control
sequencer
Figure 1. Data acquisition system. (Reproduced from Handbook of Measurement Science, Volume One, Sydenham, 1983 John Wiley
& Sons Ltd.)
D/A Process
Register Actuator
converter parameter
Computer data bus
D/A Process
Register converter Actuator
parameter
Decoder
& control
Control
Figure 2. Data distribution system. (Reproduced from Handbook of Measurement Science, Volume One, Sydenham, 1983 John
Wiley & Sons Ltd.)
voltage, or a combination of these. The amplifier, in order period of time by the multiplexer switch. During this
to convert such signals into a high-level voltage, may be connection time, a sample-hold circuit acquires the signal
one of several specialized types. voltage and then holds its value while an A/D converter
The amplifier is frequently followed by a low-pass converts the value into digital form. The resultant digital
active filter that reduces high-frequency signal components, word goes to a computer data bus or to the input of a
unwanted electrical interference noise, or electronic noise digital circuit.
from the signal (see also Article 133, Amplifiers and Fil- Thus the analog multiplexer, together with the sample-
ters for DAS, Volume 3). The amplifier is sometimes also hold, time shares the A/D converter with a number of
followed by a special nonlinear analog function circuit that analog input channels (see also Article 135, Sample-hold
performs a nonlinear operation on the high-level signal. Circuits, Volume 3). The timing and control of the com-
Such operations include squaring, multiplication, division, plete DAS is done by a digital circuit called a programmer-
rms conversion, log conversion, or linearization. sequencer, which in turn is under the control of the com-
The processed analog signal next goes to an analog puter. In some cases, the computer itself may control the
multiplexer, which switches sequentially between a number entire DAS.
of different analog input channels (see also Article 134, While this is perhaps the most commonly used
Analog Multiplexers, Volume 3). Each input is in turn DAS configuration, there are alternative ones. Instead
connected to the output of the multiplexer for a specified of multiplexing high-level signals, low-level multiplexing
Data Acquisition Systems (DAS) in General 939
is sometimes used with the amplifier following the are activated sequentially by a decoder and control circuit,
multiplexer. In such cases, just one amplifier is required, which is under computer control.
but its gain may have to be changed from one channel to The D/A converter outputs then drive actuators that
the next during multiplexing. Another method is to amplify directly control the various process parameters such as
and convert the signal into digital form at the transducer temperature, pressure, and flow. Thus, the loop is closed on
location and send the digital information in serial form the process and the result is a complete automatic process
to the computer. Here, the digital data must be converted control system under computer control.
to parallel form and then multiplexed onto the computer
data bus.
FURTHER READING
3 BASIC DATA DISTRIBUTION SYSTEM The general principles of DAS are found explained in vari-
ous titles.
The data distribution portion of a feedback control system, Austerlitz, H. (2003) Data Acquisition Techniques Using PCs,
illustrated in Figure 2, is the reverse of the data acquisition Academic Press, San Diego.
system. The computer, based on the inputs of the data acqui- Mihura, B. (2001) LabVIEW for Data Acquisition, Prentice Hall
sition system, must close the loop on a process and control it PTR, Upper Saddle River NJ.
by means of output control functions. These control outputs James, K. (2000) PC Interfacing and Data Acquisition: Tech-
are in digital form and must, therefore, be converted into niques for Measurement, Instrumentation and Control, Newnes,
analog form in order to drive the process. The conversion Oxford.
is accomplished by a series of D/A converters as shown Park, J. and MacKay, S. (2003) Practical Data Acquisition for
(also often called DAC’s). Each D/A converter is coupled Instrumentation and Control Systems, Newnes, Oxford.
to the computer data bus by means of a register, which Taylor, H.R. and Taylor, J.R. (1997) Data Acquisition for Sensor
stores the digital word until the next update. The registers Systems, Chapman & Hall, London.
This article is based in part on a chapter by E.L. Zuch originally published in Handbook of Measurement Science, P.H. Sydenham and
R. Thorn, Wiley, 1982–92.
133: Amplifiers and Filters for DAS
Gerd Wöstenkühler
Hochschule Harz, Wernigerode, Germany
high impedance at both its differential inputs and the gain
1 Operational and Instrumentation Amplifiers 940 is set by a precision resistor located elsewhere in the ampli-
2 Common-mode Rejection Principle and fier circuit. One type of instrumentation amplifier circuit is
Application 940 shown in Figure 2. Notice that no gain-setting resistors are
3 Specialized Amplifiers Used in DAS 941 connected to either of the input terminals. Instrumentation
amplifiers have the following important characteristics:
4 Filters for Use with DAS 941
Further Reading 942 • high impedance differential inputs;
• low input offset voltage drift;
• low input bias currents;
1 OPERATIONAL AND • gain easily set by means of one or two external resistors;
• high common-mode rejection ratio.
INSTRUMENTATION AMPLIFIERS
The front end of a data acquisition system extracts the
desired analog signal from a physical parameter by means 2 COMMON-MODE REJECTION
of a transducer and then amplifies and filters it. An amplifier PRINCIPLE AND APPLICATION
and filter are critical components in this initial signal
processing. The amplifier must perform one or more of the
following functions: boost the signal amplitude, buffer the The common-mode rejection ratio CMRR is an impor-
signal, convert a signal current into a voltage, or extract a tant parameter of differential amplifiers. An ideal differ-
differential signal from common-mode noise. ential input amplifier responds only to the voltage dif-
To accomplish these functions, requires a variety of dif- ference between its input terminals and does not respond
ferent amplifier types. The most popular type of electronic at all to any voltage that is common to both input ter-
amplifier is an operational amplifier (opamp), which is a minals (common-mode voltage). In nonideal amplifiers,
general purpose gain block having differential inputs. The however, the common-mode input signal causes some out-
opamp may be connected in many different closed-loop put response even though small compared to the response
configurations to make it provide a range of functions; a to a differential input signal.
few are shown in Figure 1. The gain and bandwidth of The common-mode rejection ratio is defined as the ratio
the circuits shown depend on the external resistors con- of differential and common-mode responses. The CMRR
nected around the amplifier. An operational amplifier is a of an amplifier is the ratio of differential voltage gain
good choice in general where a single-ended signal is to be to common-mode voltage gain and is generally expressed
amplified, buffered, or converted from current to voltage. in dB:
In the case of differential signal processing, the instru- AD
CMRR = 20 log10 (1)
mentation amplifier is a better choice since it maintains ACM
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Amplifiers and Filters for DAS 941
R R2
R1
I
− E1 −
+ E = −IR + −R2
E2 = E
R1 1
Current to voltage conversion Inverting voltage gain
E1 +
− E2 = 1+
R2
E −
R2 R1 1
E1 + E2 = E1
R1
Unity gain buffer
Noninverting voltage gain
Figure 1. Operational amplifier configurations. (Reproduced from Handbook of Measurement Science, Volume One, Sydenham, 1983
John Wiley & Sons Ltd.)
+VS An isolation amplifier is used to amplify a differential
signal, which is superimposed on a very high common-
RO mode voltage, perhaps of several hundred or even of
I1 I1
several thousand volts. The isolation amplifier has the
− characteristics of an instrumentation amplifier with a very
EO high common-mode input voltage capability. Another spe-
+
cial amplifier, the chopper stabilized amplifier, is used to
∆E accurately amplify microvolt level signals to the required
RO
EO 2RO amplitude. This amplifier employs a special switching sta-
RG G= =
∆E RG bilizer, which gives extremely low input offset voltage
I1 I1
drift. Another useful device, the electrometer amplifier, has
ultralow input bias currents, generally less than one picoam-
pere and is used to convert extremely small signal currents
−VS into a high-level voltage.
Figure 2. Simplified instrumentation amplifier circuit. (Repro-
duced from Handbook of Measurement Science, Volume One, 4 FILTERS FOR USE WITH DAS
Sydenham, 1983 John Wiley & Sons Ltd.)
A low-pass filter frequently follows the signal process-
where AD is differential voltage gain and ACM is common- ing amplifier to reduce signal noise. Low-pass filters are
mode voltage gain. CMRR is a function of frequency used for the following reasons: to reduce man-made elec-
and, therefore, also a function of the impedance balance trical interference noise, to reduce electronic noise, and
between the two amplifier’s input terminals. At even mod- to limit the bandwidth of the analog signal to less than
erate frequencies, CMRR can be significantly degraded by half the sampling frequency in order to eliminate fre-
small unbalances in the source series resistance and shunt quency folding. When used for the last reason, the filter
capacitance. is called a presampling filter or an antialiasing filter (see
also Article 138, Sampling Theory Relevant to DAS,
Volume 3).
Man-made electrical noise is generally periodic, as, for
3 SPECIALIZED AMPLIFIERS USED IN example, in power line interference, and is sometimes
reduced by means of a special filter such as a notch filter.
DAS
Electronic noise, on the other hand, is random noise with
noise power proportional to bandwidth and is present in
There are several other special amplifiers that are useful in transducer resistances, circuit resistances, and in amplifiers
conditioning the input signal in a data acquisition system. themselves. It is reduced by limiting the bandwidth of the
942 Elements: C – Data Acquisition and Processing Systems
system to the minimum required to pass the desired signal
components (see also Article 181, Amplitude Modulated
Signals: The Lock-in Amplifier, Volume 3; Article 180, 0
3 dB
Bandwidth Reduction of Baseband DC Signals, Vol-
Attenuation (dB)
5
ume 3; Article 179, Input Connections; Grounding and 4-Pole Bessel
Shielding, Volume 3; Article 176, Signals and Signal-to-
10
noise Ratio, Volume 3; Article 178, Noise Matching and 4-Pole
Butterworth
Preamplifier Selection, Volume 3; and Article 121, Sig-
15
nals in the Presence of Noise, Volume 2). 4-Pole
Chebyshev
Grounding and shielding of amplifiers is an important
20
topic in amplifier design – see Article 177, Grounding
and Shielding, Volume 3. 25
No filter does a perfect job of eliminating noise or 0.1 0.2 0.3 0.4 0.50.6 0.8 1 2 3 4 5
other undesirable frequency components, and therefore Normalized frequency
the choice of a filter is always a compromise. Ideal fil-
Figure 3. Some practical low-pass filter characteristics. (Repro-
ters, frequently used as analysis examples, have flat pass- duced from Handbook of Measurement Science, Volume One,
band response with infinite attenuation at the cutoff fre- Sydenham, 1983 John Wiley & Sons Ltd.)
quency, but are mathematical filters only and not physically
realizable.
of operation considerably. Fortunately, the user is buffered
In practice, the systems engineer has a choice of cut-
from the mathematics and usually can work from data
off frequency and attenuation rate. The attenuation rate
sheets to select hardware and software settings for the IC
and resultant phase response depend on the particular filter
modules provided.
characteristic and the number of poles in the filter func-
tion. Some of the more popular filter characteristics used
include the Butterworth, Chebyshev, Bessel, and elliptic FURTHER READING
alternatives. In making this choice, the effect of overshoot
and nonuniform phase delay must be carefully considered. Electronic operational amplifiers are covered as a single topic in
Figure 3 illustrates some practical low-pass filter response many titles; the following may be useful.
characteristics (see also Article 125, Outline of Purpose
of Analog Data Filters, Volume 2). Coughlin, R.F. and Driscoll, F.F. (2001) Operational Amplifiers
Passive RLC filters are seldom used in signal processing and Linear Integrated Circuits, Prentice Hall College Division,
Upper Saddle River, NJ.
applications today chiefly due to the undesirable charac-
teristics of inductors. Active filters are generally used now Franco, S. (2001) Design with Operational Amplifiers and Analog
Integrated Circuits, McGraw-Hill, Dubuque, Iowa.
since they permit the filter characteristics to be accurately
set by precision and stable resistors and capacitors. Induc- Jerald, G., Graeme, J.G. and Graeme, J.E. (1997) Optimizing Op
Amp Performance, McGraw-Hill, New York.
tors, with their undesirable saturation and temperature drift
characteristics, are thereby eliminated. Also, because active McGraw-Hill Science/Engineering/Math.
filters use operational amplifiers, the problems of insertion Mancini, R. (2003) Op Amps for Everyone, Newnes, Boston.
loss and output loading are also eliminated. Morrison, R. (1970) DC Amplifiers in Instrumentation, Wiley-
Filters are often implemented in digital form. These are Interscience, New York.
based on the analog design concepts but the digitization, Stanley, W.D. (2002) Operational Amplifiers with Linear Inte-
which introduced sampling issues, increases the complexity grated Circuits, Prentice Hall, Upper Saddle River, NJ.
This article is based in part on a chapter by E.L. Zuch originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
134: Analog Multiplexers
Gerd Wöstenkühler
Hochschule Harz, Wernigerode, Germany
CMOS driver controls the gates of the parallel-connected
1 Analog Multiplexer Operation 943 P-channel and N-channel MOSFETs. Both switches turn
2 Analog Multiplexer Characteristics 943 on together with the parallel connection giving relatively
uniform on-resistance over the required analog input volt-
3 Analog Multiplexer Applications 944
age range. The resulting on-resistance may vary from about
Further Reading 945 50 to 2 k depending on the multiplexer; this resistance
increases with temperature.
1 ANALOG MULTIPLEXER OPERATION
2 ANALOG MULTIPLEXER
Analog multiplexers are the circuits that time-share an CHARACTERISTICS
analog-to-digital (A/D) converter among a number of dif-
ferent analog channels. Since the A/D converter in many Because of the series resistance, it is common practice to
cases is the most expensive component in a data acquisition operate an analog multiplexer into a very high load resis-
system, multiplexing analog inputs to the A/D converter tance such as the input of a unity gain buffer amplifier
is an economical approach. Usually, the analog multi- shown in Figure 1. The load impedance must be large
plexer operates in a sample-hold circuit (see also Arti- compared with the switch on-resistance and any series
cle 135, Sample-hold Circuits, Volume 3), which holds source resistance in order to maintain high transfer accu-
the required analog voltage long enough for A/D conver- racy. Transfer error is the input to output error of the
sion. multiplexer with the source and load connected; error is
As shown in Figure 1, an analog multiplexer consists expressed as a percent of input voltage.
of an array of parallel electronic switches connected to a Transfer errors of 0.1 to 0.01% or less are required in
common output line. Only one switch is turned on at a time. most data acquisition systems. This is readily achieved
Popular switch configurations include 4, 8, and 16 channels by using operational amplifier buffers with typical input
that are connected in single single-ended or dual differential impedances from 108 to 1012 . Many sample-hold circuits
configurations. also have very high input impedances.
The multiplexer also contains a decoder-driver circuit Another important characteristic of analog multiplex-
that decodes a binary input word and turns on the appropri- ers is break-before-make switching. There is a small time
ate switch. This circuit interfaces with standard TTL inputs delay between disconnection from the previous channel
and drives the multiplexer switches with the proper con- and connection to the next channel, which assures that
trol voltages. For the 8-channel analog multiplexer shown, two adjacent input channels are never instantaneously con-
a one-of-eight decoder circuit is used. nected together.
Most analog multiplexers today employ the CMOS Settling time is another important specification for analog
switch circuits based on the one shown in Figure 2. A multiplexers; it is the same definition given for amplifiers
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
944 Elements: C – Data Acquisition and Processing Systems
Rs
1
−
Vs 2
Input channel no.
3 +
4 Amplifier
5 Rin
6
7
8
Enable A1 A2 A3 En. On-chan
X X X 0 None
0 0 0 1 1
0 0 1 1 2
Decoder 0 1 0 1 3
driver 0 1 1 1 4
1 0 0 1 5
1 0 1 1 6
A1 A2 A3 1 1 0 1 7
Channel address 1 1 1 1 8
Figure 1. Analog multiplexer circuit. (Reproduced from Handbook of Measurement Science, Volume One, Sydenham, 1983 John
Wiley & Sons Ltd.)
In C3
I3
To N P
decoder Rs S1 R on
Out Vs C1 I1 I2 C2 RL
Figure 2. CMOS analog switch circuit. (Reproduced from Hand-
book of Measurement Science, Volume One, Sydenham, 1983 Figure 3. Equivalent circuit of analog multiplexer switch.
John Wiley & Sons Ltd.) (Reproduced from Handbook of Measurement Science, Volume
One, Sydenham, 1983 John Wiley & Sons Ltd.)
except that it is measured from the time the channel is
switched on. Throughput rate is the highest rate at which
3 ANALOG MULTIPLEXER
a multiplexer can switch from channel to channel with the
output settling to its specified accuracy. Crosstalk is the APPLICATIONS
ratio of output voltage to input voltage with all channels
connected in parallel and off; it is generally expressed as Analog multiplexers are employed in two basic types of
an input to output attenuation ratio in decibels. operation; low level and high level. In high-level multiplex-
As shown in the representative equivalent circuit of ing, the most popular type, the analog signal is amplified
Figure 3, analog multiplexer switches have a number of to the 1 to 10 V range ahead of the multiplexer. This has
leakage currents and capacitances associated with their the advantage of reducing the effects of noise on the signal
operation. These parameters are specified on data sheets during the remaining analog processing. In low-level multi-
and must be considered in the operation of the devices. plexing, the signal is amplified after multiplexing; therefore,
Leakage currents, generally in picoamperes at room tem- great care must be exercised in handling the low-level sig-
perature, become troublesome only at high temperatures. nal up to the multiplexer. Low-level multiplexers generally
Capacitances affect crosstalk and settling time of the use two-wire differential switches in order to minimize
multiplexer. noise pickup. Reed relays, because of essentially zero series
Analog Multiplexers 945
Instrumentation mode voltage and is then further processed up to A/D
amplifier conversion.
In order to realize large numbers of multiplexed chan-
Vs C Out
nels, it is possible to connect analog multiplexers in parallel
using the enable input to control each device. This is
called single-level multiplexing. The output of several mul-
RG
VCommon-mode tiplexers can also be connected to the inputs of another to
expand the number of channels; this method is double-level
multiplexing.
Figure 4. Flying-capacitor multiplexer switch. (Reproduced from
Handbook of Measurement Science, Volume One, Sydenham,
FURTHER READING
1983 John Wiley & Sons Ltd.)
The general principles of DAS are found explained in various
titles. Multiplexing is usually covered in these books.
resistance and absence of switching spikes are frequently
employed in low-level multiplexing systems. They are also Austerlitz, H. (2003) Data Acquisition Techniques Using PCs,
useful for high common-mode voltages. Academic Press, San Diego.
A useful specialized analog multiplexer is the flying- James, K. (2000) PC Interfacing and Data Acquisition: Tech-
capacitor type. This circuit, shown as a single channel niques for Measurement, Instrumentation and Control, Newnes,
in Figure 4 has differential inputs and is particularly use- Oxford.
ful with high common-mode voltages. The capacitor con- Mihura, B. (2001) LabVIEW for Data Acquisition, Prentice Hall
nects first to the differential analog input, charging up PTR, Upper Saddle River, NJ.
to the input voltage, and is then switched to the dif- Park, J. and MacKay, S. (2003) Practical Data Acquisition for
ferential output, which goes to a high input impedance Instrumentation and Control Systems, Newnes, Oxford.
instrumentation amplifier. The differential signal is, there- Taylor, H.R. and Taylor, J.R. (1997) Data Acquisition for Sensor
fore, transferred to the amplifier input without the common Systems, Chapman & Hall, London.
This article is based in part on a chapter by E.L. Zuch originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
135: Sample-hold Circuits
Gerd Wöstenkühler
Hochschule Harz, Wernigerode, Germany
sample holds, the metal oxide semiconductor, MOS type
1 Role and Operation of Sample-hold Circuit capacitor is frequently used.
in DAS 946 A2 is the output amplifier that buffers the voltage on
2 Sample-hold Unit As a Data Recovery Filter 946 the hold capacitor. It must, therefore, have extremely low
input bias current, and for this reason a FET input amplifier
3 Specialized Sample-hold Circuits 947
is required.
4 Sample-hold Characteristics 947 There are two modes of operation for a sample hold:
Further Reading 948 sample mode or tracking mode, when the switch is closed;
and hold mode, when the switch is open. Sample holds are
usually operated in one of two basic ways. The device can
continuously track the input signal and be switched into the
1 ROLE AND OPERATION OF hold mode only at certain specified times, spending most of
SAMPLE-HOLD CIRCUIT IN DAS the time in tracking mode. This is the case for a sample hold
being employed as a deglitcher at the output of a digital-
to-analog (D/A) converter, for example, see Article 141,
Sample-hold circuits are the devices that store analog infor-
Digital-to-Analog (D/A) Converters, Volume 3.
mation and reduce the aperture time of an analog-to-digital
Alternatively, the device can stay in the hold mode most
(A/D) converter (see also Article 138, Sampling The-
of the time and go to the sample mode just to acquire a
ory Relevant to DAS, Volume 3; Article 141, Digital-to- new input signal level. This is the case for a sample hold
Analog (D/A) Converters, Volume 3). used in a data acquisition system following the multiplexer
A sample hold is simply a voltage-memory device in (see also Article 132, Data Acquisition Systems (DAS)
which an input voltage is acquired and then stored on a high in General, Volume 3).
quality capacitor. A popular circuit is shown in Figure 1.
A1 is an input buffer amplifier with a high input
impedance so that the source, which may be an analog 2 SAMPLE-HOLD UNIT AS A DATA
multiplexer (see also Article 134, Analog Multiplexers, RECOVERY FILTER
Volume 3), is not loaded. The output of A1 must be capable
of driving the hold capacitor with stability and enough drive A common application for sample-hold circuits is in data
current to charge it rapidly. S is an electronic switch, gen- recovery, or signal reconstruction, filters. The problem is to
erally field effect transistor, FET, which is rapidly switched reconstruct a train of analog samples into the original signal;
on or off by a driver circuit that interfaces with transistor when used as a recovery filter, the sample hold is known
transistor logic (TTL) inputs. as a zero-order hold. It is a useful filter because it fills in
C is a capacitor with low leakage and low dielectric the space between samples, providing data smoothing.
absorption characteristics; it is a polystyrene, polycarbon- As with other filter circuits, the gain and phase compo-
ate, polypropylene, or Teflon type. In the case of hybrid nents of the transfer function are of interest. By an analysis
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Sample-hold Circuits 947
A1 S A2 sample-hold reconstruction filter is frequently followed by
+1 +1 another conventional low-pass filter.
Input C Output
Switch 3 SPECIALIZED SAMPLE-HOLD
driver
CIRCUITS
In addition to the basic circuit of Figure 1, several other
Sample
control
sample-hold circuit configurations are frequently used.
Figure 3 shows two such circuits that are closed-loop cir-
Figure 1. Popular sample-hold circuit. (Reproduced from Hand- cuits as contrasted with the open-loop circuit of Figure 1.
book of Measurement Science, Volume One, Sydenham, 1983 Figure 3(a) uses an operational integrator and another
John Wiley & Sons Ltd.) amplifier to make a fast, accurate inverting sample hold.
A buffer amplifier is sometimes added in front of this cir-
based on the impulse response of a sample hold and use of cuit to give high input impedance. Figure 3(b) shows a high
the Laplace transform, the transfer function is found to be input impedance, noninverting sample-hold circuit.
The circuit in Figure 1, although generally not as accurate
as those in Figure 3, can be used with a diode-bridge switch
π ×f
sin π×f to realize ultrafast acquisition sample holds.
1 fs −j
fs
G0 (f ) = e (1)
fs π ×f
fs 4 SAMPLE-HOLD CHARACTERISTICS
where fs is the sampling frequency. This function contains A number of parameters are important in characterizing
the familiar (sin x)/x term plus a phase term, both of which sample-hold performance. Probably the most important
are plotted in Figure 2. of these is the acquisition time. The definition is simi-
The sample hold is, therefore, a low-pass filter with a lar to the definition of settling time for an amplifier (see
cutoff frequency slightly less than fs /2 and a linear phase also Article 133, Amplifiers and Filters for DAS, Vol-
that results in a constant delay time of T /2, where T is the ume 3). It is the time required, after the sample command
time between samples. Notice that the gain function also is given, for the hold capacitor to charge to a full-scale
has significant response lobes beyond fs . For this reason, a
R C
1
fs R S
0.636 +1 −
fs
Input Output
Gain (Go(f ))
+
1
2fs 0.212
0.127
fs
fs (a)
0 R
fs /2 fs 3fs/2 2fs 5fs/2 3fs
0 A2
−
Phase angle (rod)
S
A1 +1 Output
−π
Input +
C
−2π
(b)
−3π
Figure 3. Accurate closed-loop sample-hold circuits: (a) inver-
Figure 2. Gain and phase components of zero-order hold transfer ting and (b) noninverting. (Reproduced from Handbook of Mea-
function. (Reproduced from Handbook of Measurement Science, surement Science, Volume One, Sydenham, 1983 John Wiley
Volume One, Sydenham, 1983 John Wiley & Sons Ltd.) & Sons Ltd.)
948 Elements: C – Data Acquisition and Processing Systems
voltage change and remain within a specified error band Aperture delay is the time elapsed from the hold com-
around final value. mand to when the switch actually opens; it is generally
Several hold-mode specifications are also important. much less than a microsecond.
Hold-mode droop is the output voltage change per unit Aperture uncertainty or aperture jitter is the time varia-
time when the sample switch is open. This droop is caused tion, from sample-to-sample, of the aperture delay. It is the
by the leakage currents of the capacitor and switch, and limit on how precise is the point in time of opening the
the output amplifier bias current. Hold-mode feedthrough switch. Aperture uncertainty is the time used to determine
is the percentage of input signal transferred to the out- the aperture error due to rate of change of the input sig-
put when the sample switch is open. It is measured nal (see also Article 138, Sampling Theory Relevant to
with a sinusoidal input signal and caused by capaci- DAS, Volume 3).
tive coupling. Several of the above specifications are illustrated in the
The most critical phase of sample-hold operation is diagram of Figure 4.
the transition from the sample mode to the hold mode. Sample-hold circuits are simple in concept, but generally
Several important parameters characterize this transition. difficult to fully understand and apply. Their operation is
full of subtleties, and they must, therefore, be carefully
Sample-to-hold offset or step error is the change in output
selected and then tested in a given application. Their sim-
voltage from the sample mode to the hold mode, with
plicity is a strong reason they are not always supplanted by
a constant input voltage. It is caused by the switch still
a digital storage device alternative.
transferring some charge onto the hold capacitor as it
turns off.
FURTHER READING
The general principles of DAS are found explained in various
titles. As the sample and hold device is usually needed in a DAS,
Sample-to-hold Hold-mode
droop
the general titles usually cover them.
offset
Input Austerlitz, H. (2003) Data Acquisition Techniques Using PCs,
signal
Academic Press, San Diego.
Output
signal James, K. (2000) PC Interfacing and Data Acquisition: Tech-
Aperture niques for Measurement, Instrumentation and Control, Newnes,
delay
Oxford.
Sample Mihura, B. (2001) LabVIEW for Data Acquisition, Prentice Hall
control
Track Hold PTR, Upper Saddle River, NJ.
Park, J. and MacKay, S. (2003) Practical Data Acquisition for
Figure 4. Some sample-hold characteristics. (Reproduced from Instrumentation and Control Systems, Newnes, Oxford.
Handbook of Measurement Science, Volume One, Sydenham, Taylor, H.R. and Taylor, J.R. (1997) Data Acquisition for Sensor
1983 John Wiley & Sons Ltd.) Systems, Chapman & Hall, London.
This article is based in part on a chapter by E.L. Zuch originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
136: Quantizing Theory Relevant to DAS
Gerd Wöstenkühler
Hochschule Harz, Wernigerode, Germany
staircase function is the best approximation that can be
1 Quantizer Transfer Function 949 made to a straight line drawn through the origin and full-
2 Quantizer Resolution and Error 949 scale point; notice that the line passes through all of the
code word center points.
Further Reading 950
2 QUANTIZER RESOLUTION AND
1 QUANTIZER TRANSFER FUNCTION ERROR
At any part of the input range of the quantizer, there is a
The nonlinear transfer function shown in Figure 1 is that small range of analog values within which the same output
of an ideal quantizer with eight output states; with output code word is produced. This small range is the voltage
code words assigned, it is also that of a 3-bit A/D converter. difference between any two adjacent decision points and
The eight output states are assigned the sequence of binary is known as the analog quantization size, or quantum, Q.
numbers from 000 through to 111. The analog input range In Figure 1, the quantum is 1.25 V and is found in general
for this quantizer is 0 to +10 V. by dividing the full-scale analog range by the number of
There are several important points concerning the transfer output states. Thus,
function of Figure 1.
First, the resolution of the quantizer is defined as the FSR
number of output states expressed in bits; in this case it is Q= (1)
2n
a simple 3-bit quantizer.
The number of output states for a binary coded quantizer where FSR is the full-scale range, or 10 V in this case.
is 2n , where n is the number of bits. Thus, an 8-bit quantizer Q is the smallest analog difference that can be resolved,
has 256 output states and a 12-bit quantizer has 4096 or distinguished, by the quantizer. In the case of a 12-bit
output states. quantizer, the quantum is much smaller and is found to be
As shown in the diagram, there are 2n − 1 analog deci- FSR 10 V
sion points (or threshold levels) in the transfer function. Q= = = 2.44 mV (2)
2n 4096
These points are at voltages of +0.625, +1.875, +3.125,
+4.375, +5.625, +6.875, and +8.125 V. The decision If the quantizer input is moved through its entire range
points must be precisely set in a quantizer in order to divide of analog values and the difference between output and
the analog voltage range into the correct quantizer values. input is taken, a sawtooth error function results, as shown
The voltages 0.00, +1.25, +2.50, +3.75, +5.00, +6.25, in Figure 1. This function is called the quantizing error and
+7.50, and +8.75 V are the center points of each output it is the irreducible error that results from the quantizing
code word. The analog decision point voltages are precisely process. It can be reduced only by increasing the number
halfway between the code word center points. The quantizer of output states (or the resolution) of the quantizer, thereby
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
950 Elements: C – Data Acquisition and Processing Systems
For a given analog input value of the quantizer, the output
8 111 error will vary anywhere from 0 to ±Q/2; the error is zero
only at analog values corresponding to the code center
7 110 points. This error is also frequently called quantization
6
uncertainty or quantization noise.
101
The quantizer output can be thought of as the analog
Output states
Output code
Q
5 100 input with quantization noise added to it. The noise has
a peak-to-peak value of Q but, as with other types of
4 011
noise, the average value is zero. Its rms value, however, is
useful in analysis and can be computed from the triangular
3 010 √
waveshape to be Q/(2 × 3).
2 001
1 000
FURTHER READING
+1.25
+2.50
+3.75
+5.00
+6.25
+7.50
+8.75
+10.00
Input voltage (V) The general principles of DAS are explained in various titles.
Quantization is usually covered at different depths of explana-
+Q/ 2
Quantizer 0
tion.
Q
error −Q/ 2 Austerlitz, H. (2002) Data Acquisition Techniques Using PCs,
Academic Press.
Mihura, B. (2001) LabVIEW for Data Acquisition, Prentice Hall
Figure 1. Transfer function and error of ideal 3-bit quantizer. PTR.
(Source: reproduced by permission of Sydenham, P.H. (1983) James, K. (2000) PC Interfacing and Data Acquisition: Tech-
Handbook of Measurement Science, Volume 1 Theoretical Fun- niques for Measurement, Instrumentation and Control, Newnes.
damentals. John Wiley & Sons Ltd., Chichester.) Park, J. and MacKay, S. (2003) Practical Data Acquisition for
Instrumentation and Control Systems, Newnes.
making the quantization finer and the system increasingly Taylor, H.R. and Taylor, J.R. (1997) Data Acquisition for Sensor
sophisticated. Systems, Kluwer Academic Publishers.
This article is based in part on a chapter by E.L. Zuch originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
137: Coding for Data Converters
Gerd Wöstenkühler
Hochschule Harz, Wernigerode, Germany
(1 × 0.5) + (1 × 0.25) + (0 × 0.125) + (1 × 0.0625)
1 Natural Binary Code 951 + (0 × 0.03125) + (1 × 0.015625)
2 Other Binary Codes 952
3 Popular Bipolar Codes Used in Data = 0.828125 or 82.8125% (2)
Converters 953 of full scale for the converter. If full scale is +10 V, then
4 Binary Coded Decimal (BCD) Codes 953 the code word represents +8.28125 V. The natural binary
Further Reading 954 code belongs to a class of codes known as positive-weighted
codes since each coefficient has a specific weight none of
which is negative.
1 NATURAL BINARY CODE The leftmost bit has the most weight, 0.5 of full scale,
and is called the most significant bit (MSB); the rightmost
Analog-to-digital (A/D) and digital-to-analog (D/A) con- bit has the least weight, 2−n of full scale, and is, therefore,
verters interface with digital systems by means of an appro- called the least significant bit (LSB). The bits in a code
priate digital code. While there are many possible codes to word are numbered left to right from 1 to n.
select, a few standard ones are almost exclusively used with The LSB has the same analog equivalent value as Q
data converters. (see Article 136, Quantizing Theory Relevant to DAS,
The most popular code is natural binary, or straight Volume 3), namely,
binary, which is used in its fractional form to represent FSR
a number LSB(analog value) = (3)
2n
N = a1 2−1 + a2 2−2 + a3 2−3 + · · · + an 2−n (1) Table 1 is a useful summary of the resolution, number
of states, LSB weights, and dynamic range (DR) for data
where each coefficient a assumes a value of zero or one. N converters from 1 to 20 bits resolution.
has a value between zero and one (see also Article 33, The dynamic range of a data converter in decibels (dB)
Coding Theory and its Application to Measurement, is found as follows:
Volume 1).
A binary fraction is normally written as 0.110101, DR = 20 × log 2n = 20 × n × log 2 ≈ 20 × n × 0.301
but with data converter codes, the decimal point is = 6.02 × n (4)
omitted and the code word is written as 110101.
This code word represents a fraction of the full-scale where DR is dynamic range, n is the number of bits, and
value of the converter and has no other numerical 2n is the number of states of the converter. Since 6.02 dB
significance. corresponds to a factor of 2, it is simply necessary to mul-
The binary code word 110101, therefore, represents the tiply the resolution of a converter in bits by 6.02. A 12-bit
decimal fraction converter, for example, has a dynamic range of 72.24 dB.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
952 Elements: C – Data Acquisition and Processing Systems
Table 1. Resolution, number of states, LSB weight, and dynamic −5 V and 0 to −10 V. The standard bipolar voltage ranges
range for data converters. are ±2.5 V, ±5 V, and ±10 V. Many converters offered are
Resolution Number of LSB weight Dynamic pin-programmable between these various ranges.
bits n states 2n 2−n range Table 2 shows straight binary and complementary binary
(dB) codes for a unipolar 8-bit converter with a 0 to +10 V ana-
0 1 1 0.0 log FS range. The maximum analog value of the converter
1 2 0.5 6.0 is +9.961 V, or one LSB less than +10 V. Note that the
2 4 0.25 12.0 LSB size is 0.039 V as shown near the bottom of the table.
3 8 0.125 18.1 The complementary binary coding used in some converters
4 16 0.0625 24.1
5 32 0.03125 30.1 is simply the logic complement of straight binary.
6 64 0.015625 36.1 When A/D and D/A converters are used in bipolar opera-
7 128 0.0078125 42.1 tion, the analog range is offset by half scale, or by the MSB
8 256 0.00390625 48.2 value. The result is an analog shift of the converter transfer
9 512 0.001953125 54.2 function as shown in Figure 1. Notice for this 3-bit A/D
10 1024 0.0009765625 60.2
11 2048 0.00048828125 66.2 converter transfer function that the code 000 corresponds
12 4096 0.000244140625 72.2 with −5 V; 100 with 0 V; and 111 with +3.75 V. Since the
13 8192 0.0001220703125 78.3
14 16 384 0.00006103515625 84.3 Table 2. Binary coding for 8-bit unipolar converters.
15 32 768 0.000030517578125 90.3
16 65 536 0.0000152587890625 96.3 Fraction of FS +10 V FS Straight Complementary
17 131 072 0.00000762939453125 102.3 binary binary
18 262 144 0.000003814697265625 108.4
+FS − 1 LSB +9.961 1111 1111 0000 0000
19 524 288 0.0000019073486328125 114.4
+3/4 FS +7.500 1100 0000 0011 1111
20 1 048 576 0.00000095367431640625 120.4
+1/2 FS +5.000 1000 0000 0111 1111
Source: Reproduced from Handbook of Measurement Science, Volume +1/4 FS +2.500 0100 0000 1011 1111
One, Sydenham, 1983 John Wiley & Sons Ltd. +1/8 FS +1.250 0010 0000 1101 1111
+1 LSB +0.039 0000 0001 1111 1110
An important point to notice is that the maximum value 0 0.000 0000 0000 1111 1111
of the digital code, namely all 1’s, does not correspond with Source: Reproduced from Handbook of Measurement Science, Volume
analog full scale, but rather with one LSB less than full One, Sydenham, 1983 John Wiley & Sons Ltd.
scale, or FS (1–2−n ). Therefore, a 12-bit converter with a
0 to +10 V analog range has a maximum code of 1111 1111
1111 and a maximum analog value of +10 V (1–2−12 ) ≈
+9.99756 V. In other words, the maximum analog value of 111
Output code
the converter, corresponding to all 1’s in the code, never
quite reaches the point defined as analog full scale. 110
101
2 OTHER BINARY CODES
Several other binary codes are used with A/D and D/A 100
converters in addition to straight binary. These codes are
offset binary, two’s complement, binary coded decimal 011
(BCD), and their complemented versions. Each code has a Q
specific advantage in certain applications. BCD coding, for 010
example, is used where digital displays must be interfaced
such as in digital panel meters and digital multimeters. 001
Two’s complement coding is used for computer arithmetic 000
logic operations, and offset binary coding is used with
bipolar analog measures. −5.00 −3.75 −2.50 −1.25 0 +1.25 +2.50 +3.75 +5.00
Not only the digital codes are standardized with data Input voltage (V)
converters but also the analog voltage ranges. Most con- Figure 1. Transfer function for bipolar 3-bit A/D converter.
verters use unipolar voltage ranges of 0 to +5 V and 0 to (Reproduced from Handbook of Measurement Science, Volume
+10 V although some devices use the negative ranges 0 to One, Sydenham, 1983 John Wiley & Sons Ltd.)
Coding for Data Converters 953
Table 3. Popular bipolar codes used with data converters.
Fraction of FS ±5 V FS Offset binary Complementary Two’s complement Sign-magnitude
offset binary binary
+FS − 1 LSB +4.961 1111 1111 0000 0000 0111 1111 1111 1111
+3/4 FS +3.750 1110 0000 0001 1111 0110 0000 1110 0000
+1/2 FS +2.500 1100 0000 0011 1111 0100 0000 1100 0000
+1/4 FS +1.250 1010 0000 0101 1111 0010 0000 1010 0000
0 0.000 1000 0000 0111 1111 0000 0000 1000 0000a
−1/4 FS −1.250 0110 0000 1001 1111 1110 0000 0010 0000
−1/2 FS −2.500 0100 0000 1011 1111 1100 0000 0100 0000
−3/4 FS −3.750 0010 0000 1101 1111 1010 0000 0110 0000
−FS + 1 LSB −4.961 0000 0001 1111 1110 1000 0001 0111 1111
−FS −5.000 0000 0000 1111 1111 1000 0000 –
Source: Reproduced from Handbook of Measurement Science, Volume One, Sydenham, 1983 John Wiley & Sons Ltd.
a
Sign magnitude binary has two code words for zero as shown here:
0+ 1000 0000
0− 0000 0000.
output coding is the same as before the analog shift, it is Table 4. BCD and complementary BCD coding.
now appropriately called offset binary coding. Fraction of FS +10 V FS Binary coded Complementary
decimal BCD
3 POPULAR BIPOLAR CODES USED IN +FS − 1 LSB +9.99 1001 1001 1001 0110 0110 0110
+3/4 FS +7.50 0111 0101 0000 1000 1010 1111
DATA CONVERTERS +1/2 FS +5.00 0101 0000 0000 1010 1111 1111
+1/4 FS +2.50 0010 0101 0000 1101 1010 1111
Table 3 shows the offset binary code together with com- +1/8 FS +1.25 0001 0010 0101 1110 1101 1010
plementary offset binary, two’s complement, and sign- +1 LSB +0.01 0000 0000 0001 1111 1111 1110
0 0.00 0000 0000 0000 1111 1111 1111
magnitude binary codes. These are the most popular codes
employed in bipolar data converters. Source: Reproduced from Handbook of Measurement Science, Volume
One, Sydenham, 1983 John Wiley & Sons Ltd.
The two’s complement code has the characteristic that
the sum of the positive and negative codes for the same
analog magnitude always produces all zeros and a carry. decimal display applications. Here 4 bits are used to repre-
This characteristic makes the two’s complement code useful sent each decimal digit. BCD is a positive-weighted code
in arithmetic computations. Notice that the only difference but is relatively inefficient since in each group of 4 bits,
between two’s complement and offset binary is the comple- only 10 out of a possible 16 states are utilized (see also
menting of the MSB. In bipolar coding, the MSB becomes Article 33, Coding Theory and its Application to Mea-
the sign bit. surement, Volume 1).
The sign-magnitude binary code, infrequently used, has The LSB analog value (or quantum, Q) for BCD is
identical code words for equal magnitude analog values FSR
except that the sign bit is different. As shown in Table 3, LSB(analog value) = Q = (5)
10d
this code has two possible code words for zero: 1000 0000
or 0000 0000. The two are usually distinguished as 0+ and where FSR is the full-scale range and d is the number of
0−, respectively. Because of this characteristic, the code decimal digits. For example, if there are three digits and
has maximum analog values of ±(FS − 1 LSB) and reaches the FSR is 10 V, the LSB value is
neither analog +FS nor −FS. 10V
LSB(analog value) = = 0.01V = 10 mV (6)
103
4 BINARY CODED DECIMAL (BCD) BCD coding is frequently used with an additional over-
CODES range bit, which has a weight equal to full scale and
produces a 100% increase in range for the A/D converter.
Table 4 shows BCD coding and complementary BCD cod- Thus, for a converter with a decimal full scale of 999, an
ing for a three-decimal digit data converter. These are the over-range bit provides a new full scale of 1999, twice that
codes used with integrating type A/D converters employed of the previous one. In this case, the maximum output code
in digital panel meters, digital multimeters, and other is 1 1001 1001 1001. The additional range is commonly
954 Elements: C – Data Acquisition and Processing Systems
referred to as 1/2 digit, and the resolution of the A/D in Article 33, Coding Theory and its Application to Measure-
converter in this case is 31/2 digits. ment, Volume 1.
Likewise, if this range is again expanded by 100%, a
Austerlitz, H. (2003) Data Acquisition Techniques Using PCs,
new full scale of 3999 results and is called 33/4 digits
Academic Press, San Diego.
resolution. Here, two over-range bits have been added and
James, K. (2000) PC Interfacing and Data Acquisition: Tech-
the full-scale output code is 11 1001 1001 1001. When BCD
niques for Measurement, Instrumentation and Control, Newnes,
coding is used for bipolar measurements, another bit, a sign Oxford.
bit, is added to the code and the result is sign-magnitude
Mihura, B. (2001) LabVIEW for Data Acquisition, Prentice Hall
BCD coding. PTR, Upper Saddle River, NJ.
Park, J. and MacKay, S. (2003) Practical Data Acquisition for
FURTHER READING Instrumentation and Control Systems, Newnes, Oxford.
Taylor, H.R. and Taylor, J.R. (1997) Data Acquisition for Sensor
The general principles of DAS are found explained in various Systems, Chapman & Hall, London.
titles. Coding methods are usually covered. Coding is covered
This article is based in part on a chapter by E.L. Zuch originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
138: Sampling Theory Relevant to DAS
Gerd Wöstenkühler
Hochschule Harz, Wernigerode, Germany
waveform, and the amplitude error is
1 Aperture Time 955
d
2 Sample-holds and Aperture Error 955 V = ta (A sin(ωt))t=0 = ta Aω (2)
dt
3 Sampled-data System and Sampling Theorem 956
4 Frequency Folding and Aliasing 957 The resultant error as a fraction of the peak-to-peak full-
scale value is
Further Reading 957
V
ε= = πf ta (3)
2A
As an example from this result, the aperture time required
1 APERTURE TIME to digitize a 1 kHz signal to 10-bits resolution can be
found. The resolution required is one part in 210 or approx-
Conversion time is frequently referred to as aperture time. imately 0.001.
In general, aperture time refers to the time uncertainty (or
ε 0.001
time window) in making a measurement and results in an ta = ≈ ≈ 320 × 10−9 (4)
πf 3.14 × 103
amplitude uncertainty (or error) in the measurement if the
signal is changing during this time. The result is a required aperture time of just 320 ns! A
As shown in Figure 1, the input signal to the analog-to- 1 kHz signal is not a particularly fast one, yet it is difficult
digital (A/D) converter changes by V during the aperture to find a 10-bit A/D converter to perform this kind of
time ta in which the conversion is performed. The error can conversion at a reasonable price! Fortunately, there is a
be considered as an amplitude error or a time error; the two relatively simple and inexpensive way around this dilemma
are related as follows: by using a sample-hold circuit.
dV (t)
V = ta (1)
dt 2 SAMPLE-HOLDS AND APERTURE
where dV (t)/dt is the rate of change with time of the ERROR
input signal.
It should be noted that V represents the maximum error A sample-hold circuit samples the signal voltage and then
due to signal change, since the actual error depends on how stores it on a capacitor for the time required to perform the
the conversion is done. At some point in time within ta , the A/D conversion. The aperture time of the A/D converter
signal amplitude corresponds exactly with the output code is, therefore, greatly reduced by the much shorter aperture
word produced. time of the sample-hold circuit. In turn, the aperture time
For the specific case of a sinusoidal input signal, the of the sample-hold is a function of its bandwidth and the
maximum rate of change occurs at the zero crossing of the switching time.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
956 Elements: C – Data Acquisition and Processing Systems
∆V
V(t )
(a)
ta
Figure 1. Aperture time and amplitude uncertainty. (Reproduced (b)
from Handbook of Measurement Science, Volume One, Sydenham,
1983 John Wiley & Sons Ltd.)
Figure 2 is a useful graph of (4). It gives the aper- (c)
ture time required for converting sinusoidal signals to a
maximum error less than one part in 2n where n is the res-
olution of the converter in bits. The peak-to-peak value of
the sinusoid is assumed to be the full-scale range of the (d)
A/D converter. The graph is useful for selecting a sample-
hold by aperture time or an A/D converter by conversion
time. Figure 3. Signal sampling: (a) signal, (b) sampling pulses,
(c) sampled signal, and (d) sampled and held signal. (Reproduced
from Handbook of Measurement Science, Volume One, Sydenham,
3 SAMPLED-DATA SYSTEM AND 1983 John Wiley & Sons Ltd.)
SAMPLING THEOREM
In data acquisition and distribution systems, and other periodic basis as illustrated in Figure 3. The train of sam-
sampled-data systems, analog signals are sampled on a pling pulses in Figure 3(b) represents a fast-acting switch,
which connects to the analog signal for a very short
time and then disconnects for the remainder of the sam-
106 pling period.
The result of the fast-acting sampler is identical with the
result of multiplying the analog signal by a train of sam-
4
bi
pling pulses of unity amplitude, giving the modulated pulse
ts
105
train of Figure 3(c). The amplitude of the original signal is
6
bi
ts
preserved in the modulation envelope of the pulses. If the
switch-type sampler is replaced by a switch and capacitor
8
104
bi
ts
(a sample-hold circuit), then the amplitude of each sample
Aperture time (ns)
10
is stored between the samples and a reasonable recon-
bi
ts
struction of the original analog signal results, as shown in
103
Figure 3(d).
12
bi
A common use of sampling is in the efficient use of
ts
13 14 1 16
data processing equipment and data transmission facilities.
bi bit 5 b bi
ts s its ts
102 A single data transmission link, for example, can be used
to transmit many different analog channels on a sampled,
time-multiplexed basis, whereas it would be uneconomical
10 to devote a complete transmission link to the continuous
transmission of a single signal.
Likewise, a data acquisition and distribution system is
1 used to measure and control the many parameters of a
1 10 102 103 104 105 process control system by sampling the parameters and by
Sinusoidal frequency (Hz) updating the control inputs periodically. In data conversion
Figure 2. Graph of aperture error for sinusoidal signals. (Repro- systems, it is common to use a single, expensive A/D
duced from Handbook of Measurement Science, Volume One, converter of high speed and precision, and then multiplex
Sydenham, 1983 John Wiley & Sons Ltd.) a number of analog inputs into it.
Sampling Theory Relevant to DAS 957
An important fundamental question to answer about Signal
sampled-data systems is this: ‘How often must I sample
an analog signal in order not to lose information from it’?
It is obvious that all useful information can be extracted
if a slowly varying signal is sampled at a rate such that
little or no change takes place between samples. Equally Alias
Sampling
obvious is the fact that the information is being lost if pulses frequency
there is a significant change in signal amplitude between
samples. Figure 5. Alias frequency caused by inadequate sampling rate.
The answer to the question is contained in the well- (Reproduced from Handbook of Measurement Science, Volume
known sampling theorem, which may be stated as follows: One, Sydenham, 1983 John Wiley & Sons Ltd.)
if a continuous bandwidth-limited signal contains no fre-
quency components higher than fc , then the original signal signal, the folded part of the spectrum causes distortion in
can be recovered without distortion if it is sampled at a rate the recovered signal, which cannot be eliminated by filter-
of at least 2fc samples per second. ing the recovered signal.
From Figure 4, if the sampling rate is increased such
that fs − fc > fc , then the two spectra are separated and
4 FREQUENCY FOLDING AND ALIASING the original signal can be recovered without distortion.
This demonstrates the results of the sampling theorem that
fs > 2fc . Frequency folding can be eliminated in two ways:
The sampling theorem can be demonstrated by the fre- first, by using a high enough sampling rate, and second, by
quency spectra illustrated in Figure 4. Figure 4(a) shows filtering the signal before sampling to limit its bandwidth
the frequency spectrum of a continuous bandwidth-limited to fs /2.
analog signal with frequency components out to fc . When It must be appreciated that in practice there is always
this signal is sampled at a rate fs , the modulation process some frequency folding present owing to high-frequency
shifts the original spectrum out to . . . , −3fs , −2fs , −fs , fs , signal components, noise, and nonideal presample filtering.
2fs , 3fs , . . . , in addition to the one at the origin. A portion The effect must be reduced to negligible amounts for a
of this resultant spectrum is shown in Figure 4(b). particular application by using a sufficiently high sampling
If the sampling frequency fs is not high enough, part rate. The required rate, in fact, may be much higher than
of the spectrum centered about fs will fold over into the the minimum indicated by the sampling theorem.
original signal spectrum. This undesirable effect is called The effect of an inadequate sampling rate on a sinusoid is
frequency folding. In the process of recovering the original illustrated in Figure 5; an alias frequency in the recovered
signal results (see also Article 31, Discrete Signal Theory,
V
Volume 1). In this case, sampling at a rate slightly less
than twice per cycle gives the low-frequency sinusoid
shown by the dotted line in the recovered signal. This
alias frequency can be significantly different from the
original frequency. From the figure, it is easy to see that
f
(a) O fc if the sinusoid is sampled at least twice per cycle, as
required by the sampling theorem, the original frequency
is preserved.
Frequency folding
V
FURTHER READING
The general principles of DAS are explained in various titles.
O fs−fc fc fs f s+ f c
They usually contain information on sampling effects. IC
(b) fs/2 converter chip manufacturers offer technical data on their
products.
Figure 4. Frequency spectra demonstrating the sampling the-
orem: (a) continuous signal spectrum and (b) sampled signal
spectrum. (Reproduced from Handbook of Measurement Science, Austerlitz, H. (2003) Data Acquisition Techniques Using PCs,
Volume One, Sydenham, 1983 John Wiley & Sons Ltd.) Academic Press, San Diego.
958 Elements: C – Data Acquisition and Processing Systems
Mihura, B. (2001) LabVIEW for Data Acquisition, Prentice Hall Park, J. and MacKay, S. (2003) Practical Data Acquisition for
PTR, Upper Saddle River NJ. Instrumentation and Control Systems, Newnes, Oxford.
James, K. (2000) PC Interfacing and Data Acquisition: Tech- Taylor, H.R. and Taylor, J.R. (1997) Data Acquisition for Sensor
niques for Measurement, Instrumentation and Control, Newnes, Systems, Chapman & Hall, London.
Oxford.
This article is based in part on a chapter by E.L. Zuch originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
139: Analog-to-Digital (A/D) Converters
Gerd Wöstenkühler
Hochschule Harz, Wernigerode, Germany
The obvious advantage of the tracking A/D converter
1 Counter Type A/D Converter 959 is that it can continuously follow the input signal and
2 Successive-approximation A/D Converters 959 give updated digital output data if the signal does not
change too rapidly. Also, for small input changes, the
3 Parallel (Flash) A/D Converters 960
conversions can be quite fast. The converter can be oper-
Further Reading 961 ated in either the track or hold modes by a digital
input control.
1 COUNTER TYPE A/D CONVERTER
2 SUCCESSIVE-APPROXIMATION A/D
Analog-to-digital (A/D) converters, also called ADCs or CONVERTERS
encoders, employ a variety of different circuit techniques
to implement the conversion function. As with digital-to- By far, the most popular A/D conversion technique in gen-
analog (D/A) converters, however, relatively few of the eral use for moderate to high-speed applications is the
many originally devised circuits are widely used today. Of successive-approximation type A/D. This method falls into
the various techniques available, the choice depends on the a class of techniques known as feedback type A/D con-
resolution and speed required. verters, to which the counter type also belongs. In both
One of the simplest A/D converters is the counter, or cases, a D/A converter (see also Article 141, Digital-to-
servo type. This circuit employs a digital counter to control Analog (D/A) Converters, Volume 3) is in the feedback
the input of a D/A converter (see also Article 141, Digital- loop of a digital control circuit, which changes its out-
to-Analog (D/A) Converters, Volume 3). Clock pulses are put until it equals the analog input. In the case of the
applied to the counter and the output of the D/A is stepped successive-approximation converter, the DAC is controlled
up one LSB (least significant bit) at a time. A comparator in an optimum manner to complete a conversion in just
compares the D/A output with the analog input and stops n steps, where n is the resolution of the converter in
the clock pulses when they are equal. The counter output bits.
is then the converted digital word. The operation of this converter is analogous to weighing
While this converter is simple, it is also relatively slow. an unknown on a laboratory balance scale using standard
An improvement on this technique is shown in Figure 1 weights in a binary sequence such as 1/2, 1/4, 1/8, . . . , 1/2n kg.
and is known as a tracking A/D converter, a device com- The correct procedure is to begin with the largest standard
monly used in control systems. Here an up/down counter weight and proceed in order down to the smallest one.
controls the DAC, and the clock pulses are directed to The largest weight is placed on the balance pan first; if it
the pertinent counter input depending on whether the D/A does not tip, the weight is left on and the next largest weight
output must increase or decrease to reach the analog is added. If the balance does tip, the weight is removed
input voltage. and the next one added. The same procedure is used for
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
960 Elements: C – Data Acquisition and Processing Systems
Analog
input D/A converter ¾FS
Analog
Digital input
Comparator
output
data ½FS
Track/
hold Up Output code: 10110111
Up/down
counter
Down ¼FS
Clock
Figure 1. Tracking-type A/D converter. (Reproduced from Clock period: 1 2 3 4 5 6 7 8
Handbook of Measurement Science, Volume One, Sydenham,
1983 John Wiley & Sons Ltd.) Figure 3. D/A output for 8-bit successive-approximation con-
version. (Reproduced from Handbook of Measurement Science,
Volume One, Sydenham, 1983 John Wiley & Sons Ltd.)
the next largest weight and so on down to the smallest.
After the nth standard weight has been tried and a decision circuitry, in particular the D/A and comparator, are critical
made, the weighing is finished. The total of the standard for high-speed performance.
weights remaining on the balance is the closest possible
approximation to the unknown.
In the successive-approximation A/D converter illus- 3 PARALLEL (FLASH) A/D CONVERTERS
trated in Figure 2, a successive-approximation register
(SAR) controls the D/A converter by implementing the For ultrafast conversions required in video signal processing
weighing logic just described. The SAR first turns on the and radar applications where up to 8-bits resolution is
MSB (most significant bit) of the DAC and the compara- required, a different technique is employed; it is known
tor tests this output against the analog input. A decision as the parallel (also flash, or simultaneous) method and is
is made by the comparator to leave the bit on or turn it illustrated in Figure 4. This circuitry employs 2n − 1 analog
off after which bit 2 is turned on and a second compari- comparators to directly implement the quantizer transfer
son is made. After n comparisons, the digital output of the function of an A/D converter.
SAR indicates all those bits that remain on and produce the The comparator trip-points are spaced 1 LSB apart by
desired digital code. The clock circuit controls the timing of the series resistor chain and voltage reference. For a given
the SAR. Figure 3 shows the D/A converter output during analog input voltage, all comparators that are biased below
a typical conversion. the voltage turn on and all those biased above it remain
The conversion efficiency of this technique means that off. Since all comparators change state simultaneously, the
high-resolution conversions can be made in a very short quantization process is a one-step operation.
time. For example, it is possible to perform a 10-bit con- A second step is required, however, since the logic out-
version in 1 µs or less. Of course, the speed of the internal put of the comparators is not in binary form. Therefore,
R I
Analog
input D/A converter Ref
Digital
Comparator output
data
Successive
approximation Clock
register
Figure 2. Successive-approximation A/D converter. (Reproduced from Handbook of Measurement Science, Volume One, Sydenham,
1983 John Wiley & Sons Ltd.)
Analog-to-Digital (A/D) Converters 961
resolutions. A 4-bit converter, for example, requires only
Reference
Input
15 comparators, but an 8-bit converter needs 255. For this
reason, it is common practice to implement an 8-bit A/D
2n − 1
3R /2 with two 4-bit stages as shown in Figure 5.
Comparators
The result of the first 4-bit conversion is converted back
R to analog by means of an ultrafast 4-bit D/A and then
subtracted from the analog input. The resulting residue is
then converted by the second 4-bit A/D and the two sets of
R
Decoder
Binary data are accumulated in the 8-bit output register. Converters
output
of this type achieve 8-bit conversions at rates of 20 MHz
R and higher, while single-stage 4-bit conversions can reach
50 to 100 MHz rates.
R /2 FURTHER READING
A/D converters are usually covered in the general titles in DAS.
Manufacturers provide data sheets and application notes for their
Figure 4. 4-bit parallel A/D converter. (Reproduced from Hand- own products.
book of Measurement Science, Volume One, Sydenham, 1983
John Wiley & Sons Ltd.)
Austerlitz, H. (2003) Data Acquisition Techniques Using PCs,
Academic Press, San Diego.
James, K. (2000) PC Interfacing and Data Acquisition: Tech-
+ niques for Measurement, Instrumentation and Control, Newnes,
Analog Oxford.
4-Bit 4-Bit − 4-Bit
input
A/D D/A A/D Mihura, B. (2001) LabVIEW for Data Acquisition, Prentice Hall
PTR, Upper Saddle River, NJ.
Park, J. and MacKay, S. (2003) Practical Data Acquisition for
Instrumentation and Control Systems, Newnes, Oxford.
8-Bit register
Taylor, H.R. and Taylor, J.R. (1997) Data Acquisition for Sensor
Systems, Chapman & Hall, London.
Bit 1 2 3 4 5 6 7 8
Output data Converters are covered in
Figure 5. Two-stage parallel 8-bit A/D converter. (Reproduced Hoeschele, D.F. (1994) Analog-to-Digital and Digital-to-Analog
from Handbook of Measurement Science, Volume One, Sydenham, Conversion Techniques, Wiley-Interscience, New York.
1983 John Wiley & Sons Ltd.)
Jespers, P.G.A. (2001) Integrated Converters: D to A and A to
D Architectures, Analysis and Simulation, Oxford University
an ultrafast decoder circuit is employed to make the logic Press, Oxford.
conversion to binary. The parallel technique reaches the Moscovici, A. (2001) High Speed A/D Converters – Understand-
ultimate at high speed because only two sequential opera- ing Data Converters Through SPICE, Kluwer Academic Pub-
tions are required to make the conversion. lishers, Boston.
The limitation of the method, however, is in the Razavi, B. (1995) Principles of Data Conversion System Design,
large number of comparators required for even moderate IEEE Press, New York.
This article is based in part on a chapter by E.L. Zuch originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
140: Integrating Type A/D Converters
Gerd Wöstenkühler
Hochschule Harz, Wernigerode, Germany
up to overflow. At this point, the control circuit switches
1 Indirect A/D Conversion 962 the integrator to the negative reference voltage, which is
2 Dual-slope A/D Conversion 962 integrated until the output is back to zero. Clock pulses are
counted during this time until the comparator detects the
3 Charge-balancing A/D Conversion 962
zero crossing and turns them off.
Further Reading 964 The counter output is then the converted digital word.
Figure 2 shows the integrator output waveform where T1
is a fixed time and T2 is the time proportional to the input
1 INDIRECT A/D CONVERSION voltage. The times are related as follows:
Vin
A class of analog-to-digital (A/D) converters, known as T2 = T1 (1)
Vref
integrating type, operates by an indirect conversion method.
The unknown input voltage is converted into a time period, The digital output word, therefore, represents the ratio of
which is then measured by a clock and counter. A num- the input voltage to the reference voltage.
ber of variations exist on the basic principle such as Dual-slope conversion has several important features.
single-slope, dual-slope, and triple-slope methods. In addi- First, conversion accuracy is independent of the stability
tion, there is another technique – completely different – of the clock and integrating capacitor so long as they are
that is known as the charge-balancing, quantized-feedback constant during the conversion period. Accuracy depends
method or sigma-delta modulator. only on the reference accuracy and the integrator circuit
The most popular of these methods are dual-slope and linearity. Second, the periodic noise rejection of the con-
charge balancing; although both are slow, they have excel- verter can be infinite if T1 is set to equal the period of the
lent linearity characteristics with the capability of rejecting noise. To reject 60 Hz power noise, therefore, requires that
input noise. Because of these characteristics, integrating T1 be 16.667 ms or its multiples.
type A/D converters are almost exclusively used in digital
panel meters, digital multimeters, and other comparatively
slow measurement applications.
3 CHARGE-BALANCING A/D
CONVERSION
2 DUAL-SLOPE A/D CONVERSION
The charge-balancing, or quantized-feedback, method of
The dual-slope technique, shown in Figure 1, is perhaps the conversion is based on the principle of generating a pulse
best known. Conversion begins when the unknown input train with a frequency proportional to the input voltage and
voltage is switched to the integrator input; at the same then counting the pulses for a fixed period of time. This cir-
time, the counter begins to count clock pulses and counts cuit is shown in Figure 3. Except for the counter and timer,
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Integrating Type A/D Converters 963
C C
Input Switch R1
R Vin −
− Comparator
I1
+
+ R2 Integrator
Integrator I2
−Vref Comparator Precision
S1 pulse
generator
Clock Control logic −Vref
Counter Timer
Counter
Digital
output
Digital output
Figure 3. Charge-balancing A/D converter. (Reproduced from
Figure 1. Dual-slope A/D converter. (Reproduced from Hand- Handbook of Measurement Science, Volume One, Sydenham,
book of Measurement Science, Volume One, Sydenham, 1983 1983 John Wiley & Sons Ltd.)
John Wiley & Sons Ltd.)
has the following relationship:
Integrator output voltage
1 Vin R2
Full-scale f = (2)
conversion τ Vref R1
Half-scale
conversion where τ is the pulse width and f the frequency.
A higher input voltage, therefore, causes the integrator
Quarter-scale
conversion to ramp up and down faster, producing higher frequency
output pulses. The timer circuit sets a fixed time period
for counting and the number of pulses within this period
represents the digital output.
T1 T2
(Fixed time) (Measured time)
If the output of the comparator is synchronized with a
clock signal and directly used to control the switch S1 as the
Figure 2. Integrator output waveform for dual-slope A/D con- pulse generator and more than one clock period is needed
verter. (Reproduced from Handbook of Measurement Science, to balance the current of maximum input voltage over two
Volume One, Sydenham, 1983 John Wiley & Sons Ltd.)
clock periods, the circuit becomes a sigma-delta modula-
tor. The synchronized comparator output signal generates
the circuit is a voltage-to-frequency (V/F) converter that a sequence of 1 and 0, and the output pulse density bal-
generates an output pulse rate proportional to input voltage. ances the input voltage. The counter for the digital output
The circuit operates as follows. A positive input voltage counts the periods n where the switch is connected to the
causes a current to flow into the operational integrator reference voltage and its relation to the used number N
of clock periods in the time window represents the digital
through R1 . This current is integrated, producing a negative
output signal.
going ramp at the output. Each time the ramp crosses zero,
the comparator output triggers a precision pulse generator, Vin R2
n=N (3)
which puts out a constant width pulse. Vref R1
The pulse output controls switch S1 , which connects R2
to the negative reference for the duration of the pulse. Dur- Like the dual-slope converter, the circuit also integrates
ing this time, a pulse of current flows out of the integrator- input noise, and if the timer is synchronized with the
summing junction, producing a fast, positive ramp at the noise frequency, infinite rejection results. Figure 4 shows
integrator output. This process is repeated, generating a the noise rejection characteristic of all integrating type
train of current pulses, which exactly balances the input A/D converters with rejection plotted against the ratio of
current – hence the name charge balancing. This balance integration period to noise period.
964 Elements: C – Data Acquisition and Processing Systems
30 Austerlitz, H. (2003) Data Acquisition Techniques Using PCs,
Academic Press, San Diego.
Mihura, B. (2001) LabVIEW for Data Acquisition, Prentice Hall
PTR, Upper Saddle River, NJ.
Noise rejection (dB)
20 James, K. (2000) PC Interfacing and Data Acquisition: Tech-
niques for Measurement, Instrumentation and Control, Newnes,
Oxford.
Park, J. and MacKay, S. (2003) Practical Data Acquisition for
Instrumentation and Control Systems, Newnes, Oxford.
10
Taylor, H.R. and Taylor, J.R. (1997) Data Acquisition for Sensor
T = Integration
period
Systems, Chapman & Hall, London.
Tn = Noise period
Converters are covered in the following:
0
0.5 1.0 2.0 3.0 4.0 5.0
T/Tn Hoeschele, D.F. (1994) Analog-to-Digital and Digital-to-Analog
Conversion Techniques, Wiley-Interscience, New York.
Figure 4. Noise rejection for integrating type A/D converters. Jespers, P.G.A. (2001) Integrated Converters: D to A and A to
(Reproduced from Handbook of Measurement Science, Volume D Architectures, Analysis and Simulation, Oxford University
One, Sydenham, 1983 John Wiley & Sons Ltd.) Press, Oxford.
Moscovici, A. (2001) High Speed A/D Converters – Understand-
ing Data Converters Through SPICE, Kluwer Academic Pub-
FURTHER READING lishers, Boston, MA.
The general principles of DAS are found explained in various
titles. These often cover converters.
This article is based in part on a chapter by E.L. Zuch originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
141: Digital-to-Analog (D/A) Converters
Gerd Wöstenkühler
Hochschule Harz, Wernigerode, Germany
2 WEIGHTED CURRENT SOURCE D/A
1 Introduction to D/A Converters 965 CONVERTER
2 Weighted Current Source D/A Converter 965
3 R –2R D/A Converter 966 The most popular D/A converter design in use today is the
4 Multiplying and Deglitching D/A Converters 968 weighted current source circuit illustrated in Figure 2. An
Further Reading 970 array of switched transistor current sources is used with
binary weighted currents. The binary weighting is achieved
by using emitter resistors with binary related values of
R, 2R, 4R, 8R, . . . 2n−1 R. The resulting collector currents
1 INTRODUCTION TO D/A CONVERTERS are then added together at the current summing line.
The current sources are switched on or off from standard
Digital-to-analog (D/A) converters are the devices by which TTL semiconductor device inputs by means of the control
computers communicate with the outside analog world. diodes connected to each emitter. When the TTL input is
They are employed in a variety of applications from CRT high, the current source is on; when the input is low, it is
display systems and voice synthesizers to automatic test off, with the current flowing through the control diode. Fast
systems, digitally controlled attenuators, and process con- switching speed is achieved because there is direct control
trol actuators. In addition, they are key components inside of the transistor current, and the current sources never go
most analog-to-digital (A/D) converters. D/A converters are into saturation.
also referred to as DACs and are termed decoders by com- To interface with standard TTL levels, the current sources
munications engineers. are biased to a base voltage of +1.2 V. The emitter currents
The transfer characteristic of an ideal 3-bit D/A converter are regulated to constant values by means of the control
is shown in Figure 1. Each input code word produces amplifier and a precision voltage reference circuit together
a single, discrete analog output value, generally, but not with a binary transistor.
always, a voltage. Over the output range of the converter The summed output currents from all current sources that
2n different values are produced including zero; and the are on go to an operational amplifier summing junction;
output has a one-to-one correspondence with input, which the amplifier converts this output current into an output
is not true for A/D converters. voltage. In some D/A converters, the output current is used
There are many different circuit techniques used to to directly drive a resistor load for maximum speed, but
implement D/A converters, but a few popular ones are the positive output voltage in this case is limited to about
widely used today. Virtually all D/A converters in use are of +1 V.
the parallel type where all bits change simultaneously upon The weighted current source design has the advantage of
application of an input code word; serial type D/A convert- simplicity and high speed. Both PNP and NPN transistor
ers, on the other hand, produce an analog output only after current sources can be used with this technique although the
receiving all digital input data in sequential form. TTL interfacing is more difficult with NPN sources. This
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
966 Elements: C – Data Acquisition and Processing Systems
FS A difficulty in implementing the higher resolution D/A
converter designs to this concept is that a wide range of
emitter resistors is required and very high value resistors
cause problems with both temperature stability and switch-
¾ FS ing speed. To overcome these problems, weighted current
sources are used in identical groups, with the output of
each group divided down by a resistor divider as shown in
Analog output
Figure 3.
½ FS The resistor network, R1 through R4 , divides the output
Q of group 3 down by a factor of 256 and the output of
group 2 down by a factor of 16 with respect to the output
of group 1. Each group is identical, with four current
¼ FS sources of the type shown in Figure 2, having binary
current weights of 1, 2, 4, 8. Figure 3 also illustrates the
method of achieving a bipolar output by deriving an offset
current from the reference circuit, which is then subtracted
000 001 010 011 100 101 110 111
from the output current line through resistor RO . This
Input code current is set to exactly one half of the full scale output
current.
Figure 1. Transfer characteristic of ideal 3-bit D/A converter.
(Source: Reproduced from Handbook of Measurement Science,
Volume One, Sydenham, 1983 John Wiley & Sons Ltd.) 3 R –2R D/A CONVERTER
technique is used in most monolithic, hybrid, and modular A second popular technique for D/A conversion is the
D/A converters. R –2R ladder method. As shown in Figure 4, the network
TTL Input data
+Vs 1 2 3 n
R1 R R 2R 4R 2n −1R
+1.2V
R2
Iout
R3 RF
−
Vout
+
−
RR
+
−VREF
Figure 2. Weighted current source D/A converter. (Source: Reproduced from Handbook of Measurement Science, Volume One,
Sydenham, 1983 John Wiley & Sons Ltd.)
Digital-to-Analog (D/A) Converters 967
Current Current Current
source source source
group 3 group 2 group 1
R2 R4
RF
R1 R3 −
VOUT
RO +
−VREF
Figure 3. Current dividing the outputs of weighted current source groups. (Source: Reproduced from Handbook of Measurement
Science, Volume One, Sydenham, 1983 John Wiley & Sons Ltd.)
consists of series resistors of value R and shunt resistors of into two equal parts since it sees equal resistances in either
values 2R. The bottom of each shunt resistor has a single- direction. Likewise, the current flowing down the ladder to
pole double-throw electronic switch, which connects the the right continues to divide into two equal parts at each
resistor to either ground or to the output current summing resistor junction.
line. As in the previous circuit, the output current summing The result is binary weighted currents flowing down
line goes to an operational amplifier that converts current each shunt resistor in the ladder. The digitally controlled
to voltage. switches direct the currents to either the summing line or to
The operation of the R –2R ladder network is based on ground. Assuming all bits are on, as shown in the diagram,
the binary division of current as it flows down the lad- the output current is
der. Examination of the ladder configuration reveals that
at point A, looking to the right, one measures a resistance
of 2R; therefore, the reference input to the ladder has a V 1 1 1 1
IOUT = REF + + +··· + n (1)
resistance of R. At the reference input, the current splits R 2 4 8 2
A
IIN
R R R
+ 2R 2R 2R 2R 2R
VREF
RF
−
V
+
Figure 4. R –2R ladder D/A converter. (Source: Reproduced from Handbook of Measurement Science, Volume One, Sydenham, 1983
John Wiley & Sons Ltd.)
968 Elements: C – Data Acquisition and Processing Systems
which is a binary series. The sum of all currents is then obtained for high resolution D/A converters by using laser-
trimmed thin film resistor networks.
VREF
IOUT = (1 − 2−n ) (2)
R
4 MULTIPLYING AND DEGLITCHING
−n D/A CONVERTERS
where the 2 term physically represents the portion of the
input current flowing through the 2R terminating resistor
to ground at the far right. The R –2R ladder method is specifically used for multiply-
The advantage of the R –2R ladder technique is that ing D/A converters. With these converters, the reference
only two values of resistors are required, with the resultant voltage can be varied over the full range of ±Vmax with the
ease of matching or trimming and excellent temperature output being the product of the reference voltage and the
tracking. In addition, for high speed applications, relatively digital input word. Multiplication can be performed in 1, 2,
low resistor values can be used. Excellent results can be or 4 algebraic quadrants.
14 Bit multiplying DAC
10 K 10 K 10 K 10 K
REF
IN
20 K 20 K 20 K 20 K 20 K 20 K
Out 1
Out 2
10 K
Feed back
Bit 1 Bit 2 Bit 3 Bit 4 Bit 14
(MSB) (LSB)
Figure 5. CMOS 14-bit multiplying D/A converters. (Source: Reproduced from Handbook of Measurement Science, Volume One,
Sydenham, 1983 John Wiley & Sons Ltd.)
+VS
I I I I RF
−
2R R R R VOUT
2R 2R 2R 2R +
Figure 6. D/A converter employing R –2R ladder with equal value switched current sources. (Source: Reproduced from Handbook of
Measurement Science, Volume One, Sydenham, 1983 John Wiley & Sons Ltd.)
Digital-to-Analog (D/A) Converters 969
Register
¾FS D/A Sample- VOUT
Analog output
Converter hold
½FS
¼FS
(a) Time (b) Control
Figure 7. (a) Output glitches and (b) deglitched D/A converter. (Source: Reproduced from Handbook of Measurement Science, Volume
One, Sydenham, 1983 John Wiley & Sons Ltd.)
1%
Settling
1 LSB
Settling
Load Data DAC
Output changes
Register valid changes
5 0 10 30 250 600
Nano seconds
Figure 8. Modular deglitched D/A converter. (Source: Reproduced from Handbook of Measurement Science, Volume One, Sydenham,
1983 John Wiley & Sons Ltd.)
If the reference voltage is unipolar, the circuit is a The advantage of the equal value current sources is obvious
one-quadrant multiplying DAC; if it is bipolar, the cir- since all emitter resistors are identical and switching speeds
cuit is a two-quadrant multiplying DAC. For a four- are also identical. This technique is used in many ultrahigh
quadrant operation the two current summing lines shown in speed D/A converters.
Figure 4 must be subtracted from each other by operational One other specialized type D/A converter used primarily
amplifiers. in CRT display systems is the deglitched D/A converter. All
In multiplying D/A converters, the electronic switches D/A converters produce output spikes, or glitches, which
are usually implemented with CMOS devices. Multiplying are most serious at the major output transitions of 14 FS,
1
DACs are commonly used in automatic gain controls, CRT 2
FS, and 34 FS as illustrated in Figure 7(a).
character generation, complex function generators, digital Glitches are caused by small time differences between
attenuators, and divider circuits. Figure 5 shows two 14-bit some current sources turning off and others turning on.
multiplying CMOS D/A converters. Take, for example, the major code transition at half scale
Another important D/A converter design takes advan- from 0111 · · · 1 to 1000 · · · 0. Here, the MSB current source
tage of the best features of both the weighted current turns on while all other current sources turn off. The small
source technique and the R –2R ladder technique. This cir- difference in switching times results in a narrow half-
cuit, shown in Figure 6, uses equal value switched current scale glitch. Such a glitch produces distorted characters on
sources to drive the junctions of the R –2R ladder network. CRT displays.
970 Elements: C – Data Acquisition and Processing Systems
Glitches can be virtually eliminated by the circuit shown Mihura, B. (2001) LabVIEW for Data Acquisition, Prentice Hall
in Figure 7(b). The digital input to a D/A converter is PTR, Upper Saddle River, NJ.
controlled by an input register while the converter output James, K. (2000) PC Interfacing and Data Acquisition: Tech-
goes to a specially designed sample-hold circuit. When the niques for Measurement, Instrumentation and Control, Newnes,
Oxford.
digital input is updated by the register, the sample-hold
is switched into the hold mode. After the D/A converter Park, J. and MacKay, S. (2003) Practical Data Acquisition for
Instrumentation and Control Systems, Newnes, Oxford.
has changed to its new output value and all glitches have
settled out, the sample-hold is then switched back into Taylor, H.R. and Taylor, J.R. (1997) Data Acquisition for Sensor
Systems, Chapman & Hall, London.
the tracking mode. When this happens, the output changes
smoothly from its previous value to the new value with no
Converters are covered in
glitches present. Figure 8 shows a modular deglitched D/A
converter, which contains the circuitry just described. Hoeschele, D.F. (1994) Analog-to-Digital and Digital-to-Analog
Conversion Techniques, Wiley-Interscience, New York.
Jespers, P.G.A. (2001) Integrated Converters: D to A and A to
FURTHER READING D Architectures, Analysis and Simulation, Oxford University
Press, Oxford.
Converters used in DAS setups are explained in most general titles
Moscovici, A. (2001) High Speed A/D Converters – Understand-
on data acquisition. Manufacturers supply application notes and
ing Data Converters Through SPICE, Kluwer Academic Pub-
data sheets for their products.
lishers, Boston, MA.
Austerlitz, H. (2003) Data Acquisition Techniques Using PCs, Razavi, B. (1995) Principles of Data Conversion System Design,
Academic Press, San Diego. IEEE Press, New York.
This article is based in part on a chapter by E.L. Zuch originally published in Handbook of Measurement Science Vol 1, P.H. Sydenham,
Wiley, 1982.
142: Z-transforms
Amar Bousbaine
University of Derby, Derby, UK
to difference equations because convolution operators are
1 Purpose and Role of the Z-transform in transformed into simple algebraic ones. In many cases, the
Measurement Systems 973 sum of finite or infinite series for the z-transform can be
2 Introduction to Z-transform Theory 973 expressed in a closed form expression, which offers a com-
3 Region of Convergence 974 pact alternative representation of signals.
Prototype continuous time filters can be converted into
4 Properties of the Region of Convergence 975
high-quality digital filters by simple mapping from the s-
5 Properties of Z-transform 975 domain to the z-domain. Consequently, the z-transform
6 Some Common Z-transform Pairs 976 has become an important tool for the analysis and design
7 The Inverse Z-transform 976 of digital filters (Article 148, General Performance of
8 Application of the Z-transform in Signal the Digital Filter, Volume 3; Article 149, Low-, High-,
Processing 978 and Band-pass Digital Filters, Volume 3; Article 150,
9 Causality and Stability 978 Finite Infinite Impulse Response (IIR) Digital Filters,
Volume 3; Article 151, Finite Impulse Response (FIR)
References 980
Digital Filters, Volume 3).
1 PURPOSE AND ROLE OF THE 2 INTRODUCTION TO Z-TRANSFORM
Z-TRANSFORM IN MEASUREMENT THEORY
SYSTEMS
The z-transform of a sequence x[n] is defined as
The z-transform is an invaluable tool for representing, ∞
analyzing, and designing discrete time signals and sys- Z{x[n]} ≡ X(z) = x[n]z−n (1)
tems. Since most measurement systems now contain sig- n=−∞
nificant digital signal processing (DSP), the z-transform is
an important element in measurement system design. Infor- where z is a complex variable.
mation about signals can be obtained by examining their If z is defined as z = rejω , where r = |z| is the module
z-transforms. In particular, the stability or instability of a of z and ω its argument, then (1) becomes
linear time invariant discrete system is characterized by the ∞
location of the poles in the z-plane. It also indicates how
X(rejω ) = x[n]r −n e−jωn (2)
the system responds to particular oscillatory components of n=−∞
the input.
Moreover, the use of z-transform techniques permits sim- Equation (2) can be interpreted as the discrete time Fourier
ple algebraic manipulations, that is, it facilitates the solution transform (DTFT) of the modified sequence, x[n]r −n .
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
974 Elements: C – Data Acquisition and Processing Systems
For r = 1, (2) reduces to the Fourier transform of x[n], properties of the ROC are illustrated in the examples
provided the latter exists. The contour r = |z| = 1 is a circle that follow.
in the z-plane of unity radius and is called the unit circle. Example 1. Right-sided sequence
The definition, (1), can be applied to the impulse response Determine the z-transform of the signal,
sequence, h[n], of a digital filter. The z-transform of h[n]
n ∞
is called the transfer function of the filter and is defined as a , n≥0
x[n] = a n u[n] = = a n z−n
∞ 0, n < 0
n=0
H (z) = h[n]z−n (3)
n=−∞ Solution: Because x[n] is nonzero only for n ≥ 0, the
sequence is known as a right-sided sequence.
From the definition, (1)
3 REGION OF CONVERGENCE ∞
∞
∞
X(z) = x[n]z−n = a n z−n = (az−1 )n
Since the z-transform is an infinite power series, it exists n=−∞ n=0 n=0
only for those values of z for which this series con- ∞
For convergence, n=0 |(az−1 )n | < ∞
verges. For a given sequence, the set of values for which Thus the region of convergence is the range of values
its z-transform converges is called the region of conver- of z for which |az−1 | < 1 or equivalently, |z| > |a|, see
gence (ROC). Figure 2. Inside the ROC, the infinite series converges to
The z-transform exists when the finite sum of (1) con-
∞
verges. A necessary condition for convergence
is absolute
1 z
summability of x[n]z−n . Since x[n]z−n = x[n]r −n , then X(z) = (az−1 )n = −1
= for |z| > |a|
n=0
1 − az z−a
∞
Example 2. Left-sided sequence
x[n]r −n < ∞ (4)
n=−∞
Determine the z-transform of the signal,
n
The region of convergence of a z-transform is important −a , n ≤ −1
x[n] = −a n u[−n − 1] =
in that without knowledge of the ROC, there is no unique 0, n≥0
relationship between a sequence and its z-transform. Hence, −1
a z-transform must always be specified with its region of = (−a)n z−n
convergence. Moreover, if the ROC of a z-transform of n=−∞
a sequence includes the unit circle, the Fourier transform
Solution: Because x[n] is nonzero only for n ≤ −1, the
of the sequence is obtained simply by evaluating the
sequence is known as a left-sided sequence.
z-transform on the unit circle (Figure 1).
From the definition, (1)
One-sided exponential signals are encountered frequently
in engineering problems because one is interested in the ∞
−1
−n
behavior of a signal either before or after a given time X(z) = x[n]z =− a n z−n
instant. With such signals, the magnitude of one or more n=−∞ n=−∞
poles determines the ROC boundaries. Some of these ∞
∞
= (a −1 z)n = 1 − (a −1 z)n
n=1 n=0
Im z = e jw
Unit circle
ROC is given by, |a −1 z| < 1 or equivalently, |z| <
|a| (see Figure 3). Inside the ROC, the infinite series
converges to
w
1 1 z
Re X(z) = 1 − = for |z| < |a|
1 − a −1 z z−a
Example 3. Two-sided sequence
Determine the z-transform of the signal,
n
1 n 1
x[n] = − u[n] − u[−n − 1]
Figure 1. The unit circle. 3 2
Z-transforms 975
Im z-plane
ROC Im
Re
X O O X
−1/3 1/2
Re
O X
a 1
Unit circle
Figure 4. Pole-zero plot and ROC for a two-sided sequence.
Property 1: The ROC is a ring or a disk in the z-plane
Figure 2. Pole-zero plot and ROC for a right-sided sequence. centered at the origin (Figures 2, 3, 4, and 6).
Property 2: The ROC cannot contain any poles, (Figures
Im z-plane 2, 3, and 4).
ROC Property 3: The Fourier transform of x[n] converges abso-
lutely if and only if the ROC of the z-transform
of x[n] includes the unit circle.
Property 4: If x[n] is a finite duration sequence, that is, a
sequence that is zero except in finite interval
Re
−∞ < M1 ≤ n ≤ M2 < ∞, then the ROC is
O
a
X
1 the entire plane, except possibly z = 0 or
z = ∞.
Property 5: If x[n] is a right-sided sequence, that is, a
sequence for n < M1 < ∞, the region of con-
Unit circle
vergence extends outward from the outermost
finite pole in X(z) to, and possibly include,
z = ∞ (Figure 2).
Property 6: If x[n] is a left-sided sequence, that is, a
Figure 3. Pole-zero plot and ROC for a left-sided sequence.
sequence for −∞ < M2 < n, the region of
Solution: From Examples 1, 2 and Table 2 the z-trans- convergence extends inward from the inner-
form is most nonzero pole in X(z) to, and possibly
include, z = 0 (Figure 3).
1 n 1 1 Property 7: A two-sided sequence is an infinite-duration
− u[n] Z −1
, ROC is |z| >
3 ←−→ 1+ 3 z
1 3 sequence that is neither right sided nor left
n sided. If x[n] is a two-sided sequence, the ROC
1 1 1
u[−n − 1] Z , ROC is |z| < will consist of a ring in the z-plane, bounded
2 ←−→ 1 − 1
z −1 2
2 on the interior and exterior by a pole and,
1 1 1 1 consistent with property 2, not containing any
X(z) = + , ROC is < |z| <
1+ 1
3
z−1 1− 1
2
z−1 3 2 poles (Figure 4).
(see Figure 4). Property 8: The ROC must be a connected region (Figure
2, 3, 4, 5, and 6).
4 PROPERTIES OF THE REGION OF
CONVERGENCE 5 PROPERTIES OF Z-TRANSFORM
The basic properties of the ROC depend on the nature of The z-transform possesses a number of properties that make
the signal (for more details see Oppenheim and Schafer it an extremely valuable tool in the study of discrete time
(1999) and Proakis and Manolakis (1996)). signals and systems. Some of the useful properties that
976 Elements: C – Data Acquisition and Processing Systems
Table 1. Properties of the z-transform.
Property Sequence z-transform ROC
x[n] X(z) ROC: R2 < |z| < R1
x1 [n] X1 (z) R1c
x2 [n] X2 (z) R2c
Linearity ax1 [n] + bx2 [n] aX1 (z) + bX2 (z) Includes R1c ∩ R2c
Time shift x[ n − n o ] z−no X(z) That of X(z), except possibly z = 0
if n0 > 0 and z = ∞ if n0 < 0
Frequency scaling a n x[n] X(a −1 z) |a|R2 < |z| < |a|R1
Time reversal x[−n] X(z−1 ) 1
R2
< |z| < R11
d
Multiplication by n nx[n] −z X(z) ROC
dz
∗
Convolution x1 [n] x2 [n] X1 (z)X2 (z) At least the intersection of R1c and
R2c
Conjugation x ∗ [n] X ∗ (z∗ ) ROC
1
Real part Re{x[n]} {X(z) + X ∗ (z∗ )} Includes ROC
2
1
Imaginary part I m{x[n]} {X(z) − X ∗ (z∗ )} Includes ROC
j2
1
Multiplication x1 [n]x2 [n] X1 (v)X2 ( vz )v −1 dv At least, R1l R2l < |z| < R1u R2u
j2π
Im
Im Im
Re Re Re
Figure 5. Intersection of the ROCs of the right-sided and left-sided sequences.
have found practical use in digital signal processing are 7.1 Direct method (using the inversion integral)
summarized in Table 1.
1
x[n] = X(z)zn−1 dz (5)
j2π c
6 SOME COMMON Z-TRANSFORM
where c represents integration along the closed contour,
PAIRS
c, in the counter-clockwise direction in the z-plane. The
contour must be chosen to lie in the region of convergence
The recovery of a time domain signal from its z-transform, of X(z).
the inverse z-transform, can often be easily evaluated by
expressing X(z) as a linear combination of simpler basic
z-transform pairs and the z-transform properties to invert a 7.2 Power series expansion method
large class of z-transform. The z-transform pairs of common
sequences are given in Table 2. The power series can be obtained by arranging the numer-
ator and the denominator of X(z) in descending powers of
z and then dividing the numerator by the dominator using
long division.
7 THE INVERSE Z-TRANSFORM Given a z-transform X(z) with its ROC, X(z) can
be
∞expanded into a power series of the form X(z) =
−n
There are three methods that are often used for finding a c
n=−∞ n z , which converges in the ROC. Then by the
sequence x[n], given its z-transform X(z). uniqueness of the z-transform, x[n] = cn for all n.
Z-transforms 977
Table 2. Some common z-transform pairs.
Signal, x[n] z-transform, X(z) Region of Convergence
1
∞
x[n] = X(z)zn−1 dz X(z) = x[n]z−n ROC
j2π −∞
δ[n] 1 All z
−m
δ[n − m] z All z
1
u[n] |z| > 1
1 − z−1
1
a n u[n] |z| > a
1 − az−1
az−1
na n u[n] |z| > a
(1 − az−1 )2
1
−a n u[−n − 1] |z| < a
1 − az−1
az−1
−na n u[−n − 1] |z| < a
(1 − az−1 )2
1 − z−1 cos(ωo )
cos(ωo n)u[n] |z| > 1
1 − 2z−1 cos(ωo ) + z−2
z−1 cos(ωo )
sin(ωo n)u[n] |z| > 1
1 − 2z−1 cos(ωo ) + z−2
1 − az−1 cos(ωo )
a n cos(ωo n)u[n] |z| > a
1 − 2az−1 cos(ωo ) + a 2 z−2
az−1 cos(ωo )
a n sin(ωo n)u[n] |z| > a
1 − 2az−1 cos(ωo ) + a 2 z−2
Im Im
Re Re
Figure 6. Unacceptable ROC.
7.3 Partial fraction expansion method given by
In the partial expansion method, the z-transform is first
M
expanded into a sum of simple partial fractions. The inverse bk z−k
z-transform of each partial fraction is then obtained from b + b1 z−1 + b2 z−2 + · · · + bM z−M k=0
tables, such as Table 1, and then summed to give the overall X(z) = 0 = N
a0 + a1 z−1 + a2 z−2 + · · · + aN z−N
inverse z-transform. ak z−k
In many practical cases of linear time invariant (LTI) k=0
systems, the z-transforms are rational functions of z−1 and (6)
978 Elements: C – Data Acquisition and Processing Systems
If the poles of X(z) are of the first order and, for If H (z) has poles at z = p1 , p2 , . . . , pM and zero at
simplicity, N = M, then X(z) can be rewritten as z = z1 , z2 , . . . , zM , then H (z) can be factored and repre-
sented as
A1 A2
X(z) = B0 + + (z − z1 )(z − z2 ) . . . (z − zM )
1 − p1 z−1 1 − p2 z−1 X(z) = K (10)
AM (z − p1 )(z − p2 ) . . . (z − pM )
+ ··· +
1 − pM z−1 where K is the gain factor.
A1 z A2 z A plot of the poles and zeros of the transfer function is
X(z) = B0 + +
z − p1 z − p2 known as the pole-zero diagram and provides a useful way
of representing and analyzing the filter. For example, for
AM z Az
M
+ ··· + = B0 + k
(7) the filter to be stable, all its poles must lie within the unit
z − pM k=1
z − pk circle. There is no restriction on the zero location.
where pk are the poles of X(z), assumed distinct, Ak are
the partial fraction coefficients, and B0 = bM /aM . 8.2 Frequency response estimation
The coefficients Ak may be obtained by multiplying both
sides of (7) by z − pk /z and setting z = pk , that is, In the design of digital filters, it is often necessary to
examine the spectrum of the filter to establish that the
X(z)
Ak = (z − pk ) (8) desired specifications are satisfied. The frequency response
z z=pk of a system can be readily obtained from the z-transform
and given by
8 APPLICATION OF THE Z-TRANSFORM ∞
∞
IN SIGNAL PROCESSING H (ej ωt ) = h[n]z−n |z=ej ωt = h[n]e−j nωt (11)
n=−∞ n=−∞
One of the most important applications of the z-transform
where H (ejωt ) is the frequency response of the system.
in DSP is in the design and analysis of the errors in dig-
In general, H (ejωt ) is complex where the modulus rep-
ital filters, (Oppenheim and Schafer, 1999) and (Proakis
resents its magnitude response and the phase response of
and Manolakis, 1996). It is used extensively to deter-
the system.
mine the coefficients of digital filters, especially Infinite
Impulse Response (IIR) filters (Article 150, Finite Infi-
nite Impulse Response (IIR) Digital Filters, Volume 3).
Another important application of the z-transform in dis- 9 CAUSALITY AND STABILITY
crete filter design is in the representation of digital filter
structures. These are often represented in the form of A causal linear time invariant system is one whose unit
block diagrams, which represent the difference equations impulse response, h[n], satisfies the condition
or equivalently the transfer functions.
h[n] = 0, n<0 (12)
8.1 Pole and zero description in the discrete time Therefore, the impulse response of a causal LTI system
systems is determined from the transfer function by using the right-
sided inverse transforms. Consequently, an LTI system is
In most practical discrete time systems, the transfer func- causal if and only if the ROC of the system function is
tion, H (z), can be expressed in terms of its poles and zeros, outside the maximum pole circle.
as shown by (9). Anticausal signals, on the other hand, have ROC that
is inside the minimum pole circle. If an LTI system is
M
stable, then the impulse response is absolutely summable
bk z−k
−1 −2
b0 + b1 z + b2 z + · · · + bM z −M and the DTFT of the impulse response exists. Hence, the
k=0
X(z) = = M ROC must include the unit circle. Therefore, an LTI system
a0 + a1 z−1 + a2 z−2 + · · · + aM z−M
ak z−k is Bounded-Input Bounded-Output (BIBO) stable if and
k=0
only if the ROC of the system function includes the unit
(9) circle.
Z-transforms 979
Example 4. A discrete time system is characterized by (b) Causal system
the following transfer function For the system to be causal, the ROC of the system
function is outside the maximum pole circle, that is,
3 − 4z−1 |z| > 3. In this case, h[n] is,
H (z) =
1 − 3.5z−1 + 1.5z−2
h[n] = 0.5n u[n] + 2(3)n u[n].
1. Specify the ROC of H (z) and determine h[n] for the
following conditions The system is unstable.
(a) the system is stable
(b) the system is causal (c) Anticausal system
(c) the system is anticausal. For the system to be anticausal, the ROC of the
system function is inside the minimum pole circle,
2. Plot the pole-zero diagram of the system. that is, |z| < 0.5. In this case, the impulse response
3. Plot the frequency response of the system. Assume a
sampling frequency of 500 Hz and a resolution of less
than 1 Hz. 1.5
Solution: 1
1. ROC and h[n]
0.5
Imaginary part
3 − 4z−1 z(3z − 4)
H (z) = −1 −2
= 2 0
1 − 3.5z + 1.5z z − 3.5z + 1.5
−0.5
The poles of H (z) are p1 = 0.5 and p1 = 3. Therefore,
the expansion of the form, (6), is −1
H (z) 3z − 4 A1 A2 −1.5
= = +
z (z − 0.5)(z − 3) z − 0.5 z − 3 −1 −0.5 0 0.5 1 1.5 2 2.5 3
3z − 4 Real part
A1 = · (z − 0.5)
(z − 0.5)(z − 3) z=0.5 Figure 7. Pole-zero diagram.
3 × 0.5 − 4
= =1
0.5 − 3
5
3z − 4
A2 = · (z − 3)
Magnitude (dB)
4
(z − 0.5)(z − 3) z=3
3
3×3−4
= =2 2
3 − 0.5 1
H (z) 1 2
∴ = + 0
0 50 100 150 200
z z − 0.5 z − 3
Frequency (Hz)
1 2
H (z) = −1
+
1 − 0.5z 1 − 3z−1 0
Phase (degrees)
−20
(a) Stable system
For the system to be stable, the ROC must include −40
the unit circle, hence 0.5 < |z| < 3. Consequently, −60
h[n] can be obtained using Table 2 and is given by
−80
0 50 100 150 200
h[n] = 0.5n u[n] − 2(3)n u[−n − 1]. Frequency (Hz)
The system is noncausal. Figure 8. Frequency response.
980 Elements: C – Data Acquisition and Processing Systems
is given by shown in Figure 8 (for more details see Ifeachor and
Jervis (2002)).
h[n] = −0.5n u[−n − 1] − 2(3)n u[−n − 1].
The system is unstable.
REFERENCES
2. Plot of pole-zero diagram
Ifeachor, E. and Jervis, B. (2002) Digital Signal Processing, A
Figure 7 shows the locations of the poles and zeros of the Practical Approach, Prentice Hall, Harlow, UK.
discrete time system in the pole-zero diagram. MATLAB
Oppenheim, A. and Schafer, R. (1999) Discrete-Time Signal
software is used for the plot shown in Figure 7 (for more Processing, 2nd edn, Prentice-Hall, Upper Saddle River, New
details see Ifeachor and Jervis (2002)). Jersy.
3. Plot of frequency response Proakis, J.G. and Manolakis, D.G. (1996) Digital Signal Process-
Figure 8 shows the frequency response of the discrete ing, Principles, Algorithms, and Applications, Prentice Hall,
time system. MATLAB software is used for the plot Upper Saddle River, New Jersy.
143: DFT and FFTs
Gerd Wöstenkühler
Hochschule Harz, Wernigerode, Germany
If the periodic signal is known, the coefficients ak and
1 Fourier Spectrum of Periodic Signals 981 bk can be calculated with an integral over one period of
2 Aliasing Distortion 982 the fundamental frequency according to (2) and (3). Quite
3 Discrete Fourier Transform 983 often t0 is set to zero or −T0 /2.
4 Fast Fourier Transform 984 t0 +T0
2
5 Windowing 985 ak = x(t) cos(kω0 t) dt k ∈ N0 (2)
T0 t0
6 Averaging in Time and Frequency Domain 986 t0 +T0
2
Further Reading 988 bk = x(t) sin(kω0 t) dt k∈N (3)
T0 t0
The harmonic functions in (1) can be substituted according
1 FOURIER SPECTRUM OF PERIODIC to the Euler’s theorem shown in (4) and (5).
SIGNALS
1 jϕ
cos(ϕ) = (e + e−jϕ ) (4)
Each periodic signal with x(t) = x(t + T0 ) can be 2
described as a sum of harmonic functions shown in (1) 1
(see also Article 27, Signals in the Frequency Domain, sin(ϕ) = (e jϕ − e−jϕ ) (5)
j2
Volume 1). The time shift T0 represents the fundamen-
tal frequency ω0 = 2πf0 = 2π/T0 . The frequencies of the Using this description, the time function x(t) can be
harmonic functions are the fundamental frequency and its described with (6).
multiples like kω0 with k ∈ N. The even components of the
∞
periodic signal are specified with the cosine functions and a0 1 1
the odd components with the sine functions. The amplitudes x(t) = + (ak − jbk )e jkω0 t + (ak + jbk )e−jkω0 t
2 k=1
2 2
are the values of the variables ak and bk for the kth fre-
(6)
quency while the component a0 /2 represents the DC-value.
The term 0.5(ak − jbk ) can be substituted by a complex
∞ value ck while k has a positive value. The DC-value a0 /2
a0
x(t) = + [ak cos(kω0 t) + bk sin(kω0 t)] (1) can be substituted by c0 and the term 0.5(ak + jbk ) with a
2 k=1 positive value of k can be substituted by a complex value
c−k . The last part can be rewritten as 0.5(ak − jbk ) while k
A typical example with the components of a square wave has a negative value and so it becomes ck for k < 0. These
function is shown in Figure 1. It shows three harmonic substitutions are shown in (7) to (10).
functions and the sum of these functions and the desired
square wave function, which can be achieved when all 1
harmonic components would be used. ck = (a − jbk ) k ∈ N {1, 2, 3, . . . , +∞} (7)
2 k
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
982 Elements: C – Data Acquisition and Processing Systems
2 For k = 1
For k = 1, 3, 5
1 x(t ) 1
0
x(t )
For −T1 0 T1 T0
−1 k=3 t
2 2
For k = 5
−2 2T1
−1 0 t 1 2 T0
ak
Figure 1. Square wave and 3 harmonics and sum of harmonics.
5 6 7
1 0 1 2 3 k
c0 = a0 (8)
2
T1
1
c−k = (ak + jbk ) k ∈ N {1, 2, 3, . . . , +∞} (9) T0
2
Re{ck}
1
ck = (ak − jbk ) −k ∈ N {−1, −2, −3, . . . , −∞}
2 −7 −6 −5 5 6 7
(10)
−3 −2 −1 0 1 2 3
With this substitution, (1) becomes very compact and can k
be rewritten as (11).
Figure 2. Even pulse train, Fourier coefficients ak and complex
Fourier coefficients ck .
∞
x(t) = ck e jkω0 t k∈Z (11)
kπT1
k=−∞ sin
T T0 T1 kπT1
ck = 1 + j0 = si + j0 (14)
The variables ck can be calculated using (7) or (10) T0 kπT1 T0 T0
with (2) and (3) for the variables ak and bk and the Euler’s T0
theorem to replace the cosine and sine functions. This leads
to (12). If x(t) is a real time function, the coefficients
ck = c∗−k (conjugate complex), based on the even cosine 2 ALIASING DISTORTION
function of the real parts and the odd sine function of the
imaginary parts. For digital signal processing, it is necessary to convert an
analog function into a digital representation. Therefore, the
t0 +T0 analog function is sampled with a constant sampling rate
1
ck = x(t)e−jkω0 t dt k∈Z (12) and each value is converted into a digital representation
T0 t0
(see also Article 138, Sampling Theory Relevant to DAS,
Volume 3). The function of sampling can be regarded as
The values of the Fourier coefficients ak of an even pulse a multiplication of the original analog function with an
train with the amplitude 1 follows the si-function with the impulse train of Dirac impulses as described in (15). Each
maximum amplitude of 2T1 /T0 for k ∈ N according to (13). sampled value xn is the value of the original function
The real values of the complex coefficients ck follows the x(t) at the location nTs of the time shifted Dirac impulses
si-function with the maximum value of T1 /T0 for k ∈ Z δ(t − nTs ).
according to (14). Figure 2 shows a part of an even pulse
train with T1 = T0 /4, the coefficients of ak (bk are all zero) ∞
and the real part of ck (imaginary parts of ck are all zero). xn = x(nTs ) = x(t) × s(t) = x(t) δ(t − nTs ) (15)
n=−∞
kπT1 Such a multiplication in the time domain is equivalent
sin
2T T0 2T1 kπT1 to a convolution in the frequency domain. The spectrum of
ak = 1 = si ; bk = 0 (13)
T0 kπT1 T0 T0 the original signal can be described with the frequencies
T0 up to a maximum frequency or cutoff frequency fc . On the
DFT and FFTs 983
basis of the description of a real-time function with complex
Fourier coefficients, described in the previous section for a
periodic signal, the frequency components cover an area
between −fc to +fc . The frequency of a Dirac impulse x
train has only frequency components at fs = 1/Ts and its
multiples like n fs = n/Ts . On the basis of the convolution,
we can find the complex frequency values of the original
analog signal around all frequencies of the Dirac impulse
train. We get a periodic spectrum as shown in Figure 3.
The complete information of the analog signal can
be found around each frequency component n fs of the
Dirac impulse train. Using an ideal filter, we can select 0 1 2 3 4 5 6 7 8 9 10
a frequency area of, for example, −fc to +fc . In this t
case, we get the original analog function back as shown TS
with the dotted line in Figure 3. This works as long as the
frequency areas of the reproduced original spectrum do not Figure 5. Sampled sine function with alias distortion.
overlap. When the sampling frequency is not high enough
(fs < 2fc ), the spectrums will overlap. This effect is called original signal. This effect is called aliasing effect of the
frequency folding (see also Article 138, Sampling Theory sampling process or aliasing distortion. To avoid this, the
Relevant to DAS, Volume 3 and Article 31, Discrete frequency of the sampling rate has to fulfil (16). To meet
Signal Theory, Volume 1). In this case, the ideal filter this condition, you have to use a sampling rate that is high
would select not only the original frequency components enough or you have to reduce the frequency area of the ana-
but also the overlapping part of the reproduced frequency log signal using analog low pass filters as antialias filters.
components as shown in Figure 4.
The selection of an ideal filter for the frequency compo- fs ≥ 2fc (16)
nents between −fc and +fc causes a deviation within the
reproduced analog signal. The frequencies f of the original A simple example of the aliasing effect is shown in
spectrum appear at the frequency fs − f . If these frequen- Figure 5, where a sine function is sampled with an unsuit-
cies are in the overlapping part, they will be interpreted as able sampling rate. It can be seen that the indicated sample
frequency components with values of fs − f . So you get points suggest a frequency component that is not part of
frequency components with different frequencies out of the the original function.
X( f )
Ideal low-pass filter 3 DISCRETE FOURIER TRANSFORM
Periodic Original Periodic
If we use a digital system to convert the sampled time
extension spectrum extension function into the frequency domain, (12) of the previous
section becomes (17). The complex spectral component X k
−fs −fc 0 fc fs
is the sum of all sampled time values xn multiplied with an
f
arrow within the complex plane adjusted with its number
Figure 3. Spectrum of a sampled signal with fs > 2fc . of periods k to the corresponding time position n within the
used time window of N samples.
Ideal low-pass filter
N−1
X(f ) Xk = xn e−j2πnk/N N ∈N (17)
n=0
Periodic Original Periodic Periodic
extension spectrum extension extension Figure 6 illustrates the function of this equation. On
the left-hand side, you see a sine function, which can be
−fs −fc 0 fc fs 2fs
described with N samples from 0 to N − 1. The dimension
f
of the real part is the horizontal line and the time is the
Figure 4. Spectrum of a sampled signal with fs < 2fc with vertical line. You can imagine the imaginary part, which
frequency folding. would be orthogonal to these two dimensions. The function
984 Elements: C – Data Acquisition and Processing Systems
Z (n) = xn . e−j2πnk/N
xn Im{Z (n)} Im{Z (n)} Im{Z (n)}
1
−1 1
1
−1 Re{Z (n)} Re{Z (n)} Re{Z (n)}
−1 k=0 k=1 k=2
Im{Z (n)} Im{Z (n)} Im{Z (n)}
Re{Z (n)} Re{Z (n)} Re{Z (n)}
N
n
k=3 k=4 k=5
Figure 6. Graphical interpretation of the DFT.
e−j2πnk/N twists the time function with k revolutions at with the number of periods k and the used time position
the end where the N th sample would be. The other six n within the used time window of N samples according
diagrams show the projection of this twisted sine function to (18).
into the complex plane where the real part is the horizontal 1
N−1
line and the imaginary part is the vertical line. If k = 0, xn = X e j2πnk/N N ∈N (18)
N k=0 k
the function is not twisted and the projection describes
a horizontal line as shown in the top left diagram. The
Figure 7 shows the inverse Fourier transform with the
sum X 0 of all components is zero, which means that the
values determined in Figure 6. In the complex plane, the
sine function has no DC component. If k = 1, the resulting
two components with the amplitude N /2 are shown. For
projection is shown in the top middle diagram. It describes
the time position n = 0, the sum of both components
a symmetric function and the sum X1 of all components is
is zero and this is the value of x0 . The value of xN/16
zero, which means that the analyzed sine function has no
where n = N /16 (n ∈ N), the component of X2 will rotate
component with 1 period in the time window. If k = 2, the
counterclockwise with 45◦ or π/4 and the component of
resulting projection is shown in the top right diagram. It
XN−2 will rotate clockwise with 45◦ or π/4. The sum of
describes a circle around −j0.5 with two revolutions and
both components yields xN/16 = sin(π/4) ≈ 0.707. On the
the sum X 2 of all components is −jN /2, which means that
basis of the symmetric frequency components of a real time
the analyzed sine function has a component with 2 periods
function, all values xn have an imaginary part of zero.
in the time window. Nevertheless, all other values of k
up to N − 1 will produce a sum with the value of zero
except k = N − 2 where the sum has the value +jN /2. 4 FAST FOURIER TRANSFORM
Using (17), the sine function with two periods in the time
window is transformed into a description with X 2 = −jN /2 A central operation of the Fourier transform and the
and XN−2 = +jN /2. inverse Fourier transform is a complex multiplication with
There exist slightly different equations for this trans- sine and cosine terms. For the calculation of all values,
formation. Some have used average values, which is the there are around N 2 complex multiplications and additions
sum divided by the number of elements and so the val- necessary. For a large number N of sample points, the
ues of the spectrum become independent of the number calculation time of the frequency components becomes
of samples. In this case, all following equations and scal- quite long and so several attempts have been made to reduce
ing might be different based on this factor. The equations the amount of necessary computational time for solving
in this article have used the more often described equa- this equation.
tions of the discrete Fourier transform where the factor The used factor of e2πnk/N is a revolving arrow around
1/N is used in the inverse Fourier transform. This is the the unit circle in the complex plane. The basic step size
sum of all complex frequency components X k adjusted is e2π/N . If N has the value of 2 to the power of ν
DFT and FFTs 985
Im{Xk} where ν ∈ N, you get the same positions of this arrow at
different combinations of nk /N . Figure 8 shows this effect
N
e−jπ/4 for N = 16 samples for 4 different values of k.
XN −2
A very efficient algorithm using this effect was published
Re{Xk} by Cooley and Tukey in 1965. Their algorithm needs about
N log2 (N ) multiplications and additions. This is the reason
−N N
why this algorithm is N/ log2 (N ) faster. Table 1 shows this
X2
ejπ/4
factor for different numbers of samples. For N = 16 384
−N
samples, the relation of the time difference is nearly the
xn
same as 1 h for the discrete Fourier transform (DFT) down
to 3 s for the fast Fourier transform (FFT). On the basis of
−1 N 1
16
this, it is highly recommended to use N = 2ν with ν ∈ N
samples for data acquisition.
N
4
5 WINDOWING
One problem of data acquisition is to find the right size
N of the time window with N samples to truncate the
2 sampled function. It is not a problem if it covers one
period or its multiples of a periodic signal, but that is
usually not the case. The used segment of the sampled
Table 1. Acceleration of the FFT algorithm compared
with the DFT algorithm.
N = 2ν ν Acceleration (tDFT /tFFT ≈ N/ν)
N 16 4 4
256 8 32
n 1024 10 102.4
16 384 14 1170.3
Figure 7. Graphical interpretation of the inverse DFT.
N = 16 n = 2, 10 N = 16
n=5 n=4 n=3 k=1 k=2
n=6 n=2 n = 3, 11 n = 1, 9
n=7 n=1
n=8 n=0 n = 4, 12 n = 0, 8
n=9 n = 15
n = 10 n = 14 n = 5, 13 n = 7, 15
n = 11 n = 12 n = 13 n = 6, 14
N = 16 N = 16
n = 12 n = 1 k=3 n = 1, 5, 9, 13
n=7 k=4
n=2 n=6
n = 13 n = 11
n=8 n=0 n = 2, 6, 10, 14 n = 0, 4, 8, 12
n=3 n=5
n = 14 n = 10
n=9 n = 15
n=4 n = 3, 7, 11, 15
Figure 8. Arrow positions around the unit circle used for the FFT algorithm.
986 Elements: C – Data Acquisition and Processing Systems
time signal can be regarded as a multiplication of the x gH
original analog signal with a single pulse selecting the time 1 1
window. The spectrum of the samples is the convolution
of the spectrum of the time function with the spectrum
of the single pulse (time window or truncation window). 0 n 0 n
In this case, the spectrum is continuous and the DFT
(or FFT) algorithm determines values of the continuous
spectrum at the frequencies f0 = 1/T0 and its multiples −1 −1
up to (N − 1)f0 . The behavior of the continuous function
between these points is unknown. I would prefer a different
view. If we use the discrete spectrum of the DFT (mean xH
values are zero), we get a description of a periodic 1
signal in the time domain, where the N samples cover
one period. In fact, the continuous signal of the analog
function between these samples is unknown, but better 0
n
to imagine. If this periodic signal describes the signal
you want to analyze, the complex spectral values can be
used. The problem with the periodic extension is to get a −1
continuous periodic signal at the edges of the used period.
Figure 9(a) shows two sampled sine waves of 20 periods Figure 10. Function of the Hanning weighting window.
and Figure 9(b) shows 20.5 periods in the time window.
If we extend this period, we get the original signal in of both spectrums (see also Article 31, Discrete Signal
the case in Figure 9(a) and based on this we get a very Theory, Volume 1). An often used weighting function
good spectral description indicating 20 periods in the time is the Hanning window, where each sampled value is
window. If we extend the period in Figure 9(b), we get weighted with a value according to (19).
a nonharmonic connection between these periods. This
produces a distribution or leakage distortion of the spectral 1 n × 2π
gHn = 1 − cos
line as shown in the amplitude spectrum in Figure 9(b). 2 N
The reason is the nonharmonic connection of the periodic n ∈ {0, 1, 2, 3, . . . , N − 1} (19)
extension.
There have been several attempts at creating a weighting Figure 10 shows the behavior in the time domain.
function or truncation window function of the used time In this case, the periodic extended signal of the sampled
window. Each can be regarded as a multiplication of this time window has got an amplitude modulation with the
window with the time signal. The result is the convolution frequency f0 . This creates side bands if we have only
multiple periods in our time segment of N samples as
x x shown in Figure 11(a). If we have fractions of such a signal
1 1 as shown on the right-hand side in Figure 10, we get only a
small leakage distortion of the original spectrum at 20.5 f0
of the spectrum. It is much easier to analyze such a spectral
0 n 0 n description.
−1 −1
X DFT X DFT 6 AVERAGING IN TIME AND
N N
2 2 FREQUENCY DOMAIN
N N Quite often the signal we want to analyze is covered in
4 4
noise. This means that we have a signal, for example, a
square wave function, we want to analyze and disturbing
0 0 noise. One data acquisition process with N samples pro-
0 10 20 30 40 50 60 k 0 10 20 30 40 50 60 k
(a) (b) vides a spectrum, which does not give us the necessary
accuracy. If the process is continuous, and if we can mea-
Figure 9. Amplitude spectrums of sampled sine functions. sure several time windows with the same conditions, we
DFT and FFTs 987
X X
N N
4 4
0 0
(a) 0 10 20 30 40 50 60 k (b) 0 10 20 30 40 50 60 k
Figure 11. Amplitude spectrums of sampled sine functions with Hanning weighting window.
If you cannot use a synchronization with the square wave
0.4 signal, the phase of all components of our square wave
function varies and we can only average the amplitude spec-
trum of all measured time windows as described in (20). If
0.2
we use only the amplitude spectrum |Xk |, we cannot apply
the inverse DFT to regenerate the time signal based on
x 0.0 the missing phase of the frequency components. Using this
average function, we get the spectrum of the signal added
−0.2 with the average value of the noise. The standard devia-
tion of the noise components around its mean value will
be reduced. Such a spectrum is shown in Figure 13 with
−0.4
an average over 100 time windows (M = 100). The fre-
0 200 400 600 800 1024 quency components of the square wave signal can be seen
(a) n easily.
1
M
|X k | = |X | k ∈ {0, 1, 2, 3, . . . , N − 1} (20)
60 M m=1 km
If we can get a synchronized time window for data
40 acquisition, for example, with a separate trigger signal,
Xk the phase of the spectral components of our signal will
not change in the spectrum and we can apply the average
20
60
0
0 10 20 30 40 50 60 70 80 90 100
(b) k
40
Figure 12. Square wave signal and noise in the time and
frequency domain. Xk
can use an average algorithm. Let us assume that the signal 20
has a stable amplitude and frequency, and that the noise has
a spectrum where the values in each time segment varies
around a mean value. The behavior of such spectral com- 0
ponents can be described with a mean value and a standard 0 10 20 30 40 50 60 70 80 90 100
k
deviation. Figure 12 shows a measured square wave signal
of 5 periods with noise sampled with 1024 samples and the Figure 13. Averaged amplitude spectrum of a square wave signal
spectrum up to k = 100. and noise.
988 Elements: C – Data Acquisition and Processing Systems
0.2 algorithm to the time signal. The phase of the noise signal
is not synchronized and on the basis of this effect the
value of those spectral components will be reduced. The
0.1 average algorithm can be applied to the time signal (21)
or to the complex spectral values (22). Figure 14 shows
a time signal and its spectrum of 100 time windows
x 0.0 (M = 100) as presented in Figure 12. The remaining noise
becomes much smaller and the frequency components
of the square wave signal can be analyzed to higher
−0.1 frequency components compared with the spectrum shown
in Figure 13.
−0.2
1
M
0 200 400 600 800 1024
(a) n xn = x n ∈ {0, 1, 2, 3, . . . , N − 1} (21)
M m=1 nm
1
M
60 Xk = X k ∈ {0, 1, 2, 3, . . . , N – 1} (22)
M m=1 km
40
Xk FURTHER READING
Romberg, T.M., Black, J.L. and Ledwidge, T.J. (1996) Signal
20
Processing for Industrial Diagnostics, John Wiley & Sons,
Chichester.
Sydenham, P.H. (1983) Handbook of Measurement Science,
0 Volume 1 Theoretical Fundamentals, John Wiley & Sons,
0 10 20 30 40 50 60 70 80 90 100 Chichester.
(b) k
Figure 14. Averaged square wave signal and noise in the time
and frequency domain.
144: DSP Chip Sets
Iain Paterson-Stephens
University of Derby, Derby, UK
Without really going into the mathematics of this relation-
1 Elements of Digital Signal Processing (DSP) 989 ship, it can be seen that the process of digital filtering a
2 Basic DSP Functions 990 signal requires a number of addition , subtraction, and
3 Implementation of DSP Functions in multiplication a(m)y(i − m) operations. In the relationship,
Microelectronic Form 991 the values b(k) and a(m) represent two tables of filter
coefficients stored in memory. These will be multiplied by
4 Programmable Chip Sets 992
tables of past input x(i − k) and output y(i − m) samples
5 Application Specific Integrated Circuits
respectively. Each time a new input sample arrives it will
(ASICS) 993
be placed on the top of the input sample table and the old-
6 Introduction to Gate Arrays 993
est sample removed. Similarly, new output samples will be
7 Processors in General 994 placed onto the top of the output sample table and the old-
8 Measurement System Chip Sets 995 est removed; hence, both data tables always have a fixed
References 996 length for any given filter design. Figure 1 depicts the basic
Further Reading 996 arrangement for a single table of coefficients and past out-
put samples.
The process outlined above indicates the features that will
be required on any general purpose DSP device. It should be
1 ELEMENTS OF DIGITAL SIGNAL able to efficiently access and manage its memory areas and
PROCESSING (DSP) simultaneously handle the flow of data into and out of its
data ports. For real-time operation, the DSP device will also
Most Digital Signal Processing (DSP) devices incorporate require a fast arithmetic unit, special memory addressing
similar functions and features, the reason for this is clear modes, and good support for interfacing to external I/O.
when we consider the fundamental algorithms for which Another example of a common DSP algorithm is the
they are designed to process. The majority of DSP algo- Fast Fourier Transform (FFT), which is essentially an effi-
rithms currently in widespread use fall into the following cient method for calculating the Discrete Fourier Transform
categories: Filtering of signals (convolution), Transforma-
(DFT). The FFT was first described in 1963 by Cooley and
tion of signals (e.g. time to frequency domain) or Correla-
Tukey, and has become widely used for obtaining the fre-
tion (comparison of signals), (Ifeachor and Jervis, 2001).
quency domain representation of time domain signals. The
Considering the first of these terms; ‘Digital filtering’,
mathematical relationship used to describe the DFT is given
generally refers to a numerical operation governed by the
in (2).
following relationship
K
M
N−1 j 2πnm
y(i) = b(k)x(i − k) − a(m)y(i − m) (1) X(K) = x(n)WNnK , where WNnK = e− N (2)
k=0 m=1 n=0
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
990 Elements: C – Data Acquisition and Processing Systems
M
errors will not occur under normal operation. Most DSP
∑a (m) × y (i − m) devices incorporate a number of arithmetic features that
m=1
Last ease this process such as saturation arithmetic and over-
output flow detection.
sample
a1 yi
a2 Multiply tables & Add yi − 1 2 BASIC DSP FUNCTIONS
a3 yi − 2
Following the previous discussion about the nature of
typical DSP algorithms, it will be apparent that a number
of characteristics are desirable for any DSP device, which
is used to implement real-time DSP algorithms. Some of
these fundamental DSP device functions are summarized
aM − 1 yi − M + 1 in the following section.
aM yi − M
2.1 The ‘multiply accumulate’ operation (MAC)
Oldest
sample
discarded
Each of the fundamental signal processing algorithms dis-
cussed in the previous section have operated upon data
Figure 1. Filter coefficients and samples. samples by calculating a running sum (accumulation) of a
series of products. A DSP device must therefore be able to
It can be seen that the relationship for the DFT again calls accumulate a series of multiplication results, this is known
for storage of a list of coefficients WNnK (often referred to as the Multiply Accumulate, or MAC operation. The MAC
as Twiddle factors) in memory along with a list of sample is one of the most fundamental and distinct features that
data. The frequency resolution of the DFT is determined should be found on any dedicated DSP device.
by the number of sample points used in the calculation
and hence this directly determines the number of sample
points/twiddle factor values that need to be stored in mem- 2.2 Efficient data access–specialized memory
ory. The requirement for efficient data access and memory addressing
management is therefore a requirement for this DSP appli-
cation as are a fast arithmetic unit and specialist memory The DSP must be able to access its memory areas as
addressing modes. efficiently as possible so that the processor does not waste
Correlation, Convolution, and Power Spectral Density are time waiting for data to be fetched. In the case of the MAC
all examples of functions related to the Discrete Fourier operation, which typically runs repeatedly very quickly
Transform. These common signal processing functions can for short bursts of time, it is important that an efficient
all be executed in the digital domain using schemes similar MAC operation is not held up while waiting for data
to that shown in the equation for the DFT, (2). to be fetched from memory. Typically, DSP devices will
In a typical system using a DSP device, the sampled provide a number of different methods for fetching data
data will arrive with a regular period at the input to the and determining its address within memory. The FFT, for
DSP. On the arrival of each data sample, the DSP will example, makes use of so-called bit-reversed addressing.
carry out the calculations described by the algorithm and The output of an FFT algorithm is usually ‘bit reversed’,
the resulting data will be presented to the DSP output, that is, the data points get out of sequence in a predictable
again at a regular period. Clearly, the DSP must be quick way as a result of the FFT operation. This is rectified by
enough to complete algorithm processing before the next accessing the data points using a bit-reversed addressing
new data sample arrives. This characteristic is essential for scheme that most DSP devices provide.
any device that will be used successfully to implement DSP Another DSP specific addressing mode is the circular
algorithms in real time. buffer, depicted in Figure 2, in which an area of memory
It is apparent, from the expressions for the DFT and dig- can be made to appear as a continuous loop. Typically, data
ital filtering, that the DSP device must carry out numerous pointers are used to indicate the current data input location.
arithmetic operations. In addition, extra steps must usually Each time a data transfer takes place, the pointer is shifted
be taken by the DSP programmer to ensure that calculation one place further round the buffer.
DSP Chip Sets 991
Data element Data element though the processor clock is only 200 MHz. In this case,
‘out’ pointer ‘in’ pointer the DSP executes eight parallel operations per instruction
Direction of cycle. Superscalar operation is utilized heavily in most DSP
x (n − m) x (n) rotation devices, an example is the common MAC operation in
x (n
which two instructions are executed in parallel, that is,
‘m ’ sample the Multiply and Accumulate. Usually, the MAC is run
− 1)
period
Delay time in parallel with other operations such as data fetches and
)
−2
(T ) address pointer updates all of which operate simultaneously
x(n
in the same instruction cycle. The result is that hundreds
x (n − 4) x (n − 3) of millions of operations can (in principle) be performed
per second.
Figure 2. DSP circular buffer.
2.6 Hardware repeat loop
2.3 Deterministic performance
Programmable DSP devices have instruction sets that con-
When a DSP algorithm is to be performed in real time tain unique features for DSP algorithm work. One particular
using a DSP device, it is important that the programmer is feature, which is not generally available on a micropro-
able to predict the amount of processor time the function cessor, is hardware support for instruction looping. This
will take to complete, that is, DSP processing must be recognizes that DSP algorithms consist of compact code
deterministic. If the processor operation is not deterministic, segments that need to run very efficiently at full speed.
then problems can arise when large amounts of data are For example, in a filtering application, the MAC instruc-
being processed in real time. Real-time performance is tion will be run repeatedly in short bursts each time oper-
a term used to describe many DSP systems; it can be ating on different data sets. Without hardware looping, this
defined as process would be very inefficient as a loop counter variable
would need to be maintained. Using a DSP device equipped
Completing the required processing task quickly enough with a zero overhead loop facility, the loop counter is main-
so that system performance can continue without being tained in hardware, easing the burden on the processor
hindered by the processing overhead. core. Also, the instruction is fetched from memory once
and run many times thus giving a further potential speed
increase.
2.4 Extensive interfacing capabilities
Real-time DSP systems must be able to process signals 3 IMPLEMENTATION OF DSP
that come from real-world sources such as data from FUNCTIONS IN MICROELECTRONIC
an analog to a digital converter. The DSP device must FORM
therefore incorporate extensive interfacing capabilities so
that transducers, data acquisition devices, memory or a It is possible to perform DSP functions in real time using
system host can be connected efficiently. Most DSP devices a standard microprocessor as opposed to a dedicated DSP
provide on-chip interface components that require minimal device. In fact, many of the techniques pioneered on early
external circuitry and only limited intervention by the DSP devices have now been adopted for use on general
main core of the DSP processor. The benefit of this is purpose microprocessors. In addition, many microprocessor
that processor time is not wasted interacting with slow and microcontroller manufacturers quote the ability of their
peripherals device to perform DSP functions, after all, there is nothing
sacred about a DSP function – it is merely a mathemati-
cal relationship. The issue is whether the device is capable
2.5 Superscalar operation of performing the function quickly enough for the intended
application so as to give real-time performance. Typically, a
Superscalar mode of operation is a term often used in standard microprocessor will not be able to efficiently per-
relation to DSP devices. Superscalar operation means that form DSP functions such as a MAC operation, simultaneous
the device can execute multiple instructions in parallel. access of two data tables in memory or highly efficient bit-
This explains why a DSP can be rated at 1600 MIPS even reversal addressing. In such a case, it would be necessary
992 Elements: C – Data Acquisition and Processing Systems
for the programmer to devise some sort of software scheme Certain manufacturers have gained a reputation for their
to produce these functions, although this will never be as devices within particular industries, usually because they
quick as the dedicated DSP hardware equivalent. have incorporated some additional targeted features. For
Take the humble multiply operation as an example; all example, Analog Devices and Motorola DSP devices have
general purpose DSP devices are provided with dedicated gained wide acceptance in the professional audio prod-
hardware to perform a multiply that typically provides cal- uct market, whereas devices from Texas Instruments are
culation results within one processor cycle. In comparison, used extensively within telecoms applications. In the case
a typical microprocessor carries out its multiply operation of the pro-audio application, this has possibly been due
by a binary long multiplication process. When it encounters to the development of low-cost 24- and 32-bit proces-
a multiply instruction, an internal sequence of operations, sors within the product range of Analog Devices and
called a microcode, is invoked. This microcode performs Motorola.
the multiplication as a sequence of shifts and adds on suc- All programmable DSP chip sets have a range of soft-
cessive clock cycles until the result is complete. Since the ware development options and tools available from the
microcode has many steps, the operation requires many manufacturer. Often, DSP code is written using the C
clock cycles to perform. The resulting overhead for mul- or C++ programming language, which helps to ensure
tiplication on a microprocessor could be approximately portability between different DSP platforms, see Figure 3.
80 processor clock cycles to perform a 16-bit multiplica- C/C++ is often the first choice for developing signal
tion. Both devices can perform a multiply; its just that the processing functions; however, device specific assembly
DSP device is generally quicker (Bateman and Paterson- language is commonly used for low level operations that
Stephens, 2003). Of course, money is also a factor; if you deal directly with I/O devices and other hardware ele-
are prepared to pay for an expensive microprocessor, then ments. Assembly language is also commonly used for
you are likely to get more hardware features such as a time-critical operations in which optimum performance
dedicated multiply unit. cannot be achieved using C code, for example, proces-
sor interrupts associated with signal I/O are typically
written using assembly language (Bateman and Paterson-
4 PROGRAMMABLE CHIP SETS Stephens, 2003).
Unfortunately, no standard DSP programming language
The programmable, or general purpose DSP device, typi- has yet emerged so it is necessary for the designers to famil-
cally encompasses all, and more, of the features that have iarize themselves in great detail with the specifics of each
been presented in the preceding section. The programmable new device that is used. Although many DSP manufacturers
DSP may be reconfigured to suit an extremely wide range produce a family of different devices covering fixed/floating
of different applications. Manufacturers of these devices point, different performance levels (MIPS) and I/O capabil-
usually provide software development tools and evalua- ities – often each device within a family will have its own
tion/development platforms that help engineers to design individual quirks, which prevent easy movement of DSP
and develop new applications. Often manufacturers will code from one device to another.
provide extensive customer support, design services, access
to third party vendors, and ready coded software libraries
downloadable from the Internet. It has been the case that
general purpose programmable DSP devices have been used
in small volume applications where only tens of thousands,
or fewer units are expected to be sold. DSP devices are
produced by a number of manufacturers including, in no
particular order, those identified in the following list by:
• Lucent Technologies
• Texas Instruments
• Philips
• Analog Devices
• Motorola
• ST Microelectronics
• Harris Semiconductors Figure 3. C/C++ program development for the TI C6x DSP
• Microchip processor.
DSP Chip Sets 993
5 APPLICATION SPECIFIC INTEGRATED designs. FPGAs in particular offer a number of potential
CIRCUITS (ASICS) advantages over the traditional programmable DSP device
(Goslin, 1997). The FPGA offers the opportunity to opti-
This type of device is expensive to develop and produc- mize silicon use to a specific application in much the same
tion is only viable for high volume applications where way as an ASIC. Additional system functions that would
sales targets are in the hundreds of thousands of units. conventionally exist outside the DSP device can poten-
Application Specific Integrated Circuit (ASIC) based DSP tially be brought onto the FPGA and hence minimization
designs are expensive to develop because the level of sim- of circuit board space can be achieved. A Gate Array is
ulation and testing required before going to fabrication is essentially a grid of unconnected logic cells that can be
usually more rigorous – obviously, with a programmable linked together as required, by the use of an appropriate
device it is possible to make (software) changes to the tool, into a functioning circuit. The process of configuring
design after the system has been manufactured, whereas an the circuitry usually fuses a mesh of links between the logic
ASIC design, once fabricated, is fixed – mistakes are con- cells. FPGA devices are available in a range of sizes up to
sequently very expensive. Development tools sets and test complex devices containing tens of thousands of logic ele-
jigs for ASIC are also somewhat specialized and are conse- ments. One of the advantages of the FPGA design route is
quently expensive. The ASIC option does however provide that the configuration of circuitry can be very application
a cost effective route if the sales volumes are high since specific, for example, if the designer knows that the appli-
an ASIC can be developed to provide the specific functions cation will only require 13-bit arithmetic, then that is all
and features required for an application and as such excess that needs to be designed. Also, greater use of parallelism
cost for unused features can be minimized (Bateman and can be achieved with an FPGA than a programmable DSP.
Paterson-Stephens, 2003). The trend for general purpose For example, if a very fast convolution (filtering) processor
DSP devices on the other hand is toward the integration of is required that contains say 16 filter taps, the filter would
large amounts of memory, extensive, and varied I/O facil- require 16 multiplies and 16 adds (MACS) per new input
ities and numerous other features. While this is useful for sample. In a programmable DSP application, a fast single
the general case, it is likely that many of these features will MAC unit will be used within a hardware loop to perform
not be required in a large number of applications.
the filter function in just over 16 instruction cycles. For
It is interesting that manufacturers of some general pur-
the FPGA approach, it would be possible to incorporate
pose DSP devices are now providing the option for user
16 independent MAC units within hardware design, which
specified device features so that a device can be made up
can all operate simultaneously within one system cycle.
from a specific processing core and selected on-chip periph-
The greater potential use of parallelism in the hardware
erals. ASIC devices also offer good reliability because they
design can therefore offer greater performance than that is
usually only perform the specific tasks for which they have
achievable on a general purpose programmable DSP device
been designed. Further more expensive silicon area can be
(Goslin, 1997).
minimized and because many different system functions
Development and production costs are significantly lower
can be brought onto the ASIC itself, the final system design
than the ASIC route and for many engineers the devel-
can be optimized in terms of the circuit board (PCB) space
opment process is more familiar compared to the use of
required. One further issue relating to the minimization of
circuit functions in silicon is that ASIC-DSP designs used a programmable DSP device. Because the design of a
for low power applications, such as mobile telephony, for DSP function within an FPGA comes down to a descrip-
example, can be optimally low power for the specific appli- tion of hardware, for example, multipliers, delays and so
cation, that is, no excess or unused circuitry means that on, many engineers are already familiar with the devel-
power consumption can be kept to the minimum. opment process and tools. Typically, a graphical CAD-
One example of an ASIC device is a surround sound based approach to design can be used or the use of
decoder chip used for consumer audio applications devel- an appropriate Hardware Descriptor Language (HDL).
oped and produced by consumer product manufacturers All potential DSP functions can be defined in terms of
such as Yamaha or Sony. Clearly, in this type of appli- a description of hardware. Many manufacturers provide
cation worldwide distribution of large numbers of units downloadable HDL libraries that define typical DSP func-
is predicted. tions such as Convolution, Correlation, and Frequency
Transforms. Also, a number of programmable DSP cores
from manufacturers such as Texas Instruments and Ana-
6 INTRODUCTION TO GATE ARRAYS
log Devices can be purchased in HDL language form for
Gate Arrays and specifically Field Programmable Gate integration into an FPGA design (Bateman and Paterson-
Arrays (FPGA) are finding many applications in new DSP Stephens, 2003).
994 Elements: C – Data Acquisition and Processing Systems
7 PROCESSORS IN GENERAL by independent processor buses. An example of this is
DSP devices from Motorola, which make use of a Dual-
There is an important device architecture difference between Harvard Architecture with two independent data spaces,
a classic microprocessor compared to the DSP device, X and Y, as well as a program memory space. Using
that is, the overall structure of the memory itself within Harvard Architecture enables program and data memory
the device. Most microprocessors are designed around the to be accessed simultaneously, giving a speed advantage
Von-Neuman architecture, shown in Figure 4, in which the over a conventional microprocessor architecture (Bateman
program instructions and data items share the same memory and Paterson-Stephens, 2003). DSP devices usually have
space and hence are accessed via the same address and data sophisticated Address Generation Units (AGUs), which
buses. This strategy can be slow because the CPU needs to generate data addresses using specific addressing modes.
fetch each instruction (op-code) to be executed followed by Modulo addressing is unique to DSP devices, and is use-
fetching any associated data (operands) before it can start ful for automatically implementing circular buffer memory
processing. Each of these tasks takes up precious time, that constructs and bit-reversed addressing accesses with no
is, processor clock cycles. software overhead. The ability to auto increment and decre-
Most DSP devices use a style of processor architecture ment values in the address registers is also useful. The term
known as the Harvard Architecture (shown in Figure 5). software overhead refers to an additional code, which would
Using Harvard Architecture the program and data items need to be in place to perform a function if the specific
are stored in separate memory spaces, and are accessed feature was not available.
Von-Neuman Internal address bus
Architecture Computer
ALU
register file Instruction
Shared
decode
program
and CPU I/O devices and data
control
memory
ALU
Internal data bus
Figure 4. Von-Neuman Architecture Computer.
Address X address bus
Harvard
genera-
Architecture Y address bus
tion
Computer Program address bus
unit
ALU
X data memory
register file Instruction r
Y data memory
Program memory
decode
and DSP I/O devices
control
ALU
X data bus
Y data bus
Program data bus
Figure 5. Harvard Architecture Computer.
DSP Chip Sets 995
Clock
cycle 1 2 3 4 5 6 7 8 9
Fetch F1 F2 F3
Decode D1 D2 D3
Execute E1 E2 E3
Instruction cycle one Instruction cycle two Instruction cycle three
Figure 6. Nonpipelined operation.
Clock
cycle 1 2 3 4 5 6 7 8 9
Fetch F1 F2 F3 F4 F5 Etc.
Decode D1 D2 D3 D4 D5 Etc.
Execute E1 E2 E3 E4 E5 Etc.
.
Instruction cycle one Instruction cycle four
Instruction cycle two Instruction cycle five
Instruction cycle three Etc.
Figure 7. Pipelined operation of DSP instructions.
DSP and Reduced Instruction Set Computer (RISC) hooked up with I/O hardware for simple no fuss sys-
architectures aim to deliver results from the processing core tem development (TI-SPRA477, 1998). A standard high-
on every instruction cycle. However, in practice the pro- speed serial interface known as I2 S has been adopted by
cessing of an instruction goes through a number of phases. most manufacturers, which allows a simple 3- or 4-wire
The instruction must first be fetched from memory along interface between Analog-to-Digital (AD) and Digital-to-
with any associated operands, then the processing core will Analog (DA) converters and the DSP device (Bateman and
need to decode or interpret the instruction before finally Paterson-Stephens, 2003). Usually, AD and DA devices can
dispatching it for execution. This scheme is shown dia- be obtained for most applications that incorporate all the
grammatically in Figure 6. necessary timing logic and anti-aliasing filters. Furthermore,
The objective of pipelining is to overlap the different it is common practice to make use of a CODEC device that
phases so that instructions can be fetched prior to the com- combines all the necessary analog input and output hard-
pletion of other instructions in the pipeline. Figure 7 shows ware, sampling logic, serial interface logic, filters, and so
a pipelined arrangement in which the different phases of on. These devices offer a very simple and cost-effective way
the three instructions previously shown in Figure 6 are now of integrating a complete measurement system into just a
allowed to overlap. It can be seen that results start coming handful of chips. The word CODEC is used to describe a
out of the core on every clock cycle. This is after a delay of device that can be used to COde and DECode data between
two clock cycles while the first instruction ripples through different formats. In the context we are considering here,
the pipeline. The ability to pipeline instructions and data the coding is from analog to digital form and the decod-
and the effectiveness of the pipeline comes largely from ing is from digital back to analog. The term CODEC can
the fact that the DSP device has a very efficient memory also be used to describe any coder/decoder device or algo-
architecture (Harvard Architecture). Many standard micro- rithm, for example, an algorithm that converts between a
processors attempt to pipeline instructions and data but it linear and compressed data format and back again could be
is doubtful that the pipeline will be as effective on a Von-
described as a CODEC. The block diagram of a standard
Neuman compared to the Harvard Architecture (Bateman
interface CODEC device is shown in Figure 8.
and Paterson-Stephens, 2003).
The CODEC shown in Figure 8 incorporates a paral-
lel port that is generally used as a low speed connec-
8 MEASUREMENT SYSTEM CHIP SETS tion through which the DSP device can send initialization
information and monitor various aspects of the CODECs
Many manufacturers produce system chip sets incorpo- operation. The sample data is usually, though not necessar-
rating DSP hardware suitable for measurement applica- ily, sent via the CODECs serial port, which is connected
tions. In fact, most general purpose DSP devices are well directly to one of the DSPs on-chip serial interfaces. The
suited to this type of application. Standard interface meth- serial connection uses a four-wire bus comprising a data
ods have been adopted that allow DSP devices to be clock, frame sync, data in, and data out lines (Jordan, 1996).
996 Elements: C – Data Acquisition and Processing Systems
SDout SDin CLK Fsync
INT Seial
Serialport
port Dither
CS interface generator
EN
R/W
L_mic
compressor
16-sample
Parallel port interface
Filters R_mic
control
A-law
u-law
FIFO
MUX
& 16-bit L_line
block
Gain
A to D R_line
Data
converters L_aux
R_aux
L_out
16-sample
Address
expander
Filters
expander
control
A-law
w
u-law
FIFO
w
& 16-bit
block
Gain
mono
A-la
u-la
D to A
converters
R_out
Oscillators Gain control Timer
XTAL1 I/O XTAL2 I/O
Figure 8. Standard interface CODEC for DSP-based measurement applications.
The CODEC requires only a minimal amount of external TI-SPRA477 (1998) TMS320C6000 McBSP Interface to the
support components such as basic I/O buffering and filters, CS4231A Multimedia CODEC, Texas Instruments.
power smoothing, and a crystal clock. The I/O buffering Jordan, M. (1996) Hardware and Software Interface Issues for
can be made using standard op-amp designs and the filters DSPs and Serial Audio CODECs – Application Notes, Crystal
only need to provide AC coupling that is, an RC filter will Semiconductor Corporation.
suffice. The crystal clocks are used to provide the sample Ifeachor, E.C. and Jervis, B.W. (2001) Digital Signal Process-
rate for the device and this can be set to operate at any ing – A Practical Approach, ISBN 0 201 54413X, Addison-
Wesley, UK.
one of a number of standard rates (Crystal Semiconductor
Corporation,1994). Because the CODEC incorporates its Crystal Semiconductor Corporation (1994) CS4231A Multimedia
CODEC Applications Guide, USA.
own sample rate generator, it is not necessary for the DSP
to provide this from its internal clock source. Instead, it is Goslin, R.G. (1997) A Guide to Using FPGAs for Application
Specific Digital Signal Processor Performance, Xilinx Inc,
common practice in this type of application for the CODEC USA.
serial port to be configured such that the CODEC generates
the data clock and frame sync signals and the DSP simply
synchronizes itself to the CODEC. FURTHER READING
Xilinx Logicore (2002) MAC FIR V3.0 Data Sheet, Xilinx Inc,
REFERENCES USA.
Bateman, A. and Paterson-Stephens, I. (2003) The DSP Hand-
book – Applications, Design and Implementation, ISBN: 0-201-
39851-6, Prentice Hall, UK.
145: DSP Tools
Iain Paterson-Stephens
University of Derby, Derby, UK
design has been translated, it is compiled and tested on the
1 Tool Sets in Support of DSP 997 target device.
2 Overview of Selected Small- to-large The testing and verification process has also advanced
Mainline DSP Tools 997 somewhat in recent years so it is now possible to moni-
tor, in real time, the operations taking place inside the DSP
3 Choice of DSP Tools for an Application 998
device. This allows software problems to be identified and
4 Introducing DSP Tools in an Organization 999 resolved as quickly as possible so that the overall design is
5 Limitations of DSP Tools 1001 not delayed (Bateman and Paterson-Stephens, 2003). Stan-
References 1002 dard libraries for software development are also emerging,
which means that the design can be built using standard sig-
nal processing plug-in functions; for example, in the design
of a real-time filter, the engineer can select from a library of
1 TOOL SETS IN SUPPORT OF DSP standard filter algorithms, add the necessary coefficients to
define the filter response, and then finally glue the applica-
Design for DSP applications, as with any design, must go tion together with a small amount of extra software that will
through a number of stages and iterations before the final manage data flow and other aspects of DSP ‘housekeeping.’
product can be released. For DSP designs, a number of key All this sounds very attractive and in principle the tools are
stages will need to be considered; at a basic level, these can all here and available now to carry out the task as described.
be summarized as algorithm development and design, hard- However, in reality, the initial cost of setting up this sort
ware implementation and software coding and these must of design approach can be high not only in terms of equip-
be carried out in the context of the end application (Bate- ment and software costs but also in terms of a high initial
man and Paterson-Stephens, 2003). There are, of course, learning curve. The final choice of approach will often be
many factors that will influence the path taken for each of determined by other factors not least of all the context of
these design stages, development cost and time to market the application, that is, whether the design is likely to be
being the main driving factors. Many tools are available, mass produced and sold at a low cost or it is a one-off for
which will help the engineer to realize a particular design as research purposes.
quickly as possible and at minimum cost. In particular, tools
are available, which will enable algorithms to be simulated 2 OVERVIEW OF SELECTED SMALL-
at block diagram level in order to verify the design concept
TO-LARGE MAINLINE DSP TOOLS
(examples include Matlab and Simulink, or Hyperception),
see Figure 1. The block diagram can then be automatically 2.1 Visualization tools–algorithm
translated to a high-level language (HLL), such as C or design/verification
Ada. The block diagram algorithm to HLL translation can
be target specific, that is to say the code generated can be Matlab
optimized for a particular DSP device (Real-time workshop • Command line and interpreted-code simulation
running under Simulink) (MathWorks–A, 1998). Once the interface
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
998 Elements: C – Data Acquisition and Processing Systems
• Integrated design suite used for software and hard-
ware debugging
• Mixed C/C++ and Assembly level programming
and debugging
• Easy integration with Simulink target coder
• In-circuit emulation support – real-time debugging
• Support for customized host-control interface
designs
• Memory data graphing support – useful for signal
processing apps.
Code Composer Studio (Texas Instruments)
• Can be used with entire range of DSPs from Texas
Instruments
• Integrated design suite used for software and hard-
ware debugging
• Mixed C/C++ and Assembly level programming
and debugging
• Real-time mode debugging – in-circuit emulation
• DSP/BIOS Plug-ins for fast application develop-
ment/code reuse
• Real-time data exchange to host platform (PC).
Figure 1. Simulink application modeling screen shot.
• Matrix-based programming language 3 CHOICE OF DSP TOOLS FOR AN
• Extensive Toolbox support for a range of signal-
processing tasks
APPLICATION
• Complex graphing/plotting capabilities
At the basic level, there are only four things that are
• Massive library of support documentation and third
required to start developing DSP applications: (i) a low
party books
specification host computer – usually windows PC; (ii) a
• Fast simulations.
text editor for writing software applications using assem-
Simulink bly code; (iii) a low-cost DSP development board that
connects to the PC via a parallel or serial cable, see TI-
• ‘Graphical front end’ for Matlab – intuitive user
SPRU191 (1996), and (iv) the basic software tools usu-
interface
ally supplied with the development board (Bateman and
• All the power of Matlab with extra support for fast
Paterson-Stephens, 2003).
simulations
Assuming the user already owns the PC and a simple text
• Extensive range of toolboxes and support libraries
editor, the rest can usually be purchased for around $100,
• Real-time mode simulation possible
and this may or may not include a suitable power supply.
• Compiled code blocks for very fast simulations
Let us start by looking at the last two items from this list,
• Interfacing capabilities – able to connect algorithms
that is, the basic hardware and software tools.
to hardware
• Compilation of algorithms for target hardware
‘Real-time Workshop’. 3.1 Basic hardware–a low-cost development
board
2.2 DSP system development tools Most manufacturers of DSP devices offer a low-cost devel-
opment board, which is intended to be used for a basic
Visual DSP++ (Analog Devices) evaluation of the technology, see TI-SPRU191 (1996) as
an example of this. These boards are usually freestanding
• Can be used with entire range of Analog Devices and connect to the PC via a slow data link, for example, a
DSPs serial or parallel connection, see Figure 2.
DSP Tools 999
Figure 2. Simple DSP development system.
With the board comes a comprehensive set of software • Hex conversion utility: Converts assembled and linked
tools for assembling, linking, debugging, and download- DSP code into a form suitable for an EPROM
ing code. Typically, the development board will include programmer.
data converters so that an audio bandwidth signal can be
An overview of the steps taken during DSP software
passed into and out of the DSP device, and generally these
development is shown in Figure 3. The diagram shows
are suitable for voice or music applications thus enabling
more steps than might typically be required in a basic
the user to evaluate algorithms using real signals in real
project; however, these are shown here for completeness
time.
(TI-SPRU102C, 1998). At intermediate stages within the
diagram, the expected input and output file types are shown.
3.2 Basic software–a low-cost development
environment 3.3 More advanced tools
In order to get started with DSP software development, a Although the basic setup previously described is fine for the
number of basic tools will be needed. These will allow beginner who needs to gain a low-level understanding of
software routines to be written, converted to a form under- the technology and processes involved, the more advanced
standable to the DSP, and then downloaded to the target designer working on a serious commercial project will
device. Usually, these tools are supplied with a develop- require development tools that offer many more features
ment system as part of the complete package. The following (TI-SPRA520, 1999).
list describes the basic tools required: All DSP manufacturers offer an extensive range of DSP
development tools, for example, Code Composer Studio
• Basic text editor: This can be a very simple application offered by Texas Instruments, see Figure 4, is a fully
such as Windows notepad and is used for writing DSP integrated software package that can be used for the whole
programs in assembly language. development process in a very similar way to Microsoft
• Assembler: Available from the manufacturer of the Visual C++. Visual DSP++ integrates editor, assembler,
DSP device, it is used to convert the users’ text-based and compiler into one package along with extensive features
assembly program into a machine-readable format. to help manage projects, allow code reuse (via libraries)
• Linker: Organizes the machine-readable code generated and target debugging (TI-SPRU303, 1999). In effect, the
by the assembler so that it will match the memory advanced tool sets integrate all of the stages identified in
configuration of the target DSP. Figure 1 along with many more features.
• Debug environment: Enables software to be tested for
the particular DSP device; the debug environment may
be in the form of a simulator or an emulator. 4 INTRODUCING DSP TOOLS IN AN
• Downloader: Often a downloader is provided with DSP ORGANIZATION
development systems so that assembled and linked
programs can be transferred to the DSP develop- Many of the tools and techniques used for designing DSP
ment board. hardware and/or software will be familiar to engineers
1000 Elements: C – Data Acquisition and Processing Systems
Basic route shown C source files
highlighted in gray
C Compiler C and assembly
source files
Macro library Assembly
files source files
Assembler
RTS library Object library Object files
files files
LLinker
in ker
COFF EXE file Executable files
Listing file HEX conversion DSP debug
generator & EPROM prog environment
Figure 3. DSP program development flow.
Figure 4. Screen shot of TI Code Composer Studio DSP design interface.
already involved in the design of conventional micro- the algorithmic simulation and visualization tools, and com-
processor-based electronic products. Similarly, engineers plex mathematics. The problem often faced as far as a DSP
involved in design of analog electronic products, for exam- application design is concerned is the very broad range of
ple, control systems or filter designs, will be familiar with skills and knowledge that will be required of the engineer.
DSP Tools 1001
In addition to knowledge about the application, the DSP much regard for the subtle differences a DSP hardware
engineer should have a reasonably good grasp of assembly architecture has to offer, the DSP will perform pretty much
programming, C/C++ programming, hardware design and the same as a standard microprocessor. In order to exploit
interfacing, microprocessor architectures, DSP algorithms, fully the potential speed increases for certain applications,
and discrete mathematics. That said, the DSP engineer will the programmer must be truly aware of what the device
usually work as part of a design team where each individual has to offer. DSP features such as pipelined operation,
will have his or her own spatiality. multiple data paths and storage, Multiply and Accumulate
When introducing DSP tools into an organization, the (MAC) and other arithmetic features/modes of operation,
choice as to which tools will be the most appropriate is and so on are only accessed when requested specifically
usually predetermined by the choice of DSP device and by the programmer. When programming a DSP device, the
manufacturer. All DSP manufacturers and a number of third safest path in many ways is to use the native assembly
party organizations provide short courses on the use of their language offered by the manufacturer. Assembly language
specific tools sets and hardware design with their own DSP has the advantage that it can be very efficient and well
chips. It is often very worthwhile arranging for key staff optimized, and because it is so closely related to the specific
to attend manufacturers’ own short courses, if for no other DSP architecture, the programmer is likely to exploit the
reason than to ensure that ‘undocumented features’ of the subtleties of the DSP architecture itself. The downside to
development tools can be understood at the outset rather assembly language is that it is quite difficult to learn, it is
than allowing them to plague a design for a period of time device specific and therefore the programmers will need to
during the development process. update their knowledge for each new device, and it is not
easy to read and hence not easy to debug.
An alternative to coding using assembly language is to
5 LIMITATIONS OF DSP TOOLS use an HLL such as C or C++; this has the advantage of not
being target specific and so the skills of the programmer are
On the face of it, a DSP processor is very similar to a more generic and usable across a range of devices. C/C++
conventional microprocessor; both have address and data code is clearly an industry standard programming language
buses, a central processing core and memory, and so on. In and so many programmers will find this approach to be a
fact, if a DSP processor is programmed inefficiently without very accessible route into DSP programming. HLLs present
Figure 5. C source debugging using Code Composer Studio from Texas instruments.
1002 Elements: C – Data Acquisition and Processing Systems
their own problems, in particular that of code optimization (MathWorks-B, 1999). A pertinent example for DSP appli-
(Bateman and Paterson-Stephens, 2003). Because HLLs are cation design is the effect of fixed point processing and how
somewhat abstracted from the target hardware, the com- these are to be considered in the final design. When simu-
piled code is usually not well optimized for the device. lating a design using a visualization tool, such as Simulink,
Usually, tool manufacturers provide a range of software the simulation is performed perhaps using a 64-bit floating
switches that can be turned on/off to force the compiler point desktop PC. The design may work well in simulation;
to compile HLL code in a certain way so as to improve however, when it is translated over to the target hardware,
optimization. At times, however, the programmer will be perhaps a 16-bit fixed point processor, the performance may
required to hand code sections of HLL generated assembly be somewhat different (MathWorks-C, 1999). Simulation
language, see Figure 5. In particular, when designing rou- tools usually incorporate features to help incorporate fixed
tines that perform interfacing functions, it is necessary to point processing into a simulation, see Figure 6, and this
resort to hand-coded assembly language in order to meet can help resolve any differences.
the timing requirements of the interface. As far as limitations of design, simulation, and imple-
Simulation tools such as Matlab, Simulink, and Hyper- mentation tools are concerned, there are potential problems
ception provide a useful platform for developing algorithms and pitfalls but if the user takes the time to familiarize
and the user to interact with elements of a design before themselves with tools, many problems can be avoided.
it has been committed to a hardware or embedded soft-
ware implementation. These are intuitive tools that allow
the designer to ‘play’ with algorithms and to gain a greater REFERENCES
understanding of how the final design will operate. These
simulation tools work very well and give some good Bateman, A. and Paterson-Stephens, I. (2003) The DSP Hand-
insight into a problem; however, there are potential pitfalls book, Algorithms, Applications and Design Techniques, ISBN:
0-201-39851-6, Pearson Education/Prentice Hall, UK.
TI-SPRU191 (1996) TMS320C54x DSK Users Guide, Texas
Instruments, http://www.ti.com/.
TI-SPRU102C (1998) TMS320C54x Assembly Language Tools
Users Guide, Texas Instruments, http://www.ti.com/.
TI-SPRA520 (1999) Code Composer Studio White Paper, Texas
Instruments, http://www.ti.com/.
TI-SPRU303 (1999) TMS320C6xxx DSP/BIOSII Users Guide,
Texas Instruments, http://www.ti.com/.
MathWorks–A (1998) Guide to Rapid Prototyping for DSP
Design, The MathWorks – DSP Design Technical Examples.
MathWorks-B (1999) Using Simulink – Version III Users Manual,
The MathWorks Inc.
MathWorks-C (1999) Real-time Workshop – Users Guide, The
Figure 6. Fixed point DSP simulation tools within Simulink. MathWorks Inc.
146: Principles of DSP Hardware Design
Iain Paterson-Stephens
University of Derby, Derby, UK
of view. Single and multiprocessor architectures will be
1 General Principles of DSP Hardware Design 1003 considered.
2 Integration of DSP Hardware with Software
Element 1005
3 Suggested Process for DSP Design 1005 1.1 A generalized DSP system
4 Tools in Support of DSP Development 1006
5 Installation and Testing of DSP The block diagram shown in Figure 1 represents a standard
Developments 1007 DSP system architecture in which provision has been made
References 1007 for connecting various peripheral devices.
Looking at Figure 1, it is apparent that in addition to
processing signals coming from an external source such as
an ADC or CODEC, the DSP may need to handle events
1 GENERAL PRINCIPLES OF DSP originating from a host computer or a user control sur-
HARDWARE DESIGN face. In normal operation, the DSP runs synchronously with
the ADC or CODEC as this provides the regular sam-
Before proceeding to design any application specific digital pling interval. On arrival of each new sample, the DSP
signal processing (DSP) hardware, it is important to have carries out the necessary processing and makes the new
a detailed understanding of the intended application. This output sample available to the DAC or output side of
goes beyond just the sheer number crunching requirements the CODEC. For example, in a professional audio sys-
of the selected DSP; it is also important to fully appre- tem, the new samples would arrive at a rate of 48 000
ciate in detail the I/O requirements and how these may per second and the DSP must keep up with this flow of
impact on the algorithmic performance of the DSP. While data. If the DSP loses synchronization, then, in the case
the DSP device must be able to perform specific numerical of this audio example, an audible click would be heard,
functions very efficiently, it must also have the capability that is, data would be lost. It is, therefore, imperative that
to interface with other devices and systems so that mini- the DSP is able to synchronize with external data flow-
mum disruption occurs to the processing core (Bateman and ing into and out of its ports. Taking this audio example
Paterson-Stephens, 2003). Most DSP devices incorporate a little further, at the same time as samples arriving at
functional units that can operate with some independence 48 000 per second, the DSP may be interrupted by other
from the core and are able to carry out tasks such as external devices; the host interface may have new con-
interrupt handling, direct memory accesses (DMA), serial trol information updating a particular algorithm; the control
interfacing, and timing operations, (TI-SPRU131d, 1997). surface may have updates from the user requiring a filter
In this section, typical system architectures will be pre- response to be recalculated or signal levels to be changed.
sented in order to put many of the DSP facilities into The host and user events are not likely to be synchronous
context within different applications from a hardware point with the arrival of sampled data and yet the DSP must be
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
1004 Elements: C – Data Acquisition and Processing Systems
System User 1.1.2 Interrupt driven
Boot ROM control
clock
surface All DSP devices incorporate a strategy for interrupting the
normal processor flow. External interrupts are provided in
the form of a number of physical pins on the outside of the
Host system bus
DSP chip each of which will be allocated an individual
Host interface
Program
memory interrupt number and an associated location in program
DSP
device memory at which a service routine will be stored. The
Data interrupt service routine is usually a short program provided
memory
by the user, which carries out a task associated with a
particular interrupt.
Analog I/O For example (see Figure 2), if a press button is provided
CODEC for the user of the DSP system, the button could be con-
Test &
development nected to an interrupt input pin. When the button is pressed,
interface Communication
ports the DSP is halted from its normal flow and switches to run
(JTAG)
an alternative routine stored at a different location in mem-
Figure 1. Generalized DSP system architecture. ory, called the interrupt service routine. On completion of
the interrupt service routine, the DSP goes back to where
it was prior to the interrupt.
able to attend to these external interruptions and still main-
tain full synchronization. There are a number of hardware
1.2 Multiprocessor systems
strategies that can be used by a DSP device to handle trans-
actions with its peripherals. These are summarized briefly Real-time performance for some applications is beyond the
as follows. capabilities of a single DSP device. In this situation, the
only possibility is to use a multiprocessing configuration
in which a number of DSP devices are linked together
1.1.1 POLLED I/O (see Figure 3), and the algorithm is divided into subtasks
and distributed among them.
The DSP can POLL external devices to see if they have
new data available. In a POLLED system, the DSP simply
checks an external pin or memory location to see if a + 5v
predetermined condition is met, for example, is bit one
set high? If the condition is TRUE, then an action can DSP
User device
be taken, for example, collect data items from memory. interrupt
Although the POLLING system is very simple, it does have
the disadvantage that the DSP is continually checking for
the condition being met rather than getting on with more
useful work. Figure 2. User interrupt.
Connection to system backplane
Link ports Link ports Link ports Link ports Link ports Link ports
interface
Test bus
DSP DSP DSP DSP DSP DSP
device device device device device device
Arbitration
logic
To backplane
Bus controller Boot ROM Global memory
Figure 3. Multiprocessor DSP system design.
Principles of DSP Hardware Design 1005
Some DSP processors designed with inbuilt multipro-
cessing features are available. These include devices from DTx
Serial interface
Serial interface
Texas Instruments such as the TMS320C4x and C6xxx
DSP device
Boot ROM
DRx
family. The Analog Devices SHARC family also provides
extensive multiprocessing capability through its six serial CLK
Link Ports each with a data bandwidth of 40 Mbps. FSync
2 INTEGRATION OF DSP HARDWARE
WITH SOFTWARE ELEMENT Figure 4. Serial boot ROM using standard 3-wire interface.
The process of integrating the software and hardware ele- any boot-up data that will be required for the application.
ments of a DSP project starts with the generation of source When the DSP is powered up, the DSP device can be
code or a source file. The term source file refers to a text- configured to immediately download all of the system
based file written by the programmer, which contains a code from the serial EEPROM into the DSP’s own ‘on-
sequence of instructions; this is the starting point for many chip’ memory. The DSP then performs a soft reset and
programs written for DSP devices. The source file, if writ- immediately runs the recently downloaded code that is
ten using assembly language instructions, is then converted, sitting in its memory. One of the benefits of using a serial
using an assembler, into an object file. Object files con- EPROM like this is that they are often very compact, 8-pin,
tain the actual machine readable instructions understood devices that consume very little PCB real estate and have
by the particular DSP device targeted. Usually, an object a very simple 3-wire connection interface (see Figure 4),
file is said to be relocatable and as such cannot be used which keeps the circuit complexity to a minimum.
directly by the DSP device; instead, it must first be passed
through a linker program that produces the final executable
DSP program. Although a relocatable program contains all 3 SUGGESTED PROCESS FOR DSP
the correct machine instructions, they have not been allo- DESIGN
cated to specific memory locations within the DSP. One
of the functions of the linker is to resolve this memory Every design is different and each engineer will have his or
allocation issue so that all instructions within the final her own preferred methods, but a good starting point for any
executable program have a specific memory address. The DSP application is to first consider the basic algorithms that
benefit offered by this approach is that relocatable code will be required and to design a functional simulation of the
is not tied to a specific memory structure or even system system using Matlab and Simulink or similar visualization
architecture; therefore, the design, either hardware or soft- packages, see Figure 5. This step helps the designer to
ware element, can be updated during the life of the product
without much limitation (Bateman and Paterson-Stephens,
2003).
The name often used to reference a final executable
program is object module – note the difference with an
object file – the module is the final packaged program that
will run on the DSP device. When the object module has
been produced, it can be loaded in to the DSP device;
this may be via a debug environment hosted on a PC or
through the use of an EPROM and a HEX conversion
utility. The standard object file and object module format
used on the Texas Instruments DSP devices is the COFF
format, this stands for Common Object File Format. The
HEX conversion utility is used to convert the object module
into correctly ordered binary data, which is structured in
such a way as to suit the EPROM, EEPROM, or other
nonvolatile memory devices (TI-SPRU190b, 1998).
A common approach used in current DSP designs is to Figure 5. A screen shot showing Simulink while running a basic
use a serial EEPROM to contain the system boot code and simulation.
1006 Elements: C – Data Acquisition and Processing Systems
fully understand all of the algorithmic elements and likely In the past, DSP development tools were a rather clumsy
number crunching demands of the processor that will need collection of different applications that needed to be used
to be used. in the correct order so that the final design could be
The benefit of this initial stage is that the application realized. Nowadays, DSP development tools have been
can be considered in detail without the complications of merged such that initial coding and syntax debugging tools
sometimes unpredictable I/O hardware or the quirks of have been integrated into a complex suite usually containing
a specific DSP device. Although this stage is a purely a compiler, assembler, code debugging tool, and a project
simulated one, it is still possible to make some initial management tool (TI-SPRU269, 1998). Most of these new
predictions about the various aspects of the final design, integrated DSP development packages allow real hardware
for example: the memory storage requirements for filter to be connected such that code can be tested in real time
coefficient tables, data storage, and even code storage with ‘online’ monitoring of performance, this has greatly
requirements. eased the process of debugging complex software with
In order to select an appropriate processor and final sys- multiple events occurring in real time.
tem architecture, it is helpful to understand the algorithmic Code profiling allows the user to investigate in detail the
demands as detailed above, and the interfacing requirements timing of different sections of code within an application.
for the application must also be considered in detail. For This is particularly useful for complex real-time applica-
example, does the application require multiple channels of tions in which multiple threads of execution may be taking
fast analog I/O, a user interface, a host port, or other inter- place. The designer is able to run the application in real
facing mechanisms? time and investigate where most of the processor time is
After considering a range of issues about the application, being spent. The designer can then make choices about
its interfacing requirements and algorithmic demands, it is optimization of code within specific algorithms in order to
possible to draw up a broad system specification. This can bring the final design within the specification of the device
then be used to help the designer make choices about an and/or application.
appropriate DSP device. Once a device or at least a family Although DSP development tools such as those already
of devices has been selected, it is advisable to rework described allow very detailed analysis of the code running
the system specification and to consider in detail how on the DSP device itself, there is still a useful place for the
the performance of the algorithms will be affected by the DSP device simulator in the armory of tools available to
specific architecture of the DSP device selected (Bateman the designer. A simulator is simply a software application
and Paterson-Stephens, 2003). It should be possible, at running on a host computer that simulates all of the function
this point, to make pretty accurate predictions about the of the target DSP. Simulators do not run in real time, but
number of machine cycles that will be required for each they are used to run assembly and C code in the same
signal processing task taking into account the limitations way as the target device. One of the benefits of using the
and overheads inflicted by the I/O, memory, and other simulator is that it can be easier to test certain elements of
hardware structures on the device. an algorithm ‘offline’ without the distraction of interfacing
When an appropriate decision has been made about the components and associated software getting in the way.
device to be used, the detailed hardware design can follow. Also, the simulator usually has more detailed tools for
Usually, manufacturers provide very detailed application profiling code and assessing where efficiency gains can
notes giving recommended hardware configuration and in be made.
many cases complete system designs that can be adapted The mechanism by which DSP hardware can be mon-
to suit a particular application. This is always a good start- itored, probed, and debugged in real time is the JTAG
ing point because much of the initial design has been done, standard and its associated interface. JTAG actually stands
leaving only the application specifics requiring attention. for the Joint Test Action Group, which is a working group
Most manufacturers even provide complete hardware solu- composed mainly of leading electronic manufacturers who
tions in CAD form so that the initial design can be quickly set out to establish a common standard for in-circuit test-
imported and adapted. ing and emulation. The result was the IEEE 1149.1 JTAG
standard. Of particular interest to the DSP developer is the
ability to use the JTAG interface when testing software.
4 TOOLS IN SUPPORT OF DSP In this mode of operation, the DSP device can be single
DEVELOPMENT stepped, register values and memory areas monitored, and
forced and peripheral devices configured. Integrated pack-
One of the biggest factors in the development of a DSP ages such as Texas Instruments Code Composer Studio and
application is the suitability of the design tools available. Analog Devices Visual DSP++ are examples of hosted
Principles of DSP Hardware Design 1007
TMS 1 2 TRST electromagnetic interference and inappropriate/unexpected
use by users can all be factors.
TDI 3 4 GND The JTAG interface mentioned in the previous section
PD (VCC) 5 6 No pin (key) is also useful for in-the-field testing just as it was for the
development and debugging stage of the project. Usually,
TDO 7 8 GND a 14-pin JTAG interface connector will be included on the
TCK_RET 9 10 GND final DSP board so that in-the-field testing and problem
solving can be carried out. The IEEE 1149.1 JTAG stan-
TCK 11 12 GND
dard, in addition to the ‘software level’ debugging mode,
EMU0 13 14 EMU1 describes a set of rules by which a common test bus system
can operate throughout a system level design. The idea of
Figure 6. Standard 14-pin JTAG interface connector. the test bus system is that complex components within a
system can be linked to a common bus through which they
can be probed and tested while still in circuit. This provides
software packages that communicate with target hardware a simple mechanism by which modern systems containing a
entirely through a fast JTAG emulation port. All aspects of collection of highly integrated densely packed chips can be
software debugging, loading software, testing, and upload- tested with relative ease. The JTAG standard allows the user
ing data to the host are performed via the JTAG link. The to force conditions within a device as required to perform
JTAG support provided on the target DSP will run in par- a particular test, so, for example, using the JTAG interface,
allel with the operation of the DSP device so that, for a user could force the logic level of certain connection pins
example, the value within a target register could be moni- on a device to specific value and subsequently observe the
tored without the need to halt the DSP itself. resulting operation. This technique is often referred to as
The physical connection between the host computer and boundary scan, because the boundary conditions, that is,
the DSP’s JTAG port is made via a standard 14-pin header, those of the external connection pins can be monitored and
see Figure 6. The header carries serial data between the controlled. A subset of the JTAG standard also allows other
target device and the host via the TDO (test data out) aspects of a device’s performance and operation to be con-
and TDI (test data in) connectors. The test clock, which trolled (TI-SPRU173, 1996). Using the JTAG interface, the
is used to synchronize all data transfers, is carried on the final DSP system installation can be fully tested and eval-
TCK connection and a return clock signal is carried on the uated at software, hardware, and system levels.
TCK RET connection. Pin six acts as key so that the test
cable cannot be installed incorrectly and other connection
pins such as TMS, EMU0, and EMU1 are used to control REFERENCES
various operating modes of the interface.
Bateman, A. and Paterson-Stephens, I. (2003) The DSP Hand-
book, Algorithms, Applications and Design Techniques, ISBN:
0-201-39851-6, Pearson Education/Prentice Hall, UK.
5 INSTALLATION AND TESTING
TI-SPRU131d (1997) TMS320C54x CPU and Peripherals, Refer-
OF DSP DEVELOPMENTS ence set, Vol. 1. Texas Instruments, USA, http://www.ti.com/.
TI-SPRU190b (1998) TMS320C62xx/C67xx Peripherals Refer-
There is no substitute for ‘in-the-field’ testing of DSP ence Guide, Texas Instruments, USA, http://www.ti.com/.
designs as is the case for any other design. An application TI-SPRU173 (1996) TMS320C54x Applications Guide – Refer-
may work perfectly well in principle on the bench back ence Set, Vol. 4, Texas Instruments, USA, http://www.ti.com/.
at the design lab, but when it is taken out into the field TI-SPRU269 (1998) TMS320CC6xxx EVM Users Manual and
and tested in its intended environment, many other factors Technical Reference, Texas Instruments, USA, http://www.ti.
can affect performance. Noise induced onto power supplies, com/.
147: Ideal Digital Filter Approximation
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
systems. They are constructed from linear electrical
1 Digital Filter Categories 1008 components.
2 Response in the time and Frequency • Digital filters – these are filters formed as sampled
Domains 1008 data signals. They may be realized by hardware or by
3 Approximation to the Ideal Digital Filter 1009 suitable programming of a digital computer.
Related Articles 1011
Classification of filters according to the frequency ranges,
Further Reading 1011 which are transmitted or rejected, is as follows:
• A low-pass filter – has a passband in the low-frequency
1 DIGITAL FILTER CATEGORIES region;
• A high-pass filter – transmits only high frequency
This is the first article on digital filters. It provides an intro- input signals;
duction to their nature and how they are characterized in • A band-pass filter – has a passband in a particular
the time and frequency domains. A general bibliography on frequency band;
the subject is given at the end of this article. Articles Arti- • A band-stop filter – rejects only a particular frequency
cle 148, General Performance of the Digital Filter, Vol- band.
ume 3; Article 149, Low-, High-, and Band-pass Digital
Idealized response magnitude characteristics of various
Filters, Volume 3; Article 150, Finite Infinite Impulse
types of analog filters are sketched in Figure 1.
Response (IIR) Digital Filters, Volume 3 and Article 151,
A sampled signal has frequency components up to ω =
Finite Impulse Response (FIR) Digital Filters, Volume 3
π/T rad s−1 and, therefore, the digital filter is classi-
provide detail of types.
fied according to its effect on frequency components in
A filter is a system that selectively changes the wave-
the range–π/T < ω < π/T . Idealized response magnitude
shape, amplitude, and phase spectra of a signal in a desired
characteristics of various types of the digital filters are
manner. The purposes of the changes are to improve the
sketched in Figure 2.
quality of the signal, for example, to reduce noise, to
extract information from the signal, or to separate signals
previously combined to make an efficient use of a commu-
nication channel. 2 RESPONSE IN THE TIME AND
Filters can be classified in two groups according to FREQUENCY DOMAINS
signal form:
A filter may be described either by its frequency response,
• Analog filters – these filters process continuous sig- transfer function, H (jω), or by its impulse response, I (t).
nals and are therefore an example of continuous linear For example, the impulse responses of Figure 3 have a
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Ideal Digital Filter Approximation 1009
form of decaying oscillations at a frequency of about ω0 3 APPROXIMATION TO THE IDEAL
rad s−1 , with zero average value. The corresponding fre- DIGITAL FILTER
quency responses will therefore display strong transmission
in the region of ω = ω0 , and rejection of zero frequency
inputs. The transfer function of a filter is given by
The pole-zero approach may be used to demonstrate the
1
relationship between the magnitude and phase response of H (s) = (1)
the linear filter. The pole-zero configuration and frequency an sn + an−1 s n−1 + an−2 s n−2 + . . . a1 s + a0
response characteristic of a typical reactive filter are shown
in Figure 4.
Analog filters are composed of passive electrical ele- H ( jw)
I (t )
ments: resistors, inductors, and capacitors. The effect of
2 p/w0
practical components and terminations is to move the poles t w
and zeros slightly away from the imaginary axis. This w0
t=0 w=0
means that infinite peaks and zeros will not be observed
in practice. I (t ) H ( jw)
Response characteristics for the elementary low-pass, 2 p /w0
high-pass, and band-pass filters are shown in Figure 5. t w
The use of three types of linear electric circuit element, w0
t=0 w=0
resistors, capacitors, and inductors, enables zeros to be
placed anywhere in the s-plane and poles to be placed to Figure 3. Impulse and frequency responses of two analog
the left of the imaginary axis. The use of an active element band-pass filters.
in a filter obviates the need for inductors. Low- and high-
pass digital filters are covered in Article 149, Low-, High-,
absH ( jw)
and Band-pass Digital Filters, Volume 3. s -plane
w3 w
Lowpass Bandpass Bandstop Highpass w2 w1 w=0 w1 w2 w3
argH ( jw)
w
p/2
j( jw)
−p
w=0 w
−3p/2
Figure 1. Idealized response magnitude characteristics of various
types of analog filters. Figure 4. Characteristic of a reactive band-pass filter.
w
(a) w = 0 p /T 2p /T
w
(b) w = 0 p /T 2p /T
w
(c) w=0 p /T 2p/T
Figure 2. Idealized response magnitude characteristics of digital filters; (a) low-pass, (b) band-pass, and (c) high-pass.
1010 Elements: C – Data Acquisition and Processing Systems
(a) H ( jw) This approximates the ideal low-pass filter (Figure 6).
s -plane Im
Re
Figure 7 illustrates two types of polynomials having
w these characteristics. The denominator polynomial should
a have a magnitude close to 1.0 over the passband and as
w=0
(b) large as possible over the stopband.
Im H ( jw)
The left-hand figure shows several functions of the form
Re
w
2n
b w=0 ω
(c)
Im
f1 (ω) = 1 + (4)
H ( jw) ω0
Re
w
where n is a positive integer.
w=0 The right-hand plot of Figure 7 shows functions of the
form
Figure 5. Typical resistors–capacitor (R–C) filter characteris-
tics; (a) low-pass, (b) high-pass, and (c) band-pass. ω
f2 (ω) = 1 + ε Cn
2 2
(5)
ω0
where the factors of the denominator polynomial represent
the poles of H (s). Its frequency response is where Cn denotes the so-called Chebyshev polynomial of
order n in the range–ω0 < ω < ω0 . The value of expression
1 ε2 Cn2 (ω/ω0 ) oscillates between 0 and ε2 for any value of n.
H (s)|s=jω H (jω) = (2)
an(jω) + an−1 (jω)n−1
n The use of the polynomials, (4) and (5), gives rise to
+ an−2 (jω)n−2 + . . . a1 (jω) + a0 the well-known Butterworth and Chebyshev filters, (see
Figure 8), which are defined by the following squared
and the square of its response magnitude is given by
|H (s)|2 = H (jω)H ∗ (jω) = H (jω)H (−jω) H ( jw)
l
1
=
an (jω)n + an−1 (jω)n−1
+ an−2 (jω)n−2 + . . . a1 (jω) + a0
w
1
× (3) −w0 w=0 +w0
an (−jω) + an−1 (−jω)n−1
n
+ an−2 (−jω)n−2 + . . . a1 (−jω) + a0 Figure 6. Approximation of the ideal low-pass filter.
1 + (w/w0)2n n=4 1 + e 2 C 02(w/w0)
n=3
n=2
n=4
n=1
n=3
n=2
n=1
2.0
1 + e2
1.0 1
w w
w=0 w0 w=0 w0
Figure 7. Two types of polynomial characteristics.
Ideal Digital Filter Approximation 1011
absH (jw) 1 1
1 − (s/ω0 ) 1 + (s/ω0 )n
n
1.0
1
(1 + e 2)–1/2
=
2–1/2
1 − (s/ω )2n
for even n
|H (s)| =
2 0
(9)
1
Butterworth
1 − j(s/ω0 ) 1 + j(s/ω0 )n
n
Chebyshev
w
1
= for odd n
w=0 w0 1 + (s/ω0 )2n
Figure 8. Frequency response, magnitude characteristic of the This function has 2n poles equally spaced around a circle
low-pass Butterworth and Chebyshev filters of fifth-order. of radius ω0 in the s-plane.
For even n For odd n
(s/ω0 )2n = 1 (s/ω0 )2n = −1
s -plane
s = jw (s/ω0 )2n = ej2π (s/ω0 )2n = ejπ
π
s n = ω0n ejπ s n = ω0n ej 2
π π
si = ω0 ej n i si = ω0 ej 2n (i+1)
where i = 0, 1, . . . , 2n–1 where i = 0, 1, . . . , 2n–1
(10)
The n poles to the left of the imaginary axis define the
filter. It may be shown that the poles of the Chebyshev
filter are arranged on an ellipse whose major axis lies along
the imaginary axis in the s-plane. The pole locations of the
w0
low-pass Butterworth and Chebyshev filters are shown in
Figure 9.
Butterworth
Chebyshev
RELATED ARTICLES
Figure 9. Pole locations of the low-pass Butterworth and Cheby-
shev filters. Article 125, Outline of Purpose of Analog Data Fil-
ters, Volume 2; Article 133, Amplifiers and Filters for
DAS, Volume 3; Article 148, General Performance of
magnitude functions: the Butterworth function the Digital Filter, Volume 3; Article 149, Low-, High-,
and Band-pass Digital Filters, Volume 3.
1
|H (jω)|2 = (6)
[1 + (ω/ω0 )2n ]
FURTHER READING
and the Chebyshev function
Antoniou, A. (1993) Digital Filters: Analysis, Design and Appli-
1 cations, McGraw-Hill Companies, New York.
|H (jω)|2 = (7) Baher, H. (1993) Selective Linear-Phase Switched-Capacitor and
[1 + ε2 Cn2 (ω/ω0 )]
Digital Filters, Kluwer Academic Publishers, New York.
In the case of the Butterworth filter, we have Chen, Chi-Tsong (2000) Digital Signal Processing: Spectral Com-
putation and Filter Design, Oxford University Press, Oxford.
jω = s Datta, J., Karley, B., Lane, J. and Norwood, J. (2000) DSP Filter
|H (jω)|2 = H (jω)H (−jω) −−−→ H (s)H (−s) Cookbook, Delmar Learning, Clifton Park, New York.
Hamming, R.W. (1998) Digital Filters, Dover Publications,
1 London.
=
1 + (s/jω0 )2n Jackson, L.B. (1996) Digital Filters and Signal Processing: With
1 1 Matlab Exercises, Kluwer Academic, Dordrecht.
= (8)
1 + (−j) (s/jω0 ) 1 − (j) (s/jω0 )n
n n n Mano, M.M. (2001) Digital Design, Prentice Hall, New Jersey.
1012 Elements: C – Data Acquisition and Processing Systems
Parks, T.W. and Burrus, C.S. (1987) Digital Filter Design, Topics Rorabaugh, C.B. (1993) Digital Filter Designer’s Handbook:
in Digital Signal Processing, Wiley-Interscience, New York. Featuring C Routines, McGraw-Hill, New Jersey.
Phillips, L.C. and Nagle, T. (1994) Digital Control System Anal- Schlichtharle, D. (2000) Digital Filters: Basics and Design,
ysis and Design, Pearson Education, Upper Saddle River, NJ. Springer-Verlag, Telos, Berlin.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
This article is based in part on a chapter by J. McGhee et al originally published in Measurement Data Handling: Hardware Techniques
Vol. 2, J. McGhee et al , Technical University of Lodz, 2001.
148: General Performance of the Digital Filter
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
2 POLES AND ZEROS IN DIGITAL
1 Forms of Realization of the Digital Filter 1013 FILTER DESIGN
2 Poles and Zeros in Digital Filter Design 1013
3 Impulse Response of the Digital Filter 1014 The alternative implementation considers filtering in the
Related Articles 1014 time domain. The is done by convolution of the input signal
Further Reading 1014 with the impulse response of the appropriate filter.
A transfer function, H (z), represents a useful frequency
response characteristic. Knowing H (z), it is a simple matter
to derive the time domain formula, which describes the
operation of the filter. When considering the digital filter,
1 FORMS OF REALIZATION OF THE
we are concerned with impulse response rather than its
DIGITAL FILTER frequency spectra; we use a set of z-plane poles and zeros
of transfer function, H (z).
Digital filters can be realized in two different ways: If a digital filter and its transfer function have the poles
and zeros, as shown in Figure 1, then its response for some
frequency ω1 may be found by drawing vectors from the
• Hardware – realized by a digital logic circuit containing poles and zeros to the point z = exp(jω1 Ts ). The response
storage, delay, and addition–subtraction and multiplica- magnitude at frequency ω1 is equal to a1 a2 /b1 , where a1 ,
tion by constants. The advantage of the digital filter is a2 , and b1 are the lengths of vectors from the poles and
its speed, especially if some of the necessary operations zeros to point z.
are performed in parallel. Its very costly realization is Point A in the figure represents sinusoidal frequencies
its disadvantage. ω = 0, 2π/Ts , 4π/Ts , . . . and point B represents frequen-
• Software – as a program. The filter operates rather cies ω = π/Ts , 3π/Ts , 5π/Ts , . . . The frequency response
slowly in comparison with hardware forms, but on characteristic of any digital filter repeats indefinitely at
the other hand, this solution is more flexible and intervals in ω of 2π/T rad s−1 .
inexpensive. Figure 2(a) shows a simple low-pass filter specified by
single zero at z = −1. At frequency ω = 0, the zero vec-
There are two methods of digital filter implementation. The tor has a maximum length of 2.0, and hence, the response
first that one should consider is filtering in the frequency amplitude (Figure 2b) must also be a maximum, and it
domain. This is an important and widely used approach, becomes shorter when we move the zero vector anticlock-
which allows great flexibility in the choice of the filter wise around the unit circle. The vector vanishes at point
characteristics. The signal spectrum is simply multiplied z = −1, which corresponds to the sinusoidal frequency
by the choice of filter characteristic. ω = 2π/T . By considering the phase angle that the zero
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
1014 Elements: C – Data Acquisition and Processing Systems
Im z = exp( jw1Ts) x (n ) x (n − 3)
z-plane x (n +1)
x (n − 1)
x (n − 2)
a1
T
a3 1.0
1.0 1.0
w 1 Ts
B A Re Reversed impulse
(a) t=0 (b) response
a3
Figure 3. (a) Impulse response of the low-pass filter; (b) deriva-
tion of the filter’s output by convolution.
response. Cross multiplying point by point and the sum
terms gives
y(n) = 1 × x(n) + 1 × x(n + 1) (1)
Figure 1. The z-plane poles and zeros of digital filter.
An expression such as this is called a recurrence formula.
z-plane
The filter from the example is ‘unrealizable’, in the sense
that a given output value y(n) depends upon the coincident
input x(n) and the next input x(n + 1). This difficulty may
be solved by dividing H (z) by a term zm , where m equals
H(jw) the excess number of zeros. This is equivalent to adding
2.0
m poles at the origin of the z-plane; such poles do not
w affect the frequency response magnitude characteristic but
(a) (b) w = 0 p/T 2p/T 3p/T merely have the effect of delaying the output by m sampling
j(jw)
intervals. The new transfer function has the form
p/2 (z + 1)
3p/T w H (z) = = 1 + z−1 (2)
p/T 2p/T z
−p/2 w = 0
(c) hence the recurrence formula becomes
Figure 2. An elementary low-pass digital filter (a) pole-zero con-
figuration; (b) magnitude response characteristic; and (c) phase y(n) = x(n) + x(n + 1) (3)
response characteristic.
which means that any output sample is found by summing
vector makes with the positive real axis, we may also infer the present and previous inputs. For details, refer to Arti-
the filter’s phase response (Figure 2c). cle 147, Ideal Digital Filter Approximation, Volume 3.
The transfer function of this elementary digital filter is
H (z) = (z + 1) and equals the transform of the output
signal divided by the input. If we write it as X(z) and Y (z) RELATED ARTICLES
respectively, then Y (z) = (z + 1)X(z) = zX(z) + X(z).
Article 25, Introduction to Signals in Physical Systems,
Volume 1; Article 26, Signal Classification, Volume 1;
3 IMPULSE RESPONSE OF THE DIGITAL Article 27, Signals in the Frequency Domain, Volume 1;
Article 147, Ideal Digital Filter Approximation, Vol-
FILTER
ume 3; Article 149, Low-, High-, and Band-pass Digital
Filters, Volume 3.
The impulse response of a such filter is shown in
Figure 3(a). We convolve this impulse response with
a typical input signal having successive samples of FURTHER READING
value. . . x(n − 1), x(n), x(n + 1), x(n + 2), . . . To calcu-
late the output sample y(n), which coincides with the input Antoniou, A. (1993) Digital Filters: Analysis, Design and Appli-
samples, x(n), we lay a reversed version of the impulse cations, McGraw-Hill Companies, New York.
General Performance of the Digital Filter 1015
Baher, H. (1993) Selective Linear-Phase Switched-Capacitor and Mano, M.M. (2001) Digital Design, Prentice Hall, New Jersey.
Digital Filters, Kluwer Academic Publishers, New York. Parks, T.W. and Burrus, C.S. (1987) Digital Filter Design, Topics
Chen, Chi-Tsong (2000) Digital Signal Processing: Spectral Com- in Digital Signal Processing, Wiley-Interscience, New York.
putation and Filter Design, Oxford University Press, Oxford. Phillips, L.C. and Nagle, T. (1994) Digital Control System Analy-
Datta, J., Karley, B., Lane, J. and Norwood, J. (2000) DSP Filter sis and Design, Pearson Education, Upper Saddle
Cookbook, Delmar Learning, Clifton Park, New York. River, NJ.
Hamming, R.W. (1998) Digital Filters, Dover Publications, Rorabaugh, C.B. (1993) Digital Filter Designer’s Handbook:
London. Featuring C Routines, McGraw-Hill, New Jersey.
Jackson, L.B. (1996) Digital Filters and Signal Processing: With Schlichtharle, D. (2000) Digital Filters: Basics and Design,
Matlab Exercises, Kluwer Academic, Dordrecht. Springer-Verlag, Telos, Berlin.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
This article is based in part on a chapter by J. McGhee et al originally published in Measurement Data Handling: Hardware Techniques
Vol. 2, J. McGhee et al , Technical University of Lodz, 2001.
149: Low-, High-, and Band-pass Digital Filters
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
The output sample y(n) is found from the previous output
1 Low-pass Digital Filter 1016 and previous input. A filter, which calculates a new output
2 High-pass Digital Filter 1016 using one or more previous outputs is called recursive and
3 Analyzing a Digital Filter from a Pole-zero arises whenever a transfer function has a pole placed other
Plot 1017 than at the origin of the z-plane. It can be also written as a
nonrecursive recurrence formula
4 Building a Digital Filter from a Pole-zero
Plot 1017 y(n) = x(n − 1) + αx(n − 2) + α 2 x(n − 3)
Related Articles 1018
+ α 3 x(n − 4) + · · · · · (6)
Further Reading 1018
2 HIGH-PASS DIGITAL FILTER
1 LOW-PASS DIGITAL FILTER
A high-pass filter has the transfer function
The low-pass filter with a single pole placed at z = α is
presented in Figure 1. (z − 1) Y (z)
H (z) = = (7)
If 0 < α 1, then the pole vector will be very small (z + α) X(z)
when ω = 0, giving a large response, and relatively large Figure 2 gives its (a) pole-zero configuration, (b) magni-
when ω = π/T , giving a small response. tude response characteristic, and (c) impulse response.
The transfer function of the considered filter is From the pole-zero plot,
1 Y (z)
H (z) = = (1) (z + α)Y (z) = (z − 1)X(z) (8)
(z − α) X(z)
Hence, The recurrence formula is therefore
(z–α)Y (z) = X(z) or zY (z) − αY (z) = X(z) (2) y(n + 1) + αy(n) = x(n + 1) − x(n) (9)
The recurrence formula is therefore or
y(n + 1)–α = x(n) (3) y(n) + αy(n − 1) = x(n) − x(n − 1) (10)
which is equivalent to giving
y(n)–α(n − 1) = x(n − 1) (4) y(n) = −αy(n − 1) + x(n) − x(n − 1) (11)
or The first nine terms of the impulse response are shown
y(n) = αy(n − 1) + x(n − 1) (5) in Figure 2(c).
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Low-, High-, and Band-pass Digital Filters 1017
z -plane H 2(jw) |H ( jw)|
z -plane
1/(1 − a)
a w0T
r
− w0T
1/(1 + a) w w
(a) (b) w=0 p/T 2p/T (a) (b) w0 p/T 2 p/T
Figure 1. Elementary low-pass digital filter. (a) Pole-zero con- Figure 3. A band-pass digital filter. (a) Pole-zero configuration
figuration and (b) magnitude response characteristic. and (b) magnitude response characteristic.
z -plane |H (jw)| we express the pole position in polar coordinates; the
a 2/(1 − a) transfer function is then
(z − 1)(z + 1)
H (z) =
w
[z − r exp(jω0 T )][z − r exp(−jω0 T )]
(a) (b) w=0 p/T 2p/T z2 − 1
= (12)
z2 − rz[exp(jω0 T ) + exp(−jω0 T )] + r 2
a(1 + a)
1.0 and
t
t=0 z2 − 1
H (z) = (13)
z2 − 2rz cos ω0 T + r 2
a 2 (1 + a)
(c) −(1 + a)
From this expression, we may derive the recurrence
Figure 2. A simple high-pass digital filter. (a) Pole-zero con- formula
figuration, (b) magnitude response characteristic, and (c) impulse
response. y(n) = 2r cos(ω0 T )y(n − 1) − r 2 y(n − 2)
+ x(n) − x(n − 2) (14)
3 ANALYZING A DIGITAL FILTER FROM
A POLE-ZERO PLOT
4 BUILDING A DIGITAL FILTER FROM A
Figure 3(a) shows an example of how to analyze the digital POLE-ZERO PLOT
filter with a pole-zero configuration.
The complex conjugate pole pair placed on a circle of For hardware realization (see Figure 4), we choose the
radius r, where r is close to unity, gives rise to a band-pass following parameter values:
characteristic centered on ω0 rad s−1 (see Figure 3b). Zero-
frequency and high-frequency rejection are provided by the • radius r = 0.95
zeros at z = 1 and z = −1. As it is quite often convenient, • sampling frequency fs = 1 kHz, that is, Ts = 1 ms
x (n − 1)
x (n ) x (n − 2) y (n − 2) y (n − 1)
ADC T T T T
Filtered
analog
output
−1 0.9025 1.665
Analog
signal
input
+ ADC
Figure 4. A hardware implementation of the band-pass filter.
1018 Elements: C – Data Acquisition and Processing Systems
• center frequency ω0 = 80 Hz, giving ω0 T = 0.503 rad FURTHER READING
or 28.8◦ .
Antoniou, A. (1993) Digital Filters: Analysis, Design and Appli-
The recurrence formula becomes cations, McGraw-Hill Companies, New York.
Baher, H. (1993) Selective Linear-Phase Switched-Capacitor and
y(n) = 1.6650y(n − 1) − 0.9025y(n − 2) Digital Filters, Kluwer Academic Publishers, New York.
+ x(n) − x(n − 2) (15) Chen, Chi-Tsong (2000) Digital Signal Processing: Spectral Com-
putation and Filter Design, Oxford University Press, Oxford.
Datta, J., Karley, B., Lane, J. and Norwood, J. (2000) DSP Filter
For details, refer Article 147, Ideal Digital Filter Cookbook, Delmar Learning, Clifton Park, New York.
Approximation, Volume 3.
Hamming, R.W. (1998) Digital Filters, Dover Publications,
London.
Jackson, L.B. (1996) Digital Filters and Signal Processing: With
RELATED ARTICLES Matlab Exercises, Kluwer Academic, Dordrecht.
Mano, M.M. (2001) Digital Design, Prentice Hall, New Jersey.
Article 25, Introduction to Signals in Physical Sys- Parks, T.W. and Burrus, C.S. (1987) Digital Filter Design, Topics
tems, Volume 1; Article 146, Principles of DSP Hard- in Digital Signal Processing, Wiley-Interscience, New York.
ware Design, Volume 3; Article 147, Ideal Digital Filter Phillips, L.C. and Nagle, T. (1994) Digital Control System Anal-
Approximation, Volume 3; Article 148, General Per- ysis and Design, Pearson Education, Upper Saddle River, NJ.
formance of the Digital Filter, Volume 3; Article 150, Rorabaugh, C.B. (1993) Digital Filter Designer’s Handbook:
Finite Infinite Impulse Response (IIR) Digital Filters, Featuring C Routines, McGraw-Hill, New Jersey.
Volume 3; Article 151, Finite Impulse Response (FIR) Schlichtharle, D. (2000) Digital Filters: Basics and Design,
Digital Filters, Volume 3. Springer-Verlag, Telos, Berlin.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
This article is based in part on a chapter by J. McGhee et al originally published in Measurement Data Handling: Hardware Techniques
Vol. 2, J. McGhee et al , Technical University of Lodz, 2001.
150:Finite Infinite Impulse Response (IIR) Digital
Filters
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
filter cannot be symmetrical in form and that the IIR filter
1 The Infinite Impulse Response (IIR) 1019 cannot display pure linear-phase characteristics.
2 Design of IIR Digital Filters from Analog The advantage of its design is that the positioning of just
Filters 1020 one or a few poles inside but close to the unit circle enables
very selective filter passband characteristics to be achieved.
3 Impulse Invariance Method 1020
Many practical IIR filters are based upon analog equiv-
4 Determination of a Digital Filter from an
alents because analogous filters based upon lumped circuit
Analog Filter 1021
elements have infinite impulse response. The design prob-
5 Determination of a Digital Filter Transform lem is to find a suitable transformation for mapping the
from an Analog Filter 1022 s-plane poles and zeros of the analog filter representation
6 Bilinear Transformation Model 1022 onto the z-plane.
7 Converting an Analog Filter into a Digital There are several techniques for deriving a digital filter
Equivalent Filter 1022 from an analog one. Two of the widely used ones are as
8 Properties of Design of Digital Butterworth follows:
Filters 1023
• Impulse invariance
9 Designing a Digital Butterworth Filter 1023 • Bilinear transformation.
10 Design of a Digital Filter Using the
It is possible to choose a suitable z-plane pole-zero
Bilinear Transformation Method 1024
configuration for a filter without direct reference to analog
Related Articles 1025 designs. The least confusing point of view toward digital
Further Reading 1026 filter design is to consider the filter as being specified in
terms of angle around the unit circle rather than in terms
of analog frequencies.
A problem in determining an appropriate set of specifi-
1 THE INFINITE IMPULSE RESPONSE
cations for a low-pass digital filter is presented in Figure 1.
(IIR) Specifications for a required filter often take the form of
such a given tolerance scheme. The dashed curve represents
Any digital filter specified in terms of one or more z- the frequency response of a system that meets the prescribed
plane poles has the infinite impulse response, IIR. It is a specifications. In this case, there is a passband wherein the
recursive filter, because no infinite impulse response could magnitude of response must approximate to 1, with an error
be implemented in a purely nonrecursive operation. of ±δ1 , which can be written as
The recurrence formulae of IIR filters normally involve
relatively few terms. It is clear that the IIR of any realizable 1 − δ1 ≤ H ejω ≤ 1 + δ1 where |ω| ≤ ωp (1)
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
1020 Elements: C – Data Acquisition and Processing Systems
H(ejw) where the impulse response, ha (t), is the inverse Laplace
transform of Ha (s).
1 + d1
Alternatively, an analog system having a system function
Ha (s) can be described by the differential equation
1 − d1
Passband
N
dk ya (t) dk xa (t)
M
Transition ck = ek (5)
k=0
dt k k=0
dt k
Stopband
d2
w The corresponding rational system function for digital
0 wp ws p filters has the form
Figure 1. Tolerance limits for approximation of ideal low-pass
M
filter. bk z−k
k=0 Y (z)
H (z) = = (6)
There is also a stopband in which the magnitude response N X(z)
−k
ck z
must approximate to zero, with an error less than δ2 , for
k=0
which
jω The input and output are related by the convolution sum
H e ≤ δ where ωs ≤ |ω| ≤ π | (2)
2
+∞
The passband cut-off frequency, ωp , and the stopband y(n) = x(k)n(n − k) (7)
cut-off frequency, ωs , are given in terms of z-plane angles. −∞
The next step is to find the discrete-time linear system
or equivalently by the difference equation
whose frequency response falls within the prescribed toler-
ance.
N
M
ak y(n − k) = bk x(n − k) (8)
k=0 k=0
2 DESIGN OF IIR DIGITAL FILTERS
FROM ANALOG FILTERS In transforming an analog system to a digital system, we
must therefore obtain either H (z) or h(n) from the analog
This method is often used because analog filter design is filter design.
highly developed. Many useful analog design methods have
relatively simple closed-form design formulae. In many
applications, it is of interest to use a digital filter to simulate
3 IMPULSE INVARIANCE METHOD
the performance of an analog linear time-invariant filter.
This procedure for transforming an analog filter to a dig-
Consider an analog system function given by
ital filter design corresponds to choosing the unit-sample
response of the digital filter as equally spaced samples of
M
ek s k the impulse response of the analog filter. That is,
k=0 Ya (s)
Ha (s) = = (3) h(n) = ha (nT ) (9)
N Xa (s)
k
ck s where T is the sampling period.
k=0
The resulting relationship between the response of analog
where xa (t) is the input and ya (t) is the output of the and digital version of a typical low-pass filter is illustrated
filter, and Xa (s) and Ya (s) are their respective Laplace in Figure 2.
transforms. It is assumed that Ha (s) has been obtained To investigate the interpretation of impulse invariant
through one of the established approximation methods used design in terms of a relationship between the s-plane and
in analog filter design. The input and output of such a the z-plane, consider the system functions of the analog
system are related by the convolution integral, filter expressed in terms of the fraction expansion, so that
∞ N
Ak
ya (t) = x(τ )ha (t − τ ) dτ (4) Ha (s) = (10)
s − sk
−∞ k=1
Finite Infinite Impulse Response (IIR) Digital Filters 1021
I1(t ) H1(jw) 4 DETERMINATION OF A DIGITAL
FILTER FROM AN ANALOG FILTER
t w
As an example of the determination of a digital filter
from an analog filter by means of the impulse invariance,
consider the analog system Ha (s) given by
(a)
Sampled
s+a 0.5 0.5
H (jw) Ha (s) = = + (14)
I 2(t ) (s + a) + b
2 2 s + a + jb s + a − jb
Ts
The corresponding transfer function of the impulse
t invariant digital filter is then
w
0.5 0.5
H (z) = +
(b) 1− e−aT e−jbT z−1 1− e−aT ejbT z−1
−aT
Figure 2. The impulse invariance method. (a) The impulse and 1 − (e cos bT )z−1
= (15)
frequency response of the analog filter; (b) its digital equivalent. (1 − e−aT e−jbT z−1 )(1 − e−aT e−jbT z−1 )
I2 (t) is a sampled version of I1 (t).
from which
The corresponding impulse response is 1 − (e−aT cos bT )z−1
H (z) = (16)
N 1 − 2(e−aT cos bT )z−1 + (e−aT )2 z−2
ha (t) = Ak esk T u(t) (11)
k=1
The digital filter has one zero at the origin and a zero at
where u(t) is a continuous-time unit step function. z = e−aT cos bT (17)
The response of the digital filter is then
Figure 4 shows the s-plane pole-zero plot for Ha (s) and
N
N
the z-plane pole-zero plot of H (z), along with the corre-
h(n) = ha (nT ) = Ak e sk nT
u(n) = Ak (e sk T n
) u(n)
k=1 k=1
sponding analog and digital frequency response functions.
(12)
The system function of the digital filter H (z) is conse- s -plane 20log |H ( jΩ)|
quently given by 2 p/T
30
N
Ak a
p/T 20
H (z) = (13) 10
k=1
1 − esk T z−1 b Ω
p/T 2p/T
If the analog filter is stable, corresponding to the real −10
−p/T
part of sk less than zero, then the magnitude of esk T will be −20
less than unity. The corresponding pole in the digital filter (a) − 2 p/T
will be inside the unit circle, and consequently the digital 20log |H (e jw)|
filter will be also stable. It can be shown that strips of width z -plane 30
2π/T in the s-plane map into the entire z-plane, as depicted 20
in Figure 3. e−aT 10
bT w
p 2p
jw z-plane −10
p/T −20
s (b)
e−aTcos (bT )
−1 1
−p/T Figure 4. (a) Pole-zero plot and frequency response of a sec-
ond-order analog system; (b) Pole-zero plot and frequency
response of the discrete-time system obtained by sampling the
Figure 3. Representation of periodic sampling. impulse response of the above system.
1022 Elements: C – Data Acquisition and Processing Systems
In this case, the frequency response of the analog system w
w = 2tan−1(ΩT/2)
falls off rather slowly relative to the sampling frequency
p
and thus the effects of aliasing are apparent in the digital
frequency response. Ω
5 DETERMINATION OF A DIGITAL −p
FILTER TRANSFORM FROM AN
Figure 5. Mapping of the s-plane into z-plane using the bilinear
ANALOG FILTER transformation.
Here is determined the digital filter transform function from jΩ
the analog filter of the previous case, this time by means z -plane
of the impulse invariance. s -plane
Image of
The analog system function Ha (s) given by s = jΩ
−b j0.5 j0.5
Ha (s) = = − (18)
(s + a) + b
2 2 s + a − jb s + a + jb Image of left
half plane
The corresponding transfer function of the impulse
invariant digital filter is then
Figure 6. Mapping of the analog frequency axis onto the unit
j0.5 j0.5
H (z) = − circle using the bilinear transformation.
1 − e−aT ejbT z−1 1 − e−aT e−jbT z−1
e−aT sin bT thus for z on the unit circle, σ = 0, for which and ω are
= related by
(1 − e−aT e−jbT z−1 )(1 − e−aT e−jbT z−1 ) ω
(e−aT sin bT )z2 = tan (22)
= 2
(z − e−aT e−jbT )(z − e−aT e−jbT )
This relationship is plotted in Figure 5.
e−aT sin bT In addition to the fact that the imaginary axis in the s-
= (19)
1 − 2(e−aT cos bT )z−1 + (e−aT )2 z−2 plane maps to the unit circle in the z-plane, the left half
of the s-plane maps to the inside of the unit circle and the
The digital filter has a double zero at the origin. right half of the s-plane maps to the outside of the unit
circle, as depicted in Figure 6.
6 BILINEAR TRANSFORMATION MODEL
7 CONVERTING AN ANALOG FILTER
The bilinear transformation is an example of the so-called
frequency transformation method. A basis of the approach INTO A DIGITAL EQUIVALENT FILTER
can be clarified by an example. Consider first the function
Suppose the need is to convert an analog filter with transfer
(z − 1) function
s= where z = e sT
(20)
(z + 1) 1
Ha (s) = (23)
which is bilinear in the sense that both its numerator and s+α
denominator polynomials are linear in the variable z. into an equivalent digital filter.
In order to explain the value of this function in converting Wherever s appears in Ha (s), replace it by s from (20),
an analog filter into a digital equivalent, we need to evaluate giving us the digital filter transfer H (z) to obtain
its spectrum. This is found by putting s = jω, or z = ejωT ,
1 z+1
which gives H (z) = =
z−1 z − 1 + α(z + 1)
+α
ejωT − 1 ejωT /2 (ejωT /2 − e−jωT /2 ) z+1
s= =
ejωT + 1 ejωT /2 (ejωT /2 + e−jωT /2 ) 1 z+1
= (24)
ωT α+1 α−1
= j tan = σ + j (21) z+
2 α+1
Finite Infinite Impulse Response (IIR) Digital Filters 1023
w Ωc
p
s -plane
ws w = 2 arctan(ΩT /2)
wp 60°
Ha(jΩ)
Ωp = 2/T tan(wp /2)
Ωs = 2/T tan(ws /2)
Ω
Ωp Ωs
Figure 7. Frequency response and tolerance schemes for analog Figure 9. s-plane pole location for a third-order Butterworth
and corresponding digital filter. filter.
from which the frequency response is given by
which can be written in the form of poles
1
H (jω) = (25) π
j tan ωT /2 + α c ej N k for N odd
spk = π (28)
The complete frequency response characteristics of the c ej 2N (2k+1) for N even
analog filter are compressed into the frequency range 0 <
ω < π/T in the corresponding digital filter. Figure 7 shows where k = 0, . . . , N − 1
the mapping of an analog frequency response and its corre- The properties of a Butterworth filter in terms of poles
sponding digital frequency response and tolerance scheme. and zeros are then as follows:
• There are 2N poles equally spaced in angle on the circle
of radius c in the s-plane.
8 PROPERTIES OF DESIGN OF DIGITAL • The poles are symmetrically located with respect to the
BUTTERWORTH FILTERS imaginary axis.
• A pole never falls on the imaginary axis.
The squared magnitude function for an analog Butterworth • A pole occurs on the real axis for odd N but not for
filter is given by even N .
• The angular spacing between the poles on the circle is
1 π/N radians (for N = 3, it is indicated in Figure 9).
H (j)2 = 2N (26)
a
j • The filter is stable when poles occur on the left-half-
1+ plane part of the Butterworth circle.
jc
as sketched in Figure 8.
The roots of the denominator polynomial, the poles of 9 DESIGNING A DIGITAL
the squared magnitude function, are then at BUTTERWORTH FILTER
1 Consider the need to design a filter such that the passband
sp = (−1) 2N jc (27)
magnitude is constant within 1 dB for a frequency below
0.2π and the stopband attenuation is greater than 15 dB for
Ha(jΩ) frequencies between 0.3π and π.
1 If the passband magnitude is normalized to unity at
N=2
2 N=4 ω = 0, then we require that
N=8
20 log10 |H (j0.2π)| ≥ −1 and
Ω
0 Ω 20 log10 |H (j0.3π)| ≤ −15 (29)
Figure 8. Dependence of the Butterworth magnitude character- The filter design consists essentially of determining the
istic on the order N. parameters N and c to meet the desired specification. For
1024 Elements: C – Data Acquisition and Processing Systems
the Butterworth filter, 1.2
1.0
20 log10 |H (j)| = 10 log10 |H (j)| 2
(30)
Magnitude
0.8
0.6
then
1 0.4
10 log10 = a[dB] (31) 0.2
1 + (/c )2N
0
0 0.2p 0.4p 0.6p 0.8p p
and 2N
0
1+ = 100.1a (32) − 10
c − 20
Gain (dB)
− 30
In our example we have − 40
− 50
0.2π 2N − 60
1+ = 100.1 (33) − 70
c
− 80
0 0.2p 0.4p 0.6p 0.8p p
and 2N
0.3π 200
1+ = 101.5 (34)
Phase (degress)
150
c 100
50
The solution of these two equations leads to the value 0
N = 5.88 and c = 0.70474 rad s−1 . The parameter N , − 50
however, must be an integer and, consequently, in order − 100
− 150
for the specifications to be met or exceeded, we round N − 200
up to the nearest integer so that N = 6. 0 0.2p 0.4p 0.6p 0.8p p
Then, however, both of the passband and stopband spec-
Figure 10. Frequency response of sixth-order Butterworth filter
ifications cannot be met exactly. If we substitute N = 6 transform by impulse invariance.
in (34), we obtain c = 0.7032 rad s−1 . With this value, the
passband specification will be met exactly and the stopband It is evident from (36) that the output signal can be
specification will be exceeded for the analog filter. This derived from the input signal by a recurrence formula.
allows some margin for the aliasing in the digital filter. The frequency response of the above system is shown in
With the value of c and with N = 6, there are three pole Figure 10.
pairs in the left half of the s-plane with coordinates at Recall now that the filter was designed to exactly meet
• pole pair 1: −0.1820 ± j0.6792 the specification at the passband edge and to exceed
• pole pair 2: −0.4972 ± j0.4972 the specification at the stopband edge, and in fact this
• pole pair 3: −0.6792 ± j0.1820. is the case. This is an indication that the analog fil-
so that ter is sufficiently band limited such that aliasing presents
no problems.
0.120931
Ha (s) = (35)
(s + 0.3640s + 0.4945)
2
× (s 2 + 0.9945s + 0.4945) 10 DESIGN OF A DIGITAL FILTER USING
× (s 2 + 1.3585s + 0.4945)
THE BILINEAR TRANSFORMATION
Now consider the design of the desired filter above this METHOD
time using the impulse invariance method.
If we express Ha (s) as a partial fraction and perform a Given below is the design of a digital filter using the bilinear
transformation of the above equation, the resulting system transformation method.
function of the digital filter is The digital frequency specification must be prewarped
0.2871 − 0.4466z−1 to the corresponding analog frequencies so that with the
H (z) = frequency distortion inherent in the bilinear transformation
1 − 1.2971z−1 + 0.6949z−2
the critical analog frequencies will map to the correct
−2.1429 + 1.1454z−1
+ critical digital frequencies. For the specific filter, we require
1 − 1.0691z−1 + 0.3699z−2 that
1.8558 − 0.6304z−1 0.2π
+ (36) 20 log10 Ha j2 tan ≥ −1
(37)
1 − 0.9972z−1 + 0.2570z−2 2
Finite Infinite Impulse Response (IIR) Digital Filters 1025
and 1.2
0.3π 1.0
≤ −15
Magnitude
20 log10 Ha j2 tan (38) 0.8
2 0.6
0.4
where we have consequently assumed that T = 1. Solving 0.2
the equations with equality 0
0 0.2p 0.4p 0.6p 0.8p p
2N 0
2 tan(0.1π)
1+ = 100.1 (39) − 10
c − 20
Gain (dB)
− 30
− 40
and − 50
2N − 60
2 tan(0.15π) − 70
1+ = 101.5 (40) − 80
c 0 0.2p 0.4p 0.6p 0.8p p
200
so that
Phase (degress)
150
100
50
1 log (101.5 − 1) (100.1 − 1)
N= = 5.305 (41) 0
2 log [tan(0.15π) /tan(0.1π) ] − 50
− 100
− 150
N must be chosen as 6. If we determine c by substi- − 200
0 0.2p 0.4p 0.6p 0.8p p
tuting N = 6 into (39), we obtain c = 0.76622. For this
value of c , the passband specifications are exceeded and Figure 12. Frequency transfer of sixth-order Butterworth filter
the stopband specifications are met exactly. In the s-plane, transformed by bilinear transformation.
12 poles of the squared magnitude function are uniformly
distributed in angle on a circle of radius 0.76622, as shown with T chosen as unity, with the result that
in Figure 11.
The transfer function in the s-plane corresponding to the 0.0007378(1 − z−1 )6
H (z) =
left-half-plane poles is (1 − 1.2686z−1 + 0.7051z−2 )
× (1 − 1.0106z−1 + 0.3583z−2 )
0.20238 1
Ha (s) = (42) × (43)
(s + 0.396s + 0.5871)
2
(1 − 0.9044z−1 + 0.2155z−2 )
× (s 2 + 1.083s + 0.5871)
× (s 2 + 1.4802s + 0.5871) The magnitude and phase of the digital frequency re-
sponse is shown in Figure 12.
At ω = 0.2π, the magnitude is down 0.5632 dB and at
The transfer function H (z) for the digital filter is then
ω = 0.3π the magnitude is down exactly 15 dB.
obtained by applying the bilinear transformation to Ha (s)
It should be noted that the magnitude function in
Figure 12 falls off much more rapidly than the one in
Figure 10. This is because the bilinear transformation maps
the entire j axis of the s-plane onto the unit circle. Since
the analog Butterworth filter has sixth-order zero at s = ∞,
the resulting digital filter has a sixth-order zero at z = −1.
For further information on digital filter design, consult
p/6
the bibliography given in Article 147, Ideal Digital Filter
Approximation, Volume 3.
0.7
66
22
RELATED ARTICLES
Article 125, Outline of Purpose of Analog Data Filters,
Figure 11. s-plane locations of sixth-order Butterworth filter. Volume 2; Article 133, Amplifiers and Filters for DAS,
1026 Elements: C – Data Acquisition and Processing Systems
Volume 3; Article 147, Ideal Digital Filter Approxima- Hamming, R.W. (1998) Digital Filters, Dover Publications, Lon-
tion, Volume 3; Article 148, General Performance of the don.
Digital Filter, Volume 3; Article 149, Low-, High-, and Jackson, L.B. (1996) Digital Filters and Signal Processing: With
Band-pass Digital Filters, Volume 3. Matlab Exercises, Kluwer Academic, Dordrecht.
Mano, M.M. (2001) Digital Design, Prentice Hall, New Jer-
sey.
FURTHER READING Parks, T.W. and Burrus, C.S. (1987) Digital Filter Design,
Topics in Digital Signal Processing, Wiley-Interscience, New
Antoniou, A. (1993) Digital Filters: Analysis, Design and Appli- York.
cations, McGraw-Hill Companies, New York. Phillips, L.C. and Nagle, T. (1994) Digital Control System Anal-
Baher, H. (1993) Selective Linear-Phase Switched-Capacitor and ysis and Design, Pearson Education, Upper Saddle River,
Digital Filters, Kluwer Academic Publishers, New York. NJ.
Chen, Chi-Tsong (2000) Digital Signal Processing: Spectral Com- Rorabaugh, C.B. (1993) Digital Filter Designer’s Handbook:
putation and Filter Design, Oxford University Press, Oxford. Featuring C Routines, McGraw-Hill, New Jersey.
Datta, J., Karley, B., Lane, J. and Norwood, J. (2000) DSP Filter Schlichtharle, D. (2000) Digital Filters: Basics and Design,
Cookbook, Delmar Learning, Clifton Park, New York. Springer-Verlag, Telos, Berlin.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
This article is based in part on a chapter by J. McGhee et al originally published in Measurement Data Handling: Hardware Techniques
Vol. 2, J. McGhee et al , Technical University of Lodz, 2001.
151: Finite Impulse Response (FIR) Digital Filters
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
response, which is finite in duration, can also be symmet-
1 Outline of Digital Filter with Finite Impulse rical in form. This produces a pure linear-phase charac-
Response (FIR) 1027 teristic; there is no phase distortion. It does not have to
2 The Moving-average Filter 1027 be linear phase, but most practical designs incorporate this
3 FIR Design Based on Window Function 1027 advantage, which is not available in analog filters. The filter
is also inherently stable, since it does not involve feedback
4 Transversal Filter Design by The Window
from output to input.
Method 1028
5 Digital Filter with Linear-phase Spectrum 1029
6 Summary of Digital Filter Methods 1029
2 THE MOVING-AVERAGE FILTER
Related Articles 1030 A simple form of digital filter in which all the multipliers
Further Reading 1030 are equal is often called a moving-average filter.
As an example, consider the transversal filter in which all
multipliers are equal. We use 19 delay and 20 multipliers
equal to 0.05.
1 OUTLINE OF DIGITAL FILTER WITH The transfer function, as a sum of finite power series, is
FINITE IMPULSE RESPONSE (FIR) given by
A digital transversal filter is a filter, which has m delay (1 − z−20 ) (z20 − 1)
H (z) = = (3)
stages, (m + 1) positive or negative multipliers, and an (1 − z−1 ) z19 (z − 1)
adder or summing junction. It is nonrecursive, with a time The filter has 19 poles in origin and z = 1. The zeros
domain recurrence formula given by of the function are uniformly distributed in angle on a unit
circle in the z-plane and the angular spacing between the
y(n) = a0 x(n) + a1 x(n − 1) + a2 x(n − 2)
zeros on the circle is π/10T rad. The first transmission
+ · · · · · + am x(n − m) (1) zero is at ω = π/10T. The characteristics of this particular
design are shown in Figure 1.
and
m
y(n) = ai x(n − i) (2) 3 FIR DESIGN BASED ON WINDOW
i=0
The filter impulse response is just made up to the multi-
FUNCTION
plier coefficient sequence a0 to am , and is finite in duration. Digital filter design starts from an ideal desired frequency
The disadvantage of a nonrecursive finite impulse re- response specification such as
sponse (FIR) filter is that it requires many more delay and ∞
multiplier elements than a recursive filter. However, the Hd (ejω ) = hd (n)e−jωn (4)
filter has significant advantages. For example, an impulse −∞
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
1028 Elements: C – Data Acquisition and Processing Systems
f (t) Now follows some commonly used window functions.
−0.05
These are specified by the following equations:
Rectangular window function
t
1, 0 ≤ n ≤ N − 1
t=0 w(n) = (8)
(a) 0, elsewhere
Filter waveforms Bartlett window function
Output
2n N −1
, 0≤n≤
w(n) = N − 12n 2 (9)
2 − 2n
, ≤n≤N −1
N −1 N −1
Input
Hanning window function
t 1 1 − cos 2πn , 0≤n≤N −1
t=0 t = 19T w(n) = 2 N −1
(b)
0, otherwise
(10)
|H(jω)|
1.0 Hamming window function
0.54 − 0.46 cos 2πn , 0≤n≤N −1
w(n) = N −1
0, otherwise
(11)
w
Blackman window function
p/10T p/2T p/T
(c)
2πn
0.42 − 0.5 cos
Figure 1. A moving-average filter (a) impulse response, (b) typi- N −
1 0≤n≤N −1
cal input and output waveform, (c) frequency response magnitude w(n) = 4πn
+0.08 cos
,
characteristic.
N −1
0, otherwise
The equivalent impulse response may then be found by (12)
the inverse discrete Fourier transform, and it is used to Windowing can be applied at different stages of spec-
define the multipliers of the digital transversal filter. trum estimation:
π
1 • to the time record, called linear windowing in the
hd (n) = Hd (ejωn )ejωn dω (5)
2π time domain,
−π
• to the results of the Discrete Fourier Transform called
The impulse response will often contain an unacceptable linear windowing in the frequency domain.
number of terms. It is possible to simplify the impulse
response by ignoring the smaller terms.
4 TRANSVERSAL FILTER DESIGN BY
hd (n), 0<n<N −1 THE WINDOW METHOD
h(n) = (6)
0, elsewhere
To illustrate the techniques, now consider the design of a
Such truncation is equivalent to multiplying the function low-pass filter, which takes as its starting point an ideal
by a rectangular observation window; it causes spreading of low-pass filter characteristic G1 (ω) with a cutoff frequency
the frequency domain characteristic. An alternative solution of ω0 = π/4T (Figure 2(a)).
to the truncation problem is to select a more effective The impulse response I (t) is symmetrical about t = 0
window function, w(n), such as a Hamming, Hanning, or and of sampled (sinx)/x form (Figure 2(b)). We next decide
Keiser window. how many impulse response terms we can accommodate in
the final design, and multiply by a window function with
H (n) = h(n)w(n) (7) this number of terms.
Finite Impulse Response (FIR) Digital Filters 1029
G 1(jw) h(n)
0.6
w
(a) w=0 p/4T 0.4
I1(t )
0.2
t n
(b) t=0
10 20 30 40 50
wn −0.2
n = 10
t Figure 3. Truncated impulse response of an ideal low-pass filter.
(c) t=0 (Delay is 25 samples, total length is 51 samples and cutoff
frequency is ω0 = π/2.)
I2(t ) = w nI1(t )
The corresponding impulse response is
t
1 ω jω(n−α)
(d) t=0
e dω
2π −ω
hd (n) = (15)
sin[ωc (n − α)]
I2(t ) , n = α
(shifted) π(n − α)
t
Clearly, the impulse response has infinite duration. To
(e) t=0 create a finite duration linear phase causal filter of length
N , it is necessary to define h(n) = hd (n)w(n) and α =
|G 2(jw)|
Rectangular (N − 1)/2. Figure 3 shows a plot of h(n) for a rectangular
Hanning window where N = 51, and ωc = π/2.
w
(f) w=0 w/4p
6 SUMMARY OF DIGITAL FILTER
Figure 2. Transversal filter design by window method. METHODS
A Hamming window can be used that is defined as: It is useful to summarize the methods and techniques
presented in this part on digital data filtering. It started with
0.54 + 0.46 cos nπ/N, −N < n < n a comment on filter classifications. Several types of filters
w(n) = (13)
0, elsewhere have been distinguished according to the signal form and
This has (2N − 1) terms. Use 21 transversal multipliers frequency range.
so that N = 11. The impulse response can be shifted It continued with the description of the filter in the
forward (Figure 2(e)) so that it begins at t = 0. This makes frequency domain by its frequency response transfer func-
the filter physically releasable and converts the original tion, and in the time domain by its impulse response. The
zero-phase filter into a pure linear-phase one. The resulting Butterworth and Chebyshev filters have been presented as
frequency response magnitude characteristic is shown in examples of an ideal filter approximation.
Figure 2(f). The general performance of digital filters has been out-
lined using two different methods of digital filter realization
and two different ways of the filter implementation. The
5 DIGITAL FILTER WITH method of filter amplitude and phase spectra estimation
from its zero-pole configuration has been presented.
LINEAR-PHASE SPECTRUM
Low-pass and high-pass filters are described as typical
examples of the digital filters in the z-plane. Their mag-
This time, the design filter has to achieve a linear-phase
nitude response characteristics have been considered. The
spectrum.
problem of determining an appropriate set of specifications
The desired frequency response is defined as
−jωα with a low-pass filter as the example has been given.
e , |ω| < ωc Filters with infinite impulse response, IIR, and filters with
Hd (e−jω ) = (14)
0, elsewhere finite impulse response, have been described. Since many
1030 Elements: C – Data Acquisition and Processing Systems
practical IIR filters are based upon analog equivalents, the FURTHER READING
design problem is to find a suitable transformation for map-
ping the s-plane poles and zeros of the analog filter into the Antoniou, A. (1993) Digital Filters: Analysis, Design and Appli-
z-plane. The impulse invariance method or bilinear trans- cations, McGraw-Hill Companies, New York.
formation has been introduced. The design principles and Baher, H. (1993) Selective Linear-Phase Switched-Capacitor and
procedures for both methods have been presented. Sev- Digital Filters, Kluwer Academic Publishers, New York.
eral examples of low-pass Butterworth filter design have Chen, Chi-Tsong (2000) Digital Signal Processing: Spectral Com-
been shown. putation and Filter Design, Oxford University Press, Oxford.
The coverage also includes the concept of the moving- Datta, J., Karley, B., Lane, J. and Norwood, J. (2000) DSP Filter
average filter as an example of an FIR filter. FIR filter Cookbook, Delmar Learning, Clifton Park, New York.
design based on a window function is the most useful form. Hamming, R.W. (1998) Digital Filters, Dover Publications,
Design examples have been given. London.
For further information on digital filter design, consult Jackson, L.B. (1996) Digital Filters and Signal Processing: With
the bibliography given in Article 147, Ideal Digital Filter Matlab Exercises, Kluwer Academic, Dordrecht.
Approximation, Volume 3. Mano, M.M. (2001) Digital Design, Prentice Hall, New Jersey.
Parks, T.W. and Burrus, C.S. (1987) Digital Filter Design, Top-
ics in Digital Signal Processing, Wiley-Interscience, New
RELATED ARTICLES York.
Phillips, L.C. and Nagle, T. (1994) Digital Control System
Article 125, Outline of Purpose of Analog Data Filters, Analysis and Design, Pearson Education, Upper Saddle River,
Volume 2; Article 133, Amplifiers and Filters for DAS, NJ.
Volume 3; Article 147, Ideal Digital Filter Approxima- Rorabaugh, C.B. (1993) Digital Filter Designer’s Handbook:
tion, Volume 3; Article 149, Low-, High-, and Band-pass Featuring C Routines, McGraw-Hill, New Jersey.
Digital Filters, Volume 3; Article 150, Finite Infinite Schlichtharle, D. (2000) Digital Filters: Basics and Design,
Impulse Response (IIR) Digital Filters, Volume 3. Springer-Verlag, Telos, Berlin.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
This article is based in part on a chapter by J. McGhee et al originally published in Measurement Data Handling: Hardware Techniques
Vol. 2, J. McGhee et al , Technical University of Lodz, 2001.
Fundamentals of the Stored Program Digital
152:
Computer
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
to control almost everything. Hence, control is central to
1 Data and Information 1033 the question of data and information. More on the differ-
2 Basic Organization of the Stored Program ences between data, information, and knowledge is given
Computer 1033 in Article 13, Signals, Information and Knowledge, and
3 Purpose and Format of Macroinstructions 1035 Meaning, Volume 1.
Related Articles 1035
References 1035
2 BASIC ORGANIZATION OF THE
STORED PROGRAM COMPUTER
1 DATA AND INFORMATION
The stored program computer (Conger, 1999; Gaonkar,
Data are the basic elements used to represent the formation 1999; Karalis, 1997; Wilkinson and Makki, 1992), which
or measurement of facts, which are described by numbers, is an indispensable component in measurement systems
characters, assertions and so on, in a digital form. Data by (Lang, 1987, 1991; Morrison, 1997) is also important in
themselves have little if any meaning. Only processed data measurement systems design (Sydenham, Hancock and
gives meaningful information, as represented in Figure 1. Thorn, 1989). A common basic architecture for a stored
The raw material to gain knowledge is information, which program computer, which is shown in Figure 2, consists
is extracted from data. of three main subsystems. A central processing unit, or
Consider the example of the manufacture of a light CPU, executes instructions. The set of instructions make up
bulb to illustrate this transformation from data to infor- the instruction set. Instructions, which are to be executed
mation. The information required is whether the lamp is in a prescribed manner, form part of a program stored in
either too bright or too dull. Two pieces of information main memory.
are required to perform the assessment. In the first place, Main memory may be divided into two parts. A primary
a standard brightness is required and in the second, the memory is used to store the program to be executed as
actual brightness must be measured. Obtaining information well as the data to be operated on. This primary memory,
about the lamp’s brightness means that knowledge about the which is normally of a semiconductor form, is referred to as
lamp has been acquired and a specification of uncertainty random access memory, or RAM. It consists of a volatile,
becomes possible. In a wider context, this knowledge can read/write memory, or R/W memory, and a nonvolatile read
assist in production control so helping in running the busi- only memory, or ROM.
ness, in improving health and safety, or in making strategic Secondary memory, which is often called a backing store,
decisions about long-term planning among many other pos- may have different recording forms. Of the three kinds of
sibilities. Such a view means that information is necessary magnetic recording forms, the first kind is called the hard
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
1034 Elements: C – Data Acquisition and Processing Systems
Memory Memory
Address Address
Data
Data processing Information Fetch next Data/ Data/
Operands
instruction instruction result
CPU CPU
Operand/result
Instruction register location Instruction
Figure 1. A processor converts data to information.
(a) (b)
Figure 3. The fetch principle in (a) and the execute principle in
Input Memory Output
(b) of stored program operation.
disk drive, or HDD, which is the main operational form
Processing of secondary memory. The other two magnetic recording
ALU
forms are either removable floppy disk drives or remov-
able tape drives. Removable floppy diskettes may be used
Control for transferring programs and data from one computer to
another. Originally, they took the form of 5.25-inch platters.
Figure 2. Architecture of a stored program computer. Later, the 3.5-inch high density disk, first developed by
Memory
1st 2nd
operand operand Result
OP Operand Operand Result Instruction
code address address address address
Arithmetic and
Logic Unit
(ALU)
(a)
OP Operand Operand Result or
code address address instruction address
(b)
Memory
1st 2nd
operand operand
OP Operand Operand or
code address instruction address
Arithmetic and
Logic Unit
(ALU)
(c)
Memory 1st
operand
OP Accumulator
Operand or instruction address
code
2nd Operand
Arithmetic and
Logic Unit
(ALU)
(d)
Figure 4. The four macroinstruction formats.
Fundamentals of the Stored Program Digital Computer 1035
the Sony Corporation, became the most common form of A further reduction is possible if the destination address
removable media for backing store. is assumed. Figure 4(c) can be interpreted as ‘take operands
Unfortunately, these media are not all interchangeable or 1 and 2 from the addresses given and return the result
even compatible with the 3.5-inch, 1.44 Mb type. to operand 2 address’. This is somewhat restrictive since
Tape systems are almost all exclusively used for the memory data is overwritten. An alternative approach is to
purposes of backing up either the complete contents or include a separate register to store the result. This is called
selected parts of HDD contents. This is mainly used for the accumulator.
data security purposes. However, it can also be used as a The presence of the accumulator, as shown in Figure 4(d)
means of fast recovery of original Operating System, or OS, can be used to produce a one-address instruction format,
installations in the event of irrecoverable system crashes. where one of the source operands comes from instruction
INPUT DEVICES and OUTPUT DEVICES are interfaces and the other is the accumulator. The result is returned
to video display units, or VDUs, floppy disk drives, and to the accumulator. In practice, a register file is used for
other peripherals. providing a number of detectable registers determined by
The basic mode of operation, consisting of two separate the op-code.
control cycles as shown in Figure 3, are
• FETCH CYCLE when instructions are fetched,
• EXECUTE CYCLE when the instruction is processed. RELATED ARTICLES
More details are given in Article 154, Internal Opera-
tion of the Microprocessor, Volume 3. Article 78, Embedded Systems, Volume 2; Article 153,
Instructions are executed sequentially. The memory is Single Address Instruction Microcomputer, Volume 3;
used to store both the program and data. It should be Article 158, Microcontroller Systems, Volume 3; Arti-
noted that alternative computer architectures are possible. cle 159, Designing and Building Software for Measuring
These depend on data availability and not the order of the Systems, Volume 3.
instructions. Thus, they can operate concurrently and are
known as data flow computers.
REFERENCES
3 PURPOSE AND FORMAT OF
Conger, D. (1999) Fundamentals of Microcomputers for Technol-
MACROINSTRUCTIONS ogy Students, Prentice Hall, NJ.
Gaonkar, R. (1999) Microprocessor Architecture, Programming,
A number of instruction formats are available depending and Applications with the 8085, Prentice Hall, NJ.
on the machine design and the compromises made. The Karalis, E. (1997) Digital Design Principles and Computer Archi-
most general format is shown in Figure 4(a), where an tecture, Prentice Hall, NJ.
operational code, or op-code, is specified along with the Lang, T.T. (1987) Electronics of Measuring Systems, John Wiley
addresses of the Arithmetic and Logic Unit, or ALU, source & Sons, UK.
operands, and the destination of the result. The next instruc- Lang, T.T. (1991) Computerized Instrumentation, John Wiley &
tion address can also be specified. Thus, data transfer and Sons, UK.
control can occur together. This format is used for micro- Morrison, T. (1997) The Art of Computerised Measurement,
programming, which is a sequence of microinstructions Oxford University Press, Oxford.
used to execute one top level machine instruction. Sydenham, P.H., Hancock, N.H. and Thorn, R. (1989) Introduc-
If the program control and data processing instructions tion to Measurement Science and Technology, John Wiley &
are separated, it is possible to reduce the instruction fields Sons, UK.
by one. Figure 4(b) shows this format of separate control Wilkinson, W. and Makki, R. (1992) Digital System Design,
and processing instructions. Prentice Hall, Europe.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
153: Single Address Instruction Microcomputer
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
buffer, or ABF register, while instructions are loaded into
1 Data Width and Address Range in Computer the instruction register, or IR, for use by the control unit.
Buses 1036 Data transfer between registers occurs via an internal or
2 Instruction Set 1036 local processor bus. Control lines from the control unit are
3 External Buses – Synchronous and available for interfacing to external devices.
Asynchronous Transfers 1036
4 Overcoming Problems in the Synchronous
Protocol 1038
2 INSTRUCTION SET
Related Articles 1038 The instruction set for the processor can be divided into the
References 1038 following:
• TRANSFER: movement of data from one location to
another such as register or memory locations,
1 DATA WIDTH AND ADDRESS RANGE • ARITHMETIC AND LOGICAL: ADD, SUB, OR,
IN COMPUTER BUSES AND, NOT, and so on,
• SHIFT AND ROTATE: moving each bit in a binary
Figure 1 shows a typical microprocessor architecture con- number,
sisting of an arithmetic logic unit (ALU), accumulator and • JUMP/BRANCH: alters sequence of instructions includ-
register file, which contains the stack pointer, or SP, and the ing program control determined by conditions of the last
program counter, or PC. Microcomputers usually handle 8, ALU result,
16, 24, 32 or 64 bit wide data words. It is obviously uneco- • INPUT/OUTPUT: transfer from processor registers to
nomical to choose a 32-bit bus for use in a system intended external devices,
for 8 bits (Holdsworth, 1987; Lang, 1991; Rangan, Sarama • CONTROL: interrupts, bit test, external conditions
and Mani, 1983; Tocci and Laskowski, 1987; Wilkinson and affecting the instruction flow.
Makki, 1992). Component level buses, which are merely
extensions of the local processor bus to a back-plane, nor-
mally have 16 to 20 address lines giving 64 kB to 1 MB of 3 EXTERNAL BUSES – SYNCHRONOUS
addressable RAM. Although an 8-bit bus system may be AND ASYNCHRONOUS TRANSFERS
adequate for small process or machine tool systems, it is
totally inadequate for modern business applications, which Information must be distributed or transferred within the
usually include versatile graphic facilities. Hence, allow at system. The various bus signals, which are used for this
least 24 address lines up to say 32 or even 64. purpose, are divided into three main groups. One group,
The external data bus and external address bus are called the data bus, is used to carry data from and to
buffered by a data buffer, or DBF register and an address various locations within the system. This data needs to be
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Single Address Instruction Microcomputer 1037
Control signals
Read/write
Memory
Data bus Address bus
Control bus Address buffer
Data buffer (DBF) (ABF)
Internal/local bus
Instruction Accumulator
register
R S Register file
Arithmetic/
Control logic unit
unit (ALU) R
Stack pointer (SP)
Program counter (PC)
Single Chip Microprocessor
Figure 1. Operational organization of a microprocessor using a single address instruction format.
directed. For this reason, a second bus called the address Clock
bus is essential. The whole system needs to be controlled.
Data
A control bus carries the control signals utilized by the
system to direct, time, and order the sequential proceedings Strobe
of program execution.
Hold
The basic methods of information transfer use either
(a)
SYNCHRONOUS bus transfers or ASYNCHRONOUS bus
transfers. Reliable data transfer requires a data transfer Memory
protocol or defined set of rules. Three main problems Processor
Memory Wait state
must be overcome. When a master gates several parallel decode generator
data or address lines with the signals, lack of synchronism
causes problems. This is called skew . These problems are Address MREQ
compounded by the differences in the rise times of each WAIT
(b)
separate line signal. Moreover, the switching point of bus
transceivers must be taken into account. Figure 2. Synchronous protocol with hold and wait cycle:
In synchronous bus transfers, synchronization of master (a) signal timing diagram and (b) organization of the Z80.
and slave is timed as in Figure 2(a) using the circuit of
Figure 2(b). Information on the direction of transfer, to or
from the processor, the transfer timing, and whether it is not operate faster than the speed of the slowest device.
a memory or input/output operation, is required. Data are However, since this can be restrictive, introducing a wait
placed on the parallel bus lines. After a delay, which is signal, or ready signal, can alleviate this problem. A wait
sufficient to cover the maximum skew and rise time dif- state generator, contained within the slow memory being
ferences, a stable strobe pulse is generated. The processor accessed, activates the wait line to the processor causing
generates a request signal, together with data and address it to introduce dummy clock cycles in the middle of the
signals. When the appropriate device receives these sig- data transfer period. Hence, when the slave is slow, it gen-
nals, data is accepted or generated. The slave responds by erates a hold pulse telling the master to insert wait cycles
completing a write operation to memory or some other until the slave action is completed. Performance is limited
interface. Although the timing is known in advance and by two main factors. In the first case, a clock pulse can-
taken into account by the processor, the processor must not be changed once it has been defined. This implies that
1038 Elements: C – Data Acquisition and Processing Systems
the system cannot take advantage of faster circuits as they the processor. After first placing data on lines, the master
become available. In the second case, slow devices need to waits until its own transmitter skew effects have settled and
be synchronized. Hence, a hold pulse has to be generated asserts DS*. When the slave sees DS* asserted, it waits
very quickly. This requires fast logic so that the master can until its own receiver skew effects have settled. It then
decide whether to proceed with the next data cycle or to waits for a long enough time either to accept the data or to
insert dummy wait cycles. place new data on the relevant lines before asserting DK*.
The above information is provided by control signals, Meantime, the master holds data steady until it sees DK*
which are usually active low or ‘0’. For example, in the asserted. When this occurs, it then releases both data and
Z80, the signals are RD, WR, MREQ (memory request), DS* or waits until it has accepted the slave’s data before
IORG (input/output request). releasing DS*. The slave holds its data until it sees DS*
In the Motorola 6800 series, they are VMA or valid released. Finally, it then removes its data before releasing
memory address, R/W, E or φ. E or φ is the general DK* in preparation for the next cycle. The data transfer
processor clock with which timing is synchronized. rate is automatically adjusted to allow the fastest possible
rate for the receiver and transmitter modules involved. The
disadvantage of this method is that extra control signals are
4 OVERCOMING PROBLEMS IN THE required. Propagation delays of these extra control signals
SYNCHRONOUS PROTOCOL may slow the speed of processor operation.
In the case of the Motorola 6800, both synchronous
Problems that occur in the synchronous protocol are over- and asynchronous signals are present. The asynchronous
come by the inclusion of full handshaking signals: data are AS (address strobe), UDS upper data strobe D08-
strobe (DS*) and data acknowledge (DK*). The operations D15, LDS lower data strobe D01-D07, R/W, DTACK data
proceed in the sequence outlined in Figure 3(a). Two con- acknowledge. The synchronous signals are similar to those
trol lines are utilized. One from the processor initiates the of the 6800.
start of data transfer, while the other, from the destination,
which is the memory in this case, triggers completion of
the transfer. Thus, the total transfer time is always dictated
by the returning signal, whose absence would indicate a
RELATED ARTICLES
failure in the peripheral device. Compare this to WAIT
state generation in which failure to generate is ignored by Article 152, Fundamentals of the Stored Program Dig-
ital Computer, Volume 3; Article 154, Internal Opera-
tion of the Microprocessor, Volume 3.
DS*
DK*
Master data
REFERENCES
Slave data
Holdsworth, B. (1987) Microprocessor Engineering, Butterworth
Slave accepts data Master accepts data acknowledge
(a) and Co., London.
Lang, T.T. (1991) Computerized Instrumentation, John Wiley &
AS, UUS, LDS, R/W Sons, UK.
DTACK Rangan, C.S., Sarama, G.R. and Mani, V.S. (1983) Instrumenta-
tion Devices and Systems, Tata McGraw-Hill, New Delhi.
Processor Memory/
(b) input/output Tocci, R.J. and Laskowski, L.P. (1987) Microprocessors and
Microcomputers, Prentice Hall International.
Figure 3. Asynchronous protocol: (a) signal timing diagram and Wilkinson, W. and Makki, R. (1992) Digital System Design,
(b) organization of the Motorola 6800. Prentice Hall, Europe.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
154: Internal Operation of the Microprocessor
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
would be transferred from register A via the internal bus to
1 Fetching and Executing 1039 the data bus.
2 Register Transfer Logic, or RTL 1039
Related Articles 1040
2 REGISTER TRANSFER LOGIC, OR RTL
References 1040
The transfer of data between registers can be specified using
a register transfer notation. For example, transferring the
1 FETCHING AND EXECUTING contents of register B to register A could be indicated by
The operation of the processor can be divided into two A ←−−− B
cycles known as FETCH and EXECUTE (Conger, 1999;
Such a notation has found application in hardware
Gaonkar, 1999; Karalis, 1997; Heffer, King and Heath,
description languages, or HDLs. They can be used to
1981; (Article 153, Single Address Instruction Micro-
describe the operation of complex digital systems in which
computer, Volume 3); Wilkinson and Makki, 1992). It
other techniques such as state diagrams are inappropriate.
should be noted that all of the following operations are
Prefixing the transfer with a colon can include control
essentially register-to-register transfers. They are synchro-
signal operation
nized to the system clock. Figure 1 shows the control timing
pulses, or T states, being derived from the system clock. T1: A ←−−− B
In the FETCH cycle, instructions are brought from the
external primary memory to the internal instruction register Thus, transfer takes place when T1 is high. Also
of the processor. There are several distinct steps in this
procedure. The output address is placed on the address T1 × φ: A ←−−− B
bus, while appropriate control signals are established on
the control bus. The instruction is transferred from the indicates transfer when both T1 and φ are high.
data bus to the instruction register after a suitable delay or Arithmetic operations in the ALU can also be specified.
alternatively after WAIT states or data acknowledge. When Thus, if register B goes to the ALU R-BUS and C to the
this is completed, the program counter is incremented by 1. L-BUS, then an ADD operation gives
Subsequently, other instruction bytes are fetched if required.
To illustrate the EXECUTE cycle, consider the imple- T1: A ←−−− B + C
mentation of the Z80 instruction LD A, (100). Firstly, the
address 100 is placed onto the address lines, while the con- Internal processor operation can be specified in terms of
trol signals are placed on the control bus. Data is loaded this notation. As an example, consider the Z80 processor.
into the processor, before subsequent internal bus transfer Assume that the control signals are generated by the loading
to register A. If the instruction were LD (100), A the data of one bit registers corresponding to MI (Instruction fetch),
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
1040 Elements: C – Data Acquisition and Processing Systems
Clock, f
A write-from-a-register such as LD (100), A
T1
ET1: ABUS ←−−− ABF
T2
ET1: MREQ ←−−− 0, DBUS ←−−− DBF ←−−− A
T3
ET2: WR ←−−− 0,
T4 ET3: MREQ ←−−− 1, WR ←−−− 1
T5
T1 DBUS ←−−− Zhigh (T1 of next cycle)
Figure 1. Control state timing for the instruction execute phase. Arithmetic operations can also be described. These
include ADD A, B, which uses the following operations.
MREQ, RFSH (dynamic RAM refresh), RD, and WR ETl: ALUR ←−−− A
Processor control inputs include φ (system clock) and
WAIT . Registers are loaded on the rising edge of the ET2: ALUS ←−−− B
specified control signal, that is, T1, T3, T3, and so on. ET3: A ←−−− ALU
For instruction fetch, the following transfers would be
required; Conditional microoperations might also be required, for
example, interrogating the WAIT line until it goes to 1.
FT1: ABUS ←−−− ABF ←−−− PC, 91- ←−−− 0 Others are required for implementing conditional branch
FT1: PC ←−−− PC + 1 instructions depending on the STATUS of the ALU. In
register transfer logic (RTL), a typical statement is
FT1: MREQ ←−−− 0, RD ←−−− 0
FT3: IR ←−−− DBUS T2: IF WAIT = 0 THEN T2 ELSE T3
FT3: ABUS ←−−− RI, MREQ ←−−− 1, RD ←−−− 1, T3:
M1 ←−−− 0, RFSH ←−−− 0 The T2 cycle is repeated until WAIT = 0.
FT3: MREQ ←−−− 0, R ←−−− R + 1
FT4: MREQ ←−−− l
RELATED ARTICLES
FT1: RFSH ←−−− 1
Article 152, Fundamentals of the Stored Program Dig-
For a memory-to-register read of the form LD A, (100) ital Computer, Volume 3; Article 153, Single Address
Instruction Microcomputer, Volume 3; Article 155,
ET1: ABUS ←−−− ABF External Operation of the Microprocessor, Volume 3.
ET1: MREQ ←−−− 0, RD ←−−− 0
ET3: MREQ ←−−− 0, RD ←−−− 1, A ←−−− DBUS REFERENCES
To obtain a literal of the form LD A, n from PROGRAM Conger, D. (1999) Fundamentals of Microcomputers for Technol-
MEMORY, the following would occur after the op-code ogy Students, Prentice Hall, NJ.
had been fetched. Gaonkar, R. (1999) Microprocessor Architecture, Programming,
and Applications with the 8085, Prentice Hall, NJ.
ET1: ABUS ←−−− ABF ←−−− PC (PC points to next
Karalis, E. (1997) Digital Design Principles and Computer Archi-
instruction byte)
tecture, Prentice Hall, NJ.
ET1: PC ←−−− PC + 1 Heffer, D.E., King, G.A. and Heath, D.C. (1981) Basic Principles
and Practice of Microprocessors, Edward Arnold, London.
ET3: MREQ ←−−− 0, RD ←−−− 0
Wilkinson, W. and Makki, R. (1992) Digital System Design,
ET3: MREQ ←−−− 1, RD ←−−− 0, A ←−−− DBUS Prentice Hall, Europe.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
155: External Operation of the Microprocessor
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
interfaces for controlling the I/O device, interrogating its
1 Input and Output–Programming and Types 1041 status, interrupting the processor, and sending an interrupt
2 Interrupts and Interrupt Servicing 1042 vector. Handshake lines such as data strobe/acknowledge,
or STB, and data ready, or RDY, ensure controlled data
3 Direct Memory Access 1043
transfers at the port input.
Related Articles 1044 For SIO, the information is sent along one wire. It
References 1044 consists of a number of bits separated in time. The
Further Reading 1044 following terms, which are described in more detail
in Article 165, Introduction to Networked Instrumen-
tation, Volume 3; Article 166, Instrument Interconnec-
1 INPUT AND OUTPUT– tion, Volume 3; Article 167, Asynchronous and Syn-
chronous Interface Protocols, Volume 3; Article 168, RS
PROGRAMMING AND TYPES 232 and EIA/TIA 232 Serial Interface, Volume 3; and
Article 169, Voltage and Current Loop Transmission,
There are four parts to a microprocessor system, the Volume 3 relate to serial transmission.
microprocessor, which is an integrated circuit, or IC, cen-
tral processing unit, or the microprogram, data memory, • Simplex : data is sent in one direction only,
which is also usually fabricated in IC technology, and • Half duplex : data is sent in both directions one at a time,
the input/output circuitry. The last two topics will now • Full duplex : data is sent in both directions simultane-
be considered. ously.
Data transfer to and from the input/output devices, using
processor machine instructions, requires that checks be Serial signals may also be transmitted synchronously or
made to ascertain that the I/O device, or input–output asynchronously. In the synchronous serial interface, a block
device, has data ready before a read, and is ready to of characters is transmitted one bit at a time without stop-
accept data before data is sent. This is known as hand- bits. It is 20 to 27% faster than asynchronous transmission
shaking (Heffer, King and Heath, 1981; Holdsworth, 1987; for the same bit rate. Synchronization characters, such as
(see Article 153, Single Address Instruction Microcom- insert 3 characters are sent at the start and end of transmis-
puter, Volume 3); Tocci and Laskowski, 1987; Wilkinson sion, and periodically in between characters. The synchro-
and Makki, 1992). A status register, which is polled, can nizing characters must not be valid characters. In ASCII, or
be used to indicate the state of readiness of the inter- American Standard Code for Information Interchange, the
face (Doyle, 1999). SYN (0010110) character cannot be used. Parts available
The two types of I/O are parallel I/O, or PIO, and include USARTs, or Universal Synchronous/Asynchronous
serial I/O, or SIO. In PIO, data is sent in parallel at 1 bit Receiver/Transmitters and SIOs.
per line. Common interfaces such as PIOs and versatile The ASCII coded signal shown in Figure 1 is a
interface adapters (VIAs) have the necessary processor standard asynchronous format. Transmission rates up to
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
1042 Elements: C – Data Acquisition and Processing Systems
One character Next character
INTREQ
Data
MSB
Micro- Device External
LSB
processor interface device
INTACK
0 0 1 0 0 1 0 1 1 1
Start bit Stop bit
ASCII (R)
Parity bit
[1010010] Data bus
Figure 1. An ASCII coded signal. (a)
INTREQ INTACK
9600 bps or 960 characters per second are common for Received here xxxxxx
video terminals. Parts used include universal asynchronous xxxxxx At the end of
receiver/transmitters, or UARTs, or asynchronous commu- current instruction
nications interface adapters, or ACIAs, for microprocessors.
2 INTERRUPTS AND INTERRUPT
SERVICING
Return address
Programmed I/O has the disadvantage that the processor EI stored on stack
must continually poll the I/O status register to see if RETI
data has to be sent or received. This slows down the (b)
processor. An alternative is to interrupt the processor during
Figure 2. Interrupts in microprocessors: (a) interrupt request
normal execution, causing it to service the I/O device.
signals; and (b) interrupt handling.
Special signals, such as interrupt request and interrupt
acknowledge, are needed. Figure 2 shows the signaling
lines and the interrupt handling scheme. • Polled interrupts: All interrupt requests are wire-OR’ed
Interrupts are suited to unsynchronized I/O and randomly together for a single interrupt request, or INTREQ.
occurring I/O. It is especially suited to the multiprocess- The processor acknowledgment involves polling the
ing environment, where several programs are interleaved. devices to see which has sent INTREQ. This includes
Interrupts can also be used to respond to internal processor the following:
hardware conditions such as ALU overflow.
An interrupt service routine, or ISR, is an integral part 1. The use of an INT FLAG on each device, which
of every program, which allows the sequence of program is set on INT. Processor checks flags in order
operation to be suspended temporarily in an orderly manner.
of priority.
When one of the number of possible requests for interrupt
2. The use of an INTERRUPT REQUEST REGIS-
servicing is acknowledged, the ISR may be initiated.
TER with bits set for each device. Set order of
Interrupt requests may be due to the following:
priority of bits in register.
• An unconditional jump to a known location. This may
be fixed in the processor design and activated as non- • Daisy chained interrupts: One of the interrupt control
maskable interrupts. Several locations may be available. lines passes through each device.
• In vectored interrupts, a jump address is provided to
the processor by 1. Daisy chained interrupt acknowledge: As shown
in the circuit of Figure 3, the highest priority is
1. defined memory location holding SR address, closest to the processor that receives the request
2. interrupting device providing SR address as a directly. The interrupting device blocks the inter-
vector. rupt acknowledge, or INTACK, from the processor
Multiple interrupts may be caused if several devices to devices with lower priority. Vectored ISRs can
are used to interrupt the processor. A priority interrupt be used to identify the interrupting device.
number must be given to each so that simultaneous inter- 2. Interrupt request: The lowest priority device is
rupts can be resolved. Priority can be established by closest to the processor. There is a common ack-
the following: nowledge to all devices.
External Operation of the Microprocessor 1043
INTREQ
INTACK Device 1 Device 2
interface interface
To next device
INTREQ 1 INTACK 1 INTREQ 2 INTACK 2
Decreasing priority
Figure 3. Handling interrupts using the daisy chained method.
INTREQS
Device
DMA interface
INTREQ Micro- DMA Device
RAM
processor controller interface
INTMASK
Data
Address
Control
Figure 4. Added flexibility using an interrupt mask. Figure 5. Use of a DMA controller.
Figure 4 illustrates the added flexibility of INTERRUPT the processor by receipt of the starting address in memory
MASKING. Higher priority devices can be selectively
plus the number of bytes to be transferred. The controller
enabled immediately after accepting an INTREQ.
first captures the system data, address, and control buses,
Interrupt controllers are devices that can
followed by transfer of the required number of bytes
• handle multiple interrupts outside the processor, before signaling the processor that transfer is complete. The
• can support eight levels cascadable to more, processor, which relinquishes the bus on receipt of a bus
• can also supply interrupt vectors, request, or BREQ, at the end of the current instruction,
• can operate with fixed or rotating priority. cannot use the bus again until transfer is complete. To allow
the processor to continue operating, the DMA controller can
Rotating priority means that after the highest priority ‘steal’ processor cycles, called processor stealing, to allow
interrupt is serviced, it is then given the lowest priority. program execution.
The next to highest becomes highest and so on. Priority The controller can operate in a variety of ways to
therefore rotates thus preventing an interrupt dominating at
the expense of others, which might not be serviced at all.
• fetch setup addresses and byte counts held in exter-
This technique is also used in multiple bus-master systems.
nal memory,
• take data from source and compare it with the contents
of a preloaded DMA register to find a data match,
3 DIRECT MEMORY ACCESS
• detect when a data match is not found. This is useful
in memory testing where the same word is loaded into
Processor controlled transfers, such as memory to I/O,
each location. If a different word is detected, an error
means that the processor cannot do other things. The
is present at that location.
speed of transfer is limited by the execution times of
machine instructions. For large amounts of data, this is time
consuming. A direct memory access, or DMA, controller, DMA is useful for devices such as magnetic disk memory
in the structural arrangement given in Figure 5, may be units where a data block must be transferred as a constant
used as an attractive alternative. It is programmed from stream and at a rate dictated by the shaft speed of the disk.
1044 Elements: C – Data Acquisition and Processing Systems
RELATED ARTICLES Holdsworth, B. (1987) Microprocessor Engineering, Butterworth,
London.
Tocci, R.J. and Laskowski, L.P. (1987) Microprocessors and
Article 78, Embedded Systems, Volume 2; Article 152, Microcomputers, Prentice Hall International.
Fundamentals of the Stored Program Digital Computer,
Wilkinson, W. and Makki, R. (1992) Digital System Design,
Volume 3; Article 154, Internal Operation of the Prentice Hall, Europe.
Microprocessor, Volume 3; Article 158, Microcontroller
Systems, Volume 3.
FURTHER READING
REFERENCES Lang, T.T. (1987) Electronics of Measuring Systems, John Wiley
& Sons, UK.
Doyle, L. (1999) Computer Peripherals, Prentice Hall, NJ, Lang, T.T. (1991) Computerized Instrumentation, John Wiley &
USA. Sons, UK.
Heffer, D.E., King, G.A. and Heath, D.C. (1981) Basic Principles Rangan, C.S., Sarama, G.R. and Mani, V.S. (1983) Instrumenta-
and Practice of Microprocessors, Edward Arnold, London. tion Devices and Systems, Tata McGraw-Hill, New Delhi.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
156: Memory Management in the Microprocessor
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
2 PRIMARY AND SECONDARY MEMORY
1 Introduction to Memory Management WORD BLOCK TRANSFER
Systems 1045
2 Primary and Secondary Memory Word Block In more sophisticated management systems, automatic
Transfer 1045 replacement of memory blocks in primary memory with
3 Replacement Algorithms 1046 blocks obtained from secondary memory is desirable. Such
4 Segmented System 1047 systems require both the actual address used by the memory
Related Articles 1048 units and the program address produced by the processor.
Further Reading 1048 Translation between the two addresses is achieved by hard-
ware and software. During program transfer, actual memory
addresses may change but the program addresses used by
the processor will remain the same. The two types of sys-
1 INTRODUCTION TO MEMORY tems deal with memory in the form of pages of memory or
segments of memory.
MANAGEMENT SYSTEMS In paged or virtual memory systems, the actual address
is the real address and the program address is the virtual
When running large programs or when operating on data address. A page is a memory block of fixed length of 128,
files that do not fit completely into the available primary 256, 512, 1024 words. An address is composed of a page
memory, a memory management scheme must be imple- number and a line number. Thus, there are two fields in the
mented to control the transfer of information between the address as shown in Figure 1(a).
system’s secondary, or disk memory, and primary or ran- Memory management units, or MMUs, must provide
dom access memory (RAM). hardware to translate the virtual page number into a real
In simple systems such as microprocessors running page number. This hardware works only on the primary
MSDOS, a technique known as OVERLAYING is applied memory. For secondary memory, software replacement
where existing programs are overwritten. A large software algorithms are required to transfer pages from secondary
package is divided into a MAIN FILE and several OVER- to primary. The hardware checks to see if the page is in
LAY FILES, which overwrite the main file. It may be primary memory. If not, software must search secondary
necessary to restore the main file from time to time. Thus, memory from which data is transferred to either a blank
a common problem is caused by interaction between pro- page or overwrites an existing one. Note the number of
gram sections in different overlay files, which can cause virtual addresses can exceed the actual number of available
disk threshing owing to excessive disk to memory trans- addresses in both memories as shown in Figure 1(b).
fers. Also, errors in different overlays can be difficult to Address translation is achieved using one of two tech-
detect. niques. In direct mapping, as shown in Figure 2(a), all
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
1046 Elements: C – Data Acquisition and Processing Systems
Field 1 Field 2
Page Line
5 bits 9 bits (Many more in practice)
32 pages 512 words
(a)
Real page addresses
8 9 10 11
Virtual page addresses
12 23 30 0
0 1 2 3
12 13 14 15
9 3 24 31
18 2 29 4
4 5 6 7
16 17 18 19
16 14 19 -
7 15 8 27
Main memory of 8 pages
20 21 22 23
25 - - 5
24 25 26 27
10 26 1 11
28 29 30 31
- 22 13 17
(b) Secondary memory of 24 pages
Figure 1. Memory management: (a) the two-field address structure and (b) translation of virtual to real page number.
Virtual address the real page addresses are stored in high speed RAM
Page Line
whose addresses correspond to the virtual page addresses
and whose contents provide the real page addresses. The
Use bit Same line numbers line numbers are the same in both the real and virtual page.
Protection bit
Associative mapping uses a content addressable memory,
or CAM, as shown in Figure 2(b). Here, a location is iden-
Real page tified by its contents. The input data word is applied to the
Real address
inputs of the CAM, which simultaneously compares data in
(a) Page Line all memory locations. The location containing the data is
flagged as output. This memory, which uses bipolar technol-
Virtual address ogy, is fast, with access times of typically 30 ns. However, it
Page Line is expensive. It is linked to a RAM, which contains the real
addresses. Thus, the processor sends the virtual address; it
Use bit
Protection bit
is compared with all others, and if a match is found, the
real address is accessed from the RAM.
Virtual Real
Real address
3 REPLACEMENT ALGORITHMS
Page Line
UB PB The transfer of data between primary and secondary memo-
Compare with all Content addressable memory ries can be achieved using a number of software algorithms.
(b) stored virtual addresses
In random replacement, pages to be replaced in primary
Figure 2. Address translation methods: (a) the direct mapping memory are chosen at random. This is not very good, since
method and (b) the associative mapping method. most programs access locations in proximity to each other.
Memory Management in the Microprocessor 1047
First-in-first-out-replacement (FIFO) is an effective strat- information is required for sector identification. For a large
egy for replacement. The incoming page replaces the page page size, transfer times are long and there is an increase in
that is resident longest in memory. A rotating pointer iden- unused space at the end of each page. This is called internal
tifies the real page in primary memory on a list where fragmentation.
replacement has to take place. If a page fault occurs, which
means the page is not in primary memory, the virtual page
address at the pointer location is replaced by the new vir- 4 SEGMENTED SYSTEM
tual page address. The counter is then incremented to the
next location. Note that in a direct mapping scheme, the
As shown in Figure 3, the memory space is divided into
contents of the new virtual page address would be loaded
different sizes of blocks, or segments, at contiguous loca-
with the real memory address of the pointer. In an associa-
tions. This technique may be better when programs and
tive scheme, the new virtual page address would be stored
data are generated in various sizes. Segments can over-
at the location corresponding to the real memory address,
lap, with the segment address consisting of a number plus
in essentially the same way.
a displacement, or offset, within the segment. The offset
First-in-not-used-first-out-replacement is a modification
and segment are added together to form the real address.
of the first-in-first-out-replacement, described above. It
The logical address is the virtual address and the physical
avoids unnecessary transfers of recently used data. A use
address is the real address.
bit is included, which is set if a page is referenced after
Translation of logical addresses uses a segment table and
initial loading (use bit = ‘0’). When a page fault occurs,
direct mapping, as shown in Figure 4. The segment table
the use bit at the current pointer location is checked. If it
can include protection bits in the form of a length field
is 1, the bit is reset and the pointer incremented to the next
location, where the bit is again checked. If the bit is zero,
the location is loaded with the virtual address. Transfer of Physical memory
pages between primary and secondary memory then occurs
Free
and the counter is incremented.
Least recently used replacement, or LRU replacement, Program 1
makes use of a list of the virtual page addresses held in
Free
primary memory. The list is updated by shifting every
existing list line down by one when each new page is Program 2
inserted at the top of the list as it is brought from primary. Program 3
The page at the bottom of the list is then replaced. If,
however, the page accessed is already on the list, it is
Shared data
moved to the top of the list, its preceding pages are moved
one location down, while those after remain unchanged.
Note that the actual pages are not reordered in primary Local data
memory. If many pages are accessed, then implementation
is difficult. In one scheme, a counter is held for each page
and regularly incremented. If the page is accessed, it is Figure 3. Division of memory into contiguous blocks.
reset. The page with the highest count is replaced. It is
expensive in hardware. In an alternative scheme, associative
Logical address
memory is used to record the primary memory pages. It is
used for the top 16 or 32 entries while the rest are held in Segment (S) Offset
the RAM. This scheme essentially implements the list.
Some general points are worth emphasizing. The use bit
Segment table
can be incorporated into the virtual address translation table
together with protection bits; for a read-only page, retain
S
page in primary memory. The page size is typically 512
words and is a compromise. A small page means that (a) the Base Physical address
transfer time is short and (b) there is a large selection of
Address of first location
different pages in primary memory. Hence, a large page in physical memory
table is required. The minimum page size is constrained by
disk sector size. For small pages, an increased proportion of Figure 4. Addition of segment number to segment table.
1048 Elements: C – Data Acquisition and Processing Systems
Logical address
Segment (S) Offset
Segment table Segment table
pointer (STP)
STP
Length Base Physical address
Flag indicates that the
segment is in primary memory
Protection bits
Figure 5. Translation of logical to real address.
Next Length
For the first fit algorithm, a word that points to the next
Pointer
free space in memory and the length of the current space,
Free as shown in Figure 6, is stored at the beginning of each free
space in memory. A pointer, P, points to the first free space.
If a replacement is required, the next/length word is read. If
Segment the length is greater than the segment to be inserted, transfer
takes place with the next/length word inserted at the end.
Meantime, the length is changed to indicate the remaining
Next Length space. The next in the word of the previous free space is
Free changed to point to the new next. If the length is less than
the new segment, the space is ignored and the next space is
checked until one of sufficient length is found. To prevent
external fragmentation, a minimum residual free space can
Segment be specified.
Next Length
RELATED ARTICLES
Figure 6. Implementation of the first fit algorithm.
Article 78, Embedded Systems, Volume 2; Article 152,
so that referencing beyond a segment is prevented. Several Fundamentals of the Stored Program Digital Computer,
segment tables can also be used. Volume 3; Article 153, Single Address Instruction
In Figure 5, the segment number, S, is added to the Microcomputer, Volume 3; Article 158, Microcontroller
segment table pointer (STP) to reach the correct address in Systems, Volume 3.
the segment table. A flag is used to indicate that the segment
is in primary memory. The offset is compared with the
FURTHER READING
segment length and if the offset is greater, a segment error
signal is generated. Otherwise, the segment base address is Gaonkar, R. (1999) Microprocessor Architecture, Programming,
added to the offset to access the physical memory location. and Applications with the 8085, Prentice Hall, NJ.
Replacement algorithms, although similar to the pages Karalis, E. (1997) Digital Design Principles and Computer Archi-
system, must take account of the varying segment lengths. tecture, Prentice Hall, NJ.
Care must also be taken to prevent external fragmentation
by leaving a lot of unusable spaces.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
157: Data Acceleration in Computers
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
or instructions, which depends upon the architecture. All of
1 Single Thread Systems 1049 the variables determining Tprog depend upon the architecture
2 Acceleration of Main Memory at the of the system used while the indexing also depends upon
Architectural Level 1050 the program.
3 Cache Memory 1051 Acceleration may be achieved at the manufacturing
or realization level of the processors. The most com-
4 Operation of the Cache 1052
monly used manufacturing technique is the Very Large
5 Memory Fetch Policies 1053 Scale Integration, or VLSI, process. For example, the Intel
6 Cache Mapping Schemes – Placement 80386 IC operating at 16 MHz has 275 000 transistors
Policies 1053 (100 mm2 ), the MC68030 integrates an MC68020 CPU with
7 Cache Mapping Schemes – Direct Mapping 1053 an MC68851 MMU.
8 Cache Mapping Schemes – Associative and It is also possible to implement acceleration techniques
Set-associative Mapping 1053 at the system implementation level. The various elements
9 Cache Mapping Schemes – Replacement involved in data manipulations together with their typical
Policies and Memory Update 1054 access times and capacities are shown in Figure 1. A
Related Articles 1055 number of terms should be explained first. Latency is
the total time between the start of an operation and the
availability of a result. The bandwidth is the number of
operations per second, expressed as say 40 Mbytes/s over
1 SINGLE THREAD SYSTEMS a bus or 2 MFlops/s for floating-point operations. The time
taken to respond to a memory request is called the memory
In many systems, where only one processor is used for latency time.
program execution, there is only one program counter. Caches improve memory latency provided they are high-
Presently, there is a tendency to include coprocessors in speed (∼50 ns). Memory bandwidth is also important. For
dense program execution systems. When this is the case, example, a memory produces one byte in 100 ns and 8 bytes
the program execution time, Tprog , is then in 800 ns. Another produces the first 2 bytes in 200 ns and
thereafter 2 bytes in 50 ns giving a total for the 8 bytes
N
of 350 ns. Which is better? This introduces the concept
Tprog = Tc × Ii Ci (1)
of pipelining, or interleaving memory paths. Parallelism
i=1
where wide data paths or multiple paths are used, is also
where Tprog is the program time, Tc is the cycle time, important. Execution Latency is the delay associated with
which depends on the architecture, Ii is the number of arithmetic operations. Having other processing units such
times instruction i is executed, which depends on both the as multipliers and coprocessors can solve this problem.
program and the architecture, and Ci is the number of cycles Execution bandwidth is again effected by pipelining and
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
1050 Elements: C – Data Acquisition and Processing Systems
Main reduced instruction set computers, or RISC, so that the
Registers Disks
memory numbers of unique instructions, addressing modes, and
Access time 5–10 ns 250 –1000 ns 5 – 60 ms
data types are lower. Eliminate large microprogrammed
capacity 512 bytes 128k –16 Mbytes 10 – 800 Mbytes controllers. Migrate complex functions to the compiler and
away from execution time. It is possible to accelerate main
Figure 1. Illustration of memory latency for registers, main memory using
memory, and disks.
• multiple memory paths,
parallelism. Pipelining ensures that independent operations • interleaving partitions,
and instructions are processed simultaneously. This means
– low-level interleaving,
that execution of the next instruction should start before
– high-level interleaving.
the first has finished. For example, adding two floating-
point numbers should be executed in the manner illustrated Multiple memory paths use the Harvard architecture
in Figure 2. of Figure 3(a) instead of the von Neumann architecture,
where instructions and data are mixed and share a common
memory. The Harvard uses two memory paths, one for
2 ACCELERATION OF MAIN MEMORY instructions and the other for data. It is usually implemented
AT THE ARCHITECTURAL LEVEL internally with data and instruction caches, while externally
it appears as a von Neumann machine.
This is illustrated in Figure 3. It is based upon using Memory may be divided into multiple physically inde-
special purpose processors for signal processing where pendent modules, as shown in Figure 3(b) and (c). Memory
the ALU and multiplier are on a single chip. Another interleaving partitions are then utilized with either low-
approach used reduces the hardware complexity by using order interleaving, illustrated in Figure 3(b), or high-order
CE2 AE2 AF2 N2 CE2 AE2 AF2 N2
CE1 AE1 AF1 N1 CE1 AE1 AF1 N1
Time (cycles) Time (cycles)
Figure 2. Principles of execution latency.
Address bus Address bus
Instruction memory CPU Data memory
(a)
n −m m m n −m
Address in module Module Module Address in module
Decoder Decoder
0.....2m − 1 0.....2m − 1
Module 2m − 1 Module 0 Module 2m − 1 Module 0
(b) (c)
Figure 3. Acceleration of main memory using accelerating structures at the architectural level: (a) multiple memory paths, (b) low-order
interleaving, and (c) high-order interleaving memory.
Data Acceleration in Computers 1051
Module Cache Primary Secondary
All Access 1 Access 2 Processor
(high speed) memory memory
Out M0 M1 ................ M2m−1 M0 Figure 5. Structural principle of cache memory.
t (Transfer of access 1) 2 t Time
bipolar memory resident between the primary memory and
Figure 4. Speeding up memory access. the processor as shown in Figure 5 can reduce this problem.
The cache of limited capacity (512 words), which con-
interleaving, illustrated in Figure 3(c). Low-order interleav- tains the current processor instructions and data, is updated
ing puts consecutive addresses in consecutive modules. It as required, by reading from the primary memory. Out-
supports spatial locality and is good for vector elements puts from the processor, which are then copied to the
and successive instructions. However, the failure of a mod- primary memory, can also update the cache. The cache
ule is catastrophic. In high-order interleaving, the data is in works because the principle of locality applies, that is,
one module. In this method, shown in Figure 4, the address instructions, and to a lesser extent data, are accessed locally
space is split into M contiguous parts. It does not support to each other, and program loops are common. Thus, there
spatial locality since all are successive, easy to expand, and are many cache references to each primary reference. If
have fault tolerance. Low order is preferred in memories the first reference is obtained from primary memory and
such as caches because it has spatial locality. transferred to cache in time M s, and there then follows n
To speed up memory access, all the modules have to references by the processor to cache in access time m s, the
process memory requests simultaneously as in Figure 4. average access time, Ta , is
For low-order interleaving, simultaneous access is possible.
High-order n − m bits are sent to the modules at the same (M + n × m)
Ta = (2)
time with buffering of the output from each module. The n
low-order bits activate a multiplexer to select each module
output in turn with the second memory access overlapping Note the assumption that the output from primary goes
with the multiplexer accesses. to cache first before being accessed by the processor. In
If all M words are accessed, memory access time is practice, output could go to both. Also, note that write
divided by M, the total number of modules. For nonsequen- operations to cache must be copied to primary memory
tial access to different modules, this scheme breaks down. at some stage. This will slow average transfer times.
Concurrent Access overcomes the low efficiency of non-
sequential access by offering to each module different For M = 400 ns, m = 50 ns and
high-order addresses, which are processed at the same n = 10, then Ta = 90 ns rather than 400 ns.
time. Buffers now hold the memory addresses for each
module allowing each to operate independently. A maxi- To match the speed of data transfer to the cache, the
mum speedup of M is again possible and the modules can primary memory is interleaved so that a number of words
be accessed at random. However, a memory control unit are accessed concurrently. The memory is divided into a
is required. number of modules, usually a power of 2, and the primary
addresses are numbered across the module as shown in
Figure 6. The number of modules is chosen to match the
3 CACHE MEMORY
4 memory modules
These are high-speed buffers between the main memory and
the processor memory. Their success depends on locality.
If the entire working set of a process can be placed 8 9 10 11
into the cache, a substantial increase in performance can 4 5 6 7
be achieved. 0 1 2 3
Purposes of Cache Memory are now discussed. There
is usually a mismatch in the speed between transfers in
primary and secondary memories, the latter being much
Data is read or written concurrently/sequentially by cache
slower. There is also a speed mismatch between primary
memory access and processor operation times because the Figure 6. Numbering of primary addresses across the mem-
processor can operate faster. Having a cache of high-speed ory modules.
1052 Elements: C – Data Acquisition and Processing Systems
cache to which each word is transferred sequentially. Thus,
Cache
for the example above, 8 memory blocks would be required Send VA Real address RA
CPU MMU
having an access time per word to cache of M = 50 ns. For
n = 10 the average access time, Ta , is now equal to 55 ns. Memory
An important parameter in cache operations is the prob- (a)
ability that the required word is in cache and is of the order
of 0.8. A hit occurs when a cache location contains the VA VBA RBA
CPU MMU Memory
required data, otherwise a miss results. The hit ratio, h, is
D
the number of times the required
word is found in cache Compare OK
h= D -displacement
the total number of references RBA
the number of hits Cache
Line of data
= (3)
the total number of references (b)
Example A cache operates with a hit ratio of 0.85. Using Real address
the memory parameters as before with M = 400 ns and
m = 50 ns, calculate the access time, Ta . Assume that the Search cache tag memory
cache is loaded before the processor is accessed.
Solution: No
Cache hit?
(15 × 400 + 15 × 50 + 85 × 50)
Ta = = l00 ns (4) Yes
100
Update Get line Send real address Select cache entry
line status from cache to main memory for replacement
4 OPERATION OF THE CACHE Receive line
A TAG is a memory address. Each entry in cache consists Select requested byte Store line in cache
of a line of data, which can be several memory words, as
represented in Figure 7. For an M-way interleaved memory, Output to CPU
a cache line can be M words or 2M words or 4M words. (c)
TAGS are matched in the address field. If a hit, then select
byte in line and update line status, which reflects reference, Figure 8. Operation of the cache: (a) serial, (b) parallel, and
for example, write. If a miss, send the real address to (c) typical flow of instructions.
main memory. During read from main memory, the cache
selects a line in cache to be replaced. For virtual memory, a is needed to hold RBA . These assume that the cache is in
memory management unit, or MMU, is needed for address the real address space, which is the most common.
translation for cache and memory access. The flow diagram for cache operation is given in
The translation can be performed serially as shown in Figure 8(c). When the real address is specified, the cache
Figure 8(a). The alternative is to do it in parallel, as in TAG memory is searched. If there is a hit, the line status
Figure 8(b). This is called D-displacement. An extra field is updated and the line is fetched from the cache. Subse-
quently, the operation, which is straightforward, selects the
TAG memory Data memory
requested bytes and sends them to the CPU. If there is a
Address cache miss, the real address is sent to main memory and
TAG 0 Line 0
the cache entry is selected for replacement. The main mem-
TAG 1 Line 1
ory receives the real address, the receive line is activated
Cache memory and the line is stored in cache, and the requested bytes
TAG N Line N are selected.
Miss Data In cache design, two hit ratios are of interest. The cold-
Miss Data
start hit ratio applies to an initially empty cache, while the
warm-start hit ratio applies when a process already has data
Figure 7. Cache memory TAG principles of cache operation. in the cache.
Data Acceleration in Computers 1053
Line size and cache size are the two most important TAG + Index
TAG Index Primary
factors regarding hit ratio. Long lines improve the spatial Cache
memory
locality but it takes time to move them. For microproces- Index
sors, typical sizes are 16 to 64 bytes. A large number of TAG Data
lines improves the temporal locality up to a point. When
greater than the working set, this will not result in a Compare
decrease in the miss ratio.
‘0’ ‘1’
Access Access
5 MEMORY FETCH POLICIES primary memory data
(a)
There are various strategies in determining what lines
should be fetched from main memory and loaded into the TAG Block b Word
cache. Demand fetching fetches the line when it is needed.
This is always implemented to some degree, but a measure
b
of prefetching can improve matters. Prefetching tries to
TAG Data Data Data Data
fetch lines, which are going to be needed in the near future.
In high-speed caches, it is the line immediately following
Compare
the fetched line i, that is, line i + l. Hence, fetch line 1
and prefetch line i + l. There are a number of policies Word select
that can be implemented. One is always prefetch, which
(b)
tries to prefetch a line whenever line i is referenced. This
increases memory traffic by about 80%. An alternative is
Memory address
to prefetch-on-a-miss, that is, when a cache miss occurs, Primary
line i is fetched and line i + l is prefetched. Another more Cache memory
associative memory
common method is tagged-prefetch, which prefetches on a
miss and on the first reference to a prefetched line. This Address Data
increases memory traffic up by 10 to 20% for the same
miss ratio as always prefetch.
‘0’ ‘1’
6 CACHE MAPPING SCHEMES – Access Access
PLACEMENT POLICIES primary data
(c)
Cache is smaller than main memory, so that a few lines
map into each cache line. It is necessary to decide where Figure 9. Mapping schemes for cache memory: (a) direct type
for noninterleaved, (b) interleaved, and (c) associative type.
to put data lines in the cache. Primary memory is hardware
mapped into the cache by three main ways. consisting of 1 MB with a 20-bit address, and a 16 kB cache
with 16 bytes per line. Thus, there are 64 klines in main
7 CACHE MAPPING SCHEMES – DIRECT memory and 1024 lines in the cache. In direct mapping,
MAPPING line i is placed in line (i modulo 1024) in the cache. The
advantage of this method is that associative comparison is
Direct mapping is shown in Figure 9(a). A TAG field in not needed and the replacement of lines is trivial. However,
the processor address is used to identify a page in primary the cache can only contain one of the 64, or 26, lines that
memory, while an INDEX field finds the word within the map into each cache line one at a time.
page. Thus, words with different tags (pages) but the same
index will share the same location in cache. To see if a
word is in cache, the tags are compared. If the tags are
8 CACHE MAPPING SCHEMES –
the same, the word is read, otherwise the word is accessed ASSOCIATIVE AND SET-ASSOCIATIVE
from primary memory. For interleaved memories, a block of MAPPING
words can be transferred to cache as shown in Figure 9(b).
Note that common blocks in different pages occupy the Associative mapping, shown in Figure 9(c), uses a content
same cache locations. Take as an example a main memory addressable memory, or CAM, to compare the incoming
1054 Elements: C – Data Acquisition and Processing Systems
address with all the cache addresses simultaneously. If the tags within each set are then associatively mapped. Note
address is present, the data is read, otherwise it is accessed that each set can contain any one of 256 lines. However,
to the cache from primary memory and overwrites data. since each set has 4 lines, then 4 out of these 256 can be
This scheme is expensive since words are stored in the stored at the same time. This is better than direct mapping.
CAM. An alternative is to store a block so that the block Since the associative search is only over L elements, rather
address is compared associatively reducing storage require- than M, it is much less costly than a full-associative cache.
ments. In both cases, during block or word replacement, the Note also that for S = l, the cache is fully associative, while
existing block or word must be chosen either on a random for L = l (one line per set) the cache is directly mapped.
first-in-first-out or least recently used basis.
Fully associative cache is used when any line can reside
at any location and is found by comparing the tag con- 9 CACHE MAPPING SCHEMES –
tained in the address field with the tags held in the cache. REPLACEMENT POLICIES AND
It has the highest hit ratio but may be slower due to the
MEMORY UPDATE
comparator hardware required for the matching. It is expen-
sive to implement, but this could change with developments
Replacement policies: When a cache miss occurs and
in VLSI.
the cache is full, the question of which line to replace arises.
Set-associative mapping, as illustrated in Figure 10, all-
This requires a cache replacement policy. The same tech-
ows a block of data with the same index but different
niques that were used for management of the main memory
tags to be stored. It combines the advantages of direct
can be applied. However, since usage of the cache operates
mapping with fully associative placement. To find the
at maximum speed, such techniques must be implemented
required word, the tags are compared associatively and the
in hardware. Also, for simplicity, fixed sized partitions are
required word is found. The cache is partitioned into a
used with each process occupying either the whole or a
number of sets, S, containing a number of lines, L. The
fixed area of cache. There are two approaches. One is based
sets are directly mapped while the lines in each set are
on the usage of line and the other ignoring the usage. LRU
associatively mapped. Thus, for example, given that the
takes account of when the line was last referenced and has
number of lines per set is chosen to be L = 4, then the
the highest hit ratio but is costly to implement. Random and
number of sets S = M/L = 256, where M is the number
FIFO policies are two nonusage based methods. Statistics
of lines in main memory. To find a line, the middle 8 bits
show that the influence of placement is greater than that
of the address are used to find the set of line i such that it
of replacement. For direct mapping, replacement is trivial.
is placed in set (i modulo S) or (i modulo 256). The 8-bit
For a fully associative cache, replacement is difficult.
Real address
8 8 4 Main memory update: If data in cache is replaced by
TAG Set Byte a write operation, information will be lost if the line has
been modified and not updated in the main memory. It
is necessary to ensure that any data written to cache is
TAG Line 0
Line 0 also written to main memory so that I/O operations such
Line 1
Set 0
Line 1 as DMA can operate on primary memory. There are two
Line 2 commonly used update policies.
Line 3
Line 4 Write-through: In this approach to memory update,
Line 5 every write operation from the processor always transmits
Set 1
Line 255
changes to the cache and main memory. Although this takes
Line 256
time, there are usually some reads to cache. Hence, writes
to primary memory can take place while the cache reads
3 – 254
Line 16 383
Sets
take place. This method is good for multiprocessor systems
Line 16 384 with multiple caches.
Line 1020
Set 255
Line 1021 Copy-back also called write-back : There are three ver-
Line 65 535
Line 1022 sions of this method. In the first two, a cache line is only
Line 1023 transmitted to main memory if it is to be replaced in the
cache. In always-copy-back, no account is taken of whether
Figure 10. Cache mapping using the set-associative method. the line has been modified. In flagged-copy-back, a line is
Data Acceleration in Computers 1055
copied only when it has been modified as indicated by a not specified whether the line is copied to cache. On copy-
‘dirty flag’. This policy means that the main memory can back the written line is usually copied to the cache on a
contain ‘stale’ data, which will be lost if there is a proces- write-miss.
sor or cache failure. The third method of writing to cache is
known as write-back . Here, updating occurs during block
replacement time. A tag bit indicates when a block has
been updated. If a block has to be transferred into an exist- RELATED ARTICLES
ing block location, the old block is checked to see if it
has been changed. If so, it is transferred to the primary Article 152, Fundamentals of the Stored Program Dig-
memory. ital Computer, Volume 3; Article 153, Single Address
Optimization of the cache is an important design con- Instruction Microcomputer, Volume 3; Article 154,
sideration. Factors to be taken into account are hit ratio, Internal Operation of the Microprocessor, Volume 3;
access time on a hit, and delay time on a miss. A large Article 155, External Operation of the Microproces-
cache gives a low miss ratio but a high cost. Note that if sor, Volume 3; Article 156, Memory Management in the
there is a write-miss, with the write-through policy, it is Microprocessor, Volume 3.
Dr Joe McGhee unfortunately passed away before his material was finalised. He will be remembered by the Measurement community.
158: Microcontroller Systems
Joseph McGhee
Formerly of University of Strathclyde, Glasgow, UK
Interfacing to these peripherals is achieved using a bus
1 Single Chip Microprocessors 1056 oriented connection consisting of a control bus, an address
2 Microprocessor-based Data Acquisition 1056 bus, and a data bus as shown in Figure 2. The external
3 Introduction to Special Purpose Digital peripheral ICs (integrated circuits) handle input–output,
Processors 1058 I/O, operations, communications using Asynchronous
Communications Interface Adapter, or ACIA, and external
4 Bit-slice Devices and Elements 1059
memory of both the read-only memory, or ROM, and
5 The First Bit-slice Processor AMD 2901 1060
random-access memory, or RAM, types.
Related Articles 1061 The microcontroller unit, or MCU, which combines all
Further Reading 1061 these functions into one IC, thus simplifies the system
design. This reduces costs and board space. A typical block
diagram for an MCU is shown in Figure 3. The MCU exam-
1 SINGLE CHIP MICROPROCESSORS ple given in the figure, which is the Motorola 68HC11A8,
is one version of the very popular 68HC family of micro-
A typical single chip microprocessor has the architecture controllers. This controller, which is manufactured using
given in Figure 1. The instruction set is based on the the high density complementary metal oxide semiconduc-
standard 6800 instructions with additional functions. The tor, or HCMOS, IC technology, is an advanced 8-bit MCU
central processing unit CPU has seven registers available with sophisticated on chip peripheral capabilities. Although
to the programmer as shown in Figure 1. The two 8-bit the nominal bus speed is 2 MHz, it may operate at ultralow
accumulators, A and B, can be used by some instructions frequencies for reduced power applications. Note that the
as a single 16-bit accumulator called the D register, which central subsystem is still a microprocessor.
allows a set of 16-bit operations even though the CPU is The MCU I/O capability is grouped into ports. A port
technically an 8-bit processor. may be configured as a general I/O, similar to the VIA, or
Two index registers are available that may be used along may have a specific function, such as serial communica-
with register D to perform fast 16-bit arithmetic operations. tions, analog to digital converter, or ADC, input. Ports B
Two 16-bit by 16-bit divide instructions are also included. and C may operate in single chip mode, where the MCU
application requires no additional external memory, or in
2 MICROPROCESSOR-BASED DATA expanded mode, where external memory may be added.
ACQUISITION The latter requires the definition of an address bus, by com-
bining ports B and C, and a data bus. Note that the data
Microprocessors and microcontrollers form the heart bus is multiplexed with low byte address lines to reduce
of all modern acquisition systems. The microprocessor, the number of IC pins. The versatility of this configura-
which generally performs control, data handling, and tion allows the device to interface, or be configured, to suit
data processing functions, requires external peripherals. many applications.
Handbook of Measuring System Design, edited by Peter H. Sydenham and Richard Thorn.
2005 John Wiley & Sons, Ltd. ISBN: 0-470-02143-8.
Microcontroller Systems 1057
7 Accumulator A 0 7 Accumulator B 0 AS
15 Double accumulator D 0 D
15 Index register X 0 IX
15 Index register Y 0 IY
15 Stack pointer 0 SP
15 Program counter 0 PC
Condition code register S X H I N Z V C CCR
Carry
Overflow
Zero
Negative
I Interrupt mask
Half carry (from bit 3)
X Interrupt mask
Stop disable
Figure 1. The programming registers in a microprocessor or microcontroller.
Microprocessor
ACIA Input− Output (I/O) RAM ROM/PROM/EPROM
Figure 2. A microprocessor controlled data acquisition system.
The MCU also includes a 16-bit free running timer with with 1 start bit, 8 or 9 data bits, and a stop bit, as previ-
three input capture lines, five output-compare lines, and a ously illustrated in Article 155, External Operation of the
real time interrupt function. An 8-bit pulse accumulator sub- Microprocessor, Volume 3. An on chip baud rate genera-
system can also count external events or measure external tor derives standard baud rate frequencies from the MCU
periods. oscillator. The SCI system can be used to connect a CRT
The version shown here also has an onboard 8-channel or personal computer to the MCU. Alternatively, several
analog multiplexor, or MUX, feeding an 8-bit ADC. This widely distributed MCUs can use their SCI subsystems to
allows up to eight individual analog sensors to be moni- form a serial communication network. This clearly allows
tored by one MCU. Conversion is initiated under software remote data processing/information extraction as well as
control. local data capture.
The device also includes an asynchronous serial com- Self-monitoring circuitry is included on chip to pro-
munications interface, or SCI, and a separate synchronous tect against system errors. A computer operating properly
serial peripheral interface, or SPI. Adding suitable line watchdog system, or COP, protects against software fail-
drivers allows a communication capability directly from ures. A clock monitor system generates a system reset in
the MCU for RS 232/RS 422, which are described in Arti- case the clock is lost or runs too slow. An illegal op-code
cle 168, RS 232 and EIA/TIA 232 Serial Interface, Vol- detection circuit provides a nonmaskable interrupt if an
ume 3. The SCI generates standard communication formats illegal op-code is detected.
1058 Elements: C – Data Acquisition and Processing Systems
Pulse accumulator PAI PA7
OC2 PA6
ROM−8k Bytes OC3 PA5
Port A
Timer OC4 PA4
OC5 PA3
IC1 PA2
Periodic interrupt
IC2 PA1
RAM−256 Bytes COP watchdog IC3 PA0
SS PD5
SCK PD4
Data direction D
SPI
EEPROM−512 Bytes MOSI PD3
Port D
MISO PD2
TxD PD1
PE7 SCI
RxD PD0
PE6
PE5
PE4
PE3 A/D
PE2 converter
PE1 M68HC11 CPU
PE0
VREFH
VREFL
Address/Data bus
Reset
Interrputs
XIRQ
IRQ
(VPPBULK) Handshake I/O
XTAL
EXTAL Oscillator
Data direction C
E Port C
Port B
Power
MODA
(LIR) Mode
select
STRB
STRA
MODB
PC7
PC6
PC5
PC4
PC3
PC2
PC1
PC0
PB7
PB6
PB5
PB4
PB3
PB2
PB1
PB0
(VSTBY)
VDD
VSS
AD7
AD6
AD5
AD4
AD3
AD2
AD1
A15
A14
A13
A12
A11
A10
RW
AS
A9
A8
Figure 3. The Motorola 68HC11A8 microcontroller.
Two software controlled power saving modes of oper- 1. digital systems are less prone to error;
ation, WAIT and STOP, are available to conserve power. 2. numerical accuracy depends on the number of vari-
These modes make the 68HC11 family especially attractive ables, or bits, used;
for automotive and battery driven applications. 3. information can be both numerical and nonnumerical;
4. a structured approach can be used by breaking the main
3 INTRODUCTION TO SPECIAL task into a series of subtasks;
PURPOSE DIGITAL PROCESSORS 5. a general purpose system (computer) can be speci-
fied, which can be programmed for a particular task;
Digital systems of increasing complexity are playing a vital and
role in the processing of information. The reasons for this 6. VLSI technology has made the realization of complex
are as follows: systems cost-effective.
Microcontroller Systems 1059
From an engineering point of view, the description and lengths, and higher reliability due to minimization of the
design of digital systems require the development of suit- number of interconnections. From an economic point of
able models. For simple digital sequential networks, a view, large scale use of a device is desirable (cf appli-
systems model based on the state description of the system cation specific integrated circuits or ASICs). This has led
can be used with its concepts of output function and state to microprocessors/computers on a chip and special pur-
transition function. For other systems, performing com- pose bit-slice elements, which are the building blocks of
plex computations, an algorithmic model, which breaks computer systems.
down a complex digital computation into a sequence of Bit-slice elements are fixed width computer hardware
primitive computations, becomes attractive. The algorithms elements that range over 4, 8, 16, and 32 bits. A bit
themselves can be fixed by the hardware, or if storage is slice forms the basis of large computer systems, since they
available they can be changed by programming. Such algo- can be paralleled up. For speed, they are implemented
rithmic models can be used for the specification, simulation, in Bipolar Technology, which includes ECL, LSTTL, and
and implementation of complex digital systems. ASTTL. Examples are 74LS481, MC10800, (ECL), and
The algorithmic model can be used at several levels AMD 2901 (LSTTL). They can also be used for other
of the system depending on the nature of the primitive applications such as signal processing. To be realized as
computations. These range from (a) the hardware level, a digital system, they require control logic. The use of bit
where logical operations are implemented by hardwiring slices allows partitioning of the system into a set of LSI
networks of switching components; (b) the firmware level, building blocks of fixed words of width 4, 8, and so on to
where primitive computations involving hardware-like log- implement words in vertical slices of 8, 16, 32, and so on.
ical operations are implemented as microprograms; and This approach provides a reduction in the pin-out of the
(c) the software level, where language statements are exe- chip, a realization in LSI or VLSI, and a universal building
cuted in the computer. block.
A typical computer system consisting of memory, con- Bit-slice elements give the advantages of flexibility and
trol, and processing elements has already been described speed when designing large systems with high performance.
in Article 152, Fundamentals of the Stored Program The trend has been to introduce higher levels of integration
Digital Computer, Volume 3. The data, control, process- into bipolar families so that word length increases to 8,
ing, and I/O structures of this system are all important. 16, and 32 bits. In the latter case, the ASS832 from TI
Arithmetic processing elements in the form of bit-slice com- (Texas Instruments) uses Advanced Schottky technology to
ponents will be now introduced. obtain a 75-ns instruction execution. The structure of bit-
slice elements may be organized to allow operation in either
serial or parallel modes as in Figure 4. The serial modes are
4 BIT-SLICE DEVICES AND ELEMENTS very similar to the conventional Von Neumann architecture.
In the parallel units, many of the acceleration mechanisms
Large scale integration, or LSI, of digital logic functions already described in Article 157, Data Acceleration in
has led to higher switching speeds, due to shorter lead Computers, Volume 3 are used.
D in
Registers
MUX
D in
ALU Carry in
Carry out Ci Co Ci Co Ci Co
1 2 3 4
Accumulator
D out
Shift (b)
D out
(a)
Figure 4. Architecture for typical bit-slice processors: (a) serial and (b) parallel.
1060 Elements: C – Data Acquisition and Processing Systems
A comparison with microprocessors is appropriate. In the with 500 gates. It has a 4-bit wide slice and an 18-word
first place, microprocessors are not bit-slice devices. They RAM. Any two of its registers are selectable on A and B
are LSI stored program computers, whose hardware con- data busses by a 4-bit control input. Note that the A register
sists of registers, ALU (arithmetic logic unit) and control is read only, while the B register is read and write. Its ALU,
unit, with fixed instruction sets. Since microprocessors are which has three arithmetic and five logical instructions,
made in MOS and CMOS technology, they are generally R and S, has inputs for connection to R-A reg., D bus
slower than bit-slice elements, which are fabricated in bipo- (external, 0 S-A reg, B reg, Q reg, 0).
lar technology. The Q reg, which is designed for multiplication on
the basis of shift and add, can be fed from the ALU
or from itself with a shift up or down. It can shift the
5 THE FIRST BIT-SLICE PROCESSOR ALU output before loading into B reg. The ALU output
AMD 2901 can go to the external Y Bus, Q reg., or the Register
Array with a shift, and can also output A reg contents to
Although this was one of the first slices, it is an important theY bus.
one since its general structure, which is shown in Figure 5, There are nine control lines for controlling (a) ALU
established the architectural structure of future devices. Function (b) ALU Source Data (c) ALU Destination. Other
It was used in the DEC system 2020 and Data General control lines allow Y bus data enable ALU carry-in, ALU
Nova 4. It is a 40-pin LSI chip implemented in LSTTL carry-out, and carry look ahead signals.
8 7 6 5 4 3 2 1 0
Destination ALU ALU
control functions source
Macroinstruction decode
RAM0 RAM Shift RAM3
O0 O3
Clock (CP) O Shift
A B Data in
(Read) A Address CP
Address F 0
RAM
(16 Addressable Registers) O Register
B
B Address CP 0
You might also like
- Principal of Industrial Instrumentation Compress100% (1)Principal of Industrial Instrumentation Compress573 pages
- Electronics Engineering: Electronic Measurements and InstrumentationNo ratings yetElectronics Engineering: Electronic Measurements and Instrumentation16 pages
- Jack H. Dynamic System Modeling and Control MODEL2 4100% (1)Jack H. Dynamic System Modeling and Control MODEL2 4943 pages
- Measurement System Characteristics and Statistical Analysis Unit INo ratings yetMeasurement System Characteristics and Statistical Analysis Unit I11 pages
- Fundamentals of Instrumentation and Measurement94% (36)Fundamentals of Instrumentation and Measurement555 pages
- Module 1 Introduction To Measurement SystemsNo ratings yetModule 1 Introduction To Measurement Systems15 pages
- MA3354 Discrete Mathematics Lecture Notes 2No ratings yetMA3354 Discrete Mathematics Lecture Notes 2221 pages
- Electronics Measurements and Instrumentation100% (1)Electronics Measurements and Instrumentation74 pages
- 01 Introduction - To - Instruments - and - Their - Representation - 2No ratings yet01 Introduction - To - Instruments - and - Their - Representation - 278 pages
- Abu-Mahfouz I. Instrumentation. Theory and Practice Part 1... Measurements 2022100% (1)Abu-Mahfouz I. Instrumentation. Theory and Practice Part 1... Measurements 2022147 pages
- Ee6404 Measurements and InstrumentationNo ratings yetEe6404 Measurements and Instrumentation69 pages
- 04 - Measurement & Instrumentation - EC - Postal Theory - 2025No ratings yet04 - Measurement & Instrumentation - EC - Postal Theory - 202516 pages
- UNIT 1 - Introduction To Measurement SystemNo ratings yetUNIT 1 - Introduction To Measurement System55 pages
- Imaster NCE V100R020C10 REST NBI User Guide 10100% (1)Imaster NCE V100R020C10 REST NBI User Guide 10110 pages
- Measurement and Instrumentation - Chapter - 1No ratings yetMeasurement and Instrumentation - Chapter - 118 pages
- Birla Institute of Technology and Science, Pilani Pilani Campus AUGS/ AGSR DivisionNo ratings yetBirla Institute of Technology and Science, Pilani Pilani Campus AUGS/ AGSR Division4 pages
- Designing Controls For The Process Industries (2023)100% (5)Designing Controls For The Process Industries (2023)600 pages
- Flyer Theoryanddesignformechanicalmeasurements SixtheditionNo ratings yetFlyer Theoryanddesignformechanicalmeasurements Sixthedition10 pages
- Iare Ece Vi-Emi Emi-Lecture Notes-Word NewNo ratings yetIare Ece Vi-Emi Emi-Lecture Notes-Word New185 pages
- Course Outline of ME202 Measurements - GIKINo ratings yetCourse Outline of ME202 Measurements - GIKI4 pages
- (SV) Topic 1 Introduction To MeasurementNo ratings yet(SV) Topic 1 Introduction To Measurement18 pages
- Chapter NO.01 Introduction, Pressure - Temperature MeasurementNo ratings yetChapter NO.01 Introduction, Pressure - Temperature Measurement82 pages
- Lecture Notes: Mrs. R. Rani, Associate ProfessorNo ratings yetLecture Notes: Mrs. R. Rani, Associate Professor93 pages
- Ui / Ux Design Terms, Explanations & RolesNo ratings yetUi / Ux Design Terms, Explanations & Roles19 pages
- Measurement and Instrumentation: Theory and ApplicationNo ratings yetMeasurement and Instrumentation: Theory and Application12 pages
- Eee 2407 - Instrumentation Nov-Mar 2019-2No ratings yetEee 2407 - Instrumentation Nov-Mar 2019-287 pages
- ANSI ISA-5.1-2022 Instrumentation Symbols and Identification100% (9)ANSI ISA-5.1-2022 Instrumentation Symbols and Identification126 pages
- Syllabus For Instrumentation Engineering (IN) : Linear AlgebraNo ratings yetSyllabus For Instrumentation Engineering (IN) : Linear Algebra3 pages
- Syllabus For Instrumentation Engineering (IN) : Linear AlgebraNo ratings yetSyllabus For Instrumentation Engineering (IN) : Linear Algebra3 pages
- Instrumentation Devices Ans Systems - 1 by K.c.bhuyanNo ratings yetInstrumentation Devices Ans Systems - 1 by K.c.bhuyan70 pages
- Multilayer Integrated Film Bulk Acoustic Resonators - Yafei Zhang, Da ChenNo ratings yetMultilayer Integrated Film Bulk Acoustic Resonators - Yafei Zhang, Da Chen158 pages
- Seborg - Process Dynamics and Control 4th Ed 2017 Solutions PDF90% (10)Seborg - Process Dynamics and Control 4th Ed 2017 Solutions PDF572 pages
- Process Plant Design and Simulation Handbook100% (12)Process Plant Design and Simulation Handbook468 pages
- Cause and Effect Diagram For Process Plants100% (3)Cause and Effect Diagram For Process Plants26 pages
- Calibration and Electronic Measurements Unit - INo ratings yetCalibration and Electronic Measurements Unit - I9 pages
- Standards and Regulations With Regards To Safety Glass For Use in BuildingsNo ratings yetStandards and Regulations With Regards To Safety Glass For Use in Buildings5 pages
- IN-Instrumentation Engineering Engineering MathematicsNo ratings yetIN-Instrumentation Engineering Engineering Mathematics2 pages
- Preface 2001 Measurement and Instrumentation Principles Third EditionNo ratings yetPreface 2001 Measurement and Instrumentation Principles Third Edition3 pages
- Introduction To Process Engineering and Design @ProcessEng100% (7)Introduction To Process Engineering and Design @ProcessEng761 pages
- Multi Functional Anemometer: 7.battery ReplacementNo ratings yetMulti Functional Anemometer: 7.battery Replacement3 pages
- Multi Functional Anemometer: 7.battery ReplacementNo ratings yetMulti Functional Anemometer: 7.battery Replacement3 pages
- No Touch Exit Sensor k1-1 / k1-2: Installation Sensing RangeNo ratings yetNo Touch Exit Sensor k1-1 / k1-2: Installation Sensing Range1 page
- Introduction To Process Engineering and Design, Second Edition PDF92% (13)Introduction To Process Engineering and Design, Second Edition PDF1,138 pages
- Designing Controls For The Process Industries PDF60% (5)Designing Controls For The Process Industries PDF429 pages
- Corrosion of Mild Steel in The Presence of Elemental SulfurNo ratings yetCorrosion of Mild Steel in The Presence of Elemental Sulfur14 pages
- Fundamentals of Instrumentation and More by NJATC PDF100% (13)Fundamentals of Instrumentation and More by NJATC PDF195 pages
- Risk Based Inspections For Highway StructuresNo ratings yetRisk Based Inspections For Highway Structures33 pages
- AGC Safety Glass Technical Information 2008 enNo ratings yetAGC Safety Glass Technical Information 2008 en10 pages
- 20 MHZ To 500 MHZ If Gain Block Adl5531: Features Functional Block DiagramNo ratings yet20 MHZ To 500 MHZ If Gain Block Adl5531: Features Functional Block Diagram12 pages
- High Frequent Total Station Measurements For The Monitoring of Bridge VibrationsNo ratings yetHigh Frequent Total Station Measurements For The Monitoring of Bridge Vibrations8 pages
- Conformiq Creator 2 Self Study Material 1No ratings yetConformiq Creator 2 Self Study Material 1114 pages
- Ka5Q-Series: KA5Q0765RT/KA5Q12656RT/KA5Q1265RF/ KA5Q1565RF Fairchild Power Switch (FPS)No ratings yetKa5Q-Series: KA5Q0765RT/KA5Q12656RT/KA5Q1265RF/ KA5Q1565RF Fairchild Power Switch (FPS)16 pages
- How To Get The Cherry in Castle Excellent (The Unofficial Way) - Tech FairyNo ratings yetHow To Get The Cherry in Castle Excellent (The Unofficial Way) - Tech Fairy1 page
- Valves, Piping and Pipelines Handbook, Third Edition98% (53)Valves, Piping and Pipelines Handbook, Third Edition696 pages
- Control System Engineering Reference Manual96% (28)Control System Engineering Reference Manual144 pages
- Salesforce Org To Org Opportunity AggregationNo ratings yetSalesforce Org To Org Opportunity Aggregation10 pages
- Data Sheet 68570379 Ignition Module IM 3.2No ratings yetData Sheet 68570379 Ignition Module IM 3.23 pages
- U (K 2) ) - Matlab Code That Initiates The Viterbi Algorithm, Runs An Outer Loop For Dif-/nNo ratings yetU (K 2) ) - Matlab Code That Initiates The Viterbi Algorithm, Runs An Outer Loop For Dif-/n2 pages
- ANSI/ISA-5.1-2009 Instrumentation Symbols and Identification100% (1)ANSI/ISA-5.1-2009 Instrumentation Symbols and Identification128 pages
- Giving Bodies Back to Data: Image Makers, Bricolage, and Reinvention in Magnetic Resonance TechnologyFrom EverandGiving Bodies Back to Data: Image Makers, Bricolage, and Reinvention in Magnetic Resonance TechnologyNo ratings yet
