Signal Conditioning & PC-Based Data Acquisition Handbook

Signal Conditioning
&
PC-Based Data Acquisition
Handbook

By

Table of Contents

Chapter Title Page

1 Introduction 1

2 Analog-to-Digital Conversion 5

3 Multiplexing 19

4 Temperature Measurement 25

5 Strain & Acceleration 39

6 General Amplification 55

7 Noise Reduction and Isolation 71

8 Digital and Pulse Train Signal Conditioning 79

9 Product Selection Guide 91

Index 117
i
Preface Signal Conditioning Handbook
Preface
Any industry that performs testing or monitoring has an array of transducers specifically
designed for its particular measurement requirements. The intent of this handbook is to
introduce the reader to the most commonly used transducer interfaces and to provide prac-
tical information for dealing with the most frequently encountered transducers and their
associated signals.
Transducers are available for measuring many physical quantities, such as temperature,
pressure, strain, vibration, sound, humidity, flow, level, velocity, charge, pH, and chemical
composition, among others. In most cases, the transducer manufacturer provides applica-
tion notes on the transducers use and principles of operation. The main questions to
consider when selecting a transducer are:
What are the electrical characteristics (amplitude, frequency, source impedance) of
the transducers output?
What kind of power supply/excitation is required?
What is the transducers specified accuracy?
Over what range of amplitude and frequency are the measurements accurate?
How is the transducer calibrated?
How can transducer accuracy and calibration be verified?
In what environment (temperature, humidity, vibration, pressure) is the trans-
ducer able to operate?
It is generally unwise to view a transducer as a black box that provides a specified output
for a certain input. Knowing how a transducer works is imperative to making reliable
measurements.
1
Chapter 1 Introduction
Chapter 1
Introduction
2
This handbook is intended for newcomers to the field of data acquisition and signal
conditioning. Emphasis is given to general discussions of ADC measurement and the
signal conditioning requirements of selected transducer types. For more detailed de-
scriptions of the signal conditioning schemes discussed, contact IOtech for Applica-
tions Notes and Product Specification Sheets. Additional information on IOtech prod-
ucts is available in Chapter 9, a product selection guide.
Most measurements begin with a transducer, a device that converts a measurable physi-
cal quality, such as temperature, strain, or acceleration, to an electrical signal. Trans-
ducers are available for a wide range of measurements, and come in a variety of shapes,
sizes, and specifications. This book is intended to serve as a primer for making measure-
ments by interfacing transducers to a computer using signal conditioning.
Si gnal condi ti oni ng converts a
transducers signal so that an analog-
to-digital converter (ADC) can mea-
sure the signal. Signal conditioning
can include amplification, filtering,
differential applications, isolation, si-
multaneous sample and hold (SS&H),
current-to-voltage conversion, volt-
age-to-frequency conversion, linear-
ization and more. Signal condition-
ing also includes excitation or bias for
transducers that require it.
Figure 1.01 depicts a generic data ac-
quisition signal-conditioning configu-
ration. The transducer is connected to the input of the signal conditioning electronics.
The output of the signal conditioning is connected to an ADC input. The ADC con-
verts the analog voltage to a digital signal, which is transferred to the computer for
processing, graphing, and storage.
Analog-to-Digital Conversion. Chapter 2 includes a discussion of the four basic ADC
types, as well as issues such as accuracy, noise reduction, and discrete sampling consid-
erations. Topics such as input and source impedance, differential voltage measure-
ments, simultaneous sample and hold, selectable input ranges, multiplexing, and isola-
tion are also discussed. A section on discrete sampling considerations, which covers
aliasing, windowing, fast Fourier transforms (FFTs), standard Fourier transforms, and
digital filtering, is also included.
Fig. 1.01: Generic signal-conditioning scheme
Signal
Conditioning
Computer
Transducer
ADC
3
Multiplexing. Chapter 3 includes a discussion on multiplexing, current measurements,
and the associated issues such as simultaneous sample and hold, input buffering, and
methods of range and gain selection.
Signal Conditioning by Chapter. Signal conditioning for a wide variety of transducer types,
including temperature, strain, force, torque, pressure, and acceleration, is discussed in Chapter
4 through Chapter 8. The temperature measurement section in Chapter 4 describes the
principles of operation, signal conditioning, linearization, and accuracy of thermocouples,
RTDs, and integrated circuits (ICs).
The strain gage section in Chapter 5 discusses the Wheatstone bridge, as well as the use
of strain gages in quarter-bridge, half-bridge, and full-bridge configurations. The use of
strain gages in load cells is described, along with excitation and signal conditioning
requirements. The piezoelectric transducers (PZTs) section in Chapter 5 describes the
use of these devices in voltage and charge amplification configurations and with low-
and high-impedance transducers. The pressure transducer section covers both
strain-diaphragm transducers used for quasi-static pressure measurements and PZT-based
pressure transducers used in dynamic measurements.
Chapter 6 begins with a brief review of general amplification. This section then
describes data acquisition front ends, source impedance and multiplexing, filters,
and single-ended and differential measurements. The measurement of high volt-
ages and DC and AC currents is also discussed.
Chapter 7 discusses noise reduction and isolation, including specific methods of
isolation, such as magnetic, optical, and capacitive.
This handbook also describes digital signal conditioning in Chapter 8, including speed
and timing issues. Also, covered in this chapter are frequency measurements, pulse
counting, and pulse timing. The frequency of a signal can be measured using two
methods: conversion to a voltage that is read by an ADC, or gated pulse counting. Both
methods are described, as is the use of counters for timing applications.
Chapter 9, IOtechs Product Selection Guide, features PC-based data acquisition sys-
tems, signal conditioning options, and temperature measurement instruments.
4
5
Chapter 2 Analog-to-Digital Conversion
Chapter 2
Analog-to-Digital Conversion....
6
Fig. 2.01: 2-bit parallel converter
This chapter examines general considerations for analog-to-digital converter (ADC) mea-
surements. Discussed are the four basic ADC types, providing a general description of
each while comparing their speed and resolution. Issues such as calibration, linearity,
missing codes, and noise are discussed, as are their effects on ADC accuracy.
This chapter also includes information on simultaneous sample and hold (SS&H) and
selectable input ranges. Finally, this chapter contains a section on discrete sampling,
which includes Fourier theory, aliasing, windowing, fast Fourier transforms (FFTs), stan-
dard Fourier transforms, and digital filtering.
ADC Types
An ADC converts an analog voltage to a digital number. The digital number represents the
input voltage in discrete steps with finite resolution. ADC resolution is determined by the
number of bits that represent the digital number. An n-bit ADC has a resolution of 1 part in
2
n
. For example, a 12-bit ADC has a resolution of 1 part in 4096 (2
12
=4,096). Twelve-bit
ADC resolution corresponds to 2.44 mV for a 10V range. Similarly, a 16-bit ADCs resolu-
tion is 1 part in 65,536 (2
16
=65,536), which corresponds to 0.153 mV for a 10V range.
Many different types of analog-to-digital converters are available. Differing ADC types
offer varying resolution, accuracy, and speed specifications. The most popular ADC
types are the parallel (flash) converter, the successive approximation ADC, the voltage-
to-frequency ADC, and the integrating ADC. Descriptions of each follow.
Parallel (Flash) Converter
The parallel converter is the simplest
ADC implementation. It uses a refer-
ence voltage at the full scale of the in-
put range and a voltage divider com-
posed of 2
n
+1 resistors in series, where
n is the ADC resolution in bits. The
value of the input voltage is deter-
mined by using a comparator at each
of the 2
n
reference voltages created in
the voltage divider. Figure 2.01 depicts
a 2-bit parallel converter.
Flash converters are very fast (up to
500 MHz) because the bits are deter-
mined in parallel. This method requires a large number of comparators, thereby
limiting the resolution of most parallel converters to 8 bits (256 comparators). Flash
converters are commonly found in transient digitizers and digital oscilloscopes.
Comparators
R/2
R
R/2
R
R
+
Binary
output
Encoder
V
ref
V
input
7
Fig. 2.03: Voltage-to-frequency ADC
Successive Approximation ADC
A successive approximation ADC em-
ploys a digital-to-analog converter
(DAC) and a single comparator. It ef-
fectively makes a bisection or binomial
search by beginning with an output of
zero. It provisionally sets each bit of
the DAC, beginning with the most sig-
nificant bit. The search compares the
output of the DAC to the voltage being
measured. If setting a bit to one causes
the DAC output to rise above the input
voltage, that bit is set to zero. A dia-
gram of a successive approximation
ADC is shown in Figure 2.02.
Successive approximation is slower than flash conversion because the comparisons must
be performed in a series, and the ADC must pause at each step to set the DAC and wait
for it to settle. However, conversion rates over 200 kHz are common. Successive ap-
proximation is relatively inexpensive to implement for 12- and 16-bit resolution. Con-
sequently, they are the most commonly used ADCs, and can be found in many PC-based
data acquisition products.
Voltage-to-Frequency ADC
Figure 2.03 depicts the voltage-to-fre-
quency technique. Voltage-to-fre-
quency ADCs convert an input voltage
to an output pulse train with a fre-
quency proportional to the input volt-
age. Output frequency is determined
by counting pulses over a fixed time in-
terval, and the voltage is inferred from
the known relationship.
Voltage-to-frequency conversion has a
high degree of noise rejection, because
the input signal is effectively integrated
over the counting interval. Voltage-to-
frequency conversion is commonly used
to convert slow and often noisy signals.
Digital
pulse train
V
in
Voltage-to-frequency
converter
Timing
circuitry
Digital
outputs
Pulse
counter
Fig. 2.02: Successive approximation ADC
Digital
outputs
Comparator
DAC
Control
logic &
registers
+
V
input
8
Fig. 2.04: Integration and discharge of an integrating ADC
It is also useful for remote sensing applications in noisy environments. The input volt-
age is converted to a frequency at the remote location, and the digital pulse train is
transmitted over a pair of wires to the counter. This eliminates the noise that can be
introduced in the transmission of an analog signal over a long distance.
Integrating ADC
A number of ADCs use integrating tech-
niques, which measure the time to
charge or discharge a capacitor to de-
termine input voltage. Figure 2.04
shows Dual-slope integration, a com-
mon integration technique. Using a
current that is proportional to the in-
put voltage, a capacitor is charged for
a fixed time period. The average input
voltage is determined by measuring the
time required to discharge the capaci-
tor using a constant current.
Integrating the ADC input over an interval reduces the effect of noise pickup at the AC
line frequency if the integration time is matched to a multiple of the AC period. For this
reason, it is commonly used in precision digital multimeters and panel meters. Twenty-
bit accuracy is not uncommon. The disadvantage is a relatively slow conversion rate (60
Hz maximum, slower for ADCs that integrate over multiple line cycles).
Summary of ADC Types
Figure 2.05 summarizes the previously discussed ADC types and their resolution
and speed ranges.
Integration
time
Discharge
time
Fixed discharge
current
=
IV
input
V
c
a
p
a
c
i
t
o
r
V
ref
V
input
T
d
T
i
T
d
T
i
ADC Type Table
ADC Type Typical Typical
Resolution Speed
Parallel Converter 4-8 bit 100 kHz-500 MHz
Successive Approximation 8-16 bit 10 kHz-1 MHz
Voltage-to-Frequency 8-12 bit 1-60 Hz*
Integrating 12-24 bit 1-60 Hz*
* With line cycle rejection
Fig. 2.05: Summary of ADC types
9
Accuracy
Accuracy is an important consideration when selecting an ADC for use in test and measure-
ment applications. The following section provides an in-depth discussion of accuracy con-
siderations, pertaining to resolution, calibration, linearity, missing codes, and noise.
Accuracy vs. Resolution
The accuracy of a measurement is influenced by a variety of factors. If each independent
error is
i
, the total error is
total
=

i
2
i
This calculation includes errors resulting from the transducer, noise pickup, ADC quan-
tization, gain, offset, and other factors.
ADC resolution error is termed quantization error. In an ideal ADC, any voltage in the
range that corresponds to a unique digital code is represented by that code. The error in
this case is half of the least significant bit (LSB) at most. For a 12-bit ADC with a 10V
range, this error is 2.44 mV (0.0244%). There are three common methods of specifying
a contribution to ADC error: the error in least significant bits (LSBs), the voltage error
for a specified range, and the percent-of-reading error. It is important to recognize that
most ADCs are not as accurate as their specified resolution, because quantization error
is only one potential source of error. Nonetheless, the accuracy of a good ADC should
approach its specified resolution.
For more information concerning accuracy, refer to the Calibration section, which fol-
lows. For an in-depth discussion of errors arising in particular transducers, refer to
Chapters 3 through 8.
Figure 2.06 illustrates common error types encountered when using a 3-bit ADC. If the
manufacturer provides calibration procedures, offset and gain errors can usually be re-
duced to negligible levels, as discussed below. However, errors in linearity and missing
codes will contribute to the overall error.
Calibration
There are several common methods for calibrating an ADC. In hardware calibration,
the offset and gain of the instrumentation amplifier that serves as the ADC front end is
adjusted with trim pots. (The gain of the ADC can also be adjusted by changing the
reference voltage.) In hardware/software calibration, digital-to-analog converters that
null the offset and set the full scale voltages are programmed via software. In software
calibration, there is no hardware adjustment. Calibration correction factors are stored
in the nonvolatile memory of the data acquisition system or in the computer and used
to convert the reading from the ADC.
10
Even if an ADC is calibrated at the factory, it will need to be calibrated again after a period
of time (typically six months to a year, but possibly more frequently for ADCs of greater
resolution than 16 bits). Variations in the operating temperature can also affect instrument
calibration. Calibration procedures vary but usually require either a known reference source
or a meter of greater accuracy than the device being calibrated. Typically, offset is set via a
0V input, and gain is set via a full-scale input.
In many measurements, the voltage is not the physical quantity under test. Consequently,
it may be preferable to calibrate the complete measurement system rather than its indi-
vidual parts. For example, consider a load cell for which the manufacturer specifies the
output for a given load and excitation voltage. One could calibrate the ADC and combine
this with the manufacturers specification and a measurement of the excitation voltage;
however, this technique is open to error. Specifically, three distinct error sources are pos-
sible in this technique: error in the ADC calibration, error in the manufacturers specifica-
tions, and error in the measurement of the excitation voltage. To circumvent these error
sources, one can calibrate the measurement system using known loads and obtain a direct
relationship between load and ADC output.
4
Ideal
Input volts
A
D
C

O
u
t
p
u
t
0 6 10 2 4 8
6
4
2
0
7
1
5
3
Gain error
Input volts
A
D
C

O
u
t
p
u
t
0 6 10 2 4 8
6
4
2
0
7
1
5
3
4
Linearity error
Input volts
A
D
C

O
u
t
p
u
t
0 6 10 2 4 8
6
4
2
0
7
1
5
3
Offset error
Input volts
A
D
C

O
u
t
p
u
t
0 6 10 2 4 8
6
4
2
0
7
1
5
3
Missing code
Input volts
A
D
C

O
u
t
p
u
t
0 6 10 2 4 8
6
4
2
0
7
1
5
3
Fig. 2.06: Common ADC error sources. The straight line is ideal output from an ADC with infinite-bit
resolution. The step function shows the indicated error for a 3-bit ADC.
11
Linearity
If input voltage and ADC output deviate from the diagonal lines in Figure 2.06 more than
the ideal step function, the result is ADC error that is nearly impossible to eliminate by
calibration. This type of ADC error is referred to as nonlinearity error. If nonlinearity is
present in a calibrated ADC, the error is often largest near the middle of the input range, as
shown in Figure 2.06. The nonlinearity in a good ADC should be 1 LSB or less.
Missing Codes
Some ADCs have missing codes. In Figure 2.06, the ADC does not provide an output of
four for any input voltage. This error can result in a significant loss in resolution and
accuracy. A quality ADC should have no missing codes.
Noise
Many users are surprised by noise encountered when measuring millivolt signals or
attempting accurate measurements on larger signals. Investing in an accurate ADC
is only the first step in accurately measuring analog input signals. Controlling noise
is imperative.
Many ADCs reside on cards that plug into a PC expansion bus, where electrical noise can
present serious problems. Expansion bus noise often far exceeds the ADCs sensitivity
resulting in significant loss of measurement accuracy. Placing the ADC outside the PC is
often a better solution. An ADC in an external enclosure can communicate with the
computer over an IEEE 488 bus, serial port, or parallel port. If an application requires
placement of the ADC within the computer, the noise level should be tested by connect-
ing the ADC input to signal common and observing deviations in ADC output. (Con-
necting the ADC input to signal common isolates the cause of the noise to the circuit
card. More careful diagnostics are necessary when using an external voltage source be-
cause noise can arise from the external source and from the input leads.)
Noise Reduction and Measurement Accuracy
One technique for reducing noise and ensuring measurement accuracy is with isolation,
which also eliminates ground loops. Ground loops occur when two or more devices in
a system, such as a measurement instrument and a transducer, are connected to ground
at different physical locations. Slight differences in the actual potential of each ground
results in a current flow from one device to the other. This current, which often flows
through the low lead of a pair of measurement wires, generates a voltage drop which can
directly lead to measurement inaccuracies and noise. If at least one device is isolated,
such as the measurement device, then there is no path for the current flow, and thereby
no contribution to noise or inaccuracy.
12
Protection
Many data acquisition systems utilize solid-state multiplexing circuitry in order to
very quickly scan multiple input channels. These solid-state multiplexers are among
the most susceptible circuitry to overload voltages, which can commonly occur in a
system. Typically, multiplexers can only accept up to 20 or 30 volts before damage
occurs. Other solid-state devices in a measurement system include input amplifiers
and bias sources, both of which are also susceptible to damage from over voltage.
Isolation is one of a number of techniques used to protect sensitive solid-state cir-
cuitry in a data acquisition system.
Although isolation does not protect against excessive normal-mode input voltage (volt-
age across a pair of inputs), it does protect against excessive common-mode voltage. It
accomplishes this by eliminating the potentially large current that would otherwise flow
from the signal source to the data acquisition system, as a result of the common mode
voltage. By eliminating this current flow, the possibility of damage is eliminated.
High common-mode voltage measurements
It is often necessary to measure a small voltage which is residing on another, much larger
voltage. For example, if a thermocouple is mounted to one terminal of a battery, then
the measurement device must be capable of measuring the microvolt output of the ther-
mocouple while rejecting the battery voltage. If the common-mode voltage is less than
10-15V, a differential measurement via an instrumentation amplifier will read the ther-
mocouple voltage while ignoring the battery voltage. If the common-mode voltage is
higher than 10-15V, an isolation method is generally required.
There are several isolation methods
with the common characteristic of a
high common-mode voltage from in-
put to output. Each channel can have
an isolation amplifier, or, a group of
channels that are not isolated from
each other can be multiplexed and
digitized by an analog-to-digital con-
verter before the digital data is isolated
from the remainder of the system.
Actual isolation barriers can be opti-
cal, magnetic, or capacitive. The most
common are optical schemes in which Fig. 2.07: Optically-coupled isolation amplifier
Photo
diode
Photo
diode
LED
R
Input
ground
Output
ground
+
V
input
V
output
LED
13
infrared emission from an LED is detected by a photodiode on the opposite side of a
quartz barrier. Figure 2.07 illustrates an optically-coupled isolation amplifier. Optocou-
plers can be used to transmit pulse trains in which frequency or pulse width vary with
analog signal magnitude or which contain numerical data in serial pulse trains. It is
even possible to transmit analog information by varying LED current as in Figure 2.07.
Magnetic barriers using transformers and capacitive barriers are generally internal and
used in monolithic or hybrid isolation amplifiers.
Frequency Coupled Isolation. In frequency coupled isolation, a high frequency carrier
signal is inductively or capacitively coupled across the isolation barrier. The signal is
modulated on the input side and demodulated on the output side to reproduce the
original input signal.
Isolated ADC. When using an isolated ADC, the ADC and accompanying signal condi-
tioning are floated. The input signal is converted to a digital signal by the ADC and the
interface for transferring the digital code is digitally isolated. See Chapter 6 for a de-
tailed discussion.
Discrete Sampling Considerations
The Nyquist sampling theorem says that if a signal only contains frequencies less than
cutoff frequency f
c
, all the information in the signal can be captured by sampling at 2f
c
.
The upshot of this is that capturing a signal with maximum frequency component f
max
requires sampling at a rate of at least 2f
max
. In practice, for working in the frequency
domain, it is best to set the sampling rate between five and ten times the signals highest
frequency component. However, for viewing waveforms in the time domain, it is not
uncommon to sample 10 times the frequency of interest. One reason is to retain accu-
racy at the signals higher frequency components.
Aliasing
Aliasing can also be seen in the time domain. Figure 2.09 shows a 1-kHz sine wave
sampled at 800 Hz. The apparent frequency of the sine wave is much too low. Figure
2.08 shows the result of sampling the same 1-kHz sine wave at 5 kHz. The sampled wave
appears to have the correct frequency.
Aliasing is the main reason to sample at a rate higher than the Nyquist frequency.
Aliasingthe generation of false, low-frequency signalsoccurs when an ADCs sam-
pling rate is too low. Input signals are seldom bandwidth limited with zero amplitude
higher than f
max
. A signal with frequency components higher than one-half the sam-
pling frequency will cause the amplitude to appear below one-half the sampling fre-
quency in the Fourier transform. This is called aliasing, and it can cause inaccuracies in
sampled signal and also in the Fourier transform.
14
Aliasing is illustrated in Figure 2.10, which shows a square waves Fourier transform.
For the purpose of illustration, assume that the experiment has been designed to pro-
vide only frequencies of under 2 kHz. Ideally, a Fourier transform of a 500-Hz square
wave contains one peak at 500 Hz, and another at 1500 Hz, which is one third the
height of the first peak. In Figure 2.10, however, higher frequency peaks are aliased
into the Fourier transforms low-frequency range.
The use of a low-pass filter at 2 kHz, as
shown in Figure 2.11, removes most
of the aliased peaks. Low-pass filters
used for this purpose are often called
anti-aliasing filters.
When the sampling rate is increased to
four times the highest frequency of in-
terest, the Fourier transform in the range
of interest looks even better. Although
a small peak remains at 1,000 Hz, it is
probably the result of an imperfect
square wave rather than an effect of alias-
ing. See Figure 2.12.
Fig. 2.08: When sampled at just over 2 times the
frequency of the sine wave, the frequency content of
the signal is retained
Fig. 2.09: When sampling too slowly, the acquired
waveform erroneously represents the real sine wave
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 100 200 300 400 500 600
Actual Sine Wave Sampled at about 2 times fundamental
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 100 200 300 400 500 600
Actual Sine Wave Sampled at about 1/2 times fundamental
Frequency [Hz]
V
o
l
t
s
0 500 1000 1500 2000
1
0.4
0.2
0
0.1
0.3
0.5
0.6
0.7
0.8
0.9
Fig. 2.10: Fourier transform of a 500-Hz square wave
sampled at 4 kHz with no filtering
15
Windowing
Windowing is the multiplication of the input signal with a weighting function to
reduce spurious oscillations in a Fourier transform. Real measurements are per-
formed over finite time intervals. In contrast, Fourier transforms are defined over
infinite time intervals. As such, the Fourier transform of sampled data is an ap-
proximation. Consequently, the resolution of the Fourier transform is limited to
roughly 1/T, where T is the finite time interval over which the measurement was
made. Fourier transform resolution can only be improved by sampling for a longer
interval.
Using a finite time interval also causes spurious oscillations in the Fourier trans-
form. From a mathematical viewpoint, spurious oscillations are caused by the sig-
nal being instantaneously turned on at the beginning of the measurement and then
suddenly turned off at the end of the measurement. Figure 2.13 illustrates an ex-
ample of spurious oscillations.
Implementing window functions can help minimize spurious oscillations of a signal.
Multiplying the sampled data by a window function that rises gradually from zero de-
creases the spurious oscillations at the expense of a slight loss in triggering resolution.
There are many possible window functions, all of which involve trade-offs between
amplitude estimation and frequency resolution.
Frequency [Hz]
V
o
l
t
s
0 500 1000 1500 2000
1.4
0.2
0
0.4
0.6
0.8
1
1.2
Fig. 2.11: Fourier transform of 500-Hz square
wave sampled at a 4 kHz with low-pass filter cutoff
at 2 kHz
Frequency [Hz]
V
o
l
t
s
0 500 1000 1500 2000
1.6
0.4
0
0.6
0.8
1
1.2
1.4
0.2
Fig. 2.12: Fourier transform of 500-Hz square
wave sampled at 8 kHz with low-pass filter cutoff
at 2 kHz
16
Fast Fourier Transforms
The fast Fourier transform (FFT) is so common today that FFT has become an impre-
cise synonym for Fourier transforms in general. The FFT is a digital algorithm for com-
puting Fourier transforms of data discretely sampled at a constant interval. The FFTs
simplest implementation requires 2
n
samples. Other implementations accept other spe-
cial numbers of samples. If the data set to be transformed has a different number of
samples than required by the FFT algorithm, the data is often padded with zeros to
achieve the required number. This leads to inaccuracies, but they are often tolerable.
Fig. 2.16: Three common window types
Hamming Hanning Blackman
0.5
1
1.5
2
2.5
0
Fig. 2.15: When multiplied by a window function,
the frequency information is preserved while
minimizing the affect of irregularities at the
beginning and end of the sample segment
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Fig. 2.14: Without window function, abrupt
beginning and end points of waveform produce
erroneous frequency information
Fourier transform of a sine wave using a window function
Fourier transform of a sine wave with no window function
Frequency [Hz]
A
m
p
l
i
t
u
d
e
450 480 540 550 460 470 490 500 510 530 520
1
0.8
0.6
0.4
0.2
0
1.2
Fig. 2.13: A Fourier transform with window
function and without window function
17
Standard Fourier Transforms
A standard Fourier transform (SFT) can be used in applications where the number of
samples cannot be arranged to fall on one of the special numbers required by an FFT.
The SFT is also useful for applications that cannot tolerate the inaccuracies introduced
by padding with zeros, a handicap of the FFT.
The SFT is also suited to applications where the data is not sampled at evenly spaced
intervals or where sample points are missing. Finally, the SFT can be used to provide
more closely spaced points in the frequency domain than can be obtained with an FFT.
(In an FFT, adjacent points are separated by half the sampling frequency. Points arbi-
trarily close in frequency can be obtained using an SFT.)
There are many standard numerical integration techniques available for computing SFTs
from sampled data. Whichever technique is selected for the problem at hand, it will
probably be much slower than an FFT of a similar number of points. This is becoming
less of an issue, however, as the speed of modern computers increases.
Digital Filtering
Digital filtering is accomplished in
three steps. First, the signal must be
subjected to a Fourier transform.
Then, the signals amplitude in the
frequency domain must be multiplied
by the desired frequency response. Fi-
nally, the transferred signal must be
inverse Fourier transformed back into
the time domain. Figure 2.17 shows
the effect of digital filtering on the
noisy signal. Note that the solid line
represents the unfiltered signal, while
the two dashed lines represent differ-
ent digital filters.
Digital filtering is advantageous because the filter itself can be easily tailored to any
frequency response without introducing phase error. However, one disadvantage of
digital filtering is that it cannot be used for anti-aliasing.
Analog Filtering
In contrast to digital filtering, analog filtering can be used for anti-aliasing, but it is more
difficult to change the frequency response curves, since all analog filters introduce some
element of phase error.
Time [seconds]
V
o
l
t
s
1.546
1.534
1.536
1.538
1.54
1.542
1.544
0 0.05 0.1 0.15 0.25 0.2
50Hz low-pass digital filter
5Hz low-pass digital filter
Unfiltered signal
Fig. 2.17: The effect of digital filtering on the noisy signal
18
Sampling Hints
There are several important sampling hints to observe when designing an application.
These hints are not absolutes nor do they guarantee optimal results. However, they do
provide a useful starting point for planning frequency analysis of a physical process.
These sampling hints include:
A Fouri er transforms hi ghest meani ngful frequency i s one-hal f of the
sampling frequency
The sampling rate should be at least three to five times the highest frequency
of interest
An anti-aliasing low-pass filter is typically required; the cutoff frequency should
be close to the signals highest frequency of interest
Digital filtering can be used to smooth the data or to remove noise in a speci-
fied range after acquisition; however, aliasing can only be prevented with an
analog low-pass filter
If the phase relationship between multiple signals is important, a simulta-
neous sample and hold circuit should be used (see multiplexing in Chapter 3)
Fourier transform resolution is inversely proportional to measurement time;
acquiring data over a longer period of time results in narrower peaks in the
Fourier transform.
19
Chapter 3 Multiplexing
Chapter 3
Multiplexing
20
Multiplexing
A multiplexer is a switch that allows a
single ADC to measure many input
channels. This switch can be imple-
mented via relays or solid-state switches.
A relay is a mechanical switch, so rates
are relatively slow (less than 1 kHz for
reed relays, which are the fastest type),
but large voltages and high isolation
(several kV) can be achieved. The cur-
rent capacity of a relay is determined by
its size and contact type. Currents of 3A
are typical in relays used in laboratory
instruments. Much larger currents can
be switched with the larger relays com-
mon in industrial applications.
Solid-state switches are much faster than relays, and speeds of several MHz are com-
mon. However, these small devices are easily destroyed by voltages larger than 25V,
and they are a poor choice for isolated applications. Solid-state switches typically sup-
port currents less than 1 mA.
A multiplexing scheme such as that shown in Figure 3.01 eliminates the high cost of
having multiple ADCs. A programmable gain amplifier (PGA) allows each channel to
have a different gain and input range. Multiplexing reduces the rate at which data can
be acquired from an individual channel, because multiple channels are scanned se-
quentially. For example, an ADC that can sample a single channel at 100 kHz is limited
to a 12.5-kHz per-channel sampling rate when sampling eight channels.
IOtechs 100-kHz and 1-MHz data acquisition systems are examples of systems that use
software selectable channel and gain sequencing. IOtechs 100-kHz systems provide a
512-location scan sequencer that allows you to select, via software, each channel and its
associated input amplifier gain for both the built-in channels and the expansion channels.
The sequencer circuitry circumvents a major limitation encountered with many plug-in
data acquisition boardsa drastic reduction in the scan rate for expansion channels. All
channels are scanned, including expansion channels, at 100 kHz (10 s/channel). Digital
inputs can also be scanned using the same scan sequence employed for analog inputs,
enabling the time correlation of acquired digital data to acquired analog data. These prod-
ucts permit each scan group, containing up to 512 channel/gain combinations, to be re-
peated immediately or at programmable intervals of up to 12 hours. Within each scan
group, consecutive channels are measured at a fixed 10 s/channel rate. Figure 3.02 illus-
trates a 512-location scan sequencer operating in a 100-kHz data acquisition system.
Analog
inputs
PGA
ADC
Mux
control lines
Mux
Gain
control lines
Fig. 3.01: The multiplexer (Mux) is a fast switch that
directs different input signals to the ADC for digitizing
21
#2 #4 #7 #2 D #18 #19 #16
x1 x8 x8 x2 x100 x10 x1000
Uni Uni Bi Uni Bi Bi Uni
Channel
1
Gain
2
Unipolar or Bipolar
3
4 5
1. Unipolar or bipolar operation can be programmed for each channel dynamically by the sequencer
2. Gain can be programmed for each channel dynamically by the sequencer
3. Channels can be sampled dynamically by the sequencer
4. Expansion channels are sampled at the same rate as on-board channels
Scan group
All channels within a scan group are
measured at a fixed scan rate
Programmable
Constant
scan rate
t
t
Fig. 3.02: 512-location scan sequencer example
Unfortunately, multiplexing may introduce problems. A high source impedance can com-
bine with stray capacitance to cause settling problems and crosstalk between channels. Mul-
tiplexer impedance can also lead to signal degradation. A solid state multiplexer can have an
impedance of tens of Ohms, whereas a relay typically has a resistance of less than 0.01 Ohm.
Sequence vs. Software Selectable Ranges
Most data acquisition system implementations permit different input ranges, although
the manner in which they do so varies considerably. Some data acquisition systems
allow the input range to be switched or jumper selected on the circuit board. Others
provide software selectable gain; this is more convenient, but a distinction should be
made between data acquisition systems whose channels must all have the same gain,
and systems that allow you to sequentially select the input range of each channel. It is
often advantageous to have different input ranges on different channels, especially when
measuring signals from different transducers. Thermocouples and strain gages require
input ranges of tens of millivolts, while other transducers might output several volts.
A data acquisition system with a software selectable range can be used to measure differ-
ent ranges on different channels at a relatively slow rate by issuing a software command
to change the gain between samples.
There are two problems with this technique. First, it is slow; issuing a software
command to change the gain of a PGA can take tens or hundreds of milliseconds,
lowering the sample rate to several Hz. Second, the speed of this sequence is often
indeterminate, due to variants in the PC instruction cycle times. So cycling through
it continuously will give samples with an uneven (and unknown) spacing in time.
This complicates time-series analysis and makes FFT analysis impossible since the
FFT requires evenly spaced samples.
22
As discussed earlier, a better implemen-
tation provides a sequencer that con-
trols the channel selection and gain.
The maximum acquisition rate can be
achi eved wi th sequence-sel ectabl e
ranges, whereas the acquisition rate
will slow considerably with software-
selectable ranges if channels require
different ranges. IOtechs 100-kHz
data acquisition systems provide a
512-location scan sequencer that al-
lows you to select each channel and
associated input amplifier gain at
random. These products permit each
scan group to be repeated immedi-
ately or at programmable intervals.
Input Buffering
The impact of source impedance and stray capacitance can be estimated via a simple
formula: the time constant associated with a source impedance (R) and stray capaci-
tance (C) is T=RC.
For example, suppose you want to know the maximum tolerable input impedance
for a 100-kHz multiplexer. The time between measurements on adjacent channels
in the scan sequence is 10 s. In a
length of time T=RC, the voltage er-
ror decays by a factor of 2.718. Re-
ducing the error to 0.005% requires
waiting 10T.
Consequently, a fixed time of 10 s be-
tween scans (T
scan
) and an error of 0.005%
would appear to require a time of T=1
s. In a typical multiplexed data acqui-
sition system, this would yield errors due
to insufficient settling time. The differ-
ence can be explained as follows. Most
100-kHz converters require 2 s (T
samp
)
to sample the input signal. Subtracting
this from the scan time results in a set-
tling time of: T
settle
=T
scan
-T
samp
, T
settle
=8 s.
Fig. 3.03: Multiplexer scheme with sequence selectable gain
Analog
inputs
PGA
16
to
1
Mux
x1, x2, x4, & x8
To data
acquisition
system
2
4
Expansion
channel addressing
Expansion
gain addressing
2
4
Hardware
sequencer
CLK
Triggering
circuitry
Fig. 3.04: Buffering signals before the multiplexer
increases accuracy, especially with high-impedance
sources or fast multiplexing
Inputs
PGA
ADC
Buffer
Buffer
Buffer
Buffer
Mux
23
Fig. 3.05: DBK45 4-channel, low-pass filtering with simultaneous
sample and hold card block diagram
Channel address lines
Sample enable
Mux
To data
acquisition
system
Switched bias resistors
(per channel)
Ch 4
IA S/H
Ch 3
IA S/H
Ch 2
IA S/H
Ch 1
IA S/H
LPF
LPF
LPF
LPF
If we assume a typical 16-bit data acquisition system, the internal settling time
(T
int
) may be 6 s. The external settling time may then be computed as follows:
T
ext
= T
settle
2
-T
int
2
, T
ext
=5.29 s.
For a 16-bit data acquisition system with 100 picofarad of input capacitance
(C
in
) and a multiplexer resistance (R
mux
) of 100 Ohms, the maximum external resis-
tance is calculated as follows: R
ext
=(T
ext
/C
in
1n(2
16
))-R
mux
, R
ext
=4,670 Ohms.
The above examples are simplified and do not included any effects due to multiplexer
charge injection or inductive reactance in the measurement wiring. In actual practice
the practical upper limit on source resistance is between 1,500 and 2,000 Ohms.
Most input signals have source impedances of less than 1.5 KOhm, so such a maximum
source impedance is usually not a problem. However, faster multiplexer rates require lower
source impedances. For example, a 1-MHz multiplexer in a 12-bit system requires a source
impedance of under 1.5 KOhm. If the source impedance exceeds this value, buffering is
necessary to obtain accurate measurements. A buffer is an amplifier with high input im-
pedance and very low output impedance. Placing a buffer on each channel between the
transducer and the multiplexer eliminates inaccuracies by preventing the multiplexers stray
capacitance from having to discharge through the impedance of the transducer. IOtechs
WaveBook, LogBook, Personal Daq, and select signal conditioning options are all high-
speed devices that use this configuration. This arrangement is illustrated in Figure 3.04.
Simultaneous Sample and Hold (SS&H)
A multiplexed ADC measurement introduces a time skew among channels, which is intol-
erable in some applications. Employing simultaneous sample and hold (SS&H) on multiple
channels remedies time-skew
probl ems. Si mul taneous
sample and hold requires that
each channel be equipped with
a buffer that samples the sig-
nal at the beginning of the scan
sequence. The signal at the
buffer output is held at the
sampled value while the mul-
tiplexer switches through all
channels and the ADC digitizes
the readings. In a good simul-
taneous sampl e and hol d
implementation, all channels
are sampled within 100 ns of
each other.
24
Figure 3.05 shows a common scheme for simultaneous sample and hold; this design is used
on IOtechs DBK45 simultaneous sample and hold card, an expansion option for IOtechs
100-kHz data acquisition systems. Each input signal passes through an instrumentation
amplifier (IA) and into a sample and hold buffer (S/H). When the sample enable line goes
high, each S/H samples its input signal and holds it while the multiplexer switches through
the readings. This scheme ensures that all the samples are taken within 50 ns of each other,
even with up to sixty-four DBK45s connected to a single instrument. A system configured
with sixty-four DBK45s would provide 256 simultaneous channels.
Current Measurements
Many transducers can be configured to
output a 4-20 mA current with a lin-
ear relationship to the quantity being
measured. A shunt resistor converts
the current to a voltage for measure-
ment by an ADC. Figure 3.06 illus-
trates current measurement. If the cur-
rent (I) is passed through a resistance
(R), the voltage across the resistor is
V=IR. Consequently, a 500 Ohm re-
sistor is sufficient to map a 20 mA full-
scale current into a 10 VFS voltage for
measurement with a data acquisition
system. Using a transducer configured
for current output can yield high noise immunity and accuracy, particularly if there is a
large distance between the data acquisition system and the transducer. Long leads pick
up noise and suffer an ohmic voltage drop.
The current value inferred from measuring the voltage across a shunt resistor is only as
accurate as the resistor. Common resistor accuracies are 5%, 1%, 0.1%, and 0.01%.
Resistors are commonly specified according to their accuracy and stability when ex-
posed to changing temperatures. All resistors change with temperature. This phenom-
enon can be used to measure temperature, as discussed in Chapter 4, however, this
effect is undesirable when measuring current. Many resistor manufacturers make an
effort to lower the effects of changing temperatures on resistance. This specification
means that the actual resistance is within the specified tolerance of the given resis-
tance. In addition, a 0.1% resistor usually has a lower temperature coefficient and
better long-term stability than a resistor with lower accuracy, due to its construction
technique, which is usually wire wound or metal film vs. carbon composition.
Fig. 3.06: Current measurement using a shunt resistor
and a differential amplifier
100k Ohm
R
+
ADC
i
25
Chapter 4 Temperature Measurement
Chapter 4
Temperature Measurement
26
This chapter examines various transducers for measuring temperature: thermocouples, and
RTDs. It also discusses the required signal conditioning, as well as various techniques for
optimizing the accuracy of your temperature measurement.
Thermocouples
Thermocouples are probably the most
widely used and least understood of all
temperature measuring devices. Ther-
mocouples provide a simple and effi-
cient means of temperature measure-
ment by generating a voltage that is a
function of temperature. All electrically
conducting materials produce a thermal
electromotive force (emf or voltage dif-
ference) as a function of the tempera-
ture gradients within the material. This
is called the Seebeck effect. The amount
of the Seebeck effect depends on the
chemical composition of the material
used in the thermocouple.
When two different materials are connected to create a TC, a voltage is generated.
This voltage is the difference of the voltages generated by the two materials. In
principle, a thermocouple can be made from almost any two metals. In practice,
several thermocouple types have become de facto standards because they possess
desirable qualities, such as highly predictable output voltages and large voltage-to-
temperature ratios. Some common thermocouple types are J, K, T, E, N28, N14, S, R,
and B. Figure 4.01 depicts a Type T thermocouple. In theory, the temperature can
be inferred from such a voltage by consulting standard tables or using linearization
algorithms. In practice, this voltage cannot be directly used, because the connec-
tion of the thermocouple wires to the measurement device constitutes a thermo-
couple providing another thermal emf that must be compensated for; cold junction
compensation can be used for this purpose.
Cold Junction Compensation. Figure 4.02 shows a thermocouple that has been placed
in series with a second thermocouple, which has been placed in an ice bath. This ice
bath is called the reference junction. And when properly constructed, an ice bath can
produce an extremely accurate 0C temperature. This reference junction is required
because all thermocouple emf tables, as published by NIST (National Institute of Stan-
dards and Technology), are referenced to the emf output of a thermocouple which is
Fig. 4.01: Illustration of a Type T thermocouple
Metal A (copper)
Metal B (constantan)
+
-
Type T
e
AB
27
held at a temperature of 0.00 degrees
Celsius. In figure 4.02, the chromel/
copper and alumel/copper thermo-
couples in the ice bath have a 0.0V
contribution to the voltage measured
at the meter. The voltage read is en-
tirely from the chromel/alumel ther-
mocouple. The copper wires con-
nected to the copper terminals on the
meter do not constitute a thermo-
couple because they are the same
metal connection and both terminals
are the same temperature.
Maintaining an ice bath and an ad-
ditional reference thermocouple for
every thermocouple probe is not practical in most systems. If we know the tempera-
ture at the point where we connect to our thermocouple wires, ( the reference junc-
tion ), and if both connecting wires are at the same temperature, then the ice bath
can be eliminated. In such a case, the opposing thermoelectric voltage generated by
the nonzero temperature of the reference junction can simply be added to the ther-
mocouple emf. This is cold junction compensation (CJC).
CJC is an essential part of accurate thermocouple readings. CJC must be implemented
in any system that has no ice-point reference junction. The technique works best if the
CJC device is close to the terminal blocks that connect the external thermocouples, and
if there are no temperature gradients in the region containing the CJC and terminals.
Thermocouple Linearization. Thermocouple voltage is proportional to, but not lin-
early proportional to, the temperature at the thermocouple connection. There are
several techniques for thermocouple linearization. Analog techniques can provide
a voltage proportional to temperature from the thermocouple input. In addition, a
voltage measurement can be made with an ADC and the temperature looked up in a
table, as in the above example. To speed this process, the lookup table can be stored
in computer memory and the search performed with an algorithm. Thermocouple
linearization can also be accomplished using a polynomial approximation to the
temperature versus voltage curve. Figure 4.03 on the next page illustrates a portion
of the NIST Polynomial Coefficients table for some common thermocouples.
Fig. 4.02: Thermocouple circuit with ice bath
+
-
+
-
Copper
Copper
Copper
V
Volt meter
Ice Bath
Copper
Alumel
V
1
Chromel
28
Temperature conversion equation: T =a
0
+a
1
x +a
2
x
2
+... +a
n
x
2
Nested polynomial form: T =a
0
+x(a
1
+x(a
2
+x(a
3
+x(a
4
+a
5
x)))) (5th order)
NIST Polynomial Coefficients
160C to 400C
0.5C
7th order
a
0
a
1
a
2
a
4
a
3
a
5
a
6
a
7
a
8
a
9
0C to 760C
1C
5th order
Type J
Iron(+) vs.
Constantan()
0.048868252
19873.14503
218614.5353
264917531.4
11569199.78
2018441314
0C to 1370C
0.7C
8th order
Type K
Nickel-10%
Chromium(+) vs.
Nickel5% ()
(Aluminum Silicon)
0.226584602
24152.10900
67233.4248
860963914.9
2210340.682
4.83506E +10
1.18452E +12
1.38690E +13
6.33708E +13
0C to 1000C
0.5C
8th order
Platinum-13%
Rhodium(+) vs.
Platinum()
0.263632917
179075.491
48840341.37
4.82704E +12
1.90002E +10
7.62091E +14
7.20026E +16
3.71496E +18
8.03104E +19
Type R
0C to 1750C
1C
9th order
Platinum-10%
Rhodium(+) vs.
Platinum()
0.927763167
169526.5150
31568363.94
1.63565E +12
8990730663
1.88027E +14
1.37241E +16
6.17501E +17
1.56105E +19
1.69535E +20
Type S
Copper(+) vs.
Constantan()
0.100860910
25727.94369
767345.8295
9247486589
78025595.81
6.97688E +11
2.66192E +13
3.94078E +14
Type T
100C to 1000C
0.5C
9th order
Type E
Nickel-10%
Chromium(+) vs.
Constantan()
0.104967248
17189.45282
282639.0850
448703084.6
12695339.5
1.10866E +10
1.76807E +11
1.71842E +12
9.19278E +12
2.06132E +13
For example, IOtechs DBK19 thermocouple card includes a software driver that con-
tains a temperature conversion library that converts raw binary thermocouple chan-
nels, and CJC information into temperature. Included DaqView software provides au-
tomatic linearization and CJC for thermocouples attached to the system.
Additional Concerns
Care must be taken when using thermocouples to measure temperature. Sources of mi-
nor error can add up to highly inaccurate readings. Additional concerns include:
Thermocouple Assembly. Thermocouples are assembled via twisting wires together, sol-
dering, or welding; if done improperly, all of these techniques can introduce errors in
the temperature measurement.
Twisting Wires Together. Thermocouple junctions should not be formed by twisting the
wires together. This will produce a very poor thermocouple junction with large errors.
Soldering. For low temperature work the thermocouple wires can be joined by solder-
ing, however, soldered junctions limit the maximum temperature that can be measured
(usually less than 200 degrees Celsius). Soldering thermocouple wires introduces a third
metal. This should not introduce any appreciable error as long as both sides of the
junction are the same temperature.
Fig. 4.03: Portion of an NIST lookup table
29
Welding. Welding is the preferred method of connecting junctions. When welding
thermocouple wires together, care must be taken to prevent any of the characteristics of
the wire from changing as a result of the welding process. These concerns are compli-
cated by the different composition of the wires being joined. Commercially manufac-
tured thermocouples are typically welded using a capacitance-discharge technique that
ensures uniformity.
Decalibration of Thermocouple Wire. Decalibration is another serious fault condition.
Decalibration is particularly troublesome because it can result in an erroneous tempera-
ture reading that appears to be correct. Decalibration occurs when the physical makeup
of the thermocouple wire is altered in such a way that the wire no longer meets NIST
specifications. This can occur for a variety of reasons including, temperature extremes,
cold-working of the metal, stress placed on the cable during installation, vibration, or
temperature gradients.
Insulation Resistance Failure. Tempera-
ture extremes can also introduce error
because the insulation resistance of the
thermocouple will often decrease expo-
nentially as the temperature increases.
This can lead to two types of errors: leak-
age resistance with an open thermo-
couple and leakage resistance with small
thermocouple wire. Figure 4.04 illus-
trates insulation resistance failure.
Leakage Resistance with an Open Ther-
mocouple. In high temperature appli-
cations, the insulation resistance can
degrade to the point where the leak-
age resistance R
L
will complete the cir-
cuit and give an erroneous reading.
Leakage Resistance with Small Thermocouple Wire. In high temperature applications using
small thermocouple wire, the insulation R
L
can degrade to the point where a virtual junc-
tion T
1
is created and the circuit output voltage will be proportional to T
1
instead of T
2
.
Furthermore, high temperatures can cause impurities and chemicals within the ther-
mocouple wire insulation to diffuse into the thermocouple metal, changing the charac-
teristics of the thermocouple wire. This causes the temperature-voltage dependence to
deviate from published values. When thermocouples are used at high temperatures,
Fig. 4.04: Illustration of a virtual junction resulting from
insulation resistance failure
To DVM
Virtual junction
To DVM
(Open)
Leakage resistance
R
s
R
s
R
s
R
s
R
L
T
1
T
2
R
L
30
the atmospheric effects can be minimized by choosing the proper protective insulation.
Due to the range of thermocouple choices, thermocouple quality is an issue. Select the
thermocouple that meets your application criteria (i.e., high or low temperature range,
proper grounding, etc.)
Other Thermocouple Measurement Issues
For high-accuracy and instrumentation-quality thermocouple measurements, issues such
as thermocouple isolation, auto-zero correction, line-cycle noise rejection, and open
thermocouple detection must be addressed.
Thermocouple Isolation. Thermocouple isolation reduces noise and the error effects of
ground loops. Especially in cases where many thermocouples are bolted directly to a
distant metallic object, like an engine block. The engine block and the thermocouple-
measurement instrument may have different ground references. Without some form of
isolation, a ground loop would be created that would cause the readings to be in error.
Auto-Zero Correction. To minimize the effects of time and temperature drift on the
systems analog circuitry, a shorted channel should be constantly measured and sub-
tracted from the reading. Although very small, this circuit drift can become a substan-
tial percentage of the low-level voltage supplied by a thermocouple.
One method of subtracting the offset
due to drift is employed by IOtechs
TempScan/1100, MultiScan/1200, and
ChartScan/1400, a highly accurate ana-
log scheme is used to automatically
subtract the system offset. Before read-
ing any channel, the internal channel
sequencer first switches to a reference
node and stores the offset error on a
capacitor. As the thermocouple chan-
nel is switched onto the analog path,
the circuit holding the offset voltage
is applied to the offset correction in-
put of a differential amplifier, nulling
out the offset automatically. Figure
4.05 illustrates this technique.
Line-Cycle Noise Rejection. Because of the relatively small voltage generated by
most thermocouples, noise is always an issue. The most pervasive source of noise is
from power lines (50 Hz or 60 Hz). Since thermocouples have a bandwidth lower
Ch x
Ch ground
A
Amplifier's
offset correction
Differential amplifier
To A/D
A
A
B B
Control for "A" muxes
A/D sample
Control for "B" muxes
Auto zero
phase
Sampling
phase
Fig. 4.05: Diagram of auto-zero correction
31
than 50 Hz, a simple per-channel fil-
tering scheme can reduce the incom-
ing AC line noise. For systems with
a multiplexed front end, an RC filter
is not recommended. Although more
complicated, an active filter made up
of an op-amp and a few passive com-
ponents can also be used; however,
this method incurs significant addi-
tional expense. Furthermore, setup
is complicated because each channel
needs to be calibrated due to gain and
offset error. For these reasons active
filters are generally not used. Figure
4.06 is a diagram of an active filter.
One unique line-cycle averaging technique for reducing noise is employed by the
TempScan/1100, MultiScan/1200 and ChartScan/1400. A phase-lock-loop circuit
monitors and synchronizes the systems internal sampling rate to the multiple of
the power line frequency. During each line cycle, the CJC is sampled along with
other input channels that are sampled 32 times and averaged. This has the effect of
reducing the line cycle noise by 82%.
Op e n The rmo c o up le De t e c t io n.
Open thermocouple detection is es-
pecially important in systems with
high channel counts. Due to age, vi-
brati on, and handl i ng, thermo-
couples can break or become highly
resistive. Open thermocouple detec-
tion can be accomplished by provid-
ing a low level current path to a ca-
pacitor across the thermocouple ter-
minals. If the thermocouple is open,
the capacitor charges up and drives
the amplifier to its rail, indicating a
fault. When the thermocouple is
properly connected, the low-level
current goes through the thermocouple rather than charging the capacitor. Figure
4.07 illustrates an open thermocouple circuit.
Fig. 4.06: Diagram of active filter
R
1
R
2
+
Fig. 4.07: Open thermocouple circuit

TC
To differential
amplifier
+0.13V
10 M
10 M
0.13V
4700 pF
One of 32 channels
TempScan/1100
open TC Detect
32
Galvanic Action. Some thermocouple insulation contains dyes that form an electrolyte
in the presence of water. This creates a galvanic action, with a resultant output hun-
dreds of times greater than the net open circuit voltage. Precautions should be taken to
shield thermocouple wires from harsh atmospheres and liquids.
Thermal Shunting. Because it takes energy to heat mass and thermocouples cannot
be constructed without mass, the thermocouple will slightly alter the temperature it
was meant to measure. The mass of the thermocouple can be minimized using
small wire, but a thermocouple made with small wire is far more susceptible to the
problems of contamination, annealing, strain, and shunt impedance. Over long dis-
tances, thermocouple extension wire should be used to minimize these effects. Com-
mercial thermocouple extension wire is constructed out of material having net open-
circuit voltage coefficients similar to specific thermocouple types. It is constructed
so that its series resistance is minimal over long distances; it can be pulled through
conduit more easily, and it is generally operable over a much lower temperature
range than premium-grade thermocouple wire. In addition to offering a practical
size advantage, extension wire is less expensive than standard thermocouple wire,
especially with platinum-based thermocouples.
Extension wire is generally specified over a narrower temperature range and is more
likely to receive mechanical stress, as such the temperature gradient across the extension
wire should be kept to a minimum. These simple precautions should ensure that tem-
perature measurements will not be affected by the use of extension wire.
Improving the Accuracy of Thermocouple Wire Calibration. Thermocouple wire is
manufactured so that it conforms to NIST specifications. These specifications can
often be enhanced by calibrating the wire on site. That is accomplished by testing
the wire at known temperatures. (Note: If the wire is calibrated to improve tempera-
ture measurements, then it becomes even more imperative that care be taken to
prevent decalibration.)
33
Resistance Temperature Detectors (RTD)
RTDs function on the basic physical principle that a metals resistance increases with
temperature. Most RTDs are simply wire-wound or thin-film resistors made of a wire
with a known resistance versus temperature relationship. Platinum is the most com-
monly used material for RTDs. RTDs are available in a wide range of accuracy specifica-
tions; the most accurate RTDs are used as temperature standards at NIST.
Platinum RTDs have resistance values ranging from 10 Ohms for a birdcage model to
ten thousand Ohms for a film model. The most common value for RTDs is 100 Ohms at
0C. Platinum wire has a standard temperature coefficient of =0.00385 //C,
which means a 100 Ohm wire has a temperature coefficient of +0.385 /C at 0C; the
value for is actually the average slope from 0C to 100C. Chemically pure platinum
wire has an of +0.00392 //C. The slope and the absolute value are relatively
small, when you consider the fact that the measurement wires leading to the sensor
may be tens of Ohms. With this in mind, even a small lead impedance can cause an
appreciable error in temperature measurement.
The following equation states that the change in resistance is equal to multiplied by the
change in temperature.
R =
In the above measurement example, a 10 Ohm lead impedance suggests a 10/0.385 @26C
error. In addition to the impedance, the lead wires temperature coefficient can also con-
tribute a measurable error.
To eliminate the possibility of error as a
result of additional resistance from the
current-carrying excitation leads, an ad-
ditional set of voltage-sensing leads
should be located as close as possible to
the sensor. (Recall that the temperature
coefficient is 0.39/C, so a 0.39 error
in the resistance measurement yields a
1C error in the measured temperature.)
This configuration is called a four-wire
RTD measurement because four wires are
required between the RTD and the mea-
surement instrumentation. Placing the
sense leads at the current supply instead
of the RTD provides a less accurate con-
figuration but only requires connecting
two wires to the RTD.
RTD
R
line
R
line
Ohmmeter
Fig. 4.08: Two-wire RTD measurement
34
The five common types of circuits,
which are used for RTD measurement
are shown in Figure 4.08 through 4.12.
Following are descriptions of each
circuits operation.
Figure 4.08 (shown on the previous
page) shows a simple two-wire resistance
measurement. The RTD resistance is
read directly from the Ohmmeter. This
connection is rarely used since the lead
wire resistance must be a known value
and it is very difficult to determine the
temperature coefficient of the lead wires.
Figure 4.09 shows a four-wire measure-
ment using a current source. The RTD
resistance is volts/current. This connec-
tion requires a high stability current
source and four lead wires.
Figure 4.10 shows a three-wire mea-
surement using a current source. The
RTD resi stance i s deri ved usi ng
Kirchhoff's voltage law as follows:
RTD in Ohms =V
a
- (2 * V
b
)/current.
The benefit of this connection over
Figure 4.09 is the elimination of one-
lead wire. This connection assumes
that the two-current carrying wires
have the same resistance.
Figure 4.11 shows a four-wire mea-
surement using a voltage source. The
RTD resi stance i s deri ved usi ng
Kirchhoff's voltage law as follows:
RTD in Ohms = V
a
-((V
a
- V
b
)+(V
c
-
V
d
))/(V
d
/R
SENSE
). The voltage source in
this circuit can vary as long as the
sense resistor (R
SENSE
) is stable. This
connection requires four-lead wires.
Fig. 4.10: Three-wireRTD measurement with a current source
RTD
R
line
V
a
V
b
R
line
R
line
High
impedance
voltmeter
Current
source
Current
source
RTD
R
line
R
line
R
line
R
line
High
impedance
voltmeter
Fig. 4.09: Four-wireRTD measurement with a current source
RTD
R
line
V
a
V
b
V
c
V
d
R
line
R
line
R
line
High
impedance
voltmeter
Voltage
source
R
sense
Fig. 4.11: Four-wire RTD measurement using a voltage
source
35
Figure 4.12 shows a three-wire measure-
ment using a voltage source. The RTD
resistance is derived using Kirchhoff's
voltage law as follows: RTD in Ohms =
V
a
-((2 * (V
a
- V
b
))+(V
c
- V
d
))/(V
d
/R
SENSE
).
The voltage source in this circuit can
vary as long as the sense resistor (R
SENSE
)
is stable. This connection assumes that
the two-current carrying wires have the
same resistance.
While the RTD is more linear than
the thermocouple, its operating range
is smaller than a thermocouples. For
maximum accuracy in RTD applica-
tions you should use curve-fitting via
the Callendar-Van Dusen equation,
which follows:
R
T
= R
0
+ R
0

[ ] (
T
T
100
1
) ) (
T
100

(
T
100
1
) ) (
T
3
100
Where:
R
T
= resistance at temperature T
Ro = resistance at T = 0 C
= temperature coefficient at T = 0 C
typically +0.00392//C
= 1.49 (typical value for .00392 platinum)
= 0, When T>0
0.11 (typical) When T<0
The exact values for coefficients , ,
and , are determined by testing the
RTD at four temperatures and solving
the resul ti ng equati ons. Thi s
Callendar-Van Dusen equation was
replaced by a 20th order polynomial
to provide a more accurate curve fit.
The plot of this equation in Figure 4.13
shows the RTD to be more linear than
the thermocouple when used below
800C (RTDs are not appropriate for
applications in which the temperature
being measured will exceed 800C).
Fig. 4.13: Type S thermocouple vs a Platinum RTD
R
e
s
i
s
t
a
n
c
e

t
e
m
p
e
r
a
t
u
r
e
c
o
e
f
f
i
c
i
e
n
t
-
R
T
D
Equivalent linearities type S thermocouple
Platinum RTD
Temperature, C
0 200 400 600 800
.390
.344
.293
12
8
4
Fig. 4.12: Three-wire measurement using a voltage
source.
RTD
V
a
V
b
V
c
(not used)
V
d
R
line
R
line
R
line
High
impedance
voltmeter
Voltage
source
R
sense
36
Another source of error in RTD measurements is resistive heating. A current, I,
through a resistance, R, dissipates power P =I
2
R. For example, a 1 mA current through
a 100 RTD generates 100 watt of power. This may seem insignificant, but it can
raise the temperature of some RTDs a significant fraction of a degree. (A typical RTD
has self-heating of 1C/mW. Smaller RTDs offer fast time response, but can have
larger self-heating errors.)
Since lower currents generate less heat; currents between 100 and 500 A are commonly
used. This lowers the power dissipation to between 10 and 25 W, which is tolerable for
most applications. Further reducing the current inhibits accurate measurement because
as the currents become smaller they are susceptible to noise and more difficult to mea-
sure. The RTDs heat can be reduced below 10 W by switching the current on only
when the measurement is made. In a multichannel system, the excitation current can
be multiplexed much like the analog inputs. For example, in a 16-channel system, the
current will only excite each RTD 1/16th of the time, reducing the power delivered to
each RTD from 100% to 6%, (P*1/X where X =the number of channels).
Two practical methods of measuring an RTD include: constant current and ratiometric.
An example of a constant current topology can be found in the TempScan/1100s
RTD scanning module, the TempRTD/16B (see figure 4.14), which provides a single
constant current source of approximately 500 amps, which is switched among its
sixteen channels. A series of front
end multiplexers direct the current
to each channel sequentially while
the measurement i s bei ng taken.
Both 3- and 4-wire connections are
supported to accommodate both
types of RTDs. By applying current
to only a single RTD at a time, errors
due to resistive heating become neg-
ligible. Advantages of the constant-
current method include circuit sim-
plicity and noise immunity. The dis-
advantage of using constant current
is the expense of building or buying
a highly stable current source.
Fig. 4.14: TempRTD/16B card in a TempScan/1100
Constant current
Ch x
To A/D
24 Channels
RTD
Ch x +1
RTD
37
Fig. 4.16: Diagram of DBK9 front end with 3-wire
configuration
RTD
Constant
voltage
Precision
resistor
V
100 RTD
V
500 RTD
V
1000 RTD
V
a
V
b
V
c
i
S
R
d
i
S
V
d
R
L
R
L
The DBK9 RTD measurement card for
IOtechs LogBook, DaqBoard, DaqBook,
and Daq PC-Card based data acquisition
products uses the ratiometric method for
measuring RTDs. A simple constant volt-
age source provides a current, I
s
, through
the RTD and resistor R
d
. Four readings,
V
a
, V
b
, V
c
, and V
d
, are taken for each
RTD channel. The current through the
RTD, I
s
, is V
d
/R
d
. The voltage across the
RTD, V
rtd
, is equal to V
b
-V
c
. The RTD re-
sistance is, therefore, V
rtd
/I
s
.
For a 3-wire connection scheme, the
voltage Va-Vc includes the voltage drop
across only one of the leads. Since the
2 extension wires to the transducer are
made of the same metal, one can as-
sume that the drop in the first wire is
equal to the drop in the second wire.
Therefore, the real voltage across the
RTD would be V
RTD
=V
a
-2(V
a
-V
b
)-V
d
. The
resistance of the RTD can be calculated
by R
RTD
=R
d
[V
RTD
/V
d
].
Fig. 4.15: Ratiometric DBK9 front end with 4-wire
configuration
RTD
Constant
voltage
Precision
resistor
V
a
V
b
V
c
V
d
i
S
R
d
i
S
V
1000 RTD
V
500 RTD
V
100 RTD
R
L
R
L
38
Small Resistance vs. Large Resistance RTDs
Small RTD Large RTD
Response Time Fast Slow
Thermal Shunting Low Poor
Self-heating Error High Low
Practical Precautions
RTDs require the same precautions that apply to thermocouples, including the use of
shields and twisted-pair wire, proper sheathing, avoiding stress and steep gradients, and
using large extension wire.
In addition, the following issues come into play: the RTD is more fragile than the
thermocouple and needs to be protected during use, and RTDs are not self-powered.
An excitation current must be provided to the device to obtain a measurement. The
current causes Joule heating within the RTD, and this changes the RTDs tempera-
ture. This self-heating appears as a measurement error and must be included in the
calculations that determine measurement. A typical value for self-heating error is
1C per milliwatt in free air. An RTD immersed in a thermally conductive medium
will distribute this heat to the medium and the resulting error will be smaller. The
same RTD that rises 1C per milli-
watt in free air will rise only 1/10C
per milliwatt in air that is flowing
at the rate of one meter per second.
Using the minimum excitation cur-
rent that provides the desired reso-
lution and using the largest physi-
cally practical RTD will help reduce
self-heating errors.
Thermal Shunting. Thermal shunting results in an altered temperature reading as a result
of introducing a measurement transducer. Thermal shunting is of greater concern with
RTDs because RTDs are larger than thermocouples.
39
Chapter 5 Strain & Acceleration
Chapter 5
Strain & Acceleration
40
This chapter examines the various transducers for measuring strain and acceleration. It
also discusses the required signal conditioning, as well as various techniques for ensur-
ing measurement accuracy.
Strain Gages
Strain gages are used in a variety of test and measurement applications. Strain gages
can measure most physical phenomena that elongate or contract a solid material. Strain-
gage techniques are also used in many accelerometer and pressure-transducer designs.
Strain is defined as the change in length per unit length. For example, if a 1m long bar
is stretched to 1.000002 m, the strain is defined as 2 microstrains.
Strain can be measured by bonding a resistive strain gage to the test material. When
the test material is strained it transfers that strain to the resistive strain gage, chang-
ing its resistance slightly. The gage factor is the fractional change in resistance
divided by the strain. For example, two microstrain applied to a gage with a gage
factor of two causes a fractional resistance change of 4x10
-6
Ohm. Common gage
resistance values are from 120 Ohm to 350 Ohm, but in some applications, resis-
tance can be as low as 30 Ohm or as high as 3000 Ohm.
The Wheatstone Bridge. To make an accurate strain measurement, very small resistance
changes must be measured. As in the previous example, if two microstrain are applied to a
gage, a change in resistance of only 4 Ohm would need to be measured. The Wheatstone
bridge converts a change in resistance into a voltage ready for ADC measurement. If all
four resistors are equal, then V
out
=0. When one of the resistors changes slightly (by a
fractional amount X), then a measurable voltage is produced.
Full-Bridge Configuration. The optimal strain-gage configuration is the full bridge.
This configuration provides the highest sensitivity and the fewest error components.
Since the full bridge configuration also provides the greatest output, noise is less of
a factor in the measurement. For these reasons, wherever possible, the full bridge
configuration is recommended.
A full bridge contains four strain gages mounted on a test member, two on the tensile
surface and two on the opposite compressive surface. As the member deflects, two gages
increase in resistance while the opposite gages decrease in resistance, maximizing the
bridges output.
In this configuration, the following equation applies: V
out
=V
exc
* X where X is the
change in resistance (R). Figure 5.01 illustrates a full-bridge configuration.
41
Potential error components such as temperature change are nulled by the bridge since
all four strain gages have the same temperature characteristics. In a full-bridge configu-
ration, the resistance in the lead wires has no effect on the accuracy of the measurement
as long as the input amplifier has a high input impedance. For example, with an input
resistance of 100 MOhm, very little current flows through the measurement leads, mini-
mizing the possible voltage drop due to lead resistance.
Half-Bridge Configuration. When physical considerations do not allow for a full bridge, a
half bridge can be employed. In this configuration, two strain gages are mounted on a test
member while two resistors are used to complete the bridge.
In this configuration, the following equation applies: V
bridge
=V
excitation
* X/2 where X is
approximately the change in resistance (R). For large R, half-bridge and quarter-
bridge configurations can introduce an additional nonlinearity error. Figure 5.02 illus-
trates a half-bridge configuration.
One potential error source with half bridges is that, because of the dissimilar tempera-
ture characteristics of bridge completion resistors and strain gages, as their temperature
changes, their resistance may not change proportionately. Furthermore, bridge comple-
tion resistors are typically not in the same physical location as the strain gages, adding to
the error induced by temperature changes.
In systems with long lead wires, it is recommended that the bridge completion resistors
be attached close to the gages; however, this may not always be practical due to environ-
mental conditions.
Fig. 5.01: Full bridge and gage locations
+
+
V
exc
R
1
R
3
R
2
R
4
V
OUT
R
1
R
4
R
2
R
3
42
Quart er-Bridg e Config urat ion. A quarter bridge uses one strain gage and three
bridge completion resistors. I n this configuration, the following equation ap-
plies: V
out
= V
exc
* X/ 4, where X is approximately the change in resistance (R).
Figure 5.03 illustrates a quarter-bridge configuration.
This configuration has the smallest output, making noise a potential problem. Further-
more, all of the error sources and drawbacks in the half-bridge configuration are also
applicable to the quarter-bridge configuration.
Excitation Source. Accurate measurement requires a quiet (low noise) and stable regu-
lated excitation source. A regulated excitation source is necessary because the output of
a strain gage is proportional to the excitation voltage. Therefore, any fluctuation in the
excitation will result in output errors.
Fig. 5.02: Half-bridge and gage locations
Strain
gages
+
Bridge
completion resistors
V
exc V
OUT
+
R
1
R
2
R
1
R
2
Fig. 5.03: Quarter-bridge and gage location
+
Strain
gage
Bridge
completion resistors
V
exc
V
OUT
R
1
R
1
43
An example of a strain-gage measurement system is IOtechs DBK43A, an eight-chan-
nel signal conditioning option for their 100-kHz PC-based data acquisition systems.
The DBK43A provides one adjustable excitation output per channel. Each channel can
be independently adjusted from 1.5V to 10.5V with a current limit of 50 mA. An exci-
tation voltage V, used with a strain gage of resistance R, requires a current of I=V/R.
The composite resistance of a Wheatstone bridge is the same as its components. For
example, a 350 Ohm bridge is made up of four 350 Ohm arms. The load current is
determined by dividing the excitation voltage by the bridge resistance, in this case
10V/350=0.029A (29 mA).
The amount of Ohmic heating in the strain gages should also be considered because a
strain gages performance can be temperature dependent. In most standard configura-
tions, the dissipated power (P=V
2
/R) is less than 100 mW in each gage. If the strain gage
is bonded to a good heat conductor such as metal, dissipated power is typically not a
problem. However, when the medium conducts heat poorly, such as wood, plastic, or
glass, care should be taken in choosing the excitation voltage. Care should also be
taken when the strain gage is unusually small or when a number of strain gages are
placed in proximity. As a general rule, the lowest possible excitation voltage which
yields satisfactory results should always be used.
Finally, one should consider Kelvin
connection when applying excitation
to the bridge. Since the excitation
leads carry a small amount of current,
there is a voltage drop of V=IR
leads
be-
tween the excitation supply and the
strain gages. IOtechs DBK43A, uses
Kelvin connection to measure and
regulate the voltage at the bridge. It
accomplishes this by measuring it
wi th a di fferent set of l eads than
those carrying the current. A Kelvin
connection is also commonly called
a 6-wire connection.
If I
e
=50 mA, R
L
=5 Ohms, and the voltage drop in one lead is 250 mV, both exc wires will
equal 500 mV. There will be no voltage drop in the sense wire because of negligible current.
Fig. 5.04: Kelvin strain-gage bridge connection
R
3
R
2
R
4
R
1
V
out
R
L
R
L
R
L
R
L
R
L
R
L
i
e
+
V
exc
V
sensor
+
i0
i0
i
e
44
Strain-Gage Signal Conditioning
Most strain gages and load cells are specified for a full scale value of weight, force, ten-
sion, pressure, torque, or deflection with an output of mV/V of excitation. For example,
a load cell with a full scale rating of 1,000 lbs might output 2 mV per volt of excitation at
full load. For an excitation of 10V, the full scale output is 20 mV.
Steps must be taken to control noise when measuring mV levels. Shielded cables, low-
pass filters, differential voltage measurement and averaging can be used to minimize
noise in strain-gage measurements.
Reading a mV signal with a 12-bit A/D converter requires an instrumentation amplifier.
Most A/D amplifiers have a 10V full scale and thus, only provide a 2.44 mV resolution
(10V/4095). To measure strain using the A/Ds full resolution, one must adjust the
instrumentation amplifiers gain so that the full scale output of the strain gage or load
cell covers the entire range of the A/D.
Many test systems are characterized by the presence of a quiescent load that results in an
output voltage, even when no force is applied. An adjustable offset can reduce this quies-
cent output voltage to zero, allowing the applied load to use the full range of the ADC.
Fig. 5.05: DBK43A strain gage module block diagram
To
A/D
Exc. Adj.
6
5
2
1
DBK43A (Typical of 8 ch.)
+Sense
Sense
+Exc
Exc
Excitation
Regulator
Bypass
Bypass
LPF
+
Scale Gain Adj. Input Gain Adj.

4
3
Shunt
Cal.
Offset
Adj.
Optional
Bridge
Completion
Resistors
45
Flexibility is the key to maximizing the performance of a strain gage system. For ex-
ample, the DBK43A provides adjustable excitation, gain and offset per channel, allowing
use of the entire dynamic range of the data acquisition system. Figure 5.05 illustrates
one-channel of the DBK43A.
Co mmo n Mo d e Re je ct io n Ra t io .
Common mode rej ecti on rati o
(CMRR) is a vital instrumentation
amplifier specification. A strain-gage
signal in a Wheatstone bridge con-
figuration is superimposed on a com-
mon mode voltage of half the exci-
tation voltage. CMRR measures how
well the amplifier rejects common
mode voltage. For example, consider
a 10V excitation of a strain gage with
2 mV/V output at full scale. An am-
plifier with a CMRR of 90 dB can in-
troduce an error of 0.158 mV, corre-
sponding to 0.8% full scale. This
may not be acceptable. A CMRR of
115 dB is much better, as it intro-
duces an error of only 9 V, which
corresponds to 0.04% of full scale.
dB=20log
10
(V
sig
/V
error
)
V
max
/V
error
=log
10

-1
(dB/20)=log
10

-1
(90/20)=31,622
V
error
=V
max
/log
10

-1
(dB/20)=5.00/31,622=0.158 mV
%error=(V
erro
/V
sig
)100=(0.158 mV/20 mV)100=0.79%
IOtechs DBK43A provides a regulated excitation source with optional Kelvin excitation.
Provision is made for on-board bridge-completion resistors when using quarter- and
half-bridge strain gages. Two instrumentation amplifiers provide input and scaling gain
adjustments. An offset adjustment permits nulling of large quiescent loads, so that the
input signal uses the full range of the data acquisition system. This allows the measure-
ment to have the full resolution of the ADC.
Many strain-gage signal conditioners provide fixed gain, offset, and excitation settings;
typically the manufacturer suggests that the user adjust the system via software. The
problem with this solution is that its generalized; fixed settings do not take advantage of
the maximum dynamic range of the A/D. This decreases the actual available resolution of
the measurement.
Gain = 200
5.02 V
5.00 V 350
347
350
350
20 mV
+10 V
+V
V
IA
V
out
= 200 (0.02)
= 4.00 V
Fig. 5.06: A strain-gage signal in a Wheatstone bridge
configuration superimposed on a common mode voltage
of half the excitation voltage
46
For example, many generic strain-gage signal-conditioning modules can be configured
with a fixed 3 mV/V rating. At 10V, the excitation, offset, and gain trimming are fixed
and no adjustments are available.
The excitation adjustment provides a means for the user to set the excitation to the
maximum allowed by the manufacturer, maximizing the bridges output. The offset ad-
justment allows the user to zero the output offset caused by a slight bridge imbalance or
quiescent deformation of the mechanical member. The gain adjustments allow the user
to set a gain that will provide a full scale output under maximum load, optimizing the
dynamic range of the A/D.
The DBK43A provides a shunt calibration
feature, illustrated in Figure 5.07, that al-
lows the user to switch user-supplied
shunt resistors into either of the two
lower legs of the bridge via software. This
is useful because in some instances, it is
not possible to provide a maximum load
to set the optimum gain. A shunt resis-
tor value can be calculated so that when
it is switched in, it will simulate a full
load, allowing an accurate gain to be set.
For any balanced bridge, there is a resis-
tance value that can be applied in paral-
lel with one of the four bridge elements
to create predictable imbalance and output voltage. For example, a 350 Ohm 2 mV/
V strain gage will deliver full output if one leg drops by 0.8% (about 2.80 Ohm) to
347.2 Ohm. A 43.75K Ohm resistance shunted across one or the other lower bridge
elements will result in full positive or full negative output.
An appropriate formula for the shunt-cal resistance value is:
R
shunt
=R
bridge
arm [V
excitation
/ 4 (V
out
)]
For example:
R
shunt
=350 [10 / 4 (0.020)] =43,750 Ohm
Many products provide calibration software. The DBK43A ships with a Windows-based
program that provides three calibration methods, on-line instructions, and a diagnostic
screen for testing the calibrated system.
R
on
350 350
43.75K
Close
to shunt
350
Voltage 20 mV
when shunted
+10 V
(+)
()
RTN
Fig. 5.07: Schematic of a shunt bridge
47
Load Cells and Torque Sensors
Strain gages are commercially available in prefabricated modules such as load cells,
configured to measure force, tension, compression, and torque. Load cell manufactur-
ers provide calibration and accuracy information. Load cells typically use a full-bridge
configuration and have four leads for bridge excitation and measurement.
Strain Diaphragm Pressure Gages. A strain diaphragm pressure gage consists of two or
four strain gages on a thin diaphragm. The strain gages are wired in a Wheatstone bridge
configuration, including bridge completion resistors if necessary, so that the pressure
gage is electrically equivalent to a load cell. The output voltage is specified as a millivolts
per volt of excitation for a full scale pressure differential across the diaphragm.
If the reference pressure is open to air, the gage compares the inlet pressure to the ambient
pressure. Ambient pressure is approximately 14.7 lbs/in
2
at sea level. If the gage is going to
be used to measure ambient pressure, it is necessary to use a sealed reference chamber with
either vacuum reference pressure (0 lbs/in
2
) or sea-level reference pressure (14.7 lbs/in
2
).
Temperature variations can affect the accuracy of strain diaphragm pressure gages. A
pressure gage with a sealed nonzero reference pressure exhibits temperature variations
in a manner consistent with the ideal gas law. Near room temperature, for example, a
5C change in the ambient temperature causes an error of 1.7% in the pressure mea-
surement. Temperature variations can also affect the performance of the strain gages
themselves. Accurate pressure measurements in environments with varying tempera-
tures require transducers that compensate for such effects.
All strain diaphragm pressure gages require a regulated excitation source. Some gages pro-
vide regulation internally, requiring the user to obtain an unregulated voltage from a power
supply. Some strain diaphragm pressure gages also employ internal signal conditioning,
which amplifies the millivolt signal from the Wheatstone bridge to a full scale voltage of 5V
to 10V; gages of this type have low-impedance output. In contrast, other pressure gages
have no internal signal conditioning, so their output impedance is equal to that of the
Wheatstone bridge (several KOhm for semiconductor strain gages), and their full scale volt-
age is millivolts. In short, one must carefully consider the kind of pressure gage to use and
be prepared to provide the necessary excitation and signal conditioning.
Common Parameters to Describe a Physical System. Two parameters used to de-
scribe a physical system are the Q factor and the natural resonant frequency W
o
.
The Q factor is the measure of a systems capacity to either absorb or deliver energy
per cycle of excitation. For applications that are lightly damped, such as a piezo-
electric accelerometer (discussed in detail later), the maximum response is very close
to W
o
and the amplification factor is Q.
48
When comparing the frequency response of different applications, the frequency axis is
often normalized with respect to the systems resonant frequency, W
o
. This normaliza-
tion facilitates qualitative comparisons and mathematical operations.
Strain-Gage Accelerometers. Strain-gage accelerometers use a test mass attached to a
cantilever. The strain gage measures the bend of the cantilever, which is a result of
acceleration. These accelerometers typically have natural frequencies of several kilo-
hertz. If a system has no additional damping, the cantilever-mass combination forms a
harmonic oscillator with a very large Q, near 10
5
.
A system with a high Q, excited near its resonant frequency will produce large oscil-
lations. This leads to inaccurate measurements, and may even result in the acceler-
ometer being ripped apart. Consequently, oil is sometimes added to increase the
damping. Figure 5.08 and figure 5.09 show the frequency response of harmonic
oscillators with Q =1 and Q =10
5
. For Q =10
5
, the accelerometer is usable with 10%
accuracy up to 1/3 of its resonant frequency. In contrast, for Q =1, the accelerom-
eter is usable with 10% accuracy to 1/2 of its resonant frequency.
Q=1
0.0 0.1 0.2 0.3 0.4 0.5
0.95
1.00
1.05
1.10
1.15
1.20
1.25
1.30
1.35
A
m
p
l
i
t
u
d
e
/
0
Q=10
5
Fi g. 5.08: Frequency response of typi cal
harmonic oscillator
Fi g. 5.09: Frequency response illustrated in
Fig. 5.08, magnified to show the low frequency
response
10
5
0.0 0.5 1.0 1.5 2.0
Q=1
A
m
p
l
i
t
u
d
e
/
0
10
4
10
3
10
2
10
1
10
0
10
-1
Q=10
5
Most strain-gage accelerometers have strain gages wired in a Wheatstone-bridge con-
figuration. Signal conditioning for a common strain-gage bridge would apply to this
type of transducer since it looks exactly like a load cell.
Piezoelectric Transducers
When subjected to a dynamic mechanical force, piezoelectric crystals generate an elec-
tric charge on opposite faces. A variety of transducer types use piezoelectric crystals to
convert mechanical quantities into electrical signals. Quartz is the most commonly
49
used material in Piezoelectric transducers (PZT). Quartz crystals that respond to either
compression or shear forces are readily available.
From an electrical point of view, a PZT looks like a capacitor with a time varying charge,
Q(t). This charge is proportional to the force on the crystal and is usually measured
using either voltage or charge amplification.
Voltage Amplifier. In Figure 5.10, the voltage amplification gain is 1. Adding feedback
resistors or increasing the number of amplification stages can result in different gains.
The amplifier converts the high-impedance voltage input to a low-impedance voltage
output. The voltage is Q*C, where Q is the charge generated across the PZT, and C is the
total capacitance over which this charge is distributed (including the PZT and cable
capacitance). Both the cable and PZT capacitance may be on the order of several hun-
dred picofarads (pF). As a result, changing the cable can change the PZTs sensitivity.
The best implementation of this configuration places the op amp close to the PZT.
Doing so minimizes cable capacitance and maximizes sensitivity. (Note: The length of
the cable beyond the amplifier does not affect sensitivity.)
Charge Amplification. The charge am-
plifier configuration shown in Figure
5.10 is extremely versatile because its
output voltage is V =Q/C. Q repre-
sents the charge generated by the PZT,
and C is the feedback capacitance. If
the op amp has a very large open-loop
gain, as most op amps do, the output
voltage will remain independent of the
cable capacitance. This allows the am-
plifier to be placed a moderate distance
(several meters) from the PZT. How-
ever, since noise increases with cable
length, cable length limits are set by
noise requirements as opposed to sen-
sitivity considerations.
The charge amplifier acts like a high-pass filter with lower corner frequency f =1/(2RC),
creating a trade-off between sensitivity and frequency response. Reducing C increases sen-
sitivity, but it also increases the lower corner frequency. (Note: In the charge amplifier
configuration, PZTs can only be used for dynamic measurements.) The leakage resistance
of most PZTs prevents static measurements, as most PZTs fall in the 10
+10
to 10
+12
Ohm
range. Such leakage resistance, combined with a capacitance of several hundred pF, gives a
time constant on the order of a few seconds.
Fig. 5.10: PZT voltage and charge amp
+
Voltage amplifier
C R PZT
+
Charge amplifier
R
PZT
C
50
Low-Impedance Transducers. Most
modern PZTs feature built-in signal-con-
ditioning circuitry. Such units have a
low-impedance output and require an
external power supply. Also known as
integrated circuit piezoelectric (ICP
)
transducers, their sensitivities are speci-
fied by the manufacturer, so the users
only concerns are interfacing the PZT to
standard voltage-measuring equipment,
and converting the data from voltage to
engineering units.
The most common application of low-
impedance PZT transducers is in accel-
erometers. Accelerometers are used to measure acceleration and mechanical vibration
in a variety of applications such as automobile safety systems, which use accelerom-
eters to deploy airbags; the space shuttle uses accelerometers to monitor lift-off accel-
eration and motion during orbit, and many industrial processes use accelerometers to
ensure vibration in machinery remains below acceptable levels.
All accelerometers use a mass to convert acceleration to force via the relationship F =ma.
This force is converted to an electric signal using a strain gage or PZT. The strain-gage
technique permits constant acceleration measurements, whereas PZT accelerometers do
not. These two common types of accelerometers resemble simple harmonic oscillators.
A low-impedance Piezoelectric accel-
erometer consists of a Piezoelectric
crystal and an electronic amplifier.
When stretched or compressed, the
crystal develops a charge variation
between its two surfaces that is re-
lated to the amount of stress, shock,
or vibration on the crystal. The am-
plifier transforms the sensors high
impedance to the output impedance
of a few hundred Ohms. Low-imped-
ance Pi ezoel ectri c transducers are
used to measure pressure and force,
as well as acceleration. The acceler-
ometer circuit requires only two wires
ICP is a registered trademark of PCB Piezotronics, Inc.
A
V
Fig. 5.11: Block diagram of a typical ICP
Fig. 5.12: ICP
Accelerometer
51
(coax or twisted pair) to transmit both power and signal. At low impedance, the
system is insensitive to externally induced or triboelectric cable noise. Sensitivity
is not affected by cable length.
Piezoelectric accelerometers are composed of a Piezoelectric crystal attached to a
seismic mass. Acceleration causes the seismic mass to exert a force on the PZT,
generating an electrical signal. The seismic mass is attached to a bolt that acts like
a stiff spring, giving the system a higher resonant frequency than Piezoresistive ac-
celerometers, typically 20 kHz. Piezoelectric accelerometers with resonant frequen-
cies as high as 120 kHz are available, giving them a usable frequency range from less
than 1 Hz to 40 kHz.
Like other PZTs, Piezoelectric accelerometers are available in both low-impedance versions,
which must be powered externally, and in high-impedance versions, which usually require
an external charge amplifier. A vibrating test fixture introduces yet another noise source for
high-impedance accelerometers. If the cable between the vibrating accelerometer and the
signal conditioning circuit moves, the shield may rub against the insulator in the cable,
generating a charge that produces additional noise. This is called the triboelectric effect.
Due to their high noise immunity and simple connectivity, low-impedance ICP acceler-
ometers are most commonly used today. These transducers need only a constant cur-
rent source for excitation.
Figure 5.13 shows a simplified accelerometer-to-signal-conditioning card connection. In
this case the signal conditioning card is IOtechs DBK4 dynamic signal-input card. The
voltage developed across R is applied to the gate of the metal-oxide-silicon-field-effect tran-
sistor (MOSFET). This MOSFET is powered from a constant current source of 2 or 4 mA.
Fig. 5.13: ICP
accelerometer attached to a typical dynamic signal-input card

Current
source
PGA
ICP Transducer
1 channel of a typical dynamic signal input card
Crystal R
Programmable
filter
Programmable
clock
Micro
controller
SSH Buffer
To data
acquisition
system
Control
lines
Filter
bypass
DC
coupled
AC coupled
2mA
4mA
52
The MOSFET circuit will bias off at approximately 12V in the quiescent state. As the system
is excited, voltage is developed across the crystal and applied to the gate of the MOSFET. This
voltage will cause linear variation in the impedance of the MOSFET, which will cause a pro-
portional change in bias voltage. This voltage change will be coupled to the DBK4 input
amplifier through the capacitor C. The low-frequency corner is controlled by the value of R
and the internal capacitance of the Piezoelectric crystal. Units weighing only a few hundred
grams can provide high level outputs up to 1 V/g with response to frequencies from 0.3 Hz to
2 kHz. Smaller units with less sensitivity respond to frequencies from 1 Hz to 35 kHz.
The constant current source provides a source-gate bias for the FET. As the gate current
of the FET changes as a result of pressure on the crystal, the voltage drop, V
ds
, changes
proportionately. This voltage is the output of the pre-amp. An AC coupling circuit, or
high-pass filter, is always necessary because there is high DC offset present at V
ds
which
is a result of the bias current. The cutoff frequency of the high-pass filter is dependent
on the application and the particular accelerometer. Figure 5.14 is a constant current
source schematic.
To eliminate the necessity of an out-board pre-amp, some accelerometer input options
have the current source and the AC coupling circuitry built-in, providing direct BNC
connection between the accelerometer and the data acquisition system.
+
Floating load
Grounded load
+
R
1
R
4
R
2
R
3
i
L
= If: Then: =
R
3
R
2
R
4
R
1
i
L
V
REF
R
2
V
REF
=
R
V
REF
i
L
i
L
+
V
REF
Load
L
o
a
d
Fig 5.14: Constant current source schematic
53
Most accelerometers provide an output that needs to be amplified and filtered prior to A/D
conversion. A programmable gain amplifier is best suited for accelerometer applications so
that the gain can be interactively adjusted by the operator for the optimal response.
Along with programmable amplification, programmable low-pass filters are highly rec-
ommended to reject unwanted high-frequency signals. These signals are typically due to
noise or higher frequency vibration components that are not germane to the applica-
tion. When developing the front-end circuitry for this type of measurement, noise rejec-
tion and bandwidth are of primary concern. As the bandwidth increases, the noise can
increase as well.
Low-pass filtering should be used in most accelerometer measurements to reduce noise
and aliasing effects. The cutoff frequency of the low-pass filter should be close to the
tests maximum useful operating frequency.
54
55
Chapter 6 General Amplification
Chapter 6
General Amplification
56
General Amplification
Data acquisition systems use many amplification circuits to accomplish their tasks. The
versatility of operational amplifiers makes them the universal analog building block for
signal conditioning. In this chapter, some uses of amplifiers in data acquisition equip-
ment are discussed.
Data Acquisition Front Ends
Data acquisition systems differ from single channel measurement instruments because
they are capable of measuring multiple channels. A voltmeter with a selector switch to
allow measurement of multiple voltages is a very simple data acquisition system. The
high degree of operator interaction required limits the performance of this simple sys-
tem, but the function is the same.
Figure 6.01 illustrates a simple data ac-
quisition system which consists of a
switching network (MUX), an instru-
mentation amplifier (IA) and an ana-
log-to-digital converter (ADC). These
three sections allow measurement of
multiple voltage channels using a
single A/D. The individual circuit
blocks each have unique capabilities
and limitations which together define
the performance of the system.
Fig. 6.01: Simple data acquisition system
Analog
inputs
IA
ADC
Mux
Digital data
Switching Networks
The switching network in an automated data acquisition system can be either relay or
solid-state, depending primarily on channel-to-channel switching speed and isolation
requirements. In high-speed systems, defined as those switching channels more than
200 times per second, solid state switching is used. Solid state switching networks are
configured with analog switches or multiplexers and are capable of switching channels
over 1,000,000 times per second. Some drawbacks to solid state switches are input volt-
age limitations, on-resistance, charge injection, and crosstalk.
Input Voltage Limitations. Input voltage limitations of analog switches are generally
defined by the connected power supply voltages, usually plus and minus 15V measured
with respect to analog common. There are also protected devices which can with-
57
stand somewhat higher voltages, up to 30-50 volts. The voltage limitations of analog
switches and multiplexers also depends on whether or not the devices are powered at
the time the input voltages are applied. As a general rule, it is better to have the analog
data acquisition system powered before connecting to live measurement voltages.
On-resistance. All switches have on-resistance. An ideal switch, which is closely
approximated by a mechanical switch or relay contact, has zero resistance when the
switch is closed. A reed relay contact can be less than 10 milli-ohms. A good analog
switch can have on-resistance of 10-100 Ohms, while an analog multiplexer can
have 100-2500 Ohms of on-resistance per channel. The on resistance adds directly
to the signal source impedance and can affect measurement accuracy.
Charge injection. Charge injection is an
undesirable characteristic of analog
switching devices, which couples a small
portion of the input gate drive voltage
to the analog input and output signals.
From the standpoint of the measure-
ment, this glitch in the output signal can
cause error and will be measured on the
input signal if the source impedance is
not low. There are circuit design tech-
niques which can be used to minimize
the effects of charge injection, but the
most effective method is to keep source
impedance as low as possible.
Channel-to-channel Crosstalk. Channel-to-channel crosstalk is another non-ideal charac-
teristic of analog switching networks, especially IC multiplexers. Crosstalk occurs if the
voltage applied to any one channel affects the accuracy of the reading from another chan-
nel. Conditions are especially optimum for crosstalk when signals of relatively high magni-
tude and high frequency are being measured, such as 4V to 5V signals connected to chan-
nel one while 100 mV signals are on an adjacent channel. Multiplexing at high frequencies
also exacerbates crosstalk because of capacitive coupling between switch channels. Again,
the best defense from crosstalk is keeping source impedances low.
Instrumentation Amplifiers
The block following the switching network in a data acquisition system is an amplifica-
tion block. This stage uses an amplifier configuration called an instrumentation ampli-
fier and has several important functions. These functions include rejection of common-
mode voltages, signal voltage amplification, minimizing the effect of multiplexer on
resistance, and properly driving the ADC input.
Fig. 6.02: Effect of charge injection on a mux output
Drive signal
Mux output
58
Common mode voltage is the voltage
measured relative to analog common
which is common to both input volt-
ages. The two input voltages could be
4.1V and 4.2V, for example; the com-
mon mode voltage is 4.1V and the dif-
ferential voltage between the two is
0.1V. Ideally the instrumentation am-
plifier (IA) ignores the common mode
voltage and amplifies the difference be-
tween the two inputs. The degree to
which the amplifier rejects common
mode voltages is given by a parameter
called common-mode rejection ratio
(CMRR). The CMRR is expressed in
decibels (dBs); a ratio of 10 is 20 dB and a ratio of 100 is 40 dB. Each successive ratio of
10 adds 20 dB. If an instrumentation amplifier with a CMMR of 80 dB is measuring a 20
mV signal in the presence of a 4V common mode voltage, the error due to the common
mode voltage would be 0.4 mV, which is still 2% of the signal. The ability of an IA to
reject high common mode voltages is sometimes misinterpreted as an ability to reject
high voltages. What is really being rejected is a voltage level having a high magnitude
relative to a small signal voltage, such as 4V relative to 20 mV.
Signal voltage ranges are frequently smaller than the range of the system ADC. For
example, a 0-100 mV signal is much smaller than the 0-5V range of a typical ADC.
To achieve the maximum practical resolution for the measurement, a gain of 50 is
needed. Instrumentation amplifiers are capable of gains in this range. In fact, in-
strumentation amplifiers are capable of gains ranging from unity to well over 10,000
in some specific applications. In multiplexed systems, the gains of IAs are usually
restricted to the range of 1-1,000.
There is a special class of instrumentation amplifiers with programmable gain which
can switch between fixed gain levels at sufficiently high speeds to allow different gains
for different input signals delivered by the input switching system. These amplifiers are
called programmable gain instrumentation amplifiers (PGIA). The same digital control
circuitry which selects the input channel can also select a gain range allowing a high
degree of flexibility in the system.
V
signal V
out
= A
v
V
signal
+
V
common-mode V
s
Fig 6.03: Typical instrumentation amplifier
59
The non-ideal on resistance of analog switches will add to the source impedance of any
signal source to create measurement error. This effect is minimized by the extremely
high input impedance of the IA. The input stage of an IA consists of two voltage follow-
ers, which have the highest input impedance of any common amplifier configuration.
The high impedance and extremely low bias current drawn from the input signal create
a minimal voltage drop across the analog switch sections resulting in a more accurate
signal reaching the input to the IA.
The output of the instrumentation amplifier has a low impedance which is ideal for
driving the ADC input. The typical ADC does not have high or constant input
impedance. It is important for the preceding stage to provide a signal with the
lowest impedance practical.
Instrumentation amplifiers have some limitations including offset voltage, gain error,
limited bandwidth, and settling time. The offset voltage and gain error can be cali-
brated out as part of the measurement, but the bandwidth and settling time are param-
eters which limit the frequencies of amplified signals and limit the frequency at which
the input switching system can switch channels between signals. A series of steady DC
voltages applied to an instrumentation amplifier in rapid succession creates a difficult
composite signal to amplify. The settling time of the amplifier is the time necessary for
the output to reach final amplitude to within some small error (often 0.01% ) after the
signal is applied to the input. In a system scanning inputs at 100 kHz, the total time
spent reading each channel is 10 S. If analog-to-digital conversion requires 8 S, set-
tling time of the input amplitude to the required accuracy must be less than 2 S.
Although it has been stated that offset voltage and gain error can be calibrated out, it is
important to understand the magnitude of the problem to allow good decisions as to
when these error corrections are necessary. An IA with an offset voltage of 0.5 mV and
a gain of X2 measuring a 2V signal will only incur an error of 1 mV in 4V on the output
which is 0.025%. An offset of 0.5 mV and a gain of X50 measuring a 100 mV signal will
incur an error of 25 mV in 5 volts or 0.5%. Gain error is similar. A stage gain error of
0.25% will have a greater overall effect as gain increases resulting in larger errors at
higher gains and minimal errors at unity gain. System software can generally handle
known calibration constants with mx +b routines but some measurements are not
critical enough to justify the effort.
60
Analog-to-Digital Converters
The analog to digital converter stage is the last link in the chain between the analog
domain and the digitized signal path. In any sampled data system, such as a multi-
plexed data acquisition system, there is a functional need for a sample-hold stage pre-
ceding the ADC. The ADC cannot correctly digitize a time varying voltage to the full
resolution of the ADC unless the voltage changes very slowly. Some ADCs have internal
sample-holds or use architecture which emulates the function of the sample-hold stage.
For the purpose of this discussion, it is assumed that the ADC block includes a suitable
sample-hold circuit to stabilize the input signal during the conversion period.
The primary parameters of ADCs in data acquisition systems are resolution and speed.
Data acquisition ADCs typically run at speeds from 20 kHz to 1 MHz and have resolu-
tions of 12- or 16-bits. ADCs have one of two types of input voltage range, unipolar or
bipolar. The unipolar type range typically runs between 0 volts and some positive or
negative voltage such as 5V. The bipolar type range will typically run from a negative
voltage to a positive voltage of the same magnitude. Many data acquisition systems
have the capability of reading bipolar or unipolar voltages to the full resolution of the
ADC, which necessitates some sort of level shifting stage to allow bipolar signals to use
unipolar ADC inputs and vice versa. A typical 12-bit, 100-kHz ADC has an input range
of -5V to +5V and a full scale count of 4,096. Zero volts corresponds to nominally 2,048
counts. If the range of 10 volts is divided by the number 4,096, the LSB (least significant
bit) magnitude of 2.44 mV is obtained. A 16-bit converter with the same range and
65,536 counts has an LSB value of 153 V.
Signal Conditioning Amplification
Many real world sensors have very low signal levels. These levels are too low for direct
application to multiplexed data acquisition system inputs with relatively low gains. Two
common examples are thermocouples and strain-gage bridges which both deliver full scale
outputs of less than 50 mV. There are two broad classes of signal conditioning circuits,
multiplexed and dedicated. In multiplexed conditioners, multiple input sources are se-
quentially switched through one amplifier signal path allowing for an economical cost per
channel with a somewhat reduced performance. The more expensive dedicated type have
amplifier stages for each signal path and multiplex the high-level signals only.
Most amplifier stages are applications of one of two basic configurations, inverting or non-
inverting. See figure 6.04 for an illustration of each. Each configuration has a simple
equation which provides the idealized circuit gain as a function of the input and feedback
resistors. There are also special cases of each configuration which make up the rest of the
fundamental building blocks, namely the unity gain follower and the difference amplifier.
61
Source Impedance &
Multiplexing
Multiplexing and a high source imped-
ance do not mix well. The reason that
low source impedance is always impor-
tant in a multiplexed system can be
explained with a simple RC circuit, see
figure 6.05. Multiplexers have a small
parasitic capacitance to analog com-
mon on all signal inputs as well as out-
puts. These small capacitance values,
when combined with source resistance
and fast sampling time, affect measure-
ment accuracy. A simple RC equiva-
lent circuit consists of a DC voltage source with a series resistance, a switch, and a
capacitor. When the switch closes at t =0, the voltage source charges the capacitor
through the resistance. When charging 100 pF through 10 kOhm, the RC time con-
stant is 1 S. In a 10 S time interval of which 2 S is available for settling time, the
capacitor will only charge to 86% of the value of the signal, introducing a major error.
If the same calculations are done with a 1,000 Ohm resistor, the capacitor will easily
charge to an accurate value in 20 time constants.
Voltage Measurement Errors
One must consider several factors when making voltage measurements. This section will
discuss a few of these, including: input and source impedance, differential voltage mea-
surements, and isolation considerations.
+
Inverting Non-Inverting
R
2
R
1
Gain = -
R
2
R
1
Gain =
R
2
R
1
+1
+
R
2
R
1
Fig. 6.04: Inverting and non-inverting amp
Fig. 6.05: Simple RC circuit
R
source
R
on AMP
+
MUX
R
62
Input and Source Impedances
Figure 6.06 depicts how system input
impedance and the transducers source
impedance combine to form a voltage
divider, which reduces the voltage be-
ing read by the ADC. The input im-
pedance of most input channels is 1M
Ohm or greater, so it is usually not a
problem if the source impedance is
low. However, some transducers (pi-
ezoelectric transducers, for example)
have high source impedances and
should be used with a special charge-
sensitive amplifier. In addition, mul-
tiplexing can greatly reduce a data ac-
quisition systems effective input im-
pedance. See Chapter 3 for detailed
discussion.
Attenuators & Multiplexing
There is a basic voltage range of 5V to
10V associated with most data acqui-
sition inputs. Voltages over 10V are
generally not readable without attenu-
ation. Simple resistive dividers can eas-
ily attenuate a voltage, which is too
high for a 5V input, but there are two
drawbacks complicating this simple
scheme. Voltage dividers have a sub-
stantially lower impedance to the
source than direct analog inputs, and,
they have very high output impedance
to the multiplexer input. Consider this
example, a 10:1 divider reading 50V.
If a 900 KOhm and 100 KOhm resistor are chosen, to provide a reasonable 1 MOhm
load to the source, the impedance seen by the analog input is a disagreeable 90.0 KOhm
which will never allow an accurate multiplexed reading. If the values are both down-
sized by a factor of 100 to allow the output impedance to be lower than 1,000 Ohms,
the input impedance seen by the measured source is 10 KOhm. Most voltage measure-
ment applications could not tolerate a 2K Ohm/volt instrument. Hence the conclu-
sion: simple attenuation is not practical with multiplexed inputs.
Fig. 6.06: Input and source impedances
R
s
R
i
Transducer
V
sig
V
ADC
= V
sig
R
i
R
s
+ R
i
Fig. 6.07: Buffered attenuator
V
IN
+
R
A
R
B
V
OUT
V
OUT
=
R
B
R
A
+ R
B
V
IN
63
Attenuators & Buffers
The answer to the voltage divider prob-
lem is unity gain buffering of the volt-
age divider output. A dedicated unity
gain buffer does not have a source im-
pedance problem with 90 KOhms. The
output impedance is very low to the
multiplexed analog input as well.
Low-Pass Filters
Low-pass filters attenuate higher fre-
quencies to varying degrees depend-
ing on the number of stages and the
magni tude of the hi gh frequency
relative to the corner frequency. There is no need for high bandwidth in an ampli-
fier stage if the signal of interest is of a much lower frequency. In fact it is a basic
intent of this design to eliminate excessive bandwidth in all circuits in order to
limit noise. One of the major benefits of individual signal conditioning stages for
low-level sensors, as opposed to multiplexed stages, is the opportunity to include
low-pass filtering on a per-channel basis in the signal path. In a multiplexed circuit (an
amplifier which is being shared by
multiple low level DC signals), the
main signal path generally cannot low-
pass filter due to the fast settling time
necessary in multiplexed systems.
The best place for low pass filtering is
in the individual signal path prior to
buffering and multiplexing. For very
small signals, amplifying with an in-
strumentation amplifier prior to filter-
ing allows an active low pass filter to
operate at levels of optimum signal-to-
noise ratio. Figure 6.10 on the follow-
ing page illustrates a typical amplifier-
filter-mux configuration.
Fig. 6.08: Filtered/buffered data acquisition system
LPF IA Buffer
LPF IA Buffer
Mux To ADC
Inputs
Fig. 6.09: Simple RC filter with buffer
+
R
C
To Mux
64
Measuring High Voltages
It was shown earlier that measurement of voltages above 5-10V with data acquisition
inputs requires attenuation and that attenuation requires buffering. A typical data ac-
quisition system would require a signal conditioning stage, which provides attenuation
and buffering, in order to accommodate high-voltage inputs.
For higher voltage measurements a fully differential attenuator, calibrated to match the
buffered differential inputs of a data acquisition system would be required. With this type
of setup, differential voltages as high as 2000 volts peak-to-peak are measurable. Figure 6.11
illustrates this type of system.
For systems with a true differential in-
put, neither the inputs low or high
end are connected to ground. In this
case, attenuating an incoming signal
requires two matched resistor dividers;
one on signal-high and one on signal-
low. The low end of each divider is
grounded. For example, to provide
Fig. 6.11: Typical data acquisition system with a high-
voltage adapter
C
A
B
5 M 50 k
10 M
5 M 50 k
10 M
High-voltage adapter Data acquisition system
Channel 1
Ch 1
Channel address lines
Switched bias resistors
(per channel)
Ch 2
Ch 3
Ch 4
IA
IA
IA
IA
Low-pass filter
per channel
Low-pass filter
per channel
Low-pass filter
per channel
Low-pass filter
per channel
Mux
2
To A/D
Fig. 6.10: Typical amplifier-filter-mux configuration
65
200:1 attenuation without compromis-
ing the high input impedance, an ac-
cessory device is required. In the case
of our typical system, the two matched
buffer amplifiers (2 op amps in the
same package), A and B, on the front
end of the data acquisition system
draw nearly identical bias current
through their inputs. If matched volt-
age dividers are placed on the high and
low signal inputs, the bias current
through the dividers will cause an iden-
tical voltage offset on the output of the
op amps A and B. Since these two out-
puts are identical, they can be thought
of as a common mode voltage that the
differential amplifier, C, rejects. See
Figure 6.11.
This scheme attenuates both the high and low signal, allowing both ends to be hun-
dreds of volts off ground. For example, the high input can be 110 VAC while the low
end has a 50 VDC offset. After attenuation, the peak voltage of the high signal is 0.77V
and the low signal is 0.25V, which is well within the readable range of the data acquisi-
tion systems differential inputs.
If a single ended measurement is attempted by attenuating only the high signal and
pulling the low signal to ground, two problems are encountered. First, if the data ac-
quisition system ground and the signal ground are dissimilar, tying them together cre-
ates a ground loop and produces error in the reading. The second error component
results from the bias currents i
b
flowing through the two op amps, A and B. The bias
current produces a voltage drop through the divider on the high input, but there is no
matching drop through the low input. The difference is presented to the differential
amplifier, C, producing an offset on its output. See Figure 6.12.
C
A
B
Data acquisition system
i
Total
i
a
i
b
i
b
Fig. 6.12: Single ended attenuation of a differential input
66
Single-Ended & Differential Measurements
Data acquisition systems have provisions for single-ended and differential input con-
nections. The essential difference between the two is the choice of the analog common
connection. Single-ended multichannel measurements require that all voltages be ref-
erenced to the same common node, which will result in measurement errors unless the
common point is very carefully chosen; sometimes there is no acceptable common
point. Differential connections cancel or ignore common mode voltages and allow
measurement of the difference between the two connected points. When given the
choice, a differential measurement is always better. The rejected common mode volt-
ages can be steady DC levels or noise spikes.
The best reason to choose single-ended
measurements is for a higher channel
count that is available in some devices.
Most data acquisition products allow
you to double the number of channels
in a differential system by selecting
single-ended operation.
There is a widely misunderstood aspect
of differential measurements of float-
ing sources. A common example is a
flashlight battery connected to a dif-
ferential input pair. The measurement
cannot be made unless an additional
connection is made to the analog com-
mon of the analog input device. This
connection is frequently made with a
10K-100K resistor. The resistor provides a bias current return path for the two inputs
which are each connected to high impedance voltage follower stages on the input of an
instrumentation amplifier. If this bias current return path is not established, the ampli-
fier will saturate and cannot measure the voltage difference between the inputs. Some
data acquisition inputs provide switched resistors to analog common for this purpose.
The reason these bias resistors are switched is that there are situations in which they are
not needed. The situation in which the bias resistors are not needed is where a DC path
already exists for bias current to flow from the amplifier inputs to analog common.
This usually occurs when the device under test is already referenced to analog common.
Noise Reduction in Differential Mode
S
1
S
2
S
1
-S
2
+
Diff
Fig. 6.13: Voltages common to both lines are rejected by
the differential amplifier
67
Differential Voltage Measurements
In a single-ended voltage measurement, the voltage on the high input lead is measured
with respect to analog common. Differential measurement utilizes a differential amplifier
to measure the difference in voltage between the two signal leads. Differential measure-
ments provide greater noise rejection;
furthermore, some measurements, most
notably strain-gage measurements, re-
quire measurement of the voltage dif-
ference between two signal leads.
Figure 6.14 depicts a differential am-
plifier configured for a thermocouple
measurement. Note the 10 KOhm re-
si stor between the l ow si de and
ground. Even though the amplifier
measures the voltage difference be-
tween the two i nputs, a path to
ground is required on at least one in-
put of the amplifier. This resistor pro-
vides the path for the amplifier bias
currents. If not present or too high
in value, the amplifier bias currents will cause the inputs to approach one of the
power supply rails and exceed the common mode range of the amplifier. In most
applications, the common mode range is typically 10V. The common-mode volt-
age produced by the bias currents is V
cm
=2i
b
R
b
To determine the minimum value for R in the case of a completely floating input, R
may be a short circuit on one side only. Otherwise, under the worst case conditions,
the vol tage error i n percent of vol tage appl i ed due to source resi stance and
R is: E(%)=100 R
S
/2R
b
. This applies to R for the balanced configurations. Typical
values for R are 10 KOhm to 10 MOhm. If both equations cannot be satisfied, the
source resistance is too high to operate within specifications.
Some amplifier inputs have built-in termination resistors for differential measurements,
but very high impedance inputs require the user to supply a termination resistor. Introduc-
ing a termination resistor into the configuration dramatically lowers the input impedance
in non-isolated systems and may load the circuits being measured.
Fig. 6.14: Differential thermocouple measurement
Thermocouple
R
b
=10k
Differential
amplifier
68
Consider the strain gage in Figure 6.15.
If the excitation voltage is referenced
to the same common as the differen-
tial amplifier, termination resistors are
not required because the strain gage
itself provides a current path to ground
for the instrumentation amplifier.
However, if the excitation voltage is
floating or referenced to a separate
common, at least one termination re-
sistor is required. A balanced termi-
nation with 100 KOhm resistance can
also be used.
Common mode rejection ratio (CMRR)
is an important specification for in-
struments making differential measurements. A large CMRR is necessary to accurately
measure a small differential signal on a large common mode voltage. The CMRR speci-
fies the maximum effect of the common mode voltage on a differential measurement.
An amplifiers CMRR is typically specified in decibels (dB); a ratio of 10 results in 20 dB,
and each additional factor of 10 results in an additional 20 dB. For example, 100 dB
corresponds to 10
5
, so an instrument with a CMRR of 100 dB measuring a signal on a
common mode voltage of 1V can have an error as large as 0.01 mV. If the common
mode voltage is larger, or if greater accuracy is desired, an amplifier with a CMRR of at
least 120 dB would be required. See Chapter 5 for detailed discussion.
Measuring High DC
Currents with Shunts
It is frequently necessary to measure
high currents with data acquisition in-
struments. Generally, DC currents are
measured by monitoring the voltage
drop across a calibrated shunt which
has a 50 mV or 100 mV drop at full
rated current. Figure 6.16 shows a dif-
ferential measurement of three shunts
which have been strategically placed
in the common sides of three motor
Fig. 6.15: Differential strain gage measurement with
switchable termination resistors
100k
100k
+
V
exc
Excitation
Fig. 6.16: Ground referenced differential measurement
CH2 L
Analog
Common
Mux Board
DBK12,13, or 14
CH2 H
CH1 L
CH1 H
CH0 L
CH0 H
Motor
Loads
S
u
p
p
l
y
Shunts
Good for a low
signal reference
Bad for a low
signal reference
69
circuits to allow inexpensive monitoring with a non-isolated, low voltage input. It is
not always possible to locate the shunts in higher voltage circuits in the electrically
convenient common leg. In these instances, isolated analog inputs are the recom-
mended solution.
Measuring AC Currents
with Current Transformers
High AC currents can be measured
with shunts also, but the inherent
danger of direct connection to AC
lines is undesirable. Isolation from
the AC line voltages is provided by
current transformers (CT) along with
a ratio reduction of the input current.
A 500:5 CT for example, has a 100:1
ratio and will have a secondary cur-
rent of 5 amperes when 500 amperes
is flowing in the primary. If a low
value load resistor such as 0.01 Ohm
is connected to the output of a cur-
rent transformer, the full load secondary current will produce a 50 mVACrms, which
can be easily read with an analog input. Although the output voltage seems low,
higher resistance values will overload the CT and reduce the accuracy of the mea-
surement. If a CT has a VA rating of 2 VA, the maximum usable load resistance
including lead wire is only 0.08 Ohms. It is also necessary to remember that an
open-circuited CT with primary current flowing can generate over 1,000 volts at
lethal current levels.
Signal Connection Methods
Measurement of low level signals depends heavily on proper wiring between the signal
sources and the data acquisition device. The amplifier stages in signal conditioning
equipment are not able to determine the difference between the signal and inadvert-
ently added noise voltage due to improper wiring practices. Whenever low signal levels
(less than 1V) are being measured, twisted pairs or shielded twisted pairs provide the best
protection against unintentional pickup of noise voltages. If a shield is used, it should
only be grounded at one end, preferably at the signal source.
V
out
= 50mV
rms
0.01
I

= 500A
rms
C.T.
ratio
100:1
Fig. 6.17: Measuring high AC currents with current
transformers
70
71
Chapter 7 Noise Reduction & Isolation
Chapter 7
Noise Reduction & Isolation
72
Noise Reduction
Controlling electrical noise is imperative because it can present problems even with the
best measurement equipment. Most laboratory and industrial environments suffer from
multiple sources of electrical noise. For example, AC power lines, heavy equipment (par-
ticularly if turned on and off frequently), local radio stations, and electronic equipment can
create noise in a multitude of frequency ranges. Local radio stations are a source of high-
frequency noise, while computers and other electronic equipment can create noise in all
frequency ranges. Creating a completely noise-free environment for test and measurement
is seldom practical. Fortunately, simple techniques such as using shielded/twisted pair
wires, averaging, filtering, and differential voltage measurement are available for control-
ling the noise in our measurements. Some techniques prevent noise from entering the
system; other techniques remove noise from the signal.
Shielding. While many techniques for
controlling noise in signals provide a
means of removing noise that is already
present, the preferred solution is simply
preventing the occurrence of noise in the
signal in the first place. This is accom-
plished through careful shielding of
equipment and test leads.
For example, loose wires, which are ef-
fective antennae for radio frequency
pickup, and can form loops for induc-
tive pickup, are a source of noise; using
shielded cables is a simple solution.
To illustrate the need for controlling
noise, Figure 7.01 shows a single-ended
voltage measurement on a shorted chan-
nel. Approximately 6 feet of wire, which
was not shielded or twisted pair, was at-
tached to a data acquisition system.
Figure 7.02 shows the noise in the
measurement of a si ngl e-ended
shorted channel, where shielded cable
is the noise reduction technique used.
The improvement over unshielded
wires is readily apparent.
Time [seconds]
V
o
l
t
s
0 0.05 0.1 0.15 0.2 0.25
-5.00E-03
-4.00E-03
-3.00E-03
-2.00E-03
-1.00E-03
0.00E+00
1.00E-03
2.00E-03
3.00E-03
4.00E-03
5.00E-03
Fig. 7.02: Single-ended measurement with shielded wire
Time [seconds]
V
o
l
t
s
0 0.05 0.1 0.15 0.2 0.25
-5.00E-03
-4.00E-03
-3.00E-03
-2.00E-03
-1.00E-03
0.00E+00
1.00E-03
2.00E-03
3.00E-03
4.00E-03
5.00E-03
Fig. 7.01: Single-ended measurement with loosewires
73
Averaging. Averaging is done in soft-
ware after several samples have been
collected. Depending on the nature
of the noise and the technique used,
averaging can reduce noise by the
square root of the number of averaged
samples, however, reducing RMS noise
significantly may require averaging
many samples. Figure 7.03 demon-
strates the voltage across the shorted
channel when 16 samples of data are
averaged together. Averaging is best
suited to lower-speed applications.
Although averaging is an effective
technique, it suffers from several drawbacks. Noise present in measurements only de-
creases as the square root of the number of measurements. Therefore, in the above
example, reducing the RMS noise to a single count by averaging alone would require
3,500 samples. As such, averaging is only suited to low-speed applications. Also, aver-
aging only eliminates random noise. It does not necessarily eliminate many types of
system noisesuch as periodic noise from switching power supplies.
Analog Filtering. A filter is an analog-circuit element that attenuates an incoming sig-
nal according to its frequency. Depending on whether or not the filter is a low- or high-
pass filter determines if the frequencies are attenuated above or below the cutoff fre-
quency. For example, as a signals frequency increases beyond the cutoff point, the
attenuation of a single-pole, low-pass
filter increases slowly. Multi-pole fil-
ters provide greater attenuation be-
yond the cutoff frequency but may
introduce phasing problems that could
affect some applications. Filter circuits
can be passive or active.
Passive. A passive filter is a circuit or
device, consisting entirely of non-am-
plifying components, typically induc-
tors and capacitors which pass one fre-
quency band while rejecting others.
Raw Signal Averaged 16 values
-5.00E-03
-4.00E-03
-3.00E-03
-2.00E-03
-1.00E-03
0.00E+00
1.00E-03
2.00E-03
3.00E-03
4.00E-03
5.00E-03
Fig. 7.03: Noise reduction using averaging
V
IN
R
V
OUT
F
-3dB
=
1
2RC
C
Fi g. 7.04: Single-pole RC filter with the formula
for -3dB frequency
74
Active. An active filter is a circuit or
device, made up of amplifying com-
ponents and sui tabl e tuni ng el e-
ments, typically resistors and capaci-
tors, which pass one frequency band
while rejecting others. Figure 7.05
compares the amplitude of a single-
pole, low-pass filter, with a three-pole
filter. Both filters are set for a 1-kHz
cutoff frequency. The three-pole fil-
ter has a much greater attenuation for
frequencies exceeding the cutoff fre-
quency. The improvement in signal
quality provided by low-pass filtering
is demonstrated in Figure 7.06, in
which a signal containing wideband
noise is passed through a three-pole filter configured for a 1-kHz cutoff frequency.
The deviation from the average signal is plotted in volts. The maximum deviation
is 6 counts, and the RMS noise is 2.1 counts.
The three-pole filter used in our ex-
ample is an active filter input with
changeable filter configurations. The
active 3-pole filter can be configured
as a Butterworth, Bessel, or Chebyshev
filter with corner frequencies up to 50
kHz. Filter properties depend on the
values of the resistors and capacitors,
which can be changed by the user. Fil-
tering can also be done with switched-
capacitor filters. This type of filter re-
quires a clock signal to set the cutoff
frequency. The primary advantage to
this type of filtering is the ease of cut-
off frequency programmability.
1-pole
3-pole
(DBK18)
0
-10
-20
-30
-40
-50
-60
-70
-80
-90
-100
-110
-120
0.1 1 10 100
F0.001
FREQUENCY IN kHz
A
M
P
L
I
T
U
D
E

I
N

d
B
V
Fig. 7.05: Amplitude vs. frequency for a typical single-
pole low-pass filter vs a three-pole filter
Raw Data DBK18 Output
-0.005
-0.004
-0.003
-0.002
-0.001
0
0.001
0.002
0.003
0.004
0.005
Fig. 7.06: Noise reduction using a typical low-pass
filter card
75
Differential Voltage Measurement. Making differential voltage measurements is an-
other means of reducing noise in analog input signals. This technique works because
often most of the noise on the high analog input lead closely approximates the noise
on the low lead. This is called common mode noise. Measuring the voltage difference
between the two leads eliminates common mode noise.
The improvement obtainable with
differential voltage measurement is
i l l ustrated i n Fi gure 7.07, whi ch
shows the same signal as Figure 7.01
using a differential input rather than
single-ended.
Isolation
Isolation can be defined as the separa-
tion of one signal from another to pre-
vent unintentional interaction be-
tween the two signals. There is a de-
gree of channel-to-channel isolation in
any multiplexed data acquisition sys-
tem; relay-based systems have galvanic
isolation and solid-state systems do
not. Galvanic isolation is the absence of any direct current (DC) path. Most isolation
methods eliminate any DC path below 100 MOhm. Three benefits of galvanic isolation
are circuit protection, noise reduction, and rejection of high common-mode voltages,
especially those created by ground loops.
Circuit Protection. Isolation separates the signal source from measurement circuitry
that may be damaged by the signal. Voltages higher than about 10V can distort data or
damage components used in data acquisition systems. High-voltage signals or signals
with high-voltage spikes should therefore be isolated. The protection can also work in
the other directionto safeguard a sensitive signal conditioner from a device failing
elsewhere in the system.
Noise Reduction. Isolation eliminates ground loops for high-gain systems and multi-
unit systems that are grounded together. The chassis for each device can reside at a
ground potential slightly different from that of the other devices. These potential dif-
ferences cause the flow of ground currents that result in measurement errors.
Time [seconds]
V
o
l
t
s
0 0.05 0.1 0.15 0.2 0.25
-5.00E-03
-4.00E-03
-3.00E-03
-2.00E-03
-1.00E-03
0.00E+00
1.00E-03
2.00E-03
3.00E-03
4.00E-03
5.00E-03
Fig. 7.07: The same signal as Figure 7.01 using a
differential input
76
Rejection of High Common-Mode Voltage. Common-mode voltage, relative to earth
ground, is present in most measurement situations. Two voltage measuring points which
are relatively close together in potential, may share a very high common-mode voltage.
In order to make a safe and accurate measurement in this situation, all data acquisition
systems must use some method of rejecting common-mode voltage. The instrumenta-
tion amplifier in a simple MUX-IA-ADC data acquisition system is only capable of reject-
ing about 10V of common-mode voltage. Common-mode voltages higher than 10V
require the use of more complicated and expensive rejection methods, such as isolation
amplifiers, which employ magnetic, optical, or capacitive isolation.
Methods of Isolation
There are many methods to achieve isolation, with the three most common being mag-
netic, optical and capacitive.
Magnetic Isolation. Magnetic isolation
using a transformer, is the most com-
mon method to isolate analog signals,
and is also capable of isolating digital
signals, although not as common.
The power supplies in most systems
use transformers to isolate the AC
line voltage from the DC voltages
generated to power the data acquisi-
tion system. A transformer provides
isolation by magnetically coupling
the AC input power to the data ac-
quisition system, thereby providing
no DC path for current to flow.
Transformer isolation is also commonly used in input amplifiers to provide isolation
from the input signal to the measurement circuitry. In this application, the input signal
is first converted to an AC signal which can be fed to the primary of the transformer.
The AC signal is either frequency, pulse width, or amplitude modulated, and then mag-
netically coupled to the secondary winding. The output is then demodulated by the
receiving circuitry before being presented to the measurement circuitry.
Finally, transformers can also be used to isolate digital signals. This method is increas-
ingly rare as the cost and performance of optical isolators improves (see next discussion).
Fig. 7.08: Block diagram of a typical isolation scheme
T/Cs
Control
signals
Digital section
Isolated
power
Analog section
A/D
Opto Opto's
10
AC
line
.1
77
Fig. 7.09: Diagram of DC-to-DC converter
+
Opto
Comparator
Sawtooth
V
IN
C
1
C
2
D
1
R
1
R
2
V
ref
Load
+
V
OUT
Optical Isolation. Optical isolation is most commonly used to transmit digital signals.
A light-emitting diode within an opto-isolator generates light when current flows through
the diode. A receiving transistor, which is light sensitive, will enable a current to flow
when it detects light. Since the only connection between the input and output is a
path for light to pass, then the device can provide isolation of several thousand volts
from input to output.
Opto-isolators are very commonly used to isolate the output of an A/D converter.
The A/Ds output is usually serial (or parallel which has been converted to serial) to
minimize the number of optos required in a system (parallel to serial conversion is
less expensive than having one opto for each bit of output from the A/D). In these
instances, the power supply to the A/D and associated input circuitry is also iso-
lated, usually via a transformer as described above.
Capacitive Isolation. A capacitor is a passive electronic device that permits AC cur-
rent to flow through it, but will not pass a DC current. As a result, it can be an
inexpensive and simple device for isolation. In order to be used, the signal which is
to be isolated must first be converted to an AC signal through some modulation
technique, and then transmitted through the capacitor to the receiving side. At the
receiving side, the AC signal must again be demodulated in order to recreate the
signal to be measured. This technique is applied within low cost isolation amplifi-
ers in which the coupling capacitor is formed by a common layer between two iso-
lated IC substrate sections. Signal isolation equipment using these specialized ICs
can achieve isolation ratings as high as 1500 VAC.
The main benefits of this approach
are simplicity, low cost and band-
widths as high as 50 kHz. Figure 7.09
illustrates a DC/DC converter often
used as a component subsection in
an isolation amplifier.
78
Fig. 7.11: Analog input block diagram
Chan 1
ON
Diff
AMP
24 Channels
To grounded
digital logic
Isolated
power
conditioning
AC
line
Floating Analog Section
Auto
zero
A/D
Digital
optical
isolation
V
1
V
2
Figure 7.10 illustrates a typical multi-
channel programmable isolation am-
plifier, in a data acquisition system that
uses three types of isolation: trans-
former, optical, and capacitive. A trans-
former-based DC-to-DC converter is
used to supply power to the isolated
side. A capacitively coupled device is
used to isolate the analog signal, while
two opto-couplers are used to transmit
the digital control signals to the float-
ing circuitry.
I Otechs ChartScan/ 1400 and
MultiScan/1200 high-speed isolated
temperature and voltage measurement
instruments, which provide a comprehensive isolation scheme, use a combination of
relays and transformers to provide 200V of isolation on a per channel basis. See Figure
7.11. Adjacent thermocouple channels can have signals that are dissimilar by 200V
and still read temperatures accurately.
To provi de thi s capabi l i ty, the
MultiScan/1200s or ChartScans ana-
log front end, including its A/D con-
verter, is isolated from system ground
vi a the AC power suppl ys trans-
former. The digital information to
and from the analog section is trans-
mi tted vi a opto-coupl ers. Each
MultiScan/1200 analog input card
has a pair of relays on each channel.
At any single moment, only one in-
put channel is attached to the ana-
log path. As each channel is switched
in, the entire analog front end floats
to the ground reference of the newly
connected channel.
Fi g. 7.10: Multichannel programmable isolated
amplifier block diagram
Programmable
attenuator
amplifier
Configuration
P
M
U
X
Hi
Lo
Isolation
amplifier
DC/DC
Converter
Opto couplers
LPF
Typical of each channel
To A/D
Barrier
Bypass
79 79
Chapter 8 Digital & Pulse-Train Signal Conditioning
Chapter 8
Digital & Pulse-Train Signal Conditioning
80
80
Digital Signals
Many applications entail interfacing digital signals to a computer. Digital signals differ
from analog signals in that digital signals represent a discrete state, whereas analog
signals represent continuous values. Digital signals are termed discrete because a digital
line is either low or high. As such, a digital signal represents Boolean values such as off
or on, zero or one. Accordingly, each digital line can serve as a single bit in the repre-
sentation of a number or character. Digital output lines commonly control relays to
power-on or power-off equipment or indicators. Digital input lines can represent the
state of a switch indicating position or status. Digital I/O lines may also be used to
communicate between instruments.
Relays. A relay is an electromechanical switch. Most relays open and close a switch
using a magnetic field produced when current is passed through a coil. Consequently,
the relay behaves like an isolator in relation to the control current. A relay will be
either open or closed when control current is not flowing. Most relays specify the
control current and voltage required to operate the switch, as well as the current and
voltage ratings for the power transmitted through the switch.
Relays are typically used to break or make circuit paths that carry high voltage or
current. A relatively small current supplied to the magnetic coil can throw a switch
capable of carrying hundreds of amps.
Relays are commonly controlled by
a digital output from a computer or
other instrument. For example, to
control a heater, an i nstrument
monitoring temperature can use its
digital output to actuate a high-volt-
age relay attached to the heater coil.
Fig. 8.01: Relay turning power on and off to an
inductive load
Relay
Relay
control
input
Inductive
load
Anti-
kickback
diode
i
+
i
V
2
81 81
When an inductive load is being switched, a diode across the load provides a path
for the current spike directly after the switch is opened while the inductors mag-
netic field is collapsing. Without the diode, arcing will occur at the relays contacts,
potentially decreasing the life of the relay. See Figure 8.01.
Relays can also provide isolation in a data acquisition system. For applications where
each channel must be allowed to float, or where individual channels have high com-
mon mode voltages, relay isolation is an economical and effective means of isolation.
Because they are an electromechanical device, relay contacts can bounce as they are
closed. As the coil is energized, the contacts close abruptly, and for a brief moment, the
contacts can rattle into place causing momentary interruptions in the current. Bounce
is typically not a problem when switching loads on and off. Data acquisition devices
utilizing relays generally provide debounce time to ensure that the relay has fully settled
before initiating a reading.
High Current/ Voltage Digital I/ O. Digital circuits are not limited to low-level TTL and
CMOS systems. For some applications, the high level may typically be 30V and the
source current may commonly be 100 mA.
I n applications where a relay coil
must be energized, a TTL or CMOS
output is rarely adequate. To provide
the necessary current and voltage, a
buffer stage must be added between
the TTL signal and the relay coil. One
example of this type of system is
IOtechs Digital488/80A digital I/O
instrument. This type of instrument
can be used to switch a 24V relay on
and off as long as an amplifier stage
i s provi ded. I n the case of the
Digital488/80A, the amplifier/attenu-
ator, consisting of a PNP transistor, a
flyback diode, and a resistor, is pro-
vided by an option card, which can
be added internally to the unit.
Fig. 8.02: This illustrates the optional amplifier module
that installs inside the Digital488/80A to accommodate
high voltages and/or current
Digital488/80A
Optional amplifier module
Digital488/80A
82
82
To energize a 24V solenoid or relay us-
ing this type of instrument, an exter-
nal 24V supply is attached to the cir-
cuit. As the internal TTL output goes
high, the transistor is biased and the
output goes low (about 0.7V). When
the TTL output is low, the transistor
does not conduct and the output goes
to 24V. Since the relay coil represents
an inductive load, the flyback diode
should be attached so that the circuit
is not damaged during switching.
Using the IOtechs Digital488/80A as an
example, Figure 8.04 shows high-volt-
age digital inputs via the HVCX1 attenu-
ator circuit. This allows voltages up to
48V to be read by the TTL circuitry.
The high-voltage signal goes into a re-
sistor voltage divider, which acts as a
signal attenuator. The resistance value,
R, provides a means for selecting the
Boolean high voltage level. The
table below shows the resistor values
for frequently used voltage levels.
Fig. 8.03: A typical digital-output signal amplifier similar
to that used in IOtechs HVCX1
Flyback pull-up
voltage applied
here
Output
Removable
27 K
pull-up
resistor
200 mA
(max)
TTL signal from
Digital488/80A
Flyback
diode
Return
+V
Relay
Coil
Power
Source
L/V
Typical Voltage Levels & Associated Resistor Values
Inputs Resistor Value
0-5V 10
0-12V 20K
0-24V 56K
0-48V 120K
Fig. 8.04: Schematic of typical attenuator input, similar
to that used in IOtechs HVCX1
Resistor in
DIP package
R
User
input
15k Ohms
To Digital488/80A
TTL input
83 83
Speed, Timing, and Scanning of Digital Inputs
There are many ways to interface digital inputs to a computer, ranging from simple to
complex. In this section we will briefly discuss software-triggered single-byte readings,
hardware-paced digital-input readings, and externally triggered digital-input readings.
Asynchronous Digital-Input Readings. In cases where the computer needs to periodi-
cally sample a digital byte or a group of bits, a software-triggered, asynchronous read is
required. Sometimes, though, the speed and timing of the digital-input readings are
important. The time between readings is likely to vary when using the software-trig-
gered single-byte method, particularly in applications implemented under a multitasking
operating system. This is because the time between readings depends on the speed of
the computer and other tasks that must be performed concurrently. Variations in time
between readings can be partially compensated for by the use of software timers, but it is
difficult to obtain a timing resolution of better than 1 ms (millisecond) on a PC.
Synchronous Digital-Input Readings. Some systems, such as IOtechs DaqBook
, LogBook
,
WaveBook
, and Personal Daq
portable data acquisition systems, offer hardware-paced,

digital-input readings. In such systems, the digital input port can be read at a user-set
frequency. For example, the DaqBook and LogBook can read their 16-bit port at up to 100
kHz, while the WaveBook operates at 1MHz. The greatest advantage of hardware-paced
digital-input readings is that they can be implemented far more quickly than can software-
triggered readings. Finally, devices such as these are able to place the digital-input port
reading into the scan group of their analog readings, providing a close correlation between
analog and digital input readings.
Externally Triggered Digital-Input Readings. Some applications use external devices that
provide digital data at a pace independent of the data acquisition system. These devices
provide a digital bit, byte, or word that needs to be read by the data acquisition system. It
is often necessary to take readings only when new data is available rather than at a predeter-
mined interval. Because of this, such external devices typically have handshaking ability.
As new digital information becomes available, the external device issues a digital transition
on a separate line such as an External Data Ready or Strobe input. To interface with a device
like this, the data acquisition system must provide an input latch, which can be controlled
by this separate external signal. Furthermore, logic must be supplied to the controlling
computer, alerting it to the fact that new data is ready to be received from the latch.
One example of a device that operates in this fashion is IOtechs Digital488 series of inter-
face products. The Digital488 has an Inhibit line, among its six handshake/control lines,
for notifying external devices that the input latch is being read. This allows the external
device to hold new digital information until the current read is successfully performed.
84
84
Digital Isolation
Digital signals are isolated for several reasons:
to protect each side of the system from an inadvertent overvoltage condition on
the other side
to facilitate communication between devices with different grounds
to prevent injury when circuits are attached to individuals in medical applications
Optical Isolation. Optical isolation involves the use of an LED or diode laser to transmit
the digital signal and a photodiode or phototransistor to receive it. This technique is
depicted in Figure 8.05. IC-sized opto-couplers provide voltage isolation in the range of
500V. Using this technique, digital devices with dissimilar grounds can be controlled
and monitored.
High
2
One opto-isolator
Control
Digital
interface
6 Address
8 Data
V
cc
47K
4N28
Low
1.8K
To data
acquisition
system
Digital
inputs
Fig. 8.05: Schematic of opto-coupler isolating a system with its own ground
Pulse-Train Signal Conditioning
In many frequency measurement applications, pulses are counted and compared against
a fixed timebase. A pulse can be thought of as digital rather than analog since only the
number of rising or falling edges are of interest. In many instances, however, the pulse-
train signal comes from an analog source, like a magnetic pick up.
IOtechs DBK7 frequency-input card provides the LogBook, DaqBook, DaqBoard, and
Daq PC-Card data acquisition systems with four-channels of frequency input via two
separate front-end circuits, one for true digital inputs and one for analog inputs. An
analog input is a signal that is time-varying but not digital by nature.
85 85
The DBK7 conditions digital inputs of
different levels in the same manner as
the Digital488/80As amplifier module
discussed earlier. The analog input cir-
cuit is used to convert a time-varying
signal into a clean digital pulse-train.
Figure 8.06 shows the schematic of
the analog-input signal-conditioning
path. The front-end RC network pro-
vides AC coupling, allowing all sig-
nals above 10 Hz to pass. The select-
able attenuator desensitizes the cir-
cuit from unwanted low level noise
by reducing the overall magnitude of
the waveform.
When using a pulse-train from a relay closure, the DBK7 provides programmable
debounce settings that allow the user to select the amount of debounce time required.
The digital circuitry performs debouncing by monitoring the conditioned pulse-train
for a sustained low or high level. If debouncing is not performed, the extra edges in
the signal will produce an erroneously high, erratic frequency reading. Se Figure 8.07.
Many transducers produce an output
whose information content is not in the
signals amplitude but rather in the
signals frequency. Sensors that measure
rotational motion and flow typically fall
into this class. Photomultiplier tubes and
charged-particle detectors can also be
used for measurements that require pulse
counting. In principle, such signals
could be sampled with an ADC, but this
results in much more data than is neces-
sary and makes the analysis cumber-
some. Direct frequency measurement is
far more efficient.
0.33 F
20 K
BNC
Attenuate
1:1 or 1:50
Filter
100 kHz
300 kHz
or 30 Hz
Fig. 8.06: The DBK7 front end for converting analog
signals to a digital pulse stream
Fig. 8.07: The DBK7 debounce feature prevents relay
bounce edges from being counted
"Bouncy" input
Sustained
low
Ignored
edges
a b c d e f
Ignored
edges
Sustained
low
Time (ms)
0 10 20 30 40 50 60
Debounced
output
86
86
Fig. 8.08: Frequency-to-voltage using pulse integration
Raw input
After initial
signal
conditioning
Integrated
output voltage
Conditioned signal
has constant pulse width
Frequency-to-Voltage Conversion
Frequency-to-voltage conversion is commonly used to measure an input frequency via an
ADC. The frequency-to-voltage converters output is a DC-level signal whose magnitude is
proportional to the input frequency. One of two methods can be employed to convert a
pulse-train or AC signal into a DC level: pulse-train integration and digital-pulse counting.
Pulse-Train Integration. For single-chan-
nel modular signal conditioning, 5B
modules can be employed. The 5B
frequency-to-voltage module is a good
example of a pulse integrating F/V
(frequency-to-voltage). A series capaci-
tor provides AC coupling, removing any
very-low frequency or DC signal. A com-
parator circuit produces a pulse with a
constant duration each time the input
signal passes through zero. The result-
ing pulse is then passed through an in-
tegrating circuit (a low-pass filter), pro-
ducing a slowly changing level propor-
tional to the input frequency, as shown
in Figure 8.08.
The response time of the frequency-to-voltage converter is slow; it is the inverse of the
cutoff frequency of the low-pass filter. The cutoff frequency of the low-pass filter should
be much lower than the input frequencies of interest, yet high enough to provide the
required response time. As the mea-
sured frequency decreases toward the
cutoff frequency, significant ripple in
the output becomes a problem, as
shown in Figure 8.09.
Integrated circuits (ICs) that provide
frequency-to-voltage conversion are
available. Because an external capaci-
tor is used to select the time constant,
one can use a single IC to measure
signals in vastly different fixed fre-
quency ranges. To change range, the
Fig. 8.09: Output ripple seen at low frequencies
Low frequency
pulses
Integrated
output
87 87
capacitor would have to be changed. Frequency-to-voltage conversion works poorly
for frequencies below 100 Hz. This is because a low-pass filter with cutoff frequency
of under 10 Hz requires an extremely large capacitor.
Digital-Pulse Counting. Digital-pulse counting can be used to ensure a DC voltage that
is proportional to the input frequency. IOtechs DBK7, frequency-input card uses digi-
tal-pulse counting in this manner. Front-end circuitry converts the incoming analog or
digital signal into a clean pulse-train, devoid of relay contact bounce, high frequency
noise, and other unwanted artifacts. A microcontroller is then used to accurately mea-
sure a total period consisting of several cycles extending over one user-selectable mini-
mal peri od; thi s mi ni mal peri od determi nes the frequency resol uti on. The
microcontroller computes the frequency from the measured period and converts the
frequency to a command that it issues to a D/A converter, which provides the DC level
to the data acquisition system
Frequency
measurement DAC
BNC
a
f
LPF
Schmitt-trigger
buffer
Channel 0
Digital input mode
Jumpers shown configured
for analog input mode
Selectable
attenuator
Low pass
filter
Analog input mode
Microprocessor-based
frequency measurement circuit
To data
acquisition
system
Fig. 8.10: DBK7 frequency-input card, signal path
The advantages of this method are that very low frequencies can be measured, the
frequency range can be very wide, the output update can be quite fast, and the fre-
quency range can be programmed, allowing the expected frequencies to use the entire
analog-to-digital converters range.
88
88
The output range of the digital-to-analog converter is +5V to -5V divided into 4,096
parts. The minimum frequency selected by the user becomes the -5V output, while the
maximum frequency becomes +5V. The output update rate is regulated by F
min
and F
max
according to the following equation:
Minimum measurement period (sec) =(4,096 x 0.5 S) x (F
max
/(F
max
- F
min
))
For example, for a system with F
max
=7 kHz and F
min
=500 Hz, the digital-to-analog
converter output would be updated every 2.2 ms, or at a rate of 453 Hz.
In this equation, 4,096 is derived from the 12-bit digital-to-analog converter used, 0.5 S is
the resolution of the DBK7s timing circuits, and F
max
/(F
max
- F
min
) is the ratio of the measure-
ment time that must be increased to achieve 12-bit accuracy over the selected range.
Frequency Measurement by Gated Pulse Counting. Gated pulse counting can be used to
measure frequencies much more accurately than can be done with frequency-to-voltage
conversion. Gated pulse counting entails counting pulses that occur over a specified
period of time. One determines the frequency by dividing the number of pulses by the
counting interval. The error can be as low as the inverse of the counting interval. (For
example, if the counting interval is two seconds, the error can be as low as 0.5 Hz.)
Many data acquisition systems include TTL-compatible counter/timer ICs that can be
used to perform gated pulse digital level input. As such, they are unsuitable for use
directly with unconditioned analog signals. Fortunately, many frequency-output de-
vices feature a TTL output option.
IOtechs DaqBook and DaqBoard family of products use the 9513 counter/timer, which
features five counter/timers. Many counter/timer ICs, including the 9513, generally use an
oscillator built into the data acquisition system, or an external oscillator. Such ICs usually
have several channels available to facilitate counting applications. Each channel features
an input, a gate, and an output. The simplest counting method only uses the input. In this
configuration, the PC is programmed to periodically read and reset the counter.
This approachs weakness is uncertainty in the timing interval. Variations occur in the execu-
tion speeds of the functions that begin and end counting. In addition, the function call that
delays execution of the program for 50 ms employs an inaccurate software timer. These two
effects can render a short-counting interval-frequency measurement useless. However, the
technique is probably sufficient for counting intervals of greater than one second.
89 89
Gating can achieve greater accuracy because the gate controls the counting interval. As
a result, frequency measurements are independent of any software timing issues. The
gate can be configured so that pulses are counted only when a high signal enters the
gate. Similarly, the gate can begin counting when it detects one pulse and can stop
counting when it detects another.
A disadvantage of gated pulse counting is that it requires an extra counter to provide the
gate. However, in multiple-channel applications, a single counter can provide the gate for
many channels. For example, in a five-channel system, four channels count while one
channel provides a gate.
Timing Applications. You can also use a data acquisition system with a counter/timer in
timing applications. This can be accomplished by connecting a clock signal to the input
of a channel and using the input signal as a gate. This method requires that the gate be
configured for counting when the gate input is high. A similar technique can be used for
measuring the length of time between two pulses by configuring the gate to begin counting
at the first pulse and to end counting at the second.
Because a 16-bit counter overflows at 65,535 counts, the maximum pulse width measur-
able with a 1-MHz clock is 65.535 ms. A longer pulse will cause the counter to overflow,
producing an inaccurate measurement. A slower clock can be used for longer pulses.
90
90

Signal Conditioning & PC-Based Data Acquisition Handbook

Uploaded by

Copyright:

Available Formats

Signal Conditioning & PC-Based Data Acquisition Handbook

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Signal Conditioning & PC-Based Data Acquisition Handbook

Uploaded by

Copyright:

Available Formats

Signal Conditioning

Fig. 4.07: Open thermocouple circuit

Scale Gain Adj. Input Gain Adj.

Fig. 5.12: ICP

accelerometer attached to a typical dynamic signal-input card

, and Personal Daq

portable data acquisition systems, offer hardware-paced,

You might also like