KUO BENJAMIN C.

KUO
utomatic
control Syste
THIRD EDITION

HI
OX
2D
-'.

PRLNIlUt
HALL
 Automatic
Control
Systems
Third Edition

BENJAMIN C. KUO
Professor of Electrical Engineering
University of Illinois at Urbana-Champaign

PRENTICE-HALL, INC., Englewood Cliffs, New Jersey

 (,^-s

Library of Congress Cataloging in Publication Data

Kuo, Benjamin C
Automatic control systems.

Includes index.
1. Automatic control. 2. Control theory.
I. Title.
TJ213.K8354 1975 629.8'3 74-26544
ISBN 0-13-054973-8

METROPOLITAN
BOROUGH OF W1GAN
DEPT. OF LEISURE
LIBRARIES
*?
10067*2
A-
Ace. No,
•: o
:

5m^?1^;,
© 1975 by Prentice-Hall, Inc.
Englewood Cliffs, New Jersey

All rights reserved. No part of this book

may be reproduced any form or by any means
without permission in
in
writing from the publisher.
\

10 9 8 7 6 5 4

Printed in the United States of America

PRENTICE-HALL INTERNATIONAL, INC., London

PRENTICE-HALL OF AUSTRALIA, PTY. LTD., Sydney
PRENTICE-HALL OF CANADA, LTD., Toronto
PRENTICE-HALL OF INDIA PRIVATE LIMITED, New Delhi

PRENTICE-HALL OF JAPAN, INC., Tokyo

Contents

Preface j x

1 . Introduction

1.1 Control Systems 1

1.2 What Feedback and What Are Its Effects ?

Is

1.3 Types of Feedback Control Systems 11

2. Mathematical Foundation 15

2. 7 Introduction 15
2.2 Complex-Variable Concept 15
2.3 Laplace Transform 18
2.4 Inverse Laplace Transform by Partial-Fraction Expansion 21
2.5 Application of Laplace Transform to the Solution of Linear Ordinary
Differential Equations 25
2.6 Elementary Matrix Theory 26
2.7 Matrix Algebra 32
2.8 z-Transform 39

~\
vi / Contents

3. Transfer Function and Signal Flow Graphs 51

3.1 Introduction 51
3.2 Transfer Functions of Linear Systems 51
3.3 Impulse Response of Linear Systems 55
3.4 Block Diagrams 58
3.5 Signal Flow Graphs 64
3.6 Summary of Basic Properties of Signal Flow Graphs 66
3.7 Flow Graphs
Definitions for Signal 67
3.8 Signal-Flow-Graph Algebra 69
3.9 Examples of the Construction of Signal Flow Graphs 71
3. 10 General Gain Formula for Signal Flow Graphs 75
3.11 Application of the General Gain Formula to Block Diagrams 80
3.12 Transfer Functions of Discrete-Data Systems 81

4. State-Variable Characterization of Dynamic Systems 95

4. Introduction to the State Concept 95

4.2 State Equations and the Dynamic Equations 97
4.3 Matrix Representation of State Equations 99
4.4 State Transition Matrix 101
4.5 State Transition Equation 103
4.6 RelationshipBetween State Equations and
High-Order Differential Equations 107
4.7 Transformation to Phase-Variable Canonical Form 109
4.8 Relationship Between State Equations and Transfer Functions 115
4.9 Characteristic Equation, Eigenvalues, and Eigenvectors 117
4.10 Diagonalization of the A Matrix (Similarity Transformation) 118
4.11 Jordan Canonical Form 123
4. 12 StateDiagram 126
4. 13 Decomposition of Transfer Functions 136
4.14 Transformation into Modal Form 141
4.15 Controllability of Linear Systems 144
4.16 Observability of Linear Systems 152
4.17 Relationship Among Controllability, Observability,

and Transfer Functions 156

4. 18 Nonlinear State Equations and Their Linearization 158
4.19 State Equations of Linear Discrete-Data Systems 161
4.20 z-Transform Solution of Discrete State Equations 165
4.21 State Diagram for Discrete-Data Systems 167
4.22 State Diagrams for Samp/ed-Data Systems 171
4.23 State Equations of Linear Time-Varying Systems 173

5. Mathematical Modeling of Physical Systems 187

5. Introduction 187
5.2 Equations of Electrical Networks 188
5.3 Modeling of Mechanical System Elements 190
5.4 Equations of Mechanical Systems 203
5.5 Error-Sensing Devices in Control Systems 208
5.6 Tachometers 219
 Contents / vii

5.7 DC Motors in Control Systems 220

5.8 Two-Phase Induction Motor 225
5.9 Step Motors 228
5.10 Tension-Control System 235
5.11 Edge-Guide Control System 237
5. 12 Systems with Transportation Lags 242
5. 13 Sun-Seeker System 243

6. Time- Domain Analysis of Control Systems 259

6. Introduction 259
6.2 Response of Control Systems
Typical Test Signals for Time 260
6.3 Time-Domain Performance of Control Systems—Steady-State Response 262
6.4 Time-Domain Performance of Control Systems— Transient Response 271
6.5 Transient Response of a Second-Order System 273
6.6 Time Response of a Positional Control System 284
6.7 Effects of Derivative Control on the Time Response of
Feedback Control Systems 295
6.8 on the Time Response
Effects of Integral Control
of Feedback Control Systems 300
6.9 Rate Feedback or Tachometer Feedback Control 302
6. 10 Control by State- Variable Feedback 305

7. Stability of Control Systems 316

7.1 Introduction 316

7.2 Stability. Characteristic Equation, and the State Transition Matrix 317
7.3 Stability of Linear Time-Invariant Systems with Inputs 319
7.4 Methods of Determining Stability of Linear Control Systems 321
7.5 Routh-Hurwitz Criterion 322
7.6 Nyquist Criterion 330
7.7 Application of the Nyquist Criterion 344
7.8 Effects of Additional Poles and Zeros G(s)H(s) on the Shape
of the Nyquist Locus 352
7.9 Stability of Multiloop Systems 356
7.10 Stability of Linear Control Systems with Time Delays 360
7.11 Stability of Nonlinear Systems—Popov's Criterion 363

8. Root Locus Techniques 375

8. Introduction 375
8.2 Basic Conditions of the Root Loci 376
8.3 Construction of the Complete Root Loci 380
8.4 Application of the Root Locus Technique to the
Solution of Roots of a Polynomial 412
8.5 Some Important Aspects of the Construction of the Root Loci 417
8.6 Root Contour— Multiple-Parameter Variation 424
8.7 Root Loci of Systems with Pure Time Delay 434
8.8 Relationship Between Root Loci and the Polar Plot 444
8.9 Root Loci of Discrete-Data Control Systems 447
viii / Contents

9. Frequency-Domain Analysis of Control Systems 459

9. Introduction 459
9.2 Frequency-Domain Characteristics 462
9.3 M p .CO p , and the Bandwidth of a
Second-Order System 464
9.4 Effects of Adding a Zero to the Open-Loop Transfer Function 467
9.5 Effects of Adding a Pole to the Open-Loop Transfer Function 471
9.6 Relative Stability— Gain Margin, Phase Margin, and Mp 473
9.7 As Related to the Slope of
Relative Stability
the Magnitude Curve of the Bode Plot 483
9.8 Constant MLoci in the G(jOi) -Plane 485
9.9 Constant Phase Loci in the G{jCO)-Plane 489
9.10 Constant M and N Loci in the Magnitude - Versus-Phase Plane—
The Nichols Chart 490
9.11 Closed-Loop Frequency Response Analysis of Nonunity Feedback Systems 496
9. 12 Frequency Domain
Sensitivity Studies in the 497

10. Introduction to Control Systems Design 504

10.1 Introduction 504

10.2 Classical Design of Control Systems 510
10.3 Phase-Lead Compensation 515
10.4 Phase-Lag Compensation 535
10.5 Lag-Lead Compensation 552
10.6 Bridged-T Network Compensation 557

11. Introduction to Optimal Control 572

11.1 Introduction 572

1 1.2 Analytical Design 574
1 1.3 Parameter Optimization 583
1 1.4 Design of System with Specific Eigenvalues—An Application of Controllability 585
1 1.5 Design of State Observers 588
11.6 Optimal Linear Regulator Design 599
11.7 Design with Partial State Feedback 615

APPENDIX A Frequency-Domain Plots 626

A. 1 Polar Plots of Transfer Functions 627

A.2 Bode Plot (Corner Plot) of a Transfer Function 633
A.3 Magnitude-Versus-Phase Plot 643

APPENDIX B Laplace Transform Table 645

APPENDIX C Lagrange's Multiplier Method 650

Index 653
Preface

The first edition of this book, published in

1962, was characterized by having
chapters on sampled-data and nonlinear control systems. The
treatment of the
analysis and design of control systems was all classical.
The two major changes in the second edition, published in 1967, were the
inclusion of the state variable technique and the integration
of the discrete-data
systems with the continuous data system. The chapter on
nonlinear systems
was eliminated in the second edition to the disappointment of some
users of
that text. At the time of the revision the author felt that
a comprehensive treat-
ment on the subject of nonlinear systems could not be made effectively
with
the available space.
The third edition is still written as an introductory text for a senior course
on control systems. Although a great deal has happened in the area
of modern
control theory in the past ten years, preparing suitable material
for a modern
course on introductory control systems remains a difficult task.
The problem is
a complicated one because it is difficult to teach the topics concerned with new
developments in modern control theory at the undergraduate level. The unique
situation in control systems has been that many of
the practical problems are
still being solved in the industry by the
classical methods. While some of the
techniques in modern control theory are much more powerful
and can solve
more complex problems, there are often more restrictions when it comes
to
practical applications of the solutions. However, it should be recognized that
a modern control engineer should have an understanding of the classical
as
well as the modern control methods. The latter will enhance
and broaden one's
perspective in solving a practical problem. It is the author's
opinion that one
should strike a balance in the teaching of control systems theory at
the beginning
x / Preface

emphasis
and intermediate levels. Therefore in this current edition, equal
is

placed on the classical methods and the modern control theory.

A number of introductory books with titles involving modern control

have attempted to
theory have been published in recent years. Some authors
control with the modern control, but according
unify and integrate the classical
have Although such a goal is highly
to the critics and reviews, most failed.

desirable, if only from the standpoint of presentation, there does not seem to

be a good solution. It is possible that the objective may not be achieved until

new theories and new techniques are developed for this purpose. The fact
control systems, in some way, may be regarded as a science
of
remains that
learning how problem—control, in many different ways. These
to solve one
against each other,
different ways of solution may be compared and weighed
but it may not be possible to unify all the approaches. The
approach used in
method and the modern approach indepen-
this text is to present the classical
dently, and whenever possible, the two approaches are
considered as alterna-

tives, and the advantages and disadvantages of each

are weighed. Many

illustrative examples are carried out by both methods.

for not
Many existing text books on control systems have been criticized
including adequate practical problems. One reason for this is, perhaps, that
many text book writers are theorists, who lack the practical background and
that the
experience necessary to provide real-life examples. Another reason is
difficulty in the control systems area is compounded by
the fact that most real-

life problems are highly complex, and are rarely suitable as illustrative examples
is lost by simplifying
at the introductory level. Usually, much of the realism
the problem to fit the nice theorems and design techniques developed in the

text material. Nevertheless, the majority of the students taking a control system

course at the senior level do not pursue a graduate career, and they
must put
new employment. It is extremely
their knowledge to immediate use in their
important for these students, as well as those who will continue, to gain
an
actual feel ofwhat a real control system is like. Therefore, the author has
text. The
introduced a number of practical examples in various fields in this
homework problems also reflect the attempt of this text to provide more real-

life problems.
The following features of this new edition are emphasized by comparison

with the first two editions

1. Equal emphasis on classical and modern control theory.

2. Inclusion of sampled-data and nonlinear systems.
3. Practical system examples and homework problems.

The material assembled in this book is an outgrowth of a senior-level

control system course taught by the author at the University of

Illinois at

Urbana-Champaign for many years. Moreover, this book is written in a style

adaptable for self-study and reference.

Chapter 1 presents the basic concept of control systems. The definition of
feedback and its effects are covered. Chapter 2 presents mathematical founda-
 Preface / xi

tion and preliminaries. The subjects included are Laplace

transform, z-trans-
form, matrix algebra, and the applications
of the transform methods. Transfer
function and signal flow graphs are discussed in
Chapter 3. Chapter 4 intro-
duces the state variable approach to dynamical
systems. The concepts and
definitions of controllability and observability are
introduced at the early stage
These subjects are later being used for the analysis and
design of linear control
systems. Chapter 5 discusses the mathematical
modeling of physical systems.
Here, the emphasis is on electromechanical systems.
Typical transducers and
control systems used in practice are illustrated.
The treatment cannot be
exhaustive as there are numerous types of devices and
control systems. Chapter
6 gives the time response considerations of control systems. Both the classical
and the modern approach are used. Some simple design
considerations in the
time domain are pointed out. Chapters 7, 8, and 9 deal
with topics on stability,
root locus, and frequency response of control systems.
In Chapter 10, the design of control systems is discussed,
and the approach
is basically classical. Chapter 11 contains some of the optimal control subjects
which, in the author's opinion, can be taught at the
undergraduate level if time
permits. The text does contain more material than
can be covered in one
semester.
One of the difficulties in preparing this book was the weighing of what
subjects to cover. To keep the book to a reasonable length, some subjects,
which were in the original draft, had
to be left out of the final manuscript!
These included the treatment of signal flow graphs and
time-domain analysis,
of discrete-data systems, the second method of Liapunov's
stability method!
describing function analysis, state plane analysis, and
a few selected topics on
implementing optimal control. The author feels that the inclusion
of these
subjects would add materially to the spirit of the text,
but at the cost of a
higher price.
The author wishes to express his sincere appreciation to Dean W.
L.
Everitt (emeritus), Professors E. C. Jordan, O. L.
Gaddy, and E. W. Ernst,
of the University of Illinois, for their encouragement and
interest in the project!
The author is grateful to Dr. Andrew Sage of the University of Virginia
and
Dr. G. Singh of the University of Illinois for their valuable
suggestions. Special
thanks also goes to Mrs. Jane Carlton who typed a good portion
of the manu-
script and gave her invaluable assistance in
proofreading.

Benjamin C. Kuo
Urbana, Illinois
 1
Introduction

1 .1 Control Systems

In recent years, automatic control systems have assumed an increasingly

impor-
tant role in the development and advancement of modern civilization
and tech-
nology. Domestically, automatic controls in heating and air conditioning
systems
regulate the temperature and the humidity of modern homes for comfortable
living. Industrially,automatic control systems are found in numerous applica-
tions, such as quality control of manufactured products, automation,
machine
tool control, modern space technology and weapon systems, computer
systems,
transportation systems, and robotics.
Even such problems as inventory control,
socialand economic systems control, and environmental and hydrological sys-
tems control may be approached from the theory of automatic control.
The basic control system concept may be described by the simple block
diagram shown in Fig. 1-1. The objective of the system is to control the variable
c in a prescribed manner by the actuating signal e through the elements of the
control system.
In more common terms, the controlled variable is the output of the system,
and the actuating signal is the input. As a simple example, in the steering control
of an automobile, the direction of the two front wheels may be regarded
as the
controlled variable c, The position of the steering wheel is the input,
the output.
the actuating signal e. The controlled process or system in this case is composed
of the steering mechanisms, including the dynamics of the entire automobile.
However, if the objective is to control the speed of the automobile,
then the
amount of pressure exerted on the accelerator is the actuating signal, with the
speed regarded as the controlled variable.
 Chap. 1
2 / Introduction

Actuating Controlled
signal e variable c
Control
system (Output)
(Input)

Fig. 1-1. Basic control system.

There are many situations where several variables are to be controlled simul-
multivariabk
taneously by a number of inputs. Such systems are referred to as
systems.

Open-Loop Control Systems (Nonfeedback Systems)

The word automatic implies that there is a certain amount of sophistication

is usually
in the control system. By automatic, it generally means that the system
capable of adapting to a variety of operating conditions and is able to respond
to a class of inputs satisfactorily. However, not any type of control
system has
Usually, the automatic feature is achieved by feeding the
the automatic feature.
output variable back and comparing it with the command signal. When a system
does not have the feedback structure, it is called an open-loop system, which is
the simplest and most economical type of control system. Unfortunately, open-
loop control systems lack accuracy and versatility and can be used in none but
the simplest types of applications.
Consider, for example, control of the furnace for home heating. Let us

assume that the furnace is equipped only with a timing device, which controls
the on and off periods of the furnace. To regulate the temperature to the
proper level, the human operator must estimate the amount of time required
for the furnace to stay on and then set the timer accordingly. When the preset
time is up, the furnace is turned off. However, it is quite likely that the house
temperature is either above or below the desired value, owing to inaccuracy in
the estimate. Without further deliberation, it is quite apparent that this type of
control is inaccurate and unreliable. One reason for the inaccuracy lies in the

fact that one may not know the exact characteristics of the furnace. The other
factor is no control over the outdoor temperature, which has a
that one has
definite bearing on the indoor temperature. This also points to an
important
disadvantage of the performance of an open-loop control system, in that the

system not capable of adapting to variations in environmental conditions or

is

to external disturbances. In the case of the furnace control, perhaps an

experi-

enced person can provide control for a certain desired temperature in the house;

but if the doors or windows are opened or closed intermittently during the

operating period, the final temperature inside the house will not be accurately
regulated by the open-loop control.
An electric washing machine isanother typical example of an open-loop
system, because the amount of wash time is entirely determined by the judgment

and estimation of the human A operator. true automatic electric washing

machine should have the means of checking the cleanliness of the clothes con-
tinuously and turn itself off when the desired degree of cleanliness is reached.
Although open-loop control systems are of limited use, they form the basic
 Sec. 1.1 Control Systems / 3

elements of the closed-loop control systems. In general, the elements of an open-

loop control system are represented by the block diagram of Fig. 1-2. An input
signal or command applied to the controller, whose output acts as the
r is
actuating signal e; the actuating signal then actuates the controlled
process and
hopefully will drive the controlled variable c to the desired value.

Reference Actuating Controlled

input r signal e variable c
Controlled
Controller
process
(Output)

Fig. 1-2. Block diagram of an open-loop control system.

Closed-Loop Control Systems (Feedback Control Systems)

What is missing in the open-loop control system for more accurate and
more adaptable control is a link or feedback from the output to the input of
the
system. In order to obtain more accurate control, the controlled
signal c(t) must
be fed back and compared with the reference input, and an
actuating signal
proportional to the difference of the output and the input must be
sent through
the system to correct the error. A
system with one or more feedback paths like
that just described is called a closed-loop system.
Human beings are probably
the most complex and sophisticated feedback control system
in existence. A
human being may
be considered to be a control system with many inputs and
outputs, capable of carrying out highly complex operations.
To illustrate the human
being as a feedback control system, let us consider
that the objective is to reach for
an object on a desk. As one is reaching for the
object, the brain sends out a signal to the arm to
perform the task. The eyes
serve as a sensing device which feeds back continuously
the position of the hand.
The distance between the hand and the object is the error, which
is eventually
brought to zero as the hand reaches the object. This is a typical
example of
closed-loop control. However, if one is told to reach for the
object and then is
blindfolded, one can only reach toward the object by
estimating its exact posi-
tion. It is quite possible that the object may be missed by a wide margin. With
the eyes blindfolded, the feedback path is broken, and the human is operating
as an open-loop system. The example of the reaching of an object by a human
being is described by the block diagram shown in Fig. 1-3.
As another illustrative example of a closed-loop control system, Fig. 1-4

Error
Input detector
1 Controlled
command f x Error Controller
Controlled
variable
process
(brain)
Reach (arm and hand) Position
for object of hand

Fig. 1-3. Block diagram of a human being as a closed-loop control

system.
4 / Introduction Chap. 1

Rudder

Fig. 1-4. Rudder control system.

shows the block diagram of the rudder control system of a ship. In this case the
objective of control is the position of the rudder, and the reference input is
applied through the steering wheel. The error between the relative positions of
the steering wheel and the rudder is the signal, which actuates the controller
and the motor. When the rudder is finally aligned with the desired reference
direction, the output of the error sensor is zero. Let us assume that the steering
wheel position given a sudden rotation of R units, as shown by the time signal
is

in Fig. l-5(a). The position of the rudder as a function of time, depending upon
the characteristics of the system, may typically be one of the responses shown
in Fig. l-5(b). Because all physical systems have electrical and mechanical inertia,
the position of the rudder cannot respond instantaneously to a step input, but
will, rather, move gradually toward the final desired position. Often, the response

will oscillate about the final position before settling. It is apparent that for the
rudder control it is desirable to have a nonoscillatory response.

0,(0
6e W
R

-*-t *~t

(a) (b)

Fig. 1-5. (a) Step displacement input of rudder control system, (b) Typical
output responses.
 Sec. 1.1
Control Systems / 5

Error
sensor

Input ~^ Error
Controlled Output
J Controller
process

Feedback
elements

Fig. 1-6. Basic elements of a feedback control system.

The basic elements and the block diagram of a closed-loop control system
are shown in Fig. 1-6. In general, the configuration of a feedback control system
may not be constrained to that of Fig. 1-6. In complex systems there may be a
multitude of feedback loops and element blocks.
Figure l-7(a) illustrates the elements of a tension control system of a windup
process. The unwind reel may contain a roll of material such as paper or cable
which is to be sent into a processing unit, such as a cutter or a printer, and then
collects it by winding it onto another roll. The control system in this case is
to
maintain the tension of the material or web at a certain prescribed tension to
avoid such problems as tearing, stretching, or creasing.
To regulate the tension, web is formed into a half-loop by passing it
the
down and around a weighted The roller is attached to a pivot arm, which
roller.
allows free up-and-down motion of the roller. The combination of the roller and
the pivot arm is called the dancer.
When the system is in operation, the web normally travels at a constant
speed. The ideal position of the dancer is horizontal, producing a web tension
equal to one-half of the total weight W
of the dancer roll. The electric brake on
the unwind reel is to generate a restraining torque to keep the dancer in the
horizontal position at all times.
During actual operation, because of external disturbances, uncertainties
and irregularities of the web material, and the decrease of the effective diameter
of the unwind reel, the dancer arm will not remain horizontal unless some
scheme is employed to properly sense the dancer-arm position and control the
restraining braking torque.
To obtain the correction of the dancing-arm-position error, an angular
sensor used to measure the angular deviation, and a signal in proportion to
is

the error is used to control the braking torque through a controller. Figure
l-7(b) shows a block diagram that illustrates the interconnections between the
elements of the system.
 Chap. 1
6 / Introduction

Unwind reel
(decreasing dia.) Web
processing Windup reel
(increasing dia.)

Drive system
(constant web
speed)

(Current)

Reference
input
~"\ Error Electric Unwind Tension
Controller brake process

Dancer
arm

(b)

Fig. 1-7. (a) Tension control system, (b) Block diagram depicting the
basic elements and interconnections of a tension control system.

1 .2 What Is Feedback and What Are Its Effects ?

The concept of feedback plays an important role in control systems. We demon-

strated in Section 1.1 that feedback is a major requirement of a
closed-loop
control system would not be able to achieve
control system. Without feedback, a
required in most practical applications.
the accuracy and reliability that are
However, from a more rigorous standpoint, the definition and the significance
of feedback are much deeper and more difficult to demonstrate than the few
carry
examples given in Section 1.1. In reality, the reasons for using feedback
simple one of comparing the input with the output
far more meaning than the
error is merely one of the
in order to reduce the error. The reduction of system
many effects that feedback may bring upon a system. We shall now show that
 Sec. 1 .2 What Is Feedback and What Are Its Effects ? / 7

feedback also has effects on such system performance characteristics

as stability,
bandwidth, overall gain, impedance, and sensitivity.
To understand the effects of feedback on a control system, it is essential that
we examine this phenomenon with a broad mind. When feedback is deliberately
introduced for the purpose of control,
its existence is easily identified. However,
there arenumerous situations wherein a physical system that we normally rec-
ognize as an inherently nonfeedback system may turn out to have feedback
when it is observed in a certain manner. In general
we can state that whenever
a closed sequence of cause-and-effect relation exists among the variables
of a
system, feedback is said to exist. This viewpoint will inevitably admit
feedback
in a largenumber of systems that ordinarily would be identified as nonfeedback
systems. However, with the availability of the feedback and control system
theory, this general definition of feedback enables numerous systems,
with or
without physical feedback, to be studied in a systematic way once the existence
of feedback in the above-mentioned sense is established.
We shall now investigate the effects of feedback on the various aspects of
system performance. Without the necessary background and mathematical
foundation of linear system theory, at this point we can only rely on simple
static system notation for our discussion. Let us consider the simple feedback
system configuration shown in Fig. 1-8, where r is the input signal, c the output
signal, e the error, and b the feedback signal. The parameters G
and ZTmay be
considered as constant gains. By simple algebraic manipulations it is simple to
show that the input-output relation of the system is

G
M = t = FTW (l-i)

Using this basic relationship of the feedback system structure, we can uncover
some of the significant effects of feedback.

_.
i
-o
+
r e G c
- b +

^ _

H
-o

Fig. 1-8. Feedback system.

Effect of Feedback on Overall Gain

As seen from Eq. (1-1), feedback affects the gain G of a nonfeedback system
by a factor of 1 + GH. The reference of the feedback in the system of Fig. 1-8
is negative, since a minus sign is assigned to the feedback
signal. The quantity
GH may itself include a minus sign, so the general effect of feedback is that it
may increase or decrease the gain. In a practical control system, G and H are
 Chap. 1
8 / Introduction

functions of frequency, so the magnitude of 1 + GH

may be greater than 1 in
one frequency range but less than 1 in another. Therefore, feedback could
increase the gain of the system in one frequency range but decrease it in another.

Effect of Feedback on Stability

whether the system will be able to follow

Stability is a notion that describes
the input command. In a nonrigorous manner, a system is said to be unstable
if its output is out of control or increases without bound.

To investigate the effect of feedback on stability, we can again refer to the

expression in Eq. (1-1). If GH =-
1, the output of the system is infinite for any

finite input. Therefore, we may state that feedback can cause a system that is
originally stable to become unstable. Certainly, feedback is a two-edged sword;
when it is improperly used, it can be harmful. It should be pointed out, however,
that we are only dealing with the static case here, and, in general GH = — 1 is
not the only condition for instability.
It can be demonstrated that one of the advantages of incorporating feed-

back is that it can stabilize an unstable system. Let us assume that the feedback
system in Fig. 1-8 is unstable because GH =
—1. If we introduce another feed-
back loop through a negative feedback of F, as shown in Fig. 1-9, the input-
output relation of the overall system is

c G
r
~ I +GH+GF ( "

It is apparent that although the properties of G and H are such that the
inner-loop feedback system is unstable, because GH = — 1, the overall system
can be stable by properly selecting the outer-loop feedback gain F.

— o+
+ -o i

+ +
c
r e G
—o
b +- +

-o o-
H
-o o-

-o o-
F

Fig. 1-9. Feedback system with two feedback loops.

Effect of Feedback on Sensitivity

Sensitivity considerations often play an important role in the design of

control systems. Since all physical elements have properties that change with
environment and age, we cannot always consider the parameters of a control
Sec - 1-2 What Is Feedback and What Are Its Effects? / 9

system to be completely stationary over the entire operating life of the system.
For instance, the winding resistance of an electric motor changes as the tem-
perature of the motor rises during operation. In general, a good control system
should be very insensitive to these parameter variations while still able to follow
the command responsively. We shall investigate what effect feedback has on the
sensitivity to parameter variations.
Referring to the system in Fig. 1-8, we consider G as a parameter that may
vary. The sensitivity of the gain of the overall system M to the variation in G is
defined as

io
™ _~ dM/M ^- 3 >
~dGjG
where dM denotes the incremental change in M
due to the incremental change
inG; dM/M and dG/G denote the percentage change in and G, respectively. M
The expression of the sensitivity function Sg can be derived by using Eq. (1-1).
We have

io _
dM G _
SM
~lGM~l+GH
1
( M >

This relation shows that the sensitivity function can be made arbitrarily small
by increasing GH, provided that the system remains stable. It is apparent that
in an open-loop system the gain of the system will respond in a one-to-one
fashion to the variation in G.
In general, the sensitivity of the system gain of a feedback system to param-
eter variations depends on where the parameter is located. The reader may
derive the sensitivity of the system in Fig. 1-8 due to the variation of H.

Effect of Feedback on External Disturbance or Noise

All physical control systems are subject to some types of extraneous signals
or noise during operation. Examples of these signals are thermal noise voltage
in electronic amplifiers and brush or commutator noise in electric motors.
The effect of feedback on noise depends greatly on where the noise is intro-
duced into the system; no general conclusions can be made. However, in many
situations, feedbackcan reduce the effect of noise on system performance.
Let us refer to the system shown in Fig. 1-10, in which r denotes the com-
mand signal and n is the noise signal. In the absence of feedback,
0, the H=
output c is

c =GGe+Gx 2 2n (1-5)

where e = r. The signal-to-noise ratio of the output is defined as

output due to signal _

—GGe_ x
c 2 e
output due to noise G2 n
~~ l
~n
'
'

To increase the signal-to-noise ratio, evidently we should either increase

the magnitude of G, or e relative to n. Varying the magnitude of G would
have 2
no effect whatsoever on the ratio.
With the presence of feedback, the system output due to r and n acting
 Chap. 1
10 / Introduction

\h
n

+
+ + +
r e Gi e2 G2 c

b
+

__ _.

Fig. 1-10. Feedback system with a noise signal.

simultaneously is

_ Gl G 2
*
T + G,G 2 H
r + +
_|
1
b£3
G,G 2 H
n (1-7)
K
'

Simply comparing Eq. with Eq. (1-5) shows that the noise component in
(1-7)
the output of Eq. (1-7) reduced by the factor 1 + Gfi,H, but the signal com-
is

ponent is also reduced by the same amount. The signal-to-noise ratio is

output due to signal

~~
_GG i 2 rj(\ + G^G^H ) ___ g r_
(1-%}
output due to noise G 2 n/(l + G G H)
1 2
1
n

and is the same as that without feedback. In this case feedback is shown to have
no direct effect on the output signal-to-noise ratio of the system in Fig. 1-10.
However, the application of feedback suggests a possibility of improving the
signal-to-noise ratio under certain conditions. Let us assume that in the system
of Fig. 1-10, if the magnitude of G is increased to G\ and that of the input r
t

to r', with all other parameters unchanged, the output due to the input signal
acting alone is at the same level as that when feedback is absent. In other words,
we

'1-™ ^
let

= (1 ' 9)

With the increased G,, G\, the output due to noise acting alone becomes

which is smaller than the output due to n when G t

is not increased. The signal-
to-noise ratio is now

G 2 nl{\ + G\G 2 H) - n ^ + °^^> (1-11)

which is greater than that of the system without feedback by a factor of (1 +

G\G 2 H).
In general, feedback also has effects on such performance characteristics
 Seo -
1 - 3 Types of Feedback Control Systems 11
/

as bandwidth, impedance, transient response, and frequency response. These

effects will become known as one progresses into the ensuing material of this text.

1.3 Types of Feedback Control Systems

Feedback control systems may be classified in a number of ways, depending

upon the purpose of the classification. For instance, according to the method
of analysis and design, feedback control systems are classified as linear and non-
linear, time varying or time invariant. According to the types of signal found in
the system, reference is often made to continuous-data and discrete-data systems,

or modulated and unmodulated systems. Also, with reference to the type of

system components, we often come across descriptions such as electromechanical
control systems, hydraulic control systems, pneumatic systems, and biological
control systems. Control systems are often classified according to the main pur-
pose of the system. A
positional control system and a velocity control system
control the output variables according to the way the names imply. In general,
there are many other ways of identifying control systems according to some
special features of the system. It is important that some of these more common
ways of classifying control systems are known so that proper perspective is
gained before embarking on the analysis and design of these systems.

Linear Versus Nonlinear Control Systems

This classification is made according to the methods of analysis and design.

Strictly speaking, linear systems do not exist in practice, since all physical sys-
tems are nonlinear to some extent. Linear feedback control systems are idealized
models that are fabricated by the analyst purely for the simplicity of analysis
and design. When the magnitudes of the signals in a control system are limited
to a range in which system components exhibit linear characteristics (i.e., the
principle of superposition applies), the system is essentially linear. But when the
magnitudes of the signals are extended outside the range of the linear operation,
depending upon the severity of the nonlinearity, the system should no longer be
considered linear. For instance, amplifiers used in control systems often exhibit
saturation effect when their input signals become large; the magnetic field of
a motor usually has saturation properties. Other common nonlinear effects
found in control systems are the backlash or dead play between coupled gear
members, nonlinear characteristics in springs, nonlinear frictional force or tor-
que between moving members, and so on. Quite often, nonlinear characteristics
are intentionally introduced in a control system to improve its performance or
provide more effective control. For instance, to achieve minimum-time control,
an on-off (bang-bang or relay) type of controller is used. This type of control is
found in many missile or spacecraft control systems. For instance, in the attitude
control of missiles and spacecraft, jets are mounted on the sides of the vehicle
to provide reaction torque for attitude control. These jets are often controlled
in a full-on or full-off fashion, so a fixed amount of air is applied from a given
jet for a certain time duration to control the attitude of the space vehicle.
 Chap. 1
12 / Introduction

For linear systems there exists a wealth of analytical and graphical tech-

niques for design and analysis purposes. However, nonlinear systems are very
difficult to treat mathematically, and there are no general methods
that may be

used to solve a wide class of nonlinear systems.

Time-Invariant Versus Time-Varying Systems

When the parameters of a control system are stationary with respect to

time during the operation of the system, we have a time-invariant system. Most
physical systems contain elements that drift or vary with time to some extent.
If the variation of parameter is significant during the period of operation,
the

system is termed a time-varying system. For instance, the radius of the unwind
reel of the tension control system in Fig. 1-7 decreases with time as the material
is being transferred to the windup reel. Although a time-varying system without
nonlinearity is still a linear system, its analysis is usually much more complex
than that of the linear time-invariant systems.

Continuous-Data Control Systems

continuous-data system is one in which the signals at various parts of

A
the system are all functions of the continuous time variable t. Among all con-
tinuous-data control systems, the signals may be further classified as ac or dc.
Unlike the general definitions of ac and dc signals used in electrical engineering,

ac and dc control systems carry special significances. When one refers to an

ac control system it usually means that the signals in the system are modulated
by some kind of modulation scheme. On the other hand, when a dc control
system is referred to, it does not mean that all the signals in the system are of
the direct-current type; then there would be no control movement. dc control A
system simply implies that the signals are unmodulated, but they are still ac by
common definition. The schematic diagram of a closed-loop dc control system
is shown in Fig. 1-11. Typical waveforms of the system in response to a step

Error
^^^
6*^) "r detector
Reference Controlled
variable
input

6,

Fig. 1-11. Schematic diagram of a typical dc closed-loop control system.

 Sec. 1.3
Types of Feedback Control Systems / 13

function input are shown in the figure. Typical components

of a dc control sys-
tem are potentiometers, dc amplifiers, dc motors, and dc
tachometers.
The schematic diagram of a typical ac control system is shown in Fig. 1-12.
In this case the signals in the system are modulated;
that is, the information is
transmitted by an ac carrier signal. Notice that the output
controlled variable
still behaves similar to that of the
dc system if the two systems have the same
control objective. In this case the modulated signals
are demodulated by the
low-pass characteristics of the control motor. Typical
components of an ac
control system are synchros, ac amplifiers, ac motors,
gyroscopes, and acceler-
ometers.
In practice, not control systems are strictly of the ac or the dc type.
all
A
system may incorporate a mixture
of ac and dc components, using modulators
and demodulators to match the signals at various points of the system.

Synchro
transmitter

a-c servomotor

Reference
input

0.

Fig. 1-12. Schematic diagram of a typical ac closed-loop control system.

Sampled-Data and Digital Control Systems

Sampled-data and digital control systems differ from the

continuous-data
systems in that the signals at one or more points of the
system are in the form
of either a pulse train or a digital code. Usually,
sampled-data systems refer to
a more general class of systems whose signals are
in the form of pulse data,
where a digital control system refers to the use of a digital computer
or controller
in the system. In this text the term "discrete-data
control system" is used to
describe both types of systems.
In general a sampled-data system receives data
or information only inter-
mittently at specific instants of time.
For instance, the error signal in a control
system may be
supplied only intermittently in the form of pulses, in
which case
the control system receives no information about the
error signal during the
periods between two consecutive pulses. Figure 1-13
illustrates how a typical
sampled-data system operates. A
continuous input signal r(t) is applied to the
 Chap. 1
14 / Introduction

Input
r(t) eg) y
Sampler
e *c > Data
hold
(filter)
hit) Controlled
process
c(f)

Fig. 1-13. Block diagram of a sampled-data control system.

is sampled by a sampling device, the

sampler, and
system. The error signal e(t)
sampler a sequence of pulses. The sampling rate of the sam-
the output of the is

pler may or may not be uniform. There are many

advantages of incorporating
sampling in a control system, one of the most easily understood of these being
that sampling provides time sharing of an expensive equipment among several
control channels.
Because digital computers provide many advantages in size and
flexibility,

computer control has become increasingly popular in recent years.

Many air-

borne systems contain digital controllers that can pack several thousand
discrete
1-14 shows the
elements in a space no larger than the size of this book. Figure
basic elements of a digital autopilot for a guided missile.

Attitude
Digital
of
coded
Digital-to- missile
input Digital Airframe
analog
computer
converter

, ,

Analog-to-

con\ erter

Fig. 1-14. Digital autopilot system for a guided missile.

 2
Mathematical Foundation

2.1 Introduction

The study of control systems relies to a great extent on the use of applied
mathe-
matics. For the study of classical control theory, the prerequisites include such
subjects as complex variable theory, differential equations, Laplace transform,
and z-transform. Modern control theory, on the other hand, requires
consider-
ably more intensive mathematical background. In addition
to the above-men-
tioned subjects, modern control theory is based on the foundation
of matrix
theory, set theory, linear algebra, variational calculus, various
types of mathe-
matical programming, and so on.

2.2 Complex-Variable Concept

Complex-variable theory plays an important role in the analysis and design

of
control systems. When studying linear continuous-data systems, it is essential
that one understands the concept of complex variable and functions
of a complex
variable when the transfer function method is used.

Complex Variable

A complex variable j is considered to have two components: a real compo-

nent a, and an imaginary component co. Graphically, the real
component is
represented by an axis in the horizontal direction, and the imaginary
component
is measured along a vertical axis, in the complex j-plane. In other words,
a complex variable is always defined by a point in a complex plane
that has'
a a axis and ayco axis. Figure 2-1 illustrates the complex j-plane,
in which any

15
 Chap. 2
16 / Mathematical Foundation

/co s-plane

OJ]
i

°\

Fig. 2-1. Complex j-plane.

arbitrary point, s = su is denned by the coordinates a == a, and co = a>„ or

simply Si = ffj +y'coi.

Functions of a Complex Variable

s if for
The function said to be a function of the complex variable
G(s) is
value (or there are corresponding
every value of s there is a corresponding
values) of G(s). Since s is defined to have real and imaginary parts, the function
G(s) is also represented by its real and
imaginary parts; that is,

G(j) = ReG+yImC (2-1)

represents the imaginary

where Re G denotes the real part of G(s) and Im G
Thus, the function G(s) can also be represented
by the complex G-
part of G
measures
plane whose horizontal axis represents Re G
and whose vertical axis
every value of s (every point in the s-
the imaginary component of G{s). If for
plane) there is only one corresponding
value for G(s) [one corresponding point

in the G^-plane], G(s) is said to be

a single-valued function, and the mapping
points in the G(s)-plane is
(correspondence) from points in the j-plane onto
are many functions for
described as single valued (Fig. 2-2). However,
there
plane to the complex-variable plane is
which the mapping from the function

/co
. / ImG
x-plane

S, =0, +/C0, G 0)-plane

ReG

a, a

Gfri)

mapping from the s-plane to the G(»-plane.

Fig. 2-2. Single-valued
 Sec 2 2 -
-
Complex-Variable Concept / 17

not single valued. For instance, given the function

G(J)= <2
" 2)
,-(7TT)
it is apparent that for each value of s
there is only one unique corresponding
value for G(s). However, the reverse is not true; for instance,
the point G(s)
= oo is mapped onto two points, s and j =
1, in the j-plane. =—
Analytic Function

A function G(s) of the complex variable s is called an analytic function in

a region of the s-plane if the function and all its derivatives exist in
the region.
For instance, the function given in Eq. (2-2) is analytic at every point
in the s-
plane except at the points s and s =
-1. At these two points the value =
of the function is infinite. The function G(s) s 2 is analytic at every point = +
in the finite .s-plane.

Singularities and Poles of a Function

The singularities of a function are the points in the j-plane at which the func-
tion or its derivatives does not exist. A pole is the most common type of singu-
larity and plays a very important role in the studies of the classical control
theory.
The definition of a pole can be stated as: If a function G(s) is analytic and
single valued in the neighborhood of s except at s it is said to have a pole of
,
t ,
t

order r at s = s,if the limit

lim [0 - s,) r
G(s)]

has a finite, nonzero value. In other words, the denominator of G(s)

must include
the factor (s s,)
r
—so when s , =
s„ the function becomes infinite. If r 1, =
the pole at j =
s, is called a simple pole. As an example,
the function

°
G(s) W= l0(s
+ 2) n " xi
3>
s(s + IX* + 3)* (2

has a pole of order 2 at s = -3 and simple poles at s and s =

1. It can =-
also be said that the function is analytic in the j-plane except at these poles.

Zeros of a Function

The definition of a zero of a function can be stated as: If the function G(s)
is analytic at s =s t, it is said to have a zero of order r at s s l if the limit =
!S [(* ~
J '>" ,<7 W] ( 2 -4)

has a finite, nonzero value. Or simply, G(s) has a zero of order r at s s, ifl/G(s) =
has an rth-order pole at s = s,. For example, the function in Eq. (2-3) has a simple
zero at s = —2.
If the function
under consideration is a rational function of s, that is,
a quotient of two polynomials of s, the total number of poles
equals the total
number of zeros, counting the multiple-order poles and zeros, if the poles
and
 . , r. ^ • Chap. 2
18 / Mathematical Foundation

function in Eq. (2-3) has

zeros at infinity and at zero are taken into account. The
four finite poles at s =
0, -1, -3, -3; there is one finite zero at s -2, but =
there are three zeros at infinity, since

limGO) = lim^=0
S
(2
"5
)
s-«. s-<*>

Therefore, the function has a total of four poles and

four zeros in the entire s-

plane.

3-5
2.3 Laplace Transform

The Laplace transform is one of the mathematical tools used for the solution
of ordinary linear differential equations. In comparison with the classical method
transform method has the
of solving linear differential equations, the Laplace
following two attractive features

1. The homogeneous equation and the particular integral are solved

in one operation.
The Laplace transform converts the differential equation into an
2.

algebraic equation in s. It is then possible to manipulate the algebraic

equation by simple algebraic rules to obtain the solution in the
s

domain. The final solution is obtained by taking the inverse Laplace

transform.

Definition of the Laplace Transform

Given the function /(f) which satisfies the condition

r\f(t)e-°'\dt<oo (2-6)
J

some finite real a, the Laplace transform of /(f) is defined as

for

F(s)= \~ f(t)e-"dt (2-7)

or
m= £[/(')]

which
(2

a complex
-g
)

The variable s is referred to as the Laplace operator, is

variable; that is, s = a +jco. The defining equation of Eq. (2-7) is also known
evaluated from to oo
as the one-sided Laplace transform, as the integration
is

This simply means that all information contained in /(f) prior to t = is ignored

or considered to be zero. This assumption does not place

any serious limitation
linear system problems, since
on the applications of the Laplace transform to
reference often chosen at the instant
in the usual time-domain studies, time is

t = 0. Furthermore, for a physical system when an input is applied at t 0, =

the response of the system does not start sooner than t 0; that is, response =
does not precede excitation.
The following examples serve as illustrations on how Eq. (2-7) may be used

for the evaluation of the Laplace transform of a function /(f).

Sec. 2.3 Laplace Transform / 19

Example 2-1 Let/0) be a unit step function that is defined to have a constant value
of unity for t > and a zero value for / < 0. Or,

/(/) = u.(t) (2-9)

Then the Laplace transform of f(t) is

F(s) = £[u (t)] =s

j~ us «)e-" dt

e~" (2-10)
s

Of course, the Laplace transform given by Eq. (2-10) is valid if

j |
u£t)e-" dt |
= r | e-« | dt < co

which means that the real part of s, a, must be greater than zero. However, in practice,
we simply refer to the Laplace transform of the unit step function as lis, and rarely do
we have to be concerned about the region in which the transform integral converges
absolutely.

Example 2-2 Consider the exponential function

/(,) = e-", t 2;

where a is a constant.
The Laplace transform of/(?) is written

F(s) =
=/: e--'e-" dt s +a s +a (2-11)

Inverse Laplace Transformation

The operation of obtaining /(?) from the Laplace transform F(s) is termed
the inverse Laplace transformation. The inverse Laplace transform of F(s) is
denoted by

f(t) = Z-Vis)] (2-12)

and is given by the inverse Laplace transform integral

/(0 = /
' me"ds (2-13)
2^7
where c is a real constant that is greater than the real parts of all the singularities
of F(s). Equation (2-13) represents a line integral that is to be evaluated in
the j-plane. However, for most engineering purposes the inverse Laplace trans-
form operation can be accomplished simply by referring to the Laplace trans-
form table, such as the one given in Appendix B.

Important Theorems of the Laplace Transform

The applications of the Laplace transform in many instances are simplified

by the utilization of the properties of the transform. These properties are
presented in the following in the form of theorems, and no proofs are given.
 .

20 / Mathematical Foundation Chap. 2

1 Multiplicationby a Constant
The Laplace transform of the product of a constant k and a time func-
tion f{t) is the constant k multiplied by the Laplace transform of f{t);
that is,

£[kf(t)] = kF(s) (2-14)

where F{s) is the Laplace transform of f{t).

2. Sum and Difference
The Laplace transform of the sum {or difference) of two time functions
is the sum (or difference) of the Laplace transforms of the time func-
tions; that is,

£[fi(0 ± UO] = F,(s) ± F (s) 2 (2-1 5)

where F {s)
t
and F 2 {s) are the Laplace transforms of fit) and/2 (r),
respectively.
3. Differentiation
The Laplace transform of the first derivative of a time function f(t)
iss times the Laplace transform of f(f) minus the limit of f(t) as t
approaches 0-\-; that is,

df(ty = sF(s) - lim /(/)

dt (2-16)

= sF(s) - /(0+)
In general, for higher-order derivatives,

[d'fW] _ = s"F(s) -- lim s»-if(t) + s»~

ldl^-\- .

L df J (-0 +

n -- ^r /(o+) - 5"" 2
/ <u (o+) - ..-/ "-"(0+)
1 (
s F(s)

(2-17)

4. Integration

The Laplace transform of the first integral of a function fit) with

respect to time is the Laplace transform of f(t) divided by s; that is,

F{s)
S\ f(j)dr (2-18)
£ s

In general, for «th-order integration,

£
J
rr...r.
J
...
o Jo
fir) dx A,
o
. . . dt n
^ (2-19)

5. Shift in Time
The Laplace transform of f{t) delayed by time T is equal to the Laplace
Tl
transform of f{t) multiplied by e~ ; that is,

£[fit - T)us {t - T)] = e- T *F{s) (2-20)

where u s {t — T) denotes the unit step function, which is shifted in

time to the right by T.
 Sec 2 4
-
-
Inverse Laplace Transform by Partial-Fraction Expansion / 21

6. Initial-Value Theorem
If the Laplace transform of fit) is F(s), then

lim f(t) = lim sF(s) (2-21)

if the time limit exists.

7. Final-Value Theorem
If the Laplace transform of fit) is F(s)and ifsF(s) is analytic on the
imaginary axis and in the right half of the s-plane, then

lim f(t) = lim sFis) (2-22)

The final-value theorem is a very useful relation in the analysis and design
of feedback control systems, since it gives the final value of a time function
by
determining the behavior of its Laplace transform as s tends to zero. However,
the final-value theorem not valid
if sFis) contains any poles whose real part
is
is zero or positive, which equivalent to the analytic requirement of sFis)
is

stated in the theorem. The following examples illustrate the care that
one must
take in applying the final-value theorem.

Example 2-3 Consider the function

F(s)
s(s 2 +s+2)
Since sFis) is analytic on the imaginary axis and in the right half of the .s-plane, the
final-value theorem may be applied. Therefore, using Eq. (2-22),

5
lim fit) = lim sFis) = lim , ,
,
„ = -1 f2-231

Example 2-4 Consider the function

F& =
J2"^p (2-24)

which is known to be the Laplace transform of /(f) = sin cot. Since the function
sFis)
has two poles on the imaginary axis, the final-value theorem cannot be
applied in this
case. In other words, although the final-value theorem would
yield a value of zero as
the final value of fit), the result is erroneous.

2.4 Inverse Laplace Transform by Partial- Fraction Expansion 71 '

In a great majority of the problems in control systems, the

evaluation of the
inverse Laplace transform does not necessitate the use of the inversion
integral
of Eq. (2-13). The inverse Laplace transform operation involving
rational func-
tions can be carried out using a Laplace transform table and partial-fraction
expansion.
When the Laplace transform solution of a differential equation is a rational
function in s, it can be written

X^ =
W) ^
 Chap 2
22 / Mathematical Foundation -

where P(s) and Q(s) are polynomials of s. It is assumed that the order of Q(s)
in s is greater than that of P(s). The polynomial Q(s) may be written

Q(s) = s
n
+ a^"' 1
+ . . . +a n- t
s + a„ (2-26)

where a,, an are real coefficients. The zeros of Q(s) are either real or in
. . . ,

complex-conjugate pairs, in simple or multiple order. The methods of partial-

multiple-
fraction expansion will now be given for the cases of simple poles,
order poles, and complex poles, of X(s).

Partial-Fraction Expansion When All the Poles of X(s) Are Simple and Real

If all the poles of X(s) are real and simple, Eq. (2-25) can be written

X( S) = M P^ ,
,
(2-27)

where the poles — s — s 2 — s„ are considered to be real numbers in the

t , , . . . ,

present case. Applying the partial-fraction expansion technique, Eq. (2-27) is

written

X(vS') = -^-
s + i,
+ -4*-
s + s
+ • • + T^V
s s„ -1--
<2
- 28 )
2

The coefficient, K„ (i = 1, 2, . . . , n), is determined by multiplying both sides

of Eq. (2-28) or (2-27) by the factor (s + s ) and then setting s equal to —s,.
t

To K
find the coefficient sl , for instance, we multiply both sides of Eq. (2-27) by

(j + Ji) and let s =

— j, ; that is,

Pis,
K s\
(S + s )?& (s 2 — s t )(s 3 — Si)... (s„ — Si)
(2-29)

As an illustrative example, consider the function

(s + l)(,s + 2)(s + 3)

which is written in the partial-fractioned form

The coefficients K_ u K_ 2 and K_ 3 , are determined as follows:

+3
*-. = + W*)],-i = ( 2-~lX3 - l) _i
[('
: (2-32)

~ 2) + 3 =
*-2 = + 2)X(s)} ^ = (1- 5(
7 (2-33)
2X3 - 2)
[(.s s 2

5(--3) + 3
= [(* + 3)*(j)],-, (1- -6 ._
,-= (2-34)
3X2 - 3)
A"- 3

Therefore, Eq. (2-31) becomes

W) = tti + s +7 2 - t^
j + 3 ^ 6
(
2 - 35 >
 Sec. 2.4 Inverse Laplace Transform by Partial-Fraction Expansion
/ 23

Partial-Fraction Expansion When Some Poles of X(s) Are of Multiple Order

If r of the n poles of X(s) are identical, or say, the pole at s = —s, is of
multiplicity r, X(s) is written

/>(*)
*(*) A*)
(2-36)
g(5) (5 + s t Xs + **)...(* + *,)'(* + O
Then JT(j) can be expanded as

= ^^ *. AT,,

+s +
X(s) :

...+
J + s t s 2 s + s„

I
< — (« — r) terms of simple poles —
(2-37)
A, A2 A,
+
(* + J,)'

[
« — / terms of repeated poles — > |

The n - which correspond to simple poles, K

r coefficients,
n K, 2 K , , . .

may be evaluated by the method described by Eq. (2-29). The determination

of
the coefficients that correspond to the multiple-order
poles is described below.
Ar = [(S + *,)'*(*)],__„ (2-38)

Ar-^-Ks + s^Xis)] (2-39)

(2-40)

(2-41)

Example 2-5 Consider the function

1
X{s)
s{s + l)3(j + 2) (2-42)

Using the format of Eq. (2-37), X(s) is written

= *° + s-^-^ 4- ^' A
X(s)
J + 2 + s + +~(s +
-l
'

l '
;
z
+ ^3 ,

l)2T-(7+Tp
•

(2-43)

Then the coefficients corresponding to the simple poles

are
K = [sX(s)] s=0 =$ (2-44)
K- 2 = [(j + 2)-Y(j)],__ 2 = i-
(2-45)
and those of the third-order poles are
As =[(j + j
1) JT(j)] |,._,= -1 (2-46)

"^ ls=-i <wLt(j -{- 2).

~(2j + 2) =
(2-47)
s 2 (y + 2)2
and
1 d1 1 d[ -2(s + l)
Al
=2T^^ + 1)
3
^)] -i U
2 dsls 2 (s
(. + 2) 2
- 1 2(J + 1) 2(s + 1)
(2-48)
^ sHs + 2)3
, ,

is +
"+"
2y F(JT2p.
 : :

24 / Mathematical Foundation Chap. 2

The completed partial-fraction expansion is

X® = 27 + 2(7+2) ~ 7TT ~ (s + l) 3
(2 " 49)

Partial-Fraction Expansion of Simple Complex-Conjugate Poles

The partial-fraction expansion of Eq. (2-28) is valid also for simple complex-
conjugate poles. However, since complex-conjugate poles are more difficult to

handle and are of special interest in control-systems studies, they deserve sepa-
rate treatment here.
Suppose that the rational function X(s) of Eq. (2-25) contains a pair of
complex poles
5=— tx+jcu and s= — a— jco
Then the corresponding coefficients of these poles are

K-.+j. = (s + a - jco)X(s) U_ a+ w ,
(2-50)

K...J. = (s + a + jco)X(s) \... . Ja m (2-51)

Example 2-6 Consider the function

'« = ^+2^ + atf) (2 - 52)

Let us assume that the values of C and ©„ are such that the nonzero poles of X(s) are
complex numbers. Then X(s) is expanded as follows

*« = T + , + « 1% + , f« +% (2
" 53
>

where
a = Ccu„ (2-54)

and
oo = co„Vl - C 2 (2-55)

The coefficients in Eq. (2-53) are determined as

K = sX(s)\ , = s \ (2-56)

K-x+Ja> = (j + a-y(»)A'(j)|,._, +./

(2-57)
_ co? _ G)„ j(e+i/2)
2joo{— a +ye>) 2co
where

= tan-i[-f] (2-58)

Also,
*-«-/«, = (* +a +yo))A'(j)|,._ < ,. yto

(2-59)
— 2jco(— a — /a)) 2co

The complete expansion is

1 to r e--" s+ */2 > ejw+*n) i

 Sec 2 5 -
Application of Laplace Transform to the Solution /
-
25

Taking the inverse Laplace transform on both sides of the last equation gives

x(t) = 1 + 5k (g-yw+*/2) e (-«+y<»)( _|_ e ne+nm e {-*-Mt\

(2-61)
= 1 + ~e-" sia(COt - 6)
or

x(t) = + 6 "^' Sin(£0 "v/1

~ £ 2 ' ~ 6) <2
" 62
*
vi-C 2 >

where 6 is given by Eq. (2-58).

2.5 Application of Laplace Transform to the Solution of Linear

Ordinary Differential Equations

With the aid of the theorems concerning Laplace transform given in Section
2.3 and a table of transforms, linear ordinary differential equations can be
solved by the Laplace transform method. The advantages with the Laplace
transform method are that, with the aid of a transform table the steps involved
are all algebraic, and the homogeneous solution and the particular integral
solution are obtained simultaneously.
Let us illustrate the method by several illustrative examples.

Example 2-7 Consider the differential equation

d 2
x(t) ?dx(t) „ , ...
-d$T + 3 ~^- +
.
,

2*(f)
.

= 5u,(t) (2-63)

where u s (t) is the unit step function, which is defined as

The initial conditions are x(0+) = -1 and * (I) (0+) = dx(t)/dt |,. 0+ = 2. To solve
the differential equation we first take the Laplace transform on both sides of Eq. (2-63);
we have

s*X(s) - sx(0+) - *<»(0+) + 3sX(s) - 3*(0+) + 2*0) = — (2-65)

Substituting the values of *(0+) andx (1) (0+) into Eq. (2-65) and solving for X(s), we
get

~ sl ~ s + 5
w=
*Y(s) - ~ ja ~ s + 5 n
+ 3s + 2) ~ s(s + 1)0 + 2)
Kt\
(2_66)
j(j*

Equation (2-66) is expanded by partial-fraction expansion to give

* *)= 2!-j4t+20T2)
( (2 " 67)

Now taking the inverse Laplace transform of the last equation, we get the complete
solution as

x(t) = \ - 5e-< + le- 1 '

t > (2-68)
The first term in the last equation is the steady-state solution, and the last two
terms are the transient solution. Unlike the classical method, which requires separate
 Z

26 / Mathematical Foundation Chap. 2

steps to give the transient and the steady-state solutions, the Laplace transform method
gives the entire solution of the differential equation in one operation.
If only the magnitude of the steady-state solution is of interest, the final-value
theorem may be applied. Thus

lim x{t) = lim sX(s) = Iim ~~f~ S ~t = 4" (2-69)

(-.00 s-0 J-.0 ? T 3S + 2. I

where we have first checked and found that the function, sX(s), has poles only in the
left half of the .s-plane.

Example 2-8 Consider the linear differential equation

^P- + 34.5^- + 1000jc(O = 1000u )

s (t) (2-70)

where u s (t) is the unit step function. The initial values of x(t) and dx{t)jdt are assumed
to be zero.
Taking the Laplace transform on both sides of Eq. (2-70) and applying zero initial
conditions, we have

s 2 X(s) + 34.5sX(s) + 1000X(s) = — (2-71)

Solving X(s) from the last equation, we obtain

X s) = s(s + 34.5s + 1000)

^ 2 (2 " 72)

The poles of X(s) are at s = 0, s = -17.25 +7 26.5, and s = -17.25 -;26.5. There-
fore, Eq. (2-72) can be written as
1000
X s) = s(s +
(
+ (2 " 73)
17.25 -j26.5)(s 17.25 + J26S)
One way of solving for x(t) is to perform the partial-fraction expansion of Eq.
(2-73), giving
31.6 T e -J(«+*/2) g/<e+s/2) -i

= — + 5^63) U +
1
X(s)
17.25 -/26.5 +s+ 17.25 +726.5J
(2 " 74)

where
2
= tan-i(:=
J!)
= -56.9° (2-75)

• Then, using Eq. (2-61),

x(J) = 1 + 1.193e- I7 - 2 ='

sin(26.5f - 6) (2-76)

Another approach is to compare Eq. (2-72) with Eq. (2-52), so that

CO n = ±^1000 = ±31.6 (2-77)

and
C = 0.546 (2-78)

and the solution to x(t) is given directly by Eq. (2-62).

2' 6
2.6 Elementary Matrix Theory 1 -

In the study of modern control theory it is often desirable to use matrix notation
to simplify complex mathematical expressions. The simplifying matrix notation
may not reduce the amount of work required to solve the mathematical equa-
tions, but it usually makes the equations much easier to handle and manipulate.
 : :

Sec. 2.6
Elementary Matrix Theory / 27

As a motivation to the reason of using matrix notation, let us consider

the following set of n simultaneous algebraic equations:

«n*i + a, 2 *2 + . .
. + a x =y
ln n l

«21*1 + «22*2 + . . . + a 2n X„ = J2
(2-79)

a.i*i + a„ 2 x 2 + ...+ a„„x„ =y n

We may use the matrix equation

Ax =y (2-80)
as a simplified representation for Eq. (2-79).
The symbols A, x, and y are defined as matrices, which contain the coeffi-
cientsand variables of the original equations as their elements. In terms of matrix
algebra,which will be discussed later, Eq. (2-80) can be stated as: The product
of the matrices A andx is equal to the matrix y. The three matrices involved here
are defined to be

«21

(2-81)

«»i a»2

x2
X = (2-82)

x„

>l'

y = (2-83)

JV
which are simply bracketed arrays of coefficients and variables. Thus, we can
define a matrix as follows

Definition of a Matrix

A matrix is a collection of elements arranged in a rectangular or square array.

Several ways of representing a matrix are as follows

3 10"
A= A =(°
3
1 -2 .

\1 -2 :)
3 10
-2 A= [aj 2i3
1

In this text we shall use square brackets to represent the matrix.

 Chap. 2
28 / Mathematical Foundation

It is important to distinguish between a matrix and a determinant:

Matrix Determinant

An array of numbers or elements An array of numbers or elements

with n rows and m
columns. with n rows and n columns {always
Does not have a value, although a square),

square matrix (n = m) has a deter- Has a value,

minant.

Some important definitions of matrices are given in the following.

Matrix elements. When a matrix is written

an a, 2 a 13
a 23 (2-84)
#21 fl 22

a 31 a 32 a 31

au is identified as the element in the /th row and thejth column of the matrix.
As a rule, we always refer to the row firstand the column last.

Order of a matrix. The order of a matrix refers to the total number

of
the matrix in Eq. (2-84) has three
rows and columns of the matrix. For example,
matrix.
rows and three columns and, therefore, is called a 3 X 3 (three by three)
In general, a matrix with n rows and m columns is termed "« x m" or "n by m."

Square matrix. A square matrix is one that has the same number of rows
as columns.

Column matrix. A column matrix is one that has one column and more
than one row, that is, an m X 1 matrix, m> 1.
simply
Quite often, a column matrix is referred to as a column vector or
m rows. The matrix in Eq. (2-82) is a typical column
an m-vector if there are
matrix that is n X 1, or an n-vector.

Row matrix. A row matrix is one that has one row and more than one
column, that is, a 1 X n matrix. A row matrix can also be referred to as a row
vector.

Diagonal matrix. A diagonal matrix is a square matrix with a tj = for

all i ^j. Examples of a diagonal matrix are

Tfln "
"5 0"
a 22 (2-85)
3
a 33 .
matrix with
Unity matrix (Identity matrix). A unity matrix is a diagonal
all the elements on the main diagonal (i =j) equal to 1. unity matrix is often A
designated by I or U. An example of a unity matrix is

Tl 0"

1 = 1 (2-86)

1
 Sec. 2.6
Elementary Matrix Theory / 29

Null matrix. A null matrix is one whose elements are all equal to zero;
for example,
"0 0"
o= (2-87)

Symmetric matrix. A symmetric matrix is a square matrix that satisfies

the condition

au = <*jt (2-88)
for all i andj. A symmetric matrix has the property that if its rows are inter-
changed with its columns, the same matrix is preserved. Two examples of sym-
metric matrices are
T6 5 r
1 -4".
5 10 (2-89)
1 10 -1
Determinant of a matrix. With each square matrix a determinant
having
the same elements and order as the matrix may be defined. The
determinant of
a square matrix A is designated by

detA = A^ = |A| (2-90)

As an illustrative example, consider the matrix

1 -r
3 2 (2-91)
-1 1 oj
The determinant of A is

1 -1
|A| = 3 2 = -5 (2-92)
1 1

Singular matrix. A square matrix is said to be singular if the value of its

determinant is zero. On the other hand, if a square matrix has a nonzero deter-
minant, called a nonsingular matrix.
it is

When a matrix is singular, it usually means that not all the rows or
not all
the columns of the matrix are independent of each other.
When the matrix is
used to represent a set of algebraic equations, singularity of
the matrix means
that these equations are not independent of each
other. As an illustrative exam-
ple, let us consider the following set of equations:

2x t
— 3x2 + x = 3

-x, + x + x =2 3 (2-93)
x, - 2x + 2x =
2 3

Note that in Eq. (2-93), the third equation is equal to

the sum of the first two
equations. Therefore, these three equations are not
completely independent.
In matrix form, these equations may be represented by

AX =
 Mathematical Foundation
Chap. 2
30 /

where
2 -3 1

A= -1 1 1 (2-94)

1 -2 2

X= (2-95)

A3

and O is a 3 X 1 null matrix. The determinant of A is

2 -3 1

|A|= -1 1 1 4-3+2- 4 = (2-96)

1 -2 2

Therefore, the matrix A of Eq. (2-94) is singular. In this case the rows of A are
dependent.

Transpose of a matrix. The transpose of a matrix A is defined as the matrix

that is obtained by interchanging the corresponding rows and columns in A.
Let A
be an n X m
matrix which is represented by

A= [%]„, m
(2-97)

Then the transpose of A, denoted by A', is given by

A' = transpose of A = [a„] m> „ (2-98)

Notice that the order of A is n X m; the transpose of A has an order m X n.

Example 2-9 As an example of the transpose of a matrix, consider the matrix

"3 2 r
.0 -1 5.

The transpose of A is given by

A' = 2 -1
Ll 5.

Skew-symmetric matrix. A skew-symmetric matrix is a square matrix

that equals its negative transpose ; that is,

A= -A' (2-99)

Some Operations of a Matrix Transpose

1. (A')'=A (2-100)

2. (£A)' = AA', where k is a scalar (2-101)

3. (A + B)' = A' + B' (2-102)

4. (AB)' = B'A' (2-103)

 Sec. 2.6 Elementary Matrix Theory 31
/

Adjoint of a matrix. Let A be a square matrix of order n. The adjoint

matrix of A, denoted by adj A, is defined as

adj A= [if cofactor of det A]' (2-104)

where the ij cofactor of the determinant of A is the determinant obtained by omit-

ting the z'th row and the jth column of\A\ and then multiplying it by (—l) i+J .

Example 2-10 As an example of determining the adjoint matrix, let us consider

a 2 x 2 matrix,
first

A = «n
fll2

a2 i #22-
The determinant of A is

|A| = an «12

a%i o 22
The 1,1 cofactor, or the cofactor of the (1, 1) element of A|, | is a 22 the
; 1,2 cofactor is
— flu! the 2,1 cofactor is — a xl ; and the 2,2 cofactor is a n Thus, from Eq. (2-104),
.

the adjoint matrix of A is

1,1 cofactor 1,2 cofactor
adj A
2,1 cofactor 2,2 cofactor.

1,1 cofactor 2,1 cofactor"

(2-105)
J ,2 cofactor 2,2 cofactor.

#22 —#12
#21 auj

Example 2-1 1 As a second example

Ie of the adjoint matrix, consider

"a ii aj 2 an
#21 #22 #23 (2-106)
-O a a 32 #3 3
Then
1,1 cofactor 2,1 cofactor 3,1 cofactor"

adj A = 1,2 cofactor 2,2 cofactor 3,2 cofactor

1,3 cofactor 2,3 cofactor 3,3 cofactor_

(#22#3 3 ~ #23#32) — (#12#33 ~ #1 3#32) (#12#23 — #1 3#22)~

— (#21#33 — #23#3l) (#U#33 — #13#3l) ~ (#1 1#23 — #13#2l)
- (#21#32 — «22#3l) — (#U«32 ~ #12«3l) (#1 1#22 — #12#2l)-

(2-107)

Conjugate matrix. Given a matrix A whose elements are represented by

a tJ the conjugate of A, denoted by A, is obtained by replacing the elements of A
,

by their complex conjugates; that is,

 Chap. 2
32 / Mathematical Foundation

A= conjugate matrix of A
(2-108)
= [<y

where a tl = complex conjugate of a u .

2.7 Matrix Algebra

define matrix algebra in

When carrying out matrix operations it is necessary to
division, and other necessary
the form of addition, subtraction, multiplication,
point out at this stage that matrix operations are
operations. It is important to

defined independently of the algebraic

operations for scalar quantities.

Equality of Matrices
the
Two matrices A and B are said to be equal to each other if they satisfy
following conditions:

1. They are of the same order.

2. The corresponding elements are equal; that is,

a,j = b iJ for every /and j

For example,
a ti a 12 bn bn
A= B = (2-109)
_«2 1
fl 22. Pl\ 022_

implies that a n = b u a 12
,
= b 12 a 21
,
= b 2U and a 22 - b :

Addition of Matrices

Two matrices A and B can be added to form A + B if they are of the same
order. Then
A+B= [a l} ]„, m + [b,jl,, m =C= [c,7 L,, (2-110)

where
ClJ = au + b 'J (2-111)

and/ The order of the matrices is preserved after addition.

for all /

Example 2-12 As an illustrative example, consider the two matrices

_
" 2" 3
3

A= -1 4 B = -1 2

_ -1_ _ 1 0^

which are of the same order. Then the sum of A and

B is given by

~ 5"
3+0 + 3"
2 3

C=A+B -1-1 4 + 2 = -2 6 (2-112)

_ + 1 -1+0. _ 1 -1-
 „

Sec. 2.7 „ .
.,
Matrix Algebra / 33

Matrix Subtraction

The rules governing the subtraction of matrices are

similar to those of
matrix addition. In other words, Eqs. (2-1
10) and (2-1 1 1) are true if all the plus
signs are replaced by minus signs. Or,

C=A-B= _ l
[«,,]„ m [blJ m
= a ul,m + [-b
l iJ ] n . m (2-113)
= C
l tjh,m
where
Cti = a„ — b (2-114)
for all / and /

Associate Law of Matrix (Addition and Subtraction)

The associate law of scalar algebra still holds for matrix addition and
subtraction. Therefore,

(A + B) +C=A+ (B + C) (2-115)

Commutative Law of Matrix (Addition and Subtraction)

The commutative law for matrix addition and subtraction

states that the
following matrix relationship is true:

A+B+C=B+C+A
(2 " 116 >
=A+C+B
Matrix Multiplication

The matrices A and B may be multiplied together to form

the product AB
ifthey are conformable. This means that the number
of columns of A must equal
the number of rows of B. In other words, let

B= [*>„],,„,

Then A and B are conformable to form the product

C = AB = [a„]„ P [b,j\q m = [c J„,

, (2-H7)
ifand only ifp = q. The matrix C will have the same number of rows
the same number of columns
as A and
as B.
important to note that A and B may be conformable
It is
for AB but they
may not be conformable for the product BA, unless in
Eq. (2-117) n also equals
m. This points out an important fact that the
commutative law is not generally
valid for matrix multiplication. It is also
noteworthy that even though A and B
are conformable for both AB and BA, usually
ing references are
AB BA. In general, ^ the follow-
made with respect to matrix multiplication whenever
thev
J
exist:

AB = A postmultiplied by B
AB = B premultiplied by A
34 / Mathematical Foundation Chap. 2

Having established the condition for matrix multiplication, let us now turn
to the rule of matrix multiplication. When the matrices A and B are conformable
to form the matrix C = AB as in Eq. (2-117), the yth element of C, c„, is given by

Cii = Li a.kbkj (2-118)

for i = 1, 2, . . . , n, andj = 1, 2, . . . , m.

Example 2-13 Given the matrices

A= [fl,
7] 2 , 3 B=[6 ; ,] 3 ,i

we notice that these two matrices are conformable for the product AB but not for BA.
Thus,
"An"
Hn
«U 0n
«12 fln
«13
AB bn
-On Oil a 13-i
631- (2-119)

tfllAll + «12^21 + 013631

a2lAu + #22621 + O2363I
Example 2-14 Given the matrices
"3 -r
A 1 B=
J2 o_

we notice that both AB and BA are conformable for multiplication.

-3 -r -r
"i
AB = 1
3. 1 o_
_2 o_

"(3X1) + (-1)(2) (3)(0) + (-D(l) (3)(-l)+(-l)(0)"

(0)(1) + (1X2) (0)(0) + (1)(1) (OX-D+dXO) (2-120)

_(2)(1) + (0)(2) (2X0) + (0X1) (2)(-l) + (0)(0) _

_ -
1 -1 -3
2 1

_2 -2_
"3 -1"
"1 -11
BA 1
_2 1 Oj
_2 0_

"(1X3) + (0)(0) + (-1X2) (1)(-1) + (0)(1) h (-1X0)" (2-121)

_(2)(3) + (1)(0) + (0)(2) (2)(-l) + (1)(1) -1-

(0X0) -

n — 1"
1 1

j> -1.
Therefore, even though AB and BA may both exist, in general, they are not equal.
In this case the products are not even of the same order.
 Sec. 2.7 Matrix Algebra / 36

Although the commutative law does not hold in general for matrix multi-
plication, the associativeand the distributive laws are valid. For the distributive
law, we state that

A(B + C) = AB + AC (2-122)
if the products are conformable.
For the associative law,
(AB)C = A(BC) (2-123)
if the product is conformable.

Multiplication by a Scalar k
Multiplying a matrix A by any scalar k is equivalent to multiplying each
element of A by k. Therefore, if A= [a u ]„ m
(2-124)
kA = [ka u ]. m

Inverse of a Matrix (Matrix Division)

In the algebra for scalar quantities, when we write

ax =y (2-125)
it leads to

x=-y 1
(2-126)

or
x = a *y (2-127)
Equations (2-126) and (2-127) are notationally equivalent.
In matrix algebra, if
Ax = y (2-128)
then it may be possible to write
x = A _1 y (2-129)
where A 1
denotes the "inverse of A." The conditions that A -1
exists are:

1. A is a square matrix.
2. A must be nortsingular.
If A :
exists, it is given by

= adjA
A-' (2-130)

Example 2-15 Given the matrix

"11 «12
A= (2-131)
-Ozi a 2 2-
the inverse of A is given by
«22 — a 12
A' 1
= adj A (2-132)
dllOll — «12«21
 Chap. 2
36 / Mathematical Foundation

where for A to be nonsingular, A| | ^ 0, or

011^22
— «12«21 ¥= (2-133)

the numerator of A" we

If we pay attention to the adjoint matrix of A, which is »,

obtained by interchanging the two elements on

A
see that for a 2 x 2 matrix, adj is

signs of the elements on the off diagonal

of A.
the main diagonal and changing the

Example 2-16 Given the matrix

1 1

A= -10 2 (2-134)

L l l U
the determinant of A is
1 1

IA| = -1 "0 2 = 1
(2-135)

1 1 1

Therefore, A has an inverse matrix, and is given by

"-2 -1 2
A" 1
= 1 -2 (2-136)

lj

Some Properties of Matrix Inverse

1. AA" =A"»A 1 = I
(2-137)

2. (A" )"1 1
=A (2-138)

3. In matrix algebra, in general,

AB = AC (2-139)

does not necessarily imply B = C. The reader can easily construct

to illustrate this property. However, if A is a square

an example
nonsingular, we can premultiply both sides of Eq.
matrix, and is
(2-139) by A" Then 1
.

AAB = AAC (2-140)

°f
IB = IC (2-141)

which leads to
B =C
4. If A and B are square matrices and are nonsingular, then

(AB)" 1 =B-'A-' (2-142)

Rank of a Matrix

The rank of a matrix maximum number of linearly independent

A is the
is the order of the largest nonsingular matrix contained in
columns of A; or, it
on the rank of a matrix are as follows:
A. Several examples
 ,

Sec. 2.7
Matrix Algebra / 37

"0
1
=
"0514
o.
rank 1 rank = 2,
.0 _3 3 2
"3
9 2 "3

1 3 rank = 2, 1 2 rank
2 6 1 1

The following properties on rank are useful in the determination of the rank
of a matrix. Given annxm
matrix A,

1. Rank of A = Rank of A'.

2. Rank of A = Rank of A'A.
3. Rank of A = Rank of AA'.

Properties 2 and 3 are useful in the determination of rank;

since A'A and AA'
are always square, the rank condition can be checked by
evaluating the deter-
minant of these matrices.

Quadratic Forms

Consider the scalar function

/(x) =£2 a,jx,xj (2-143)

which is called the quadratic form. We can write this equation as

/(*) = 2 *,!>,,*,. (2-144)

Let

yt = % a,jXj (2-145)

Then Eq. (2-144) becomes

/00 = 2 X& (2-146)

Now if we define

Xi ~y\

x2 yi
x = y =

xn _ _j»_
Eq. (2-146) can be written

/(*) = x'y (2-147)

and from Eq. (2-145),

y = Ax (2-148)
where
A= [atf ],.» (2-149)
 Chap. 2
38 / Mathematical Foundation

Finally, /(x) becomes

/(x) = x'Ax (2-150)

Since the coefficient of xp s a n for i

is a„ +
j, given any ^
quadratic form as in

Eq. (2-150), we can always replace A with a symmetric matrix. In other words,

given any A, we can always define a symmetric matrix B such that

b
- bn =?ii+£», i*j (2-151)

are often used as performance indices in control

The quadratic forms
conveniences in the
systems design, since they usually lead to mathematical
design algorithms.

Definiteness

Positive definite. An n X n matrix A is said to be positive definite if all

the roots of the equation

|
AI —A = |
(2-152)

are positive. Equation (2-152) is called the characteristic equation

of A, and the
roots are referred to as the eigenvalues of A.

Positive semidefinite. The matrix A (n X n) is positive semidefinite if all

zero.
its eigenvalues are nonnegative and at least one of the eigenvalues is

Negative definite.The matrix A (n X n) is negative semidefinite if all its

eigenvalues are nonpositive and at least one of the eigenvalues is zero.

The matrix A {n x
Indefinite. n) is indefinite if some of the eigenvalues are

negative and some are positive.

An alternative way of testing the definiteness of a square matrix is to check

the signs of all the leading principal minors of the matrix. The leading
principal

minors of an n X n matrix are A

defined as follows. Given the square matrix

"an a 12 ... 0i

the n leading principal minors are the following determinants:

0n 012 013
a„ «12
a lt ^21 «22 a23
021 <*22
a 31 032 "33

Then the definiteness of A is determined as follows:

A is positive (negative) definite if all the leading principal minors
of A are

positive (negative).
A is positive semidefinite if A = | |
and all the leading principal minors of

A are nonnegative.
 Sec. 2.8 _ , __ ,
z-Jransform / 39

A is negative semidefinite if |
A| = and all the leading principal minors
of —A are nonnegative.
We may also refer to the definiteness of the quadratic
form, x'Ax.
The quadratic form, x'Ax (A is symmetric), is positive definite
(positive
semidefinite, negative definite, negative semidefinite)
if the matrix is positive A
definite (positive semidefinite, negative definite,
negative semidefinite).

2.8 z-Transform 1213

The Laplace transform is a powerful tool for the analysis and

design of linear
time-invariant control systems with continuous data.
However, for linear sys-
tems with sampled or discrete data, we may find that the
z-transform is more
appropriate.
Let us first consider the analysis of a discrete-data system
which is repre-
sented by the block diagram of Fig. 2-3. One way of
describing the discrete
nature of the signals is to consider that the input and the
output of the system
are sequences of numbers. These numbers are
spaced T seconds apart. Thus,
the input sequence and the output sequence may be
represented by r(kT) and
c(kT), respectively, k =
0, 1, 2, .... To represent these input and output
sequences by time-domain expressions, the numbers are
represented by impulse
functions in such a way that the strengths of the impulses
correspond to the
values of these numbers at the corresponding time
instants. This way, the input
sequence is expressed as a train of impulses,

'*(') = %/^T)5it - kT) (2-153)

A similar expression can be written for the output sequence.

r(kT) c(kT)
Discrete-data '('> •r^
_X. p>
(
,pK
',?(')
system

Fig. 2-3. Block diagram of a discrete-data Fig. 2-4. Block diagram of a

system -
iinite-pulsewidth sampler.

Another type of system that has discontinuous signals is the

sampled-data
system. A sampled-data system is characterized by having samplers in the system.
A sampler is a device that converts continuous data into some form of sampled
data. For example, Fig. 2-4 shows the block diagram of a typical sampler that
closes for a very short duration of/> seconds once every
seconds. This is referred r
to as a sampler with a uniform sampling period T and
a finite sampling duration p.
Figure 2-5 illustrates a set of typical input and output
signals of the sampler.
With the notation of Figs. 2-4 and 2-5, the output of the
finite-pulse-dura-
tion sampler is written

'J(0 = '(0 E [u.{t - *r) - u,(t -kT- p)] (2-1 54)

where u£t) is the unit step function.

40 / Mathematical Foundation Chap. 2

r(t) ire)

, n h
p T 2T

Fig. 2-5. Input and output signals of a finite-pulsewidth sampler.

For small p, that is, p <^T, the narrow-width pulses of Fig. 2-5 may be
approximated by flat-topped pulses. In other words, Eq. (2- 54) can be written 1

r*(t) ~ 2 r{kT)[u - kT) - u - kT - p)]

s
(t s {t
(2-155)

Multiplying both sides of Eq. (2-1 55) by l/p and taking the limit as p approaches
zero, we have

Hm — r*(r) — Urn f] —p r{kT)[u s (t

- kT) - u,{f - kT - p)]
p-0 p P^O k=o

= t, r{kT)8(t - kT)

or

lim — r*(0 - r\i) (2-156)

In arriving at this result we have made use of the fact that

6(t) = lim -L [k,(0 - u,(t ~ p)] (2-157)

J>-0 p
The significance of Eq. (2-156) is that the output of the finite-pulsewidth
sampler can be approximated by a train of impulses if the pulsewidth approaches
zero in the limit. A sampler whose output is a train of impulses with the strength
of each impulse equal to the magnitude of the input at the corresponding sam-
pling instant is called an ideal sampler. Figure 2-6 shows the block diagram of

an ideal sampler connected in cascade with a constant factor p so that the com-
bination is an approximation to the finite-pulsewidth sampler of Fig. 2-4 if p is
very small. Figure 2-7 illustrates the typical input and output signals of an ideal
sampler; the arrows are used to represent impulses with the heights representing
the strengths (or areas) of the latter.

r(t) r*(t) = V (t)

*. *

Ideal sampler

Fig. 2-6. Approximation of a finite-pulsewidth sampler by an ideal sam-

pler and a cascade constant factor.
 /

Sec. 2.8
z-Transform / 41

r*(t)

t j i A
—
T IT 3T 4T
HT
Fig. 2-7. Input and output signals of an ideal sampler.

Inview of these considerations we may now use the ideal sampler

to repre-
sent the discrete data, r(kT). This points to the fact that
the signals of the system
in Fig. 2-3 can essentially be treated as outputs
of ideal samplers.
Now we are ready to investigate the application of transform methods
to
discrete and sampled-data systems. Taking the
Laplace transform on both
sides of Eq. (2-153), we have

R*(s) = £ r(kT)e- (2-158)

The fact that Eq. (2-1 58) contains the exponential term e~ kTs
reveals the difficulty
of using Laplace transform for the general treatment of
discrete-data systems,
since the transfer function relations will no longer
be algebraic as in the continu-
ous-data case. Although it is conceptually simple to perform
inverse Laplace
transform on algebraic transfer relations, it is not a simple
matter to perform
inverse Laplace transform on transcendental functions.
One simple fact is that
the commonly used Laplace transform tables do not have entries with trans-
cendental functions in s. This necessitates the use of the
z-transform. Our moti-
vation here for the generation of the z-transform is simply
to convert trans-
cendental functions in s into algebraic ones in z. The
definition of z-transform
is given with this objective in mind.

Definition of the z-Transform

The z-transform is defined as

z = e
1
(2-159)
where s is the Laplace transform variable and J is the sampling period. Equation
(2-159) also leads to

s = — In z (2-160)

Using Eq. (2-159), the expression in Eq. (2-158) is written

R*(s = -L in z) = R(Z) = •£ r (kT)z~ (2-161)

or
R(z) = z-transform of /•*(/)

(2-162)
= [Laplace transform of r*(t)] s=1/Tlnz
 -

ap '

42 / Mathematical Foundation

Therefore, we have treated the z-transform as simply a change in variable,

z =
Ts
e .

The following examples illustrate some of the simple z-transform operations.

Example 2-17 Consider the sequence

r(kT) = e- kT , k = 0,1,2,... (2-163)

where a is a constant.
From Eq. (2-153),

#•*(/) = S e~° kT5(t - kT) (2-164)

Then
r*(s) = f; e
-' kT
e- kTl (2-165)
*=o

Multiply both sides of Eq. (2-165) by e

-<-° + ° )T
and subtract the resulting equation from
Eq. (2-165) ; now we can show easily that R*(s) can be written in a closed form,

<2
" 166)
*>(*) = !__}-<...»

for „ . ._.
|g-(a+«)r| < i (2-167)

where <T is the real part of s. Then the z-transform of /•*(/) is

itfy> - *
= I — (2-168)

forle-^z-'l < 1.

Example 2-18 In Example 2-17, if a = 0, we have

r(*D = l, * = 0,1,2,... (2-169)

which represents a sequence of numbers all equal to unity. Then

~
*•(*) = S= e kTs (2 " 170)
A

Biz) = £ z-* = 1 + z" + 2"* + z-> +

1
. • • (2-171)
k=a

This expression is written in closed form as

2 - 172)
R(z) = y4^ i*-m<i <

or
- 173
(2
^) = F=1 I
Z_1 I<1 >

functions are obtained by

In general, the z-transforms of more complex
preceding two examples. If a time
use of the same procedure as described in the
of finding its z-trans-
function r(t) is given as the starting point, the procedure
and then use Eq. (2-161) to get R(z).
form is to first form the sequence r(kT)
to send the signal r(t) through an
An equivalent interpretation of this step is

ideal sampler whose output is r*(t).

then take the Laplace transform We
and R(z) is obtained by substituting
of r*(t) to give R*(s) as in Eq. (2-158),
Ts
z for e .
 :

Sec. 2.8
z-Transform / 43

Table 2-1 Table of z-Transforms

Laplace Transform Time Function z-Transform

Unit impulse S(t)

Unit step «(/)

z- 1

StU) =2,6(t- nT) z

1 -e~ n=0 z- 1

\_
Tz
s2 -
(Z 1)2
± T*z{z + 1)
2 1(Z - 1)3
1 (-1)"
11 lim d- ( z \
s n+l -
n\
i™ n\ daAz e-°r)
1

s + a z — e-° T
1
te~ a <
Tze~' T
U+ a) 2 {z -
e-" T ) 2
a -
+ 1 — e~<"
(1 e-*r)z
s(s a) (z - l)(z - e-'T)
co z sin coT
S2 + CO 2 sin of
z2 — 2z cos coT + 1
CO
ze~"r sin coT
e~"< sin cot
(* + a) 2 + co 2 z 2 e 2<,T _ 2ze°T cos cor + 1
s
cos cot
z(z — cos coT)
S2 + CO 2
z2 — 2z cos co7" + 1
+as z2 — ze~ aT cos coT
e~a ' cos cot
(s + a) 2 + co 2 z2 — 2ze _ar coscor+ e~2or

Table 2-1 gives the z-transforms of some of the time functions commonly
used in systems analysis. A more extensive table may be found in the
litera-
ture. 12 13
-

Inverse z-Transformation

Just as in the Laplace transformation, one of the major

objectives of the
z-transformation is that algebraic manipulations can be
made first in the z-
domain, and then the final time response is determined by the inverse
z-transfor-
mation. In general, the inverse z-transformation of R(z) can yield
information
only on r(kT), not on r(t). In other words, the z-transform carries
information
only in a discrete fashion. When the time signal r(t) is sampled
by the ideal
sampler, only information on the signal at the sampling instants,
t kT, is =
retained. With this in mind, the inverse z-transformation can be effected by one
of the following three methods

1. The partial-fraction expansion method.

2. The power-series method.
3. The inversion formula.

Partial-fraction expansion method. The z-transform function R{z) is ex-

panded by partial-fraction expansion into a sum of simple recognizable terms,
 Chap. 2
44 / Mathematical Foundation

and the z-transform used to determine the corresponding r{kT). In

table is
slight difference between
carrying out the partial-fraction expansion, there is a
the z-transform and the Laplace transform procedures.
With reference to the z-
functions have the
transform table, we note that practically all the transform
into the form
term z in the numerator. Therefore, we should expand R{z)
2 - 174)
**) = r §£ a + r §£-+-" .
<

For this, we should expand R(z)jz into fractions and then multiply z across
first
illustrate this
to obtain the final desired expression. The following example will
recommended procedure.

Example 2-19 Given the z-transform function

R( ^ _ ~ e "T z T
) (2-175)
R{z > ~ (z - l)(z - e- )

it is desired to find the inverse z-transform.

Expanding R(z)lz by partial-fraction expansion, we have
R(z)
= _J
z — z —
I
e-" T
(2-176)
z 1

Thus,
n( .\ I I (2-177)

inverse z-transform of
From the z-transform table of Table 2-1, the corresponding
R(z) is found to be
r(kT) = 1 - e~'kT (2-178)

Power-series method. The z-transform R{z) is expanded into a power

the value
series in powers of z" In view of Eq. (2-161), the coefficient of z~* is
1
.

we
t = kT, or simply r(kT). For example, for the R(z) in Eq. (2-175),
of r(t) at
into a power series in powers of z' 1
by long division; then we have
expand it

21( Z ) = (i _ «-->"» + (1 - e-
2or
)z- 2 + (1 - e~ 3 °T )z-i

+ (l _ e -° kT)z- k +
" 179
(2
+ . . .
... >

or
R(z) =S (1 - e-°
kT
)z-
k ( 2_18() )
*=
ThUS '

r(kT)=l-e-» (2-18D

which is the same result as in Eq. (2-178).

Inversion formula. The time sequence r(kT) may be determined from R{z)
by use of the inversion formula,

r(kT)= ^U
2ff
<f
R{z)z k -'dz (2-182)
J J r

a contour integration along the path T, where T is

3 a circle of radius
which is
= cT centered at the origin in the z-plane, and c is of such a value that all
\z\ e

the poles of R{z) are inside the circle.

Sec. 2.8 z-Transform / 45

One way of evaluating the contour integration of Eq. (2-182) is by use of

the residue theorem of complex variable theory. Equation (2-182) may be
written

~1
r( kT ) = 2jp§ R(z)z k dz

= 2 Residues of i{(z)z*-> at the poles of R(z)z k ~ l

(2-183)

For simple poles, the residues of R(z)z k~1

at the pole z =z 3
is obtained as

k'1
Residue of R(z)z at the pole z, = (z- z J )R(z)z"- 1 |„ x , (2-184)

Now let us consider the same function used in Example 2-19. The function
R(z) of Eq. (2-175) has two poles: z 1 and z =
e~- r Using Eq. (2-183), we = .

have

r(kT) = [Residue of iJ(z)z*-» at z = 1] + [Residue of R(z)z k

- 1
at z = e~'T \

T k T k
(1 )z (1 )z
r
(z-e-»r ) ,., (z-1) ,_,^, (2-185)
= - e-" kT
1

which again agrees with the result obtained earlier.

Some Important Theorems of the z-Transformation

Some of the commonly used theorems of the z-transform are stated in the
following without proof. Just as in the case of the Laplace transform, these
theorems are useful in many aspects of the z-transform analysis.

1. Addition and Subtraction

If r x {kT) and r 2 {kT) have z-transforms Rfe) and J? 2 (z), respectively,
then
gUr^kT) ± r 2 (kT)] =R t
(z) ±R 2 (z) (2-186)

2. Multiplication by a Constant

S[ar(kT)] = as[r(kT)) = aR(z) (2-187)

where a is a constant.
3. Real Translation

$[r(kT - nT)] = z~"R(z) (2-188)

and

g[r(kT + nT)] = Ar{z) - "jj r(kT)z- k

~]
(2-189)

where n is a positive integer. Equation (2-188) represents the z-

transform of a time sequence that is shifted to the right by nT, and

Eq. (2-189) denotes that of a time sequence shifted to the left by nT.
The reason the right-hand side of Eq. (2-189) is not z"R(z) is because
the z-transform, similar to the Laplace transform, is defined only for
 . .

Chap. 2
46 / Mathematical Foundation

Eq. (2-189)
k > Thus, the second term on the right-hand side of
0.
that lost after is shifted to the
simply represents the sequence is it

left by nT.
4. Complex Translation

Z[e*°
kT
r(kT)\ = R{ze^ T ) (2-190)

5. Initial -Value Theorem

lim r(kT) = lim R(z) (2-191)

if the limit exists.

6. Final-Value Theorem

lim r{kT) = lim (1 - z' l )R{z) (2-192)

if the function, (1 -
z" ')*(*). has no poles on or outside the unit
circle centered at the origin in the z-plane, \z\= 1

these theorems.
The following examples illustrate the usefulness of

Example 2-20 Apply the complex translation theorem to find the z-transform of
/(/) = te--, t ^ 0.

Let KO = U ?>0;then

R(z) = *[«/.(01 = g(kT) = ^jjj (2-193)

Using the complex translation theorem,

F{z) = glte-'uM] = R{ze") = Jl^-.ry (2

" 194 )

Example 2-21 Given the function

0.792z 2
R( z >
-
= - l)(z 2 - 0.416z + 0.208)
(2 . 195)
(z

determine the value of r(kT) as k approaches infinity.

Since
° 7 9 2Z (2496)
(l-^-W)- z2 _ 41 6 2 + 0.208
\z\ = in the z-plane, the final-
does not have any pole on or outside the unit circle 1

value theorem of the z-transform can be applied.

S K*r> -
k
Hence,

lim
z2 _o 416 z + 0.208
92

powers of z~
=" '

l
^
This result is easily checked by expanding R(.z) in ,

R(z) = 0.792z-i

+ 0.981z" +
+
6
i.i21z-»

0.998z~ 7
+ 1.091*-'

+ . .
+ 1.013*"* + 0.986z-=
^^
apparent that the coefficients of this power series converge rapidly to the final
It is

value of unity.
 Chap. 2
Problems / 47

REFERENCES
Complex Variables, Laplace Transforms, and Matrix Algebra
1. F-B.Hildebrani}, Methods of Applied Mathematics,
Prentice-Hall Inc Enele-
wood Cliffs, N.J., 1952. ' ''

2. R. Bellman, Introduction to Matrix Analysis, McGraw-Hill Book Company,

i one,
New
i "dO.

3. B C. Kuo, £*«<?«/• Ai*™,*, an/ Systems, McGraw-Hill Book Company New

York, 1967.
4. R. Legros and A. V. J. Martin, Transform Calculus
for Electrical Engineers
Prentice-Hall, Inc., Englewood Cliffs, N.J., 1961.
5. C. R. Wylie, Jr., Advanced Engineering
Mathematics, 2nd ed., McGraw-Hill
Book Company, New York, 1960.
6 ETT Matrices ' Polynomials, and Linear
'

%I£?
lttt
'
;
Trans. Automatic Control, Vol. AC-18,
Time-Invariant Systems,"
pp. 1-10, Feb. 1973.

Partial Fraction Expansion

7. D Hazony and J. Riley, "Evaluating Residues and Coefficients of High Order

Poles, IRE Trans. Automatic Control, Vol. AC-4, pp. 132-136, Nov. 1959.
8 L Partial Fraction
tI
E^ansion of a Rational Function with
'

S u°7 »
Multiple f°uby Digital
Poles
!f.
Computer," IEEE Trans. Circuit Theory, Vol CT-11
pp. 161-162, Mar. 1964.
"
9 QdCk ChCCk °n
UNIS A
'

^ix°
IEEE ;
Trans. Automatic Control, Vol. AC-11,
Fraction Expansion
pp. 318-319, Apr. 1966.
Partial Coefficients,"

"
10 R R Partkl FraCti
° n Expami0n of RationaI F ^tions
'

wi't^r^n h
One High-Order
with B
Pole," f°:.
IEEE Trans. Automatic Control, Vol. AC-13
' p 133
Feb. 1968. '

"'
v ?A^
Vol. yrpp.
AC-16,
A S
'
l Partial FraCti ° n A, g° rithm ."
489-491, Oct. 1971.
IEEE Trans. Automatic Control,

Samp/ed-Data and Discrete-Data Control Systems

12. B. C. Kuo Analysis and Synthesis of Sampled- Data

Control Systems, Prentice-
Hall, Inc., Englewood Cliffs, N.J., 1963.
13. B C. Kuo, Discrete Data Control Systems, Science-Tech, Box 2277
Champaign, Illinois, 1970.
Station A

PROBLEMS
2.1. Find the poles and zeros of the following
functions (include the ones at infinity) •

W
(a) G(s)
() =
*Hs
5(s
+
+
+ 5)
2Xs
1)

* 2 (*
+ 1)
(b) G(s)
' W =
(s + 2)(s> + 3* + 2)
 : : :

Mathematical Foundation Chap. 2

48 /

= +
(C) G(s) f f' ^u
-Ke"
(d) G(s) =
+ 2)
:

(j + DC*
2.2. Find the Laplace transforms of the following functions
(a) g{t) = te-*>
(b) git) = cos 5/ t

(c) #(0 = e~' sin CO/

(d) git) = S sikT)5it - fcT); <*(<) = unit impulse function
*=o

2.3. Find the Laplace transforms of the functions shown in Fig. P2-3.

gU)

-
1

1
1 2 3 4 5 6 7
P
sl

(a)

irOT

(b)

Figure P2-3.

2.4. Find the Laplace transform of the following function

fO t < 1

t + 1 1<? <3
fit) = '

4 3<f<4
4 <t
2.5. Solve the following differential equation by means of the Laplace transforma-
tion:
wm^j
^
dM + 4/(0 =
dt*
J
5
dt
e
-,
uAt)

Assume that all the initial conditions are zero.

2.6. Find the inverse Laplace transforms of the following functions

l
(a) Gis)
is + 2)is + 3)
1
(b) Gis)
+ D 2 (s +
1

(.? 4)
10
(c) Gis) - s(j2
+ 4)(j + 1}

(d) Gis) - s(s2 +s 2)

+
Chap. 2 Problems / 49

2.7. Carry out the following matrix sums and differences:

"3 6" 0~
(a) 7
-5_
+ -3
_0 - 10_
(b) " 15" r 20"
-1 - -4
. 3_ _ 5.
(c)
—s s + 1 10
+
s-3
1
—
2.8. Determine if the following matrices are conformable for the products AB and
BA. Find the valid products.
(a) rii
A= B= [6 1]

_3_
(b) "2 -r 10 9"
B=
0. -1 -1 0.

2.9. Express the following set of algebraic equations in matrix form:

5x, + x2 — x = 3 1

— *i + 3*2 — x = 3 1

3*i — 7* 2 — 2*3 =
2.10. Express the following set of differential equations in the form ±(t) = Ax(f)
+ Bu(»:
*i(0 = -*i(0 + x 2 {t)
x 2 (t) = -2* 2 (0 - 3* (0 + in(t)3

x 3 (t) = -*,(?) - 5x2 (t) - 3* (f) + 3 u 2 (t)

2.11. Find the inverse of the following matrices:
(a) "
2 5"
.

JO -1^
(b) 3 -11
r
-2 1 2
_ 1 -1_
(c) r l 3 41
-l 1

_-i -1_
2.12. Determine the ranks of the following matrices:
(a) "3 2"

7 1

_0 3_
(b) "2 4 8"

Ll 2 6 3_
 : :

50 / Mathematical Foundation Chap. 2

2.13. Determine the definiteness of the following matrices:

"
(a) 2 3"
_-l 2_
(b)
"
1 5 -r
-2
_ 3 1 i_

2.14. The following sampled by an ideal sampler with a sampling period

signals are
of Determine the output of the sampler, /*(/), and find the Laplace
T seconds.
transform of/*0), F*(s). Express F*(s) in closed form.
(a) /(/) = te~*
(b) fit) = e~" sin cot
2.15. Determine the z-transform of the following functions
*
(a) G(s)
is + a)n
1
(b) G(j)
s(s + 5)
2

(c) CKs)= 1

s (s + 2) 3

(d) g(f) = t*e-*'

(e) git) = sin cot
t

2.16. Find the inverse z-transform of

G(z)
lOzjz + 1)
(z - l)(z 2 +z 1)

by means of the following methods

(a) the inversion formula
(b) partial-fraction expansion
 3
Transfer Function and

Signal Flow Graphs

3.1 Introduction

One of the most important steps in the analysis

of a physical system is the
mathematical description and modeling of the
system. A mathematical model
ot a system is essential because it allows
one to gain a clear understanding of
the system in terms of cause-and-effect
relationships among the system com-
ponents.
In general, a physical system can be
represented by a schematic diagram
that portrays the relationships and
interconnections among the system com-
ponents. From the mathematical standpoint,
algebraic and differential or
difference equations can be used to describe
the dynamic behavior of a system
In systems theory, the block diagram is
often used to portray systems of
all types. For linear systems,
transfer functions and signal flow graphs are
valuable tools for analysis as well as for design.
In this chapter we give the definition of
transfer function of a linear system
and demonstrate the power of the signal-flow-graph
technique in the analysis
of linear systems.

3.2 Transfer Functions of Linear Systems

Transfer function plays an important role in

the characterization of linear
time-invariant systems. Together with block
diagram and signal flow graph
transfer function forms the basis of representing
the input-output relationships
ot a linear time-invariant system in
classical control theory.
The starting point of defining the transfer function
is the differential
equa-

51
 . : :

62 Function and Signal Flow Graphs Chap. 3

/ Transfer

tion of a dynamic system. Consider that a linear time-invariant system is

described by the following nth-order differential equation

-
. d m r{t) , . dm '/•(?) , .
/,
dr(t) , , ,.

where c(r) is the output variable and r(t) is the input variable. The coefficients,

a au
, . . , a„ b m are constants, and n
and b , bt m. >
The differential equation in Eq. (3-1) represents a complete description of
the system between the input r(t) and the output c(t). Once the input and the
initial conditions of the system are specified, the output response may be

obtained by solving Eq. (3-1). However, it is apparent that the differential

equation method of describing a system is, although essential, a rather cum-
bersome one, and the higher-order differential equation of Eq. (3-1) is of little
practical use in design. More important is the fact that although efficient sub-
routines are available computers for the solution of high-order
on digital

differential equations, the important development in linear control theory relies

on analysis and design techniques without actual solutions of the system differ-
ential equations.
A convenient way of describing linear systems is made possible by the use
of transfer function and impulse response. To obtain the transfer function of the
linear system that is represented by Eq. (3-1), we take the Laplace transform on
both sides of the equation, and assuming zero initial conditions, we have
(a s n + a^"" 1
+ . . .+ a ^s + a„)C(s)
n

= (b s m + b.s'"-' + ... + b m .,s -L-

b m )R(s)

The transfer function of the system is defined as the ratio of C(s) to R(s);
therefore,
-

UW -Q£- b sm + b sm +... + bm . s + b m
i

r(*\ (3 _ 3)
~ R(s) ~ a + a.s"-
1 1

s"
1
+... + a . s a.
K t
-,-

Summarizing over the properties of a transfer function we state

1. A transfer function is defined only for a linear system, and, strictly,

only for time-invariant systems.

2. A transfer function between an input variable and an output variable
of a system is defined as the ratio of the Laplace transform of the
output to the Laplace transform of the input.
3. All initial conditions of the system are assumed to be zero.
4. A transfer function is independent of input excitation.
The following example is given to illustrate how transfer functions for a

linear system are derived.

Example 3-1 A series RLC network is shown in Fig. 3-1. The input voltage is

designated by e,{t). The output variable in this case can be defined as

the voltage across any one of the three network elements, or the current
 Sec. 3.2
Transfer Functions of Linear Systems / 53

R I i(t). The loop equation of the network

+ o v\M Tm is written

*,(/) =Ri(t)+L^ + ±j /(,) dt (3-4)

ei<" 'W ^) ;;C e c (t)

Taking the Laplace transform on both sides of Eq. (3-4)
-. and assuming zero initial conditions, we have

EAs) = (r+Ls+^)I(s) (3-5)

Fig. 3-1. RLC network. If we regard the current i (/) as an output variable, the transfer
function between e,{t) and i(t) is simply
./(*) _ 1 Cs
E,(s) R+Ls + (l/Cs) ~ + RCs + LCs I 2 ^'^
If the voltage across the capacitor e (t) is considered as an output, the transfer
c function
between e,(0 and e c (t) is obtained by substituting

E (s)=-~Ks)
c
(3.7)
into Eq. (3-5). Therefore,

E (s) _
c 1
-8
£,(*) 1 + RCs + LCs 2 <3 )

The definition of transfer function easily extended to a system with a

is
multiple number of inputs and outputs. A
system of this type is often referred
to as a multivariable system. In a multivariate
system, a differential equation
of the form of Eq. (3-1) may be used to describe the
relationship between a pair
of input and output. When dealing with the relationship
between one input and
one output, it is assumed that all other inputs are set to zero.
Since the principle
of superposition is valid for linear systems, the total effect
on any output variable
due to all the inputs acting simultaneously can be
obtained by adding the
individual effects.
As a
simple illustrative example of the transfer functions
of a multivariable
system, us consider the control of a turbopropeller engine.
let
In this case the
input variables are the fuel rate and the propeller
blade angle. The output
variables are the speed of rotation of the engine
and the turbine-inlet tempera-
ture. In general, either one of the outputs is
affected by the changes in both
inputs. For instance, when the blade angle of
the propeller is increased, the
speed of rotation of the engine will decrease and
the temperature usually
increases. The following transfer relations
may be written from steady-state
tests performed on the system:

C,(j) = G, ,(*)*,(*) + G 12 (s)R 2 (s) (3-9)

Ci(s) = G ,(s)R (s) + G 22 (s)R
2 l 2 (s) (3-10)
where
C,(i) = transformed variable of speed of rotation
C = transformed variable of turbine-inlet temperature
2 (s)

Ri(s) = transformed variable of fuel rate

R (s) = transformed variable of propeller blade angle
2

All these variables are assumed to be measured from

some reference levels.
54 / Transfer Function and Signal Flow Graphs Chap. 3

Since Eqs. (3-9) and (3-10) are written with the assumption that the system
is linear, the principle of superposition holds. Therefore, G n (s) represents the
transfer function between the fuel rate and the speed of rotation of the engine
with the propeller blade angle held at the reference value ; that is, R 2 (s) = 0.

Similar statements can be made for the other transfer functions.

In general, if a linear system has p inputs and q outputs, the transfer func-
tion between the j'th output and the Jth input is defined as

C (s)
t
(3-11)
G,As)
Rj(s)

with R k (s) = 0, k = l,2,...,p, k^j. Note that Eq. (3-11) is defined with
only they'th input in effect, while the other inputs are set to zero. The z'th output
transform of the system is related to all the input transforms by

Cls) = G n {s)R x
{s) +G i2 {s)R 2 {s) + ...+ G, p (s)R p (s)
(3-12)
= t Gu {s)Rj(s) (i=l,2,...,q)

where G tJ
(s) is defined in Eq. (3-11).
It is convenient to represent Eq. (3-12) by a matrix equation

C(s) = G(s)R(s) (3-13)

where
cm
C2 {s)
C(*) (3-14)

Cq{s)j
is a q X 1 matrix, called the transformed output vector;

Rii.s)

R(*) = (3-15)

RJis)}

is &p X 1 matrix, called the transformed input vector;

'G^is) G 12 (s) ... G lp (s)

G 2i (s) G 22 (s) G 2p {s)
G(s) (3-16)

G ql (s) G q2 (s) ... G„(s)_

is aq X p matrix, called the transfer function matrix.

 ec 3 3 '
'
Impulse Response of Linear Systems / 55

3.3 Impulse Response of Linear Systems

The impulse response of a linear system is defined as the output response

of the
system when the input is a unit impulse function. Therefore,
for a system with
a single input and a single output, if r(t) = d(t), the
Laplace transform of the
system output is simply the transfer function of the system, that is,

C(s) = G(s) (3-17)

since the Laplace transform of the unit impulse
function is unity.
Taking the inverse Laplace transform on both sides of Eq.
(3-17) yields

c(0 = 8(t) (3-18)

where g(t) is the inverse
Laplace transform of G(s) and is the impulse response
(sometimes also called the weighing function) of a linear
system. Therefore, we
can state that the Laplace transform of the impulse response
is the transfer function.
Since the transfer function is a powerful way of characterizing linear systems,
this means that if a linear system has zero initial conditions,
theoretically, the
system can be described or identified by exciting it with
a unit impulse response
and measuring the output. In practice, although a true
impulse cannot be
generated physically, a pulse with a very narrow
pulsewidth usually provides
a suitable approximation.
For a multivariable system, an impulse response matrix must
be defined and
is given by

g(0 = £-'[G(*)] (3-19)

where the inverse Laplace transform of G(s) implies the
transform operation
on each term of the matrix.
The derivation of G(s) in Eq. (3-3) is based on the knowledge of the system
differential equation, and the solution of C(s) from Eq. (3-3) also assumes that
R(s) and G(s) are all available in analytical forms. This
is not always possible
for quite often the input signal #(/) is not
Laplace transformable or is available
only in the form of experimental data. Under such
conditions, to analyze the
system we would have to work with the time function
r(t) and g(t).
Let us consider that the input signal r(j) shown in Fig.
3-2(a) is applied to a
linear system whose impulse response is g{t). The output response c(t)
is to be
determined. In this case we have denoted the input
signal as a function of r
which is the time variable; this is necessary since t is
reserved as a fixed time
quantity in the analysis. For all practical purposes, r(r) is assumed to extend
from minus infinity to plus infinity in time.
Now consider that the input r(r) is approximated by a sequence of pulses
of pulsewidth At, as shown in Fig. 3-2(b). In
the limit, as At approaches zero
these pulses become impulses, and the impulse at time kLx has a strength
or
area equal to At-z-^At), which is the area of the
pulse at kLx. Also, when At
decreases, k has to be increased proportionally,
so the value of &At remains
constant and equals t, which is a particular point
on the time axis. We now
compute the output response of the linear system, using
the impulse-approxi-
56 / Transfer Function and Signal Flow Graphs Chap. 3

r(T)

r(kAr)

(b)

Fig. 3-2. (a) Input signal of a linear system, (b) Input signal represented by
sum of rectangular pulses.

mated signal. When only the impulse at time x = kAx is considered, the system
response is given by
At • r(kAx)g(t - kAx) (3-20)

which is the system impulse response delayed by kAx, multiplied by the impulse
strength Ax-r(kAx). By use of the superposition principle, the total response
due to r(x) is obtained by adding up the responses due to each of the impulses
from — oo to +co. Therefore,

c{t) = lim £ r(kAx)g(t — kAx) Ax (3-21)

AT— oo
or

c{t) = ^_j(x)g(t-x)dx (3-22)

For all physical systems, output response does not precede excitation. Thus

g(t) = (3-23)

for / < 0, since the impulse function is applied at / = 0. Or

g (t - x) = t<z (3-24)
 Sec 3 3
Impulse Response of Linear Systems / 57

The output response of the system is now written

c (0 = /'__ r(T)g(t - T) <*t (3-25)

Further, if r( T ) = for T < 0, Eq. (3-25) becomes

C W = J' *<*)*(' - t) A
o
(3-26)

The expressions of Eqs. (3-25) and (3-26) are called the convolution integral
the convolution operation is denoted by the symbol *, so

. .
J
c(t) = r(t) * g(t)
(3. 27 )
is interpreted as

c(t) = r(t) convolves into g(t) (3-28)

The positions of r(t) and g{t) in the convolution operation may be inter-
changed, since basically there is no difference
between the two functions There-
fore, the convolution integral can
also be written as

c (0 = f'
g(i>(t - T ) dx

= git) * /(/) (3-29)

= git) convolves into /•(?)

The evaluation of the impulse response of linear

a system is sometimes an
important step in the analysis and design of a
class of systems known as the
adaptive control systems. In real life the
dynamic characteristics of most systems
vary to some extent over an extended
period of time. This may be caused by
simple deterioration of components due to
wear and tear, drift in operating
environments, and the like. Some systems
simply have parameters that vary
with time m
a predictable or unpredictable fashion. For
instance, the transfer
characteristic of a guided missile in flight
will vary in time because of the change
of mass of the nmsile and the change of
atmospheric conditions. On the other
hand, for a simple mechanical system with mass
and friction, the latter may be
subject to unpredictable variation either
due to "aging" or surface conditions
thus the control system designed under the
assumption of known and fixed
parameters may fail to yield satisfactory response
should the system parameters
vary. In order that the system may
have the ability of self-correction or self-
adjustment in accordance with varying
parameters and environment it is
necessary that the system's transfer characteristics
be identified continuously or
at appropriate intervals during the
operation of the system. One of the methods
of identification is to measure the impulse
response of the system so that design
parameters may be adjusted accordingly to attain
optimal control at all times
In the two preceding sections, definitions
of transfer function and impulse
response of a linear system have been presented.
The two functions are directly
related through the Laplace transformation,
and they represent essentially the
same information about the system. However,
it must be reiterated that
transfer
58 Transfer Function and Signal Flow Graphs Chap. 3
/

function and impulse response are denned only for linear systems and that the
initial conditions are assumed to be zero.

3.4 Block Diagrams 1

Because of its simplicity and versatility, block diagram is often used by control
engineers to portray systems of all types. A block diagram can be used simply to
represent the composition and interconnection of a system. Or, it can be used,
together with transfer functions, to represent the cause-and-effect relationships
throughout the system. For instance, the block diagram of Fig. 3-3 represents a
turbine-driven hydraulic power system for an aircraft. The main components of
the system include a pressure-compensated hydraulic pump, an air-driven pump,
an electronic speed controller, and a control valve. The block diagram in the
figure depicts how these components are interconnected.

Current
Controller

Turbine
Output
torque

Inlet
pressure
Control
valve
Mass
flow
Turbine
rO Load *

Hydraulic
pump
Pump torque

Load flow

Fig. 3-3. Block diagram of a turbine-driven hydraulic power system.

If the mathematical and functional relationships of all the system elements

are known, the block diagram can be used as a reference for the analytical or the
computer solution of the system. Furthermore, if all the system elements are
assumed to be linear, the transfer function for the overall system can be obtained
by means of block-diagram algebra.
The essential point is that block diagram can be used to portray nonlinear
as well as linear systems. For example, Fig. 3-4(a) shows the block diagram
of a simple control system which includes an amplifier and a motor. In the
figure the nonlinear characteristic of the amplifier is depicted by its nonlinear
 Sec. 3.4
Block Diagrams / 59

Kry

s(s+a)

(a)

(b)

Fig. 3-4. Block diagram of a simple control system, (a)

Amplifier shown
with a nonlinear gain characteristic, (b)
Amplifier shown with a linear gain
characteristic.

input-output relation. The motor is assumed to

be linear and its dynamics are
represented by a transfer function between
the input voltage and the output
displacement. Figure 3-4(b) illustrates the same
system but with the amplifier
characteristic approximated by a constant
gain. In this case the overall system
is linear, and it is now
possible to write the transfer function for
the overall
system as

E (s)
t Em (s) E{s) ~ s(s + a) (3
"3
°)

Block Diagrams of Control Systems

We shall now define some block-diagram elements used frequently

in con-
trol systems and the block-diagram algebra. One of the important
components
or a feedback control system is the sensing device that acts
as a junction point
for s lg nal comparisons. The physical components involved
are the potenti-
ometer, synchros, resolvers, differential amplifiers,
multipliers, and so on In
general, the operations of the sensing devices
are addition, subtraction mul-
tiplication, and sometimes combinations of these. The
block-diagram elements
of these operations are illustrated as shown in
Fig. 3-5. It should be pointed out
that the signals shown in the diagram of
Fig. 3-5 can be functions of time t
or
functions of the Laplace transform variable s.
In Fig. 3-4 we have already used block-diagram
elements to represent input-
output relationships of linear and nonlinear
elements. It simply shows that the
block-diagram notation can be used to represent
practically any input-output
relation as long as the relation is defined.
For instance, the block diagram of
60 / Transfer Function and Signal Flow Graphs Chap. 3

*- e = r — c +~ e = r +c

(a) Subtraction (b) Addition

*- e = rl + r2 — c +~- e =rc

(c) Addition and subtraction (d) Multiplication

Fig. 3-5. Block-diagram elements of typical sensing devices of control

systems, (a) Subtraction, (b) Addition, (c) Addition and subtraction, (d)
Multiplication.

u(t) x(t) u(f) x(0 R(s) C(s)

x =ax + bu x= f(x, u) G(s)

(a) (b) (c)

Fig. 3-6. Block-diagram representations of input-output relationships of

systems.

Fig. 3-6(a) represents a system that is described by the linear differential equa-
tion
x(t) = ax{t) + bu{t) (3-31)

Figure 3-6(b) illustrates the input-output relation of a system described by the

vector-matrix differential equation

±{t) = f [x(0, u«] (3-32)

where x(f) isan n X 1 vector and u(t) is an r X 1 vector. As another example,

Fig. 3-6(c) shows a block diagram which represents the transfer function of a
 :

Sec. 3.4
Block Diagrams / 61

linear system; that is,

C(s) = G{s)R(s) (3-33)

where G(s) is the transfer function.
Figure 3-7 shows the block diagram of a linear feedback control
system.
The following terminology often used in control systems is defined with reference
to the block diagram
/•(/)> R(s) = reference input
c(t), C(s) = output signal (controlled variable)
b(i), B(s) = feedback signal
e(t), &(s) = actuating signal
e(t), E(s) = - C(s) = error signal
R(s)

G(s) = g?^ = open-loop transfer function or forward-

path transfer function

M C(s)
= j±4 = closed-loop transfer function
(s)

H(s) = feedback-path transfer function

G(s)H(s) = loop transfer function

The closed-loop transfer function, M(s) = C(s)/R(s), can be

expressed as
a function of G(s) and H(s). From Fig. 3-7 we write

C(s) = G(s)&(s) (3-34)

and
B(s) = H(s)C(s) (3-35)
The actuating signal is written

6(j) = R(s) - B(s) (3-36)

Substituting Eq. (3-36) into Eq. (3-34) yields

C(s) = G(s)R(s) - G(s)B(s) (3-37)

Substituting Eq. (3-35) into Eq. (3-37) gives

C(s) = G(s)R(s) - G(s)H(s)C(s) (3-38)

Solving C(s) from the last equation, the closed-loop transfer
function of the

R(s)
/) ew C(s)
r(t)
' G(s)
.y eit)
c(t)

bit)
B(s)

H(s)

Fig. 3-7. Basic block diagram of a feedback control system.

62 / Transfer Function and Signal Flow Graphs Chap. 3

system is given by

In general, a practical control system

w- ^- i + Gis)His)
may contain many feedback loops, and
(3-39)

the evaluation of the transfer function from the block diagram by means of the
algebraic method described above may be tedious. In principle at least, the
block diagram of a system with one input and one output can always be reduced
to the basic single-loop form of Fig. 3-7. However, the steps involved in the
reduction process may again be quite involved. We shall show later that the

transfer function of any linear system can be obtained directly from its block
diagram by use of the signal-flow-graph gain formula.

Block Diagram and Transfer Function of Multivariable Systems

A is defined as one that has a multiple number of
multivariable system
inputs and outputs. Two
block-diagram representations of a multiple-variable
system with p inputs and q outputs are shown in Fig. 3-8(a) and (b). In Fig.
3-8(a) the individual input and output signals are designated, whereas in the
block diagram of Fig. 3-8(b), the multiplicity of the inputs and outputs is
denoted by vectors. The case of Fig. 3-8(b) is preferable in practice because of its
simplicity.

MO- -»- c l (t)

Multivariable
system

'
P (0- -*~c q (t)

(a)

Multivariable
i(0" -*- c(/)
system

(b)

Fig. 3-8. Block-diagram representations of a multivariable system.

Figure 3-9 shows the block diagram of a multivariable feedback control

system. The transfer function relationship between the input and the output of
the system is obtained by using matrix algebra

C(j) = G(s)£(s) (3-40)

&is) = R(s) - Bis) (3-41)

B(s) = H(j)C(j) (3-42)

 Sec. 3.4
Block Diagrams / 63

R(s)
S) 80) C&)

^
G(s)

Bfr)

Hfc)

Fig. 3-9. Block diagram of a multivariable feedback control system.

Substituting Eq. (3-42) into Eq. (3-41) and then from

Eq. (3-41) into Eq (3-40)
yields

C(s) = G(s)R(s) - G(s)H(s)C(s) (3-43)

Solving for C(s) from Eq. (3-43) gives

C(s) = [I + G(j)H(j)]- G(s)R(s) »

(3-44)

provided that I +
G(5 )H(j) is nonsingular.
should be mentioned that although the development of the
It
input-output
relationship here is similar to that of the single input-output
case, in the present
situation improper to speak of the ratio C(s)/R(s) since C(s) and R(s)
it is
are
matrices. However, it is still possible to define the closed-loop
transfer matrix as

M(j) = [I + G(*)H(s)]-»G(j) (3-45)

Then Eq. (3-44) is written

C(j) = M(a)R(j) (3-46)

Example 3-2 Consider that the forward-path transfer function matrix

and the
feedback-path transfer function matrix of the system shown
in Fig.
3-9 are

r l _j_i
s + l s
G(s,= (3-47)
1
2
2J
and
1
0"
H(s) =
Lo i_
respectively.
The closed-loop transfer matrix of the system is given by Eq. (3-45) and is evalu-
ated as follows:

1
1 1_

= s+ 1 s s + l s
I + G(s)H(s)
(3-48)
2
1 s+j,
1 2
'
s + 2J s + 2J
64 / Transfer Function and Signal Flow Graphs Chap. 3

The closed-loop transfer matrix is

5 + 3 _1_
5 + 2 s 5 ! 1
M(a) = [I + G( S )H(s)]-'GM = -^ (3-49)
-2 5 + 2
5 + 1 5 + 2J

where
+2s+3+
A _ sS+1S|2 , 2 __ s 2 + 5s + 2
(3-50)
S 5(5+1)
Thus
•
35 2 +95 + 4 _ ~5 1_

= s(s + 1) s{s + 1)(5 + 2)

M(5) (3-51)
52 + 55 + 2 35 + 2
5(5 + 1).

3.5 Signal Flow Graphs 2

A signal flow graph may be regarded as a simplified notation for a block dia-
gram, although it was originally introduced by S. J. Mason- as a cause-and-effect
representation of linear systems. In general, besides the difference in the physical
appearances of the signal flow graph and the block diagram, we may regard
more rigid mathematical relationships,
the signal flow graph to be constrained by
whereas the rules of using the block-diagram notation are far more flexible and
less stringent.
A signal flow graph may be defined as a graphical means of portraying the
input-output relationships between the variables of a set of linear algebraic
equations.
Consider that a linear system is described by the set of N algebraic equations

yj = S o kJ y k j = 1,2, . . . ,N (3-52)

It should be pointed out that these N equations are written in the form of cause-
and-effect relations
N
jth effect = XI (§ am fr° m ^ to ./)(^ tri cause) (3-53)
k=l
or simply
output = 2 (gain)(input) (3-54)

This is the single most important axiom in the construction of the set of algebriac
equations from which a signal flow graph is drawn.
In the case when a system is represented by a set of integrodifferential
equations, we must first transform them into Laplace transform equations and
then rearrange the latter into the form of Eq. (3-52), or

W = S G kj (s)Yk (s) 7=1,2, N (3-55)

When constructing a signal flow graph, junction points or nodes are used
ys and y k The nodes are connected together by line
to represent the variables .

segments called branches, according to the cause-and-effect equations. The

 Sec 3 5
Signal Flow Graphs / 65

branches have associated branch gains and

directions. A signal can transmit
through a branch only in the direction of the arrow.
Ingeneral, given a set of
equations such as those of Eq. (3-52) or Eq.
(3-55), the construction of the
signal flow graph is basically a matter
of following through the cause-and-
effect relations relating
each variable in terms of itself and the other
variables
For instance, consider that a linear system is represented
by the simple equation

where j, is the input variable,

y 2 the output variable, and a l2 the gain or trans-
mittance between the two variables. The
signal-flow-graph representation of
Eq. (3-56) is shown in Fig. 3-10. Notice that
the branch
°n directing from node y to node y 2 expresses the depend-
x
° * O ence of y 2 upon should be reiterated that Eq. (3-56)
>»,. It

and Fig. 3-10 represent only the dependence

of the out-
Fig. 3-10. Signal flow graph of
y2 P ut variable upon the input variable, not the reverse.
= ° 12 >''- An important consideration in the application of
signal
flow graphs that the branch between the two nodes
is
y,
and y 2 should be integrated as a unilateral amplifier
with gain a i2 , so that when
a signal of one unit is applied at the input
y u the signal is multiplied by a l2 and
a signal of strength a l2 is delivered at node>>
2 Although algebraically Eq (3-56)
.

can be rewritten

the signal flow graph of Fig. 3-10 does

not imply this relationship If Eq (3-57)
isvalid as a cause-and-effect equation in
the physical sense, a new signal flow
graph must be drawn.
As another illustrative example, consider the following
set of algebraic
equations

yi = tf12.F1 + a 32 y 3

yi = a 23 y 2 + a43 y 4

(3 " 58)
}'* = Oi*yi + a 34 y + 3 a 44 y 4

ys = a 2s y 2 +a 4 <,y 4

The signal flow

graph for these equations is constructed step
by step as shown
in Fig. 3-11, although the indicated
sequence of steps is not unique The nodes
representing the variables
y u y2 y 3 y 4 and y s are located in order from left to
, , ,

right. The equation states that 2 depends upon two

first
y signals, a liyi and
<hiy 3 the signal flow graph representing this
;
equation is
drawn as shown in
Fig. 3-1 1(a). The second equation states that
y 3 depends upon a 23 y 2 and a 43y 4
therefore on the signal flow graph of Fig. 3-1
1(a), a branch of gain a
23 is drawn
trom node y 2 to y 3 and a branch of gain « is
,
43 drawn from y t to y 3 with the
directions of the branches indicated by the
arrows, as shown in Fig 3-1 1(b)
Similarly, with the consideration of the
third equation, Fig. 3-1 1(c) is obtained
Finally, when the last equation of Eq. (3-58) is portrayed,
the complete signal
flow graph is shown in Fig. 3-1 1(d). The
branch that begins from the node
y4
66 / Transfer Function and Signal Flow Graphs Chap. 3

o O
>4 y$

(a)y 2 =a n y + a 31 y3
l

a 43

o
y\ yi ^3 y* >"5

(b)>>2 =a n y\ +032^3. ^3 =fl23> 2

, + «43>'4

O
^5

(c)y 2 =012^1 +232>'3.>'3 = a 23>'2 + 043^4. ^ =a

24>
,
2 + a 34> ,
3 + fl 44>"4

(d) Complete signal flow graph

Fig. 3-11. Step-by-step construction of the signal flow graph for Eq.
(3-58). (a) y2 = ai 2 yi + a 3 zy3- (b) yi = anyi + fl32>'3, yi = miyi
+ any*, (c) yz = a\zy\ + a 3 zy3, yi = anyi + any*, y* = auyi
+ a t,yi + cmy»,. (d) Complete signal flow graph.
3

and ends at y 4 is called a loop, and with a gain a 44 represents the dependence
,

of y 4 upon itself.

3.6 Summary of Basic Properties of Signal Flow Graphs

At this point it is best to summarize some of the important properties of the

signal flow graph.

1. A signal flow graph applies only to linear systems.

2. The equations based on which a signal flow graph is drawn must be
algebraic equations in the form of effects as functions of causes.
3. Nodes are used to represent variables. Normally, the nodes are
 Sec. 3.7
Definitions for Signal Flow Graphs / 67

arranged from left to right, following a

succession of causes and
effectsthrough the system.
4. Signals travel along branches only in
the direction described by the
arrows of the branches.
5. The branch directing from node
y k to j, represents the dependence
of the variable y, upon but not
yk , the reverse.
6. A signal y k traveling along a branch between nodes
yk and y. is
multiplied by the gain of the branch, a
delivered at node y,.
kj so that a signal
, aj k is

3.7 Definitions for Signal Flow Graphs

In addition to the branches and nodes

defined earlier for the signal flow graph
the following terms are useful for the
purposes of identification and reference.

Input node {source). An input node is a node that has only outgoing
branches. (Example: node
yi in Fig. 3-11.)

Output node (sink). An output node is a node

which has only incoming
branches. (Example: node
y 5 in Fig. 3-11.) However, this condition is not always
r eadiy
p
Mg. i ^/
b an out P ut node For instance, the signal flow
3-12(a) does not have
-

any node that satisfies

graph shown in
the condition of an output
node. However, it may be necessary to
regard nodes y 2 and/or 3 as output
nodes In order to meet the definition
y
requirement, we may simply introduce
r eS lth UDity g3inS and additional variables
, f£S
3-12(b). J
Notice that in the modified signal flow
j 2 and y 3 as shown
graph it is
, in Fig.
equivalent that the

(a) Original signal flow graph

(b) Modified signal flow graph

Fig. 3-12. Modification of a signal flow

graph so that y 2 and y z satisfy the
requirement as output nodes, (a) Original signal
flow graph, (b) Modified
signal flow graph.
 Function and Signal Flow Graphs Chap. 3
68 / Transfer

equations y 2 =y 2 and y 3 =y
are added. In general, we can state that any
3

noninput node graph

of a signal flow can always be made an output node by
the aforementioned operation. However, we cannot convert a noninput node
into an input node by using a similar operation. For instance, node y 2 of the
signal flow graph of Fig. 3-12(a) does not satisfy the definition of an input node.
If we attempt to convert it into an input node by adding an incoming branch of
unity gain from another identical node y 2 the signal flow graph of Fig. 3-13
,

would result. However, the equation that portrays the relationship at node y 2
now reads
yi = yi + <*\%y\ + a nyi (3- 59 )

which is different from the original equation, as written from Fig. 3- 12(a),

yi = a i2 yi + a 32 y 3 (3-60)

9 Vi

Fig. 3-13. Erroneous way to make the node y 2 Fig. 3-14. Signal flow graph with y 2 as an input
an input node. node.

Since the only proper way that a signal flow graph can be drawn is from a
set of cause-and-effect equations, that is, with the causes on the right side of the
equation and the effects on the left side of the equation, we must transfer y 2
if it were to be an input. Rearranging Eq. (3-60),
to the right side of Eq. (3-60)
the two equations originally for the signal flow graph of Fig. 3-12 now become

1
y% -—y 3
(3-61)
«12 "12

y3 = a 2i y 2 (3-62)

The signal flow graph for these two equations is shown in Fig. 3-14, with y2
as an input node.

Path. A any collection of continuous succession of branches

path is

traversed in the same The definition of a path is entirely general since

direction.
it does not prevent any node to be traversed more than once. Therefore, as
simple as the signal flow graph of Fig. 3-12(a) is, it may have numerous paths.

Forward path. A forward path is a path that starts at an input node and
ends at an output node and along which no node is traversed more than once.
For example, in the signal flow graph of Fig. 3-1 1(d), y is the input node, and t

there are four possible output nodes in y 2 y 3 j 4 and y s The forward path
, , , .
 Sec. 3.8 Signal-Flow-Graph Algebra / 69

between j t and y 2 is simply the branch connected between y and y 2 There are t
.

two forward paths between y L and y 3 one contains the branches from y t to y 2
;

to y 3 and the other one contains the branches from y to y 2 to y 4 (through the
,
t

branch with gain a 2i ) and then back to y 3 (through the branch with gain a 43 ).
The reader may determine the two forward paths between j>, and j 4 . Similarly,
there are also two forward paths between j, and y s .

Loop. A loop is a path that originates and terminates on the same node
and along which no other node is encountered more than once. For example,
there are four loops in the signal flow graph of Fig. 3-1 1(d). These are shown
in Fig. 3-15.

"44

o
y*

Fig. 3-15. Four loops in the signal flow graph of Fig. 3-1 1(d).

Path gain. The product of the branch gains encountered in traversing a

path is called the path gain. For example, the path gain for the path
J>i
— yz - y - y* 3 in Fig. 3-1 1(d) is a 12 a 23 a 34 .

Forward-path gain. Forward-path gain is defined as the path gain of a

forward path.

Loop gain. Loop gain is defined as the path gain of a loop. For example,
the loop gain of the loop y2 — y4 — y — y 3 2 in Fig. 3-15 is a 2i a i3 a 32 .

3.8 Signal-Flow-Graph Algebra

Based on the properties of the signal flow graph, we can state the following
manipulation and algebra of the signal flow graph.

1 . The value of the variable represented by a node is equal to the sum

of all the signals entering the node. Therefore, for the signal flow
70 / Transfer Function and Signal Flow Graphs Chap. 3

Fig. 3-16. Node as a summing point and as a transmitting point.

graph of Fig. 3-16, the value of y is equal to the sum of the signals
{

transmitted through all the incoming branches; that is,

Jl = "li}>2 + «31^3 + «4lj4 + Osijs (3-63)

2. The value of the variable represented by a node is transmitted

through all branches leaving the node. In the signal flow graph of
Fig. 3-16, we have
ye = tfi6.Fi

yn = a xl y x
(3-64)

y% = tfisJi

3. Parallel branches in the same direction connected between two

nodes can be replaced by branch with gain equal to the sum
a single
of the gains of the parallel branches. An example of this case is
illustrated in Fig. 3-17.

-X y\
a-vb +
>
c
X.
yi

Fig. 3-17. Signal flow graph with parallel paths replaced by one with a
single branch.
 Sec. 3.9 Examples of the Construction of Signal Flow Graphs / 71

a 12 a 23 "34 a
<*45
45 "56
O O » O o— » O l»
V\ vi y3 y4 ys y6

a 12<*23tf34 fl 45fl56
o » o

Fig. 3-18. Signal flow graph with cascaded unidirectional branches re-
placed by a single branch.

4. A connection of unidirectional branches, as shown in Fig.

series
3-18, can be replaced by a single branch with gain equal to the
product of the branch gains.
5. Signal flow graph of a feedback control system. Figure 3-19 shows
the signal flow graph of a feedback control system whose block
diagram is may be
given in Fig. 3-7. Therefore, the signal flow graph
regarded as a simplified notation for the block diagram. Writing the
equations for the signals at the nodes &(s) and C(s), we have

S(j) = R(s) - H(s)C(s) (3-65)

and
C(s) = G(s)&(s) (3-66)

The closed-loop transfer function is obtained from these two

equations,
C(s) _ G(s)
(3-67)
R(s) 1 + G(s)H(s)

R(s) e(s)\_ ^y <Xs) C(s)

-ms)
Fig. 3-19. Signal flow graph of a feedback control system.

For complex signal flow graphs we do not need to rely on algebraic manipu-
lation to determine the input-output relation. In Section 3.10 a general gain
formula will be introduced which allows the determination of the gain between
an input node and an output node by mere inspection.

3.9 Examples of the Construction of Signal Flow Graphs

It was emphasized earlier that the construction of a signal flow graph of a phys-
ical system depends upon first writing the equations of the system in the cause-
and-effect form. In this section we shall give two simple illustrative examples.
72 / Transfer Function and Signal Flow Graphs Chap. 3

Owing to the lack of background on systems at this early stage, we are using two
electric networks as examples. More elaborate cases will be discussed in Chapter
5, where the modeling of systems is formally covered.

Example 3-3 The passive network shown in Fig. 3-20(a) is considered to consist of
R, L, and C elements so that the network elements can be represented
by impedance functions, Z(s), and admittance functions, Y(s). The
Laplace transform of the input voltage is denoted by Ein (s) and that of the output
voltage is EJjs). In this case it is more convenient to use the branch currents and node
voltages designated as shown in Fig. 3-20(a). Then one set of independent equations
representing cause-and-effect relation is

Ii(.s) = [E Js)-E (s)]Y S

i 1 1 ( ) (3-68)

E2 (s) = [h{s) - /,(*)]Z,(j) (3-69)

/,(,) = [E1 {s) - Ea (s)]Y (s) 3 (3-70)

Ea (s) = Z«(j)/ (j) s (3-71)

YAs) Y3 (s)

£|n« E (s)

(a)

£•„(*)

YAs) Z2 (s) Y3 (s)

Fig. 3-20. (a) Passive ladder network, (b) A signal flow graph for the net-
work.

With the variables Ein(s), /,($), E2 (s), I

3 (s), and E„(s) arranged from left to right

in order, the signal flowgraph of the network is constructed as shown in Fig. 3-20(b).
It is noteworthy that in the case of network analysis, the cause-and-effect equa-

tions that are most convenient for the construction of a signal flow graph are neither the
loop equations nor the node equations. Of course, this does not mean that we cannot
construct a signal flow graph using the loop or the node equations. For instance, in
Fig. 3-20(a), if we let Ii(s) and I3 (s) be the loop currents of the two loops, the loop
equations are
 Sec. 3.9 Examples of the Construction of Signal Flow Graphs
/ 73

Etn(s) = [Zi(s) + Z2 (j)]/,(j) - Z (s)I (s) 2 3 (3-72)

= -Z Gs)/,0) + [Z (s) + Z (s) + Z4 (s)]I
2 2 3 3 (s) (3-73)
E (s) = Zt(s)I (s) 3
(3 . 74)
However, Eqs. (3-72) and (3-73) should be rearranged, since
only effect variables can
appear on the left-hand sides of the equations. Therefore,
solving for /,0s) from Ea
(3-72) and I3 (s) from Eq. (3-73), we get

71
« =Z i(s) I Z2 (s) E ^ + zjfzjs) ^ 1 (3-75)

h(s) = W
Z X {?) + z\(s) + Z4 (*) 7 '
(3
- 76
>

Now, Eqs. (3-74), (3-75), and (3-76) are in the form of

cause-and-effect equations. The
signal flow graph portraying these equations
is drawn as shown in Fig 3-21 This
exercise also illustrates that the signal flow
graph of a system is not unique.

1 Z2 (s)
Z 1
(s) + Z 2 (s) Z2 fc) + Z3 (s) + Z4 (s)
O »
E (s)

Z2 (s)
Z 1
(s) + Z 2 (s)

Fig. 3-21. Signal flow graph of the network

in Fig. 3-20(a) using the loop
equations as a starting point.

Example 3-4 Let us consider the RLC

network shown in Fig. 3-22(a). We shall
define the current /(f)
and the voltage e£t) as the dependent variables
of the network. Writing the voltage across the
inductance and the
current in the capacitor, we have the following differential
equations:

L
^W =eM ~ Ri W ~ ecO (3-77)

C dec (t) =
r .

~1T '<'> (3-78)

However, we cannot construct a signal flow graph using these two equations
since they
are differential equations. In order to arrive
at algebraic equations, we divide Eqs
(3-77) and (3-78) by L and C, respectively. When we take the Laplace transform, we
have

sl(s) = i(0+) + j-Ets) - j-m - j-E (s) c (3-79)

sEc (s)=ec (0+)+-Ll(s )

(3 _ 80)

where i(0+) is the initial current and e (0+) is the initial

c voltage at t = 0+ In these
last two equations, * e (0+), i(0+), and E
(s) are the input variables. There are
t
several
possible ways of constructing the signal flow graph
for these equations. One way is to
solve for I(s) from Eq. (3-79) and £„(*) from Eq.
(3-80); we get
74 / Transfer Function and Signal Flow Graphs Chap. 3

R L
1 WW- nfw* —
+
no
»i(')Q
) c = e c (t)

(a)

EiU) EJs)

LU + R/L)

(c)

Fig. 3-22. (a) RLC network, (b) Signal flow graph, (c) Alternative signal
flow graph.

'« = 7tW (0+) + as +W)] £lW - WTTRim EM (3 " 81)

E (s) =
c y^c(0+) + -gr/O) (3-82)

The signal flowgraph using the last equations is drawn as shown in Fig. 3-22(b).
The graph in Fig. 3-22(b) is of analytical value only. In other words,
signal flow
we can solve for I(s) and Ec (s) from the signal flow graph in terms of the inputs, e c (0+),
j'(0+), and Ei(s), but the value of the signal flow graph would probably end here. As an
alternative, we can use Eqs. (3-79) and (3-80) directly, and define I(s), E (s), sl(s), and
c

sEc (s) as the noninput variables. These four variables are related by the equations
 Sec 3.10 General Gain rormula for Signal Flow Graphs / 75

/(*) =5- »[*/(*)] (3-83)

E (s) = s-^sEds)]
c (3-84)

The significance of using s~ l

is that it represents pure integration in the time domain.

Now, a signal flow graph using Eqs. (3-79), (3-80), (3-83), and (3-84) is constructed as
shown in Fig. 3-22(c). Notice that in this signal flow graph the Laplace transform
variable appears only in the form of s' 1 Therefore, this signal flow graph may be used
.

as a basis for analog or digital computer solution of the problem. Signal flow graphs
in this form are defined in Chapter 4 as the state diagrams. 5

3.10 General Gain Formula for Signal Flow Graphs 3

Given a graph or a block diagram, it is usually a tedious task to

signal flow
solve for input-output relationships by analytical means. Fortunately, there
its

is a general gain formula available which allows the determination of the input-
output relationship of a signal flow graph by mere inspection. The general gain
formula is

V Mj^k
M = £na = *=1 (3-85)'
A Jin

where
M = gain between y in and yout
yout = output node variable
jln = input node variable
N = total number of forward paths
Mk = gain of the kth. forward path
A = - S Pmi + £ Pm ~ S Pmi +
1 i • • • (3-86)
m m m
Pmr = gain product of the mth possible combination of r
nontouching* loops
or
A= 1 — (sum of all individual loop gains) + (sum of
gain products of all possible combinations of two
nontouching loops) — (sum of the gain products
of all possible combinations of three nontouching (3-87)
loops) + . . .

A* = the A for that part of the signal flow graph which

is nontouching with the fcth forward path

This general formula may seem formidable to use at first glance. However,
the only complicated term in the gain formula is A; but in practice, systems
having a large number of nontouching loops are rare. An error that is frequently
made with regard to the gain formula is the condition under which it is valid.
It must be emphasized that the gain formula can be applied only between an

input node and an output node.

*Two parts of a signal flow graph are said to be nontouching if they do not share a com-
mon node.
76 / Transfer Function and Signal Flow Graphs Chap. 3

Example 3-5 Consider the signal flow graph of Fig. 3-19. We wish to find the
transfer function C(s)/R(s) by use of the gain formula, Eq. (3-85).
The following conclusions are obtained by inspection from the signal
flow graph

1. There is only one forward path between R(s) and C(s), and the forward-
path gain is
M, = G(s) (3-88)

2. There is only one loop; the loop gain is

Pu = -G(s)H(s) (3-89)

3. There are no nontouching loops since there is only one loop. Further-
more, the forward path is in touch with the only loop. Thus Ai = 1, and
A= 1 -/»,,= 1 + G(j)#(j).

By use of Eq. (3-85), the transfer function of the system is obtained as

C(s)
= Mi At = G(s)
(390)
R(s) A 1 + G(s)H(s)

which agrees with the result obtained in Eq. (3-67).

Example 3-6 Consider, in Fig. 3-20(b) that the functional relation between Eia and
E is to be determined by use of the general gain formula. The signal
flow graph is redrawn in Fig. 3-23(a). The following conclusions
are obtained by inspection from the signal flow graph :

1. There is only one forward path between Eia and E , as shown in Fig.

3-23(b). The forward-path gain is

M, = FiZ2 y3 Z4 (3-91)

2. There are three individual loops, as shown in Fig. 3-23(c); the loop gains
are
P n = -Z2 y, (3-92)

21 = -Z2 r3
/> (3-93)

i> 31 = -Z4 r 3 (3-94)

3. There is one pair of nontouching loops, as shown in Fig. 3-23(d); the loop
gains of these two loops are

-Z Y 2 { and -Z4 Y 3

Thus
Pu = product of gains of the first (and only) possible (3-95)
combination of two nontouching loops -= Z 2 Z4 Yx Y }

4. There are no three nontouching loops, four nontouching loops, and so

on; thus
Pm3 =0, Pm4 = 0, . .

From Eq. (3-86),

A= l -(J»„ +P +P )+P
zl 3l l2
(3-96)
= 1 + Z Y + Z Y + Z4 Y + Z Z4 Yt Y
2 { 2 3 3 2 3
 Sec. 3.10 General Gain Formula for Signal Flow Graphs / 77

(a)

O Oh
(b)

(d)

Fig. 3-23. (a) Signal flow graph of the passive network in Fig. 3-20(a). (b)
Forward path between Ein and Ea (c) Three individual loops, (d) Two
.

nontouching loops.

5. All the three feedback loops are in touch with the forward path ; thus

A, = 1 (3-97)

Substituting the quantities in Eqs. (3-91) through (3-97) into Eq. (3-85), we obtain

M x Ai y, r3 z2 z4 (3-98)
1 +Z 1 Y l +Z 2 Y + Z<Y + Z1 Zt Y Y
3 2 1 3

Example 3-7 Consider the signal flow graph of Fig. 3-22(c). It is desired to find the
relationships between /and the three inputs, E lt i(0+), and e c (fl+).
Similar relationship is desired for Ec . Since the system is linear, the
principle of superposition applies. The gain between one input and one output is deter-
mined by applying the gain formula to the two variables while setting the rest of the
inputs to zero.
The signal flow graph is redrawn as shown in Fig. 3-24(a). Let us first consider /

as the output variable. The forward paths between each inputs and / are shown in
Fig. 3-24(b), (c), and (d), respectively.
78 / Transfer Function and Signal Flow Graphs Chap. 3

Q »(0+) O ec (0+)

<(0 +)

" 1

o- -O
sl /

(c)

Fig. 3-24. (a) Signal flow graph of the RLC network in Fig. 3-22(a). (b)
Forward path between Ei and /. (c) Forward path between /(0+) and /.

(d) Forward path between e c (0+) and /.

Sec. 3.10 General Gain Formula for Signal Flow Graphs / 79

The signal flow graph has two loops the ; A is given by

A= l +#*-' 1

L ° + LC '
(3-99)

All the forward paths are in touch with the two loops; thus A, = 1 for all cases
Considering each input separately, we have

/(0+) = 0, * c (0+) = (3-100)

£, A
£-, = 0, e c (0+) = (3-101)
'(0+)
<HL)s- = 0,
i(0+) E = (3-102)
ec (0+) A x

When all three inputs are applied simultaneously, we write

(3-103)

In a similar fashion, the reader should verify that when E c is considered as the
output variable, we have

Lc j^^E, + ± s -*i(p+) + s-i/l + |r'W+) (3-104)

A
Notice that the loop between the nodes si and / is not in touch with the forward path
between e c (0+) and Ec .

Example 3-8 Consider the signal flow graph of Fig. 3-25. The following input-
output relations are obtained by use of the general gain formula:

yi + dy
a(i
(3-105)
>-i A
y_3 _ ag(l + d) + abc
(3-106)
yi A
where
A= l +eg + d + bcg + deg (3-107)

Fig. 3-25. Signal flow graph for Example 3-8.

80 / Transfer Function and Signal Flow Graphs Chap. 3

3.11 Application of the General Gain Formula to Block Diagrams

Because of the similarity between the block diagram and the signal flow graph,
the general gain formula in Eq. (3-85) can be used to determine the input-output
relationships of either. In general, given a block diagram of a linear system we
can apply the gain formula directly to it. However, in order to be able to identify
all the loops and nontouching parts clearly, sometimes it may be helpful if an

equivalent signal flow graph is drawn for a block diagram before applying the
gain formula.
To illustrate how the signal flow graph and the block diagram are related,
the equivalent models of a control system are shown Note that
in Fig. 3-26.
summing point of all

^
since a node on the signal flow graph is interpreted as a

(a)

(b)

Fig. 3-26. (a) Block diagram of a control system, (b) Equivalent signal
flow graph.
 Sec. 3.12 Transfer Functions of Discrete-Data Systems / 81

incoming signals to the node, the negative feedback paths in this case are
represented by assigning negative gains to the feedback paths.
The closed-loop transfer function of the system is obtained by applying
Eq. (3-85) to either the block diagram or the signal flow graph

C(s) G G2 G
i 3
-j- G t
Gj
(3-108)
R(s) 1 + G G H + G G H + G G G + G H2 +
l 1 1 2 3 2 X 2 3 4 G,G 4
Similarly,

E(s) __..
! + G,(? 2 ff, + G G H + G4 H
2 3 2 2
(3-109)
R(s) A
Y 3 (s) _
~ 1 + G G H + G,H
2 3 2 2
(3-110)
R(s) A
where
A= 1 + G^H, + G G H + 2 3 2 G,G 2 C 3 + G4 H + G Gi
2 1
(3-111)

3.12 Transfer Functions of Discrete-Data Systems 7 - 8

It isshown in Chapter 2 that the signals in a discrete-data or sampled-data

system are in the form of pulse trains. Therefore, the Laplace transform and the
transfer functions defined for continuous-data systems, in the ^-domain, cannot
be used adequately to describe these systems.
Figure 3-27(a) illustrates a linear system with transfer function G(s) whose
input is the output of a finite-pulsewidth sampler. As described in Section 2.8,
the finite-pulsewidth sampler closes for a short duration of p seconds once
every T seconds, and a typical set of input and output signals of the sampler is
shown in Fig. 2-5. Since for a very small pulse duration p, as compared with the
sampling period T, the finite-pulsewidth sampler can be approximated by an
ideal sampler connected in cascade with a constant attenuation p, the system
of Fig. 3-27(a) may be approximated by the system shown in Fig. 3-27(b).

r{t) r *0) c(t)

R(s)
-? (P)
p
R?(s)
G(s)
C(s)

(a)

KO c(0
R(s)
-* r*(t)
'
>-*<,t)

* P* (s)
G(s) —-
C(s)
Ideal sampler

(b)

i^iaiiGiG-uata aysicm
Fig. 3-27. (a) Discrete-data system with
wiui a iiuiic-puiscwiuin sampl
finite-pulsewidth sampler, (b)
Discrete-data system with an ideal sampler that approximates the system
:

in (a).
82 / Transfer Function and Signal Flow Graphs Chap. 3

d
,y\
2
c*(t)
_^_
|" C*(s)
T

I
c(t)
rit) >• /•*(/)

R(s)
<* '
R*(s)
G(s)
C{s)
T

Fig. 3-28. Discrete-data system with an ideal sampler.

Normally, for convenience, it is assumed that the attenuation factor p is

included in the transfer function of the process, G(s). Therefore, the block
diagram of Fig. 3-28 is considered as that of a typical open-loop discrete-data
or sampled-data system.
There are several ways of deriving the transfer function representation of
the system of Fig. 3-28. In the following we shall show two different representa-
tions of the transfer function of the system. Let us assume that r*(t), the output
of the ideal sampler 5",, is a unit impulse function. This may be obtained by
sampling a unit step function u,(t) once at t = or if r(t) is a unit impulse func-
tion.* Unless stated otherwise, the samplers considered in the remaining portion
of this text are ideal samplers.
The output of G{s) is the impulse g{i). If a fictitious ideal sampler S2 which
,

issynchronized with Sj and has the same sampling period as that of 5,, is placed
at the output of the system as shown in Fig. 3-28, the output of the sampler S z
may be written as

c*(t) = g*(t) = t g(kT)S(t ~ kT)

A=
(3-112)

where c(kT) = g(kT) is defined as the weighting sequence of the linear process
G(s). In other words, the sampled version of the impulse response or weighting
function is the weighting sequence.
Taking the Laplace transform on both sides of Eq. (3-112) yields

C*(s) = G*(s) = £[g*(f)]

(3-113)
= S g(kT)e-*T-

which is defined as the pulse transfer function of the linear process.

At this point we can summarize our findings about the description of the
discrete-data system of Fig. 3-28 as follows. When a unit impulse function is

applied to the linear process, the output is simply the impulse response of the process ;
the impulse response is sampled by a fictitious ideal sampler S2 and the output of ,

the sampler is the weighting sequence of the process. The Laplace transform of the
weighting sequence impulse train gives the pulse transfer function G*(s).

Although from a mathematical standpoint the meaning of sampling an impulse function

is questionable and difficult to define, physically, we may argue that sending a pulse through
a finite-pulsewidth sampler will retain the same identity of the pulse.
Sec. 3.12 Transfer Functions of Discrete-Data Systems / 83

Once the weighting sequence of a linear system is defined, the output of

the system, c(t),and the sampled output, c*(t), which is due to any arbitrary
input r(t), can be obtained by means of the principle of superposition.
Consider that an arbitrary input r(t) is applied to the system of Fig. 3-28
at t = 0. The output of the ideal sampler is the impulse train,

r*(f) = £ r(kT)S(t - kT) (3-114)

A=

By means of superposition, the output of the process, which is due to r*(t), is

c(t) = r(0)g(t) + r(T)g(t - T) + . . . + r(kT)g(t - kT) + . . . (3-115)

At t = kT, the last equation becomes

c(kT) = r(0)g(kT) + r{T)g[{k - 1)7] + . . . + r[(k - l)T]g(T) + r(kT)g(0)

(3-116)

where it is assumed that g(t) is zero for all t < 0, since the process is a physical
system so that its output does not precede the input.
Multiplying both sides of Eq. (3-116) by e~ kTs and taking the summation ,

from k = to k = oo, we have

£ c(kT)e- kT ° = £ r(0)g(kT)e- kT ° + ± r(T)g[(k - l)T]e~ kT ° + ...

(3-117)
+ £ r[{k - l)T]g(T)e-^ £ r(kT)g(0)e^
+ *=0
<t =

Again, using the fact that g(t) is zero for negative time, Eq. (3- 1 1 7) is simplified to

2 c(kT)e- kT < = [r(O) + r (T)e~ T ° + r{2T)e~™> + ...]£ g{kT)e~ kTs (3-118)

*=

or

£ c(kT)e-« Ts = £ r{kT)e- kT * £ g(kT) e - kT >

(3-119)

Therefore, using the definition of the pulse transfer function, the last equation
is written

C*(s) = R*(s)G*(s) (3-120)

which is the input-output transfer relationship of the discrete-data system

shown in Fig. 3-28. The z-transform relationship is obtained directly from the
definition of the z-transform. Since z = e
Ts
, Eq. (3-119) is also written

£ c(kT)z- k = £ r{kT)z' k £ g{kT)z~ k (3-121)

Therefore, defining the z-transfer function of the process as

G(z) = £ g{kT)z' k (3-122)

k=0

which implies that the z-transfer function of a linear system, C(z), is the z-
transform of the weighting sequence, gikT), of the system. Equation (3-121)
is written
C(z) = R{z)G(z) (3-123)
 „ u Chap. 3
84 / Transfer Function and Signal Flow Graphs

discrete-data system is
important to point out that the output of the
It is
However, the pulse transform of the output,
continuous with respect to time.
output, C(z), specify the values of c{t) only
C*(s) and the z-transform of the
well-behaved function between sampling
at the sampling instants. If c(t) is a
description of the true output:c{t).
instants c*(0 or C(z) may give an accurate
sampling instants, the z-
However if c(t) has wide fluctuations between the
only at the sampling instants, will
transform method, which gives information
yield misleading or inaccurate results.
can also be obtained by use of
The pulse transfer relation of Eq. (3-120)
C(s), which is given in the literature
the following relation between C*(s) and
:

C*(*) =i S C(s+jnco s) O 124 )

where co s isthe sampling frequency in radians per

second and ca s 2njT. =
continuous-data output c(t) is
From Fig. 3-28, the Laplace transform of the

C(s ) = G(s)R*(s) (3-125)

Substituting Eq. (3-125) into Eq. (3-124) gives

C*(s) = ± £ G(s + jnaJR^s + jnw,) (3-126)

We can write
R*(s + jnco ) = k=0
J r(kT)e-"^
s
+ ^
(3-127)

= E r(kT)e- kTs
fc =

since for integral A: and n,

e
-jnkTo>, = e
-j2xnk = 1 (3-128)

Thus Eq. (3-127) becomes

+ jnco =
" 129
R*(s J«*(5) (3 )
s)

this identity, Eq. (3-126) is simplified to

Using

C*(s) = i{*(4 S G(5+7«t» s)

(3-130)

= J?*WG*(*) ( 3431)
C*(j)
where
G*(j) = 4 S
J „ = — oo
G^+jnco,) (3-132)

The transfer function in z of Eq. (3-123) can again be obtained directly from Eq.
(3-131) by use of z = e
Ts
.
.

In conclusion, we note thatthe input to a linear system is sampled

when
unsampled, the Laplace transform of the continuous output
but the output is

is given by
C(s) = G(s)R*(s) (3-133)

sampled by a sampler that is synchronized with

If the continuous-data output is
 Sec. 3.12 Transfer Functions of Discrete-Data Systems / 85

and has the same sampling period as the input, the Laplace transform of the
discrete-data output is given by

C*(s) = G*(s)R*(s) (3-134)

The result in Eq. (3-133) is natural, since it is in line with the well-established
transfer relation for linear time-invariant systems. The expression in Eq. (3-134)
is obtained by use of Eqs. (3-124) and (3-132). However, can be interpreted as it

being obtained directly from Eq. (3-133) by taking the pulse transform on both
sides of the equation. In other words, in view of Eq. (3-129), we can write, from
Eq. (3-133),
C*(s) = [Gis)R*(s)]* (3-135)

= G*(s)[R*(s)]*
where Eq. (3-134) implies that
[R*(s)]* = R*(s) (3-136)

Transfer Functions of Discrete-Data Systems with Cascaded Elements

The transfer function representation of discrete-data systems with cascaded

elements is slightly more involved than that for continuous-data systems,

because of the variation of having or not having any samplers in between the
elements. Figure 3-29 illustrates two different situations of a discrete-data
system which contains two cascaded elements. In the system of Fig. 3-29(a), the
two elements are separated by a sampler S2 which is synchronized to and has

"2
(0 d(t) d*(t) c(t)
T D* G t
(s) G 2 (s)
R*(s) D(s) D*(s) C(s)

w)

(0

C*Cs)

Kt)
R(s)
** r*(t)

R*(s)
Gi(s)
dU)
D(s)
G 2 (s)
c(t)
—»~
C(s)

(b)

Fig. 3-29. (a) Discrete-data system with cascaded elements and sampler
separates the two elements, (b) Discrete-data system with cascaded ele-
ments and no sampler in between.
86 / Transfer Function and Signal Flow Graphs Chap. 3

the same period as the sampler Si. The two elements with transfer functions
G t
and G 2 (s) of the system in Fig. 3-29(b) are connected directly together. In
(s)

discrete-data systems, it is important to distinguish these two cases when deriving

the pulse transfer functions.
Let us consider first the system of Fig. 3-29(a). The output of Gi(s) is written

D(s) = Gi(s)R*(s) (3-137)

and the system output is

C(s) =G 2 (s)D*(s) (3-138)

Taking the pulse transform on both sides of Eq. (3-137) and substituting the
result into Eq. (3-138) yields

C(s) =G 2 (s)GKs)R*(s) (3-139)

Then, taking the pulse transform on both sides of the last equation gives

C*(s) = Gf(s)G*(s)R*(s) (3-140)

where we have made use of the relation in Eq. (3-136). The corresponding
z-transform expression of the last equation is

C(z) =G 2 (z)Gi(z)R(z) (3-141)

We conclude that the z-transform of two linear elements separated by a sampler

is equal to the product of the z-transforms of the two individual transfer func-
tions.
The Laplace transform of the output of the system in Fig. 3-29(b) is

C(s) = Gi{s)G 2 (s)R*(s) (3-142)

The pulse transform of the last equation is

C*(s) = [Gi(s)G 2 (s)]*R*(s) (3-143)

where

[Gi(s)G 2 (s)]* = -i f) G (.s+jmo.yG 1 (.s+jna>.)

1
(3-144)

Notice that since C?i(s) and G 2 (s) are not separated by a sampler, they have to
be treated as one element when taking the pulse transform. For simplicity, we
define the following notation

[Gi(s)G 2 (s)]* = GiG^s)

(3-145)
= G GHs) 2

Then Eq. (3-143) becomes

C*(s) = GiG^(s)R*(s) (3-146)

Taking the z-transform on both sides of Eq. (3-146) gives

C(z) = GiG 2 (z)R(z) (3-147)

where G G 2 (z) is defined as the z-transform of the product of Gi(s) and

1
G 2 (s),
and it should be treated as a single function.
 Sec. 3.12 Transfer Functions of Discrete-Data Systems / 87

It is important to note that, in general,

G Gt(s)^Gf(s)GKs)
t (3-148)
and
G.G^z) ^ G^Gziz) (3-149)

Therefore, we conclude that the z-transform of two cascaded elements with no

sampler in between is equal to the z-transform of the product of the transfer
functions of the two elements.

Transfer Functions of Closed-Loop Discrete-Data Systems

In this section the transfer functions of simple closed-loop discrete-data

systems are derived by algebraic means. Consider the closed-loop system shown
in Fig. 3-30. The output transform is

C(s) = G(s)E*(s) (3-150)

"X, c*(t)

^ C*(s)
•

«t) e(t) V e*(t)

G(s)
c(t)

R(s) _\y E(S ) j.

E*(s)
C(s)

H(s)

Fig. 3-30. Closed-loop discrete-data system.

The Laplace transform of the continuous error function is

E(s) = R(s) - H(s)C(s) (3-151)

Substituting Eq. (3-150) into Eq. (3-151) yields

E(s) = R(s) - G(s)H(s)E*(s) (3-152)

Taking the pulse transform on both sides of the last equation and solving for
E*(s) gives

E * (S) _ R*(s)
(3-153)

The output transform C(s) is obtained by substituting E*(s) from Eq. (3-153)
we have
into Eq. (3-150);

G(s)
C(s) ,R*(s) (3-154)
1 + GH*(s)
Now taking the pulse transform on both sides of Eq. (3-154) gives

G*(s)
c
^-TTmwr {s} (3-155)
88 / Transfer Function and Signal Flow Graphs Chap. 3

In this case it is possible to define the pulse transfer function between the input
and the output of the closed-loop system as
C*(s) G*(s)
(3-156)
R*(s) 1 + GH*(s)
The z-transfer function of the system is

C(z) _ G{z)
R(z) ~ + GH(z)
1
(3 " 157)

We shall show in the following that although it is possible to define a

transfer function for the closed-loop system of Fig. 3-30, in general, this may
not be possible for all discrete-data systems. Let us consider the system shown
in Fig. 3-31. The output transforms, C(s) and C{z), are derived as follows:

C(s) = G(s)E(s) (3-158)

E(s) = R(s) - H(s)C*(s) (3-159)

~>" C*(t)

T
KO ~?) e(t) c(t)

R(s)
(% G(s)
\J E(s) C(s)

c*(t)
Hds)
C*(s)
T

Fig. 3-31. Closed-loop discrete-data system.

Substituting Eq. (3-159) into Eq. (3-158) yields

C{s) = G(s)R(s) - G{s)H(s)C*{s) (3-160)

Taking the pulse transform on both sides of the last equation and solving for
C*(s), we have

= G
+ **£L
C*(s) (3-161)
,
1 GH *{s)
Note that the input and the transfer function G(s) are now combined as one
function, GR*(s), and the two cannot be separated. In this case we cannot define
a transfer function in the form of C*(s)/R*(s).
The z-transform of the output is determined directly from Eq. (3-161) to be

C(z) = r GR(z)
GH(z)
(3-162)

where it is important to note that

GR(z) = #[G(*)fl(s)] (3-163)
and
GH(z) = g[G(s)H(s)] (3-164)
Chap. 3 Problems / 89

To determine the transform of the continuous output, C(s), we substitute

C*(s) from Eq. (3-161) into Eq. (3-160). We have

C(s) = G(s)R(s) - G+gh*\s) GR * (s)

t
(3 " 165)

Although we have been able to arrive at the input-output transfer function

and transfer relation of the systems of Figs. 3-30 and 3-3 1 by algebraic means
without difficulty, for more complex system configurations, the algebraic method
may become tedious. The signal-flow-graph method is extended to the analysis
7
of discrete-data systems; the reader may refer to the literature. - 8

REFERENCES

Block Diagram and Signal Flow Graphs

1. T. D. Graybeal, "Block Diagram Network Transformation," Elec. Eng., Vol.

70, pp. 985-990, 1951.

2. S. J. Mason, "Feedback Theory Some — Properties of Signal Flow Graphs,"

Proc. IRE, Vol. 41, No. 9, pp. 1144-1156, Sept. 1953.

3. S. J. —
Mason, "Feedback Theory Further Properties of Signal Flow Graphs,"
Proc. IRE, Vol. 44, No. 7, pp. 920-926, July 1956.
4. L. P. A. Robichaud, M. Boisvert, and J. Robert, Signal Flow Graphs and
Englewood Cliffs, N.J., 1962.
Applications, Prentice-Hall, Inc.,

5. B. C. Kuo, Linear Networks and Systems, McGraw-Hill Book Company, New

York, 1967.

6. N. Ahmed, "On Obtaining Transfer Functions from Gain-Function Derivatives,"

IEEE Trans. Automatic Control, Vol. AC-12, p. 229, Apr. 1967.

Signal Flow Graphs of Sampled- Data Systems

1. B. C. Kuo, Analysis and Synthesis of Sampled- Data Control Systems, Prentice-

Hall, Inc., Englewood Cliffs, N.J., 1963.
8. B. C. Kuo, Discrete Data Control Systems, Science-Tech, Box 2277, Station A,
Champaign, Illinois, 1970.

PROBLEMS
3.1. The following differential equations represent linear time-invariant systems,
where denotes the input and
r(t) c{t) denotes the output. Find the transfer
function of each of the systems.

d^) + 3 d_^1 + 4 aW) + c(/) = 2 ^) f(/)

(a) +
(b) *%P + 104g-> + 2cit) = r(» - 2)

3.2. The block diagram of a multivariate feedback control system is shown in Fig.
P3-2. The transfer function matrices of the system are
90 / Transfer Function and Signal Flow Graphs Chap. 3

1 1

= s s + 2
G0)

1 0"
HO)
.0 1.

Find the closed-loop transfer function matrix for the system.

Rfe) /\ ») ,
G(s)
C(s)

y
i

H(s)

Figure P3-2.

3.3. A multivariable system with two inputs and two outputs is shown in Fig. P3-3.
Determine the following transfer function relationships

Ci(j) C2 (s) Ci(j) C2 (s)

*i(*) Ri(s)

Write the transfer function relation of the system in the form

C(s) = G(s)R(s)

R 2 (s)

Figure P3-3.

3.4. Draw a signal flow graph for the following set of algebraic equations:

3xi + x 2 + 5x = 3

Xi + 2x 2 — 4x = 2 3

—x 2 — x = 3
Chap. 3 Problems / 91

3.5. Draw an equivalent signal flow graph for the block diagram in Fig. P3-5. Find
the transfer function C(s)jR(s).

Figure P3-5.

3.6. Find the gains, y 6 /yi, y%ly\, and yi/y 2 for the signal flow graph shown in Fig.
P3-6.

Figure P3-6.

3.7. Find the gains y<,\y x and y 2 lyi for the signal flow graph shown in Fig. P3-7.

-0.5

Figure P3-7.
92 / Transfer Function and Signal Flow Graphs Chap. 3

3.8. In the circuit of Fig. P3-8, e s (t), e^t), and i,(t) are ideal sources. Find the value
of a so that the voltage e (t) is not affected by the source ed (t).

o+

Figure P3-8.

3.9. Are the two systems shown in Fig. P3-9(a) and (b) equivalent ? Explain.

i i

O-
yj

(a)

(b)

Figure P3-9.

3.10. Given the signal flow graph of Fig. P3-10(a) and the transfer functions G it G 2 ,

G 3 G 4 and G 5 find the transfer functions GA GB and (7C so that the three
, , , , ,

systems shown in Fig. P3-10 are all equivalent.

(a)

Figure P3-10.
 Chap. 3 Problems / 93

(b)

(c)

Figure P3-10. (Cont.)

3.11. Construct an equivalent signal flow graph for the block diagram of Fig. P3-11.
(a) Evaluate the transfer function C/R when N=
0. (b) Determine the relation
among the transfer functions G u G 2 , G 3 G4 H
u and 2 so that the output C
, , H
is not affected by the disturbance signal N.

Figure P3-11.

3.12. A multivariate system is described by the following matrix transfer function

relations
C(s) = G(s)S(s)
S(.y) = R(.r) - H(*)CO)
where

C(s) = c^sy R(s) = -R^sT

-Ri(s).
i 5 n

= s s + 1
= 'I 0"
G(s) H( S)
1
_1_ .0 0_
S -
94 Function and Signal Flow Graphs Chap. 3
/ Transfer

(a) Derive the closed-loop transfer function relationship

C(i) = M(.$)RO)
by using
M(s) = [I + G(j)H(5)]-'G(i)

(b) Draw a signal flow graph for the system and find M(s> from the signal flow
graph using Mason's gain formula.
3.13. Find the transfer function relations C(s)/R(s) and C(s)jE(s) for the system
shown in Fig. P3-13.

Figure P3-13.

3.14. Find the transfer function C(z)/R(z) of the discrete-data system shown in Fig.

P3-14.The sampling period is 1 sec.

r(t) ^ r*(t)
s(s
1

+ 2)
c(t)

Figure P3-14.

3.15. Find the z-transfer functions C(z)/R(z) of the discrete-data systems shown in

Fig. P3-15.

KO
^ r*(.t)

s
1

+ s
2
+ 2
cd)

(a)

r(t) ^ r*(t)
^
s+ 1
-X- s +
c(t)

(b)

Figure P3-15.
 4
State-Variable Characterization

of Dynamic Systems

4.1 Introduction to the State Concept

In Chapter 3 the classical methods of describing a linear system by

transfer
function, impulse response, block diagram, and signal flow graph have
been
presented. An important feature of this type of representation is that the
system
dynamics are described by the input-output relations. For instance, the trans-
fer function describes the input-output relation in the Laplace
transform do-
main. However, the transform method suffers from the disadvantage that
all
the initial conditions of the system are neglected. Therefore,
when one is in-
terested in atime-domain solution, which depends to a great deal on the past
history of the system, the transfer function does not carry all the
necessary
information. Transfer function js valuable for frequency-domain analysis and
design, as well as for stability studies.
The greatest advantage of transfer
function is compactness and the ease that we can obtain qualitative in-
in its
formation on the system from the poles and zeros of the transfer function.
An alternative to the transfer function method of describing a linear system
is the state-variable method. The state-variable representation isnot limited
to linear systems and time-invariant systems. It can be applied to nonlinear as
well as time-varying systems.
The state-variable method is often referred to as a modern approach.
However, in reality, the state equations are simply first-order differential equa-
tions, which have been used for the characterization of dynamic
systems for
many years by physicists and mathematicians.
To begin with the state-variable approach, we should first begin by de-
fining the state of a system. As the word implies, the state of a system refers to

95
 State-Variable Characterization of Dynamic Systems Cna P- 4
96 /

the past, present, and future conditions of the system. It is interesting to note
that an easily understood example is the "State of the Union" speech given by
the President of the United States every year. In this case, the entire system
encompasses all elements of the government, society, economy, and so on. In
general, the state can be described of numbers, a curve, an equation,
by a set

or something that is more abstract in nature. From a mathematical sense it is

convenient to define a set of state variables and state equations to portray sys-
tems. There are some basic ground rules regarding the definition of a state
variable and what constitutes a state equation. Consider that the set of variables,
., x„(t) is chosen to describe the dynamic
characteristics of a
Xl (t), x 2 (t), . .

system. Let us define these variables as the state variables of the system. Then
these state variables must satisfy the following conditions

1. At any time t = t , the state variables, Xi(t ), x 2 (t a ), . . . , x„(t )

define the initial states of the system at the selected initial time.

2. Once the inputs of the system for t t and the initial states defined >
above are specified, the state variables should completely define the
future behavior of the system.

Therefore, we may define the state variables as follows:

of state variables. The state variables of a system are defined as

Definition
a minimal of variables, x^t), x 2 (t),
set ., x n (t) such that knowledge of these
. .

variables at any time t plus information on the input excitation subsequently

,

t > 1
applied, are sufficient to determine the state of the system at any time
.

confuse the state variables with the outputs of a system.

One should not
An output of a system is a variable that can be measured, but a state variable
does not always, and often does not, satisfy this requirement. However, an
output variable is usually defined as a function of the state variables.

R Example 4-1 As a simple illustrative example of state vari-

-^AM 1 ables, let us consider the RL
network shown
in Fig. 4-1. The history of the network is
e(f) i(0 ) fJ£ completely specified by the initial current of the inductance,
j'(0+), at t 0. At / = =
0, a constant input voltage of ampli-

tude Ei is applied to the network. The loop equation of the

network for t > is

Fig. 4-1. RL network.

^=R .

(t) + L dm ^ )

Taking the Laplace transform on both sides of the last equation, we get

E(S ) =*h. = (R+ Ls)I(s) - Li(0+) (4-2)

Solving for I(s) from the last equation yields

m~
., E,
s{R+Ls)'r
. 1-/(0+)
R+Ls (4-3)

The current /(/) for f ^ is obtained by taking the inverse Laplace transform of both
 Sec. 4.2 State Equations and the Dynamic Equations / 97

sides of Eq. (4-3). We have

'(0 = ^ (1 - e-™1 + -) i(0+)e- R,/L (4-4)

Once the current /(/) is determined for / > 0, the behavior of the entire network is

defined for the same time apparent that the current i(t) in this
interval. Therefore, it is

case satisfies the basic requirements as a state variable. This is not surprising since an
inductor is an electric element that stores kinetic energy, and it is the energy storage
capability that holds the information on the history of the system. Similarly, it is easy
to see that the voltage across a capacitor also qualifies as a state variable.

4.2 State Equations and the Dynamic Equations

The first-order differential equation of Eq. (4-1), which gives the relationship
between the state variable i(t) and the input e{t), can be rearranged to give

This first-order differential equation is referred to as a state equation.

For a system with p inputs and q outputs, the system may be linear or
nonlinear, time varying or time invariant, the state equations of the system are
written as

^ i
=fix
=
l (f),

1,2, ... ,n
x 2 (t), ..., x„(t), r,(0, r 2 (0, • • • , r p (t)] (4-6)

where x^t), x2 (t), . . . , xn (t) are the state variables; /-,(/), r 2 (t), . .
.
, rp {t) are
the input variables; and/;, denotes the z'th functional relationship.
The outputs of the system c k (t), k = 1, 2, q, . . .
,
are related to the state
variables and the inputs through the output equation,
c k {t) = &[*,(*), x 2 (t), ..., x„(t), rM r 2 {i), ..., r p (t)] (4-7)

k=\,2,...,q
where gk denotes the kth functional relationship. The state equations and the
output equations together form the set of equations which are often called the
dynamic equations of the system.
Notice that for the state equations, the left side of the equation should
contain only thefirst derivatives of the state variables, while the right side should

have only the state variables and the inputs.

Example 4-2 Consider the RLC network shown in Fig. 4-2. Using the conven-
tional network approach, the loop equation of the network is
written

e{t) = Ri(t) +L^£ + ±[ i{t) dt (4-8)

We notice that is not in the form of a state

this equation
equation, since the last term a time integral. One method
is

of writing the state equations of the network, starting with

Fig. 4-2. RLC network. Eq. (4-8), is to let the state variables be defined as
 :

98 / State-Variable Characterization of Dynamic Systems Chap. 4

*i(0 = iff) (4-9)

Xl(t) = f /(/) dt (4-10)

Substituting the last two equations into Eq. (4-8), we have

eit)--= Rx (t)+L
1 ^
c
+ ^x 2 (t) (4-11)

Rearranging the terms in Eq. (4-11) and taking the time derivative on both sides
we have the two state equations of the network,
of Eq. (4-10),

^ = -T*'<'>-nr*<'> + r*> (4 " 12)

^- = }
*,(/) (4-13)

which are linear first-order differential equations.

We have demonstrated how the state equations of the RLC network may
be written from the loop equations by defining the state variables in a specific
way. The objective, of course, is to replace Eq. (4-8) by two first-order differential
equations. An alternative approach is to start with the network and define the
state variables according to the elements of the network. As stated in Section
4.1, we may assign the current through an inductor and the voltage across a
capacitor as state variables. Therefore, with reference to Fig. 4-2, the state
variables are defined as
*i(o = m (4-i4)

*2 (0 = efy) (4-15)

Then, knowing that the state equations would have to contain the first deriva-
tives of x^t) and x 2 (t) on the left side of the equations, we can write the equa-
tions directly by inspection from Fig. 4-2

Voltage across L: £ ^p = -Ri(t) - e e (t) + <?(0 (4

" 16 )

Current in C: C^2 = at
i(t) (4-17)

Using Eqs. (4-14) and (4-15), and dividing both sides of the last two equations
by L and C, respectively, we have

**M = -£ Xl (f)
- i-* 2 (0 + ±e(t) (4-18)

4*M = (4-19)
^ Xi (f)

which are the state equations of the RLC network. We notice that using the
two independent methods, the only difference in the results is in the definition
of the second state variable x 2 (t). Equation (4-19) differs from Eq. (4-13) by
the factor of the capacitance C.
From two simple examples we see that
these for linear time-invariant
systems, the state equations can generally be written as
 Sec. 4.3 Matrix Representation of State Equations / 99

dx£t)
dt % a u xj(0 + 2l b ik r k (i) i=l,2,....,n (4-20)

where a tJ and b ik are constant coefficients. The output equations are written

c k (0 = £d kJ Xj(t) + m^
t, e km r m {i) k=l,2,...,q (4-21)
j= 1 1

where dkj and ekm are constant coefficients.

For a linear system with time-varying parameters, the coefficients of Eqs.
(4-20) and (4-21) become time dependent.

4.3 Matrix Representation of State Equations

The dynamic equations are more conveniently expressed in matrix form. Let
us define the following column matrices

"*i(0'

x 2 (t)
x(r) (n X 1) (4-22)

where x(t) is defined as the state vector;

>i(0"
r 2 (t)

r(/) = (px 1) (4-23)

WO.
where r(/) is defined as the input vector; and

"c,(0"

cz(t)

c(0 (qX 1) (4-24)

_c,(0_

where c(r) is defined as the output vector. Then the state equations of Eq. (4-6)
can be written

M)=f[x(0,r(0] (4-25)

where f denotes an n x 1 column matrix that contains the functions fu f2 , . . .

,

f„ as elements, and the output equations of Eq. (4-7) become

c(r) = g[x(r), r(r)] (4-26)
where g denotes a q x column matrix that contains the functions
1
g u g 2 ,...,
gq as elements.
 :

State-Variable Characterization of Dynamic Systems Chap. 4

100 /

For a linear time-invariant system, the dynamic equations are written

State equation : ^ = Ax(0 + Br(r) (4-27)

Output equation: c(/) = Dx(0 + Er(/) (4-28)

where A is an n X n coefficient matrix given by

a 11 fl 12 ••• . a,
"In

d 21 #22 • • a 2n

A= (4-29)

a„, a„

B is an n X p matrix given by
t>u b 12 a.;

bn b2 2

B (4-30)

P„i b„ 2

D is a q X n matrix given by
~d u d i2 du
"21 "22 d2 „
D= (4.31)

dql dq2
and E is a q X p matrix,
'e u «i2

en e 12 "2p

E = (4-32)

Zql C«2

Example 4-3 The state equations of Eqs. (4-18) and (4-19) are expressed in matrix-

vector form as follows

-
dxi(ty r R V -\
'
L '
L r*iw" L
dt
+ e(0 (4-33)
dx 2 (t) j_
2 \x {t)
L Cdt J

Thus the coefficient matrices A and B are identified to be

_R 1

L L
A= (4-34)
J_
C
 Sec. 4.4
f State Transition Matrix / 101

B= L (4-35)

4.4 State Transition Matrix

The state transition matrix is defined as a matrix that satisfies the linear homo-
geneous state equation

^ = Ax« (4-36)

Let <£(f) be an n x n matrix that represents the state transition matrix; then it
must satisfy the equation

^ = A«KO (4-37)

Furthermore, let x(0+) denote the initial state at t = 0; then <f>(t) is also defined
by the matrix equation
x(t) = <Kr)x(0+) (4-38)

which is the solution of the homogeneous state equation for t > 0.

One way of determining fy(t) is by taking the Laplace transform on both
sides of Eq. (4-36) we have ;

sX(s) - x(0+) = AX(s) (4-39)

Solving for X(s) from the last equation, we get

- A)" >x(0+)X(s) = (si (4-40)

where it is assumed that the matrix (si — A) is nonsingular. Taking the inverse
Laplace transform on both sides of the last equation yields

x(0 = £,- » [(si - A)" ']x(0+) / > (4-41)

Comparing Eq. (4-41) with Eq. (4-38), the state transition matrix is identified
to be
(>(0 = £- [(jI-A)-']
1
(4-42)

An alternative way of solving the homogeneous state equation is to assume a

solution, as in the classical method of solving differential equations. We let
the solution to Eq. (4-36) be
x(0 = e
A
'x(0+) (4-43)

for t > 0, where e At

represents a power series of the matrix At and

e
AI
=1+ At 4- ^A*/ 2 +1A 3
/
3
+ . . . (4-44)*

It is easy to show that Eq. (4-43) is a solution of the homogeneous state

equation, since, from Eq. (4-44),

^= Ae Kt (4-45)

*It can be proved that this power series is uniformly convergent.

 .

Dynamic Systems Cha P- 4

102 / State-Variable Characterization of

Therefore, in addition to Eq. (4-42), we have obtained another expression for

the state transition matrix in

«(,(,) = eK< =I+ At + ~A 2 2

/ + 1A 3
;
3
+ . . (4-46)

Equation (4-46) can also be obtained directly from Eq. (4-42). This is left as an
(Problem 4-3).
exercise for the reader

Example 4-4 Consider the RL network of Fig. 4-1 with the input short circuited;
that is, e(t) = 0. The homogeneous state equation is written

*£=-£*) (4-47)

The solution of the last equation for t > is obtained from Eq. (4-4) by setting Ei = 0.
Thus
i(t) = e' R,/L i(0+) (4-48)

The state transition matrix in this case is a scalar and is given by

0(0 = e-*" L t > (4-49)

which is a simple exponential decay function.

Significance of the State Transition Matrix

Since the state transition matrix satisfies the homogeneous state equation,
it represents the free response of the system. In other words, it governs the
response that is excited by the initial conditions only. In view of Eqs. (4-42)
and (4-46), the state transition matrix is dependent only upon the matrix A.
As the name
implies, the state transition matrix <f>(0 completely defines the
transition of the states from the initial time t to any time t. =
Properties of the State Transition Matrix

The state transition matrix <J>(0 possesses the following properties:

1.
<f)(0)
= I the identity matrix (4-50)

Proof: Equation (4-50) follows directly from Eq. (4-46) by setting

f = 0.

2. 0-«(O = 4»(-O (4-51)

Proof: Postmultiplying both sides of Eq. (4-46) by e~ Ac

, we get

<K0e"
At
= e x 'e- A ' = I (4-52)
_1
Then premultiplying both sides of Eq. (4-52) by <J> (')> we 8 et
e- A ' = (f,-i( f ) (4-53)
Thus
<K-r) = <f»~'(0 = e~ At (4-54)

An interesting result from this property of <f>(t) is that Eq. (4-43) can be re-
arranged to read
x(0+) = <J>(-0x(0 (4-55)
 Sec. 4.5
State Transition Equation / 103

which means that the state transition process can be

considered as bilateral in
time.That is, the transition in time can take place in either
direction.
3. <J>(7 2 - 0<K>i - t ) = <f>(t 2 - t ) for any tQ , t u t2 (4-56)
Proof:

<K'2 - 'i)<K'l - t ) = A «? e
<'«-"> A(r,-»„)

=e A( "-'«)
(4.57)
= M2 - t )

This property of the state transition matrix is

important since it implies
that a state transition process can be
divided into a number of sequential
transitions. Figure 4-3 illustrates that the
transition from t t Q to t t 2 is = =

Fig. 4-3. Property of the state

transition matrix.

equal to the transition from t to t and then

lt from U to t 2 In general, of course .

the transition process can be broken up

into any number of parts
Another way of proving Eq. (4-56) is to write

= <K>2 - tMt
*(h) t ) (4-58)
*('.) = ftr, - )x(t t
) (4.59)
x(t = <K?2 - )x(t
2) tQ
) (4 _ 60)
The proper result is obtained by substituting Eq. (4-59) into Eq (4-58) and
comparing the result with Eq. (4-60).
4 -
MO]" = <Kkt) for k = integer (4-61)
Proof:
oAr
[<t>(t)Y (k terms)
— e kAt
(4-62)
= Mt)
4.5 State Transition Equation

The state transition equation is defined as

the solution of the linear nonhomo-
geneous state equation. For example, Eq.
(4-5) is a state equation of the RL
network of Fig. 4-1. Then Eq. (4-4) is the state
transition equation when the
input voltage is constant of amplitude £, for t > 0.
104 / State-Variable Characterization of Dynamic Systems Chap. 4

In general, the linear time-invariant state equation

&j& = Ax(0 + Br(r) (4-63)

can be solved by using either the classical method of solving differential equa-
tions or the Laplace transform method. The Laplace transform method is
presented in the following.
Taking the Laplace transform on both sides of Eq. (4-63), we have
sX(s) - x(0+) = AX(s) + BR(s) (4-64)

where x(0+) denotes the initial state vector evaluated at t = 0+. Solving for
X(s) in Eq. (4-64) yields

X(s) = (si - A)" 'x(0+) + (si - A)" 'BR(s) (4-65)

The state transition equation of Eq. (4-63) is obtained by taking the inverse
Laplace transform on both sides of Eq. (4-65),

x(f) = £"'[(5l - A)" ]x(0+) 1

+ JC-'Kil - A)- BR(s)]
1
(4-66)

Using the definition of the state transition matrix of Eq. (4-42), and the con-
volution integral, Eq. (3-26), Eq. (4-66) is written

x(r) = <K0x(0+) + Jf <f>(?

- T)Br(r) dx t > (4-67)

The state transition equation in Eq. (4-67) is useful only when the initial
time is defined to be at t = 0. In the study of control systems, especially dis-
crete-data control systems, it is often desirable to break up a state transition
process into a sequence of transitions, so that a more flexible initial time must
be chosen. Let the initial time be represented by ta and the corresponding initial

state by x(/ ), and assume that an input r(f) is applied for / > r„.

We start with Eq. (4-67) by setting t = t , and solving for x(0+), we get

x(0+) = <J>(-/ )x(/o)

- <K-'o) f" Wo - T)Br(t) di (4-68)

where the property on <(>(f) of Eq. (4-51) has been used.

Substituting Eq. (4-68) into Eq. (4-67) yields

x(0 = (J>(0<K->o)x('o) - <KO<K-'o) Jf°<Wo - T)Br(T)rfr

° (4-69)
,,
+ J
<(>(?- T)Br(r) di

Now using the property of Eq. (4-56), and combining the last two integrals,
Eq. (4-69) becomes
x(r) = <f>(t
- t )x(t ) + f <j)(t ~ r)Br(r) dx (4-70)

It is apparent that Eq. (4-70) reverts to Eq. (4-67) when i

a
~ 0.

Once the state transition equation is determined, the output vector can be
expressed as a function of the initial state and the input vector simply by sub-
stituting x(r) from Eq. (4-70) into Eq. (4-28). Thus the output vector is written

c(0 = Dcf>(' - t )x(t ) + f ' D<f>(' - T)Br(r) dx ' Er(f) (4-71)

 Sec. 4.5
State Transition Equation / 105

The following example illustrates the application of the state transition

equation.

Example 4-5 Consider the state equation

dt r*i(o"

dx 2 (t) + #(/)
(4-72)
-2 1
dt |*2(0
The problem to determine the state vector >
is x(0 for t when the input r(/) = 1 for
t > 0; that is, r(t) = u s (t). The coefficient matrices A and B are identified to
be
r ~0~
B= (4-73)
_ 2 -3. _1_
Therefore,
0~ " 1"
si ~s -1 "1

s _-2 -3. _2 s + 3_
The matrix inverse of (si ~ A) is

(si- A)-' = 1

+
s + 3 1"

(4-74)
3s 2 -2 5

The state transition matrix of A is found by taking the inverse Laplace transform of
the last equation. Thus

= 2e~' — e' 1 '

e~ !— e' 2 '

<K0 =<£-'[(*!- A)- >] (4-75)

-2e~' + 2e~ 2 ' —e~' + 2e~ 2t

The state transition equation for t > is obtained by substituting Eq (4-75) B and
'
#•(/) into Eq. (4-67). We have
2e~' ~ e' 2 e~' —
'
'
e~ 2 '
x(0 x(0+)
—2e-' + e~ 2 — e -< +
<
2e~ 2 \

2fi-('-') — e -2(t-t) »-(i-t) „-2<»-t)

J
-2g-(r-r) _|_ 2 e -2('-r) _ e -(f-T) _|_2e- 2 ('-'>
rfT (4-76)

2e~' — c -2 ' g-' — e

~2' '

x(0 x(0+)
-2c"' + 2e~ 2 <
-e-< + 2e" 2 '.
(4-77)
2 — c -
T j
+ <?-' — e -2t >>0
As an alternative, the second term of the state transition equation can be
obtained by
taking the inverse Laplace transform of (si - A)-'BR(s). Therefore
£~ Ksl- 1
A)-iBR(s)] =£-i 's + 3 1"

- -2 s.

-J_-l
= £-' 1
s
3s + 2 (4-78)
1_
| - e~' + \e~ 2>

e~< — e -2 '
/>0
106 / State-Variable Characterization of Dynamic Systems Chap. 4

Example 4-6 In this example we shall illustrate the utilization of the state transition
method to a system with input discontinuity. Let us consider that the
input voltage to the RL network of Fig. 4-1 is as shown in Fig. 4-4.

IE

e(t)

Fig. 4-4. Input voltage waveform for the network in Fig. 4-1.

The state equation of the network is

di(t)
rf- = -r /(/ + r«w )
(4-79)

Thus
= R l
A
L
B (4-80)
L
The state transition matrix is

#(0 = e~ R " L (4-81)

One approach to the problem of solving for ;'(?) for / > is to express the input
voltage as
e(t) = Eu s (t) + Eu s (t - ?,) (4-82)

where u s (t) is the unit step function. The Laplace transform of e(r) is

£(s)=^ (!+*-"•) (4-83)

Then
- A)-'BR(j) 1 1 E
(si
s + R/L L s
d+e-"')
(4-84)
E +<?-'")
Rs[l + (L/R)s] (1 *

Substituting Eq. (4-84) into Eq. (4-66), the current for t > is obtained:

i(t) = e- R "H(0+)u s (0 + §d - e-"*)u,(t)

(4-85)
+ -f[l - e -M-«w-]u.(t-t 1 )

Using the state transition approach we can divide the transition period into two
parts: t = to t = t,, and t = ?, to t = oo. First for the time interval, < t <, t,,

the input is

e(t) = Eu,(t) 0<t < r, (4-86)

Then

(4-87)
j

Rs[l + (L/R)s]
 :

Sec. 4.6 Relationship Between State Equations / 107

Thus the state transition equation for the time interval < < t f i is

i(0 = [e- R'^i(0+) + -fr (1 - e- R"L)^ «,(/) (4-88)

Substituting t = tx into this equation, we get

i( tl ) = e -Ru/L i( Q + ) + -j| (1 - e -*"> L ) (4-89)

The value of i(t) at t = ti is now used as the initial state for the next transition
period of tt < < co. The magnitude of the input for this interval
t is 2E. Therefore,
the state transition equation for the second transition period is

- i(0 = e-*('--"L i('i) + =£ [1 - e -*e-'.>/i] t > t, (4-90)

where i(ti) is given by Eq. (4-89).

This example illustrates two possible ways of solving a state transition problem.
In the first approach, the transition is treated as one continuous process, whereas in
the second, the transition period is divided into parts over which the input can be more
easily represented. Although the
approach requires only one operation, the second
first

method and it often

yields relatively simple results to the state transition equation,
presents computational advantages. Notice that in the second method the state at
/ = t is used as the initial state for the next transition period, which begins at t
x x .

4.6 Relationship Between State Equations and High-Order

Differential Equations

In preceding sections we defined the state equations and their solutions for
linear time-invariant systems. In general, although it is always possible to
from the schematic diagram of a system, in practice
write the state equations
the system may have been described by a high-order differential equation or
transfer function. Therefore, it is necessary to investigate how state equations
can be written directly from the differential equation or the transfer function.
The relationship between a high-order differential equation and the state equa-
tions is discussed in this section.
Let us consider that a single-variable, linear time-invariant system is de-
scribed by the following nth-order differential equation
~l
d"c{i) , dn c(i) , d"~ 2 c(t) , , dc(i) , ,.. ,,..
lA „,,.

where c(t) is the output variable and r(t) is the input.

The problem is to represent Eq. (4-91) by n state equations and an output
equation. This simply involves the defining of the n state variables in terms of
the output c(0 and its derivatives. We have shown earlier that the state vari-
ables of a given system are not unique. Therefore, in general, we seek the most
convenient way of assigning the state variables as long as the definition of state
variables stated in Section 4.1 is met.
108 / State-Variable Characterization of Dynamic Systems Chap. 4

For the present case it is convenient to define the state variables as

x,(0 = c(t)

*a(0 dt

(4-92)

*»(o 1
fife""

Then the state equations are

dx z (t)
dt
=* s (0

(4-93)

<fa»-i(0
xJLO
<//

dx„(t)
dt
= -a x,(0 - B a„_,x 2 (0 - ... - a 2 Jf»-i(0 - «i^(0 + KO
where the last state equation is obtained by equating the highest-ordered deriva-
tive term to the rest of Eq. (4-91). The output equation is simply

c{t) = x,(0 (4-94)

In vector-matrix form, Eq. (4-93) is written

dx(t)
= Ax(0 +
.

Br(t) (4-95)
dt

where x(t) is the n X 1 state vector and r(i) is the scalar input. The coefficient
matrices are

1
(« X n) (4-96)

—a n
— a„-i — a n -2 — A,-3 — fl„_4 • .
—a
0"

B (»xl) (4-97)
 Sec 4 7
- -
Transformation to Phase-Variable Canonical Form / 109

The output equation in vector-matrix form is

c(0 = Dx(r) (4-98)

where
D= [1 ... 0] (1 x ji) (4-99)

The state equation of Eq. (4-95) with the matrices A

and B defined as in
Eqs. (4-96) and (4-97) is called the phase-variable canonical form in the next
section.

Example 4-7 Consider the differential equation

^£) + 5 ^L) + ^) + 2c(/) = rCO (4100)

Rearranging the last equation so that the highest-order derivative term is equated to
the rest of the terms, we have

The state variables are defined as

*i(') = c(0
x.(0=*£> (4-102)

d 2 c(
t)
**<*) ~ ~^m
dt*

Then the state equations are represented by the vector-matrix equation of Eq. (4-95)
with
"010
1
(4-103)
-2 -1 -5
and

B= (4-104)

The output equation is

c(t) = Xl (t) (4-105)

4.7 Transformation to Phase-Variable Canonical Form

In general, when the coefficient matrices A and B are given by Eqs. (4-96) and
(4-97), respectively, the state equation of Eq. (4-95) is called the phase-variable
canonical form. It is shown in the following that any linear time-invariant sys-
tem with single input and satisfying a certain condition of controllability (see
section 4.15) can always be represented in the phase-variable canonical form.

Theorem 4-1. Let the state equation of a linear time-invariant system be

given by

4*j& = Ax(0 + Brit) (4-106)

 Dynamic Systems Chap. 4
110 / State-Variable Characterization of

where x(t) is an n X 1 state vector, A an n X n coefficient matrix, B an n X 1

coefficient matrix, and r(t) a scalar input. If the matrix

S = [B AB A 2 B ... Anl B] (4-107)

is nonsingular, then there exists a nonsingular transformation

y(f) = Px(r) (4-108)

or
x« = P-'yCO (4-109)

which transforms Eq. (4-106) to the phase-variable canonical form

y(0 =A t y(0 + B,r(0 (4-110)

where
1 ...

1 ...

1 ...
A, (4-111)

-a„ —a„
and

B 1
= (4-112)

The transforming matrix P is given by

Pt
PA t

(4-113)

_P A"-
1
1
_

where
P, = [0 1][B AB A B 2 A"- B]-
]
(4-114)

Proof: Let

Xi(f)

x(0 = (4-115)

_*.(0-
 Sec. 4.7
Transformation to Phase-Variable Canonical
Form / 111

>i(0'

y(0 =
(4-116)

U.(0.
and
Pll Pl2 Pin P.
Pi I PlZ Pin
P=
(4-117)

Pnl Pnl .
Pnn.
where
P = < [Pa Pa P in ] i = 1, 2, n
. . . , (4-118)
Then, from Eq. (4-108),

yi(t) =P il x 1
(t) + p 12 x 2 (t) + ...+ Pu xn(t)

J ^"
tim e
° n b ° th sMeS ° f the
1 6
Fn!f "rf um ,
last e 4 uation and in view of
fcqs. (4-110) and i
(4-111),

M*) =y 2 (t) = P,x(0 = P,Ax(0 + p.BKO (4-120)

pJ=
Eq
TSre^
5
^^ " *
fUnCti ° n
°
f
* W 0nly
'
in
^ (4 " 120) >

Ji(0 = y7.it) = V Jaif) 1

(4 _ 12 i)
Taking the time derivative of the last
equation once again leads to

Mt) =y 3 (t) = P,A 2x(?) (4-122)

with P AB =
t 0.

Repeating the procedure leads to

J'.-i(0 = >'.(0 = PiA"-'x(0 (4-123)

with P,A-*B = 0. Therefore, using Eq. (4-108), we have
Pt
P.A
y(f) = Px(0 = x(0 (4-124)

PjA"- 1

or
P,
P.A
P=
(4-125)

.Pi A"-'.
112 / State- Variable Characterization of Dynamic Systems Chap. 4

and P, should satisfy the condition

PjB = P,AB = . . . = P,A"- 2 B = (4-126)

Now taking the derivative of Eq. (4-108) with respect to time, we get

fit) = Px(0 = PAx(f) + PBr(0 (4-127)

Comparing Eq. (4-127) with Eq. (4-110), we obtain

A, = PAP '
(4-128)

and
B = PB
Y
(4-129)

Then, from Eq. (4-125),

-
P,B ' '0'
P AB
t

PB = __ (4-130)

_P,A"- B. 1
_1 .

Since Pj is an 1 X n row matrix, Eq. (4-130) can be written

P,[B AB A 2 B ... A"-»B] = [0 ... 1] (4-131)

Thus P! is obtained as

P,=[0 ... 1][B AB A B 2

... A-^B]- 1

(4-132)
= [0 ... lJS" 1

if the matrix S = [B AB A B ...

2 A""'B] is nonsingular. This is the
condition of complete state controllability. Once P! is determined from Eq.
(4-132), the transformation matrix P is given by Eq. (4-125).

Example 4-8 Let a linear time-invariant system be described by Eq. (4-95) with

"1 -1
B (4-133)
-1

It is desired to transform the state equation into the phase-variable canonical form.
Since the matrix
ri o~
S = [B AB] = '

(4-134)
-1

is nonsingular, the system may be expressed in the phase-variable canonical form.

Therefore, Pi is obtained as a row matrix which contains the elements of the last row
of S -1 ; that is,

P,=[l -1] (4-135)

Using Eq. (4-125),

(4-136)
_PjA_
 Sec. 4.7
Transformation to Phase-Variable Canonical Form
/ 113

Thus
"0 1
A, =PAP' (4-137)
1

B! = PB = ro (4-138)
i

The method of defining state variables by inspection

as described earlier
with reference to Eq. (4-91) is inadequate
when the right-hand side of the
differential equation also includes the
derivatives of r(/). To illustrate the point
we consider the following example.

Example 4-9 Given the differential equation

^^ + 5
~dk + ~dT +
L
2c ^-^dr + 2r(t) (4-139)

it is desired to represent
the equation by three state equations. Since
the right side of
the state equations cannot include any
derivatives of the input r(t), it is necessary
to
include /•(?) when defining the state variables.
Let us rewrite Eq. (4-139) as

d3c(0 dr(t) ,d*c(t) dc(t) „,, „,

(4-140)

The state variables are now defined as

*i(0 = c(t) (4-141)

x2 (0=^ (4-142)

(4-143)

Using these last three equations and Eq. (4-140), the state equations are
written

dt
=**(»

^
dx 3 (t)
= *.(0+r(0
.
(4-144)

= ~ 2x iW ~ , .

x 2(0 - 5x 3 (t) -
~3J~ 3r(f)

In general, it can be shown that for the «th-order differential equation

dt-
+ dr->
+ +a"- i + a« c W
'
~dt

*.^ + dt"
... + *._ 1
^ +M0
dt
(4-145)

the state variables should be defined

as
114 / State-Variable Characterization of Dynamic Systems Chap. 4

x,(t) = c(t) - V(0

x 2 (t) = ^f>-h x
r{t)

xAt) = **M - h 2 r{t)

(4-146)

*„(0 = ^^-A 1
.-.'(0

where
A, = — a,6
bj,

h2 = {b — a b —
2 2 ) flj/71

>h = (* ™ a b - * A -
3 3 ) aJh (4-147)

h„ = (b„ — a„b ) — a„_ x h x — a„_ 2 h 2 — . . . -a 2 hn _2 - aj\ n _ x

Using Eqs. (4-146) and (4-147), we resolve the «th-order differential equa-
tion in Eq. (4-145) into the following n state equations:

^
d
= x 2 (r) + MO
=x 3 (t) + h 2 r(t)
-^f>

(4-148)

^M = xm (t) + h^r(0

^1 = -a„x,(t) - a„.,x 2 {t) - ... - fl 2 x._,(0 a iX „{t) + h„r(0

The output equation is obtained by rearranging the first equation of Eq. (4-146):

c(t) = x,(0 + b r{t) (4-149)

Now if we apply these equations to the case of Example 4-9, we have

a, = 5 b = b = 2 3

a = 2 6, = 1

a = 2
3
b = 2 1

/j,= 6; — tfj^o =
h = (b — a b — Vi =
2 2 2 ) •

hi = (b — a b — a h — ajx =
3 3 ) 2 x 2 3

When we substitute these parameters into Eqs. (4-146) and (4-147), we have
 Sec. 4.8 Relationship Between State Equations and Transfer Functions / 115

the same results for the state variables and the state equations as obtained in
Example 4-9.
The disadvantage with the method of Eqs. (4-146), (4-147), and (4-148)
is that these equations are difficult and impractical to memorize. It is not ex-

pected that one will always have these equations available for reference. How-
ever, we shall later describe a more convenient method using the transfer
function.

4.8 Relationship Between State Equations and Transfer Functions

We have presented the methods of describing a linear time-invariant system

by transfer functions and by dynamic equations. It is interesting to investigate
the relationship between these two representations.
In Eq. (3-3), the transfer function of a linear single-variable system is

defined in terms of the coefficients of the system's differential equation. Similar-

ly, Eq. (3-16) gives the matrix transfer function relation for a multivariate
system that has p inputs and q outputs. Now we shall investigate the transfer
function matrix relation using the dynamic equation notation.
Consider that a linear time-invariant system is described by the dynamic
equations

^ = Ax(0 + Br(?) (4-150)

c(/) = Dx(t) + Er(/) (4-151)

where
x(/) =n X 1 state vector

r (0 = P X I input vector

c(t) =q X \ output vector

and A, B, D, and E are matrices of appropriate dimensions.
Taking the Laplace transform on both sides of Eq. (4-150) and solving
for X(s), we have
X(s) = (si - A)" >x(0+) + (si ~ A)-'BR(s) (4-1 52)

The Laplace transform of Eq. (4-151) is

Q» = DX(j) + ER(s) (4-153)

Substituting Eq. (4-152) into Eq. (4-153), we have

C(s) = D(sl ~ A)-»x(0+) + D(jI - A)-'BR(5) + ER(s) (4-154)

Since the definition of transfer function requires that the initial conditions be
set to zero, x(0+) =
0; thus Eq. (4-154) becomes

C(s) = [D(sl - A)-'B + E]R(s) (4-1 55)

Therefore, the transfer function is defined as

G(s) = D(sl - A)~'B + E (4-156)

which is a q x p matrix. Of course, the existence of G(*) requires that the
matrix (si — A) is nonsingular.
 :

116 / State-Variable Characterization of Dynamic Systems Chap. 4

Example 4-10 Consider that a multivariable system is described by the differential

equations

dt*
+4 dt
iCl (4-157)

^+^ +2
dt
c1 +2c 2 =/- 2 (4-158)

The state variables of the system are assigned as follows

Xi = Ci (4-159)

x2 = dci
-^ (4-160)

x3 = c2 (4-161)

These state variables have been denned by mere inspection of the two differential

equations, as no particular reasons for the definitions are given other than that these
are the most convenient.
Now equating the first term of each of the equations of Eqs. (4-157) and (4-158) to
the rest of the terms and using the state-variable relations of Eqs. (4-159) through
(4-161), we arrive at the following state equation and output equation in matrix form:

dxi
1 xi
dt
ri
dx 2
dt
= 0-4 3 x2 + 1
Li- z J
(4-162)

dx 3
-1 -1 -2 x3 1
ldt_
~Xi~
pi~| ri 01
x2 Dx (4-163)
_C2. _o 1_
_*3_

To determine the transfer function matrix of the system using the state-variable
formulation, we substitute the A, B, D, and E matrices into Eq. (4-156). First, we form
the matrix (si — A),
~s -1
(sI-A)= j +4 -3 (4-164)

1 1 s + 2_
The determinant of (si — A) is
|jI-A| = s3 +6s 2 + 1] s + 3 (4-165)

Thus
~s z +6s + 11 +2
s 3 "

(.51 A) -| jI _ A -3 s(s + 2) 3i f4-1 fifi^

-(s + 1)
|

. -(* + 4) s(s+4)_

The transfer function matrix is, in tllis case,

GCs) = T>(sl - A)-'B

" s+2 3 (4-167)
1
~ S* + 6s 2 + Us + 3 _-(« + !) i(5 + 4)J
Using the conventional approach, we take the Laplace transform on both sides of
Eqs. (4-157) and (4-158) and assume zero initial conditions. The resulting transformed
 E

Sec. 4.9 Characteristic Equation. Eigenvalues, and Eigenvectors / 117

equations are written in matrix form as

sis + 4) -3 nrc,(j)-i
= VR^s)' (4-168)
_ s + 1 s + 2j[_C 2 is)j \_R 2 (s)_

Solving for C(s) from Eq. (4-168), we obtain

C(.r) = G(j)R(i) (4-169)

where
~
Sis + 4) -3
Gis) (4-170)
_ s +1 s + 2_
and the same result as in Eq. (4-167) is obtained when the matrix inverse is carried out.

4.9 Characteristic Equation, Eigenvalues, and Eigenvectors

The characteristic equation plays an important part in the study of linear

systems. It can be defined from the basis of the differential equation, the trans-
fer function, or the state equations.
Consider that a linear time-invariant system is described by the differential
equation

d"~ c u n d°~ 2 c
l
dc
df df-
"
ai
,

dF^
+ ,

• • • + ,

an -\ £ +ac ,

(4-171)

'df df
By defining the operator p as

P
k = Mi A; = 1,2, ... ,m (4-172)
dt
Eq. (4-171) is written

a 2 pn . .. + a^^ + ajc
(4-173)
= ib P" + *,/>""' + • • • + 1?-'P + b„)r
Then the characteristic equation of the system is defined as

s" + a^"- + 1
a 2 s"- 2 + . . . + o„-,J + an = (4-174)
which is setting the homogeneous part of Eq. (4-173) to zero. Furthermore, the
operator p is replaced by the Laplace transform variable s.

The transfer function of the system is

° - <&) - fro*" + M"" + 1

... + b . iS + b„ u -«
R(s) ~ s» + a^ +
G(s) n ,
{S)
~ . . . + a _ lS + a„
n
(4 '
175)

Therefore, the characteristic equation is obtained by equating the denominator

of the transfer function to zero.
From the state-variable approach, we can write Eq. (4-156) as

G( , ) = D adj(5l-A) B + E
si — A
I |

(4-176)
_ D[adj(sl - A)]B + |
jl -A |

UI-AI
118 / State-Variable Characterization of Dynamic Systems Chap. 4

Setting the denominator of the transfer function matrix G(s) to zero, we get
the characteristic equation expressed as

|
*I —A = | (4-177)

which is an alternative form of Eq. (4-174).

Eigenvalues

The roots of the characteristic equation are often referred to as the eigen-
values of the matrix A. It is interesting to note that if the state equations are
represented in the phase-variable canonical form, the coefficients of the char-
acteristic equation are readily given by the elements in the last row of the
elements of the A matrix.
That is, if A is given by Eq. (4-96), the characteristic
equation is readily given by Eq. (4-174).
Another important property of the characteristic equation and the eigen-
values is that they are invariant under a nonsingular transformation. In other
words, when the A matrix is transformed by a nonsingular transformation
x = Py, so that
A = P"'AP (4-178)

then the characteristic equation and the eigenvalues of A are identical to those
of A. This is proved by writing
si- A = ^I-P'AP (4-179)
or
si- A = ^""P-P-'AP (4-180)

The characteristic equation of A is

\sl- AlHsP-'P-P-'API
(4-181)
= |P-'(jI- A)P|
Since the determinant of a product is equal to the product of the determinants,
Eq. (4-181) becomes
UI- A| = ]P-MI^I- A||P|
= |*I-A|
Eigenvectors

The n X 1 vector p, which satisfies the matrix equation

a,I - A)p, = (4-183)

where X, is the rth eigenvalue of A, is called the eigenvector of A associated with

the eigenvalue X t . Illustrative examples of how the eigenvectors of a matrix
are determined are given in the following section.

4.10 Diagonalization of the A Matrix (Similarity Transformation)

One of the motivations for diagonalizing the A matrix is that if A is a diagonal

matrix, the eigenvalues of A, X X2 x , , , X„, all assumed to be distinct, are
located on the main diagonal; then the state transition matrix e K ' will also be
Sec. 4.10 Diagonalization of the A Matrix / 119

diagonal, with nonzero elements given by e Xl ', e M

its e*-'. There are other , . . . ,

reasons for wanting to diagonalize the A matrix, such as the controllability of

a system (Section 4. 1 5). We have to assume that all the eigenvalues of A are
distinct, since, unless it is real and symmetric, A cannot always be diagonalized
if it has multiple-order eigenvalues.

The problem can be stated as, given the linear system

x(0 = Ax(0 + Bu(f) (4-184)

where x(t) is an n- vector, u(t) an r- vector, and A has distinct eigenvalues X u

X2 , X„, it is desired to find a nonsingular matrix P such that the transfor-
. .
. ,

mation
x(0 = Py(?) (4-185)
transforms Eq. (4-184) into

y(0 = Ay(r) + ru(f) (4-186)

with A given by the diagonal matrix

A. ...

h ...

23 ... (n X n) (4-187)

... K
This transformation is also known as the similarity transformation. The state
equation of Eq. (4-186) is known as the canonical form.
Substituting Eq. (4-185) into Eq. (4-184) it is easy to see that

P 'AP (4-188)
and
P'B (n X r) (4-189)

In general, there are several methods of determining the matrix P. We show

in the following that P can be formed by use of the eigenvectors of A; that is,

P= [Pi P2 P 3 ...pj (4-190)

where p, (/ = 1,2, ... ,n) denotes the eigenvector that is associated with the
eigenvalue k t
. This is proved by use of Eq. (4-183), which is written

AiPi = Ap,- i= 1,2,. . . ,n (4-191)

Now forming the n x n matrix,

UiPi A 2p 2 . . . A„p„] = [Ap, Ap 2 . . . Ap„]

(4-192)
= A[p, p2 . . .
p„]
or

[Pi P2 • P„]A = A[p, p2 P.J (4-193)

Therefore, if we let

P= [Pi P2 • P.] (4-194)

120 / State-Variable Characterization of Dynamic Systems Chap. 4

Eq. (4-193) gives

PA = AP (4-195)
or
A = P-AP (4-196)

which is the desired transformation.

If the matrix A is of the phase-variable canonical form, it can be shown that
the P matrix which diagonalizes A may be the Vandermonde matrix,

1 1 ... 1

K
X\ X\ XI
(4-197)

X„~\

where X u X 2 , , X n are the eigenvalues of A.

Since it has been proven that P contains as its columns the eigenvectors
of A, we shall show that the rth column of the matrix in Eq. (4-197) is the
eigenvector of A that is associated with X„i= 1,2,..., n.

Let

Pn
P, (4-198)

Pin.

be the rth eigenvector of A Then .

U,i - A)p, == (4-199)

or
_
~x t
-1 Pn
x t
-1 Pn
x< -1
(4-200)

-1
_0„ a„-! a n -i a„- 3 x t + Cl\_ -Pin

This equation implies that

XiPn — Pn = °
XiPn — Pn = °
(4-201)

XtPi,„-l — Pin= °
fl»/»/i + a n-\Pn + (X t + a )p =
t tm
 Sec. 4.10 Diagonalization of the A Matrix / 121

Now we arbitrarily let pn = 1. Then Eq. (4-201) gives

Pn = X,

Pn = tf

(4-202)

1
Pt,n-1 xr
Pin Ar 1

which represent the elements of the rth column of the matrix in Eq. (4-197).
Substitution of these elements of p, into the last equation of Eq. (4-201) simply
verifies that the characteristic equation is satisfied.

Example 4-11 Given the matrix

1 on
A= i (4-203)
6 -11 -6_
which is the phase- variable canonical form, the eigenvalues of A are X = — 1, t

A 2 = —2, X 3 = —3. The similarity transformation may be carried out by use of the
Vandermonde matrix of Eq. (4-197). Therefore,
"1 1 1" 1 1

P = X\ "I A3 -2 -3 (4-204)
_Aj Ai A3 4 9

The canonical-form state equation is given by Eq. (4-186) with

"-1 "a,

A =piAP -2 A2 (4-205)
A3.

Example 4-12 Given the matrix

1 -r
A= 11 6 (4-206)
11 5j
itcan be shown that the eigenvalues of A are Xi = -1, k 2 = -2, and A = -3. It
3
isdesired to find a nonsingular matrix P that will transform A into a diagonal matrix
A, such that A = P~ 1 AP.
We shall follow the guideline that P contains the eigenvectors of A. Since A is not
of the phase-variable canonical form, we cannot use the Vandermonde matrix.
Let the eigenvector associated with a, = —1 be represented by

pn
Pi = Pn (4-207)

Then pi must satisfy

a,i - 1^1 == (4-208)

122 / State-Variable Characterization of Dynamic Systems Chap. 4

or
~X 1 -1 1 "
Pu
6 A, + 11 -6 Pi\ (4-209)

_6 11 A -5_t J>31.

The last matrix equation leads to

—Pu —Pzi + P3l =0

+ 10p 21 — 6p 31 =
6pu (4-210)

6p n + 11^21 — 6/»31 =0
from which we get^ 2 i = and/?n = p 3i Therefore, we . can letpn = /> 3 =
i 1, and
get

Pi (4-211)

For the eigenvector associated with A 2 = —2, the following matrix equation
must be satisfied
"A 2 -1 1 "
P12
6 A2 + ll -6 P11 (4-212)

6 11 A 2 -5. _/>32.

or
— 2?12 —^22 + />32 =
6p n + 9p — 6p 32 = 22 (4-213)

6/>i2 + HP22 — 7/? = 32

In these three equations we let/>i 2 = then/» 2 = 2 andp 32 =

1 ; 2 4. Thus
1

P2 = 2 (4-214)

4
Finally, for the eigenvector p 3 , we have
~X 3 -1 1
"
Pl3
6 A3 + 11 -6 P2 3 (4-215)

_6 11 A 3 -5. -P33.
or
— 3/> 13 — Pu + P33 =0
6p 13 + Sp 23 - 6p 33 = (4-216)

6pi + Up 2 — 8/? 33 =0
3 3

Now if we arbitrarily let p i3 = 1, the last three equations give P2i = 6 and p 33 =9.
Therefore,

p3 = (4-217)

The matrix P is now given by

"1
1 r
P = [Pi P2 P 3] 2 6 (4-218)

_1 4 9_
 Sec. 4.11 Jordan Canonical Form / 123

It is easy to show that

"A, 1 on
A=P 1
AP A2 -2 (4-219)

_0 A3 -3_

4.11. Jordan Canonical Form

In general when the A matrix has multiple-order eigenvalues, unless the

matrix symmetric and has real elements, it cannot be diagonalized. However,
is

there exists a similarity transformation

A = P'AP (« x ri) (4-220)

such that the matrix A is almost a diagonal matrix. The matrix A is called the
Jordan canonical form. Typical Jordan canonical forms are shown in the fol-
lowing examples

[A, 1

A, 1

A= A, (4-221)

A2
_0 A3
"A: 1

A,

A= A2 (4-222)

A3
_0 A4
The Jordan canonical form generally has the following properties:

1. The elements on the main diagonal of A are the eigenvalues of the

matrix.
2. All the elements below the main diagonal of are zero. A
3. Some of the elements immediately above the multiple-ordered
eigenvalues on the main diagonal are Is, such as the cases illustrated
by Eqs. (4-221) and (4-222).
4. The Is, together with the eigenvalues, form typical blocks which
are called Jordan blocks. In Eqs. (4-221) and (4-222) the Jordan
blocks are enclosed by dotted lines.
5. When the nonsymmetrical A
matrix has multiple-order eigenvalues,
its eigenvectors are not linearly independent. For an n X n A, there
is only r (r < n) linearly independent eigenvectors.
6. The number of Jordan blocks is equal to the number of independent
eigenvectors, r. There is one and only one linearly independent
eigenvector associated with each Jordan block.
7. The number of 1 s above the main diagonal is equal ton — r.
124 / State-Variable Characterization of Dynamic Systems Chap. 4

The matrix P is determined with the following considerations. Let us as-

sume that A has q distinct eigenvalues among n eigenvalues. In the first place,
the eigenvectors that correspond to the first-order eigenvalues are determined
in the usual manner from
(A I f
- A)p, = (4-223)

where A, denotes the /th distinct eigenvalue, / = l , 2, . . .

, q.

The eigenvectors associated with an mth-order Jordan block are deter-

mined by referring to the Jordan block being written as

"Ay 1 ...

A, 1 ...

(m X m) (4-224)

... A,- 1

.0 Xj_

where A, denotes the y'th eigenvalue.

Then the following transformation must hold:
"A y 1 ... 0"

Ay 1

[Pi Vz PJ A[p, p2 PJ (4-225)

o kj 1

Lo •• Ay.

or
AyPi = Ap,
p + AyP 2 = Ap 2

P2 + AyP 3
= Ap 3
(4-226)

Pm-i + A,p m = Ap m
The vectors Pi, p 2 , • •
, pm are determined from these equations, which can
also be written
(Ayl - A)Pl =
(Ayl - A)P = -P, 2

(Ayl - A)p = -P 3 2
(4-227)

(Ayl — A)p m = — P m _l
Example 4-13 Given the matrix
"0 6 -5"
A= 1 2 (4-228)

3 2 4_
Sec. 4.11 Jordan Canonical Form / 125

the determinant of Al — A is
A -6 5
IAI -Al -1 A -2 A3 - 4A 2 + 5A - 2
(4-229)
-3 -2 A -4
= (A - 2)(A - l) 2

Therefore, A has a simple eigenvalue at A! =2 and a double eigenvalue at A 2 = 1.

To find theJordan canonical form of A involves the determination of the matrix
that A = P AP. The eigenvector that is associated with X = 2 is deter-
_1
P such t

mined from
(A,I - A)p, = (4-230)
Thus
2 -6 5"
/>n
-1 2 -2 ^21 = (4-231)
-3 -2 -2_, _/>31_

Setting pn = 2 arbitrarily, the last equation gives p zl - 1 and pi i

-2. There-
fore,
2"

Pi (4-232)

For the eigenvector associated with the second-order eigenvalue, we turn to Eq.
(4-227). We have (the two remaining eigenvectors are p 2 and p 3 )
(A 2 I - A)p 2 = (4-233)
and
(A 2 I - A)p 3 = -p 2 (4-234)
Equation (4-233) leads to
1 -6 5"
P\i
-1 1 -2 Pn (4-235)
-3 -2 -3_ _/>3 2_

Setting p 12 = 1 arbitrarily, wehave/? 22 = — ^ andp 32

1"

Pz (4-236)

Equation (4-234), when expanded, gives

_
1 -6 - 5" Pl3 -r
-1 2 P23 = 3
7 (4-237)
-3 3_ _P33_ L
5
7J
from which we have
" 1"
Pl3
P3 Pl3 = -n (4-238)

_P33_
_46
Thus

-7 -n (4-239)

-t -li
126 / State-Variable Characterization of Dynamic Systems Chap. 4

The Jordan canonical form is now obtained as

~2 0"

A = P'AP 1 1 (4-240)

_0 1_

Note that in this case there are two Jordan blocks and there is one element of unity
above the main diagonal of A.

4.12 State Diagram

The signal flow graph discussed in Section 3.5 applies only to algebraic equations.
In this section we introduce the methods of the state diagram, which represents
an extension of the signal flow graph to portray state equations and differential
equations. The important significance of the state diagram is that it forms a
close relationship among the state equations, state transition equation, com-
puter simulation, and transfer functions. A state diagram is constructed fol-
lowing all the rules of the signal flow graph. Therefore, the state diagram may
be used for solving linear systems either analytically or by computers.

Basic Analog Computer Elements

Before taking up the subject of state diagrams, it is useful to discuss the

basic elements of an analog computer. The fundamental linear operations that
can be performed on an analog computer are multiplication by a constant,
addition, and integration. These are discussed separately in the following.

Multiplication by a constant. Multiplication of a machine variable by a

constant is done by potentiometers and amplifiers. Let us consider the opera-
tion
x 2 (t) = a Xl (t) (4-241)

where a is a constant. If a lies between zero and unity, a potentiometer is used

to realize the operation of Eq. (4-241). An operational amplifier is used to simu-
late Eq. (4-241)if a is a negative integer less than —1. The negative value

of a considered due to the fact that there is always an 180° phase shift between
is

the output and the input of an operational amplifier. The computer block
diagram symbols of the potentiometer and the operational amplifier are shown
in Figs. 4-5 and 4-6, respectively.

*i(0 x 2 (t) xAt) x 2 «)

<z>
x 2 (t) = ax l
(t) 0<a<\ x 2 {t)=ax {t)
x

Fig. 4-5. Analog-computer block-diagram Fig. 4-6. Analog-computer block diagram

symbol of a potentiometer. of an operational amplifier.
Sec. 4.12 State Diagram / 127

Algebraic sum of two or more variables. The algebraic sum of two or more
machine variables may be obtained by means of the operational amplifier.
Amplification may be accompanied by algebraic sum. For example, Fig. 4-7

*i(0
x 2 (t)
*- x,(t)
x 3 (t)

x 4 (0 = a, x l (t) + a 2 x 2 (t) + a } x 3 (t)

Fig. 4-7. Operational amplifier used as a summer.

illustrates the analog computer block diagram of a summing operational am-

plifier which portrays the following equation

x 4 (t) = a^At) + a 2 x 2 {t) +a 3 x 3 (t) (4-242)

Integration. The integration of a machine variable on an analog computer

is achieved by means of a computer element called the integrator. If x t (t) is the
output of the integrator with initial condition Xi0 ) given at t = t and x 2 (t)
is the input, the integrator performs the following operations

x,0) = f ax 2 {x)
•> to
dz + JCi(f ) a < 1 (4-243)

The block diagram symbol of the integrator is shown in Fig. 4-8. The integrator
can also serve simultaneously as a summing and amplification device.

*i('o)

x 2 (t)
-»- x,(t)

*i( f )= J ax 2 (r)dT + *,(?<))

Fig. 4-8. Analog computer block diagram of an integrator.

We shall now show that these analog computer operations can be portrayed
by signal flow graphs which are called state diagrams because the state vari-
ables are involved.
First consider the multiplication of a variable by a constant; we take the
Laplace transform on both sides of Eq. (4-241). We have

X (s) = aX^s)
2 (4-244)

The signal flow graph of Eq. (4-244) is shown in Fig. 4-9.

128 / State-Variable Characterization of Dynamic Systems Chap. 4

*i(0

*2(0 x 4 (t)

a
X2 (s) XAs)

x,(0 x 2 (t)

^i(s) X2 (s)

Fig. 4-9. Signal-flow-graph Fig. 4-10. Signal-flow-graph rep-

representation of x 2 (t) = resentation of Xi(i) = aix\{t)
axi(t) or X (s) =
2 aX\(s). + a 2 xz(t) + ajXi(i) orX4 (s) =
aiX^s) + a 2 X 2 (s) + ai X3 (s).

For the summing operation of Eq. (4-242), the Laplace transform equation
is

X (s) =
4 «,*,(*) + a 2 X2 (s) + a s X3 (s) (4-245)

The signal-flow-graph representation of the last equation is shown in Fig. 4-10.

It isimportant to note that the variables in the signal flow graphs of Figs. 4-9
and 4-10 may be in the time domain or the Laplace transform domain. Since
the branch gains are constants in these cases, the equations are algebraic in
both domains.
For the integration operation, we take the Laplace transform on both sides
of Eq. (4-243). In this case the transform operation is necessary, since the signal-
flow-graph algebra does not handle integration in the time domain. We have

X 2 (T)dT\ + *i(*o)
x,(t )
= a£ I x 2 {x) dx — x 2 (t) dx (4-246)
w o Jo

a£ I x 2 (r) dx

However, since the past history of the integrator is represented by x 2 (t ), and

the state transition starts from t = t x 2 (x) =0 for < x < t a Thus Eq.
, .

(4-246) becomes

X^s) = ^^ + ^5) x > to (4-247)

It should be emphasized that Eq. (4-247) is defined only for the period x > t .

Therefore, the inverse Laplace transform of X^s) in Eq. (4-247) will lead to
x,(0 of Eq. (4-243).
Equation (4-247) is now algebraic and can be represented by a signal
flow graph as shown in Fig. 4-11. An alternative signal flow graph for Eq.
(4-247) is shown in Fig. 4-12.
Sec. 4.12 State Diagram / 129

*i('o)
Q *i('o)

w 1 < 1

a as- 1
O— > <> Xi (s)
X2 is) Xds) X2 (s)
Fig. 4-11. Signal-flow-graph representation of Fig. 4-12. Signal-flow-graph rep-
Xi(s) = [aX 2 (s)ls] + [xiOo)/*] resentation of Xi(s) = [aX 2 (s)/s]
+ [xi(.to)ls].

Thus we have established a correspondence between the simple analog

computer operations and the signal-flow-graph representations. Since, as shown
in Fig. 4-12, these signal-flow-graph elements may include initial conditions
and can be used to solve state transition problems, they form the basic elements
of the state diagram.
Before embarking on several illustrative examples on state diagrams, let
us point out the important usages of the state diagrams.

1. A state diagram can be constructed directly from the system's

differential equation. This allows the determination of the state
variables and the state equations once the differential equation of
the system is given.
2. A state diagram can be constructed from the system's transfer
function. This step is defined as the decomposition of transfer
functions (Section 4.13).
3. The state diagram can be used for the programming of the system
on an analog computer.
4. The state diagram can be used for the simulation of the system on
a digital computer.
5. The equation in the Laplace transform domain
state transition
may be obtained from the state diagram by means of the signal-
flow-graph gain formula.
6. The transfer functions of a system can be obtained from the state
diagram.
7. The state equations and the output equations can be determined
from the state diagram.

The details of these techniques are given below.

From Differential Equation to the State Diagram

When
a linear system is described by a high-order differential equation, a
statediagram can be constructed from these equations, although a direct
approach is not always the most convenient. Consider the following differential
 :

130 / State-Variable Characterization of Dynamic Systems Chap. 4

equation
d"c „ d" l
c , dc ,
._
(4-248)
,
,

In order to construct a state diagram using this equation, we rearrange the equa-
tion to read
d"c _ d"- c l
dc
-a„. iw -a„c-rr (4-249)
dt"

o O o o < o o
n~l c
R s"C s C s"- 2 C

(a)

(b)

„("-!)/" „(n-2>, (1) (f c ( ? o)

)

(c)

Fig. 4-13. State diagram representation of the differential equation of Eq.

(4-248).
Sec. 4.12 State Diagram / 131

Let us use the following symbols to simplify the representation of the

derivatives of c:

=
C< "
W '=1.2,...,b (4-250)

Frequently, c in the literature, is used to represent dcjdt.

Now the variables r, c, c (1> , c (2) , c
(n>
are represented by nodes arranged
as shown in Fig. 4-1 3(a). In terms of Laplace transform, these variables are
denoted by R{s), C(s), sC(s), s 2 C(s), s"C(s), respectively.
. . . ,

As the next step, the nodes in Fig. 4- 13(a) are connected by branches to
portray Eq. (4-249). Since the variables c and c _I> are related through in-
(,'> ('

tegration with respect to time, they can be interconnected by a branch with gain
-1
s and the elements of Figs. 4-11 and 4-12 can be used. Therefore, the com-
plete state diagram is drawn as shown in Fig. 4- 13(c).
When the differential equation is that of Eq. (4-145), with derivatives of
the input on the right side, the problem of drawing the state diagram directly
is not so straightforward. We shallshow later that, in general, it is more con-
venient to obtain the transfer function from the differential equation first and
then obtain the state diagram through decomposition (Section 4.13).

Example 4-14 Consider the following differential equation:

g + 3| + 2c =r (4-251)

Equating the highest-ordered term of Eq. (4-251) to the rest of the terms, we have
d 2c

dT^- 2c - 3 dl
_ , dc
(4-252)

Following the procedure outlined above, the state diagram of the system is shown in
Fig. 4-14.

O c (1) a +)
9*o+)

" 1

Fig. 4-14. State diagram for Eq. (4-251).

From State Diagram to Analog Computer Block Diagram

was mentioned earlier that the state diagram is essentially a block dia-
It

gram programming of an analog computer, except for the phase reversal

for the
through amplification and integration.
 :

132 / State-Variable Characterization of Dynamic Systems Chap. 4

Example 4-15 An analog computer block diagram of the system described by Eq.
(4-251) is shown in Fig. 4-15. The final practical version of the com-

puter block diagram for programming may be somewhat different

from shown, since amplitude and time scaling may be necessary.

Fig. 4-15. Analog-computer block diagram for the system described by

Eq. (4-251).

From State Diagram to Digital Computer Simulation

The solution of differential equations by FORTRAN
on the digital com-
puter has been well established. However, from the standpoint of programming,
a convenient way of modeling a system on the digital computer is by CSMP
(Continuous System Modeling Program). 32 In many respects CSMP serves the
same purpose computer program, except that the scaling problem
as an analog
is practically eliminated. state diagram or the state equations form a
The
natural basis for the solution by CSMP. The following examples illustrate
typical CSMP statements for the mathematical equations listed:

Mathematical Equations CSMP Statements

c = a x
1 i + a 2x 2 C= A\ * XI + A2 * X2
x.
Y= JT1/2.

Xl = J
f' *,(t) dx + x,(0) XI = INTGRL(Z2, JT10)

Example 4-16 From the state diagram of Fig. 4-14, the following equations are
written

c = f t dt (4-253)

c=\cdt (4-254)

c = r - 3c - 1c (4-255)
 ,

Sec. 4.12 State Diagram / 133

Let the variables of these equations be denoted by

c =C c(0) = CO
c = CI <?(0) = CIO
c = C2
r =R
on the CSMP. Then the main program of the CSMP representation of the system is

given as follows:
C = INTGRL (CI, CO) (4-256)

CI= INTGRL (C2, CIO) (4-257)

C2 = R - 3. * CI - 2. * C (4-258)

From State Diagram to the State Transition Equation

We have shown earlier that the Laplace transform method of solving the
state equation requires the carrying out of the matrix inverse of (si — A).
With the state diagram, the equivalence of the matrix inverse operation is carried
out by use of the signal-fiow-graph formula.
The state transition equation in the Laplace transform domain is
X(s) = (jI - A)- !x(r J) + (jI - A)" 'BR(i) t > t (4-259)

Therefore, the last equation can be written directly from the state diagram by
use of the gain formula, with X^s), i = 1, 2, . . . , n, as the output nodes, and
*;('<0> * = 1,2, ... ,n, and Rj(s),j = 1, 2, . .
. p, as the input nodes.

Example 4-17 Consider the state diagram of Fig. 4-14. The outputs of the inte-
grators are assigned as state variables and the state diagram is
redrawn as shown in Fig. 4-16.

Fig. 4-16. State diagram for Eq. (4-251).

Applying the gain formula to the state diagram in Fig. 4-16, with X (s)
t and
X (s) as output nodes, *i(/J), x
2 2 (*o)> anc* R(s ) as input nodes, we have
"' 3;rl) -2 r -2
Xl (s) =J (1
+
— — >- + _L
v.(t*\
*l('o ) +— ^ XlOZ) + ^ *fr) (4-260)

-2s~
X (s) =
2 *iM)+V **('£) +V*(s) (4-261)
1 34 / State-Variable Characterization of Dynamic Systems Chap. 4

where
A= 1 +3s~ +2s~ 2 i
(4-262)

After simplification, Eqs. (4-260) and (4-261) are presented in matrix form:
1

1
s + 3 1 xiin) (s + l)(s + 2)
(j + IX* + 2)
+ R(s) (4-263)
Xz(s) -2 s xAn) 5
L(j + iXj + 2)J

The state transition equation for f > ? is obtained by taking the inverse Laplace
transform on both sides of Eq. (4-263).
Consider that the input r{t) is a unit step function applied at t = t . Then the
following inverse Laplace transform relationships are identified

£-» (-1) = u,{t - to) t>t (4-264)

£" '
(jTTi)
= «- "" «A' -h)
< )
t>t a (4-265)

The inverse Laplace transform of Eq. (4-263) is

~
2e~ (,- '» ) — e -2 - *' ' ' g-c-«o) — e -2d-r )

xiCt)
Mt). 2e -( '~'°'
+ 2e~ 2(t_ ' o) g-c-'o) -|_ 2e~ 2( '~'» ) x 2 (t$)_
(4-266)
t u s {t
- t ) - c" <'-'•>
+ £ e- 2 <'-'°>"

+ (l-lo) p-2(t-lo)
?> t

The reader should compare this result with that of Eq. (4-77), obtained for t > 0.
From State Diagram to Transfer Function

The transfer function between an input and an output is obtained from the
state diagram by setting all other inputs and all initial states to zero.

Example 4-18 Consider the state diagram of Fig. 4-16. The transfer function
C(s)/R(s) is obtained by applying the gain formula between these
two nodes and setting x^{t J) = and Xi(t J) == 0. Therefore,

C(s) 1
t_ (4-267)
R(s) s2 + 3s + 2
The characteristic equation of the system is

s2 + 3s + 2 = (4-268)

From State Diagram to the State Equations

When the state diagram of a system is already given, the state equations
and the output equations can be obtained directly from the state diagram by
use of the gain formula. Some
clarification seems necessary here, since the
determined from the state diagram by use of
state transition equations are
the gain formula. However, when writing the state transition equations from
the state diagram, the state variables, Xj(s), i = 1 , 2, . . . , n, are regarded as the
output nodes, and the inputs, Rj(s),j = 1,2, ... ,p, and the initial states,

xi(t )-> ' = 1» 2, regarded as the input nodes. Furthermore, the state
. .
.
, n, are
transition equations, as written directly from the state diagram, are necessarily
in the Laplace transform domain. The state transition equations in the time
domain are subsequently obtained by taking the inverse Laplace transform.
 Sec. 4.12 State Diagram / 135

The left side of the state equation contains the first-order time derivative
of the state variable x (t). The right side of the equation contains the state
t

variables and the inputvariables. There are no Laplace operator s or initial

state variables in a state equation. Therefore, to obtain state equations from
the state diagram, we should disregard all the initial states and all the integrator
branches with gains s~ l To avoid confusion, the initial states and the branches
.

with the gain s' 1 can actually be eliminated from the state diagram. The state
diagram of Fig. 4-16 is simplified as described above, and the result is shown in
Fig. 4-17. Then, using x, and x 2 as output nodes and *,, x 2 and , r as input nodes,

Fig. 4-17. State diagram of Fig. 4-16 with the initial states and the integra-
tor branches eliminated.

and applying the gain formula between these nodes, the state equations are
written directly:

(4-269)
^ = -2x
dx 2
dt
x
- 3x 2 +

Example 4-19 As another example of illustrating the determination of the state

equations from the state diagram, consider the state diagram shown
in Fig. 4-1 8(a). This illustration will emphasize the importance of
using the gain formula. Figure 4-1 8(b) shows the state diagram with the initial states

o-
r

(a)

Fig. 4-18. (a) State diagram.

136 / State-Variable Characterization of Dynamic Systems Chap. 4

O
r *3

(b)

Fig. 4-18 (Cont.). (b) State diagram in (a) with all initial states and inte-
grators eliminated.

and the integrators being eliminated. Notice that in this case the state diagram in Fig.
4-18(b) still contains a loop. Applying the gain formula to the diagram of Fig. 4-18(b)
withii,;c 2 and x 3 as the output node variables and r,x\,x 2 and x 3 as the input
, ,

nodes, the state equations are determined as follows

dxi
1 Xi
dt

dx 2 — (a 2 + a 3) 1 O «2
x2 (4-270)
dt 1 +aa 3 1 + aa a 3
dx 3
*3
dt

4.13 Decomposition of Transfer Functions

Up until this point, various methods of characterizing a linear system have

been presented. be useful to summarize briefly and gather thoughts at
It will

this point, before proceeding to the main topics of this section.

It has been shown that the starting point of the description of a linear

system may
be the system's differential equation, transfer function, or dynamic
equations. Itis demonstrated that all these methods are closely related. Further,

the state diagram is shown to be a useful tool which not only can lead to the
solutions of the state equations but also serves as a vehicle of translation from
one type of description to the others. A block diagram is drawn as shown in
Fig. 4-19 to illustrate the interrelationships between the various loops of de-
scribing a linear system. The block diagram shows that starting, for instance,
with the differential equation of a system, one can get to the solution by use of
the transfer function method or the state equation method. The block diagram
also shows that the majority of the relationships are bilateral, so a great deal
of flexibility exists between the methods.
 :

Sec. 4.13 Decomposition of Transfer Functions / 137

Differential Dynamic
equations equations

1
'
,

State
transition
equation
^

Transfer State
function diagram

Fig. 4-19. Block diagram showing the relationships among various

methods of describing linear systems.

One
step remains to be discussed. This involves the construction of the
statediagram from the transfer function. In general, it is necessary to establish
a better method than using Eqs. (4-146) through (4-148) in getting from a
high-order differential equation to the state equations.
The process of going from the transfer function to the state diagram or the
state equations is called the decomposition of the transfer function. In general,
ways of decomposing a transfer function direct decomposi-
there are three basic :

tion,cascade decomposition, and parallel decomposition. Each of these three

schemes of decomposition has its own advantage and is best suited for a par-
ticular situation.

Direct Decomposition

The direct decomposition scheme is applied to a transfer function that

isnot in factored form. Without loss of generality, the method of direct de-
composition can be described by the following transfer function

C(s) a s2 +as+a 2
(4-271)
R(s) b s* +bs+b x 2

The objective is to obtain the state diagram and the state equations. The fol-
lowing steps are outlined for the direct decomposition:

1. Alter the transfer function so that it has only negative powers

of s. This
accomplished by multiplying the numerator and the
is

denominator of the transfer function by the inverse of its highest

power For the transfer function of Eq. (4-271), we multiply
in j.

the numerator and the denominator of C{s)jR{s) by s~ 2 .

2. Multiply the numerator and the denominator of the transfer func-

tion by a dummy variable X(s). Implementing steps 1 and 2, Eq.
(4-271) becomes

C(s) _ a + a s + a s~
t
'
2
2
X{s)
(4-272)
R(s) b + b,s' + bh-*~
1
s~ 2
2
X{s)

3. The numerators and the denominators on both sides of the transfer

 .

138 / State-Variable Characterization of Dynamic Systems Chap. 4

function resulting from steps 1 and 2 are equated to each other,

respectively. From Eq. (4-272) this step results in

C(s) = (fl a,s~ a 2 s- 2 )X(s) (4-273)

R(s) = (b + *,*-» b t s-*)X(s) (4-274)

4. In order to construct a state diagram using these two equations,

they must first be in the proper cause-and-effect relation. It is ap-

parent that Eq. (4-273) already However,

satisfies this prerequisite.

Eq. (4-274) has the input on the and must be rearranged.

left side

Dividing both sides of Eq. (4-274) by b and writing X(s) in terms of

the other terms, we have
b2
X{s) 'X(s) (4-275)
ba o b

The state diagram is now drawn in Fig. 4-20 using the expressions in
Eqs. (4-273) and (4-275). For simplicity, the initial states are not drawn on the
diagram. As usual, the state variables are defined as the outputs of the integra-
tors.

C(s)
R(s)

Fig. 4-20. State diagram for the transfer function of Eq. (4-271) by direct

decomposition.

Following the method described in the last section, the state equations are
written directly from the state diagram:

dxi 1 Xl
dt (4-276)
-b 2
+
dx 2 -6i x2
1

.dt b J

The output equation is obtained from Fig. 4-20 by applying the gain for-
mula with c(t) as the output node and x^t), x 2 {t), and r(t) as the input nodes.

H*-^)*' + (*'-*£)** Hi' (4 - 277)

 : : :

Sec. 4.13 Decomposition of Transfer Functions / 139

Cascade Decomposition

Cascade decomposition may be applied to a transfer function that is in

the factored form. Consider that the transfer function of Eq. (4-271) may be
factored in the following form (of course, there are other possible combinations
of factoring)

g£i = ^l±Iil±Ji (4-278)

b s + p s + p
\
R(s) t 2

where z u z 2 ,p u and p 2 are real constants. Then it is possible to treat the func-
tion as the product of two The state diagram of
first-order transfer functions.
each of the first-order transfer functions is realized by using the direct de-
composition method. The complete state diagram is obtained by cascading the
two first-order diagrams as shown in Fig. 4-21. As usual, the outputs of the

R(.s)

~Pi
Fig. 4-21. State diagram of the transfer function of Eq. (4-278) by cascade
decomposition.

integrators on the state diagram are assigned as the state variables. The state
equations are written in matrix form

dx-i r ~
a
-Pi Pi x i
dt b
= + (4-279)
ctx^ a
-Pi •*2
Ldt \ _*oJ
The output equation is

c = (z 2 p 2 )x +(z,
l
- Pi)x + 2 % (4-280)

The cascade decomposition has the advantage that the poles and zeros of
the transfer function appear as isolated branch gains on the state diagram.
This facilitates the study of the effects on the system when the poles and zeros
are varied.

Parallel Decomposition

When the denominator of a transfer function is in factored form, it is

possible toexpand the transfer function by partial fractions. Consider that a
second-order system is represented by the following transfer function

Cjs)_ Pis)
(4-281)
R(s) (s + Pl )(s + p 2)

where P(s) is a polynomial of order less than 2. We assume that the poles p x
140 / State-Variable Characterization of Dynamic Systems Chap. 4

and p 2 may be complex conjugate for analytical purposes, but it is difficult to

implement complex coefficients on the computer.
In this case if p and p 2 are equal it would not be possible to carry out a
x

partial-fraction expansion of the transfer function of Eq. (4-281). With p and t

p2 being distinct, Eq. (4-281) is written

C(s) Kr K,
(4-282)
R(s) s + Pi s +p 2

where K x
and K
2 are constants.
The diagram for the system is formed by the parallel combination of
state
the state diagram representation of each of the first-order terms on the right
side of Eq. (4-282), as shown in Fig. 4-22. The state equations of the system are
written
dxC
dt
—P\
= (4-283)
dx 2
-Pi X%
\_dt _

X!(f +)

R(s)

Fig. 4-22. State diagram of the transfer function of Eq. (4-28 1) by parallel
decomposition.

The output equation is

c = [K, K 2] (4-284)

One of the advantages of the parallel decomposition is that for transfer

functions with simple poles, the resulting A matrix is always a diagonal matrix.
Therefore, we can consider that parallel decomposition may be used for the
diagonalization of the A matrix.
When a transfer function has multiple-order poles, care must be taken
that the state diagram, as obtained through the parallel decomposition, con-
tain the minimum number of integrators. To further clarify the point just made,
consider the following transfer function and its partial-fraction expansion:
 Sec. 4.14 Transformation into Modal Form / 141

an 2s 2 + 6s +5 1
+ s+l
1 1
(4-285)
R(s) (s + iy(s + 2) (s + iy
Note that the transfer function is of the third order, and although the total order
of the terms on the right side of Eq. (4-285) is four, only three integrators should
be used in the state diagram. The state diagram for the system is drawn as shown
in Fig. 4-23. The minimum number of three integrators are used, with one in-

Fig. 4-23. State diagram of the transfer function of Eq. (4-285) by

paralled decomposition.

tegrator being shared by two channels. The state equations of the system are
written

-1 1 *1
dt

dx 2
dt
0-1 x2 + 1 (4-286)

dx 3
dt
0-2 x3 1

Therefore, the A matrix is of the Jordan canonical form.

4.14 Transformation into Modal Form

When the A matrix has complex eigenvalues it may not be possible to trans-
form it into a diagonal matrix with real elements. To facilitate computer com-
putation, it is desirable to avoid matrices with complex elements. When A has

complex eigenvalues, in general, the matrix can be transformed into a non-

diagonal matrix which is called the modal form by the transformation

A = P"'AP (4-287)

Let us assume that A is 2 x 2 and has eigenvalues A, = a + jco and X 2 = a

— jco. Then the modal-form matrix is given by
142 / State-Variable Characterization of Dynamic Systems Chap. 4

CO
A= (4-288)
— CO
The elements of the P matrix may be determined by brute force using Eqs.
(4-287) and (4-288). If A has m real distinct eigenvalues in X u X 2 X m and , . . . , ,

n sets of complex-conjugate eigenvalues in X, = a ± jcoi, i = 1 2, «, t

, . . . ,

the modal-form matrix is given by

A, .. . .

x2 . .

Am .. o
A=
•

(4-289)
.. A x
..

A 2 .. o

.. A„
where

(4-290)
\—CO, On
If they'th complex eigenvalue pair is of multiplicity m, then A y
is written

Tj I ...

o r, i

o o r, (m X m blocks) (4-291)

where
ooo
" Oj COj~
(4-292)
i-COj CTj]

ri oi
(4-293)
Lo lj

The modal-form matrix in Eq. (4-289) is easily modified for real and multiple-
order eigenvalues by use of the Jordan canonical form.
Although the modal-form matrix is not diagonal and does not represent
decoupling of the states from the standpoint of the state diagram, it still has
the components of the eigenvalues as its matrix elements.
To determine the transformation matrix P for the matrix A of Eq. (4-288),
we let

P= [Pl pj (4-294)

where p and p 2 are 2 x

(
1 vectors. Equation (4-287) is written

a co
[Pi P2] A[p, p2 ] (4-295)
-co a _
Sec. 4.14 Transformation into Modal Form / 143

or
ffPi — cop = Ap! 2 (4-296)

toPi + cq>2 = Ap 2 (4-297)

These two equations are expressed in matrix equation form,

<rl -coll ~pr "A 0~

Pi
(4-298)
col rf_ J>2_ _0 A_ _P 2

where I denotes a 2 x 2 identity matrix.

Let qi and q 2 denote the eigenvectors that are associated with the two
complex-conjugate eigenvalues, X 1 a jco and X 2 = +
a —ja>, respectively. =
Then, according to the definition of eigenvectors, q t and q 2 must satisfy

(^+yco)q 1
=Aq 1 (4-299)

(<r - jco)q 2
= Aq 2 (4-300)
Let
q, = <X, (4-301)

q2 = «2 (4-302)

Then Eqs. (4-299) and (4-300) become

(<7+ ycoXa, + yp,) = A(a, + #,) (4-303)

{a - jco)(a + yp = A(a + j%)

2 2) 2 (4-304)

Equating the real and imaginary parts in the last two equations, we have

ctI —col "«i "A 01 r«ii

(4-305)
_coI ctI LpJ A_ LpJ
and
a\ coT "«2~ A 0" "«2~
(4-306)
_— col <rl_ LP 2 J _0 A_ LpzJ
Comparing Eq. (4-305) with Eq. (4-298), we have the identity

P=[P. P2] = [«i Pi] (4-307)

The significance of this result is that the transformation matrix P is formed

by taking the real and imaginary components of the eigenvector of A associated
with Xi = a + jco.
Example 4-20 Consider the state equation

i = Ax + Br (4-308)
where
1"
B=
-2 -2
The eigenvalues ofA are A i
= —1 + j and X = — —/ The eigenvectors are
2 l

1 l
qi q2
_-i -J.
or

qi = a! +/Pi q2 = a 2 +/P2 = + J -1
144 / State-Variable Characterization of Dynamic Systems Chap. 4

Therefore,
i
P = [*i Pt] = r °i (4-309)
_-i i-

A= P-!AP = r-i i~|

(4-310)
-i ~i_

T = piB = (4-311)

The original state equation of Eq. (4-308) is transformed to

y = Ay + Tr (4-312)

Then the state transition matrix is given by

~ cos sin i
${t) = eM = e~<
t
(4-313)
—sin t cos /

4.15 Controllability of Linear Systems

The concepts of controllability and observability introduced first by Kalman 24

play an important role in both theoretical and practical aspects of modern
control theory. The conditions on controllability and observability often govern
the existence of a solution to an optimal control problem, particularly in mul-
tivariable systems. However, one should not associate these conditions with
the concept of stability. Furthermore, not all optimal control problems require

that the system be controllable and/or observable in order to achieve the con-
trol objectives. These points will be clarified as the subjects are developed. In
this section we shall first treat the subject of controllability.

General Concept of Controllability

The concept of controllability can be stated with reference to the block

diagram of Fig. 4-24. The process G is said to be completely controllable if

Control u(0 State x(t)

Fig. 4-24. Linear time-invariant system.

every state variable of G can be affected or controlled to reach a certain objec-

tive in finite time by some unconstrained control u(t). Intuitively, it is simple
to understand that any one of the state variables is independent of the control
if

u(r), there would be no way of driving this particular state variable to a desired
state in finite time by means of a control effort. Therefore, this particular state
is said to be uncontrollable, and as long as there is at least one uncontrollable

state, the system is said to be not completely controllable, or simply uncon-

trollable.
 Sec. 4.15 Controllability of Linear Systems / 145

?*i('o+) ?*i (>o+)

u(t)
O c

Fig. 4-25. State diagram of a system that is not state controllable.

As asimple example of an uncontrollable system, Fig. 4-25 illustrates the

state diagram of a linear system with two state variables. Since the control
i/(0 affects only the state x^t), x 2 (t) is uncontrollable. In other words, it would
be impossible to drive x 2 (t) from an initial state x 2 (t ) to a desired state x 2 (tf)
in a finite time interval tf — ta by any control u(t). Therefore, the entire system
is said to be uncontrollable.
The concept of controllability given above refers to the states and is some-
times referred to as the state controllability. Controllability can also be defined
for the outputs of a system, so there is a difference between state controllability
and output controllability.

Definition of Controllability (State Controllability)

Consider that a linear time-invariant system is described by the following

dynamic equations :

x(0 = Ax(0 + Bu(0 (4-314)

c(0 = Dx(0 + Eu(0 (4-315)

where
x(0 =n x 1 state vector

u(0 = r X 1 input vector

c(0 = p x 1 output vector

A=n X n coefficient matrix

B=n X r coefficient matrix

D =p x n coefficient matrix

E =p x r coefficient matrix

The state x{i) is said to be controllable at t =

/„ // there exists a piecewise
continuous input u(r) that will drive the state to any final state x(t for a finite
f)
time {t f —t ) >
0. If every state x(t ) of the system is controllable in a finite time

interval, the system is said to be completely state controllable or simply state

controllable.
146 / Stale-Variable Characterization of Dynamic Systems Chap. 4

The following theorem shows that the condition of controllability depends

on the coefficient matrices A and B of the system. The theorem also gives one
way of testing state controllability.

Theorem 4-2. For the system described by the state equation of Eq. (4-314)
tobe completely state controllable, it is necessary and sufficient that the following
n X nr matrix has a rank of n:

S = [B AB A 2
B. . . A"-'B] (4-316)

Since the matrices A and B are involved, sometimes we say that the pair
[A, B] is controllable, which implies that S is of rank n.

Proof: The solution of Eq. (4-314) is

x(?) = <j>(? - t )x(t ) + f ftt - t)Bu(t) dx (4-317)

•> to

for t > tQ . Without losing any generality we can assume that the desired final
state for some finite t
f > t is x(t f ) = 0. Then Eq. (4-317) gives

x('o) =~ f" Mo - t)Bu(t) dx (4-318)

From the Caley-Hamilton's theorem, 21

A* = 2 a* m A
m for any k (4-319)

Then the state transition matrix can be written

<K0 = e
A '= S A"t
*=o k} icl
(4-320)
= Zj TT 2-1 &km™-
k= K- m=0
or
n-l
2 a km U
oo ,k
ftf) = "t A" (4-321)
m-0 k=0 ft!

Thus <£(?) can be written in the form

<K0 = 2 «»«A" (4-322)

Substituting Eq. (4-322) into Eq. (4-318) and rearranging, we have

x(*- ) = - 2 Am B f"
a m (f - t)u(t) </t (4-323)

Let

U m = r« m 0o-^)u(T)^T
J
(4-324)
to

Then Eq. (4-323) becomes

x('o) = - 2 A m BU m (4-325)
771 =

which is written in matrix form:

Sec. 4.1 5 Controllability of Linear Systems / 147

x('o) = -[» AB A B 2
. . . A""'B]U (4-326)

= -SU
where
U= [U U....U,.,]' (4-327)

Equation (4-325) represents n equations with nr unknowns, and the con-

trollability problem may be interpreted as: Given any initial state x(t ), find
the control vector a(t) so that the final state is x(^) = for finite f — This t t .

implies that given x(f ) and the matrix S, solve U from Eq. (4-326). Therefore,
the system is completely state controllable if and only if there exists a set of n

linearly independent column vectors in S. For a system with a scalar input,

r = 1, the matrix S is square; then the condition of state controllability is that
S must be nonsingular.
Although the criterion of state controllability given by Theorem 4-2 is
quite straightforward, it is not very easy to implement for multiple-input sys-
tems. Even with r = 2, there are In columns in S, and there would be a large
number of possible combinations of n x n matrices. A practical way may be
to use one column of B at a time, each time giving an n X n matrix for S. How-
ever, failure to find an S with a rank for n this way does not mean that the sys-
tem is uncontrollable, until all the columns of B are used. An easier way would
be to form the matrix SS', which is n X n then if SS' is nonsingular, S has ;

rank n.

Example 4-21 Consider the system shown in Fig. 4-25, which was reasoned earlier to
be uncontrollable. Let us investigate the same problem using the
condition of Eq. (4-316).
The state equations of the system are written, from Fig. 4-25,

r«ki(Oi
dt
-2 1 X iW 1

dx 2 (t)
+ u(t) (4-328)
-1 x%{t)
I dt J

Therefore, from Eq. (4-316),

"1 -2"
S = [B AB] = (4-329)
_o o.

which is singular, anc the sys tem

. is not state cc>ntro llab e.

Example 4-22 Determine the state controllability of the system described by the
state equation

dxM
dt
1 *i(0
dx 2 Q)
+ "(0 (4-330)
-1 xi(0
. dt J
From Eq. (4-316),
1
S = [B AB] (4-331)
1

which is nonsingular. Therefore, the system is completely state controllable.

148 / State-Variable Characterization of Dynamic Systems Chap. 4

Alternative Definition of Controllability

Consider that a linear time invariant system is described by the state equation

i(0 = Ax(0 + Bu(?) (4-332)

If the eigenvalues of A are distinct and are denoted by X„ i = \, 2, . . . , n,

then there exists an nth-order nonsingular matrix P which transforms A into a

diagonal matrix A such that
"A, ...

A,
P'AP = (4-333)

0.
Let the new state variable be

y = P-'x (4-334)

Then the state equation transformed through P is

y = Ay + Tii (4-335)

where
r = p-'B (4-336)

The motivation for the use of the similarity transformation is that the
states of the system of Eq. (4-335) are decoupled from each other, and the only
way the states are controllable is through the inputs directly. Thus, for state
controllability, each state should be controlled by at least one input. Therefore,
an alternative definition of state controllability for a system with distinct
eigenvalues is: The system is completely state controllable if T has no rows that

are all zeros.

It should be noted that the prerequisite on distinct eigenvalues precedes
the condition of diagonalization of A. In other words, all square matrices with

distinct eigenvalues can be diagonalized. However, certain matrices with

multiple-order eigenvalues can also be diagonalized. The natural question is:

Does the alternative definition apply to a system with multiple -order eigenvalues
but whose A matrix can be diagonalized? The answer is no. We must not lose
sight of the original definition on any state x(t ) is
state controllability that
brought to any state \(t f ) in finite time. Thus the question of independent con-
trol must enter the picture. In other words, consider that we have two states
which are uncoupled and are related by the following state equations:

^ = a Xi (t) + bMO (4-337)

^ = ax&) b 2 u{t) (4-338)

Sec. 4.15 Controllability of Linear Systems / 149

This system is apparently uncontrollable, since

~b, ab
S = [B AB]
l
(4-339)
bz ab 2

is singular. Therefore, just because A is diagonal, and B has no rows which

are zeros does not mean that the system is controllable. The reason in this case
is that A has multiple-order eigenvalues.
When A has multiple-order eigenvalues and cannot be diagonalized, there
is P which transforms A into a Jordan canonical form
a nonsingular matrix
A = P'AP. The condition of state controllability is that all the elements of
r = P _1 B that correspond to the last row of each Jordan block are nonzero.
The reason behind this is that the last row of each Jordan block corresponds
to a state equation that is completely uncoupled from the other state equations.
The elements in the other rows of T need not all be nonzero, since the cor-
responding states are all coupled. For instance, if the matrix A has four eigen-
values, X it X u X u Xi, three of which are equal, then there is a nonsingular P
which transforms A into the Jordan canonical form
A, 1
0~

A, 1
P AP = '

!
(4-340)
A,

A2 J
Then the condition given above becomes self-explanatory.

Example 4-23 Consider the system of Example 4-21. The A and B matrices are,
respectively,
"-2 r
B=
T
L o -i_ l_oJ

Let us check the controllability of the system by checking the rows of the matrix T.
It can be shown that A is diagonalized by the matrix

p=
l r
_o
Therefore,

r = 'B = 1 -r r = "1"
(4-341)
_o i_ _o_ _0_
The transformed state equation is

~-2 0~
no y(0 + u(t) (4-342)
_0_

Since the second row of T is zero, the state variable yi(t), or x 2 (0» *s uncontrollable,
and the system is uncontrollable.

Example 4-24 Consider that a third-order system has the coefficient matrices

1 2 -r "0

A= 1 B=
1 -4 3 1
150 / State-Variable Characterization of Dynamic Systems Chap. 4

Then
-1 -41
S = [B AB A 2
B] (4-343)
1 3 8_

Since S is singular, the system is not state controllable.

Using the alternative method, the eigenvalues of A are found to be Ai = 2,
A2 = 2, and A 3 = 1. The Jordan canonical form of A is obtained with
1
0"

1 (4-344)
1 1 2_
Then
[2 1 01
A = P'AP 2 (4-345)

_0 1_

T = P'B (4-346)

Since the last row of T is zero, the state variable y3 is uncontrollable. Since x2 =y 3,

this corresponds to x 2 being uncontrollable.

Example 4-25 Determine the controllability of the system described by the state
equation

= o r
i(0 x(0 + u(t) (4-347)
-l o
We form the matrix
1"
S = [B AB] (4-348)
1

which is nonsingular. The system is completely controllable.

Let us now check the controllability of the system from the rows of T. The
eigenvalues of A are complex and are k =j and X 2 = x —j. With the similarity trans-
formation,
~
"1 1
P
J -i-

A =
P ]
AP = 7 "

_o — /_
and

r= p-'b =
L2/J
Since all the rows of T are nonzero, the system is controllable.
In general, when the eigenvalues are complex, which occurs quite frequently in
control systems, it is more difficult to work with complex numbers. However, we may
Sec. 4.15 Controllability of Linear Systems / 151

use the modal form so that only real matrices are dealt with. In the present problem
A may be transformed to the modal form

a CO" r (4-349)
_—a> a_ .-1 o_
by the transform matrix

P =
"1
-r
_1 i_
Then

r=p 'B = 2"

1
-1-

Since the modal form A implies that the states are coupled, the condition of control-
lability is that not all the rows of T are zeros.

Output Controllability 1 *

The condition of controllability defined in the preceding sections is referred

only to the states of a system. Essentially, a system is controllable if every
desired transition of the states can be effected in finite time by an unconstrained
control. However, controllability defined in terms of the states is neither neces-
sary nor sufficient for the existence of a solution of the problem of controlling
the outputs of the system.

Definition of output controllability. A system is said to be completely

output controllable if there exists a piecewise continuous function u(t) that will
drive the output y(? ) at t = t to any final output y(f r ) for a finite time (t f — t )

>0.

Theorem 4-3. Consider that an nth-order linear time-invariant system is

described by the dynamic equations of Eqs. (4-314) and (4-315). The system is
completely output controllable if and only if the p X (n T)r matrix +
T = [DB DAB DA 2 B. . . DA nl B E] (4-350)

is of rank p. Or, T has a set of p linearly independent columns.

'

The proof of this theorem is similar to that of Theorem 4-2.

Example 4-26 Consider a linear system whose input-output relationship is described

by the differential equation

d 2 c(t)
dt 2
2
dc(t)
~dT + c{t)
, .
_ du{i)
~ -dT m (4-351)

The state controllability and the output controllability of the system will be investi-
gated. We shall show that the state controllability of the system depends upon how the
state variables are defined.
Let the state variables be defined as

x x
= c

x2 = c — u
 . :

152 / State-Variable Characterization of Dynamic Systems Chap. 4

The state equations of the system are expressed in matrix form as

=
"
o r i
+ (4-352)
_*2_ _-i -2_ X%- -l

The output equation is

C = X\ (4-353)
The state controllability matrix is
~ 1 r
S = B [
AB] = (4-354)
_-l i.

which is singular. The system is not state controllable.

From the output equation, D= [1 0] and E= 0. The output controllability
matrix is written
T = [DB DAB E] = [1 -1 0] (4-355)

which is of rank 1, the same as the number of output. Thus the system is output con-
trollable.
Now let us define the state variables of the system in a different way. By the
method of direct decomposition, the state equations are written in matrix form
"0"
*1
=
"
o r
["1 + (4-356)
_*2_ .-1 -2. L*2_ _1_

The output equation is

C = Xi +x 2 (4-357)

The system is now completely state controllable since

= =
"0 r
S [B AB] (4-358)
H _i —
-2,
ZJ
which is nonsingular.
The system is still output controllable since

T = [DB DAB E] = [1 -1 0] (4-359)

which is of rank 1

We have demonstrated through this example that given a linear system, state
controllability depends on how the state variables are defined. Of course, the output
controllability is directly dependent upon the assignment of the output variable. The
two types of controllability are not at all related to each other.

4.16 Observability of Linear Systems

The concept of observability is quite similar to that of controllability. Essen-

tially, a system is completely observable if every state variable of the system
affects some of the outputs. In other words, it is often desirable to obtain in-
formation on the state variables from measurements of the outputs and the
inputs. If any one of the states cannot be observed from the measurements of
the outputs, the state is said to be unobservable, and the system is not complete-
ly observable, or is simply unobservable. Figure 4-26 shows the state diagram of
a linear system in which the state x 2 is not connected to the output c in any way.
Once we have measured c, we can observe the state x since x, = c. However, x ,

x 2 cannot be observed from the information on c. Thus the system is

the state
described as not completely observable, or simply unobservable.
 Sec. 4.16 Observability of Linear Systems / 153

?*2('0+) ?-M'o+)

u(t)
O c

Fig. 4-26. State diagram of a system that is not observable.

Definition of observability. Given a linear time-invariant system that is

described by the dynamic equations of Eqs. (4-3 14) and (4-315), the state \(t )
is said to be observable if given any input u(f), there exists a finite time t tQ
f >
such that the knowledge ofu(t)for t <t <
t ; the matrices A, B, D, and E; and
f
the output c(t) for t <t<t
f are sufficient to determine x(t ). If every state of
f we say that the system is completely ob-
the system is observable for a finite t ,

servable, or simply observable.

The following theorem shows that the condition of observability depends
on the coefficient matrices A and D of the system. The theorem also gives one
method of testing observability.

Theorem 4-4. For the system described by the dynamic equation of Eqs.
(4-314) and (4-315) to be completely observable, it is necessary and sufficient
that the following n x np matrix has a rank of n:

V= [D' A'D' (A') 2 D\ . . (A')"- 1 D'] (4-360)

The condition is also referred to as the pair [A, D] being observable. In

particular, if the system has only one output, is an 1 X n matrix; Dof Eq. V
(4-360) is an n X n square matrix. Then the system is completely observable if V
is nonsingular.

Proof: Substituting Eq. (4-317) into Eq. (4-315), we have

c(/) = B(f>(t - t )x(t ) +D f <f>(?

- t)Bu(t) dx + Eu(?) (4-361)

Based on the definition of observability, it is apparent that the observability

of x(f) depends essentially on the first term of the right side of Eq. (4-361).

With u(t) = 0, Eq. (4-361) becomes

c(0 = D$(r - t )x(t ) (4-362)

Making use of Eq. (4-322), Eq. (4-362) becomes

c(0 = S a m (0DA-x(r ) (4-363)

m=
 :

154 / State-Variable Characterization of Dynamic Systems Chap. 4

or
D
DA
DA 2

c(0 = (a I a I t «„-,!) xOo) (4-364)

DA"
Therefore, knowing the output c(t) over the time interval t < <
t t
f , x(t )
is uniquely determined from Eq. (4-364) if and only if the matrix

D
DA
DA 2

(np X w)

DA" 1

has rank n. Or the matrix

V= [D' A'D' (A') 2 D . . . (A')"-'D'] (4-365)

has a rank of n.

Comparing Eq. (4-360) with Eq. (4-316) and the rank condition, the
following observations may be made

1. Controllability of the pair [A, B] implies observability of the pair

[A', B'].
2. Observability of the pair [A, B] implies controllability of the pair
[A', B'].

Example 4-27 Consider the system shown in Fig. 4-26, which was earlier defined to
be unobservable. The dynamic equations of the system are written
directly from the state diagram.

'-2 0" ~3"

= + (4-366)
-1_ L*2_ _1_

~x{~
t [1 0] (4-367)
L*2j
_*2_
Therefore,

D= [1 0] D'

-2 0"
A'D'
-1
and, from Eq. (4-360),
"1 -2'
V= [D' A'D] = (4-368)

Since V is singular, the system is unobservable.

 :

Sec. 4.16 Observability of Linear Systems / 155

Example 4-28 Consider the linear system described by the following dynamic
equations

1 -1" Xl
(4-369)
L*2. 1 1. Xl.

ci 1
(4-370)
-1 1 Lxil
For the test of observability, we evaluate

AD'
r i n ri -ii
r
i
°i (4-371)
_-i i- J) i_ _-l 2_

The observability matrix becomes

smes
"i --l 1 0"
V )' AD] = (4-372)
_o l -1 2_

Since V has a rank of 2, which is the number of inputs, the system is completely
observable.

Example 4-29 Let us consider the system described by the differential equation of
Eq. (4-351), Example 4-26. In Example 4-26 we have shown that
state controllability of a system depends on how the state variables
are defined. We shall now show that the observability also depends on the definition
of the state variables. Let the dynamic equations of the system be defined as in Eqs.
(4-352) and (4-353),
o r
A= D=[l 0]
-1

Then
V= [D' AD] (4-373)

and thus the system is completely observable.

Let the dynamic equations of the system be given by Eqs. (4-356) and (4-357).
Then
1"
D= [l 1]
-1 -2
Then
"1 -1
V= [D' AD] = ']

1 -1
which is singular. Thus the system is unobservable, and we have shown that given the
input-output relation of a linear system, the observability of the system depends on
how the state variables are defined. It should be noted that for the system of Eq.
(4-351),one method of state variable assignment, Eqs. (4-352) and (4-353) yields a
system that is observable but not state controllable. On the other hand, if the dynamic
equations of Eqs. (4-356) and (4-357) are used, the system is completely state control-
lable but not observable. There are definite reasons behind these results, and we shall
investigate these phenomena further in the following discussions.

Alternative definition of observability. If the matrix A has distinct eigen-

values, it can be diagonalized as in Eq. (4-333). The new state variable is

y = P-'x (4-374)
156 / State-Variable Characterization of Dynamic Systems Chap. 4

The new dynamic equations are

y = + Tu
Ay (4-375)

c == Fy + Eu (4-376)
where
F = DP (4-377)

Then the system is completely observable ifF has no zero columns.

The reason behind the above condition is that if the /th (j = 1,2, ... ,n)
column of F contains all zeros, the state variable ys will not appear in Eq.
(4-376) and is not related to the output z{i). Therefore, y, will be unobservable.
In general, the states that correspond to zero columns of F are said to be un-
observable, and the rest of the state variables are observable.

Example 4-30 Consider the system of Example 4-27, which was found to be un-
observable. Since the A
matrix, as shown in Eq (4-366), is already a
diagonal matrix, the alternative condition of observability stated
above requires that the matrix D = [1 0] must not contain any zero columns. Since
the second column of D is indeed zero, the state x 2 is unobservable, and the system
is unobservable.

4.17 Relationship Among Controllability, Observability,

and Transfer Functions

In the classical analysis of control systems, transfer functions are often used
Although controllability
for the modeling of linear time-invariant systems.
and observability are concepts of modern control theory, they are closely
related to the properties of the transfer function.
Let us focus our attention on the system considered in Examples 4-26 and
4-29. It was demonstrated in these two examples that the system is either not
state controllable or not observable, depending on the ways the state variables
are defined. These phenomena can be explained by referring to the transfer
function of the system, which is obtained from Eq. (4-35 1). We have
C(s) _ s + 1 _ + _
s 1 1 , 78
/oj (
.
<.<-•>
,

U(s) s* + 2s + 1 ~(s + l) s+l 2

which has an identical pole and zero at s = — The following theorem gives
1 .

the relationship between controllability and observability and the pole-zero

cancellation of a transfer function.

Theorem 4-5. If the input-output transfer function of a linear system has

pole-zero cancellation, the system will be either not state controllable or un-
observable, depending on how the state variables are defined. If the input-output
transfer function of a linear system does not have pole-zero cancellation, the
system can always be represented by dynamic equations as a completely con-
trollable and observable system.

Proof: Consider that an «th-order system with a single input and single
output and distinct eigenvalues is represented by the dynamic equations

x(?) = Ax(t) + Bu(t) (4-379)

c(t) = Dx(0 (4-380)

Sec. 4.17 Relationship Among Controllability, Observability and Functions / 157

Let the A matrix be diagonalized by an n x n Vandermonde matrix P,

1 1 1 ... 1

/] ki A3 ... A„

X\ XI X\ A? (4-381)

1n-l
.Ai A1n-l
2
ln-1
A3 . . .
].-
A„

The new state equation in canonical form is

y(0 = Ay(?) + Tu(t) (4-382)

where A= P"'AP. The output equation is transformed into

c (0 = Fy(0 (4-383)

where F = DP. The state vectors x(f) and y(f) are related by
x(r) = Py(f) (4-384)

Since A is a diagonal matrix, the /th equation of Eq. (4-382) is

MO = *#&) + ytff) (4-385)

where a, is the /th eigenvalue of A and y, is the /th element of T, where T is an
n X 1 matrix in the present case. Taking the Laplace transform on both sides
of Eq. (4-385) and assuming zero initial conditions, we obtain the transfer
function relation between Y£s) and U(s) as

YJis) Jl. U(s) (4-386)

S — A,-

The Laplace transform of Eq. (4-383) is

C(s) = FY(5) = DPY(i) (4-387)

Now if it is assumed that

D= [rf 1 d2 ... d„] (4-388)
then
F = DP = [/ I f2 /J (4-389)
where
/, = </, + d 2 X, + ... + djrr (4-390)

for i = 1,2, ... ,n. Equation (4-387) is written as

Gto = [/i h /JY(j)

[/. h U(s) (4-391)

f,7,
U(s)
s — X,
158 / State-Variable Characterization of Dynamic Systems Chap. 4

For the nth-order system with distinct eigenvalues, let us assume that the
input-output transfer function is of the form

U(s) (s - A,)(j - A2) . . . - A„)

which is expanded by partial fraction into

where a denotes the residue of C(s)jU(s) at s = A,-.

v$>
=
&^ (4 - 393)

It was established earlier that for the system described by Eq. (4-382) to

be state controllable, all the rows of T must be nonzero that is, y, for ; ^
i = 1, 2, ...,«. If C(s)/U(s) has one or more pairs of identical pole and zero,

for instance in Eq. (4-392), a = A 1; then in Eq. (4-393), a = 0. Comparing

t x

Eq. (4-391) with Eq. (4-393), we see that in general

(r,=/,y, (4-394)

Therefore, when a, = 0, y, will be zero if/, =£ 0, and the state y, is uncontrol-

lable.
For observability, it was established earlier that F must not have columns
containing zeros. Or, in the present case, /, ^ for i = 1 , 2, . . . , n. However,
from Eq. (4-394),

f = (4-395)
t
f t

When the transfer function has an identical pair of pole and zero at a, = X„
a, = 0. Thus, from Eq. (4-395), /, = if y, ^ 0.
4.18 Nonlinear State Equations and Their Linearization

When a dynamic system has nonlinear characteristics, the state equations of

the system can be represented by the following vector-matrix form:

d\(t) __
f[x(r), r(0] (4-396)
dt

where x(t) represents the n X 1 state vector, r(t) the p x 1 input vector, and
f [x(r), r(f)] denotes an n X 1 function vector. In general, f is a function of the
state vector and the input vector.
Being able to represent a nonlinear and/or time-varying system by state
equations is a distinct advantage of the state-variable approach over the transfer

function method, since the latter is defined strictly only for linear time-invariant
systems.
As a simple illustrative example, the following state equations are non-
linear :

^
j ^
= x,(/) + x\{t)

(4 - 397)
*$> = x,(0 + r{t)
 j

Sec. 4.18 Nonlinear State Equations and Their Linearization / 159

Since nonlinear systems are usually difficult to analyze and design, it would
be desirable to perform a linearization whenever the situation justifies.
A linearization process that depends on expanding the nonlinear state
equation into a Taylor series about a nominal operating point or trajectory
is now described. All the terms of the Taylor series of order higher than 1 are
discarded, and linear approximation of the nonlinear state equation at the
nominal point results.
Let the nominal operating trajectory be denoted by x (f), which corresponds
to the nominal input r (?) and some fixed initial states. Expanding the nonlinear
state equation of Eq. (4-396) into a Taylor series about x(f) = x (f) and neglect-
ing all the higher-order terms yields

x (t)=f (x
i i
,i ) S—
+ j=x (7/Xx, r)
53-
OXj
(Xj - x 0J )
(4-398)
* Sfjjx, r)
0y - r 0J )

i = 1, 2, . . . , n. Let
Axi X, Oj (4-399)
and
Ar, r, — r 0/ (4-400)
Then
Ax, = x, X 0! (4-401)
Since
x oi = /i(x o> r o) (4-402)
Equation (4-398) is written

dMx,r)
~~
^ An (4-403)
'
j=x dx, io,r, j" i or

The last equation may be written in the vector-matrix form

Ax = A* Ax + B*Ar (4-404)
where
'df 9fx df{
dx x
dx 2 dx n
dh
A* = dx x dx 2 dx„ (4-405)

El dL df
dx t dx 2 dx„

dfx dli df~

dr, dr z "
drp

df2 d_h df2

B* dr t dr 2 '
dr„
(4-406)

dh dL d_L
dr, dr 2 dr a
 1

160 / State-Variable Characterization of Dynamic Systems Chap. 4

where should be reiterated that A* and B* are evaluated at the nominal point.
it

Thus we have linearized the nonlinear system of Eq. (4-396) at a nominal

operating point. However, in general, although Eq. (4-404) is linear, the ele-
ments of A* and B* may be time varying.
The following examples serve to illustrate the linearization procedure just

described.

Example 4-31 Figure 4-27 shows the block diagram of a control system with a
saturation nonlinearity. The state equations of the system are

*i=fi=xi Xi (4-407)

X 2 = f2 = u (4-408)

It
*l
x = Ax + Bu

Fig. 4-27. Nonlinear control system.

where the input-output relation of the saturation nonlinearity is represented by

u = (1 - e-*i*'i) SGN jf, (4-409)

where
+1 jc, >
SGN xi =
[

(4-410)
-1 x x
<0
Substituting Eq. (4-409) into Eq. (4-408) and using Eq. (4-403), we have the
linearized state equation
At
Aii = -M-Ax 2 = Ax 2 (4-411)
ax 2

Ax 2 = ^Ax
ax
{
= Ke-V'^Axi (4-412)
t

where x 01 denotes a nominal value of x Notice that the last two equations are linear
x .

and are valid only for small signals. In vector-matrix form, ihese linearized state
equations are written as
~Ax{ o r "Axi"
(4-413)
Ax 2 _
a Ax 2 .

where
a = Ke- K \*"\ = constant (4-414)

It is of interest to check the significance of the linearization. If x oi is chosen to

be at the origin of the nonlinearity, x 01 = 0, then a = K; Eq. (4-412) becomes

Ax 2 =KAxi (4-415)

Thus the linearized model is equivalent to having a linear amplifier with a constant
 Sec. 4.19 State Equations of Linear Discrete- Data Systems / 161

gain K. On the other hand, if x , is a large number, the nominal operating point will
lieon the saturated portion of the nonlinearity, and a = 0. This means that any small
variation in x, (small Ax,) will give rise to practically no change in Ax
2 .

Example 4-32 In the last example the linearized system turns out to be time invari-
ant. In general, linearizationof a nonlinear system often leads to a
linear time-varying system. Consider the following nonlinear system:

*i = ^t (4-416)

x2 = «x, (4-417)

We would like to linearize these equations about the nominal trajectory [x ,(0,
X02O)], which is the solution of the equations with the initial conditions x,(0) = x 2 (0)
= 1 and the input u{t) = 0.
Integrating both sides of Eq. (4-41 7), we have
Xi = x 2 (0) = 1 (4-418)
Then Eq. (4-416) gives
xi = -/ + 1 (4-419)
Therefore, the nominal trajectory about which Eqs. (4-416) and (4-417) are to be
linearized is described by

•*oi(0 = ~t + 1 (4-420)

x Q i(t) = 1 (4-421)

Now evaluating the coefficients of Eq. (4-403), we get

dxi
= V
dx
1 =
2 <?x. du
= Xi

Equation (4-403) gives

Ax, = -3-Ax
X02 2 (4-422)

Ax 2 = «oAx, + x ,Ai/ (4-423)

Substituting Eqs. (4-420) and (4-421) into Eqs. (4-422) and (4-423), the linearized
equations are written as

Ax, "0 2" Ax, " "

_Ax 2
+ A« (4-424)
_o o_ _Ax 2 _ J -'_
which is a set of linear state equations with time-varying coefficients.

4.19 State Equations of Linear Discrete- Data Systems

Similar to the continuous-data systems case, a modern way of modeling a

discrete-data system is by means of discrete state equations. As described earlier,
when dealing with discrete-data systems, we often encounter two different
situations. The first one is that the components of the system are continuous-
data elements, but the signals at certain points of the system are discrete or
discontinuous with respect to time, because of the sample-and-hold operations.
In this case the components of the system are still described by differential
equations, but because of the discrete data, a set of difference equations may be
 1

162 / State- Variable Characterization of Dynamic Systems Chap. 4

generated from the original differential equations. The second situation involves
systems that are completely discrete with respect to time in the sense that they
receive discrete data only, such as in the case of a digital controller
and send out
or digital computer. Under this condition, the system dynamics should be
described by difference equations.
Let us consider the open-loop discrete-data control system with a sample-
and-hold device, as shown in Fig. 4-28. Typical signals that appear at various
points in the system are also shown in the figure. The output signal, c(t),

rit) (t) h(t) c(t)

J\-J- Zero-order
hold
G

r*(t) ' \

i: -«*- <TTTv *- 1
T IT 3T AT 5T6T IT
^liJ-

*-

Fig. 4-28. Discrete-data system with sample-and-hold.

 :: :

Sec. 4.19 State Equations of Linear Discrete-Data Systems / 163

ordinarily is a continuous-data signal. The output of the sample-and-hold, h{t),

is a train of steps. Therefore, we can write

h(kT) = r(kT) k = 0, 1 , 2, . . . (4-425)

Now we let the linear process G be described by the state equation and out-
put equation
ML = Ax(0 + Bh(t) (4-426)

c(t) = Dx(0 + EA(0 (4-427)

where x(/) is the state vector and h{i) and c{t) are the scalar input and output
signals, respectively. The matrices A, B, D, and E are coefficient matrices which
have been defined earlier. Using Eq. (4-70), the state transition equation of the
system is written

x(t) = tft - t )\(t ) + ("

<f>(t
- x)Bh(x) dx (4-428)
for t > t .

If we are interested only in the responses at the sampling instants, just as

in the case of the z-transform solution, we let t = (k + l)rand t = kT. Then
Eq. (4-428) becomes
+ VT
x[(* + l)r] = 4>(T)x(kT) + |^ $[{k +l)T- t]BA(t) d-c (4-429)

where §{t) is the state transition matrix as defined in Section 4.4.

Since h(t) is piecewise constant, that is, h{kT) = r(kT) for kT<t<
(k + \)T, the input function A(t) in Eq. (4-429) can be taken outside the integral
sign. Equation (4-429) is written
+Ur
x[(* + \)T] = ${T)x{kT) + |2 *K* + l)r - t]B dx r(kT) (4-430)
or

x[(fc + l)r] = tt7>(*r) + Q(T)r(kT) (4-431)

where
9(T) =
J kT
^k + 1} T~ T]B dZ (4 " 432 >

Equation (4-431) is of the form of a linear difference equation in vector-matrix

form. Since represents a set of first-order difference equations,
it it is referred
to as the vector-matrix discrete state equation.
The discrete state equation in Eq. (4-431) can be solved by means of a
simple recursion procedure. Setting k =
0, 1, 2, ... in Eq. (4-431), we find that
the following equations result

k = 0: = (K7>(0) + 9(7X0)
x(T) (4-433)
* = 1 x(2r) = <K7>(T) + e(7XT) (4-434)
k = 2: x(3J) = (f>(T)x(2T) + 0(7X27) (4-435)

k= k- 1: x(kT) = (K7>p - 1)T] + *(T)r[(k - \)T] (4-436)

164 / State-Variable Characterization of Dynamic Systems Chap. 4

Substituting Eq. (4-433) into Eq. (4-434), and then Eq. (4-434) into Eq. (4-435),
.
.
, and so on, we obtain the following
.
solution for Eq. (4-431):

x(kT) = <f>*(r>x(0) + S p-'-KTWIW.iT) (4-437)

(=0

Equation (4-437) is defined as the discrete state transition equation of the discrete-
data system. It is interesting to note that Eq. (4-437) is analogous to its continu-
ous counterpart in Eq. (4-67). In fact, the state transition equation of Eq.
(4-67) describes the state of the system of Fig. 4-28 with or without sampling.
The discrete state transition equation of Eq. (4-437) is more restricted in that it
describes the state only at t = kT (k = 0, 1, 2, . . .), and only if the system has
a sample-and-hold device such as in Fig. 4-28.
With kT considered as the initial time, a discrete state transition equation
similar to that of Eq. (4-70) can be obtained as
1

x[(k + N)T] = 4>

N (T)x(kT)
+ 2 4>»-'-\T)B(r)r[(k + i)T] (4-438)

where N is a positive integer. The derivation of Eq. (4-438) is left as an exercise

for the reader.
The output of the system of Fig. 4-28 at the sampling instants is obtained by
substituting t = AT and Eq. (4-437) into Eq. (4-427), yielding

c(kT) = Dx(kT) + Eh(kT)

*-i (4-439)
= D4>*(7>(0) + D 2 p-'-KTWTWT) + Kh{kT)

important advantage of the state-variable method over the z-transform

An
method is that it can be modified easily to describe the states and the output

between sampling instants. In Eq. (4-428) if we let / (k - A)T, where = <

A< 1 = kT, we get
and t

f()H-A)T
x[(k + A)T] = ftA7X*r) + <j>P + A)r - r]B dx r(kT) ^^
\ kT

= 4>(AT)x(kT) + 9(A7>(A:7)
By varying the value of A between and 1, the information between the sam-
pling instants completely described by Eq. (4-440).
is

One of the interesting properties of the state transition matrix $(t) is that

+*(T) = ftkT) (4-441)

which is proved as follows.

Using the homogeneous solution of the state equation of Eq. (4-426),

we have
x(t) = #t - foM'o) (4-442)

Let t = kT and ta = 0; the last equation becomes

x{kT) = ftfcTXO) (4-443)

Also, by the recursive procedure with = {k +

t \)T and t = kT, k = 0, 1,
 Sec. 4.20 z-Transform Solution of Discrete State Equations / 165

2, . .
.
, Eq. (4-442) leads to
x(kT) = <J>*(T)x(0) (4-444)

Comparison of Eqs. (4-443) and (4-444) gives the identity in Eq. (4-441).
In view of the relation of Eq. (4-441), the discrete state transition equations
of Eqs. (4-437) and (4-438) are written

x(kT) = (j)(A:r)x(0) + 2 tf(* - - l)r]9(7>0T) '"

(4-445)

x[(A: + N)T] = $(NT)x(kT) + l]' «(# - - l)T]Q(T)r[(k + Or] i (4-446)

These two equations can be modified to represent systems with

respectively.
multiple inputs simply by changing the input r into a vector r.

When a linear system has only discrete data throughout the system, its
dynamics can be described by a set of discrete state equations

x[(k + 1)7] = Ax(kT) + Br(A:r) (4-447)

and output equations
c(kT) = Dx(kT) + Er(kT) (4-448)

where A, B, D, and E are coefficient matrices of the appropriate dimensions.

Notice that Eq. (4-447) is basically of the same form as Eq. (4-431). The only
difference in the two situations is the starting point of the system representation.
In the case of Eq. (4-431), the starting point is the continuous-data state equa-
tions of Eq. (4-426); $(T) and 8(7") are determined from the A and B matrices
of Eq. (4-426). In the case of Eq. (4-447), the equation itself represents an out-
right description of the discrete-data system, which has only discrete signals.
The solution of Eq. (4-447) follows directly from that of Eq. (4-431).
Therefore, the discrete state transition equation of Eq. (4-447) is written

x(kT) = A*x(0) + £ A*-'-'BrOT)

i =
(4-449)

where
A* = AAAA...A (4-450)

4.20 z-Transform Solution of Discrete State Equations

The discrete state equation in vector-matrix form,

x[(k + 1)7"] = Ax(kT) + Br(&r) (4-451)

can be solved by means of the z-transform method. Taking the z-transform on

both sides of Eq. (4-451) yields

zX(z) - zx(0+) = AX(z) + BR(z) (4-452)

Solving for X(z) from the last equation gives

X(z) = (zl - A)" »zx(0-r-) + (zl - A)" 'BR(z) (4-453)

166 / State-Variable Characterization of Dynamic Systems Chap. 4

The inverse z-transform of the last equation is

x(kT) = g- '[(zl - A)" z]x(0)

l
+ g-*[(zl - A) »BR(z)] (4-454)

In order to carry out the inverse z-transform operation of the last equation, we
write the z-transform of A* as

g(A") = f] A*z"* =I+ Az" 1

+A 2
z" 2 + . • (4-455)
k=

Premultiplying both sides of the last equation by Az -1 and subtracting the

result from the last equation, we get

(I- Az-')S(A*) = I (4-456)

Therefore, solving for g{A k ) from the last equation yields

g(A k ) = (I - Az" 1
)" 1
= (zl - A)"'z (4-457)
or
A* = g
l
[{zl -A )
-1
^ (4-458)

Equation (4-458) also represents a way of finding A* by using the z-transform

method. Similarly, we can prove that

5-»t(2l - A)-'BR(z)] = S A^-'-'BrOT) (4-459)

Now we substitute Eqs. (4-458) and (4-459), into Eq. (4-454) and we have the
solution for x(kT) as
k-\
x(kT) = A*x(0) + 2 A*-'-'Br(ir) (4-460)
= i

which is identical to the expression in Eq. (4-449).

Once a discrete-data system is represented by the dynamic equations of
Eqs. (4-447) and (4-448), the transfer function relation of the system can be
expressed in terms of the coefficient matrices.
Setting the initial state x(0+) to zero, Eq. (4-453) gives

X(z) = (zl - A)-»BR(z) (4-461)

When this equation is substituted into the z-transformed version of Eq. (4-448),
we have
C(z) = [D(zl - A)" 'B + E]R(z) (4-462)

Thus the transfer function matrix of the system is

G(z) = D(zl - A)-'B + E (4-463)

This equation can be written

G(z) = D[adj(zI-A)]B + lzI-A|E {A _ m)

The characteristic equation of the system is defined as

|zI-A| = (4-465)

In general, a linear time-invariant discrete-data system with one input and

one output can be described by the following linear difference equation with
constant coefficients:
 . .

Sec. 4.21 State Diagrams for Discrete-Data Systems / 167

c[(k + 1071 + fl,c[(fc + n- + n - 2)71 +

l)r] + a 2 c[(k . .

+ a _ lC [(k + l)r] + a c(kT) n n

(4-466)
= VP + m)T] + VP + m- 1)7] + . .

+ b m . A(k + 1)T] + b m r{kT) n>m

Taking the z-transform on both sides of this equation and rearranging terms,
the transfer function of the system is written

C{z) b zm + Z>,z"-' + ... + b m „ z + b m ,. ,

l
46? ;
R(z) z' + a^"" +... + o,. z + a,
1
1
v

The characteristic equation is defined as

z" + fljz"- 1 + . . . + a„_ ,z + a„ = (4-468)

Example 4-33 Consider that a discrete-data system is described by the difference

equation

c(k + 2) + 5c(k + 1) + 3c(/c) = r(k + 1) + 2r(/t) (4-469)

Taking the z-transform on both sides of the last equation and assuming zero initial

conditions yields
z^C{z) + 5zC(z) + 3C(z) = zR(z) + 2R(z) (4-470)

From the last equation the transfer function of the system is easily written

R(z) z2 + 5z + 3 ^ *' l)
The characteristic equation is obtained by setting the denominator polynomial of the
transfer function to zero,
z1 + 5z + 3 = (4-472)

The state variables of the system are arbitrarily defined as

Xl (k) = c(k) (4-473)

XiQc) = x,{k + 1) - r(k) (4-474)

Substitution of the last two relations into the original difference equation of Eq.
(4-469) gives the two state equations of the system as

Xl (k + 1)= x 2 (k) + r{k) (4-475)

x 2 (k + 1) = -3^!^) - 5x 2 (k) - 3r{k) (4-476)

from which we have the A matrix of the system,

1"
(4-477)
-3 -5_
The same characteristic equation as in Eq. (4-472) is obtained by using zl |
— A| = 0.

4.21 State Diagrams for Discrete-Data Systems

When a discrete-data system is described by difference equations or discrete

state equations, a discrete state diagram may be constructed for the system.
Similar to the relations between the analog computer diagram and the state
diagram for a continuous-data system, the elements of a discrete state diagram
168 / State-Variable Characterization of Dynamic Systems Chap. 4

resemble the computing elements of a digital computer. Some of the operations

of a digital computer are multiplication by a constant, addition of several
machine variables, time delay, or shifting. The mathematical descriptions of
these basic digital computations and their corresponding z-transform expres-
sions are as follows:

1. Multiplication by a constant:

x 2 (kT) = ax^kT) (4-478)

X (z) = aX,(z)
2 (4-479)
2. Summing:
x 2 (kT) = x (kT) + Xi (kT) (4-480)

X (z)
2
= X (z) + Z,(z) (4-481)

3. Shifting or time delay:

x 2 (kT)= Xi [(k + l)T) (4-482)

X (z) = zX,{z) - zx,(0+)

2 (4-483)
or
X^z) = z- X {z) +
l
2 *,(<)+) (4-484)

The state diagram representations of these opera-

a
X (z) o- -OX (z)
tions are illustrated in Fig. 4-29. The initial time t = 0+
{ 2
in Eq. (4-484) can be generalized to t = rj. Then Eq.
X2 (z) = aX l
(z) (4-484) is written

Z,(z) = z' l
X (z) + x, {n)
2 (4-485)
X (z)
which represents the discrete-time state transition for
time greater than or equal to fj.

X2 (z) Example 4-34 Consider again the difference equation in

Eq. (4-469), which is

c(k + 2) + 5c(k + 1) + 3c(k) = r(k + 1) + 2r(k)

(4-486)
X t
(z)
One way of constructing the discrete state diagram for the
X2 (z) = XQ (z) + X x
(z) system is to use the state equations. In this case the state
equations are available in Eqs. (4-475) and (4-476) and these
are repeated here:
9*i(0+)
x,(k + 1) = Xl (k) + r(k) (4-487)

x 2 (k + 1) = -3 Xl (k) - 5x 2 (k) - 3r(k) (4-488)

X2 (z)o O^(r) Using essentially the same principle as for the state diagrams
for continuous-data systems, the state diagram for Eqs. (4-487)
and (4-488) is constructed in Fig. 4-30. The time delay unit
X (z) = z-iX 2 (z) + x 1 (0+)
1
z
_1
is used to relate x^k + The state variables
1) to Xi(k).

Fig. 4-29. Basic elements of a discrete will always appear as outputs of the delay units on the state
state diagram. diagram.
 Sec. 4.21 State Diagrams for Discrete-Data Systems / 169

R{z) C{z)

-3

diagram of the system described by the difference

Fig. 4-30. Discrete state
equation of Eq. (4-486) or by the state equations of Eqs. (4-487) and
(4-488).

As an alternative, the state diagram can also be drawn directly from the differ-
ence equation by means of the decomposition schemes. The decomposition of a discrete
transfer function will be discussed in the following section, after we have demon-
strated some of the practical applications of the discrete state diagram.
state transition equation of the system can be obtained directly from the
The
statediagram using the gain formula. Referring to {z) and X
z (z) as the output nodes
x
X
and to xi(0+), * 2 (0+), and R(z) as input nodes in Fig. 4-30, the state transition
equations are written in the following vector-matrix form:

1
"1 +5z-' "*i(o+r 1 rz-i(l +5z-')-3z- 2
R{z) (4-489)
~ A -3Z- 1
1 _* 2 (0+)_ A|_ _ 3z -i_3 z -2
_ .

where
A= 1 + 5z-' + 3z- (4-490)

The same transfer function between R(z) and C(z) as in Eq. (4-471) can be obtained
directly from the state diagram by applying the gain formula between these two nodes.

Decomposition of Discrete Transfer Functions

The three schemes of decomposition discussed earlier for continuous-data

systems can be applied to transfer functions of discrete-data systems without
the need of modification. As an illustrative example, the following transfer
function is decomposed by the three methods, and the corresponding state

diagrams are shown in Fig. 4-31

<2f) ^ + 2
(4.491)

Equation (4-491) is used for direct decomposition after the numerator and the
denominator are both multiplied by z~ 2 For cascade decomposition, the trans-
.
 x 2 (0+)
o

Riz) C(z)

-3
(a) Direct decomposition

x, (0 +)

R(z) C(z)

(b) Cascade decomposition

*i(0+)

RU)

(c) Parallel decomposition

Fig. 4-31. State diagrams of the transfer function C(z)/i?(z) = (z + 2)/(z 2

+ 5z + 3) by the three methods of decomposition, (a) Direct decomposi-
tion, (b) Cascade decomposition, (c) Parallel decomposition.

170
 O :

Sec. 4.22 State Diagrams for Sampled- Data Systems / 171

fer function is first written in factored form as

C{z) _ z + 2
(4-492)
R(z) (z + 4.3)(z + 0.7)
For the parallel decomposition, the transfer function is first fractioned by
partial fraction into the following form
C(z) 0.64 , 0.36
R(z) ~ z + 4.3 + 0.7
(4-493)

4.22 State Diagrams for Sampled-Data Systems

When a discrete-data system has continuous-data as well as discrete-data

elements, with the two types of elements separated by sample-and-hold devices,
a special treatment of the state diagram is necessary if a description of the
continuous-data states is desired for all times.
Let us first diagram of the zero-order hold. Consider
establish the state
that the input of the zero-order hold is denoted by e*{i) which is a train of
impulses, and the output by h{t). Since the zero-order hold simply holds the
magnitude of the input impulse at the sampling instant until the next input
comes along, the signal h{t) is a sequence of steps. The input-output relation
in the Laplace domain is written

= 1 -e-*-
H(s) -E*(s) (4-494)

In the time domain, the relation is simply

h(t) = e(kT+) (4-495)

for kT<t<(k + l)T.
In the state diagram notation, we need the relation between H(s) and
e(kT+). For this purpose we take the Laplace transform on both sides of Eq.
(4-495) to give

e(kT+) H(s) = e(fcr+) (4-496)

O —
for kT < t < (k + \)T. The state diagram representa-
Fig. 4-32. State diagram rep- tion of the zero-order hold is shown in Fig. 4-32. As
resentation of the zero-order an illustrative example on how the state diagram of a
hold. sampled-data system is constructed, let us consider the
system shown in Fig. 4-33. We shall demonstrate the
various available ways of modeling the input-output relations of the system.
First, the Laplace transform of the output of the system is written

C(s) = w (4-497)
s s + 1

Fig. 4-33. Sampled-data system.

 — : )

172 / State- Variable Characterization of Dynamic Systems Chap. 4

Taking the z-transform on both sides of the last equation yields

(4-498)

Given information on the input e(t) or e*(t), Eq. (4-498) gives the output
response at the sampling instants.
A state diagram can be drawn from Eq. (4-498) using the decomposition
technique. Figure 4-34 illustrates the discrete state diagram of he system through t

O x, (0 +

e(kT+)

Fig. 4-34. Discrete state diagram of the system in Fig. 4-33.

decomposition. The discrete dynamic equations of the system are written

directly from this state diagram

*,[(* + 1)7]= e- T x,(kT) + (1 - e- T)e{kT) (4-499)

c(kT) = x (kT)
x
(4-500)

Therefore, the output response of the system can also be obtained by solving
the difference equation of Eq. (4-499).
If the response of the output c(t) is desired for all t, we may construct the
state diagram shown in Fig. 4-35. This state diagram is obtained by cascading
the state diagram representations of the zero-order hold and the process G(s).

x (kT+)
{

s
o

,' 1

o » o *>

e(kT+) His)

Fig. 4-35. State diagram for the system of Fig. 4-33 for the time interval
kT<t<(k + 1)T.
 :

Sec. 4.23 State Equations of Linear Time-Varying Systems / 173

To determine c(t), which is also *i(f)> we must first obtain X^s) by applying
the gain formula to the state diagram of Fig. 4-35. We have

*•(*> = TT~^ e(kT+) + TT~^ Xl(kT) (4 " 501)

for kT < t < (& + l)r. Taking the inverse Laplace transform of the last equa-
tion gives

*i(0 = [1 - e- ( '- kT>

]e(kT+) + (
e- «-*"jc,(Jtr) (4-502)

kT< < (k +
t \)T. It is interesting to note that in Eq. (4-502) t is valid for
one sampling period, whereas the result in Eq. (4-499) gives information on
x,(r) only at the sampling instants. It is easy to see that if we let t = {k + Y)T
in Eq. (4-502), the latter becomes Eq. (4-499).

4.23 State Equations of Linear Time-Varying Systems

When a linear system has time-varying elements, it can be represented by the

following dynamic equations

^ = A(t)x(t) + B(/)r« (4-503)

c(?) = D(0x(r) + E(r)r(?) (4-504)

where
x(r) =n X 1 state vector

r (0 =P x 1 input vector

c (0 =9 x 1 output vector

and E(t) are coefficient matrices of appropriate dimensions.

A{t), B(/), D(?),
The elements of these coefficient matrices are functions of t.
Unlike in the time-invariant case, time-varying differential equations
generally do not have closed-form solutions. Let us investigate the properties
of a time-varying system by considering a scalar homogeneous state equation,

^ = a(t)x(t) (4-505)

This equation can be solved by first separating the variables,

and then integrating both

^ = a{t)dt (5-506)

sides to get

In x(t) - In x(t Q ) = a(z) dx (4-507)

J'
Therefore,

x{t) = exp
£ a{x) dx\x(t a ) (4-508)

where t denotes the initial time.

Just as in the time-invariant situation,we can define a state transition
matrix for the time-varying state equation. For the scalar case under considera-
 .

174 / State-Variable Characterization of Dynamic Systems Chap. 4

tion, the state transition matrix is

<f>(t, t 9 )
= exp f a(x) dx (4-509)

Notice that for the time-varying case, the state transition matrix depends upon
t and t not simply t — ?
, -

For the vector-matrix state equation

±(t) = A(0x(r) (4-510)

it is simple to show that the solution can be written

x(r) = ftr, t )x(t ) (4-511)

where $(t, t ) is the state transition matrix that satisfies Eq. (4-510). However,
the problem is how to find <f>(t, t ) in general. The question is: Is §{t, t ) related
to the A(f) matrix through the following relationship ?

<J>(?, t ) = exp H" A(t) dx (4-512)

To answer the posed question, let us expand the right side of Eq. (4-512)
into a power series,

exp T P A(t) dx) =I+ f A(t) dx +^ [ A(t) dx f

A(<r) </* + ...

(4-513)

Taking the derivative on both sides of the last equation with respect to time,
we have

*-[T.
A(t) dx A(0
i m L' A(«t) do + j- \ A(t) rft A(f) + . .

(4-514)

Multiplying both sides of Eq. (4-513) by A(t), we have

A(0 exp [" ('

A(t) dx = A(0 + A(0 j" A(t) dx (4-515)

By comparison of Eqs. (4-514) and (4-515), we see that

^ exp P A(t) dx = A(0 exp \ f A(t) Jt (4-516)

J

or
${t, / ) = A(0<K?, * o) (4-517)
dt
if and only if

A(r) f A(t) dx = ' A(t) dx A(t) (4-518)

J to Jf la

that is,

A(0 and f A(t) rfr

J to

commute.
The requirement that A(r) and its integral commute is evidently a very
stringent condition. Therefore, in general, Eq. (4-512) will not be valid.
 :

Sec. 4.23 State Equations of Linear Time-Varying Systems / 175

Most of the properties of the time-invariant state transition matrix,

4>(t — t ), can be extended to the time- varying case. These are listed as follows

1. <KMo) = I-

2. <f.-
1
(?,? ) = <f.(/ ,0.
3. ${t 2 ?,) <K?,,
, t ) = ${t 2 , t ) for any t , t u t2 .

Solution of the Nonhomogeneous Time-Varying State Equation

Disregarding the problem of finding the state transition matrix §{t, t )

for the moment, we shall solve for the solution of the nonhomogeneous state
equation of Eq. (4-503).
Let the solution be
x(f) = ${t, to)f[{t) (4-519)

where an n x 1 vector, and <^(/, t ) is the state transition matrix that

r\(t) is

Eq. (4-517). Equation (4-519) must satisfy Eq. (4-503). Substitution of

satisfies

Eq. (4-519) into Eq. (4-503) yields

4>U, toMO + W, t )r\(t) = A(0ft/, toMt) + B(0«(?) (4-520)

Substituting Eq. (4-517) into Eq. (4-520) and simplifying, we get

¥t, *oM) = B«u(0 (4-521)

Thus
f|(?) = <f»"'(f, t ) B(f)u(0 (4-522)
and

¥0 = f *
_i
(t, ?o)B(t)u(t) dx + tl(f„) (4-523)

The vector r\(t ) is obtained from Eq. (4-519) by setting t = r . Thus substitut-
ing Eq. (4-523) into Eq. (4-519), we have

x(f) = W, t )x(t ) + $(r, t ) f $-»(*, ? )B(t)u(t) rft (4-524)

Since

W, r )*-'(T, r ) = W, /„)<K'o, t) = <K', t) (4-525)

Eq. (4-524) becomes

x(r) = ftf, r )x(r ) + f'

$(t, t)B(t)u(t) dt (4-526)

which is the state transition equation of Eq. (4-503).

Discrete Approximation of the Linear Time-Varying System

In practice, not too many

time-varying systems can be solved by using
Eq. (4-526), since $(t, not readily available. It is possible to discretize
t ) is

the system with a time increment during which the time-varying parameters
do not vary appreciably. Then the problem becomes that of solving a set of
linear time-varying discrete state equations. One method of discretizing the
system is to approximate the derivative of x(r) by
176 / State- Variable Characterization of Dynamic Systems Chap. 4

i(t) ~ -L{x[(k + l)T] - x(kT)} kT<t<(k+ l)T (4-527)

where T is a small time interval. The state equation of Eq. (4-503) is approxi-
mated by the time-varying difference equation

x[(k = A*{kT)x(kT) + B*(kT)r(kT)

+ l)T] (4-528)

over the time interval, kT < < (k + l)T, where

t

A.*(kT) = TXQcT) + I

B*(kT) = TB(kT)

Equation (4-528) can be solved recursively in much the same way as in the time-
invariant case, Eqs. (4-433) through (4-437).

REFERENCES

State Variables and State Equations

1. L. A. Zadeh, "An Introduction to State Space Techniques," Workshop on

Techniques for Control Systems, Proceedings, Joint Automatic Control Confer-
ence, Boulder, Colo., 1962.

2. B. C. Kuo, Linear Networks and Systems, McGraw-Hill Book Company, New

York, 1967.

3. D. W. Wiberg, Theory and Problems of State Space and Linear Systems (Schaum's
Outline Series), McGraw-Hill Book Company, New York, 1971.

State Transition Matrix

4. R. B. Kirchner, "An Explicit Formula for e Al ," Amer. Math. Monthly, Vol. 74,
pp. 1200, 1204, 1967.

5. W. Everling, "On the Evaluation of e A by ' Power Series," Proc. IEEE, Vol. 55,

p. 413, Mar. 1967.

6. T. A. Bickart, "Matrix Exponential: Approximation by Truncated Power

Series," Proc. IEEE, Vol. 56, pp. 872-873, May 1968.

7. T. M. Apostol, "Some Explicit Formulas for the Exponential Matrix e At ," Amer.
Math. Monthly, Vol. 76, pp. 289-292, 1969.

8. M. Vidyasagar, "A Novel Method of Evaluating eA! in Closed Form," IEEE

Trans. Automatic Control, Vol. AC-15, pp. 600-601, Oct. 1970.
A IEEE
9. C. G. Cullen, "Remarks on Computing e '," Trans. Automatic Control,
Vol. AC-16, pp. 94-95, Feb. 1971.

10. J. C. Johnson and C. L. Phillips, "An Algorithm for the Computation of the
Integral of the State Transition Matrix," IEEE Trans. Automatic Control, Vol.
AC-16, pp. 204-205, Apr. 1971.
Chap. 4 References / 177

11. M. Healey, "Study of Methods of Computing Transition Matrices," Proc. IEE,

No. 8, pp. 905-912, Aug. 1973.
Vol. 120,

Transformations

12. C. D. Johnson and W. M. Wonham, "A Note on the Transformation to Canon-

ical (Phase Variable) Form," IEEE Trans. Automatic Control, Vol. AC-9, pp.
312-313, July 1964.

13. I.H. Mufti, "On the Reduction of a System to Canonical (Phase- Variable)
Form," IEEE Trans. Automatic Control, Vol. AC-10, pp. 206-207, Apr. 1965.
14. M. R. Chidambara, "The Transformation to (Phase- Variable) Canonical Form,"
IEEE Trans. Automatic Control, Vol. AC-10, pp. 492-495, Oct. 1965.

15. L. M. Silverman, "Transformation of Time- Variable Systems to Canonical

(Phase- Variable) Form," IEEE Trans. Automatic Control, Vol. AC-11, pp. 300-
303, Apr. 1966.

16. W. G. Tuel, Jr., "On the Transformation to (Phase- Variable) Canonical Form,"
IEEE Trans. Automatic Control, Vol. AC-11, p. 607, July 1966.
17. D. G. Luenberger, "Canonical Forms for Linear Multivariate Systems,"
IEEE Trans. Automatic Control, Vol. AC-12, pp. 290-293, June 1967.
18. S. J. Asseo, "Phase- Variable Canonical Transformation of Multicontroller Sys-
tems," IEEE Trans. Automatic Control, Vol. AC-13, pp. 129-131, Feb. 1968.

19. B. Ramaswami and Ramar, "Transformation to the Phase- Variable Canonical

Form," IEEE Trans. Automatic Control, Vol. AC-13, pp. 746-747, Dec. 1968.
20. W. B. Rubin, "A Simple Method for Finding the Jordan Form of a Matrix,"
IEEE Trans. Automatic Control, Vol. AC-17, pp. 145-146, Feb. 1972.
21. K. Ogata, State Space Analysis of Control Systems, Prentice-Hall, Inc., Engle-
wood Cliffs, N.J., 1967.

State Diagram

22. B. C. Kuo, "State Transition Flow Graphs of Continuous and Sampled Dynamic
Systems," WESCON Convention Records, 18.1, Aug. 1962.

Controllability and Observability

23. Y. C. Ho, "What Constitutes a Controllable System?" IRE Trans. Automatic

Control, Vol. AC-7, p. 76, Apr. 1962.

24. R. E. Kalman, Y. C. Ho, and K. S. Narendra, "Controllability of Linear

Dynamical Systems," Contribution to Differential Equations, Vol. 1, No. 2, pp.
189-213, 1962.

25. E. G. Gilbert, "Controllability and Observability in Multivariate Control

Systems,"/. SIAM Control, Vol. 1, pp. 128-151, 1963.

26. L. A. Zadeh and C. A. Desoer, Linear System Theory, McGraw-Hill Book

Company, New York, 1963.
27. R. E. Kalman, "Mathematical Description of Linear Dynamical Systems," /.
Soc. Ind. Appl. Math., Vol. II, No. 1, Ser. A, pp. 151-192, 1963.
 .

178 / State- Variable Characterization of Dynamic Systems Chap. 4

28. E. Kreindler and P. Sarachik, "On the Concept of Controllability and Observ-
ability of Linear Systems," IEEE Trans. Automatic Control, Vol. AC-9, pp.
129-136, Apr. 1964.

29. A. R. Stubberud, "A Controllability Criterion for a Class of Linear Systems,"

IEEE Trans. Application and Industry, Vol. 83, pp. 411-413, Nov. 1964.

30. R. W. Brockett, "Poles, Zeros, and Feedback: State Space Interpretation,"

IEEE Trans. Automatic Control, Vol. AC-10, pp. 129-135, Apr. 1965.
31. R. D. Bonnell, "An Observability Criterion for a Class of Linear Systems,"
IEEE Trans. Automatic Control, Vol. AC-11, p. 135, Jan. 1966.

CSMP {Continuous System Modeling Program)

32. System/360 Continuous System Modeling Program (360A-CX-16X) User's Manual,
Technical Publications Dept., International Business Machines Corporation,
White Plains, N.Y.

PROBLEMS

4.1. Write state equations for the electric networks shown in Fig. P4-1

) e(t)

4.2. The following differential equations represent linear time-invariant systems.

Write the dynamic equations (state equations and output equation) in vector-
matrix form.
Chap. 4 Problems / 179

(a) «*!£)
dt 1
+ 3*0 c(t) = r(t)

(b)

(c) 3
^
rf
2
c(Q
4

rfc(/)
+
5c(t)

c{t)
=
=
r(t)

dr(t)

rfM£)
(d) 2 "+
,

-c(t) c(T)dT = /•(/)

^
<ft
f.
(e)
rf/
2 +1_u
"
6^ +rfr
5c(0 = ,«H0 + 2^<fr
,(/)

4.3. Using Eq. (4-42), show that

((>(0 = I + Ar Klf2
. -r
l
A 3
/
3

ji^
,

T\ J,,

4.4. The state equations of a linear time-invariant system are represented by

iO) = Ax(0 + B«(0

Find the state transition matrix <(»(/) for the following cases:

(a) 0" "1"

B =
-1 -2_ _1_
(b) r "0"
B=
-2 -3_ _1_
(c) -2 0" "10"
B =
-2_ _ 1_

(d) -1 1
0"
ro
-1 1 B=
-1 l

4.5. Find the state transition equations for the systems described in Problem 4.4
for t > 0. It is assumed that x(0+) is given and u(t) is a unit step function.
4.6. Given the state equation

±(t) = Ax(0 + B«(0

where
~0 r
A= 1 -1 B
_2 1 0_
find the transformation such that the state equation becomes
y(0 = A,x(/) + B!«(0
where A! and B ( are in the phase-variable canonical form.
4.7. For the state equation given in Problem 4.6, if

B=

can the state equation be transformed into the phase- variable form? Explain.
4.8. Given the state equations of a linear time-invariant system as

x(/) = Ax(?) + B«(0

 :

180 / State-Variable Characterization of Dynamic Systems Chap. 4

where
1
0~ ro
1 B=
-2 -3. l

determine the transfer function relation between X(s) and U(s). Find the
eigenvalues of A.

4.9. For a linear time-invariant system whose state equations have coefficient
matrices given by Eqs. (4-111) and (4-112) (phase-variable canonical form),
show that
1

adj (si - A)B =

and the characteristic equation is

s" + ais" -1
+ a 2 s" -2 + . . . + aa -is + a„ == 0.

4.10. A closed-loop control system is described by

±(t) = Ax(0 + Bu(0

u(/) = -Gx(0

where x(t) = //-vector, u(/) = /--vector, A is n x n, B is n x r, and G is the

r x n feedback matrix. Show that the roots of the characteristic equation are
eigenvalues of A — BG. Let

1
0"

1
A= B=
1

-2 -5 -10_
G= [£i g2 gi g*\

Find the characteristic equation of the closed-loop system. Determine the

elements of G so that the eigenvalues of A — BG are at — 1, —2, —1 —jl, and
—1 +/1. Can all the eigenvalues of A — BG be arbitrarily assigned for this
problem?
4.11. A linear time-invariant system is described by the following differential equa-
tion :

d 2 c(t) Ml) c(t) = r(f)

dt 2 dt

(a) Find the state transition matrix <J>(/).

(b) Let c(0) = 1, t(0) = 0, and r(t) = u s (t), the unit step function; find the
state transition equations for the system.
(c) Determine the characteristic equation of the system and the eigenvalues.

4.12. A linear multivariable system is described by the following set of differential

equations
 :

Chap. 4 Problems / 181

rf'cKQ dc,(rt
rf/
2 + ,

rf/
+ 2c,(0-2c 2 (0=r,(0

dt 2
ci(0 + c 2 (0 = r 2 (f)

(a) Write the state equations of the system in vector-matrix form. Write the
output equation in vector-matrix form.
(b) Find the transfer function between the outputs and the inputs of the system.
4.13. Given the state transition equation ±(t) = Ax(/), where
A=
jo a
a and co are real numbers.
(a) Find the state transition matrix <^(/).

(b) Find the eigenvalues of A.

(c) Find the eigenvectors of A.
4.14. Given the state equations of a linear system as

±(t) = Ax(/) + B«(0

where
1 on ro
1 B=
-6 -11 -6J |_1.

The eigenvalues of A are A j = —1,X 2 = -2, X 3 = —3. Find a transformation

x(t) = Py(f) that will transform A into a diagonal matrix A = diag [Xi X 2 X 3 ],
4.15. Given a linear system with the state equations described by
1(0 = Ax(f) + Bm(0
where
'
1
0'
r°i
1 B=
.-25 -35 11_ _1_
The eigenvalues are X x
= —I, X 2 = —5, X 3 =
—5. Find the transformation
x(/)= Py(/) so that A is transformed into the Jordan canonical form. The
transformed state equations are

HO = Ay(0 + T«(0
Find A and T.
4.16 Draw state diagrams for the following systems
(a) ±(0 = Ax(/) + Bu(t)
-3 2 0" ro
A = -1 -1 1 B=
-5 -2 -1 l

(b) ±(r) Ax(f) + Bu(f). Same A as in part (a) but with

~o r
B

4.17. The block diagram of a feedback control system is shown in Fig. P4-17.
(a) Write the dynamic equations of the system in vector-matrix form.
 :

182 / State-Variable Characterization of Dynamic Systems Chap. 4

Figure P4-17.

(b) Draw a state diagram for the system.

(c) Find the state transition equations for the system. Express the equations in
matrix form. The initial states are represented by x(t'b), and the input r{t)
is a unit step function, u s {t — / ), which is applied at t == t .

4.18. Draw state diagrams for the following transfer functions by means of direct
decomposition
10
(a) G(s) =
s3 + 5s 2 + 4s + 10
6(s + 1)
< b> °W = s(s + 1F+ 3)
Write the state equations from the state diagrams and express them in the
phase-variable canonical form.
4.19. Draw state diagrams for the following systems by means of parallel decom-
position :

(a) G(s)
6(s+ 1)
~
s(s + 2)(s + 3)
d
<i»3P + «^> 5c(0 = 2
-f- +
r W
Write the state equations from the state diagrams and show that the states are
decoupled from each other.
4.20. Draw state diagrams for the systems in Problem 4.19 by means of cascade
decomposition.
4.21. Given the transfer function of a linear system,

iQfr + i)
G(s) =
UW (s + 2¥(s + 5)

Draw state diagrams for the system using three different methods of decom-
position. The state diagrams should contain a minimum number of integrators.

4.22. The state diagram of a linear system is shown in Fig. P4-22.

Figure P4-22.
Chap. 4 Problems / 183

(a) Assign the state variables and write the dynamic equations of the system.
(b) Determine the transfer function C(s)/R(s).
4.23. Draw state diagrams for the electric network shown in Fig. P4-1.

4.24. The state diagram of a linear system is shown in Fig. P4-24.

Figure P4-24.

(a) Assign state variables on the state diagram; create additional nodes if

necessary, as long as the system

is not altered.

(b) Write the dynamic equations for the system.

4.25. Given the state equation

±(0 = Ax(?)
where
-2 1 0"

-2 1

.0 -2_
(a) Find the eigenvalues of A.
(b) Determine the state transition matrix <J)(0-

4.26. Given the state equation

±(/) = Ax(t) + B«(0

where
1 on ro
1 B =
-2 -4 -3J |_1.

The eigenvalues of A are Ai = —1, A 2 = —1 — j\, A = —1 +jl. Find the 3

transformation x(?) = Py</) which transforms A into the modal form

"-1 0"
-1 1 =P"iAP
-1 -1_
4.27. Given the linear system

±(0 = Ax(0 + B«(0

where «(/) is generated by state feedback,
u(t) = -Gx(0
The state transition matrix for the closed-loop system is

<J>(/)
= e (A-BG)« = £-![(,,! _ a + BG)- 1
]

Is the following relation valid ?

e (A-BG)t _ gAtg-BGt
 :

184 / State- Variable Characterization of Dynamic Systems Chap. 4

where
e At = £-l[(5I _ A)"']
e -BGr = £-i[(sl + BG)" 1
]

Explain your conclusions.

4.28. Determine the state controllability of the system shown in Fig. P4-28.

Figure P4-28.

(a) a = 1, b = 2, c = 2, and d=\.

(b) Are there any nonzero values for a, b, c, and (/such that the system is not
completely state controllable?

4.29. Figure P4-29 shows the block diagram of a feedback control system. Determine
the state controllability and observability of the system by the following
methods, whenever applicable

J
\X
x c
It
\ 1
\

.9 + 2
;

'
'
+
3
x2
1
I

Fig ure P4 -29.

(a) Conditions on the A, B, D, and E matrices.

(b) Transfer function.
(c) Coupling of states.

4.30. The transfer function of a linear system is given by

C(s) s +a
R(s)~ s3 +6s 2 + Us + 6

(a) Determine the value of a so that the system is either uncontrollable or

unobservable.
(b) Define the state variables so that one of them is uncontrollable.
(c) Define the state variables so that one of the states is unobservable.

4.31. Consider the system described by the state equation

±(f) = Ax(0 + B«(0

 :

Chap. 4 Problems / 185

where
"
o r "1"
B =
L-l a_ L*J
Find the region in the a-versus-6 plane such that the system is completely
controllable.

4.32. Draw the state diagram of a second-order system that is neither controllable nor
observable.

4.33. Determine the conditions on b\, b2 , d u and d2 so that the following system is

completely state controllable, output controllable, and observable

±(0 = Ax(0 + Bk(/)

~i r
B
_o i_

c(0 = Dx(0
D= [«/, d2 ]
4.34. The block diagram of a simplified control system for the Large Space Telescope
(LST) is shown in Fig. P4-34. For simulation and control purposes, it would
be desirable to represent the system by state equations and a state diagram.

Gimbai Control moment

controller gyro dynamics

Vehicle
K,
Kp s +
~\+ K p
s + Kj 1 H
position

Command K, -J s JG s J„s 2
1

Vehicle
dynamics

KN 1
s

Figure P4-34.

(a) Draw a state diagram for the system and write the state equations in vector-
matrix form.
(b) Find the characteristic equation of the system.
(c) A modern control design scheme, called the state feedback, utilizes the
concept of feeding back every state variable through a constant gain. In
this case the control law is described by

e = r - gjXi — g 2 x 2 — g]X - g x Xi 3

Find the values of gi, g 2 gi, and g^ such that the eigenvalues of the overall
,

system are at s = — 100, —200, —1 +/1, —1 —jl. The system parameters

are given as = 600, Kj = 9700, JG = 2, Jv = 10 5 Kp = 216, and
H ,

KN = 300. All units are consistent.

 —
186 / State-Variable Characterization of Dynamic Systems Chap. 4

4.35. The difference equation of a linear discrete-data system is given by

c[(k + 2)T] + 0.5c[(A: + 1)T] + 0.1c(kT) = 1

(a) Write the state equations for the system.

(b) The initial conditions are given as c(0) = 1 and c(T) = 0. Find c{kT) for
k = 2, 3, . .
.
, 10 by means of recursion. Can you project the final value of
c(kT) from the recursive results?
4.36. Given the discrete-data state equations,

Xl (k + 1) = 0.1x 2 (k)
x 2 (k + 1) = -x^k) + 0Jx 2 (k) + r(k)
find the state transition matrix </>(&).

4.37. A discrete-data system is characterized by the transfer function

C(z) Kz
R(z) (z - l)(z 2 -z 3)

(a) Draw a state diagram for the system.

(b) Write the dynamic equation for the system in vector-matrix form..

4.38. The block diagram of a discrete-data control system is shown in Fig. P4-38.

tO —^T
>it) e(t) - e*(0 c(t)
z.o.h.
2Q +
s(s
0.5)
+ 0.2)
— *-

Figure P4-38.

(a) Draw a state diagram for the system.

(b) Write the state equations in vector-matrix form,

x[(fc + 1)71 = <J>(r)x(/cD + B(T)r(kT)

(c) Find <^(T) when T= 0.1 sec.
 5
Mathematical Modeling

of Physical Systems

5.1 Introduction

One of the most important tasks in the analysis and design of control systems is

the mathematical modeling of the systems. In preceding chapters we have intro-

duced a number of well-known methods of modeling linear systems. The two
most common methods are the transfer function approach and the state-variable
approach. However, in reality most physical systems have nonlinear charac-
teristics to some extent. A physical system may be portrayed by a linear mathe-
matical model only if the true characteristics and the range of operation of the
system justify the assumption of linearity.
Although the analysis and design of
linear control systems have been well
developed, their counterparts for nonlinear systems are usually quite complex.
Therefore, the control systems engineer often has the task of determining not
only how to accurately describe a system mathematically, but, more important,
how to make proper assumptions and approximations, whenever necessary, so
that the system may be adequately characterized by a linear mathematical model.
important to point out that the modern control engineer should place
It is

special emphasis on the mathematical modeling of the system so that the analysis
and design problems can be adaptable for computer solutions. Therefore, the
main objectives of this chapter are

1. To demonstrate the mathematical modeling of control systems and

components.
2. To demonstrate how the modeling will lead to computer solutions.

187
188 / Mathematical Modeling of Physical Systems Chap. 5

The modeling of many system components and control systems will be

However, the emphasis is placed on the approach to
illustrated in this chapter.
the problem, and no attempt is being made to cover all possible types of systems
encountered in practice.

5.2 Equations of Electrical Networks

The classical way of writing network equations of an electrical network is the

loop method and the node method, which are formulated from the two laws of
Kirchhoff. However, although the loop and node equations are easy to write,
they are not natural for computer solutions. A more modern method of writing
network equations is the state-variable method. We shall treat the subject briefly
in this section. More detailed discussions on the state equations of electrical
networks may be found in texts on network theory. 12
Let us use the RLC network of Fig. 5-1 to illustrate the basic principle of

R L
+ V\AA< T1KP
HO
+
e(t) C -J? e c (t)

Fig. 5-1. RLC network.

writing state equations for electric networks. It is relatively simple to write the
loop equation of this network:

= L d-^l + R d
e(t)
-f + -L q (t) (5-1)

where q(t) is the electric charge and is related to the current i(t) by

q{t)=\' i{-c)dx (5-2)

It is shown in Chapter 4 that the second-order differential equation in

Eq. (5-1) can be replaced by two first-order differential equations called the
state equations. In this case it is convenient to define the state variables as

Xl (t) = ^P = e c (t) (5-3)

where ec (t) is the voltage across the capacitor and

* 2 (0 = ^ = /(0 = C^) (5-4)

Substituting Eqs. (5-3) and (5-4) into Eq. (5-1) yields

e(t) = L **M + Rx 2 (t) + Xl (t) (5-5)

 :

Sec. 5.2 Equations of Electrical Networks / 189

Thus, from Eqs. (5-4) and (5-5) the state equations of the network are

dx x
(t)

dt
£*,<0 (5-6)

Rx 1

^r - l
x,(t) 2 (t) e(t) (5-7)

A more direct way of arriving at the state equations is to assign the current
in the inductor L, i(t), and the voltage across the capacitor C, e c (t), as the state
variables. Then the state equations are written by equating the current in C and
the voltage across L in terms of the state variables and the input source. This way
the state equations are written by inspection from the network. Therefore,

Current in C: C^4^ = i(t) (5-8)

dt

Voltage across L L di(f)_ -e c (t) - Ri(t) + e(t) (5-9)

Since x^t) = e c (t) and x 2 (t) = i(i), it is apparent that these state equations are
identical to those of Eqs. (5-6) and (5-7).
In general, it is appropriate to assign the voltages across the capacitors and
currents in the inductors as state variables in an electric network, although there
are exceptions. 12
One must recognize that the basic laws used in writing state equations for
electric networks arestill the KirchhofFs laws. Although the state equations in

Eqs. (5-8) and (5-9) are arrived at by inspection, in general, the inspection method
does not always work, especially for complicated networks. However, a general
method using the theory of linear graphs of network analysis is available. 1

Example 5-1 As another example of writing the state equations of an electric net-
work, consider the network shown in Fig. 5-2. According to the
foregoing discussion, the voltage across the capacitor e c and the cur-
rents in the inductors /, and i2 are assigned as state variables, as shown in Fig. 5-2.

Fig. 5-2. Network in Example 5-1.

The state equations of the network are obtained by writing the voltages across the
inductors and the currents in the capacitor in terms of the three state variables. The
state equations are

^p- = -RMi) ~ e {t) + c e(.t) (5-10)

dh(t)
L2 -Rihit) + e (t)
c (5-11)
dt
190 / Mathematical Modeling of Physical Systems Chap. 5

c^ = m-m (5-12)

Rearranging the constant coefficients, the state equations are written in the following
canonical form:
1
diAtY] _*i o -+- hit) 1
dt u
dhU)
dt
-£ 1
hit)
+ h e(t) (5-13)

de c (t)
edt)
dt C c

5.3 Modeling of Mechanical System Elements 3

Most feedback control systems contain mechanical as well as electrical com-

ponents. From a mathematical viewpoint, the descriptions of electrical and
mechanical elements are analogous. In fact, we can show that given an electrical
device, there is usually an analogous mechanical counterpart, and vice versa.
The analogy, of course, is a mathematical one that is, two systems are analo-
;

gous to each other they are described mathematically by similar equations.

if

The motion of mechanical elements can be described in various dimensions

as translational, rotational, or a combination of both. The equations governing
the motions of mechanical systems are often directly or indirectly formulated
from Newton's law of motion.

Translational Motion

The motion of translation is defined as a motion that takes place along a

straight line. The variables that are used to describe translational motion are
acceleration, velocity, and displacement.
Newton's law of motion states that the algebraic sum offorces acting on a
rigid body in a given direction is equal to the product of the mass of the body and
its acceleration in the same direction. The law can be expressed as

2 forces = Ma (5-14)

where M
denotes the mass and a is the acceleration in the direction considered.
For translational motion, the following elements are usually involved:

1. Mass: Mass is considered as an indication of the property of an

element which stores the kinetic energy of translational motion. It is
analogous to inductance of electrical networks. If denotes the W
weight of a body, then M is given by

M W (5-15)
g
where g is the acceleration of the body due to gravity of the accelera-
tion of free fall. Three consistent sets of units for the elements in Eqs.
(5-14) and (5-15) are as follows:
 :

Sec. 5.3 Modeling of Mechanical System Elements / 191

Units Mass M Weight W Acceleration Force

MKS newtons/m/sec 2 newton m/sec 2 newton

CGS dynes/cm/sec 2 dyne cm/sec 2 dyne
British lb/ft/sec 2 (slug) lb ft/sec 2 lb

Figure 5-3 illustrates the situation where a force is

y(t)
acting on a body with mass M. The force equation is

M fit)
written

fit) Ma(t) = M<?M M-dv(t)

r
(5-16)
dt
Fig. 5-3. Force-mass system.
where y(t) represents displacement, v(t) the velocity, and
a{t) is the acceleration, all referenced in the direction of the
applied force.
2. Linear spring : A linear spring in practice may be an actual spring or
the compliance of a cable or a belt. In general, a spring is considered
to be an element that stores potential energy. It is analogous to a
capacitor in electric networks. In practice, all springs are nonlinear
to some However, if the deformation of a spring
extent. is small, its

behavior may be approximated by a linear relationship,

f{t) = Ky{t) (5-17)

where K is the spring constant, or simply stiffness. The three unit

systems for the spring constant are

Units K
MKS newtons/m
CGS dynes/cm
British lb/ft

Equation (5-17) implies that the force acting on the spring is directly
proportional to the displacement (deformation) of the
spring. The model representing a linear spring element
y{t)
K is shown in Fig. 5-4.
-
AAAAA/v- fit) If the spring is preloaded with a preload tension of
T, then Eq. (5-17) should be modified to

Fig. 5-4. Force-spring system f{t)-T=Ry{t) (5-18)

Friction for translational motion. Whenever there is motion or tendency

of motion between two elements, frictional forces exist. The frictional forces
encountered in physical systems are usually of a nonlinear nature. The charac-
teristics of the frictional forces between two contacting surfaces often depend

on such factors as the composition of the surfaces, the pressure between the
surfaces, their relative velocity, and others, so that an exact mathematical
description of the frictional force is difficult. However, for practical purposes,
frictional forces can be divided into three basic catagories: viscous friction,
 : :

192 / Mathematical Modeling of Physical Systems Chap. 5

static friction, and Coulomb friction. These are discussed separately in detail in
the following.

1. Viscous friction. Viscous friction represents a retarding force that is

a linear relationship between the applied force and velocity. The

schematic diagram element for friction is often repre-
sented by a dashpot such as that shov/n in Fig. 5-5. The
y(t)
mathematical expression of viscous friction

H
is

fit)
f(f) = B®& (5-19)

Fig. 5-5. Dashpot for viscous friction.

where B is the viscous frictional coefficient. The dimen-
sions of B in the three unit systems are as follows

Units B
MKS newton/m/sec
CGS dyne/cm/sec
British lb/ft/sec

Figure 5-6(a) shows the functional relation between the viscous

frictional force and velocity.
2. Static friction. Static friction represents a retarding force that tends
to prevent motion from beginning. The static frictional force can be
represented by the following expression

f{t) = ±(F,),_ (5-20)

where (^)^, is defined as the static frictional force that exists only
when the body is stationary but has a tendency of moving. The sign
of the friction depends on the direction of motion or the initial

direction of velocity. The force-velocity relation of static friction is

illustrated in Fig. 5-6(b). Notice that once motion begins, the static
and other frictions take over.
frictional force vanishes,
3. Coulomb friction. Coulomb friction is a retarding force that has a

f f
+ F.

Slope = B

(b) (c)

Fig. 5-6. Functional relationships oflinear and nonlinear frictional forces,

(a) Viscous friction, (b) Static friction, (c) Coulomb friction.
Sec. 5.3 Modeling of Mechanical System Elements / 193

constant amplitude with respect to the change in velocity, but the

sign of the frictional force changes with the reversal of the direction
of velocity. The mathematical relation for the Coulomb friction is
given by

<5 - 2,)
/w-'-GF/ISD
where F c is the Coulomb friction coefficient. The functional descrip-
tion of the friction to velocity relation is shown in Fig. 5-6(c).

Rotational Motion

The rotational motion of a body may be defined as motion about a fixed

axis.The variables generally used to describe the motion of rotation are torque;
angular acceleration, a; angular velocity, <y; and angular displacement, 6. The
following elements are usually involved with the rotational motion.

Inertia. Inertia, /, is considered as an indication of the property of an

element which stores the kinetic energy of rotational motion. The inertia of a
given element depends on the geometric composition about the axis of rotation
and its density.
For instance, the inertia of a circular disk or a circular shaft about its geo-
metric axis is given by
J = \Mr 2
(5-22)

where M is the mass of the disk or shaft and r is its radius.

Example 5-2 Given a disk that is 1 in. in diameter, 0.25 in. thick, and weighing 5
oz, its inertia is

j _ 1 Wr 2 ^ 1 (5oz)(l in) 2
2 g 2 386 in/sec 2 (5-23)
= 0.00647 oz-in-sec 2

Usually the density of the material is given in weight per unit volume. Then, for a
circular disk or shaft it can be shown that the inertia is proportional to the fourth
power of the radius and the first power of the thickness or length. Therefore, if the
weight W is expressed as
W = p(nr 2 h)
(5-24)

where p is the density in weight per unit volume, r the radius, and h the thickness or
length, then Eq. (5-22) is written

j = 1 pn^ = omQ6p h/ . 4
(525)

where h and r are in inches.

For steel, p is 4.53 oz/in 3 Eq. ; (5-25) becomes

J = 0.0184 hr 4 (5-26)

For aluminum, p is 1.56 oz/in 3 ; Eq. (5-25) becomes

/ = 0.00636 hr* (5-27)

 : :

194 / Mathematical Modeling of Physical Systems Chap. 5

When a torque is applied to a body with inertia /,

T(-^ as shown in Fig. 5-7, the torque equation is written

» <' >
no = Mt) = J ^r = J ~$P (5-28)

The three generally used sets of units for the quan-

Fig. 5-7. Torque-inertia system, tities in Eq. (5-28) are tabulated as follows

Units Inertia Torque Angular Displacement

MKS kg-m 2 newton-m radian

CGS g-cm 2 dyne-cm radian
English slug-ft 2 lb-ft

or lb-ft-sec 2
oz-in-sec 2 oz-m radian

The following conversion factors are often found useful

Angular displacement.

1 rad = 11? = 57.3°

Angular velocity.

1 rpm = ?£
oU
= 0.1047 rad/sec
1 rpm = 6 deg/sec

Torque.
1 g-cm = 0.0139 oz-in

1 lb-ft = 192 oz-in

1 oz-in = 0.00521 lb-ft

Inertia.
1 g-cm 2 = 1.417 X 10" 5 oz-in-sec 2

1 lb-ft-sec
2
=192 oz-in-sec 2

1 oz-in-sec 2 = 386 oz-in 2

1 g-cm-sec 2 = 980 g-cm 2
1 lb-ft-sec
2
= 32.2 lb-ft 2

Torsional spring. As with the linear spring for translational motion, a

torsional spring constant K, in torque per unit angular
displacement, can be devised to represent the compliance
K 7X0
of a rod or a shaft when it is subject to an applied
-M/WW torque. Figure 5-8 illustrates a simple torque-spring
0(0 system that can be represented by the following equa-
tion:
Fig. 5-8. Torque-torsional spring
system. 7X0 = KB{t) (5-29)
 :

Sec. 5.3 Modeling of Mechanical System Elements / 195

The dimension for K is given in the following units

Units K
MKS newton-m/rad
CGS dyne-cm/rad
British oz-in/rad

If the torsional spring is preloaded by a preload torque of TP, Eq. (5-29) is

modified to
T(t) -TP= Kd(t) (5-30)

Friction for rotational motion. The three types of friction described for
translational motion can be carried over to the motion of rotation. Therefore,
Eqs. (5-19), (5-20), and (5-21) can be replaced, respectively, by their counter-
parts :

no _ n dftt)
dt
(5-31)

no ±{F )d=0
s (5-32)

(5-33)

where B is the viscous frictional coefficient in torque per unit angular velocity,
C\)<t«o is the static friction, and Fc is the Coulomb friction coefficient.

Relation Between Translational and Rotational Motions

In motion control problems it is often necessary to convert rotational mo-
tion into a translational one. For instance, a load may be controlled to move
along a straight line through a rotary motor and screw assembly, such as that
shown in Fig. 5-9. Figure 5-10 shows a similar situation in which a rack and

no, HO -x(0

Motor W

Fig. 5-9. Rotary-to-linear motion-control system.

x(0

W
/VWWWWWWWWWWWWWVAA

Drive
motor

Fig. 5-10. Rotary-to-linear motion-control system.

 r

196 / Mathematical Modeling of Physical Systems Chap. 5

Fig. 5-11. Rotary-to-linear motion-control system.

pinion is used as the mechanical linkage. Another common system in motion

control is the control of a mass through a pulley by a rotary prime mover, such
as that shown in Fig. 5-11. The systems shown in Figs. 5-9, 5-10, and 5-11 can
all be represented by a simple system with an equivalent inertia connected
directly to the drive motor. For instance, the mass in Fig. 5-1 1 can be regarded as
a point mass which moves about the pulley, which has a radius r. Disregarding
the inertia of the pulley, the equivalent inertia that the motor sees is

/=M/- 2 = —g 2
(5-34)

If the radius is r, the equivalent inertia which the

of the pinion in Fig. 5-10
motor by Eq. (5-34).
sees is also given
Now consider the system of Fig. 5-9. The lead of the screw, L, is defined as
the linear distance which the mass travels per revolution of the screw. In prin-
ciple, the two systems in Fig. 5-10 and 5-11 are equivalent. In Fig. 5-10, the
distance traveled by the mass per revolution of the pinion is 2nr. Therefore,
using Eq. (5-34) as the equivalent inertia for the system of Fig. 5-9,
2
7 = fte) (5 " 35)

where in the British system

/= inertia (oz-in-sec 2
)

W — weight (oz)
L = screw lead (in)
g = gravitational force (386.4 in/sec
2
)

Mechanical Energy and Power

Energy and power play an important role in the design of electromechanical

systems. Stored energy in the form of and potential energy controls the
kinetic
dynamics of the system, whereas dissipative energy usually is spent in the form of
heat, which must be closely controlled.
The mass or inertia of a body indicates its ability to store kinetic energy.
The kinetic energy of a moving mass with a velocity v is
Wk = \Mv r
(5-36)

The following consistent sets of units are given for the kinetic energy rela-
tion:
Sec. 5.3 Modeling of Mechanical System Elements / 197

Units Energy Mass Velocity

MKS joule or newton/m/sec 2 m/sec

newton-m
CGS dyne-cm dyne-cm-sec 2 cm/sec
British ft-lb lb/ft/sec 2 ft/sec
(slug)

For a rotational system, the kinetic energy relation is written

W = ya>
k
2
(5-37)
where / is the moment of intertia and co the angular velocity. The following
units are given for the rotational kinetic energy:

Units Energy Inertia Angular Velocity

MKS joule or kg-m 2 rad/sec

newton-m
CGS dyne-cm gm-cm 2 rad/sec
British oz-in oz-in-sec 2 rad/sec

Potential energy stored in a mechanical element represents the amount of

work required to change the configuration. For a linear spring that is deformed
by y in length, the potential energy stored in the spring is

W = \Ky*
p (5-38)
where K is the spring constant. For a torsional spring, the potential energy
stored is given by
Wp = \KQ* (5-39)
Whendealing with a frictional element, the form of energy differs from the
previous two cases in that the energy represents a loss or dissipation by the sys-
tem in overcoming the frictional force. Power is the time rate of doing work.
Therefore, the power dissipated in a frictional element is the product of force
and velocity; that is,

P=fi> (5-40)
Since /= Bv, where B is the frictional coefficient, Eq. (5-40) becomes

P = Bv* (5-41)
The MKS unit for power is in newton-m/sec or watt; for the CGS system it is
dyne-cm/sec. In the British unit system, power is represented in ft-lb/sec or
horsepower (hp). Furthermore,
1 hp = 746 watt
(5-42)
= 550 ft-lb/sec
Since power is the rate at which energy is being dissipated, the energy dis-
sipated in a frictional element is

W = BJv
d
2
dt (5-43)
 .

198 / Mathematical Modeling of Physical Systems Chap. 5

Gear Trains, Levers, and Timing Belts

A gear train, a lever, or a timing belt over pulleys is a mechanical device

that transmits energy from one part of a system to another in such a way that
force, torque, speed, and displacement are altered. These devices may also be
regarded as matching devices used to attain maximum power transfer. Two
gears are shown coupled together in Fig. 5-12. The inertia and friction of the
gears are neglected in the ideal case considered.
The relationships between the torques 7^ and T2 angular displacements ,

0, and 2 and the teeth numbers JV", and N2 of the gear train are derived from
> ,

the following facts:

1 The number of teeth on the surface of the gears is proportional to the

radii r, and r 2 of the gears; that is,

r,N z =r N 1 l
(5-44)

2. The distance traveled along the surface of each gear is the same.
Therefore,
6',!•, = 62 r2 (5-45)

3. The work done by one gear is equal to that of the other since there
is assumed to be no loss. Thus

T 1 1
=T 2 62 (5-46)
JV,

Ti ,
a
Ft

r,,«, N,

*
i T, 0,

4
Tiy d7
M^4 <C2 "2
JV,
JV,

Fig. 5-12. Gear train. Fig. 5-13. Gear train with friction and inertia.

If the angular velocities of the two gears, co 1 and co 2 , are brought into the
picture, Eqs. (5-44) through (5-46) lead to

(5-47)
T2 0t #2 CO! r2

In practice, real gears do have inertia and friction between the coupled gear
teeth which often cannot be neglected. An equivalent representation of a gear
train with viscous friction, Coulomb friction, and inertia considered as lumped
elements is shown in Fig. 5-13. The following variables and parameters are
defined for the gear train:
 :

Sec. 5.3 Modeling of Mechanical System Elements / 199

T= applied torque
6 1, 6 2 = angular displacements
Ti,T2 = torque transmitted to gears
/, J2
, = inertia of gears
N u N — number of teeth
2

F,i, F = Coulomb friction coefficients

c2

B u B = viscous frictional coefficients

2

^ ^
The torque equation for gear 2 is written

T2 (t) = J2 +B 2 + Fc2 -^ (5-48)

The torque equation on the side of gear 1 is

r(0 = /, ^°- + *, *M> + F M- + 7\(?) ei (5-49)

By the use of Eq. (5-47), Eq. (5-48) is converted to

™ w™
T (a -N, T (t _ (n.VjJi rfwo (N y B
- 2
,
- \wj -&- + ) W ST + n F
t

2
>
ddti)
Nl
if
<>
e2
|^y
,
(s 50)
(5 m
Equation (5-50) indicates that it is possible to reflect inertia, friction, (and com-
pliance) torque, speed, and displacement from one side of a gear train to the
other.
Therefore, the following quantities are obtained when reflecting from gear
2 to gear 1

Inertia: (£i)V2

Viscous frictional coefficient : (--^1 B2

Torque: ^T 2

Angular displacement: -jrr-d 2

N 2

Angular velocity: ~co 2

Jy 2

c° 2
Coulomb frictional torque: -rrFc2
N 2
,

|a>2l
.

If torsional spring effect were present, the spring constant is also multiplied by
(NJN2 ) 2
in reflecting from gear 2 to gear 1. Now, substituting Eq. (5-50) into
Eq. (5-49), we get

where
HO = /,. ^0 + B U ^M + T P (5-51)

= Ji +
Ju J 2 (5-52)
(jff
2
Bu = B + xB 2 (5-53)
(jfy
200 / Mathematical Modeling of Physical Systems Chap. 5

Example 5-3 Given a load that has inertia of 0.05 oz-in-sec 2 and a Coulomb fric-

tion torque of 2 oz-in, find the inertia and frictional torque reflected
through a 1 : 5 gear train (Ni/N2 = ^ with N 2 on the load side). The
reflected inertia on the side of JV, is (£)
2x 0.05 = 0.002 oz-in-sec 2 . The reflected
Coulomb friction is (^)2 = 0.4 oz-in.

Timing belts and chain drives serve the same purposes as the gear train
except that they allow the transfer of energy over a longer distance without using
an excessive number of gears. Figure 5-14 shows the diagram of a belt or chain
drive between two pulleys. Assuming that there is no slippage between the belt
and the pulleys, it is easy to see that Eq. (5-47) still applies to this case. In fact,
the reflection or transmittance of torque, inertia, friction, etc., is similar to that of
a gear train.

*~/i

T2 , 02

Fig. 5-14. Belt or chain drive. Fig. 5-15. Lever system.

The shown in Fig. 5-15 transmits translational motion and

lever system
force in thesame way that gear trains transmit rotational motion. The relation
between the forces and distances is

A h (5-55)

Backlash and Dead Zone

Backlash and dead zone usually play an important

x{t)
and similar mechanical linkages. In a
role in gear trains
\ |
great majority of situations, backlash may give rise to
|
* .HO
undesirable oscillations and instability in control systems.
~*~ 2 "*~
2 In addition, it has a tendency to wear down the mechan-
Output ) ical elements. Regardless of the actual mechanical ele-
ments, a physical model of backlash or dead zone between

Fig. 5-16. Physical model of backlash an input and an output member is shown in Fig. 5-16.
between two mechanical elements. The model can be used for a rotational system as well as
Sec. 5.3 Modeling of Mechanical System Elements / 201

for a translational system. The amount of backlash is b/2 on either side of the
reference position.
In general, the dynamics of the mechanical linkage with backlash depend
upon the relative inertia-to-friction ratio of the output member. If the inertia of
the output member is very small compared with that of the input member, the
motion is controlled predominantly by friction. This means that the output
member will not coast whenever there is no contact between the two members.
When the output is driven by the input, the two members will travel together
until the input member reverses its direction; then the output member will

stand still until the backlash is taken up on the other side, at which time it is

assumed that the output member instantaneously takes on the velocity of the
input member. The transfer characteristic between the input and the output
displacements of a backlash element with negligible output inertia is shown in
Figure 5-17. To illustrate the relative motion between the input and the output

Fig. 5-17. Input-output characteristic of backlash with negligible output

inertia.

members, let us assume that the input displacement is driven sinusoidally with
respect to time. The displacements and velocities of the input and output mem-
bers are illustrated as shown in Fig. 5-18. Note that the reference position of the
two members is taken to be that of Fig. 5-16, that is, with the input member
starting at the center of the total backlash. For Fig. 5-18, it is assumed that
when motion begins, the input member is in contact with the output member on
the right, so that x(0) = and y(0) =
—b/2. At the other extreme, if the friction
on the output member it may be neglected, the inertia of the
is so small that
output member remains in contact with the input member as long as the accel-
eration is in the direction to keep the two members together. When the accelera-
tion of the input member becomes zero, the output member does not stop
immediately but leaves the input member and coasts at a constant velocity that is
equal to the maximum velocity attained by the input member. When the output
member has traversed a distance, relative to the input member, equal to the full
width of the backlash, it will be restrained by the opposite side of the input mem-
202 / Mathematical Modeling of Physical Systems Chap. 5

Fig. 5-18. Displacementand velocity waveforms of input and output

members of a backlash element with a sinusoidal input displacement.

Fig. 5-19. Input-output displacement characteristic of a backlash element

without friction.

ber. At that time the output member will again assume the velocity of the input
member. The transfer characteristic between the input and the output displace-
ment of a backlash element with negligible output friction is shown in Fig.
5-19. The displacement, velocity, and acceleration waveforms of the input and
output members, when the input displacement is driven sinusoidally, is shown in
Fig. 5-20.
In practice, of course, the output member of a mechanical linkage with
backlash usually has friction as well as inertia. Then the output waveforms in
response to a sinusoidally driven input displacement should lie between those of
Figs. 5-18 and 5-20.
 Sec. 5.4 Equations of Mechanical Systems / 203

Displacement

Velocity

Acceleration

Output
Input

and acceleration waveforms of input

Fig. 5-20. Displacement, velocity,
and output members of a backlash element when the input displacement
is driven sinusoidally.

5.4 Equations of Mechanical Systems 3

The equations of a linear mechanical system are written by first constructing a

model of the system containing interconnected linear elements, and then the
system equations are written by applying Newton's law of motion to the free-
body diagram.

Example 5-4 Let us consider the mechanical system shown in Fig. 5-2 1(a). The
free-body diagram of the system is shown in Fig. 5-21 (b). The force

, d 2 y(t)

y«)

*-/(/)

(a) (b)

Fig. 5-21. (a) Mass-spring-friction system, (b) Free-body diagram.

 : :

204 / Mathematical Modeling of Physical Systems Chap. 5

equation of the system is written

(5-56)

This second-order differential equation can be decomposed into two first-order state
equations, using the method discussed in Chapter 4. Let us assign x x = y and jc 2 =
dy/dt as the state variables. Then Eq. (5-56) is written

dxj(t)
**(/) (5-57)
dt

It is not
T =
difficult to see that this
-i^<)-M^> + ]>>
mechanical system is analogous to a series
(5-58)

RLC
electric network. With this analogy it is simple to formulate the state equations direct-
ly from the mechanical system using a different set of state variables. If we consider
that mass is analogous to inductance, and the spring constant K is analogous to the
inverse of capacitance, 1/C, it is logical to assign v(t), the velocity, and fk (t), the force
acting on the spring, as state variables, since the former is analogous to the current in
an inductor and the latter is analogous to the voltage across a capacitor.
Then the state equations of the system are

Force on mass M Mp = _ Mt) -A(0 -i-/(0 (5-59)

Velocity of spring
l dfk (t) _ v{t) (5-60)
K dt

Notice that the first state equation is similar to writing the equation on the voltage
across an inductor; the second is through a capacitor.
like that of the current
This simple example further illustrates the points made in Chapter 4 regarding
the fact that the state equations and state variables of a dynamic system are not unique.

Example As a second example of writing equations for mechanical systems,

5-5
consider the system shown in Fig. 5-22(a). Since the spring is deformed
when it is subjected to the force /(f), two displacements, Vi and y 2 ,

must be assigned to the end points of the spring. The free-body diagrams of the system
are given in Fig. 5-22(b). From these free-body diagrams the force equations of the
system are written

y 2 U)
Md y 2 (t)
y 2 U)
yi(0 dt 2 yi(t)

VA K
CF M -nptfs-
K fU) dy 2 (t)
M
K(y -y 2 )
-nnnn-
/(»
B l

~dT-

(a) (b)

Fig. 5-22. Mechanical system for Example 5-5. (a) Mass-spring-friction

system, (b) Free-body diagrams.
 Sec. 5.4 Equations of Mechanical Systems / 205

fit) = K\y x (t) - y 2 (t)] (5-61)

K[yi(t) (5-62)

Now let us write the state equations of the system. Since the differential equation of
the system is already available in Eq. (5-62), the most direct way is to decompose this
equation into two first-order differential equations.
Therefore, letting x x (t) =y 2 (t) and x 2 (t) = dy 2 (t)/dt, Eqs. (5-61) and (5-62) give

dxi(t)
dt
=x 2 (t) (5-63)

dt
-£*(') + 3^/(0 (5-64)

As an alternative, we can assign the velocity v(t) of the body with mass M as one
state variable, and the force fk {t) on the spring as the other state variable, so we have

dv(i) B ,,, . 1 -,.

(5-65)

and
/*(') =/(') (5-66)

One may wonder at this point if the two equations in Eqs. (5-65) and (5-66) are
correct as state equations, since seems that only Eq. (5-65) is a state equation, but we
it

do have two state variables in v(t) and fk (t). Why do we need only one state equation
here, whereas Eqs. (5-63) and (5-64) clearly are two independ-
ent state equations? The situation is better explained (at least
for electrical engineers) by referring to the analogous electric
network of the system, shown in Fig. 5-23. It is clear that
although the network has two reactive elements in L and C
and thus there should be two state variables, the capacitance
in this case is a "redundant" element, since e c (t) is equal to
the applied voltage e(t). However, the equations in Eqs. (5-65)
and (5-66) can provide only the solution to the velocity of M
Fig. 5-23. Electric network analogous once /(f) is specified. If we need to find the displacement y^it)
to the mechanical system of Fig. 5-22. at the point where f(t) is applied, we have to use the relation

ydt) = &jp + y 2 (t) = ££ + P v(t) dx + y 2 (0+) (5-67)

where .^(O-l-) is thedisplacement of the body with mass M. On the other hand,
initial

we can from the two state equations of Eqs. (5-63) and (5-64), and then
solve for y 2 (t)
yi(t) is determined from Eq. (5-61).

Example 5-6 In this example the equations for the mechanical system in Fig. 5-24(a)
are to be written. Then we are to draw state diagrams and derive
transfer functions for the system.
The free-body diagrams for the two masses are shown in Fig. 5-24(b), with the
reference directions of the displacements y and y 2 as indicated. The Newton's force
t
 :

206 / Mathematical Modeling of Physical Systems Chap. 5

^^^^^

K-,

M-,

y 2 (f)

K, *i(>-i

M,

Vlit)

no /u)

(a) (b)

Fig. 5-24. Mechanical system for Example 5-6.

equations for the system are written directly from the free-body diagram:

(5-68)

(5-69)

dt* dt

We may now decompose these two second-order simultaneous differential equations

into four state equations by defining the following state variables

x, = yi (5-70)

Xl
dy\ _ dx±
(5-71)
dt dt

x 3 =yi (5-72)

= dy 2 dx 3
xt (5-73)
dt dt

Equations (5-71) and (5-73) form two state equations naturally; the other two are
obtained by substituting Eqs. (5-70) through (5-73) into Eqs. (5-68) and (5-69), and
rearranging we have
dxi
dt
= x2 (5-74)
 :

Sec. 5.4 Equations of Mechanical Systems / 207

dx 2
dt
Xi) ~ Ml {Xz ~ Xi) + Wm x
(5-75)

dx 3
Xi (5-76)
dt

+ AT2
dt -M 2
Xl + W 2
X2
A",

M 2
"
*33
A/ 2
1
•(£, + Bi )x i (5-77)

If we are interested in the displacements y x

and j^, the output equations are
written
yi(t) = *i(0 (5-78)

*« = *s(0 (5-79)

The state diagram of the system, according to the equations written above, is
drawn as shown in Fig. 5-25. The transfer functions Y (s)IF(s) and Y2 (s)/F(s) are x

By/Mj

Fig. 5-25. State diagram for the mechanical system of Fig. 5-24.

obtained from the state diagram by applying Mason's gain formula. The reader
should verify the following results (make sure that all the loops and nontouching loops
are taken into account)

Y M + + B )s + + K2
mx (s) 2 s* (Bi
A
2 (K t )
(5-80)

MO
F(s)
B t s +K x
(5-81)

where
A = MM x 2 s* + [M (B + B2 + B M2 - Bfis* + [M (K + K2 ) + K M2
X X ) X S X X
(5-82)
+B 1 (B l + B2 ) - BiK^s 2 + [K B 2 + B (K + K2 )]s + K K2 X X X X

The state equations can also be written directly from the diagram of the mechanical
system. The state variables are assigned as v = dy /dt, v 2 = dy 2 /dt, and the forces on t x

the two springs, /x! and fK2 Then, if we write the forces acting on the masses and the
.

velocities of the springs, as functions of the four state variables and the external force,
the state equations are

Force on M x : M x
dv
~di
x
-B lVl +B v -fK1 +f
x 2 (5-83)
208 / Mathematical Modeling of Physical Systems Chap. 5

Force on M 2 : M z -%* = fi,^i — (2?i B2 )v 2 +/ki —fx: (5-84)

dfKi
Velocity on K t
.

: "^- = #,(*>i v2 ) (5-85)

Velocity on K2 :
^p = K v 2 2 (5-86)

Example The rotational system shown in Fig. 5-26(a) consists of a disk mounted
5-7
on a shaft that is fixed at one end. The moment of inertia of the disk
about its axis is /. The edge of the disk is riding on a surface, and
the viscous friction coefficient between the two surfaces is B. The inertia of the shaft
is negligible, but the stiffness is K.

<s^^^^ ^^^^

(a) (b)

Fig. 5-26. Rotational system for Example 5-7.

Assume that a torque is applied to the disk as shown; then the torque or moment
equation about the axis of the shaft is written from the free-body diagram of Fig.
5-26(b):

T(t) = J 4%P + B^ + K0(t) (5-87)

Notice that this system is analogous to the translational system of Fig. 5-21. The state
equations may be written by defining the state variables as x-i(t) = 6(t) and dx x {f)\dt
= x 2 (t). The reader may carry out the next step of writing the state equations as an
exercise.

5.5 Error-Sensing Devices in Control Systems 4 '

In feedback control systems it is often necessary to compare several signals at a

certain point of a system. For instance, it is usually the case to compare the
reference input with the controlled variable; the difference between the two
signals is called the error. The error signal is then used to actuate the system.
The block-diagram notation for the algebraic sum of several signals is defined in
Fig. 3-5. In terms of physical components, an error-sensing device can be a simple
potentiometer or combination of potentiometers, a differential gear, a trans-
former, a differential amplifier, a synchro, or a similar element. The mathematical
modeling of some of these devices is discussed in the following.
Sec. 5.5 Error-Sensing Devices in Control Systems / 209

Potentiometers. Since the output voltage of a potentiometer is proportional

to the shaft displacement, when a voltage is applied across its fixed terminals,
the device can be used to compare two shaft positions. In this case one shaft
may be fastened to the potentiometer case and the other to the shaft of the
potentiometer. When a constant voltage is applied to the fixed terminals of the
potentiometer, the voltage across the variable and the reference terminals will
be proportional to the difference between the two shaft positions. The arrange-
ment shown in Fig. 5-27(b) is a one-potentiometer realization of the error-sens-
ing devicesshown in Fig. 5-27(a). A more versatile arrangement may be obtained
by using two potentiometers connected in parallel as shown in Fig. 5-27(c). This

(a)

u(0

(b)

6
4 o_

-e(0-

(c)

Fig. 5-27. (a)Block-diagram and signal-flow-graph symbols for an error

sensor, (b) Position error sensor usingone potentiometer, (c) Position
error sensor using two potentiometers.
 t o Load

Gear train

T
Input
e(f)
dc amplifier e a (t) (
M
>U
shaft

Permanent-magnet
(a) dc motor

Fig. 5-28. (a) Direct current control system with potentiometers as error
detectors, (b) Typical waveforms of signals in the control system of (a).

210
Sec. 5.5 Error-Sensing Devices in Control Systems / 211

allows comparison of two remotely located shaft positions. The applied voltage
v(t)can be ac or dc, depending upon the types of transducers that follow the
error sensor. If v(t) is a dc voltage, the polarity of the output voltage e(t) deter-
mines the relative position of the two shafts. In the case of an ac applied voltage,
the phase of e(t) acts as the indicator of the relative shaft directions. In either
case the transfer relation of the two error-sensor configurations can be written

e(t) = K [O (t) -
s r 6 c (t)] (5-88)
where
=
e(t) error voltage, volts

K =
s sensitivity of the error sensor, volts per radian

The value of K, depends upon the applied voltage and the total displacement
capacity of the potentiometers. For instance, if the magnitude of v(t) is V volts
and each of the potentiometers is capable of rotating 10 turns, Ks = F/2O71 V/rad.
A simple example that illustrates the use of a pair of potentiometers as an
error detector is shown in Fig. 5-28(a). In this case the voltage supplied to the
error detector, v(t), is a dc voltage. An unmodulated or dc electric signal, e(t),
proportional to the misalignment between the reference input shaft and the
controlled shaft, appears as the output of the potentiometer error detector. In
control system terminology, a dc signal usually refers to an unmodulated signal.
On the other hand, an ac signal in control systems is modulated by a modulation
process. These definitions are different from those commonly used in electrical
engineering, where dc simply refers to unidirectional and ac indicates alternating.
As shown in Fig. 5-28(a), the error signal is amplified by a dc amplifier
whose output drives the armature of a permanent-magnet dc motor. If the sys-
tem works properly, wherever there is a misalignment between the input and the
output shafts, the motor will rotate in such a direction as to reduce the error to
a minimum. Typical waveforms of the signals in the system are shown in Fig.
5-28(b). Note that the electric signals are all unmodulated and the output dis-
placements of the motor and the load are essentially of the same form as the
error signal. Figure 5-29(a) illustrates a control system which could serve essen-
tially the same purpose as that of the system of Fig. 5-28(a) except that ac signals

prevail. In this case the voltage applied to the error sensor is sinusoidal. The
frequency of this signal is usually much higher than the frequency of the true
signal that is being transmitted through the system. Typical signals of the ac
control system are shown in Fig. 5-29(b). The signal v(t) is referred to as the
carrier signal whose frequency is co c or ,

v(t)=V sin co c t (5-89)

Analytically, the output of the error sensor is given by

e(t) = K,0JMt) (5-90)

where 0,(0 is the difference between the input displacement and the load dis-
placement, 0/0 r (?)
= —
c (f). For the 0/0 shown in Fig. 5-29(b), e(t) becomes

a suppressed-carrier modulated'signal. A reversal in phase of e(t) occurs whenever

212 / Mathematical Modeling of Physical Systems Chap. 5

6c (t) Load

Gear train

Two-phase
ac motor
(a)

Fig. 5-29. (a) AC control system with potentiometers as error detector, (b)
Typical waveforms of signals in the control system of (a).

the signal crosses the zero-magnitude axis. This reversal in phase causes the ac
motor to reverse in direction according to the desired sense of correction of the
error 6£t). The name "suppressed-carrier modulation" stems from the fact that
when a signal 6 e (t) is modulated by a carrier signal v(t) according to Eq. (5-90),
Sec. 5.5 Error-Sensing Devices in Control Systems / 213

the resultant signal e(t) no longer contains the original carrier frequency co c . To
illustrate this, let us assume that 8£t) is also a sinusoid given by

0,(0 = sin o>X?) (5-91)

where normally, co s <C co c By use of familiar trigonometric

. relations, substitut-
ing Eqs. (5-89) and (5-91) into Eq. (5-90) we get

e(t) = $K s V[cos(co c - co s )t ~ cos (co c + co s )t] (5-92)

Therefore, no longer contains the carrier frequency co c or the signal fre-

e(t)

quency co s but it does have the two side bands co c + co s and co c — co s

, .

Interestingly enough, when the modulated signal is transmitted through the

system, the motor acts as a demodulator, so that the displacement of the load
will be of the same form as the dc signal before modulation. This is clearly seen
from the waveforms of Fig. 5-29(b).
should be pointed out that a control system need not contain all-dc or
It

components. It is quite common to couple a dc component to an ac com-

all-ac
ponent through a modulator or an ac device to a dc device through a demodula-
tor.For instance, the dc amplifier of the system in Fig. 5-28(a) may be replaced
by an ac amplifier that is preceded by a modulator and followed by a demodula-
tor.

Synchros. Among the various types of sensing devices for mechanical

shaft errors, themost widely used is a pair of synchros. Basically, a synchro is a
rotary device that operates on the same principle as a transformer and produces
a correlation between an angular position and a voltage or a set of voltages.
Depending upon the manufacturers, synchros are known by such trade names
as Selsyn, Autosyn, Diehlsyn, and Telesyn. There are many types and different
applications of synchros, but in this section only the synchro transmitter and the
synchro control transformer will be discussed.

Synchro Transmitter

A synchro transmitter has a Y-connected stator winding which resembles

the stator of a three-phase induction motor. The rotor is a salient-pole, dumb-
bell-shaped magnet with a single winding. The schematic diagram of a synchro
transmitter is shown in Fig. 5-30. A single-phase ac voltage is applied to the

Stator

Slip
*1 rings
— ,
Stator
ac voltage Rotor -
R-,

(a) (b)

Fig. 5-30. Schematic diagrams of a synchro transmitter.

214 / Mathematical Modeling of Physical Systems Chap. 5

rotor through two slip rings. The symbol G is often used to designate a synchro
transmitter, which is sometimes also known as a synchro generator.
Let the ac voltage applied to the rotor of a synchro transmitter be

e r (t) =E t sin co c t (5-93)

When the rotor is in the position shown in Fig. 5-30, which is defined as the
electric zero, the voltage induced across the stator winding between S 2 and the
neutral n is maximum and is written

e s,Sf) = KE r sin cot (5-94)

where A" is a proportional constant. The voltages across the terminals 5[« and
S3 n are

e Sm (t) = KE r cos 240° sin cot = — 0.5KE, sin cot (5-95)

e sm(0 = KE r cos 120° sin cot = —0.5KE, sin cot (5-96)

the terminal voltages of the stator are

e SiS , = e Sl „ - es = ,„ —\.5KE r sin cot

(5-97)
e Sl s, = e Stn — e Ss =„ 1.5KE, sin cot

e S,Si = e S,n e Sin ~ (5-98)

The above equations show that, despite the similarity between the construc-
tion of the stator of a synchro and that of a three-phase
machine, there are only single-phase voltages induced in
the stator.
Consider now that the rotor of the synchro trans-
mitter is allowed to rotate in a counterclockwise direc-
tion, as shown in Fig. 5-31. The voltages in each stator
winding will vary as a function of the cosine of the rotor
displacement 6; that is, the voltage magnitudes are

ESl „ = KE r cos (6 - 240°) (5-99)

Fig. 5-31. Rotor position Es „ = KE r cos 9 (5-100)

of a synchro transmitter.
ESin = KE r cos (0 - 120°) (5-101)

The magnitudes of the stator terminal voltages become

ESlS = ESin - ESm = ,/TKE, sin (9 +

.
240°) (5-102)

*J~TKE r sin (9 120°) (5-103)

Es ,s< ^fJKE T sin 9 (5-104)

A plot of these terminal voltages as a function of the rotor shaft position is

shown in Fig. 5-32. Notice that

each rotor position corresponds to one unique
set of stator voltages. This leads to the use of the synchro transmitter to identify
angular positions by measuring and identifying the set of voltages at the three
stator terminals.
Sec. 5.5 Error-Sensing Devices in Control Systems / 215

Volts

Rotor position
6 (degrees)

Fig. 5-32. Variation of the terminal voltages of a synchro transmitter as

a function of the rotor position. 9 is measured counterclockwise from
the electric zero.

Synchro Control Transformer

Since the function of an error detector is to convert the difference of two

shaft positions into an electrical signal, a single synchro transmitter is apparently
inadequate. A typical arrangement of a synchro error detector involves the use
of two synchros: a transmitter and a control transformer, as shown in Fig. 5-33.

Synchro Control
transmitter transformer
R, Rotor
Output voltage
proportional to
sin (0, - dc )

Fig. 5-33. Synchro error detector.

For small angular deviations between the two rotor positions, a proportional
voltage generated at the rotor terminals of the control transformer.
is

Basically, the principle of operation of a synchro control transformer is

identical to that of the synchro transmitter, except that the rotor is cylindrically
shaped so that the air-gap flux is uniformly distributed around the rotor. This
feature is essential for a control transformer, since its rotor terminals are usually
connected to an amplifier or similar electrical device, in order that the latter
sees a constant impedance. The change in the rotor impedance with rotations of
the shaft position should be minimized.
216 / Mathematical Modeling of Physical Systems Chap. 5

The symbol CT is often used to designate a synchro control transformer.

Referring to the arrangement shown in Fig. 5-33, the voltages given by
Eqs. (5-102), (5-103), and (5-104) are now impressed across the corresponding
stator terminals of the control transformer. Because of the similarity in the
magnetic construction, the flux patterns produced in the two synchros will be
the same if all losses are neglected. For example, if the rotor of the transmitter
is in its electric zero position, the fluxes in the transmitter and in the control
transformer are as shown in Fig. 5-34(a) and (b).

When the rotor of the control transformer is in the position shown in Fig.
5-34(b), the induced voltage at its rotor terminals is zero. The shafts of the two
synchros are considered to be in alignment. When the rotor position of the con-
trol transformer is rotated 180° from the position shown, its terminal voltage is

again zero. These are known as the two null positions of the error detector. If
the control transformer rotor is at an angle a from either of the null positions,

Synchro Control
transmitter transformer
°
-~~|
L/X a = °° or 180

/ /f ! ! I \ \ Rotor voltage =

Flux pattern

(b)

Control transformer

Control
transformer

Flux
pattern

Rotor pattern proportional Rotor voltage is at

to sin a maximum

(c) (d)

Fig. 5-34. Relations among flux patterns, rotor positions, and the rotor
voltage of synchro error detector.
Sec. 5.5 Error-Sensing Devices in Control Systems / 217

such as that shown in Fig. 5-34(c) and (d), the magnitude of the rotor voltage is
proportional to sin a. Similarly, it can be shown that when the transmitter shaft
is in any position other than that shown in Fig. 5-34(a), the flux patterns will
shift accordingly, and the rotor voltage of the control transformer will be pro-
portional to the sine of the difference of the rotor positions, a. The rotor voltage
of the control transformer versus the difference in positions of the rotors of the
transmitter and the control transformer is shown in Fig. 5-35.

Rotor
voltage Vr

360° a=(dr -dc )

Fig. 5-35. Rotor voltage of control transformer as a function of the dif-

ference of rotor positions.

From Fig. 5-35 it is apparent that the synchro error detector is a nonlinear
device. However, for small angular deviations of up to 15 degrees in the vicinity
of the two null positions, the rotor voltage of the control transformer is approxi-
mately proportional to the difference between the positions of the rotors of
the transmitter and the control transformer. Therefore, for small deviations,
the transfer function of the synchro error detector can be approximated by a
constant Ks :

K
where
^e~rri "
(5 105)

E= error voltage
Br= shaft position of synchro transmitter, degrees
9C= shaft position of synchro control transformer, degrees
9e= error in shaft positions
Ks = sensitivity of the error detector, volts per degree

The schematic diagram of a positional control system employing a synchro

error detector is shown in Fig. 5-36(a). The purpose of the control system is to
make the controlled shaft follow the angular displacement of the reference input
shaft as closely as possible. The rotor of the control transformer is mechan-
ically connected to the controlled shaft, and the rotor of the synchro trans-
mitter is connected to the reference input shaft. When the controlled shaft is

aligned with the reference shaft, the error voltage is zero and the motor does not

turn. When an angular misalignment exists, an error voltage of relative polarity

218 / Mathematical Modeling of Physical Systems Chap. 5

Two-phase
induction
motor

Reference Controlled
input output

(a)

or r Ks
e
ac
Motor
em
Gear
ec

amplifier load
+ ^-
1

(b)

Fig. 5-36. (a) Alternating current control system employing synchro

error detector, (b) Block diagram of system in (a).

appears at the amplifier input, and the output of the amplifier will drive the
motor in such a direction as to reduce the error. For small dev iations between the
controlled and the reference shafts, the synchro error detector can be represented
by the constant Ks given by Eq. (5-105). Then the linear operation of the posi-
tional control system can be represented by the block diagram of Fig. 5-36(b).
From the characteristic of the error detector shown in Fig. 5-35, it is clear that
Ks has opposite signs at the two null positions. However, in closed-loop systems,
only one of the two null positions is a true null; the other one corresponds to an
unstable operating point.
Suppose that, in thesystem shown in Fig. 5-36(a), the synchro positions are
close to the true null and the controlled shaft lags behind the reference shaft; a
positive error voltage will cause the motor to turn in the proper direction to cor-
rect this lag. But if the synchros are operating close to the false null, for the same
lag between 9 r and 6 C the error voltage is negative and the motor is driven in
,

the direction that will increase the lag. A larger lag in the controlled shaft posi-
 Sec. 5.6 Tachometers / 219

tion will increase the magnitude of the error voltage still further and cause the
motor to rotate in the same direction, until the true null position
is reached.

In reality, the error signal at the rotor terminals of the synchro control
transformer may be represented as a function of time. If the ac signal applied to
the rotor terminals of the transmitter is denoted by sin a> c t, where co c is known
as the carrier frequency, the error signal is given by

e(t) = Ks O (i) sin a

e c t (5-106)

Therefore, as explained earlier, e(t) is again a suppressed-carrier modulated

signal.

5.6 Tachometers 4 - 5

Tachometers are electromechanical devices that convert mechanical energy into

electrical energy. The device works essentially as a generator with the output
voltage proportional to the magnitude of the angular velocity.
In control systems tachometers are used for the sensing of shaft velocity and
for the improvement of system performance. For instance, in a control system
with the output displacement designated as the state variable *, and the output
velocity as the state variable x2 , variable may be excessed to by
the first state

means of a potentiometer while x 2 monitored by a tachometer.

is

In general, tachometers may be classified into two types: ac and dc. The
simplified schematic diagrams of these two versions are
shown in Fig. 5-37. For the ac tachometer, a sinusoidal
o voltage of rated value is applied to the primary winding.
A secondary winding is placed at a 90° angle mechan-
ically with respect to the primary winding. When the

± rotor of the tachometer is stationary, the output voltage

'

at the secondary winding is zero. When the rotor shaft is

rotated, the output voltage of the tachometer is closely
o proportional to the rotor velocity. The polarity of the
dependent upon the direction of rotation,
Fig. 5-37. Schematic diagram of a
volta
p is

tachometer. The input-output relation of an ac tachometer can be

represented by a first-order differential equation

e l (t) =K (5-107)
!
^jp
where e,{t) is the output voltage, 9{t) the rotor displacement, and K, is defined as
the tachometer constant, usually represented in units of volts per rpm or volts
per 1000 rpm. The transfer function of an ac tachometer is obtained by taking the
Laplace transform of Eq. (5-107); thus

§g> = K,s (5-108)

A
dc tachometer serves exactly the same purpose as the ac tachometer
described above. One advantage of the dc tachometer is that the magnetic field
of the device may be set up by permanent magnet, and therefore no separate
 :

220 / Mathematical Modeling of Physical Systems Chap. 5

excitation voltage is required. In principle, the equations of Eqs. (5-107) and

(5-108) are also true for a dc tachometer.
A dc tachometer can also replace an ac tachometer in a control system by
use of a modulator to convert its dc output signal into ac. Similarly, a dc
tachometer can be replaced by an ac one if a phase-sensitive demodulator is used
to convert the ac output to dc.

5.7 DC Motors in Control Systems

Direct current motors are one of the most widely used prime movers in industry.
The advantages of dc motors are that they are available in a great variety of
types and sizes and that control is relatively simple. The primary disadvantage
of a dc motor relates to its brushes and commutator.
For general purposes, dc motors are classified as series-excited, shunt-
excited, and separately-excited, all of which refer to the way in which the field
is excited. However, the characteristics of the first two types of motor are highly

nonlinear, so for control systems applications, the separately-excited dc motor

is the most popular.
The separately-excited dc motor is divided into two groups, depending
on whether the control action is applied to the field terminals or to the arma-
ture terminals of the motor. These are referred to di% field controlled and armature
controlled, where usually the rotor of the motor is referred to as the armature
(although there are exceptions).
In recent years, advanced design and manufacturing techniques, have pro-
duced dc motors with permanent-magnet fields of high field intensity and rotors
—
of very low inertia motors with very high torque-to-inertia ratios, in other
words. It is possible to have a 4-hp motor with a mechanical time constant as
low as 2 milliseconds. The high torque-to-inertia ratio of dc motors has opened
new applications for motors in computer peripheral equipment such as tape
drives, printers, and disk packs, as well as in the machine tool industry. Of
course, when a dc motor has a permanent-magnetic field it is necessarily arma-
ture controlled.
The mathematical modeling of the armature-controlled and field -controlled
dc motors, including the permanent-magnet dc motor, is discussed in the follow-
ing.

Field-Controlled DC Motor
The schematic diagram of a field-controlled dc motor is shown in Fig. 5-38.

The following motor variables and parameters are defined

ea {t) armature voltage

Ra armature resistance
0(f) air-gap flux
e b (t) back emf (electromotive force) voltage
Kb back emf constant
Sec. 5.7 DC Motors in Control Systems / 221

/. = constant

Fig. 5-38. Schematic diagram of a field-controlled dc motor.

K, torque constant
4(f) armature current
i (t)
f field current
e f (t) field voltage
Tm (t) torque developed by motor
Jm rotor inertia of motor
Bm viscous frictional coefficient
OJt) angular rotor displacement

To carry out a linear analysis, the following assumptions are made:

1 The armature current is held constant, /'„ = /„.

2. The air-gap flux is proportional to the field current; that is,

0(0 = Kf i {t)
f (5-109)

3. The torque developed by the motor is proportional to the air-gap

fluxand the armature current. Thus

TJt) = Km IJ(t) (5-H0)

where Km is a constant. Substituting Eq. (5-109) into Eq. (5-110)

gives
TJt) = Km Kf IJf {t) (5-111)
If we let

Kt — Km Kf I <t
(5-112)

be the torque constant, Eq. (5-111) becomes

TJf) =K t
i
f {t) (5-113)

Referring to the motor circuit diagram of Fig. 5-38, the state variables are
assigned as if (t), cojt), and OJt). The first-order differential equations relating
these state variables and the other variables are written

L/-^ = -RMt) + e f (t) (5-114)

222 / Mathematical Modeling of Physical Systems Chap. 5

Jn
dcoJt) = _ BmCOm(t) + TJf) (5-115)
dt

dBJf) = co m (t) (5-116)

dt

By proper substitution, the last three equations are written in the form of state
equations:
~
'
dif {t) -
R f
'
1
i/O
dt

d(Q m (t) Kj Bm ~.

dt J J
coJJ) + efy) (5-117)

ddjt)
1 [9Jf)
. dt _

The state diagram of the system is drawn as shown in Fig. 5-39. The transfer

o Q O

"s-l

ULf Ki/Jm
o-
e
f
-R L ~B m Um
f' f

Fig. 5-39. State diagram of a field-controlled dc motor.

function between the motor displacement and the input voltage is obtained from
the state diagram as

fl-fr) _ K,
(5-118)
Ef (s) Lf Jm s 3 + (R,JM + Bm Lf )s + R B n s 2
a

or
Om(s) _ K t
(5-H9)
Ef(s) R a Bm s(l + T m s)(l + x f 8)
where

r_ = -=P- = mechanical time constant of motor

xf = -=£
K
= field electrical time constant of motor
f

Armature-Controlled DC Motor
The schematic diagram of an armature-controlled dc motor is shown in
Fig. 5-40. In this case for linear operation it is necessary to hold the field current
of the motor constant. The torque constant K t
relates the motor torque and the
Sec. 5.7 DC Motors in Control Systems / 223

+o » WvV

Constant field
current
o

Fig. 5-40. Schematic diagram of an armature-controlled dc motor.

armature current, i a (t) ; thus

TJt) = K,Ut) (5-120)

where K
t
is a function of the air-gap flux, which is constant in this case. Of
course, in the case of a permanent-magnet motor, (f>
is constant also. The back
emf voltage is proportional to the motor speed,

=K dOJf) K
e„(t) t b co m (t) (5-121)
dt

With reference to Fig. 5-40, the state equations of the armature-controlled

dc motor are written

diJLO I Ra Kb

dt
r
La ao
dco m (t)
= Bm
dt
co m (0 + ej,t) (5-122)

dBJf)
1 OJt)
dt _

The state diagram of the system is drawn as shown in Fig. 5-41 The transfer .

function between the motor displacement and the input voltage is obtained from
the state diagram as

e m (s) K,
(5-123)
E (s)
a La Jm s> + (R a Jm + B m L )s +
a
2
(Kb K t
+ R Bm )s
a

Although a dc motor is basically an open-loop system, as in the case of

the field-controlled motor, the state diagram of Fig. 5-41 shows that the arma-
ture-controlled motor has a "built-in" feedback loop caused by the back emf.
Physically, the back emf voltage represents the feedback of a signal that is pro-
portional to the negative of the speed of the motor. As seen from Eq. (5-123),
the back emf constant, Kb represents an added term to the resistance and fac-
,

tional coefficient. Therefore, the back emf effect is equivalent to an "electrical

friction," which tends to improve the stability of the motor.
It should be noted that in the English unit system, K, is in lb-ft/amp or

oz-in/amp and the back emf constant Kb is in V/rad/sec. With these units, Kt

and K b are related by a constant factor. We can write the mechanical power
224 / Mathematical Modeling of Physical Systems Chap. 5

- Kb IL a
Fig. 5-41. State diagram of an armature-controlled dc motor.

developed in the motor armature as

P = e (t)i
b a (t) watts

e b {t)ia {t) (5-124)

~ hp
746

Substituting Eqs. (5-120) and (5-121) into Eq. (5-124), we get

K T ddjjt)
p ~ 146K,m dt
h
hp (5-125)

This power is equal to the mechanical power

P ~ Tm de m (t) hp (5-126)
550 dt

Therefore, equating Eqs. (5-125) and (5-126), we have

K = t
0.737 K
b (5-127)

We can also determine the constants of the motor using torque-speed

curves. A typical set of torque-speed curves for various applied voltages of a dc
motor is shown in Fig. 5-42. The rated voltage is denoted by E At no load, r.

Tm = 0, the speed is given by the intersect on the abscissa for a given applied
voltage. Then the back emf constant K b is given by

K»-a (5-128)
r

where in this case the rated values are used for voltage and angular velocity.
 Two-Phase Induction Motor / 225

Speed co m rad/sec

Fig. 5-42. Typical torque-speed curves of a dc motor.

When the motor is stalled, the blocked-rotor torque at the rated voltage is

designated by T (t). Let

, _
~~
blocked-rotor torque at rated voltage
~~
T _ (5-129)
rated voltage ~Er
Also,
K,
T {t) = K ijj) = ^E
t r (5-130)

Therefore, from the last two equations, the torque constant is determined:

K< = kR a (5-131)

5.8 Two-Phase Induction Motor 6,7

For low-power applications in control systems ac motors

are preferred over dc motors because they are more
Control
phase rugged and have no brushes to maintain. Most ac motors
used in control systems- are of the two-phase induction

r^WS type, which generally are rated from a fraction of a

watt up to a few hundred watts. The frequency of the
motor is normally at 60, 400, or 1000 Hz.
A schematic diagram of a two-phase induction
motor is shown in Fig. 5-43. The motor consists of a

6 e l (t) o
stator with two distributed windings displaced 90 elec-
trical degrees apart. Under normal operating conditions
Reference phase
in control applications, a fixed voltage from a constant-
Fig. 5-43. Schematic diagram of a two-phase voltage source is applied to one phase, the fixed or
induction motor. reference phase. The other phase, the control phase, is
226 / Mathematical Modeling of Physical Systems Chap. 5

energized by a voltage that is 90° out of phase with respect to the voltage of
the fixed phase. The control-phase
voltage is usually supplied from a servo
amplifier,and the voltage has a variable amplitude and polarity. The direction of
rotation of the motor reverses when the control phase signal changes its sign.
Unlike that of a dc motor, the torque-speed curve of a two-phase induction
motor is quite nonlinear. However, for linear analysis, it is generally considered
an acceptable practice to approximate the torque-speed curves of a two-phase
induction motor by straight lines, such as those shown in Fig. 5-44. These curves
are assumed to be straight lines parallel to the torque-speed curve at a rated
control voltage (E2 = E = rated value), and they are equally spaced for equal
t

increments of the control voltage.

E 2 > E, > E 22 > E n > E M

,

Speed cj m

Fig. 5-44. Typical linearized torque-speed curves of a two-phase induc-

tion motor.

The state equations of the motor are determined as follows. Let k be the
blocked-rotor torque at rated voltage per unit control voltage; that is,

blocked-rotor torque at E2 = E, _ 7\,

(5-132)
rated control voltage E x
~~ E 1

Let m be a negative number which represents the slope of the linearized torque-
speed curve shown in Fig. 5-44. Then

blocked-rotor torque _
— T (5-133)
no-load speed fi

For any torque Tm , the family of straight lines in Fig. 5-44 is represented by the
equation
Tm {t) = mmjf) + ke t (t) (5-134)

where cojf) is the speed of the motor and e 2 (t) the control voltage. Now, if we
Sec. 5.8 Two-Phase Induction Motor / 227

designate co m (t) as a state variable, one of the state equations may be obtained
from

Jm ^P=-B n o, m (t) + Tm (t) (5-135)

Substituting Eq. (5-134) into Eq. (5-135) and recognizing that m (t) is the
other state variable, we have the two state equations

d9 m {t)
dt
= <o m (t) (5-136)

±&>-U-K». + $. M (5-137)

The state diagram of the two-phase induction motor is shown in Fig. 5-45. The

(m-B m )/Jm
Fig. 5-45. State diagram of the two-phase induction motor.

transfer function of the motor between the control voltage and the motor dis-
placement is obtained as

e m (s) k
(5-138)
E 2 (s) (Bm - m)s[\ + JJ(B m - m)s]
or
m (s) _ Km
(5-139)
E 2 (s) s(l + r m s)
where

Km = B k = motor gain constant

m -m
(5-140)

=
T„
Bm -m = motor time constant (5-141)

Since m
is a negative number, the equations above show that the effect of

the slope of the torque-speed curve is to add more friction to the motor, thus
improving the damping or stability of the motor. Therefore, the slope of the
228 / Mathematical Modeling of Physical Systems Chap. 5

torque-speed curve of a two-phase induction motor is analogous to the back emf

effect of a dc motor. However, if m is a positive number, negative damping
occurs for m > Bm , it can be shown that the motor becomes unstable.

5.9 Step Motors 8

A step motor is an electromagnetic incremental actuator that converts digital

pulse inputs to analog output shaft motion. In a rotary step motor, the output
shaft of the motorrotates in equal increments in response to a train of input
pulses. When properly controlled, the output steps of a step motor are equal in
number to the number of input pulses. Because modern control systems often
have incremental motion of one type or another, step motors have become an
important actuator in recent years. For instance, incremental motion control is
found in all types of computer peripheral equipment, such as printers, tape
drives, capstan drives, and memory-access mechanisms, as well as in a great
variety of machine tool and process control systems. Figure 5-46 illustrates the

Fig. 5-46. Paper-drive mechanism using a step motor.

application of a step motor in the paper-drive mechanism of a printer. In this

case the motor coupled directly to the platen so that the paper is driven a
is

certain increment at a time. Typical resolution of step motors available com-

mercially ranges from several steps per revolution to as much as 800 steps per
revolution, and even higher.
Step motors come in a variety of types, depending upon the principle of
operation. The two most common types of step motors are the variable-reluc-
tance type and the permanent-magnet type. The complete mathematical analysis
of these motors is highly complex, since the motor characteristics are very non-
linear. Unlike dc and induction motors, linearized models of a step motor are
usually unrealistic. In this section we shall describe the principle of operation
and a simplified mathematical model of the variable-reluctance motor.
The variable-reluctance step motor has a wound stator and an unexcited
rotor. The motor can be of the single- or the multiple-stack type. In the multiple-
stack version, the stator and the rotor consist of three or more separate sets of
teeth. The separate sets of teeth on the rotor, usually laminated, are mounted on
the same shaft. The teeth on all portions of the rotor are perfectly aligned.
Figure 5-47 shows a typical rotor-stator model of a motor that has three sepa-
Sec. 5.9 Step Motors / 229

Stator "C

Rotor "A'

Fig. 5-47. Schematic diagram of the arrangement of rotor and stator teeth
in a multiple-stack, three-phase variable-reluctance step motor. The motor
is shown to have 12 teeth on each stack or phase.

rate sections on the rotor, or a three-phase motor. A variable-reluctance step

motor must have at least three phases in order to have directional control. The
three sets of rotor teeth are magnetically independent, and are assembled to one
shaft, which is supported by bearings. Arranged around each rotor section is a
stator core with windings. The windings are not shown in Fig. 5-47. Figure 5-48
is a schematic diagram of the windings on the stator. The end view of the
stator
of one phase, and the rotor, of a practical motor is shown in Fig. 5-49. In this
 Phase A Phase B Phase C

h <
. '* < . h <
»

~ »
» »
: ;
_ 1 . \
,
; i

_ ,,

Fig. 5-48. Schematic diagram of a multiple-stack three-phase variable-

reluctance step motor.

Fig. 5-49. End view of the stator of one phase of a multiple-stack variable-
reluctance step motor.

230
Sec. 5.9 Step Motors / 231

case the rotor is shown at a position where its teeth are in alignment with that of
the particular phase of the stator.
The rotor and stator have the same number of teeth, which means that the
tooth pitch on the stator and the rotor are the same. To make the motor rotate,
the stator sections of the three-phase motor are indexed one-third of a tooth
pitch, in the same direction. Figure 5-50 shows this arrangement for a 10-tooth

"Phase B
Fig. 5-50.Rotor and stator teeth arrangements of a multiple-stack three-
phase variable-reluctance step motor. The rotor has 10 teeth.

rotor. Therefore, the teeth on one stator phase are displaced 12° with respect to
the stator phase. Here the teeth of phase C of the stator are shown to be aligned
with the corresponding rotor teeth. The teeth of phase A
of the stator are dis-
placed clockwise by 12° with respect to the teeth of phase C. The teeth of phase
B of the stator are displaced 12° clockwise with respect to those of phase A, or
12° counterclockwise with respect to those of phase C. It is easy to see that a
minimum of three phases is necessary to give directional control. In general,
four-and five-phase motors are also common, and motors with as many as eight
phases are available commercially. For an n-phase motor, the stator teeth are
displaced by l/« of a tooth pitch from section to section.
The operating principle of the variable-reluctance stepping motor is straight-
forward. Let any one phase of the windings be energized with a dc signal. The
magnetomotive force setup will position the rotor such that the teeth of the
rotor section under the excited phase are aligned opposite the teeth on the excited
phase of the stator. This is the position of minimum reluctance, and the motor
is in a stable equilibrium.
232 / Mathematical Modeling of Physical Systems Chap. 5

If phase C is energized in Fig. 5-50, the rotor would be (in steady state)
positioned as shown. It can also be visualized from the same figure that if the dc

signal is switched to phase A, the rotor will rotate by 12°, clockwise, and the
rotor teeth will be aligned opposite the teeth of phase A of the stator. Continu-
ing in the same way, the input sequence CABCAB will rotate the motor clock-
wise in steps of 12°.
Reversing the input sequence will reverse the direction of rotation. That is,

the input sequence CBACB will rotate the motor in the counterclockwise direc-
tion in steps of 12°.
The steady-state torque curve of each phase is approximately as shown in
Fig. 5-51. The 0° line represents the axis of any tooth of the energized stator
phase. The nearest rotor tooth axis will always lie within 18° on either side of
The corresponding starting torque exerted when this phase is energized
this line.
can be seen in Fig. 5-51. The arrows mark the direction of motion of the rotor.

Torque

Fig. 5-51. Torque curve for one phase of a step motor.

Let positive angular displacements represent clockwise motion. Suppose

also that phase C has been excited for a long time. This means that the initial
condition of the rotor will be as shown in Fig. 5-50. If phase A is now energized
and Fig. 5-51 represents the torque variation of phase A, the initial position of
the rotor teeth will be at —12°. As soon as phase A is energized, the rotor will
finally settle after some oscillations, assuming that the inertia and friction are
such that there is no overshoot beyond the 18° point.
It may be noticed that the position, ±18°, also represents an equilibrium
point. This is because in that position the deflecting torque is zero. It is, however,
a position of unstable equilibrium since the slightest shift from this position will
send the motor straight to 0°.
8°
If on energizing one phase the rotor happens to lie exactly at the ± 1

point, theoretically it will stay there. In practice, however, there will always be
some mechanical imperfections in construction, and the resulting asymmetry
will prevent any locking at the unstable point.
We now look upon the stepping motor from a single-step point of view and
try to develop the equations that govern its performance. Several assumptions
will be made initially to simplify the development. Subsequent modifications
Sec. 5.9 Step Motors / 233

may be made if any of these assumptions are found to be invalidated. We start

by writing the equation for the electrical circuit of the stator winding. Let
e(t) = applied voltage per phase

R= winding resistance per phase

L(0) = winding inductance per phase

i(t) — current per phase

8(t) = angular displacement

The voltage-current equation of one stator phase is written

e(t) = Ri{t) + ±[iL(d)]

(5 ' 142)
= RKt)
M d
+ L{e)f +iji ue)t

or

= +
e(t) Ri(t) L(9)
f+
t
/ JJgUp)
f (
(5-143)

The term, L(9)(di/dt) represents transformer electromotive force, or self-

induced electromotive force, and the term i[dL(6)/d9](d6/dt) represents the
back emf due to the rotation of the rotor. We have assumed above that the
inductance is a function of 9(t), the angular displacement, only. No dependence
on the current has been assumed. This will reflect directly on our procedure for
getting the torque developed by the motor.
The energy in the air gap can be written
W = \L{9)i\t) (5-144)

From elementary electromechanical energy-conversion principles, we know

that the torque in a single excited rotational system is given by

T=£s [W(i,0)] (5-145)

where W is the energy of the system expressed explicitly in terms of i(i) and
6{t). Therefore,

T = \i\t)j [L{e)] (5-146)

d
This torque is then applied to the rotor and the equation of motion is
obtained as

T = jJ^ + Bj-f (5-147)

where Jm is the rotor inertia and Bm the viscous frictional coefficient. Jm and Bm
also may include the effects of any load.
To complete the torque expression of Eq. (5-146), we need to know the
form of the inductance L(d). In practice the motor inductance as a function of
displacement may be approximated by a cosine wave; that is,

L(6) = Z,, + L2 cos rd (5-148)

234 / Mathematical Modeling of Physical Systems Chap. 5

where L x and L 2 are constants and r is the number of teeth on each rotor section.
Substituting Eq. (5-148) into Eq. (5-146), we get
T = -%L 2 ri 2 (t) sin rQ = -Ki\t) sin r6 (5-149)

which is the sinusoidal approximation of the static torque curve of Fig. 5-51.
Now let us apply these equations to a three-phase motor. Let the equilib-
rium position be the situation when phase A is energized. Then the inductance
and torque for phase A are given by

LA = L, +L 2 cos rO (5-150)

TA = -Ki\ sin rd (5-151)

^K>

eB ~ s bB

=K> /
RB + sL B

ebB

~o

Fig. 5-52. Block diagram of the variable-reluctance step motor.

 :

Sec. 5.10 Tension-Control System / 235

respectively. For the = 10, phase B has its

10-teeth rotor considered earlier, r
equilibrium position 12° behind the reference point, assuming that the sequence
ABC represents forward motion. Similarly, the equilibrium position of phase C
is 12° ahead of the reference point. Therefore, the inductances and the torques
for phases B and C are written as follows

LB = L, +L 2
- 120°)
cos (100 (5-152)

Lc = L, +L
cos (100 + 120°)
2 (5-153)

TB = -Ki\{t) sin (100 - 120°) (5-154)

Tc = -Kil(t) sin (100 + 120°) (5-155)

The electrical circuits of the three phases are isolated so that each phase has its

differential equation of the form of Eq. (5-143).

The total torque developed on the rotor is the algebraic sum of torques
of the three phases. Thus

T = TA + TB + Tc (5-156)

The nonlinearity of the torque equations precludes the use of linearized models
for the portrayal of a step motor. Therefore, realistic studies of a step motor
using the equations presented above can be made only through computer simula-
tion. A block-diagram representation of the motor, which may be used for
analog or digital computer simulation, is shown in Fig. 5-52.

5.10 Tension-Control System

The problem of proper tension control exists in a great variety of winding and
unwinding industrial processes. Such industries as paper, plastic, and wire all
have processes that involve unwinding and rewinding processes. For example,
in the paper industry, the paper is first wound on a roll in a form that is nonsale-
able, owing to nonuniform width and breaks. This roll is rewound to trim edges,
splice breaks, and slit to required widths. Proper tension during this rewinding
is mandatory for several reasons slick paper will telescope if not wound tightly
:

enough and the width will vary inversely as the tension. Conversely, during
storage, a roll wound at varying tensions has internal stresses that will cause it

to explode. Similar examples could be cited, but the need for proper tension
control is relatively simple to understand.
Most rewind systems contain an unwind roll, a windup roll driven by a
motor, and some type of dancer and/or pinch-roller assemblies between the two.
Some systems employ spring-with-damper idlers with feedback to motor drives
to control tension. Some use tension-measuring devices and feedback to a
motor-generator or brake on the unwind reel to hold tension at a constant
value.
In this section a specific type of tension-control system for unwind processes
is investigated.As shown in Fig. 5-53, the system has a dc-motor-driven windup
reel. The tension of the web is controlled by control of the armature voltage ea {i)
of the motor.
236 / Mathematical Modeling of Physical Systems Chap. 5

+ o

Pinch
rolls

Fig. 5-53. Tension-control system for a winding process.

The mathematical modeling of the system is conducted by writing the

equations of the dc motor.
Armature:

eJLQ = R.Ut) + L **p + K co m {t)

a h (5-157)

where
K = b back emf constant of motor
=
co m (t) angular velocity of motor

Torque equation:

TJf) = ^Vnetojt)] + BmeCOm (t) + nrT(f) (5-158)

where
r= effective radius of windup reel
Tm {t) = motor torque = K,i (t) a

n = gear-train ratio
T(t) = web tension
jme = Jm n JL = equivalent inertia at motor shaft
-\-
2

Bme = Bm + n BL = equivalent viscous friction coefficient

2

at motor shaft
=
JL effective inertia of windup reel
BL = viscous friction coefficient of windup reel

Since the web material is taken up by the windup reel as the process pro-
ceeds, the inertia JL and the radius r of the windup reel increase as functions of
time. This explains the reason the derivative of Jme co m is taken in Eq. (5-158).
Furthermore, if h denotes the thickness of the web,

dr h
(5-159)

Jr=&> (5-160)

Thus

where
dJ,
Ti = W
i „ 3 dr
Tt
(5-161)

p = mass density of the web material per width

 Sec. 5.11 Edge-Guide Control System / 237

Assume now that the web material has a coefficient of elasticity of C and
that Hooke's law is obeyed. Then

£%p = C[v w (t) - »,(/)] (5-162)

where v s {t) is the web velocity at the pinch rolls. Assuming that the pinch rolls
are driven at constant speed, v,(f) = constant = V Also, s.

vjt) = roojt) = nrajf) (5-163)

It is apparent now that because r and JL are functions of time, Eq. (5-158)
is a time-varying nonlinear differential equation. However, if the web is very
thin, h ~ 0, we may consider that over a certain time period r and JL are con-
stant. Then, the linearized state equations of the system are written

^dF = - r>(0 " T aJf) + T

m a
e - (t) (5 " 164)

%^ = *"
}
jt-Ut)
*^ me
- j^cojf)
me
- "17X0
me
(5-165)

^ = Cnrmjt) - CV S (5-166)

5.11 Edge- Guide Control System

Whenever there is a tension-control problem there is a desire for edge-guide con-

trol. Therefore, the problem of edge guiding has become very important in
modern paper manufacturing, steel strip processing lines, flexographic and
textile industries, and similar fields. To maintain optimum sheet quality and

maximum process line speed, the moving web must be maintained at a constant
lateral-edge position. In general, there are many different ways of measuring and
tracking the edge of a moving web. However, to achieve stable and accurate
edge guiding, a feedback control system should be used.
The schematic diagram of an edge-guide system using the pivot-roll method
is shown in Fig. 5-54. The pivot roll is controlled to rotate about the pivot

point in guiding the direction of the web. The source of controlling the motion
of the pivot roll may be a dc motor coupled to a lead screw or rack and pinion,
or a linear actuator.
Figure 5-55 shows the side view of the edge-guide system. The axes of the
rollers 1 and 2 are assumed to be fixed or uncontrolled. S^ and S 2 represent

sensors that are placed at the indicated points to sense the centering of the web
at the respective points.
Let
v(t) = linear velocity of web
^r(0 = initial error of web position in the z direction at roll 1

Z;(0 = error of web position in the z direction at the leading

side of the pivot roll (see Figs. 5-54 and 5-55)
 Roll 2

Fig. 5-54. Schematic diagram of an edge-guide control system.

Fig. 5-55. Side view of an edge-guide system.

238
Sec. 5.1 Edge-Guide Control System / 239

Assuming that there is no slippage when the web moves over the pivot roll,

the following equation is written for v(t) and z x

(t):

^2 = t<0tana
at
(5-167)

If the angle a is small, from Fig. 5-54 it is seen that

tan a ~ z «(0 - *i(0 (5. 1 68)

m,
Thus Eq. (5-167) can be written approximately as

dzgt) = z R (t) - z.(f)

v(t) (5 . 169)
ox in 1

If the linear velocity of the web is constant, v(t) = v, and Eq. (5-169) is
written

Taking the Laplace transform on both

^ at + £*»<'>
Wj = £*«<'>
m x

sides of Eq. (5-170), the transfer

(5 " 170)

relation between z (t) and z R (t) is

t

^rr\
Z x(j) = T-r
+ Ti5
1 — (5
" 171 )

where Tj = mjv.
Assuming that there is no stretching in the web, from Fig. 5-55,

zz(t) - Zl(t) _ z R (t) - z x

{t)
(5-172)
w 2 m t

or

zz(t) = z t
(t) + ^[z R (t) - z,(0] (5-173)
m.
n
Taking the Laplace transform on both sides of Eq. (5-173) and solving for
ZR {s), we have
Z ^= 1 + \lZ)r,s
Z ^ (5
" 174
>

Substitution of ZR (s) from Eq. (174) into Eq. (171) gives the transfer rela-
tion between Z x
(s) and Z (s),2

z i(J ) = \ (5-175)
Z 2 (s) + (mJm^TtS 1
v

When the pivot roll is rotated by an angle L from its reference position, the
error z 3 (t) will be affected by approximately D sin 9L (t). With no slippage of the
web over the pivot roll, the error z 3 (t) due to the error z {f) is written x

z 3 (t) = Zl -T)- (t DOAt) (5-176)

where T= nD/2v, and for small L {t), sin L (t) is approximated by 9 L{t). Take
the inverse Laplace transform on both sides of Eq. (5-176) which yields
Z,(j) = e- T 'Z t (s) - D6 L (s) (5-177)

Similar to the relationship between zR and z x ,

the transfer function between
240 / Mathematical Modeling of Physical Systems Chap. 5

z 3 {t) and z 5 (t) is

Z 5 (s) 1
(5-178)
Z 3 (s) 1 + r3s

where t 3 = m 3 /v. Also, in analogy to Eq. (5-174), the transfer relation between
z 3 and z 4 is
Z 4 (s) = 1 + (mjm )z 3 3s
(5-179)
Z {s)
3 1 +rs 3

Now consider that the drive motor of the system is an armature-controlled

dc motor with negligible inductance. The equation of the motor armature is

= Ejjs) - sK b OJs)
Us) (5-180)

where Ia{s) is the armature current, Kh the back emf constant, R a the armature

resistance, and 6m (s) the motor displacement. The torque is given by

Tm(s) = KJJ^s) (5-181)

where K t
is the torque constant.
The torque equation of the motor and load is

TJs) = (Jm s 2 + Bm s)9 m (s) + LF(s) (5-182)

where
Jm = inertia of motor
Bm = viscous frictional coefficient of motor
L— lead of screw
F(s) = transform of force in the x direction
and
F(s) = 2^(-^ + B^ s + klWl(s)
2
(5-183)

where
JL = inertia of pivot roll about pivot point
BL = viscous frictional coefficient at pivot point
KL = spring constant at pivot roll due to tension of web
Combining Eqs. (5-182) and (5-183), we have
BJs) 1
(5-184)
TJs) /me s 2 + B me s + Kme
where
' me " m (5-185)
\2nr)

B
B,me *~m
™ + (4)^ (5-186)

K- = Kl (5-187)
(^)
Also,
X(s) = r9 L {s) (5-188)

9 L (s) = £*„(,) (5-189)

A block diagram of the overall system is drawn as shown in Fig. 5-56 using
T3
J*
O
5

241
242 / Mathematical Modeling of Physical Systems Chap. 5

Eqs. (5-175), (5-177), (5-179), (5-180), (5-181), (5-184), and (5-189). The blocks
with transfer functions Hp {s) and H (s) represent possible locations of control-
c

lers of the edge-guide system. The design problem may involve the determination
of the transfer functions Hp (s) and H c (s), so that for a given error z 2 the error
z3 is minimized.

5.12 Systems with Transportation Lags

Thus far a great majority of the systems considered have transfer functions that
are quotients of polynomials. However, in the edge-guide control system of
Section 5.11, the relation between the variables z x
(t) and z 3 (t) is that of a pure
time delay. Then Z t
(s) and Z 3 (s)
are related through an exponential transfer
function e' Ts In general, pure time delays
. may be encountered in various types
of systems, especially systems with hydraulic, pneumatic, or mechanical trans-
missions. In these systems the output will not begin to respond to an input until
after a given time interval. Figure 5-57 illustrates examples in which transporta-

Metering
point ^> Solution /I

— —s (
S~^
) »
— Valve

—
N

[* d — ^-\ Solution B

(a)
Thickness
measuring gauge

Roller ( o
II 3)
zs
~*— Steel plate

Roller ( o
5)
\« d M
(b)

Fig. 5-57. Physical systems with transportation lags.

tion lags are observed. Figure 5-57(a) outlines an arrangement in which two
different fluids are to be mixed in appropriate proportions. To assure that a
homogeneous solution is measured, the monitoring point is located some dis-
tance from the mixing point. A transportation lag therefore exists between the
mixing point and the place where the change in concentration is detected. If
the rate of flow of the mixed solution is v inches per second and dis the distance
between the mixing and the metering points, the time lag is given by

T= d -

sec (5-190)

If it is assumed that the concentration at the mixing point is c(t) and that it

is reproduced without change T seconds later at the monitoring point, the

 :

Sec. 5.13 Sun-Seeker System / 243

measured quantity is

6(0 = c(t - T) (5-191)

The Laplace transform of the last equation is

B{s) = T
e~ *C(s) (5-192)

Thus the transfer function between b(t) and c(i) is

The arrangement shown

m =
in Fig. 5-57(b)
e
~ Ts

may be
(5 - 193)

thought of as a thickness
control of the rolling of steel plates. As in the case above, the transfer function
between the thickness at the rollers and the measuring point is given by Eq.
(5-193).
Other examples of transportation lags are found in human beings as control
systems where action and reaction are always accompanied by pure time delays.
The operation of the sample-and-hold device of a sampled-data system closely
resembles a pure time delay; it is sometimes approximated by a simple time-lag
term, e~ Ts .

In terms of state variables, a system with pure time delay can no longer be
described by the matrix-state equation

^p- = Ax(?) + Bu(0 (5-194)

A general state description of a system containing time lags is given by the fol-
lowing matrix differential-difference equation

^ UL
t A,Ht -
= t=l 77) ± Bju(t -
+ j=l Tj) (5-195)

where T, and Tj are fixed time delays. In this case Eq. (5-195) represents a
general situation where time delays may exist on both the inputs as well as the
states.

5.13 Sun-Seeker System

In this section we shall model a sun-seeker control system whose purpose is to

control the attitude of a space vehicle so that sun with high ac-
it will track the
curacy. In the system described, tracking of the sun in one plane is accomplished.
A schematic diagram of the system is shown in Fig. 5-58. The principal elements
of the error discriminator are two small rectangular silicon photovoltaic cells
mounted behind a rectangular slit in an enclosure. The cells are mounted in
such a way that when the sensor is pointed at the sun, the beam of light from the
slit overlaps both cells. The silicon cells are used as current sources and con-

nected in opposite polarity to the input of an operational amplifier. Any differ-

ence in the short-circuit current of the two cells is sensed and amplified by the
operational amplifier. Since the current of each cell is proportional to the illu-
mination on the an error signal will be present at the output of the amplifier
cell,

when the light from the slit is not precisely centered on the cells. This error
244 / Mathematical Modeling of Physical Systems Chap. 5

Error discriminator

dc tachometer

Fig. 5-58. Schematic diagram of a sun-seeker system.

voltage, when fed to the servoamplifier, will cause the servo to drive the system
back into alignment. A description of each part of the system is given as follows.

Coordinate System

The center of the coordinate system is considered to be at the output gear

of the system. The reference axis is taken to be the fixed frame of the dc motor,
and all rotations are measured with respect to this axis. The solar axis or the line
from the output gear to the sun makes an angle d r with respect to the reference
axis, and 9 denotes the vehicle axis with respect to the reference axis. The objec-
tive of the control system is to maintain the error between 6 r and O a, near ,

zero:
a = 6r - 6 (5-196)

Figure 5-59 shows the coordinate system described.

Error Discriminator

When the vehicle is aligned perfectly with the sun, a = 0. and Ia = Ib = I,

or Ia — Ib = 0. From the geometry of the sun's rays and the photovoltaic cells
shown in Fig. 5-58, we have
oa
w + L tan a
^
= -^- (5-197)

Ltana (5-198)
Sec. 5.13 Sun-Seeker System / 245

Center of
output gear Fixed axis of
dc motor frame

Fig. 5-59. Coordinate system of the sun-seeker system.

where oa denotes the width of the sun's ray that shines on cell A, and ob is the
same on cell B, for a given a. Since the current Ia is proportional to oa, and Ib
to ob, we have

IT 1
/ + ^tan a (5-199)

and

= I-
t 2LI t
I„ -jp- tan a (5-200)

for < tan a < W/2L. For W/2L < tan a < (C - W/2)/L, the sun's ray com- is

pletely on cell A, and I„ = 21, Ib = 0. For (C - W/2)/L < tan a < (C + W/2)/L,
J„ decreases linearly from 21 to zero. = I = for tan a > (C + W/2)/L.
/„ b

Therefore, the error discriminator may be represented by the nonlinear charac-

teristic of Fig. 5-60, where for small angle a, tan a has been approximated by a
on the abscissa.

Fig. 5-60. Nonlinear characteristic of the error discriminator. The abscissa

is tan a but is approximated by a for small values of a.
246 / Mathematical Modeling of Physical Systems Chap. 5

Operational Amplifier

The schematic diagram of the operational amplifier is shown in Fig. 5-61.

Summing the currents at point G and assuming that the amplifier does not draw
any current, we have
e
/. -h- 7f
+ -SLjjrI* = (5-201)

-VWV-

o WW-

R
-JvVW-

r 1

1-
Fig. 5-61. Operational amplifier.

Since e = —Aee , eg = —e /A, Eq. (5-201) becomes

/«- / =
' + (i + i + i> "
(5 202)

If A approaches infinity, as in operational amplifiers, A may reach 10 6 ; then

Eq. (5-202) is written
e = -R F(I - a I„) (5-203)

This equation gives the transfer relationship between Ia — Ib and e .

Servoamplifier

The gain of the servoamplifier is —K a. With reference to Fig. 5-58, the

output of the servo amplifier is expressed as

(5-204)
= -K e a s

.
Tachometer
The output voltage of the tachometer eT is related to the angular velocity of
the motor through the tachometer constant KT that is, ;

eT = KT co m (5-205)

The angular position of the output gear is related to the motor position through
Thus
the gear ratio l/n.

On = 9, (5-206)
u

00

247
248 / Mathematical Modeling of Physical Systems Chap. 5

Armature-Controlled DC Motor

The armature-controlled dc motor has been modeled earlier. The equations

are
= RJa + e„
ea (5-207)

e = K„co m
b (5-208)

Tm = KJ. (5-209)

Tm = J^ + Bco m (5-210)

where / and B are the inertia and viscous friction coefficient seen at the motor
shaft. The inductance in the motor armature is neglected in Eq. (5-207).
A block diagram that characterizes all the functional relations of the system
is shown in Fig. 5-62.

REFERENCES
State-Variable Analysis of Electric Networks

1. B. C. Kuo, Linear Circuits and Systems, McGraw-Hill Book Company, New

York, 1967.

2. R. Rohrer, Circuit Analysis: An Introduction to the State Variable Approach,

McGraw-Hill Book Company, New York, 1970.

Mechanical Systems

3. R. Cannon, Dynamics of Physical Systems, McGraw-Hill Bock Company, New

York, 1967.

Control System Components

4. W. R. Ahrendt, Servomechanism Practice, McGraw-Hill Book Company, New

York, 1954.

5. J.E. Gibson and F. B. Tuteur, Control System Components, McGraw-Hill Book

Company, New York, 1958.

Two-Phase Induction Motor

6. W. A. Stein and G. J. Thaler, "Effect of Nonlinearity in a 2-Phase Servomotor,"

AIEE Trans., Vol. 73, Part II, pp. 518-521, 1954.
7. B. C. Kuo, "Studying the Two-Phase Servomotor," Instrument Soc. Amer. J.,

Vol. 7, No. 4, pp. 64-65, Apr. 1960.

Step Motors

8. B. C. Kuo (ed.), Proceedings, Symposium on Incremental Motion Control Systems

and Devices, Part I, Step Motors and Controls, University of Illinois, Urbana, 111.,

1972.

9. B. C. Kuo (ed.), Proceedings, Second Annual Symposium on Incremental Motion

Control Systems and Devices, University of Illinois, Urbana, 111., 1973.
Chap. 5 Problems / 249

10. B. C. Kuo, Theory and Applications of Step Motors, West Publishing Company,
St. Paul, Minn., 1974.

PROBLEMS

5.1. Write the force or torque equations for the mechanical systems shown in Fig.
from the force or torque equations.
P5-1. Write the state equations

B,
Af,

B
777777m777m777m7777mm7m7777MM77M7M777,
M,
B,
F(t)

(a)

*- Fit)

WW///////////////////

(b)

(c)

Figure P5-1.

5.2. Write a set of state equations for the mechanical system shown in Fig. P5-2.
On the first try, one will probably end up with four state equations with the
state variables defined as 2, 0} 2 , 0i, and ©,. However, it is apparent that there
are only three energy-storage elements in /j, K, and /2 so the system has a
,

minimum order of three.

K
no, h h

6l , co t 62 ,
cj 2

Figure P5-2.
250 / Mathematical Modeling of Physical Systems Chap. 5

(a) Write the state equations in vector-matrix form with the state variables
defined as above.
(b) Redefine the state variables so that there are only three state equations.
(c) Draw state diagrams for both cases.
(d) Derive the transfer function (o 2 (s)/T(s) for each case, and compare the
results.
(e) Determine the controllability of the system. Does the fact that the system
can be modeled by three state equations mean that the four-state model is
uncontrollable? Explain.
5.3. For the system shown in Fig. P5-3, determine the transfer function E (s)/Tm (s).
The potentiometer rotates through 10 turns, and the voltage applied across the
potentiometer terminals is E volts.

+
—E
I

I
i_
Tm it)
Potentiometer

B2 = viscous friction coefficient of

potentiometer contact

Figure P5-3.

5.4. Write the torque equations of the gear-train system shown in Fig. P5-4. The
moments of inertia of the gears and shafts are Jit J2 and J 3 Tit) , . is the applied
torque. N denotes the number of gear teeth. Assume rigid shafts.
Chap. 5 Problems / 251

5.5. Figure P5-5 shows the diagram of an epicyclic gear train.

Figure P5-5.

(a) Using the reference directions of the angular velocity variables as indicated,
write algebraic equations that relate these variables.
(b) Draw a signal flow graph to relate among the inputs 0i a and ©i and the
output co 6
.

(c) Find the transfer function relation between (o s and (O x and (O a .

5.6. The block diagram of the automatic braking control of a high-speed train is

shown in Fig. P5-6a, where

^ e
.
Amplifier
K
eb
Brake
M Train
v(t)

>

Tachometer

(a)

(Sec)

(b)

Figure P5-6.
 :

252 / Mathematical Modeling of Physical Systems Chap. 5

V, = voltage representing desired speed

v = velocity of train
K = amplifier gain = 100
M = mass of train = 5 x 10* lb/ft/sec 2

K, = tachometer constant = 0.15 volt/ft/sec

e, = K,v

The force characteristics of the brake are shown in Fig. P5-6b when e b = 1 volt.

(Neglect all frictional forces.)

(a) Drawa block diagram of the system and include the transfer function of
each block.
(b) Determine the closed-loop transfer function between V, and velocity v of
the train.
(c) If the steady-state velocity of the train is to be maintained at 20 ft/sec,

what should be the value of Vr ?

5.7. Figure P5-7 illustrates a winding process of newsprint. The system parameters
and variables are defined as follows

+ o

Figure P5-7.

ea= applied voltage

R = armature resistance of dc motor
L = armature inductance of dc motor
4 = armature current
K = back emf of dc motor
b

Tm = motor torque = Km ia

Jm = motor inertia
Bm = motor friction coefficient
JL = inertia of windup reel
co m = angular velocity of dc motor
co = angular velocity of windup reel
TL = torque at the windup reel
r = effective radius of windup reel
V„ = linear velocity of web at windup reel
T = tension
V„ = linear velocity of web at input pinch rolls

Assume that the linear velocity at the input pinch rolls, V s, is constant. The
elasticity of the web material is assumed to satisfy Hooke's law; that is, the dis-
tortion of the material is directly proportional to the force applied, and the
proportional constant is K (force/displacement).
 :

Chap. 5 Problems / 253

(a) Write the nonlinear state equations for the system using i„, co m , and T as
state variables.
(b) Assuming that r is constant, draw a state diagram of the system with ea and
V s as inputs.

5.8. Write state equations and output equation for the edge-guide control system
whose block diagram is shown in Fig. 5-56.
5.9. The schematic diagram of a steel rolling process is shown in Fig. P5-9.

Two-phase
induction motor

Om U),Tm {t)

Gear box and

linear actuator

Steel
plate

bit)-- Ks dt)
ait)-- «M0
Figure P5-9.

(a) Describe the system by a set of differential-difference equation of the form

of Eq. (5-195).
(b) Derive the transfer function between c(t) and r{t).

5.10. Figure P5-10a shows an industrial process in which a dc motor drives a capstan
and tape assembly. The objective is to drive the tape at a certain constant speed.
Another tape driven by a separate source is made to be in contact with the
primary tape by the action of a pinch roll over certain periods of time. When the
two tapes are in contact, we may consider that a constant frictional torque of
TF is seen at the load. The following system parameters are defined

e, = applied motor voltage, volts

4 = armature current, amperes

= back emf voltage = K com volts
et, b ,

K = back emf constant = 0.052 volt/rad/sec

b

Km = torque constant = 10 oz-in./amp

Tm = torque, oz-in.
254 / Mathematical Modeling of Physical Systems Chap. 5

'm 'm "m ^_

00

< :
^
J
>
e
•
1

Integral
control
e,
G l
(s)
um
G 2 (s)

1 volt/rad/s

Feedback transducer

(b)

Figure P5-10.

8m = motor displacement, rad

= motor speed, rad/sec
(O m
Ra = motor resistance = Q 1

Jm = motor inertia = 0.1 oz-in./rad/sec 2 (includes capstan inertia)

Bm = motor viscous friction = 0.1 oz-in./rad/sec
KL = spring constant of tape = 100 oz-in./rad (converted to rotational)
JL = load inertia = 0.1 oz-in./rad/sec 2
(a) Write the equations of the system in vector-matrix state equation form.
(b) Draw a state diagram for the system.
(c) Derive the transfer function for C0 L (s) with E (s) and TF (s) as inputs. t

(d) If a constant voltage e,(t) = 10 V is applied to the motor, find the steady-
state speed of the motor in rpm when the pinch roll is not activated. What
is the steady-state speed of the load ?
(e) When the pinch roll is activated, making the two tapes in contact, the
constant friction torque TF is 1 oz-in. Find the change in the steady-state
speedcojr, when e, = 10 V.
(f) To overcome the effect of the frictional torque 7> it is suggested that a
closed-loop system should be formed as shown by the block diagram in
Chap. 5 Problems / 255

Fig. P5-10(b). In this case the motor speed is fed back and compared with
the reference input. The closed-loop system should give accurate speed
control, and the integral control should give better regulation to the fric-
tional torque. Draw
a state diagram for the closed-loop system.
(g) Determine the steady-state speed of the load when the input is 1 V. First
consider that the pinch roll is not activated, and then is activated.
5.11. This problem deals with the attitude control of a guided missile. When traveling
through the atmosphere, a missile encounters aerodynamic forces that usually
tend to cause instability in the attitude of the missile. The basic concern from the
flight control standpoint is the lateral force of the air, which tends to rotate the
missile about its center of gravity. If the missile centerline is not aligned with
the direction in which the center of gravity C is traveling, as shown in Fig. P5-1

Figure P5-11.

with the angle 6 (9 is also called the angle of attack), a side force is produced by
the resistance of the air through which the missile is traveling. The total force
Fa may be considered to be centered at the center of pressure P. As shown in
Fig. P5-11, this side force has a tendency to cause the missile to tumble, espe-
cially if the point P
is in front of the center of gravity C. Let the angular accelera-

tion of the missile about the point C, due to the side force, be denoted by Ct F .

Normally, a F is directly proportional to the angle of attack 6 and is given by

a, F = ad
where a is a constant described by

Kr is a constant that depends on such parameters as dynamic pressure, velocity

of the missile, air density, and so on, and

J= missile moment of inertia about C

d =
t distance between C and P
The main object of the flight control system is to provide the stabilizing action
to counter the effect of the side force. One of the standard control means is to
256 / Mathematical Modeling of Physical Systems Chap. 5

use gas injection at the tail of the missile to deflect the direction of the rocket
engine thrust T, as shown in Fig. P5-1 1
(a) Write a torque differential equation to relate among T,d,9, and the system
parameters. Assume that S is very small.
(b) Assume that T is constant and find the transfer function 9(s)/S(s) for small
S.
(c) Repeat (a) and (b) with the points C and P interchanged.
5.12. (a) Draw a state diagram for the tension-control system of Fig. 5-53, using the
state equations of Eqs. (5-164), (5-165), and (5-166).

(b) Write the relation among Ea (s), Vs and , T(s), with EJs) and Vs as inputs and
T(s) as the output.

5.13. The following equations describe the motion of an electric train in a traction
system
x(t) = v{t)

i,(t) = -k(v) - g(x) + T(t)

where
x(t) = linear displacement of train
v(t) — linear velocity of train
k{v) = train resistance force [odd function off, with the properties
k(0) = and dk(v)\dv > 0]
g{x) = force due to gravity for a nonlevel track or due to curvature of track
T{t) = tractive force
The electric motor that provides the traction force is described by the following
relations
e(t) = Kb ^>(t)v{t) + Ria {t)

T(t) = Km <f>{t)ia (t)

where
R = armature resistance
= magnetic flux = Kf f (i)
<j>(t) i

e(t) = applied voltage

Km Kb = proportional constants
,

a {t) = armature current

i

i/(f) = field current

(a) Consider that the motor isa dc series motor so that a {t) = i(t); g(x) = 0,
i

k{v) = Bv(t), and R = 0. The voltage eU) is the input. Show that the system
is described by the following set of nonlinear state equations

x(t) = v(t)

(b) Consider that ia (t) = i

f (t) is the input and derive the state equations of the

system. g{x) — 0, k(v) = Bv(t).

(c) Consider that <f>(t) is the input, g(x) — 0, k(v) = Bv{t), and derive the state

equations of the system.

5.14. Figure P5-14 shows a gear-coupled mechanical system.
(a) Find the optimum gear ratio n such that the load acceleration, <X L , is

maximized.
(b) Repeat part (a) when the load drag torque TL is zero.
 t t

Chap. 5 Problems / 257

N,
=
1 \
Jl

J 1

V
«1

=
/
•4

V
N, EE

Figure P5-14.

5.15. (a) Write the torque equations of the system in Fig. P5-15 in the form
e + j-'k8 = o
where 8 is a 3 x 1 vector that contains all the displacement variables, 9
U
2 and 8 3
, J is the inertia matrix and K contains all the spring constants.
.

Determine J and K.

7 7
%
Figure P5-15.

(b) Show that the torque equations can be expressed as a set of state equations
of the form
i = Ax
where
o
A=r- --L!.i
_-j-'k o_ ;

(c) Consider the following set of parameters with consistent units: A", 1000, =
K2 = 3000, Ji = 1, J2 = 5, J3 = 2, and K =
3 1000. Find the matrix A.
5.16. Figure P5-16 shows the layout of the control of the unwind process of a cable
reel with the object of maintaining constant linear cable velocity. Control
is
established by measuring the cable velocity, comparing it with a reference
signal, and using the error to generate a control signal. A tachometer is used to
sense the cable velocity. To maintain a constant linear cable velocity, the
angular reel velocity 9 R must increase as the cable unwinds; that is, as the
effective radius of the reel decreases. Let

D= cable diameter = 0.1 ft

W = width of reel = 2 f
jR = effective radius of reel (empty reel) = 2 f
 t

258 / Mathematical Modeling of Physical Systems Chap. 5

w
v:

Motor 'OrIM^
Tachometer

+ + +
em Amplifier e ~ «/ +
- -

Figure P5-16.

Rf = effective radius of reel (full reel) = 4 f

R= effective radius of reel

= moment of inertia of reel = 18 J? 4 -
JR 200 ft-lb-sec 2
= linear speed of cable, ft/sec
vR
e, = output voltage of tachometer, volts

Tm (t) = motor torque, ft-lb

emit) = motor input voltage, volts
K = amplifier gain
Motor inertia and friction are negligible. The tachometer transfer function is

E,{s) 1

VR(s) 1 + 0.5s
and the motor transfer function is

Tm (s) 50
Em {s) s + 1

(a) Write an expression to describe the change of the radius of the reel R as a
function of 6 R .

(b) Between layers of the cable, R and JR are assumed to be constant, and the
system is considered linear. Draw a block diagram for the system and indi-
cate all the transfer functions. The input is e r and the output is vR .

(c) Derive the closed-loop transfer function VR (s)/E (s).

r
 6
Time-Domain Analysis
of Control Systems

6.1 Introduction

Since time
is used as an independent variable in most control systems, it is usu-

of interest to evaluate the time response of the systems. In the analysis

ally
problem, a reference input signal is applied to a system, and the performance of
the system is evaluated by studying the response in the time domain. For
instance, if the objective of the control system is to have the output variable
follow the input signal as closely as possible, it is necessary to compare the input
and the output as functions of time.
The time response of a control system is usually divided, into two parts:
the transient response and the steady-state response. If c(t) denotes a time
response, then, in general, it may be written

c(t) = c,(t) + c„(t) (6-1)

where
c,(t) = transient response
c n(t) = steady-state response

The definition of the steady state has not been entirely standardized. In circuit
analysis it is sometimes useful to define a steady-state variable as being a constant
with respect to time. In control systems applications, however, when a response
has reached its steady state it can still vary with time. In control systems the
steady-state response is simply the fixed response when time reaches infinity.
Therefore, a sine wave is considered as a steady-state response because its behav-

ior is fixed for any time interval, as when time approaches infinity. Similarly, if
a response is described by c(t) = t, it may be defined as a steady-state response.

259
260 / Time-Domain Analysis of Control Systems Chap. 6

Transient response is defined as the part of the response that goes to zero
as time becomes large. Therefore, c,(t) has the property of

lim c,(f) = (6-2)

It can also be stated that the steady-state response is that part of the response
which remains after the transient has died out.
All control systems exhibit transient phenomenon to some extent before a
steady state reached. Since inertia, mass, and inductance cannot be avoided
is

in physical systems, the responses cannot follow sudden changes in the input
instantaneously, and transients are usually observed.
The transient response of a control system is of importance, since it is part

of the dynamic behavior of the system; and the deviation between the response
and the input or the desired response, before the steady state is reached, must
be closely watched. The steady-state response, when compared with the input,
gives an indication of the final accuracy of the system. If the steady-state
response of the output does not agree with the steady state of the input exactly,
the system is said to have a steady-state error.

6.2 Typical Test Signals for Time Response of Control Systems

Unlike many electrical circuits and communication systems, the input excitations
to many practical control systems are not known ahead of time. In many cases,
the actual inputs of a control system may vary in random fashions with respect
to time. For and speed of the
instance, in a radar tracking system, the position
target to be tracked may an unpredictable
vary in manner, so that they cannot
be expressed deterministically by a mathematical expression. This poses a prob-
lem for the designer, since it is difficult to design the control system so that it
will perform satisfactorily to any input signal. For the purposes of analysis and
design, it is necessary to assume some basic types of input functions so that the
performance of a system can be evaluated with respect to these test signals. By
selecting these basic test signals properly, not only the mathematical treatment
of the problem is systematized, but the responses due to these inputs allow the
prediction of the system's performance to other more complex inputs. In a
design problem, performance criteria may be specified with respect to these test
signals so that a systemmay be designed to meet the criteria.
When the response of a linear time-invariant system is analyzed in the fre-
quency domain, a sinusoidal input with variable frequency is used. When the
input frequency is swept from zero to beyond the significant range of the system
characteristics, curves in terms of the amplitude ratio and phase between input
and output are drawn as functions of frequency. It is possible to predict the
time-domain behavior of the system from its frequency-domain characteristics.
To facilitate the time-domain analysis, the following deterministic test
signals are often used.

Step input function. The step input function represents an instantaneous

change in the reference input variable. For example, if the input is the angular
 Sec. 6.2 Typical Test Signals for Time Response of Control Systems / 261

position of a mechanical shaft, the step input represents the sudden rotation of
the shaft. The mathematical representation of a step function is

f >0
K0 = (6-3)
o t <0
where R is a constant. Or
r{t) = Ru s (t)
(6-4)

where u s (t) is the unit step function. The step function is not defined at t = 0.

The step function as a function of time is shown in Fig. 6-l(a).

(a) (b)

Fig. 6-1. Basic time-domain test signals for control systems, (a) Step
function input, r(t) = Ru s (t). (b) Ramp function input, r(t) — Rtu s {t).
(c) Parabolic function input, r(t) = Rt 2 u s (t).

Ramp input function. In the case of the ramp function, the signal is con-
sidered to have a constant change in value with respect to time. Mathematically,
a ramp function is represented by

\Rt t>0
K0 = L ."„ (6
-5
)
[0 ?<0
or simply
/(f) - Rtu s (0 (6-6)

The ramp function is shown in Fig. 6- 1(b). If the input variable is of the form of
the angular displacement of a shaft, the ramp input represents the constant-
speed rotation of the shaft.

Parabolic input function. The mathematical representation of a parabolic

input function is

, ,
\Rt 2 t >
262 / Time-Domain Analysis of Control Systems Chap. 6

or simply

<0 = Rt*u£t) (6-8)

The graphical representation of the parabolic function is shown in Fig. 6-l(c).

These test signals all have the common feature that they are simple to de-
scribe mathematically, and from the step function to the parabolic function they
become progressively faster with respect to time. The step function is very useful
as a test signal since its initial instantaneous jump in amplitude reveals a great
deal about a system's quickness to respond. Also, since the step function has, in
principle, a wide band of frequencies in its spectrum, as a result of the jump dis-
continuity, as a test signal it is equivalent to the application of numerous
sinusoidal signals with a wide range of frequencies.
The ramp function has the ability to test how the system would respond to
a signal that changes linearly with time. A parabolic function is one degree faster
than a ramp function. In practice, we seldom find it necessary to use a test signal
faster than a parabolic function. This is because, as we shall show later, to track
or follow a high-order input, the system is necessarily of high order, which may
mean that stability problems will be encountered.

6.3 Time-Domain Performance of Control Systems

Steady-State Response

In this section we shall discuss the typical criteria used for the measurement of
the time-domain performance of a control system. The time response of a control
system may be characterized by the transient response and the steady-state
response or, alternative, by a performance index that gives a qualitative
as an
measure on the time response as a whole. These criteria will be discussed in the
following.

Steady-State Error

It was mentioned earlier that the steady-state error is a measure of system

accuracy when a specific type of input is applied to a control system. In a phy-
sical system, because of friction and the nature of the particular system, the
steady state of the output response seldom agrees exactly with the reference
input. Therefore, steady-state errors in control systems are almost unavoidable,
and in a design problem one of the objectives is to keep the error to a minimum
or below a certain tolerable value. For instance, in a positional control system,
it is desirable to have the final position of the output be in exact correspondence

with the reference position. In a velocity-control system, the objective is to have

the output velocity be as close as possible to the reference value.
If the reference input r{t) and the controlled output c(t) are dimensionally
the same, for example, a voltage controlling a voltage, a position controlling a
position, and so on, and are at the same level or of the same order of magnitude,
the error signal is simply

e(t) = r(t) - c{t) (6-9)

Sec. 6.3 Time-Domain Performance of Control Systems / 263

However, sometimes it may be impossible or inconvenient to provide a

reference input that is at the same level or even of the same dimension as the
controlled variable. For instance, it may be necessary to use a low-voltage
source for the control of the output of a high-voltage power source; for a
velocity-control system it is more practical to use a voltage source or position
input to control the velocity of the output shaft. Under these conditions, the
error signal cannot be defined simply as the difference between the reference
input and the controlled output, and Eq. (6-9) becomes meaningless. The input
and the output signals must be of the same dimension and at the same level

before subtraction. Therefore, a nonunity element, H(s), is usually incorporated

in the feedback path, as shown in Fig. 6-2. The error of this nonunity-feedback

r(t)
/\ e(0
> G(s)
cit)

R(s) as)

bit)
H(s)
Bis)

Fig. 6-2. Nonunity feedback control system.

control system is defined as

e (0 = r{t) b(t) (6-10)

or
&{s) = R(s) - B(s) = R(s) - H(s)C(s) (6-1 1)

For example, if a 10-v reference is used to regulate a 100-v voltage supply, H

is a constant and is equal to 0.1. When the output voltage is exactly 100 v, the
error signal is

f(0= 10-0.1-100 = (6-12)

As another example let us consider that the system shown in Fig. 6-2
is a

velocity-control system in that the input r(t) is used as a reference to control the
output velocity of the system. Let c(t) denote the output displacement. Then,
we need a device such as a tachometer in the feedback path, so that H(s) = K,s.
Thus the error in velocity is defined as

e (0 = r{t) - b{t)

dc(t) (6-13)
r{t) - k;-
dt

The error becomes zero when the output velocity dc(i)\dt is equal to r(i)\K t
.

The steady-state error of a feedback control system is defined as the error

when time reaches infinity; that is,
steady-state error = e„ = lim e(t) (6-14)
264 / Time-Domain Analysis of Control Systems Chap. 6

With reference to Fig. 6-2, the Laplace-transformed error function is

8
^ = r+W)W) (6 ' 15)

By use of the final-value theorem, the steady-state error of the system is

e ss = lim e(t) = lim s&(s) (6-16)

where s&(s) is to have no poles that lie on the imaginary axis and in the right half
of the s-plane. Substituting Eq. (6-15) into Eq. (6-16), we have

e ss = lim , ,
*fffr). > (6-17)

which shows that the steady-state error depends on the reference input R(s)
and the loop transfer function G(s)H(s).
Let us first establish the type of control system by referring to the form of
G(s)H(s). In general, G(s)H(s) may be written

W
G(s)H(s) - KV+T s){l + T s)...{l + T m
i 2 s)
(6m
{t 1V)
{) ~ s'(l + T s)(l + T s)
a b . . . (1 + T s) n

where K and all the Ts are constants. The type of feedback control system refers
to the order of the pole of G(s)H(s) at s = 0. Therefore, the system that is de-
scribed by the G(s)H(s) o(Eq. (6-1 8) is of type/, wherey = 0, 1,2, ... The values .

of m, and the Ts are not important to the system type and do not affect the
n,
value of the steady-state error. For instance, a feedback control system with

G <*M*> = /+^(1°+1) (6 " 19)

is of type 1, since j = 1.

Now let us consider the effects of the types of inputs on the steady-state
error. We shall consider only the step, ramp, and parabolic inputs.

Steady-state error due to a step input. If the reference input to the control
system of Fig. 6-2 is a step input of magnitude R, the Laplace transform of the

input is R/s. Equation (6-17) becomes

sR(s) • R R_
"
i:

™ 1 + G(s)H(s) ~~ ™ 1 + G(s)H(s)
~~
1 + lim G(s)H(s)
,
K
fi ?m
'

For convenience we define

Kp = lim G(s)H(s) (6-21)

where K„ is the positional error constant. Then Eq. (6-20) is written

e
" = TTK, (6 " 22)

We see that for e ss to be zero, when the input is a step function, Kp must be
infinite. If G(s)H(s) is described by Eq. (6-18), we see that for K„ to be infinite,
j must be at least equal to unity; that is, G(s)H(s) must have at least one pure
integration. Therefore, we can summarize the steady-state error due to a step
Sec. 6.3 Time-Domain Performance of Control Systems / 265

input as follows:
R = constant
type system :
-" ~ + 1 K,
type 1 (or higher) system: e ss =
Steady-state error due to a ramp input. If the input to the control system
of Fig. 6-2 is

r(t) = Rtu,(t) (6-23)

where R is a constant, the Laplace transform of r(t) is

(6-24)

Substituting Eq. (6-24) into Eq. (6-17), we have

Mm R R
ss
V"o s + sG(s)H(s) ~ lim sG(s)H(s)
(6-25)

If we define

K =v lim sG(s)H(s) = velocity error constant (6-26)

Eq. (6-25) reads

(6-27)
K.

which is the steady-state error when the input is a ramp function. A typical e s
due to a ramp input is shown in Fig. 6-3.

c(t) i

Reference input e ss = R/Kv

r(l) = Rtu s (t)

Fig. 6-3. Typical steady-state error due to a ramp input.

Equation (6-27) shows that for e ss to be zero when the input is a ramp
function, Kv must be infinite. Using Eq. (6-18) and (6-26),
K =
v lim sG(s)H(s)
s-0
= lim
s-0 S
~
J
j= 0,1,2,. (6-28)

Therefore, in order for K v to be infinite, j must be at least equal to two, or the

system must be of type 2 or higher. The following conclusions may be stated
 :

266 / Time-Domain Analysis of Control Systems Chap. 6

with regard to the steady-state error of a system with ramp input:

type system e ss = oo

type 1 system: e ss = -£- = constant

type 2 (or higher) system: e ss =

Steady-state error due to a parabolic input. If the input is described by

r(t) = \u
2
s {t) (6-29)

the Laplace transform of r(t) is

R(s) = -J (6-30)

The steady-state error of the system of Fig. 6-2 is

R
" (6-31)
lim s 2 G{s)H(s)

Defining the acceleration error constant as

K =
a lim s 2 G(s)H(s) (6-32)

the steady-state error is

e ss = (6-33)
£
The following conclusions can be made with regard to the steady-state error of
a system with parabolic input:
type system: e ss = oo

type 1 system: e ss = oo

type 2 system: e„ = — = constant

type 3 (or higher) system : e ss =
As a summary of the error analysis, the relations among the error constants,
the types of the system, and the input types are tabulated in Table 6-1. The
transfer function of Eq. (6-18) is used as a reference.

Table 6-1 Summary of the Steady-State Errors Due to Step, Ramp,

and Parabolic Inputs

Step Ramp Parabolic

Type of Input, Input, Input,
System _ R _ R ^ SS
_ R
js v- &ss ^**
J
-is
Ap Ay Aa
\ if \

1 ~r Jy p
is
A;, A
T^~
a

"
K e„ =
1 +K
1 oo K e„ = e ss = ^
2 oo oo K e ss = e ss = e ss =
3 oo oo oo e ss = e 3S = c„ =
Sec. 6.3 Time-Domain Performance of Control Systems / 267

should be noted that the position, velocity, and acceleration error con-
It

stants are significant in the error analysis only when the input signal is a step
function, a ramp function, and a parabolic function, respectively.
It should be noted further that the steady-state error analysis in this section
isconducted by applying the final-value theorem to the error function, which is
defined as the difference between the actual output and the desired output signal.
In certain cases the error signal may be defined as the difference between the
output and the reference input, whether or not the feedback element is unity.
For instance, one may define the error signal for the system of Fig. 6-2 as

e(t) = lit) - c(t) (6-34)

Then
_ 1 + G(s)[H(s) - l]
R() (6 . 35)
()

and
+ G(s)[H(s) - l]
e -lims l R(s) (6-36)

It should be kept in mind that since the steady-state error analysis discussed
here relies on the use of the final-value theorem, it is important to first check to
see if sE(s) has any poles on the ja> axis or in the right half of the j-plane.
One is, of course, that they do
of the disadvantages of the error constants
not give information on the steady-state error when inputs are other than the
three basic types mentioned. Another difficulty is that when the steady-state
error is a function of time, the error constants give only an answer of infinity,
and do not provide any information on how the error varies with time. We shall
present the error series in the following section, which gives a more general repre-
sentation of the steady-state error.

Error Series

In this section, the error-constant concept is generalized to include inputs

of almost any arbitrary function of time. We start with the transformed error
function of Eq. (6-15),

S ^> = r+Mm (6 " 37)

or of Eq. (6-35), as the case may be.

Using the principle of the convolution integral as discussed in Section 3.3,

the error signal e(t) may be written

€ (r) = w e (t)r{t - t) dx (6-38)

J" _

where we (r) is the inverse Laplace transform of

W^= +G \s)H(s) l
< W9 >

which is known as the error transfer function.

If the first n derivatives of r(t) exist for all values of /, the function r(t — t)
268 / Time-Domain Analysis of Control Systems Chap. 6

can be expanded into a Taylor series; that is,

r(t - T) = r(t) - xr(t) + |if(/) - ^r(t) + ... (6-40)

where r(t) first derivative of r(t) with respect to time.

represents the
Since considered to be zero for negative time, the limit of the con-
r(t) is

volution integral of Eq. (6-38) may be taken from to t. Substituting Eq. (6-40)
into Eq. (6-38), we have

= w.(T)[r(0 - *H?) + ff (0 - yj'C) + • . rfr

J .]
(6-41)

= lit) w e (r) dx - r(t) xw e (x) dx + r(t) f ^w.(i) dx

J J

As before, the steady-state error is obtained by taking the limit of e(t) as t

approaches infinity; thus

e„ = lim e(t) = lim e,(f) (6-42)

where e£t) denotes the steady-state part of e (t) and is given by

e.(t) = r,(t)\ w e (x)dx-f s (t)\ xw e (x)dx + r s (t)\ ~^\{x)dx

(6-43)

- ^(0 ^w,(x)dz + •••

and r,(t) denotes the steady-state part of r(t).

Let us define

C = I
w e {x) dx
J

Ci = — xw£x) dx
J

C2 = x 2 w e (x) dx (6-44)
J

C„ = (-l)TT"W
J
(TVT e

Equation (6-42) is written

e,(0 =C r,(t) + C,r,(0 + ^(f) + . . . + ^r«(/) + . . . (6-45)

which is called the error series, and the coefficients, C C C2

, ]; , . . . , C„ are
defined as the generalized error coefficients, or simply as the error coefficients.
The error coefficients may be readily evaluated directly from the error
transfer function, W,(s). Since W,(s) and w e (x) are related through the Laplace
transform, we have
W. is) = P w,(t>-" rft (6-46)
J
Sec. 6.3 Time-Domain Performance of Control Systems / 269

Taking the limit on both sides of Eq. (6-46) as s approaches zero, we have

lim
s^o
W e (s)
= lim
i-o f w (T)e"" dx
e
•> o (6-47)
=C
The derivative of W£s) of Eq. (6-46) with respect to s gives

—-f^ = —
ds
Tw e (r)e " di
Jo (6-48)
= de'"
from which we get

C, = lim 4EM- (6-49)

.-o ds

The rest of the error coefficients are obtained in a similar fashion by taking suc-
cessive differentiation of Eq. (6-46) with respect to s. Therefore,

Cz = lim
,^o
^M ds 2
(6-50)

C = 3 lim«£) 3
(6-51)
v
ds
*^o

The following examples illustrate the general application of the error series
and its advantages over the error constants.

Example 6-1 In this illustrative example the steady-state error of a feedback control
system will be evaluated by use of the error series and the error coef-
ficients. Consider a unity feedback control system with the open-loop
transfer function given as

G(s) = j^-j (6-53)

Since the system is of type 0, the error constants are p K

= K, Kv = 0, and a = 0. K
Thus the steady-state errors of the system due to the three basic types of inputs are as
follows

unit step input, u,(t): e ss =

1 R
unit ramp input, tu s (t) e ss = oo

unit parabolic input, t

2
u s (t): e ss = oo

Notice that when the input is either a ramp or a parabolic function, the steady-
state error is infinite in magnitude, since it apparently increases with time. It is apparent
that the error constants fail to indicate the exact manner in which the steady-state
function increases with time. Therefore, ordinarily, if the steady-state response of this

system due to a ramp or parabolic input is desired, the differential equation of the sys-
tem must be solved. We now show that the steady-state response of the system can
actually be determined from the error series.
 . ^

270 / Time-Domain Analysis of Control Systems Chap. 6

Using Eq. (6-39), we have for this system

^> = r+W) = rFXTT (6 " 54)

The error coefficients are evaluated as

C = lim W.(s)
j^.0
= Kjr-x-f
+ I
(6
- 55 >

c
<=il-T = (rTTF (6 " 56)

C = ™ ~d^- = + *)3
>
l (6 " 57)
(1

Although higher-order coefficients can be obtained, they will become less significant

as their values will be increasingly smaller. The error series is written

<?.(0 = YTK r*W + (1 +K)* W+ (1 +^) 3

>
"''
(?) + • •
• (6 " 58)

Now let us consider the three basic types of inputs.

1 When the input signal is a unit step function, r s (t) =u s (t), and all deriva-
tives of r,(f) are zero. The error series gives

*,(0 = — (6-59)

which agrees with the result given by the error-constant method.

2. When the input signal a unit ramp function, r s (t) = tu s (t), r s {t) = u s (t),
is

and all higher-order derivatives of r s (t) are zero. Therefore, the error series
is

«- (0 = I TTT-F
_1+ K t' + TTZT^I
(1 +KY
«-(') (6
- 60)

which indicates that the steady-state error increases linearly with time.
The error-constant method simply yields the result that the steady-state
error is infinite but fails to give details of the time dependence.
3. For a parabolic input, r£t) = (t
2
/2)u s (t), r s (t) = tu s (t), r s (t) = u s {t), and
all higher derivatives are zero. The error series becomes

*.(0 = [y^k T + (TTkT2 - inhcy] «' w ' (6 " 61)

In this case the error increases in magnitude as the second power of /.

4. Consider that the input signal is represented by a polynomial of t and an
exponential term,

r(f) = [«o +a + t t ^+ e-°«~jujf) (6-62)

where a , a it a 2 and a 3 are constants. Then,

,

r,(f) = [a + <M + ^]«,(0 (6-63)

fit) = («i + a 2 t)u s (t) (6-64)

rXt) = a 2 u (t) s (6-65)

In this case the error series becomes

K K
e,0) = 1
1

+ K n(t) + (1 +K) i
f M- (i +^)3 ^(f) (6
" 66
>
 —

Sec. 6.4 Time-Domain Performance of Control Systems / 271

Example 6-2 In this example we shall consider a situation in which the error con-
stant is totally inadequate in providing a solution to the steady-state
error. Let us consider that the input to the system described in Example
6-1 is a sinusoid,
r(t) = sin (D t (6-67)

where (D = 2. Then
r,(i) = sin (D t

rjit) = (D cos (D t
(6-68)
r,(t) = -Oil sin (D t

r (t) = -G>1 cos (D

s t

The error series can be written

e,(t) = [c - 2y C0§ + 4T »o
- • •
•]
sin °V + [Ci©o - -yfcoS + • •
•]
cos (D t

(6-69)

Because of the sinusoidal input, the error series is now an infinite series. The con-
vergence of the series is important in arriving at a meaningful answer to the steady-

state error. It is clear that the convergence of the error series depends on the value of
(D and K. Let us assign the value of K to be 100. Then
C =
YTH = 00099
Cl = (1 +K? = 00098
K -
c* = ~ (i +Ky = 0000194
6K
C = (i
3 „ +"
„, =
xy 5.65 x 10" 8

Thus, using only the first four error coefficients, Eq. (6-69) becomes

e£t) =i ro.0099
L
+ a0029 194 -4lJ sin It + 0.0196 cos It _
(6 70 )
= 0.01029 sin It + 0.0196 cos It
or
e,(0 ~ 0.02215 sin {It + 62.3°) (6-71)

Therefore, the steady-state error in this case is also a sinusoid, as given by Eq. (6-71).

6.4 Time-Domain Performance of Control Systems

Transient Response

The transient portion of the time response is that part which goes to zero as
time becomes large. Of course, the transient response has significance only when
a stable system is referred to, since for an unstable system the response does not
diminish and is out of control.
272 / Time-Domain Analysis of Control Systems Chap. 6

The transient performance of a control system is usually characterized by

the use of a unit step input. Typical performance criteria thai are used to charac-
terize the transient response to a unit step input include overshoot, delay time,
rise time, and settling time. Figure 6-4 illustrates a typical unit step response of

Maximum
overshoot

Fig. 6-4. Typical unit step response of a control system.

a linear control system. The above-mentioned criteria are defined with respect to
the step response:

1. Maximum overshoot. The maximum overshoot is defined as the

largest deviation of the output over the step input during the tran-
sient state. The amount of maximum overshoot is also used as a
measure of the relative stability of the system. The maximum over-
shoot is often represented as a percentage of the final value of the
step response; that is,

per cent maximum overshoot = maximum overshoot X 100%

final value
(6-72)

2. Delay time. The delay time Td is defined as the time required for the
step response to reach 50 per cent of its final value.
3. Rise time. The rise time T r is defined as the time required for the
from 10 per cent
step response to rise to 90 per cent of its final value.
Sometimes an alternative measure is to represent the rise time as a
 Sec. 6.5 Transient Response of a Second-Order System / 273

reciprocal of the slope of the step response at the instant that the
response is equal to 50 per cent of its final value.
4. Settling time. The settling time T s
is defined as the time required for
the step response to decrease and stay within a specified percentage
of its final value. A frequently used figure is 5 per cent.

The four quantities defined above give a direct measure of the transient
characteristics of the step response. These quantities are relatively easy to mea-
sure when a step response is already plotted. However, analytically these quan-
tities are difficult to determine except for the simple cases.

Performance Index

Since the general design objective of a control system is to have a small

overshoot, fast rise time, short delay time, short settling time, and low steady-
state error,it is advantageous to use a performance index that gives a measure

of the overall quality of the response. Let us define the input signal of a system
as r(t) and the output as c(t). The difference between the input and the output
is defined as the error signal, as in Eq. (6-9). Sometimes r(t) is referred to as
the desired output.
In trying to minimize the error signal, time integrals of functions of the error
signal may be used as performance indices. For example, the simplest integral
function of the error is

dt (6-73)
J

where / is used to designate performance index. It is easy to see that Eq. (6-73)

isnot a practical performance index, since minimizing it is equivalent to mini-

mizing the area under e{t), and an oscillatory signal would yield a zero area and
thus a zero /. Some of the practical integral performance indices are

\"\e{f)\dt
J o
[°te{t)dt
Jo
f e\t)dt
JO
and there are many others. The subject of the design of control systems using
performance indices is covered in Chapter 11.

6.5 Transient Response of a Second-Order System

Although true second-order control systems are rare in practice, their analysis
generally helps to form a basis for the understanding of design and analysis
techniques.
Consider that a second-order feedback control system is represented by the
state diagram of Fig. 6-5. The state equations are written
"x (0i r o i
1
= "*i(0"
+ r(t) (6-74)
Ji(t)J \_-<ol -2Cco„,
where £ and co„ are constants.
274 / Time-Domain Analysis of Control Systems Chap. 6

x 2 (0+)/s x,(0+)/s
Q

Fig. 6-5. State diagram of a second-order feedback control system.

The output equation is

c(0 = colx^t) (6-75)

Applying the gain formula to the state diagram of Fig. 6-5, the state transi-
tion equations are written

's + 2Cco„ 1 *i(0+)" 1

+ R(s) (6-76)
L -col s_\ix 2 (0+)
where
A= s2 + 2£co„s + col (6-77)

The inverse Laplace transform of Eq. (6-76) is carried out with the help of
the Laplace transform table. For a unit step function input, we have

*l(0 sin (co*s/l —C 2

1 + y/) —
co n
sin awT- £ 2
1 *i(0+)
Vi -C 2

x 2 (t) J — co„ sin g>„V — 1 £

2
1 sin (eo„V 1 — £
2
1 + $) .* 2 (0+)J

1
(6-78)
e^-'sinKVl -C 2 '-
col I yi-c :
<t>\\

+ r>0
1 fo "'
— =e" sin co,, VI -C 2 '
w„*J\ C
where

y/ = tan ,
yi-c 2
(6-79)
c

= tan
-,yi-c 2
(6-80)

Although Eq. (6-78) gives the complete solution of the state variables in
terms of the initial states and the unit step input, it is a rather formidable-looking
expression, especially in view of the fact that the system is only of the second
order. However, the analysis of control systems does not rely completely on the
evaluation of the complete state and output responses. The development of
linear control theory allows the study of control system performance by use of
Sec. 6.5 Transient Response of a Second-Order System / 275

the transfer function and the characteristic equation. We shall show that a great
deal can be learned about the system's behavior by studying the location of the
roots of the characteristic equation.
The closed-loop transfer function of the system is determined from Fig. 6-5.

C(£)
= col
(6
K
. 81
)
}
R(s) s
2
+ 2£gv + col
The characteristic equation of the system is obtained by setting Eq. (6-77) to
zero ; that is,

A= s
2
+ 2Cco„s + col = (6-82)

For a unit step function input, R(s) I Is, the output response of the system =
isdetermined by taking the inverse Laplace transform of

C(s) = f' (6-83)

s{s
2
+ 2£co„s + col)
Or, c(t) is determined by use of Eqs. (6-75) and (6-78) with zero initial states

-CoJnf 2
c(0 = + co„*/l — 2 - tan -,-v/l-C > (6-84)
7r=c sSin
1 * t
C
-c J
It is interesting to study the relationship between the roots of the characteristic
equation and the behavior of the step response c(t). The two roots of Eq. (6-82)
are

su s2 = -fco„ ±jco„ a/1 -C

(0-83)
= —a.±jco
The physical significance of the constants £, co n , a, and co is now described
as follows:
As seen from Eq. (6-85), a =£a>„, and a appears as the constant that is
multiplied to t in the exponential term of Eq. (6-84). Therefore, a controls the
and decay of the time response. In other words, a controls the
rate of rise
"damping" of the system and is called the damping constant or the damping
factor. The inverse of a, 1/a, is proportional to the time constant of the system.
When the two roots of the characteristic equation are real and identical we
call the system critically damped. From damping Eq. (6-85) we see that critical
occurs when f = 1. Under this condition the damping factor is simply a = co„.
Therefore, we can regard £ as the damping ratio, which is the ratio between the
actual damping factor and the damping factor when the damping is critical.
co„ is defined as the natural undamped frequency. As seen from Eq. (6-85),
when the damping is zero, £ =
0, the roots of the characteristic equation are
imaginary, and Eq. (6-84) shows that the step response is purely sinusoidal.
Therefore, con corresponds to the frequency of the undamped sinusoid.
Equation (6-85) shows that

co = aw^ - C
2
(6-86)

However, since unless C = 0, the response of Eq. (6-84) is not a periodic func-
tion. Therefore, strictly, co is not a frequency. For the purpose of reference co is
sometimes defined as conditional frequency.
276 / Time-Domain Analysis of Control Systems Chap. 6

j
l'u

Root s-plane

)
;,

/
X w„ « = «» >/!--r 2
e \
i \
N
/ "

•*-a = $u„

X
Root

Fig. 6-6. Relationship between the characteristic equation roots of a

second-order system and a, C, co„, and to.

Figure 6-6 illustrates the relationship between the location of the charac-
teristic equation roots and a, C> cu„, and ca. For the complex-conjugate roots
shown, a>„ is the radial distance from the roots to the origin of the s-plane. The
damping factor a is the real part of the roots; the conditional frequency is the
imaginary part of the roots, and the damping ratio £ is equal to the cosine of
the angle between the radial line to the roots and the negative real axis; that is,

cos 9 (6-87)

Figure 6-7 shows the constant-cw„ loci, the constant-^ loci, the constant-a
loci, and the constant-cu loci. Note that the left-half of the j-plane corre-
sponds to positive damping (i.e., the damping factor or ratio is positive), and the
right-half of the j-plane corresponds to negative damping. The imaginary axis
corresponds to zero damping (a = 0, £ = 0). As shown by Eq. (6-84), when
the damping is positive, the step response will settle to its constant final value
because of the negative exponent of e~ Ca*. Negative damping will correspond
to a response that grows without bound, and zero damping gives rise to a sus-
tained sinusoidal oscillation. These last two cases are defined as unstable for
linear systems. Therefore, we have demonstrated that the location of the charac-
teristic equation roots plays a great part in the dynamic behavior of the transient
response of the system.
The effect of the characteristic-equation roots on the damping of the second-
 :

Sec. 6.5 Transient Response of a Second-Order System / 277

s-plane
\ ' /to
s-plane

^ \\ f = o

XyA J
Positive >N Negative
damping / damping

1 i A
Positive /
/]
1 Negative
damping / J damping

»' f2>fl
1 r = o

(b)

/CO
/OJ
,
s-plane s-plane

Positive Negative
co 2
damping damping

00,

-t- a
-«2 -Oij -«3
-to,

Positive Negative -co 2

damping damping

a2 >a >0
l
<*
3

(c) (d)

Fig. 6-7. (a) Constant natural undamped frequency loci, (b) Constant
damping ratio loci, (c) Constant damping factor loci, (d) Constant condi-
tional frequency loci.

order system is further illustrated by Figs. 6-8 and 6-9. In Fig. 6-8, co„ is held

constant while the damping ratio C is varied from oo to +oo. The following —
classification of the system dynamics with respect to the value of £ is given

<C< 1 : s u s2 = -Ccon ± j(O *J\~

n £
2
underdamped case
critically damped case
C>i overdamped case
C = o Si, s 2 = ±/cu„ undamped case
C<o s t ,
s2 = — Ca>„ ± jco„^/l — C
2
negatively damped case
 . . /oo

x-plane
r=o
_^
<?<!/' X 0>f>-l

r>i ?=
5l
*-
t- f<-I I

/
f-

yo>r>-i

r =o

Locus of roots of Eq. (6-82) when a>„ is held constant while the
Fig. 6-8.
ratio is varied from -°o to
damping +oo.

,. /"
c(,t)

s-plane

-X *-

f>l

/OJ

x-plane

f=l

x-plane /«
c(/)
X

o<r<i

*-f

in the x-plane.
Fig. 6-9. Response comparison for various root locations

278
 Sec. 6.5
Transient Response of a Second-Order
System / 279

/co c(t)

s-plane

>~ct 1

f =

, . /CO

s-plane
X

0>f>-l

/CO

s-plane

r<-i

Fig. 6-9 (Cont.).

Figure 6-9 illustrates typical step responses

that correspond to the various root
locations.
In practical applications only stable
systems are of interest. Therefore
the cases when f is positive are of particular interest. In
Fig. 6-10 is plotted the
variation of the unit step response described
by Eq. (6-84) as a function of the
normalized time co„t, for various values of
the damping ratio
f. It is seen that
the response becomes more oscillatory
as £ decreases in value When r
there is no overshoot in the step response;
i >
that is, the output never exceedsThe
value of the reference input.
The exact relation between the damping ratio and
the amount of overshoot
can be obtained by taking the derivative
of Eq. (6-84) and setting the result to
zero. Thus
dc(t) -Zcon t
tco»e
dt
sin (cot — t4)

-CcOnt (6-88)
~
/i _^ CQa ^/1 ~ 2
£ cos (a>t — 4>) t>0
280 / Time-Domain Analysis of Control Systems Chap. 6

Fig. 6-10. Transient response of a second-order system to a unit step func-

tion input.

where
(j> = tan
-, yi-c 2
(6-89)

Equation (6-88) is simplified to

dc(t) co„
e
-&»,<
si n q^ ^i _ £2, i > o (6-90)
Jl -
"
2
dt C
Therefore, setting Eq. (6-90) to zero, we have t = oo and

c»„Vl-C 2 ' = ™ n = 0,1,2,... (6-91)

 Sec. 6.5 Transient Response of a Second-Order Systems / 281

or
nn

The first maximum value of the step response c(t) occurs at n = 1. Therefore,
the time at which the maximum overshoot occurs is given by

•max /, .., (6-93)

In general, for all odd values of n, that is, n

1 3, 5, Eq. (6-92) gives= , . . . ,

the times at which the overshoots occur. For all

even values of n, Eq. (6-92)
gives the times at which the undershoots occur, as
shown in Fig. 6-11. It is
interesting to note that, although the maxima and
the minima of the response
occur at periodic intervals, the response is a damped sinusoid
and is not a
periodic function.

c(t)

v7~p vT^p" vT^T1 vT^F

Fig. 6-11. Step response illustrating that the
maxima and minima occur at
periodic intervals.

The magnitudes of the overshoots and the undershoots can be

obtained by
substituting Eq. (6-92) into Eq. (6-84). Thus

<\0 Im.x or mm = 1 H r
Vi,

-C 2
==- sin ( nn — tan" 1
^ZI^-)[
il
"=1,2,3,...
(6-94)
or

C(OI««orml. = + 1 (-I)""'*— «"f=F

(6-95)
The maximum overshoot is obtained by letting n = 1 in Eq. (6-95). Therefore,

maximum overshoot = c max — 1 = e -'C/-/T=c

r
(6-96)
and
per cent maximum overshoot = lOOer*^ 1 ^ 5
(6-97)
282 / Time-Domain Analysis of Control Systems Chap. 6

0.4 0.6 0.8 1.0 1.2

Damping ratio f

Fig. 6-12. Per cent overshoot as a function of damping ratio for the step
response of a second-order system.

Note that for the second-order system, the maximum overshoot of the step
response is only a function of the damping ratio. The relationship between the
per cent maximum overshoot and damping ratio for the second-order system
is shown in Fig. 6-12.
From Eqs. (6-93) and (6-94) it is seen that for the second-order system under
consideration, the maximum overshoot and the time at which it occurs are all

exactly expressed in terms of £ For the delay time, rise time, and settling
and co„.

time, however, the relationships are not so simple. It would be difficult to deter-
mine the exact expressions for these quantities. For instance, for the delay time,
we would have to set c(t) = 0.5 in Eq. (6-84) and solve for /. An easier way
would be to plot co„td versus £ as shown in Fig. 6-13. Then, over the range of
< £ < 1 .0 it is possible to approximate the curve by a straight line,
^~l+0.7£ (6-98)

Thus the delay time is

U — + 0-7C
1
(6-99)
co„

For a wider range of £, a second-order equation should be used. Then

,
id
_1 + 0.6£ + 0.15£
=
2
(6-100)
CO„

For the rise time tr , which is the time for the step response to reach from
10 per cent to 90 per cent of its final value, the exact values can again be obtained
directly from the responses of Fig. 6-10. The plot of co„t r versus £ is shown in
Fig. 6-14. In this case the rise time versus £ relation can again be approximated
by a straight line over a limited range of £. Therefore,

0.8 + 2.5£
o<£< i (6-101)
COn

A better approximation may be obtained by using a second-order equation;

 Sec. 6.5 Transient Response of a Second-Order System
/ 283

«„^ = l+0.7f

H'rf

<%) _ <
R (s) s2+2$u„s + w n2
0.5

0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6

Fig. 6-13. Normalized delay time versus f for a second-order control

system.

5.0 "i r t r

4.0

2.0

-I 1 I L J_
0.2 0.4 0.6 0.8 1.0 1.2

Fig. 6-14. Normalized rise time versus C for a second-order system.

then

,_
~ + l l-K+lAC 2
(6-102)
co„

From the definition of settling time, it is clear that the expression for the
settling time is the most difficult to determine. However, we can obtain an approx-
imation for the case of <f< 1 by using the envelope of the damped sinu-
soid, as shown in Fig. 6-15.
284 / Time- Domain Analysis of Control Systems Chap. 6

Fig. 6-15. Approximation of settling time using the envelope of the decay-
ing step response of a second-order system (0 < £ < 1).

From the figure it is clear that the same result is obtained with the approx-
imation whether the upper envelope or the lower envelope is used. Therefore,
-CcOnl.
c(t) = 1 1.05 (6-103)

Solving for co„t s from the last equation, we have

1
<oj,= -4-ln[0.05yi-C»] (6-104)

For small values of f, Eq. (6-104) is simplified to

CO n t s = -y (6-105)

or

s
- c<»„ o<c<i (6-106)

Now reviewing the relationships for the delay time, rise time, and settling
time, seen that small values of f would yield short rise time and short delay
it is

time. However, a fast settling time requires a large value for £. Therefore, a
compromise in the value of £ should be made when all these criteria are to be
satisfactorily met in a design problem. Together with the consideration on
maximum overshoot, a generally accepted range of damping ratio for satis-
factory all-around performance is between 0.5 and 0.8.

6.6 Time Response of a Positional Control System

In this sectionwe shall study the time-domain performance of a control system

whose objective is to control the position of a load that has viscous friction and
inertia. The schematic diagram of the system is shown in Fig. 6-16.
 Sec. 6.6
Time Response of a Positional Control
System / 285

dc motor

dc
amplifier = constant
; if

*
Error detector

t ^Q09

Fig. 6-16. Direct current

positional control system.

A set of potentiometers
form the error detector with
error detector sends a signal to
sensitivity The K
the dc amplifier which is
proportionafto *e
dtfference between the angular
positions of the reference input
shaft anS he
C UtPUt f he dC
° ° ""***" is USed to contro1 the armature of
dcmot
a dc TH current
motor. The ,!
the m
field of the dc motor is held
constant
The parameters of the system are given
as follows:

Sensitivity of error detector

K = 1/57.3 volt/deg =
s 1 volt/rad
Gain of dc amplifier A (variable)
Resistance of armature of motor
* = 5Q
Inductance of armature of motor L = negligible
a
Inertia of rotor of motor Jm = 10" lb-ft-sec*
3

Friction of motor shaft

Bm = negligible
Friction of load shaft
BL = 0.1 lb-ft-sec
Inertia of load
/z,= 0.11b-ft-sec 2

Gear ratio
« = NJN 1 =
Torque constant of motor TV
K = 0.5 lb-ft/amp
t

The first step in the analysis is to write the equations for

the system This
s cause-

1. Error detector:

9eQ) = e (t) - exo

r
(6-107)
e(t) = KftJit) (6-108)
2. DC amplifier:
ea {t) = Ae(t) (6-109)
 Time-Domain Analysis of Control Systems Chap. 6
286 /

3. Armature-controlled dc motor:

LjMl = - RJ {t) + a e a (t) - e b (t) (6-1 10)

e„{t) = K co m {t)b
(6-111)

where K b is the back emf constant of the motor

TJf) = Km (6
- 112 )

jj-^- = -Bme co m + (t) TJt) (6-113)

where Jme and Bme are the equivalent inertia and viscous frictional
coefficients seen by the motor, respectively, and

Jme= Jm + = lO' + 0.01(0.1) = 2

n 2 JL 3
X 10- 3 lb-ft-sec 2 (6-114)

Bme = Bm -\- n*BL = 10" lb-ft-sec 3 (6-1 1

5)

4. Output:

^ 0c(t)
=
=
co m {t)

n8Jt)
(6-116)

(6-117)

The value of the back emf constant, Kb is not given originally, but a definite
,

relationship exists between Kb and K In the British unit system, K, is given in

t
.

lb-ft/amp and the units of the back emf constant are volts/rad/sec. With these
units, Kb is related to K, through a constant ratio. The mechanical power
developed in the motor armature is {see Sec. 5.7)

Pit) = e b {t)i a {t) watts

(6-118)
= ^e b (t)i£t)
hp

Substituting Eqs. (6-111) and (6-112) into Eq. (6-118), we have

P(f) = ^TJfyoJt) hp (6-119)

Also, it is known that

Pit) = -^Tjfajt) hp (6-120)

Therefore, equating Eq. (6-119) to Eq. (6-120) gives

A, = 2£a» = 0.737A. (6-121)

/40
or
K = b 1.36*; (6-122)

Thus, given K, to be 0.5 lb-ft/amp, K„ is found to be 0.68 volt/rad/sec.

Using Eqs. (6-107) through (6-116), the state equations of the system are
written in the matrix form as follows:
 Sec. 6.6
Time Response of a Positional Control System / 287

\
dia {ty Kb nAK, [AK.1
dt r us)
dco m (t)
= Bme
dt J me •'me
co m (t) + 0,(0 (6-123)
dBJf)
1
L dt J [0„(O
The output equation is given by Eq. (6-117). The state diagram
of the system
isdrawn as shown in Fig. 6-17.

9<a (0+)A
9 B m (0 +)/s

"AK.IL.

Fig. 6-17. State diagram of the positional control system in Fig. 6-16.

When the initial states are assumed to be zero, the closed-loop transfer
function of the system is obtained as

0,(0 K AK n
s t

°M *A.4l + V>U + W) + K K b lS + K.AK,n (6 " 124)

where

?. = -£ = negligible

^ me ~n — ^ SeC

The error between the reference input and the output is denned as

0.(0 = 0,(0-0,(0 (6-125)

The open-loop transfer function of the system is obtained by use of Eq. (6-124).

G(s) = JsW = K,AK n t

°M K ^meS{l + t.j)(1 + r me s) +KK b lS

(6-126)

The state
transition equations of the system
can be obtained from the state
diagram by use of the gain formula,
in the usual fashion. However,
the main
objective of this problem is to demonstrate
the behavior of the time response of
288 / Time-Domain Analysis of Control Systems Chap. 6

the positional control system with respect to the system parameters, and it is

sufficient to assume that all the initial states of the system are zero.
Since La is negligible, t„ = 0, and the closed-loop transfer function in Eq.

(6-124) is simplified to

OJs) K,AK,n
(&l27)
d,{s) R.Jm y + (K K t l + R aB m .)s + K,AK n l
v

Equation (6-127) of the second order; thus it can be written in the standard
is

form of Eq. (6-81). The natural undamped frequency of the system is

i V KR.Jm
AK,n
w =± "
s

.
(6-128)
v

The damping ratio is

r
^
_ K K, + R„B me _ K K, + R B m
b b a ,
(6-129)
2R Jme(o„ 2jK AK,RJme n
a s

When the values of the system parameters are substituted into Eq. (6-127), the
closed-loop transfer function of the system becomes

JA -
G 5A (6-130)
l° '
VJs)-s 1 + 34.5s + 5A
Suppose that the gain of the dc amplifier is arbitrarily set at 200. The natural
undamped frequency and the damping ratio are, respectively,

©„= ±31.6rad/sec (6-131)

C = 0.546 (6-132)

The characteristic equation of the system is

s
2
+ 34.5s + 1000 = (6-133)

whose roots are

s u s2 = -17.25 ±j26.5 (6-134)

Time Response to a Step Input

Consider that the reference input is a unit step function, 9,{t) = u s (t) rad;
then the output of the system, under zero initial state condition, is

or
^=
rA =-
£'
W
4-
+ 3^7+
9 <,-o.5 4 6<»„r
1000
1000) J

^ n rn.R17m tan" 1
(6 " ,35)

fl 1 1 t -i- ir. -i- 1.53)

(6-136)
= 1 + 1.2e-
17 "' sin (26.4? + 236.8°)

The output response of the system is plotted in Fig. 6-18 as a function of the
normalized time co„t.

It is interesting to study the effects on the time response when the value of
the gainA is varied. Note that in Eq. (6-128) and in Eq. (6-129), an increase in
the gain A increases the natural undamped frequency oj>„ but decreases the
damping ratio £. For A = 1500, £ = 0.2, and ©„ = 86.2 rad/sec. The output
response to a unit step displacement input for A = 1500 is plotted as shown in
 Sec. 6.6
Time Response of a Positional Control System 289
/

Bc {t)

Fig. 6-18. Time response of the positional control system in Fig. 6-16 when
the input is a unit step displacement.

Fig. 6-18. In this case the overshoot

is very high, although the rise
time and
delay time are apparently reduced. The
settling time is also shortened by the
increase in A, although the responses in Fig.
6-18 do not show this because the
time axis is normalized by «».. In fact, in Fig. 6-18
the step response for A =
1500 seems to take a longer time to reach the steady
state than that for A 200 =
This is distorted by the normalized time axis;
for A 1500, co =
86 2 rad/sec =
as compared to a>„ =
31.6 rad/sec for A 200. =
When A is set at a relatively
low value, 13.5, the damping ratio and the
natural undamped frequency are f 2.1 and .= m =
8.22 rad/sec, respectively
fcince
£ is greater than one, the step response corresponding to A 13 5 is over- =
damped. Table 6-2 gives the comparison of the time
responses of the system for
the three different values of A used.

Table 6-2 Comparison of Transient Response of a Second-Order

Positional
Control System When the Gain Varies

Damping Maximum
Gain A Ratio C (On Overshoot Tr
Tt Ts 'max

13.5 2.1 8.22 0.325 1.02 1.51

200 0.546 31.6 0.13 0.043 0.064 0.174 0.119
1500 0.2 86.6 0.52 0.013 0.015 0.17 0.037
 5

290 / Time-Domain Analysis of Control Systems Chap. 6

When the value of A is set at 13.5 and 200, the roots of the characteristic
equation are determined and listed as follows:

A = 13.5 s u s2 = -2.1, -32.4

,4=200 j„j 2 = -17.25 ±y26.

These roots, together with the ones for A = 1500, are located in the s-plane
as shown in Fig. 6-19. In general, for any value of A, the roots of the character-
teristic equation are given by

Sl ,s 2 = -17.25
1
±-j-*/U90-20A (6-137)

A = 1500 /86

.s-plane

-2.1
= 135
^ .4 = 200 ;26.5

°°<- A A<0 ,4 = A = 13.5 A = A<0

-/26.5

Root loci

A = 1500 •
/86
A

Fig. 6-19. Root locus diagram of the system in Fig. 6-14.

Therefore, for values of A between zero and 59.5, the two roots are real and lie

on the negative and the system is overdamped. For

real axis in the s-plane,
values of A greater than 59.5, the roots are complex conjugate; the real parts of
the roots are equal to —17.25 and are not affected by the value of A. Therefore,
as A approaches infinity, the damping factor of the second-order system is always
-1
equal to 17.25 sec When A varies continuously between -co and °o, the
.

two roots of the characteristic equation trace out two continuous loci in the
s-plane, as shown in Fig. 6-19. In general, these are called the root loci of
the characteristic equation or of the control system. In control system studies a
root-locus diagram can be very useful for analysis and design once the relation-
ship between the root location and the transient behavior has been established.
In this case Fig. 6-19 shows that the second-order system is always stable for all

finite positive values of A. When A is negative, one of the roots is positive,

 :

eC Time Response
'
' of a Positional Control System 291
/

which corresponds to a time response that

increases monotonically with time
and the system is said to be unstable. The
dynamic characteristics of the transient
response as determined from the root-locus
diagram of Fig. 6-19 are summarized
as follows

Amplifier Gain Characteristic Equation Roots System Dynamics

<A< 59.5 Two negative distinct real Overdamped (£ > 1)
roots
A = 59.5 Two negative equal real Critically damped
roots (£ = n
59.5 <A < oo Two complex conjugate roots Underdamped (£ < 1)
with negative real parts
-oo <A< Two distinct real roots,one Unstable system
positive and one negative (f < 0)
Since the system under consideration is of
type 1, the steady-state error of
the system is zero for all positive values of
A, when the input is a step function
in other words, for a step input, the
positional error constant K, is to be used
Substituting the open-loop transfer function
G(s) for the system and H(s) 1 =
into Eq. (6-21), we have

K„
v = lim 5A = ° '
(6 138 )
j>
"i? 5(s + 34.5)
Therefore, the steady-state error is given by Eq. (6-22) as

e °° = T+X = ° ( 6 -139)

The unit step responses of Fig. 6-18 verify

this fact.

Time Response to a Ramp input

When a unit ramp function input 0,(0 = /«.(/) is applied to the control sys-
tem ot hig. 6-16, the output response is described by

=
0,(0 JB-'
UV + 2Cto.fi + coi) (6-140)

From the Laplace transform table, Eq. (6-140)

is written

U0 = ~ S + '
Sin [(a ^TZ=T 2
-«
aWl'-C^
' ' (6-H1)
where

= 2tan->Vl^ .
(6 142)

he = 115 20°' and 150° are sketched as

f-
Fig. I,
6-20
P re
nT T
nSeS f° r
'
A
Notice that in this case the steady-state error
of the system nonzero is
shown in

As seen from Eq. (6-141), the last term will decay

to zero as time approaches
infinity. Therefore, the steady-state response
of the system due to a unit ramp
292 / Time-Domain Analysis of Control Systems Chap. 6

4.0

e„(A = 200)

ea (A = 1500) s
3.0
A = 1500 e ss (A = 13.5)

0,(0 = tu s (t)
A = 13.5
2.0
A = 200

Fig. 6-20. Output response (normalized) of the control system in Fig. 6-16
when the input is a unit ramp function.

input is

lim 6 c {t) = lim _2f (6-143)

<w„.

It is simple to see that the steady-state error due to the unit ramp input is

2£ _ 34.5
(6-144)
a>„ 5A
which is a constant.
A more systematic method of determining the steady-state error to a ramp
input is to use the velocity error constant K From Eq. (6-26), v.

K, = lim sG(s)H(s) = lim

^3 = ^ (6-145)

Therefore, from Eq. (6-27),

1 34.5
(6-146)

which agrees with the result of Eq. (6-144).

Equation (6-146) shows that the steady-state error is inversely proportional

to the magnitude of A. However, if we choose to improve the steady-state accu-

racy of the system by increasing the forward gain, the transient response becomes
more oscillatory. This phenomenon is rather typical in all control systems. For
higher-order systems, if the loop gain of the system is too high the system may
become unstable.
 Time Response of a Positional Control System / 293

Transient Response of a Third-Order System

It was shown
in the last section that if the armature
inductance L of the
dc motor neglected, the control system is of the second
is
order and is stable for
all positive values of A. Suppose
now that we let La 0.1 henry in the sys- =
tem in Fig. 6-16 and keep the other parameters
unchanged. The armature time
constant To is now 0.02 sec. The closed-loop transfer function given by
y Eq
4 '

(6-124) is now
£&) = __ 0.05,4
9 r (s) U.005<1 + + 2s) + (6 " 147 >
or
0.02j)(1 34s + 0.05,4

*«(*) 250,4
0M js + 50.5** + 1725^ + 250,4 (6
" 148 )

The open-loop transfer function is

G (A
° {S) _ flcfr) 250,4
~W)~ s(s> + 50.5, + 1725) (6
" 149 )

The characteristic equation of the system is

s3 + 50.5s 2 + \725s + 250A = (6-150)

It is apparent that when the armature inductance
is restored, the system is now
of the third order.
If we let A = 13.5, the closed-loop transfer function of Eq. (6-147) becomes

£G) = 1

ris) (1 + 0.48 j)(l + 0.0298j + 0.000616* 2 )

(6_151 )

The characteristic equation has a real root at

s -2.08 and two complex roots =
at 24.2 ±j 32.2. For a unit step function input the output response is
fl e (0 = - 1 I.06e-""
0.0667e-»-*« sin(32.2< +
1.88) (6-152) +
In Eq. (6-1 52), since the time constant
of the pure exponential term is more
than 10 times greater than that of the damped
sinusoid term, the transient re-
sponse of 8£t) is predominantly governed
by the characteristic root at s =
-2.08. The response due to the last term of
Eq. (6-152) decreases to zero very
rapidly with the increase of t. Comparing
Eq. (6-152) with Eq. (6-136) we see
that the damping factor for the second-order
system (L 0) is 17 25 whereas =
for the third-order system (L. =
0.1) the output response is governed by
the
exponential term, which has an equivalent
damping factor of 2.08 Thus the
third-order system will have a slower rise
time. This result is expected, since
the presence of inductance will slow
down the buildup of the armature current
thus slowing down the torque development
of the motor. However, higher induc-
tance will cause a higher overshoot in the
step response, as shown in Fig 6-21
With La - 0.1 H, and A = 348, the characteristic equation of Eq. (6-150)
J is
factored as

(*+50.5)(j* + 1725) = (6-153)

Thus the two imaginary roots at s = ±;41 5 The
characteristic equation has
response corresponding to these imaginary
roots is an undamped sinusoid. The
 l.O 1 1 1 1 i

=0A
1.4 -
/~\ /L "

-
1.2 \ ^=0.01^.^^

1.0

0.8

0.6
v/ ^ -

0.4 -

-
0.2

i
i i i i

0.1 0.2 0.3 0.4 0.5 0.6

Time (second)

Fig. 6-21. Unit step response of the positional control system of Fig. 6-16
with varying inductance, A = 200.

s-plane

A = 348
A= 13.5 /41.5

A= Q

<*•*- A A = 13.5 A =0

25.25 2.08

-/41.5
A = 348

Fig. 6-22. Root loci of the characteristic equation of Eq. (6-150) when A
is varied from zero to infinity.

294
 Sec. 6.7 Effects of Derivative Control on the Time Response / 295

frequency of the sinusoidal response is 41.5 rad/sec.

When the roots of a charac-
teristicequation lie on the imaginary axis of the s-plane,
such as in the present
situation, the linear system is said to be on
the verge of instability.
Figure 6-22 shows the root loci of the characteristic
equation of Eq (6-150)
when A is varied from zero to infinity. For all values
of A greater than 348 the
two complex roots are found in the right-half
of the *-plane, and, with time
the step response of the third-order system
will increase without bound.
From this illustrative example we have learned that a
second-order system
isalways stable as long as the loop gain is finite
and positive; third- and higher-
order systems may become unstable if the loop
gain becomes high.

6.7 Effects of Derivative Control on the Time Response of

Feedback Control Systems

The control systems considered thus far in

this chapter are all of the propor-
tional type, in that the system develops
a correcting effort that is proportional
to themagnitude of the actuating signal only. The illustrative
example given in
Section 6.6 shows that a proportional type
of control system has the limitation
or disadvantage that it is often difficult
to find a proper forward path gain so that
the steady-state and the transient responses
satisfy their respective requirements
Often, in practice, a single gain parameter
is seldom sufficient to meet the
design
requirements on two performance criteria.
It is logical
to perform other operations, in addition
to the proportional
control, on the actuating
signal. In terms of signal processing,
we may perform
a time derivative of the actuating signal.
Figure 6-23 shows the block diagram

Qs)

Fig. 6-23. Feedback control system with derivative control.

of a typical second-order feedback control

system with derivative control added
to the proportional control. In this case the
constant x d represents the amount
of derivative control used in proportion
to the ordinary proportional control
The open-loop transfer function of the system is now

" 154 )
E(s)-s(s + 2tco„) <6

Analytically, Eq. (6-154) shows that the derivative control is equivalent to the
addition of a simple zero at s = -i/ T, to the open-loop transfer function.
296 / Time-Domain Analysis of Control Systems Chap. 6

de(t)
dt

(c)

Fig. 6-24. Waveforms of c(/), e(t), and de(t)ldt showing the effect of deriva-
tive control, (a) Step response, (b) Error signal, (c) Time rate of change of
error signal.

The effect of the derivative control on the transient response of a feedback

control system can be investigated by referring to Fig. 6-24. Let us assume that
the unit step response of a proportional type of system is shown in Fig. 6-24(a).

The corresponding error signal e(t) and the time derivative of e{t) are as shown
in Fig. 6-24(b) and (c), respectively. Notice that under the assumed case the step
response exhibits a high peak overshoot. For a system that is driven by a motor
of some kind, this large overshoot is due to the excessive amount of torque
developed by the motor in the time interval < t < t u during which the error
signal is positive. For the time interval f < t < t 3 the error signal is negative,
, ,

and the corresponding motor torque is negative. This negative torque tends to
reverse the direction of motion of the output, thus causing c(t) to undershoot
during t 3 < t < t 5 During the time interval t 3 < t < t 5 the motor torque is
.

again positive, thus tending to reduce the undershoot in the response caused by
the negative torque in the previous interval. Since the system is assumed to be
stable, the error amplitude is reduced with each oscillation, and the output
eventually is settled to its final desired value.
Considering the explanation given above, we can say that the contributing
 Sec 6 7 -
Effects of Derivative Control on the Time Response
'
/ 297

factors to a high overshoot are as follows: (1) The positive correcting torque
in the interval < < is too large, and (2) the retarding torque in the time
/ /,

interval t, < < is inadequate. Therefore, in order to reduce the overshoot

t t2

in the step response, a logical approach is, to decrease the amount of positive
correcting torque and to increase the retarding torque. Similarly, in
the time
interval t 2 t < <
r 4 , the negative corrective torque should be
reduced, and the
retarding torque, which is now in the positive direction, should be increased in
order to improve the undershoot.
The derivative control as represented by the system of Fig. 6-23 gives pre-
cisely the compensation effect described in the last paragraph. Let us consider
that the proportional type of control system whose signals are
described in Fig.
6-24 is now modified so that the torque developed by the motor
is proportional
to the signal e(t) +
r d de(t)/dt. In other words, in addition to the error signal, a
signal that is proportional to the time rate of change of error
is applied to the
motor. As shown in Fig. 6-24(c), for t < <
r„ the time derivative of e(t) is
negative; this will reduce the original torque developed due
to e(f) alone. For
r, <t < r2 both e?(r) and de(t)/dt are negative, which means that the negative
,

retarding torque developed will be greater than that of the

proportional case.
Therefore, all these effects will result in a smaller overshoot. It
is easy to see that
e(/) and de(t)Jdt have opposite signs in the time interval
r2 t t3 therefore, < < ;

the negative torque that originally contributes to the

undershoot is reduced also!
Since de(t)jdt represents the slope of e{i), the derivative control
is essentially
an anticipatory type of control. Normally, in a linear system, if the
slope of e(t)
or c(f) due to a step input is large, a high overshoot will
subsequently occur.
The derivative control measures the instantaneous slope of e(t), predicts the
large
overshoot ahead of time, and makes a proper correcting effort
before the over-
shoot actually occurs.
It is apparent that the derivative control will
affect the steady-state error of
a system only if the steady-state error varies with time. If
the steady-state error
of a system constant with respect to time, the time derivative of this
is
error is
zero, and the derivative control has no effect on the
steady-state error. But if the
steady-state error increases with time, a torque is again developed
in proportion
to de(t)jdt,which will reduce the magnitude of the error.
Consider that the positional control system of Fig. 6-16 is
modified by
replacing the dc amplifier with derivative control so that
the open-loop transfer
function is now

ee (s) _
-
saq. + td s) " 155
9e (s) s(s + 34.5) (6 )

Figure 6-25 illustrates the unit step responses of the

closed-loop system
when A =
13.5 and r d =
0.01, 0.1, and 1.0. Notice that in the case of the low
value for A, an increasing value of t has the effect of slowing
d down the response
and the damping is increased. Figure 6-26 illustrates the
unit step responses
when A =
1500. In this case the damping is noticeably improved by
the deriva-
tive control. When xd = 0, Fig. 6-18 shows that the overshoot is 52 per cent
for
A = 1500, whereas with % d = 0.01, the overshoot is reduced to approximately
14 per cent. When %d = 0.1, the overshoot is completely eliminated, but the step
 ,4 = 13.5
1.0 Td = O.OlX^s.

0.8
// ^=0.1
// A = 13.5
^ Tj=1.0
0.6 A= 13.5

0.4 ~
II S

0.2

V i i 1
I

1.0 2.0 3.0 4.0

(Seconds)

Fig. 6-25. Step responses of the positional control system of Fig. 6-16 with
derivative control; A = 13.5.

^=0.01

0.02 0.03 0.04 0.05 0.08

(Seconds)

Fig. 6-26. Step responses of the positional control system of Fig. 6-16 with
derivative control ; A = 1 500.

298
 Sec 6 7 -
Effects of Derivative Control on the Time Response
-
/ 299

response is slow in reaching its final value. Figure 6-26 also shows the step
response for r d = 1.0.
The effect of derivative control on the transient response of a feedback
control system can also be studied by referring to the open-loop
transfer func-
tion of Eq. (6-154). The corresponding closed-loop transfer
function of the
system is

(6_156)
*(*) s
2
+ (2fa„ + xi(ol)s + oil
The characteristic equation of the system is

s
2
+ (2£c0„ + t„co1;)s + © = 2
(6-157)
Notice that the derivative control has the effect of increasing the coefficient
of
the s term by the quantity Td coJ. This means that the damping
of the system is
increased. Using the values as represented by Eq. (6-155),
the characteristic
equation becomes

s
2
+ (34.5 + 5Ard)s + 5A = (6-158)
Figure 6-27 shows the loci of the roots of Eq. (6-158) when A 13.5 and x d is =
varied from to oo. The improvement on the system damping
due to the deriva-
tive control is illustrated by the root loci of Fig. 6-28
with A set at 1500. Notice
that for small values of r„, the roots of the characteristic
equation are still com-
plex, although they are farther away from the imaginary
axis than those when
Td = 0. For large values of z d , the roots become real.

/w

i-plane

' Td rd =0 rd =Q Td ='
-* fc **-
32.4 -2.1

Fig. 6-27.Root loci of the characteristic equation of positional control

system with derivative control,* 2 + (34.5 5Ax d)s 5A +
0,A = 13.5. + =
It should be pointed out that although the derivative control
as fashioned
by the scheme shown in Fig. 6-23 generally improves the damping
of a feedback
control system, no considerations have been given
here on the practicability of
the configuration. In practice, however, the transfer function
of (1 + xd s) cannot
be physically realized by passive elements. Furthermore,
derivative control has
the characteristics of a high-pass
filter which tends to propagate noise and
dis-
turbances through the system. Practical controllers which
are used to improve
300 / Time-Domain Analysis of Control Systems Chap. 6

-/86
s-plane

*d
J—< -»- a

-/86
Td =0

Fig. 6-28.Root loci of the characteristic equation of positional control

system with derivative control; s 2 + (34.5 + 5Azd)s + 5 A --= 0, <4 1500. =

the performance of control systems usually have more complex transfer func-
tions than that of the derivative controller illustrated here.

6.8 Effects of Integral Control on the Time Response of

Feedback Control Systems

The counterpart of derivative control is the integral control that introduces a

signal in proportion to the time integral of the error. Figure 6-29 illustrates the
basic scheme for the integral control applied to a second-order system. The signal
applied to the process consists of two components: one proportional to the
instantaneous error signal, the other, to the time integral of the error. The
parameter K x
is a proportional constant. The open-loop transfer function of
the system is

C(s)_ (oljs + Kj (6-159)

E(s) 2
s (s + 2Ca>„)

R(s)
>l
y.
«»
1

HgH i ,
s(j
or
+ 2?<o„)
C(s)

*1 [ dr
Jo

Integral contro

Fig. 6-29. Feedback control system with integral com rol.

 Sec. 6.8
Effects of Integral Control on the Time Response / 301

One obvious effect of the integral control is that it

increases the order of the sys-
tem by 1 More important is that it increases the type of the
.
system by 1 There-
tore, the steady-state error of the original system without integral control
is
improved by an order of 1. In other words, if the steady-state error
to a given
P U S Constant the inte Sral contr°l educes it to zero. In the
^ cmi
(6-159), the system will now have a zero
'
case of Eq
steady-state error when the input is a
ramp function. However, because the system is
now of the third order it tends
to be less stable than the original
second-order system. Tn fact, if the loop gain
ot the system is high, the system may be unstable.
Let us consider the positional control
With the
system of Fig. 6-16 again, with L =
0. integral control, the open-loop transfer
function of the system be-
comes
ec {s) _ 5A(s + K
0e(s)
~ ~s\s + 34.5)) t

(6
" 16
°)
The characteristic equation is

j3 + 34.5s 2 + 5As + 5AK, = (6-161)

The of the value of K, on the transient behavior
effect
of the system may be
investigated by plotting the roots of Eq.
(6-161) as a function of A. Figure 6-30

1 OO

t t
4 ;co
s-plane A A
s-plane

A =0 A =oo i
\ ,A =0
- 34.5 - K t
A A =0 A=0
=

- 34.5 + AT, |

1
(a) (b)

Fig. 6-30. (a) Root loci of the characteristic

equation of a feedback control
system with integral control, s *
+ 34.5*2 + SAs + 5AKi = 0, K t <
34.5. (b) Root loci of the characteristic
equation of a feedback control sys-
tem with integral control, ^
+ 34.5*2 5As + = + 5AK X 0, Ki = 34.5.
302 / Time-Domain Analysis of Control Systems Chap. 6

s-plane

A = t» .4=0 \

-*i -34.5
\ 1-34.5 + AT!

Fig. 6-30 (Cont.). (c) Root loci

(c)
T ooll

of the characteristic equation of a feedback

control system with integral control, s 3 + 34.5.S 2 + 5As — 5AK\ = 0,
K >
x 34.5.

illustrates the root loci of Eq. (6-161) for three different values of AT t and when A
varies between zero and infinity. Notice that when K l
lies between and 34.5, the
roots of the characteristic equation all lie in the left half s-plane. When K =
t

34.5, the system becomes second order and the two roots lie on the yea-axis for all
values of A between and oo, and the system is on the verge of instability. When
the value of K x
exceeds 34.5, the system with the integral control is unstable for
all values of A.
As an alternative, we can fix the value of A and show the effect of varying
K on the roots of the characteristic equation. Figure 6-3 1 shows the root loci
t

of Eq. (6-161) with A = 13.5 and K < oo. < t

To verify the analytical results obtained in the preceding sections, Fig. 6-32
shows the step responses of the system with A = 13.5 and K = 1, 10, and 34.5. x

As predicted by the root loci of Figs. 6-30 and 6-31, the response for K = 34.5 x

is a pure sinusoid.

6.9 Rate Feedback or Tachometer Feedback Control

The philosophy of using the derivative of the actuating signal to improve the
damping of a system can be applied to the output signal to achieve a similar
 s-plane

K, AT, =0
K
-32.4

Fig. 6-31. Root loci of the

characteristic equation of a feedback
control
system with integral control, ^
+ 34.5*2 + 5 As + 5AK t
= 0, A = 13.5.

1_ ATj = M). K \
=34.5
2

1.0 2.0 3.0 4.0 5.0

Time (seconds)

Fig. 6-32. Step responses of the

positional control system of Fig 6-16
with
La = and integral control, 13.5. A=

303
304 / Time-Domain Analysis of Control Systems Chap. 6

In other words, the derivative of the output signal is fed back and com-
effect.

pared with the reference input. In practice, if the output variable is mechanical
displacement, a tachometer may be used which converts the mechanical dis-
placement into an electrical signal that is proportional to the derivative of the
displacement. Figure 6-33 shows the block diagram of a second-order system
with a secondary path that feeds back the derivative of the output. The transfer
function of the tachometer is denoted by K,s, where K, is the tachometer constant,
usually expressed in volts/unit velocity.

R(s) >, «<),/\ ?) w2 *- C(s)

9 ^9 t i
s(.s + 2?oo„)

K,s

Fig. 6-33. Second-order system with tachometer feedback.

The closed-loop transfer function of the system of Fig. 6-33 is

<M col
(6-162)
R(s) s* + (2{a>„ + K,col)s + col
and the characteristic equation is

s
1
+ (2£co„ + Kfi>l)s + col = (6-163)

Comparing Eq. (6-163) with Eq. (6-157), which is the characteristic equation of
the second-order system with derivative control, we see that the two equa-
tions are of the same form. In fact, they are identical if % d is interchanged with
K,. Therefore, we conclude that the rate or tachometer feedback also im-
proves the damping of a feedback control system. However, it should be noted
that the closed-loop transfer function of Eq. (6-162) does not have a zero,
and thus the responses of the two systems will not be identical even if K,
equals x d .

The open-loop transfer function of the system with tachometer feedback is

obtained from Fig. 6-33

col
(6-164)
E(s) s(s + 2£co B + K,col)

The system is still of type 1, so the basic characteristic of the steady-state error
is not altered. However, for a unit ramp function input, the steady-state error to
 Sec. 6.10
Control by State-Variable Feedback / 305

the system of Fig. 6-23 which has the derivative

control is 2Clco„, whereas that
of the system in Fig. 6-33 is (2£ K,(a„)/co + n.

6.10 Control by State-Variable Feedback 45

One of the design techniques in modern control theory is that instead of using
controllers that have
dynamics and feedback from the output variable, flexibility
can be gained by feeding back some or all of the state
variables to control the
process. In the system with tachometer feedback,
shown in Fig. 6-33, if we
decompose the process in the forward path by direct decomposition,
we can
show that the system is actually equivalent to having state
feedback. Figure
6-34(a) shows the state diagram of the process

C(£)
= (6-165)
E(s) s(s + 2Cca„)
which is decomposed by direct decomposition. If the states
x t and x2 are phys-
ically accessible, we may
feed back these variables through individual gains,
as
shown in Fig. 6-34(b), to give closed-loop control of the process.
The closed-loop transfer function of the system in Fig. 6- 34(b)
is

C(s) col
R W s* + (2Co>, + g )s +T>
t
(6 " 166)

CO

-O
c

2f"„

(a)

-Si

(b)

Fig. 6-34. Control of a second-order system by state feedback.

306 / Time-Domain Analysis of Control Systems Chap. 6

Comparing this transfer function with that of the system with tachometer feed-
back, Eq. (6-162), wetwo transfer functions would be identical
notice that the
if g t
=
and g 2 = K,col. In fact, in selecting the feedback gains g^ and g 2 in
col ,

order to have zero steady-state error for a step input, g should equal col. The l

value of g 2 is selected to satisfy the damping requirements of the system.

The significance of this illustration is that, if all the state variables are
available for feedback, we can achieve at least the same or better control with
state feedback as with dynamic controllers and feeding back only the output.
Note that the system with tachometer feedback shown in Fig. 6-33 has only the
output variable available for feedback. If we regard the output as one of the
states, it is fed back through a unity gain. The second state needed for feedback
is actually "reconstructed" by the dynamics of the tachometer. In other words,

the tachometer acts as an "observer" (see Chapter 11), which recovers the state
variable from the output variable.
In modern control theory, certain types of design algorithm, such as the
linear regulator theory, naturally lead to state-variable feedback. Since the
eigenvalues (or the poles of the closed-loop transfer function) of a linear system
directly control the transient response of the system, it would be desirable if the
designer is able to place the eigenvalues according to the performance specifica-
tions. It is shown in Chapter 11 that if a system is completely controllable, the
eigenvalues of the system can be placed arbitrarily.
The following example gives an illustration on how the state feedback and
eigenvalue assignment affect the time response of a linear control system.

Example 6-3 Consider that the transfer function of a linear process is

2 °,
G(s)
v
'
= i£^
E(s)
= ,
s 2 (s + ,, (6-167)
1)

Figure 6-35(a) shows the state diagram of G(s), and Fig. 6-35(b) shows the state dia-
gram with feedback from all three states. The closed-loop transfer function of the sys-
tem is

= (6 " 168)
W) *
3
+ (g + 3 I)* 2 + Sis + gl
Let us assume that we desire to have zero steady-state error with the input being a unit
step function, and in addition, two of the closed-loop poles must be at s = — 1 +j
and s = —1 —j. The steady-state requirement fixes the value of gi at 20, and only
#2 and £- 3 need to be determined from the eigenvalue location.
The characteristic equation of the system is

s 3
+ (g 3 + Ds 2
+ Sis + 20 = + 1 -j)(s + 1 -/)(> + a) (6-169)

Equating the coefficients of the corresponding terms in the last equation, we get

g2 =22 and g3 = 11

and the third pole is at s — —10.

Since the complex closed-loop poles have a damping ratio of 0.707, and the third
pole is of these poles, the system acts like a second-order system.
quite far to the left

Figure 6-36 shows that the unit step response has an overshoot of 4 per cent.
 20
o- -O » O-
e

(a)

(b)

Fig. 6-35. Control of a third-order system

by state feedback.

Fig. 6-36. Unit step response of the

control system in Example 6-3.

307
308 / Time-Domain Analysis of Control Systems Chap. 6

REFERENCES

Time- Domain Analysis

1. O. L. R. Jacobs, "The Damping Ratio of an Optimal Control System," IEEE

Trans. Automatic Control, Vol. AC-10, pp. 473-476, Oct. 1965.

2. G. A. Jones, "On the Step Response of a Class of Third-Order Linear Systems,"

IEEE Trans. Automatic Control, Vol. AC-12, p. 341, June 1967.
3. R. A. Monzingo, "On Approximating the Step Response of a Third-Order
Linear System by a Second-Order Linear System," IEEE Trans. Automatic Con-
trol, Vol. AC-13, p. 739, Dec. 1968.

State- Variable Feedback

4. W. M. Wonham, "On Pole Assignment in Multi-input Controllable Linear Sys-

tems," IEEE Trans. Automatic Control, Vol. AC-12, pp. 660-665, Dec. 1967.

5. J.C. Willems and S. K. Mitter, "Controllability, Observability, Pole Allocation,

and State Reconstruction," IEEE Trans. Automatic Control, Vol. AC-16, pp.
582-595, Dec. 1971.

PROBLEMS

6.1. A pair of complex-conjugate poles in the j-plane is required to meet the various
specifications below. For each specification, sketch the region in the j-plane in
which the poles may be located.
(a) C ^ 0.707, co„ !> 2 rad/sec (positive damping)
(b) < C < 0.707 CO < 2 rad/sec (positive damping)
(c) C ^ °-5, 2 < <o„ < 4 rad/sec (positive damping)
(d) 0.5 < £ < 0.707, co„ < 2 rad/sec (positive and negative damping)

6.2. Determine the position, velocity, and acceleration error constants for the follow-
ing control systems with unity feedback. The open-loop transfer functions are
given by
50
(a) G(s) =
+ 0.1.s)(l + 2s)
(1
,

(b) G(s) = K
5(1 + 0.l5)(l + 0.5s)

(c) G(s) = z K
s(s +4s + 200)

6.3. For the systems in Problem 6.2, determine the steady-state error for a unit step
input, a unit ramp input, and an acceleration input t 2 /2.

6.4. The open-loop transfer function of a control system with unity feedback is

riA 500

Evaluate the error series for the system. Determine the steady-state error of the
system when the following inputs are applied:
Chap. 6
Problems / 309

(a) r{t) = «,(r)/2

f
2

(b) r(t) = (1 + It + t*)u,0)

Show that the steady-state error obtained from the
error series is equal to the
inverse Laplace transform of E(s) with the
terms generated by the poles of
E(s)IR(s) discarded.

6.5. In Problem 6.4, if a sinusoidal input #(/) = sin cot is applied

to the system at
/ - 0, determine the steady-state error of the system
by use of the error series
for co = 5 rad/sec. What are the limitations in the error series when r(t) is
sinusoidal?

6.6. A machine-tool contouring control system is to cut the piecewise-linear contour

shown in Fig. P6-6 with a two-axis control system.
Sketch the reference input of
each of the two-axis systems as a function of time.

t = 4.5 sec 7 sec

12 IS

6.7. A machine-tool contouring control system is to cut

a perfect circular arc with
a two-axis control system. Determine the
reference inputs of the two systems
that will accomplish this.

6.8. A step motor gives a single step response shown in Fig. P6-8 after a pulse
exci-

Step position 1

Step position
0.005 sec

Figure P6-8.
310 / Time-Domain Analysis of Control Systems Chap. 6

tation is applied. Find a linear second-order transfer function to model the

motor for this operation.

6.9. The attitude control of the missile shown in Fig. P5-1 1 is accomplished by thrust
vectoring. The transfer function between the thrust angle 5 and the angle of
attack can be represented by

(refer to Problem 5.1 1) where K and a are positive constants. The attitude con-
trol system is represented by the block diagram in Fig. P6-9.

-^c^
attitude 5 e

Or
-P~ .
k
J — k
. Gp (s)

Attitude rate
sensor K,s

Attitude
sensor Ks

Figure P6-9.

(a) In Fig. P6-9, consider that only the attitude sensor loop is in operation
{K, = 0).
Determine the performance of the overall system with respect to
the relative values of K, Ks and a. ,

(b) Consider that both loops are in operation. Determine the minimum values
oiK, and K, in terms of A' and a so that the missile will not tumble.
(c) It is desired that the closed-loop system should have a damping ratio of £

and a natural undamped frequency of w„. Find the values of K, and K, in

terms of £, co„, a, and K.
6.10. A controlled process is represented by the following state equations
Xj = X\ — 3X2

x = 5xi + u 2

The control is obtained from state feedback such that

U = —glXi — giX 2
where gi and g 2 are real constants.
(a) Find the locus in the gi versus g 2 plane on which the overall system has a
natural undamped frequency of ^/2" rad/sec.
(b) Find the locus in the gi versus g 2 plane on which the overall system has a
damping ratio of 70.7 per cent.
(c) Find the values of gi and g 2 such that £ = 0.707 and co„ = V/T rad/sec.
6.11. Given a linear time-invariant system which is described by the state equations

i = Ax + Bw
Chap. 6
Problems / 311

where
1

B
-1 -2
The input u is a unit step function. Determine the equilibrium state
which
satisfies Ax + B« = 0.

6.12. Repeat Problem 6.11 when

1 0"

1 B
-2 -3
6.13. The block diagram of a missile attitude control system shown
is in Fig. P6-13
control is represented by u(t), and the dynamics of the missile are represented
by
d„(s) L
U(s) ~ Js*

e
^r\ e
Gc (s)
U
L/Js 2
8o

+ X—

Desired output

-*- t

Figure P6-13.
312 / Time-Domain Analysis of Control Systems Chap. 6

The attitude controller is represented by G c (s), and (s) is the actual heading or
output.
(a) With Gc (s) = 1, determine the response of the system, 9 a (t), when the input
6 r {t) is a unit step function. Assume zero initial conditions. Discuss the
effects of L and / on this response.
(b) Let Gc (s) = (1 + Td s), L = 10, and J = 1000. Determine the value of Td
so that the system is critically damped.
(c) It is desired to obtain an output response with no overshoot (see Fig.

P6-13); the response time may not be minimum. Let Gc (s) = 1, and the
system is controlled through the input 6 r (t), which is chosen to be of the
form shown. Determine the values of k and f so that the desired output i

is obtained. Use the same values of L and / as given before.

6.14. The block diagram of a simple servo system is shown in Fig. P6-14.

R(s)
A
V .
Amplifiei
gain K
Motor I
s
2
C(s)

as +o

Figure P6-14.

(a) For K=
10, determine the values of a and b to give an overshoot of 16
per cent and a time constant of 0.1 sec of the system response to a unit step
input. Time constant is defined here as the inverse of the damping factor.
(b) If the value of K is decreased slightly, how does it affect the damping ratio
of the system?
(c) Plot several points on the loci of roots of the system characteristic equation
as K is varied from zero to infinity.

6.15. The parameters of the positioning servo system shown in Fig. P6-15 are given
below:
JL = load inertia 1 ft-lb/rad/sec 2
BL = load viscous friction 0.00143 ft-lb/rad/sec
Jm = motor inertia 8 x 10"* ft-lb/rad/sec 2
Bm = motor viscous friction negligible
Rf = generator field resistance 50 Q
Lf = generator field inductance 5 henry
R„ = total armature resistance of generator and
motor 48.8 £2
K — t motor torque constant 0.812 ft-lb/A
Kg = generator constant 200 volts/amp
La = total armature inductance of generator and
motor negligible
(a) For an amplifier gain of K = 100, find the roots of the characteristic equa-
tion of the system. Locate these roots in the s-plane.
(b) For K= 100, evaluate the output response L (,O when 0,(0 is a unit step
displacement input. Sketch the waveform of 6 L (t)
 Chap. 6
Problems / 313

-A/VWr-
Rf 'f ~T~ '/ = :onstan1
'

WW—* , Ik + " +
Amplifier J
e Lf (g\ eg
gain K _
f 1 eb
(

? O

iflZ>

Motor to load gear

ratio n = i
50

Load to potentiometer
gearratio =1/1

Figure P6-15.

(c) Repeat parts (a) and (b) for K = 60.7.

(d) Repeat parts (a) and (b) for K = 50. How does the steady-state response
ot VL {t) compare with the reference input
0,(0 ?
6.16. The following parameters are given for the servo system shown in Fig. P6-16

motor 'f
~ constant
Demodulator dc amplifier
ed
A

60 Hz
ac
o-

Synchros

Figure P6-16.
314 / Time-Domain Analysis of Control Systems Chap. 6

K = sensitivity of error detector

s 1 volt/rad
JL = load inertia 0.05 ft-lb/rad/sec 2
BL = load viscous friction 0.005 ft-lb/rad/sec
Jm = motor inertia 0.05 ft-lb/rad/sec 2
B;= motor viscous friction negligible
Ki = motor torque constant 1 ft-lb/amp
L„ = motor armature inductance negligible
Ra = motor armature resistance ion
Kd = gain of demodulator 1 volt/volt
Gear ratio n = Ni/N 2 1: 1

(a) Write the characteristic equation of the system and determine the value of
the dc amplifier gain A for a critically damped system.
(b) For a unit ramp function input 9 r {t) = tu {t),
what should be the minimum
s

value of A 9 L (t) will follow

so that the steady-state value of the response
the reference input with a positional error not exceeding 0.0186 rad? With
this gain setting, evaluate the output response 9 L (t).

6.17. In the feedback control system shown in Fig. P6-17, the sensitivity of the syn-
chro error detector is 1 V/deg. After the entire system is set up, the transfer
function of the two-phase motor is determined experimentally as

Km
s(l + Tm s)
where Km 10 volts-sec, and Tm = 0.1 sec.

/ 2-phase
/ servomot

ec =eL ,e m =iooei

Figure P6-17.

(a) If the load on the output shaft is to be driven in its steady state at a constant
speed of 30 rpm, what is the minimum value of gain A of the amplifier in
order that the deviation between output and input positions will not exceed
3° when the steady state is reached ?
(b) The gain of the amplifier is given by A = 35. Determine the damping ratio

C and the undamped natural frequency of the system.

Chap 6
'

Problems / 315

(c) The amplifier is modified to differentiate the error signal so that the output
of the amplifier is written

e 2 {t) = Ae(t) + ATd d-^-

where A = 35. Determine the value of Td so that the damping ratio
is 40
per cent. Repeat part (a) with the modified
amplifier.
 7
Stability of Control Systems

7.1 Introduction

It was shown in Chapter 6 that the transient response of a linear feedback control
system is governed by the location of the roots of the characteristic equation.

Basically, the design of linear feedback control systems may be regarded as

a problem of arranging the location of the characteristic equation roots in such
a way that the corresponding system will perform according to the prescribed
specifications. We shall learn in Chapter 1 1 that there are conditions the system
must satisfy before its characteristic equation roots can be arbitrarily assigned.
Among the many forms of performance specifications used in the design
of control systems, the most important requirement is that the system must be

However, there are various ways of defining stability, espe-

stable at all times.
cially when we include all types of systems, such as linear and nonlinear.
In any case, the notion of stability is used to distinguish two classes of
systems: useful and useless. From a practical standpoint, a stable system may
be useful, whereas an unstable system is useless.
In essence, the analysis and design of linear control systems is centered on
stability studies. The notion of stability may be expanded to include absolute
stability and relative stability. Absolute stability refers to the condition of
stable or unstable. Once the system is found to be stable, it is of interest to
determine how stable it is, and the degree of stability is measured by relative
stability. Parameters such as overshoot and damping ratio used in relation to

the transient response in Chapter 6 provide indications of the relative stability

of a linear time-invariant system in the time domain.

316
 Sec 7 ' 2 Stability. Characteristic
'
Equation, and the State Transition Matrix 317
/

In this chapter we are concerned with the subject of absolute

stability of
control systems, which issimply a yes or no proposition.
From the illustrative examples of Chapter 6 we may summarize
the relation
between the transient response and the characteristic
equation roots as follows:

1. When all the roots of the characteristic

equation are found in the
lefthalf of the j-plane, the system responses
due to the initial con-
ditions will decrease to zero as time approaches
infinity.
2. If one or more pairs of simple roots are located
on the imaginary
axis of the j-plane, but there are no
roots in the right half of the
j-plane, the responses due to initial
conditions will be undamped
sinusoidal oscillations.
3. Ifone or more roots are found in the right half
of the j-plane, the
responses will increase in magnitude as time
increases.

In linear system theory, the last two

categories are usually defined as
unstable conditions. Note that the responses referred to in the above
conditions
are due to initial conditions only, and
they are often called the zero-input
sponses. * re-

7.2 Stability, Characteristic Equation,

and the State
Transition Matrix

We can show from a more rigorous approach that

the zero-input stability of
a linear time-invariant system is determined
by the location of the roots of
the
characteristic equation or the behavior
of the state transition matrix <h(t)
Let a linear time-invariant system be
described by the state equation

x(0 = Ax(t) + Bu(r) (7

.jx

where x(t)is the state vector and u W the input vector. For zero input
x(t) =
satisfies the homogeneous state equation x(0 =
Ax(f ) and is defined as the
equd&num state of the system. The zero-input
stability is defined as follows
If the zero-mput response x(t\ subject to
finite initial state x(t returns to the
),
emkbnum state x(t) = as approaches infinity, the system
t

tzz-jzsr
is

m0re mathematical manner

mstabie ms
type °fstaMity is
-

the foregoing definition may

is

*° k ™-
said to bistable-

*" -
A //^\
Inear Ume-mvanant system
>

said to be stable {zero input)

is
be stated-
if for any finite
uutud state x(, ) there is a positive number M
[which depends x(l)] sZhthat L
(1) \\x(t)\[<M for all t>t rj_ 2
)
and

(2) lira
t-">3
|| x(r) ||
= (7-3)
318 / Stability of Control Systems Chap. 7

where \
\
x(t) |
|
represents the norm* of the state vector x(t), or

=r 11/
»

l*(OII (7-4)
_(=1 J

The condition any

stated in Eq. (7-2) implies that the transition of state for
(> /„ as represented by the norm of the vector x(r) must be bounded. Equation
(7-3) states that the system must reach its equilibrium state as time approaches
infinity.

The interpretation of the stability criterion in the state space is illustrated

by the second-order case shown in Fig. 7-1. The state trajectory represents the
transition of x(t) for t > t from a finite initial state x(r ). As shown in the figure,
x(t ) is represented by a point that is the tip of the vector obtained from the
vector sum *i(r ) and x 2 (t ). A cylinder with radius forms the upper bound M
for the trajectory points for all t > t and as t approaches infinity, the system
,

reaches its equilibrium state x(t) = 0.

*2<>o)

Fig. 7-1. Stability concept illustrated in the state space.

Next we shall show that the definition of stability of linear time-invariant

systems given above leads to the same conclusion on the condition of the roots
of the characteristic equation. For the zero-input condition, the state transition
equation of the system is
x(f) = <J>(r - (.W,) (7-5)

where §{t — t ) is the state transition matrix.

Taking the norm on both sides of Eq. (7-5) gives

|x(OII = IIW-'o)x('o)l (7-6)

The norm of a vector is the generalization of the idea of length. ||x|| is always a real
number.
 Sec. 7.3 Stability of Linear Time-Invariant Systems with Inputs / 319

An important property of the norm of a vector is

l|x(0||<||<K'-'o)l|||x(f )|[ (7-7)

which analogous to the relation between lengths of vectors.
is

Then the condition in Eq. (7-2) requires that

\\tft l|x(f )|| be finite - OH
Thus, if x(/ ) ||is finite as postulated, <j>(t
1 1
/„) must also be finite for /
1
1

t
- 1
1
>
Similarly, Eq. (7-3) leads to the condition
that

lim || Mt-t„) = || (7. 8 )

or

!™ *'& - '«) = (7-9)

i,j = 1, 2, ,
n, where
.
<f>u
. .(t /„) is the yth element of- <f>(t
- /„).
In Eq. (4-42) the state transition matrix is written

<(>(0 = £- [(iI-A)-']
1

(7 .10)

Since si |
A| - =
is the characteristic equation
of the system, Eq (7-11)
implies that the time response of
ft/) is governed by the roots of the characteris-
tic equation. Thus the condition
in Eq. (7-8) requires that the roots
of the
characteristic equation must all have negative real parts.

7.3 Stability of Linear Time-Invariant Systems with Inputs

Although the stability criterion for linear

time-invariant
systems given in the
preceding section for the zero-input condition, we
is
can show that, in general
the stability condition for this class of
systems is independent of the inputs
alternative way of defining stability of linear
An
time-invariant systems is as follows •

A system is stable if its output is bounded for any bounded input

In other words, let c(t) be the output and r(t) the input of a
linear system
with a single input-output. If

K0l<JV<oo for/>/ (7 . 12)

then
I
c(t) | < M < oo for / > t
(7_ 13 )
However, there are a few exceptions to the foregoing
gives rise to an impulse response at / = t
definition. A differentiator
when it issubjected to a unit steo
function input. In this case the amplitude of
the input is bounded but the
amplitude of the output is not, since an impulse
is known to have a'n infinite
amplitude. Also, when a unit step function
is applied to a perfect integrator
the output is an unbounded ramp function.
However, since a differentiator and
an integrator are useful systems, they are defined as stable
all
systems and are
exceptions to the stability condition defined
above.
Weshall show that the bounded
input-bounded output definition of stabi-
lity again leads to the requirement
that the roots of the characteristic
equation
be located in the left half of the j-plane
320 / Stability of Control Systems Chap. 7

Let us express the input-output relation of a linear system by the con-

volution integral

c(t) = T r{t - z)g(r) dr (7-14)

J

where g(r) is the impulse response of the system.

Taking the absolute value on both sides of Eq. (7-14), we get

\c(t)\ = \r r{t-x)g{x)dx (7-15)

I J

Since the absolute value of an integral is not greater than the integral of
the absolute value of the integrand, Eq. (7-15) is written

\c(t)\< I"" | r(f - t)1 |

*<t) |
rfr (7-16)
J

Now if r{i) is a bounded signal, then, from Eq. (7-12),

I
c{t) |
< Jo" N
f |
g(z) dx |
= N f"
JO
[
g(r) dx
|
(7-17)

Therefore, if c{t) is to be a bounded output,

N •I
[°°\g{T)\dT<M<oo (7-18)

or

r|g(T)|^T< JP<oo (7-19)

J o

A physical interpretation of Eq. (7-19) is that the area under the absolute-value
curve of the impulse response g(t), evaluated from t = to t = oo, must be
finite.

We
shall now show that the requirement on the impulse response for
stabilitycan be linked to the restrictions on the characteristic equation roots.
By definition, the transfer function G{s) of the system and the impulse response
g{i) are related through the Laplace transform integral

G(s)= f g(t)e'"dt (7-20)

J

Taking the absolute value on the left side of Eq. (7-20) gives

\G(s)\<r\g(t)\\e-"\dt (7-21)
-I

The roots of the characteristic equation are the poles of G(s), and when s
takes on these values, \G(s)\ oo. Also, s =
a +jco; the absolute value of =
e~" is |
e~" \. Equation (7-21) becomes

oo <C F\g(t)\\e-°'\dt (7-22)

J

one or more roots of the characteristic equation are in the right half or
If
on the imaginary axis of the s-plane, a 0, and thus \e~'"\<N= 1. Thus >
 Sec. 7.4 Methods of Determining Stability of Linear
Control Systems / 321

Eq. (7-22) is written

<f N\g(t)\dt = j™\ g

Q
(t)\dt (7-23)
for Re(s) = a > 0.
Since Eq. (7-23) contradicts the stability
criterion given in Eq (7-19) we
conclude that for the system to be stable, the
roots of the characteristic equation
must all he inside the left half of the s-plane.
The discussions conducted in the preceding sections
lead to the conclusions
that the stability of linear time-invariant
systems can be determined by checking
whether any roots of the characteristic
equation are in the right half or on the
imaginary axis of the *-plane. The regions
of stability and instability in the s-
plane are illustrated in Fig. 7-2. The
imaginary axis, excluding the origin, is
included in the unstable region.

]<x>

v y s-plane

Stable ^\ / Unstable
region / region

^
Stable
region
l
Oy
K
'*
Unstable
region

\/
Fig. 7-2. Stable and unstable regions in the .s-plane.

7.4 Methods of Determining Stability of Linear Control Systems

Although the stability of linear time-invariant systems may be checked by

investigating the lm pulse response, the state transition matrix, or
by finding
the roots of the characteristic equation,
these criteria are difficult to implement
in practice. For instance, the impulse
response is obtained by taking the inverse
Laplace transform of the transfer function,
which is not always a simple task-
a similar process is required to evaluate the
state transition matrix Mt)
The
solving of the roots of a high-order
polynomial can only be carried out' by a
digital computer. In practice, therefore,
the stability analysis of a linear system
322 / Stability of Control Systems Chap. 7

is seldom carried out by working with the impulse response or the state transi-
tion matrix, or even by finding the exact location of the roots of the characteristic
equation. In general, we are interested in algorithms that are straightforward
to apply and which can provide answers to stability or instability, without
excessive computations. The methods outlined below are frequently used for
the stability studies of linear time-invariant systems.

1. Routh-Hurwitz criterion an algebraic method that provides infor-

1
:

mation on the absolute stability of a linear time-invariant system.

The criterion tests whether any roots of the characteristic equation
lie in the right half of the s-plane. The number of roots that lie on
the imaginary axis and in the right half of the .?-plane is also indi-
cated.
2. Nyquist criterion 6 : a semigraphical method that gives information
on the between the number of poles and zeros of the
difference
closed loop transfer function by observing the behavior of the
Nyquist plot of the loop transfer function. The poles of the closed-
loop transfer function are the roots of the characteristic equation.
This method requires that we know the relative location of the zeros
of the closed-loop transfer function.
3. Root locus plot (Chapter 8): represents a diagram of loci of the
characteristic equation roots when a certain system parameter
varies. When the root loci lie in the right half of the s-plane, the
closed-loop system is unstable.
4. Bode diagram (Appendix A): the Bode plot of the loop transfer
function G(s)H(s) may be used to determine the stability of the

closed-loop system. However, the method can be used only if G(s)

H(s) has no poles and zeros in the right-half s-plane.
5. Lyapunov's stability criterion: a method of determining the stability
of nonlinear systems, although it can also be applied to linear sys-
tems. The stability of the system is determined by checking on the
properties of the Lyapunov function of the system.

1" 5
7.5 Routh-Hurwitz Criterion

The Routh-Hurwitz criterion represents a method of determining the location

of zeros of a polynomial with constant real coefficients with respect to the
left half and the right half of the j-plane, without actually solving for the

zeros.The method can be applied to systems with single input and output
and with multiple inputs and outputs, as well as single- or multiple-loop
systems.
Consider that the characteristic equation of a linear time-invariant system
is of the form

F(s) = a s" + a,5"-' + a z s"' 2 + . . . + an _ x

s - an = (7-24)

where all the coefficients are real numbers.

 Sec. 7.5
Routh-Hurwitz Criterion / 323

In order that there be no roots of the last equation

with positive real parts
it is necessary but not sufficient that*

1. All the coefficients of the polynomial have the

same sign.
2. None of the coefficients vanishes.

These two necessary conditions can easily be

checked by inspection How-
ever, these conditions are not sufficient;
it is quite possible that a polynomial
with all its coefficients nonzero and of the same sign still have zeros in the right
s
halfofthej-plane.
The necessary and sufficient condition that all
the roots of Eq. (7-24) lie in
the left half of the s-plane is that the polynomial's Hurwitz determinants,
— Dk , k
1>2 n, must be all positive.
The Hurwitz determinants of Eq. (7-24) are given
by
a, a3
a3
D =
t ai D2 = D 3 do a2
«0 a2
*3

a\ o3 a5 • «2»-l
a a2 a4 • «2»-2
a, a3 • 02„-3
D„ = a a2 «2»-4
•
(7-25)
"t «2»-5

where the coefficients with indices larger than n or with negative indices are
replaced by zeros.
At first glance the application of the Hurwitz
determinants may seem to be
formidable for high-order polynomials,
because of the labor involved in evalua-
ting the determinants in Eq.
(7-25). Fortunately, the rule was simplified by Routh
mtoa tabulation, so one does not have to
work with the determinants of Eq.

in e/S •
C baSfC kWS ° f a ' gebra
' ^ f0U ° Wing rektions
«™ obse ™d for the polynomial
~= —2 all roots

^ = X products of the roots taken 2 at a time

(73
= -* Products of the roots taken 3 at a time
Zl

~ =(-!)" products of all roots

AH the ratios must be positive and nonzero unless at least one of the roots has a positive
p&rt. real
 : :

324 / Stability of Control Systems Chap. 7

The first step in the simplification of the Routh-Hurwitz criterion is to

arrange the polynomial coefficients into two rows. The first row consists of the

first, third, fifth, . . . coefficients, and the second row consists of the second,
the fourth, sixth, . . . coefficients, as shown in the following tabulation

ao az a\ a& as ...
a\ a-$ as an 09 ...

The next step is to form the following array of numbers by the indicated
operations (the example shown is for a sixth-order system)

s" ao ai <H a6
55 a\ a3 a5
a\a 2 — a az ^ a\a* — aoa; =B aia * - °° :
i-°= a6
s*
a\ a\ a\
Aai — a\B ^ Aai — a\a6 = n A xO - a x x0 _
S3 c A
A A
s2
BC - AD E_ Ca 6 - A x
06
CxO-/ixO =
c C c
1
ED — Cat, F
r.
s
E
Fa 6 - E x
s° „ a6

The array of numbers and operations given above is known as the Routh
tabulation or the Routh array. The column of ^s on the left side is used for
identification purpose.
Once the Routh tabulation has been completed, the last step in the Routh-
Hurwitz criterion is to investigate the signs of the numbers in the first column

of the tabulation. The following conclusions are drawn The roots of the poly- :

nomial are all in the left half of the s-plane if all the elements of the first column
of the Routh tabulation are of the same sign. If there are changes of signs in the
elements of the first column, the number of sign changes indicates the number of
roots with positive real parts.
The reason for the foregoing conclusion is simple, based on the require-
ments on the Hurwitz determinants. The relations between the elements in the
first column of the Routh tabulation and the Hurwitz determinants are:

a =a a

a, =D t

° D 1

E-S±
D 3

'-a
 Sec 7 5 '
Routh-Hurwitz
-
Criterion / 325

Therefore, if all the Hurwitz determinants are positive, the elements

in the first
column would also be of the same sign.
The following two examples illustrate the application of the Routh-Hurwitz
criterion to simple problems.

Example 7-1 Consider the equation

- 2)(s + 1)0 - 3) = s - 4s 2 + s +
3
6 = (7-26)
which has two negative coefficients. Thus, from the necessary condition,
we know
without applying the Routh-Hurwitz test that the equation has at least
one root with
positive real parts. But for the purpose of illustrating the
Routh-Hurwitz criterion, the
Routh tabulation is formed as follows:

s3 1 i
Sign change
*2 -4 6
Sign change

(-4)W-(6)(1) =25 Q

s° (2-5X6) -(-4)(0) = 6 Q

Since there are two sign changes in the first column of

the tabulation, the poly-
nomial has two roots located in the right half of the j-plane.
This agrees with the known
result, as Eq. (7-26) clearly shows that the two right-half plane roots =
are ats 2 and
$ = 3.
Example 7-2 Consider the equation

2s* + s + 3s + 5s +
3 2
10 = (7-27)
Since the equation has no missing terms and the coefficients are all of the same
sign, it the necessary condition for not having roots in
satisfies
the right half of the
5-plane or on the imaginary axis. However, the
sufficient condition must still be
checked. The Routh tabulation is made as follows

2 3 10
1 5
Sign change

Sign change
° )(
V 2)(5)
=-7 10

(-7)(5)_-dX10) = 643
10

Since there are two changes in sign in the first column, the equation has two roots
in the right half of the .s-plane.

Special Cases

The two illustrative examples given above are designed so

that the Routh-
Hurwitz criterion can be carried out without any
complications. However
depending upon the equation to be tested, the following
difficulties may occur
326 / Stability of Control Systems Chap. 7

occasionally when carrying out the Routh test:

1. The first element in any one row of the Routh tabulation is zero,
but the other elements are not.
2. The elements in one row of the Routh tabulation are all zero.

In the first case, if a zero appears in the first position of a row, the elements
in the next row will all become infinite, and the Routh test breaks down. This
situation may be corrected by multiplying the equation by the factor (s a), +
where a is any number,* and then carry on the usual tabulation.

Example 7-3 Consider the equation

- 1)
2
+ 2) = s* - 3s + 2 = (7-28)
2
Since the coefficient of the s term is zero, we know from the necessary condition
that at least one root of the equation is located in the right half of the j-plane. To

determine how many of the roots are in the right-half plane, we carry out the Routh
tabulation as follows

.$3 1 -3
s2 2
Sl °o

Because of the zero in the first element of the second row, the first element of the
third row is infinite. To correct this situation, we
multiply both sides of Eq. (7-28) by
the factor (s + a), where a is an arbitrary number. The simplest number that enters
one's mind is 1. However, for reasons that will become apparent later, we do not
choose a to be 1 or 2. Let a = 3; then Eq. (7-28) becomes

(s - l)\s + 2)(s + 3) - si + 3s 3 -3s 2 -7s + 6 =0 (7-29)

The Routh tabulation of Eq. (7-29) is

1 -3 6
3 -7
Sign change
-? + 7_ 2
3 ~ 3
Sign change
(- jX-7) ~ 18 = 2Q
-§
6

Since there are two changes in sign in the first column of the Routh tabulation, the
equation has two roots in the right half of the f-plane.

As an alternative to the remedy of the situation described above, we may

replace the zero element in the Routh tabulation by an arbitrary small positive

*If one chooses to use a negative number for a, the (s + a) term will contribute a root in
the right half of the j-plane, and this root must be taken into account when interpreting the
Routh tabulation.
 .:

Sec 7 5
- -
Routh-Hurwitz Criterion / 327

number e and then proceed with the Routh test. For instance, for the equation
given in Eq. (7-28), we may replace the zero element in the second row of the
Routh tabulation by e then we have ;

1 -3
Sl e 2
Sign change
~3e - 2
Sign change
s° 1

Since e is postulated to be a small positive number, (—

3e— 2)/e approaches
-2/e, which is a negative number; thus the first column of the last tabulation
has two sign changes. This agrees with the result obtained earlier.
On the
other hand, we may assume e to be negative, and we can easily verify that the
number of sign changes is still two, but they are between the first
three rows.
In the second special case, when all the elements in one row of
the Routh
tabulation are zeros, it indicates that one or more of the
following conditions
may exist

1 Pairs of real roots with opposite signs.

2. Pairs of imaginary roots.
3. Pairs of complex-conjugate roots forming symmetry about
the origin
of the s-plane.

The equation that is formed by using the coefficients

of the row just above
the row of zeros is called the auxiliary equation. The order of the auxiliary
equa-
tion is always even, and it indicates the number of root
pairs that are equal in
magnitude but opposite in sign. For instance, if the auxiliary
equation is of the
second order, there are two equal and opposite roots. For a
fourth-order aux-
iliary equation, there must be two pairs of equal
and opposite roots. All these
roots of equal magnitude can be obtained by solving
the auxiliary equation.
When a row of zeros appear in the Routh tabulation, again the test breaks
down. The test may be carried on by performing the following remedies:

1. Take the derivative of the auxiliary equation with respect to s.

2. Replace the row of zeros with the coefficients of the resultant
equa-
tion obtained by taking the derivative of the auxiliary
equation.
3. Carry on the Routh test in the usual manner with the newly
formed
tabulation.

Example 7-4 Consider the same equation, Eq. (7-28), which is used in Example
7-3.
In multiplying this equation by a factor (s
+ a), logically the first
number that comes into one's mind would be a = I. Multiplying
both sides of Eq. (7-28) by (s + 1), we have

(s - l) 2 (s + 2)(s + 1) = j* + ^ _ 3^2 _^+ 2 = (7 . 30)

 n

328 / Stability of Control Systems Chap 7

The Routh tabulation is made as follows:

^V
s4
s3

2 ax— 3) — ax—
1

--2,
-10
-3

.
2
2

^_2=0
Since the s 1 row contains all zeros, the Routh test terminates prematurely. The
difficulty in this case is due to the multiplication of the original equation, which already
has a root at s = 1, by the factor (s + 1). This makes the new equation fit special case
(2). To remedy this situation, we form the auxiliary equation using the coefficients
contained in the s 2 row, that is, the row preceding the row of zeros. Thus the auxiliary
equation is written
A(s) = -2s 2 +2=0 (7-31)
Taking the derivation of A(s) with respect to s gives

dA(s)
ds
= -4.s (7-32)

Now, the row of zeros in the Routh tabulation is replaced by the coefficients of Eq.
(7-32), and the new tabulation reads as follows:

s* 1 -3
s> 1 -1
Sign change
s2 -2 2 (coefficients of auxiliary equations)
i1 -4 [coefficients of dA(s)jds]
Sign change
s° 2

Since the preceding tabulation has two sign changes, the equation of Eq. (7-28)
has two roots in the right-half plane. By solving the roots of the auxiliary equation in
Eq. (7-31), we have
s1 = 1 or s = ±1
These are also the roots of the equation in Eq. (7-30). It should be remembered
that the roots of the auxiliary equation are also roots of the original equation, which is
under the Routh test.

Example 7-5 In this example we shall consider equations with imaginary roots.
Consider

(s + 2)(5 - 2)(.s +j)(s -JXs +s + l) 2

= s + s — 2s* — 3i - Is ~
6 s 3 2
As - 4 = (7-33)

which is known to have two pairs of equal roots with opposite signs at i = ±2 and
s = ±j. The Routh tabulation is

1 -2 -7 -4
1 -3 -4
1 -3 -4
 Sec 7
-
- 5 Routh-Hurwitz Criterion 329
/

Since a row of zeros appears prematurely, we form the auxiliary equation using the
coefficients of the s* row. The auxiliary equation is

A(s) = s* - 3s 2 - 4 = (7-34)
The derivative of A(s) with respect to 5 is

-jf = 4s -6s =
3
(7-35)

from which the coefficients 4 and -6 are used to replace the row of zeros in the Routh
tabulation. The new Routh tabulation is

1 -2 -7
1 -3 -4
1 -3 -4
4 -6 [coefficients of
Sign change
^^]
-1.5 -4
16.7
-4

Since there is one change in sign in the first column of the

new Routh tabulation,
Eq. (7-33) has one root in the right half of the j-plane. The
equal, but opposite roots
that caused the all-zero row to occur are solved from
the auxiliary equation. From
Eq. (7-34) these roots are found to be

•s
1

= +2, —2, +j, and —/

Afrequent use of the Routh-Hurwitz criterion is for a quick
check of the
stability and the simple design of a linear feedback control system. For
exam-
ple, the control system with integral control
shown in Fig. 6-29 has the charac-
teristic equation

s3 + 34.5s 2 + 7500s + 7500/^, = (7-36)

The Routh-Hurwitz criterion may be used to determine the range of AT, for
which the closed-loop system is stable. The Routh tabulation of Eq.
(7-36) is

»3 1 7500
j2 34.5 7500ATi
,., 258,750 - 7500A-,
341
s° 7500ATi

For the system to be stable, all the coefficients in the first

column of the last
tabulation should be of the same sign. This leads to
the following conditions:
258,750 - 7500/sT, .
343 l
>° (7-37)

7500*, > (
7_ 38)
From the condition in Eq. (7-37) we have
AT, < 34.5 (7.39)
 : —

330 / Stability of Control Systems Cn ap 7

and the condition in Eq. (7-38) gives

Ki > (7-40)
Therefore, the condition for stability is that A^ must satisfy

<K < x
34.5 (7-41)

Example 7-6 Consider the characteristic equation of a certain closed-loop control

system,
s3 + 3Ks 2 + (K + 2)s +4= (7-42)

It is desired to determine the range of Kso that the system is stable. The Routh tabula-
tion of Eq. (7-42) is

s3 1 (K + 2)
s2 3K 4

so 4

From the s 2 row, the condition of stability is

K>
and from the s 1 row, the condition of stability is

3K + 6K - 4 >
2
or K< -2.528 or K> 0.528

When the conditions of K> and K> 0.528 are compared, it is apparent that the
latter limitation is the more stringent one. Thus for the closed-loop system to be stable,
K must satisfy
K> 0.528

The requirement of K < —2.528 is disregarded since K cannot be negative.

It should be reiterated that the Routh-Hurwitz criterion is valid only if
the characteristic equation is algebraic and all the coefficients are real. If any
one of the coefficients of the characteristic equation is a complex number, or if
the equation contains exponential functions of s, such as in the case of a system
with time delays, the Routh-Hurwitz criterion cannot be applied.
Another limitation of the Routh-Hurwitz criterion is that it offers infor-
mation only on the absolute stability of the system. If a system is found to be
stable by the Routh's test, one still does not know how good the system is
in other words, how closely the roots of the characteristic equation roots are
located to the imaginary axis of the s-plane. On the other hand, if the system is

shown to be unstable, the Routh-Hurwitz criterion gives no indication of how

the system can be stabilized.

7.6 Nyquist Criterion 615

Thus far, two methods of determining stability by investigating the location of

the roots of the characteristic equation have been indicated

1. The roots of the characteristic equation are actually solved. This

 . . :

Sec. 7.6
Nyquist Criterion / 331

step can be carried out on a digital computer by means of a root-

finding subroutine.
2. The relative location of the roots with respect
to the imaginary
axis of the j-plane is determined by means of the Routh-Hurwitz
criterion.

These two methods are not very useful for design purposes.
It is necessary
to devise stability criteria that will enable the
analyst to determine the relative
stability of the system. It would be desirable
if the criterion indicates how the
stability of the system might be improved.
The Nyquist criterion possesses the following features that
make it desirable
for the analysis as well as the design of control
systems

1 It provides the same amount of information on the

absolute stability
of a control system as the Routh-Hurwitz criterion.
2. In addition to absolute system stability, the
Nyquist criterion in-
dicates the degree of stability of a stable system
and gives an indica-
tion of how the system stability may be improved,
if needed.
3. It gives information on the frequency-domain
response of the system.
4. It can be used for a stability study of systems with
time delay.
5. It can be modified for nonlinear systems.

We may formulate the Nyquist criterion by first using

modern control
notation, that state equations. Let a linear control system
is,
with a single input
be represented by the following set of equations:

x(?) = Ax(0 + Be(t) (7.43)

c{t) = Dx(r)
(7.44)
e(t) = - c(t)
/•(/)
(7.45)
A block diagram that gives the transfer relations
of the system is shown
in Fig. 7-3. This is known as a closed-loop
system with unity feedback. The open-
loop transfer function of the system is defined
as

G{s) = H(sl ~ A) "'I* (7 . 46)

When the feedback is nonunity, with the feedback dynamics represented
by the

-« G(s) —>—
R(s)
C(s)
"V*? (si-- A) IB D

Fig. 7-3. Block diagram of a closed-loop control system with

unity feed-
back.
332 / Stability of Control Systems Chap. 7

*~ C(s)

Fig. 7-4. Block diagram of a closed-loop control system with unity feed-
back.

transfer function H(s), the system block diagram is modified as shown in Fig. 7-4.
The closed-loop transfer function of the system of Fig. 7-4 is

CO) G(s)
(7-47)
R(s) 1 + G(s)H(s)
Let the denominator of the closed-loop transfer function be represented
byFO). Then
F(s) = +
1 G(s)H(s) = + D(sl - A)-'B#(s)
1 (7-48)

It is easy to see that the zeros of F(s) are the roots of the characteristic equation
of the system, if there is no pole-zero cancellation in G(s)H(s).
Let us assume that G(s) and H(s) are rational functions; that is, G(s)H(s)
isa quotient of two polynomials with constant coefficients. In general, F(s) can
be written
+ z,)(j + z ) (2 + z m )
F(s) :
K(s
(7-49)
2 . . .

+ pMs + p )...(s + p s J (s 2 n)

where z,(i = 1,2, ... ,m) &ndp k (k = 2, n) are either real or in complex-
1, . . . ,

conjugate pairs. Then the roots of the characteristic equations are s = —z t ,

—z 2 , . . . , —zm . For the closed-loop system to be stable, it is required that none

of these roots has a positive real part. There is no particular restriction on the
location of the poles of F(s), which are at s 0, p l7 — p z , —p„. It is = — , . .
.

important to note that the poles of F(s) are the same as those of G(s)H(s). If
any one of the poles of G(s)H(s) lies in the right half of the s -plane, the open-
loop system is said to be unstable; however, the closed-loop system can
still be

of F(s) are found in the left half of the ,?-plane. This is a

stable, if all the zeros
very important feature of a feedback control system. In preceding chapters it
has been pointed out that a high forward-path gain generally reduces the steady-
state error of a feedback control system. Consequently, it is desirable to use high
gain in multiple-loop control systems. Although this practice may result in an
unstable inner-loop system, the entire closed-loop system can be made stable
by proper design.
Before embarking on the fundamentals of the Nyquist criterion, it is es-
sential to summarize on the pole-zero relationships with respect to the system
functions.
Sec. 7.6 Nyquist Criterion / 333

1. Identification of poles and zeros:

loop transfer function zeros = zeros of G(s)H(s)
loop transfer function poles = poles of G(s)H(s)
closed-loop poles = poles of C(s)/R(s)
= zeros of F(s) = + 1 G(s)H(s)
= roots of the characteristic equation
2. The poles of F(s) are the same as the poles of the loop transfer
function G(s)H(s).
3. For a closed-loop system to be stable, there is no restriction on the
location of the poles and zeros of the loop transfer function G(s)
H(s), but the poles of the closed-loop transfer function must all be
located in the left half- of the .s-plane.

"Encircled" versus "Enclosed"

It is important to distinguish between the concepts encircled and enclosed,

which are used frequently with the Nyquist criterion.

Encircled. A point is said to be encircled by a closed path if it is found inside

the path. For example, point A in Fig. 7-5 is encircled by the closed path T,
since A is found inside the closed path, whereas point B is not encircled. In

the application of the Nyquist criterion, the closed path also has a direction
associated with it. As shown in Fig. 7-5, point A is said to be encircled by T
in the counterclockwise direction.
When considering all the points inside the closed path, we can say that
the region inside the closed path T is encircled in the indicated direction.

Fig. 7-5. Illustration of the definition of encirclement.

334 / Stability of Control Systems Chap. 7

Enclosed. A said to be enclosed by a closed path if it is

point or region is

found to lie to the left of the path when the path is traversed in the prescribed
direction. Putting it another way, enclosure is equivalent to counterclockwise

encirclement. For instance, the shaded regions shown in Fig. 7-6(a) and (b) are
considered to be enclosed by the closed path I\ In other words, point A in Fig.
7-6(a) is enclosed by I\ but point A in Fig. 7-6(b) is not. However, in Fig. 7-6(b),
point B and all the points in the region outside T are considered to be enclosed.

(a) (b)

Fig. 7-6. Definition of enclosed points and regions, (a) Point A is enclosed
by T. (b) Point A is not enclosed but B is enclosed by the locus T.

Number of encirclement and enclosure. When point A is encircled or

enclosed by a closed path, a number N may be assigned to the number of en-
circlement or enclosure, as the case may The value of N may be determined
be.
by drawing a vector from A to any arbitrary point s on the closed path r and
t

then follow the path in the prescribed direction until it returns to the
let s t

starting point. The total net number of revolutions traversed by this vector is
N. For example, point A in Fig. 7-7(a) is encircled once by T, and point B is

(a) (b)

Fig. 7-7. Definition of the number of encirclement and enclosure.

Sec. 7.6 Nyquist Criterion / 335

encircled twice, all in the clockwise direction. Point A in Fig. 7-7(b) is enclosed

once; point B is enclosed twice.

Principle of the Argument

The Nyquist criterion was originated as an engineering application of the

well-known principle of the argument in complex variable theory. The principle
is stated as follows, in a heuristic manner. Let F(s) be a single-valued rational

function that is analytic everywhere in a specific region except at a finite number

of points in the s-plane. For each point at which F(s) is analytic in the specified
region in the s-plane, there is a corresponding point in the F(s)-plane.
Suppose that a continuous closed path r, is arbitrarily chosen in the s-
plane, as shown in Fig. 7-8(a). If all the points on I\ are in the specified region

,, ](X>
ly'ImF
s-plane
F(s)-plane

*- ReF

(a) (b)

Fig. 7-8. (a) Arbitrarily chosen closed path in the s-plane. (b) Correspond-
ing locus TF in the FO)-plane.

in which F(s) is analytic, then curve rF mapped by the function F(s) into the
F(s)-plane is also a closed one, as shown in Fig. 7-8(b). If, corresponding to
point s t
in the s-plane, point F(s t ) is located in the F(s)-plane, then as the r,
locus is traversed starting from point Si in the arbitrarily chosen direction
(clockwise) and then returning to jj after going through all the points on the
T s shown in Fig. 7-8(a)], the corresponding F P locus will start from
locus [as
point F(sj) and go through points F(s 2 ) and F(s 3 ), which correspond to s 2 and
s 3 respectively, and return to the starting point, F(Ji). The direction of traverse
,

of r F may be either clockwise or counterclockwise; that is, in the same direction

or the opposite direction as that of r s depending on the particular function ,

F(s). In Fig. 7-8(b) the direction of rP is shown, for illustration purposes, to be

counterclockwise.
It should be pointed out that, although the mapping from the j-plane to
the F(j)-plane is one to one for a rational function F(s), the reverse process is

usually not a one-to-one mapping. For example, consider the function

336 / Stability of Control Systems Chap. 7

which is analytic in the finite .s-plane except at the points s = 0, — 1, and —2.
For each value of s in the finite .s-plane other than the three points 0, — 1 and ,

—2, there is only one corresponding point in the F(.s)-plane. However, for each
point in the F(V)-plane, the function maps into three corresponding points in
the .s-plane. The simplest way to illustrate this is to write Eq. (7-50) as

s{s+l)(s + 2)=-^ (7-51)

The left side of Eq. (7-51) is a third-order equation, which has three roots when
F(s) is chosen to be a constant.
The principle of the argument can be stated: Let F(s) be a single-valued
rational function that is analytic in a given region in the s-plane except at a finite
number ofpoints. Suppose that an arbitrary closed path T s
is chosen in the s-plane

so that F(s) is analytic at every point on F s; the corresponding F(s) locus mapped
in the F(s)-plane will encircle the origin as many times as the difference between
the number of the zeros and the number of poles of F(s) thai are encircled by
the s-plane locus IV
In equation form, this statement can be expressed as

N= Z - P (7-52)
where
N= number of encirclement of the origin made by the F(s)-plane
locus rF
Z= number of zeros of F(s) encircled by the s-plane locus T s
in the
s-plane

P = number of poles of F(s) encircled by the .s-plane locus T s in the

.s-plane

In general, N
can be positive (Z > P), zero (Z = P), or negative (Z < P).
These three situations will now be described.

1. N> (Z > P). If the .s-plane locus encircles more zeros than poles
of F(s) in a certain prescribed direction (clockwise or counterclock-
wise), N is a positive integer. In this case the F(s)-plane locus will
encircle the origin of the F(s)-plane N
times in the same direction
as that of 1%.
2. /V = (Z = P). If the .s-plane locus encircles as many poles as zeros,
or no poles and zeros, of F(s), the F(.s)-plane locus F F will not encircle
the origin of the F(s)-plane.
3. TV < (Z < P). If the .s-plane locus encircles more poles than zeros
of F(s) in a certain direction, TV is a negative integer. In this case the
F(.s)-plane locus, FF , will encircle the origin TV times in the opposite
direction from that of T s
.

A convenient way of determining N with respect to the origin (or any other
point) of the F(s) plane is to draw a line from the point in any direction to
Sec. 7.6 Nyquist Criterion / 337

F(s)-plane F(s)-p\ane

N=-2

F(s )-plane

N=

Fig. 7-9. Examples of the determination of N in the F(i)-plane.

infinity; the number of net intersections of this line with the F(s) locus gives the
magnitude of N. Figure 7-9 gives several examples of this method of determining
N. It is assumed that the T s locus has a counterclockwise sense.
A rigorous proof of the principle of the argument is not given here. The
following illustration may be considered as a heuristic explanation of the prin-
ciple. Let us consider the function F(s) given by

F(s) (7-53)
+ />i)C* + Pi)
where K is a positive number. The poles and zero of F(s) are assumed to be as
shown in Fig. 7- 10(a).
The function F(s) can be written

F(s) = \F(s)\/F(s)

(7-54)

\
s + Pi [
s +p 2
Us Is + Pi— js + p 2)
I
338 / Stability of Control Systems Chap. 7

x-plane

ReF

Fig. 7-10. (a) Pole-zero configuration of F(s) in Eq. (7-53) and the j-plane
trajectory 1%. (b) F(j)-plane locus, IV, which corresponds to the I\ locus
of (a) through the mapping of Eq. (7-53).

Figure 7-10(a) shows an arbitrarily chosen trajectory r, in the s-plane,

with the arbitrary point 5^ on the path. The function F(s) evaluated at s = st
is given by

F( Sl )
K(s x + zt)
(7-55)
0*1 + />i)C*i + Pi)
The factor s + z can be represented graphically by the vector drawn
t x

from — z, to s Similar vectors can be defined for {s + p,) and (s + Pi)-

x
.
x t

Thus F(s ) is represented by the vectors drawn from the given poles and zero
1

to the point s u as shown in Fig. 7-10(a). Now, if the point s- moves along the
locus r\ in the prescribed counterclockwise direction until it returns to the
starting point, the angles generated by the vectors drawn from the poles (and
Sec. 7.6 Nyquist Criterion / 339

zeros if there were any) that are not encircled by T s when Si completes one

round trip are zero; whereas the vector (j, + z,) drawn from the zero at —z u
which is encircled by I\, generates a positive angle (counterclockwise sense)
of 2n rad. Then, in Eq. (7-54), the net angle or argument of F(s) as the point
Si travels around T s once is equal to 2n, which means that the corresponding
F(s) plot must go around the origin 2n radians or one revolution in a counter-
clockwise direction, as shown in Fig. 7-10(b). This is why only the poles and
zeros of F(s), which are inside the JT, path in the j-plane, would contribute to
the value of N of Eq. (7-52). Since poles of F(s) correspond to negative phase
angles and zeros correspond to positive phase angles, the value of depends N
only on the difference between Z and P.
In the present case,
Z= 1 F=0
Thus
N = Z- P= 1

which means that the F(y)-plane locus should encircle the origin once in the same
direction as the s-plane locus. It should be kept in mind that Z and P refer only
to the zeros and poles, respectively, of F(s) that are encircled by r„ and not the
total number of zeros and poles of F(s).
In general, if there are N
more zeros than poles of F(s), which are encircled
by the j-plane locus r, in a prescribed direction, the net angle traversed by the
.F(j)-plane locus as the s-plane locus is traversed once is equal to

2n(Z -P) = 2nN (7-56)

This equation implies that the F(s)-plane locus will encircle the origin N times
in the same direction as that of r,. Conversely, if N more poles than zeros are
encircled by T s in a given prescribed direction, N in Eq. (7-56) will be negative,
,

N
and the F(s)-p\ane locus must encircle the origin times in the opposite direction
to that of IV
A summary of all the possible outcomes of the principle of the argument
is given in Table 7-1.

Table 7-1 Summary of All Possible Outcomes of the

Principle of the Argument

F(s)-Plane Locus

Sense of the Number of Encirclements Direction of

N=Z—P s-Plane Locus of the Origin Encirclement

N> Clockwise
N Clockwise
Counterclockwise Counterclockwise

N <0 Clockwise Counterclockwise

Counterclockwise
N Clockwise

Af = Clockwise No encirclement
Counterclockwise No encirclement
340 / Stability of Control Systems Chap. 7

Nyquist Path

At this point the reader should place himself in the position of Nyquist
many years ago, confronted with the problem of determining whether or not
the rational function 1 + G{s)H{s) has zeros in the right half of the j-plane.
Apparently, Nyquist discovered that the principle of the argument could be
used to solve the stability problems, if the j-plane locus, T s , is taken to be one
that encircles the entire right half of the .r-plane. Of course, as an alternative,
I\ can be chosen to encircle the entire left-half s-plane, as the solution is a
relative one. Figure 7-11 illustrates a T s locus, with a counterclockwise sense,
which encircles the entire right half of the s-plane. This path is often called the
Nyquist path.
Since the Nyquist path must not pass through any singularity of F{s),
the small semicircles shown along the^'co axis in Fig. 7-11 are used to indicate
that the path should go around these singular points of F(s). It is apparent that

s-plane

Poles of
F(s)

Fig. 7-11. Nyquist path.

Sec. 7.6 Nyquist Criterion / 341

if any pole or zero of F(s) lies inside the right half of the s-plane, it will be
encircled by this Nyquist path.
For the convenience of analysis, the Nyquist path is divided into a mini-
mum of three sections. The exact number of sections depends upon how many
of those small semicircles are necessary on the imaginary axis. For the situation
illustrated in Fig. 7-1 1, a total of eight sections needs to be denned. The order
+
of numbering these sections is entirely arbitrary. The notations j co\, j co~i, +j0 ]

+
—j0 —ja>U and —jcoj are used to identify the starting and ending points of
,

the small semicircles only.

Section 1 from s :+j°° to = +

jcot along they'co axis.
Section 2: from +jcot to +jcoT along the small semicircle around
•J =j bi-
section 3 : from 5 = jco~[ to +J0 + along the jco axis.
Section 4: from +j0 + to —y'0 + along the small semicircle around s =
Section 5 : from s = —j0 + to —year along the jco axis (mirror image
of section 3).

Section 6: from j = —jco~[ to —j cot along the semicircle around

=
—ja>i (mirror image of section 2).
s
Section 7: from s —jco\ to s = =
—j°° along the jco axis (mirror
image of section 1).

Section 8: from s = —joo to s = +j°o along the semicircle of

infinite radius.

Nyquist Criterion and the G(s)H(s) Plot

The Nyquist criterion is a direct application of the principle of the argument

when the j-plane locus is the Nyquist path. In principle, once the Nyquist path
is specified, the stability of a closed-loop system can be determined by plotting
the F(s) = 1 + G(s)H(s) locus when j takes on values along the Nyquist path,
and investigating the behavior of the F(s) plot with respect to the origin of the
F(s)-plane. This is However, since usually the
called the Nyquist plot of F(s).
function G(s)H(s) is approach is to construct the Nyquist
given, a simpler
plot of G(s)H(s), and the same result of F(s) can be determined from the be-
havior of the G(s)H(s) plot with respect to the (—l,jO) point in the G(s)H(s)-
plane. This is because the origin of the F(s)-p\anc corresponds to the ( — l,y'0)
point (or the — 1 point on the real axis) of the G(s)H(s)-plane. In addition, if

the location of the poles of G(s)H(s) is not known, the stability of the open-loop
system can be determined by investigating the behavior of the Nyquist plot of
G(s)H(s) with respect to the origin of the G(s)//(s)-plane.
Let us define the origin of the G(s)H(s)-plane or the origin of the 1 +
G(s)H(s)-p\anc, as the problem requires, as the critical point. We are inter-
ested basically in two types of stability : stability of the open-loop system, and
stability of the closed-loop system.
It is important to clarify that closed-loop stability implies that 1 G(s) +
H(s) has zeros only in the left half of the j-plane. Open-loop stability implies
that G(s)H(s) has poles only in the left half of the j-plane.
 : : :

342 / Stability of Control Systems Chap. 7

With this added dimension to the stability problem, it is necessary to define

two sets of N, Z, and P, as follows

N = number of encirclements of the origin made by G(s)H(s)

Z = number of zeros of G(s)H(s) that are encircled by the

Nyquist path, or in the right half of the j-plane
P = number of poles of G(s)H(s) that are encircled by the
Nyquist path, or in the right half of the j-plane
N_ t
= number of encirclements of the (— l,y'0) point made by G(s)H(s)
Z_ , = number of zeros of 1 +
G(s)H(s) that are encircled by the
Nyquist path, or in the right half of the j-plane

P_ i
= number of poles of G{s)H(s) that are encircled by the
Nyquist path, or in the right half of the j-plane
Several facts become clear and should be remembered at this point

Po = P-i (7-57)

since G(s)H(s) and 1 + G(s)H(s) always have the same poles. Closed-loop
stability implies or requires that

Z_! = (7-58)

but open-loop stability requires that

P = (7-59)

The crux of the matter is that closed-loop stability is determined by the

properties of the Nyquist plot of the open-loop transfer function G(s)H(s).
The procedure of applying the principle of the argument for stability studies
is summarized as follows

1. Given the control system, which is represented by the closed-loop

transfer function of Eq. (7-47), the Nyquist path is defined according
to the pole-zero properties of G(s)H(s).
2. The Nyquist plot of G(s)H(s) is constructed.
3. The values of N and JV_, are determined by observing the behavior
of the Nyquist plot of G(s)H(s) with respect to the origin and the
(— \,j0) point, respectively.
4. Once N and N- are determined, the value of P a
1 (if it is not already
known) is determined from
N =Z -P (7-60)

if Z is given. Once P is determined, P_ , =P [Eq. (7-57)], and

Z_ [ is determined from

N. = t
Z_, - />_! (7-61)

Since it has been established that for a stable closed-loop system Z_ , must
be zero, Eq. (7-61) gives
JV_, = -P_, (7-62)
Sec. 7.6 Nyquist Criterion / 343

Therefore, the Nyquist criterion may be formally stated: For a closed-

loop system to be stable, the Nyquist plot ofG(s)H(s) must encircle the (—l,jO)
point as many times as the number of poles of G(s)H(s) that are in the right half of
the s-plane, and the encirclement, if any, must be made in the clockwise direction.
In general, it is not always necessary to construct the Nyquist plot which
corresponds to the entire Nyquist path. In fact, a great majority of the practical
problems can be solved simply by sketching the part of G(s)H(s) that corresponds
to the +yco-axis of the s-plane. The condition when this simplified procedure is
possible is when P 0; that is, G(s)H(s) has no poles in the right half
of the .y-plane. In this case Eq. (7-61) becomes

JV-i = Z_, (7-63)

which means that the Nyquist plot of G(s)H(s) can encircle the (— l,j0) point
only in the counterclockwise direction, since N_ in Eq. (7-63) can only be zero i

or positive. Since counterclockwise encirclement is equivalent to enclosure, we

can determine closed-loop stability by checking whether the (— l,y'0) critical
point is enclosed by the G(s)H(s) plot. Furthermore, if we are interested only in
whether N. 1
is we need not sketch the entire Nyquist plot for G(s)H(s),
zero,
only the portion from s = +yoo to s = along the imaginary axis of the
.y-plane.

The Nyquist criterion for the P_ t

= case may be stated: If the function
G(s)H(s) has no poles in the right half of the s-plane, for the closed-loop system to
be stable, the Nyquist plot of G(s)H(s) must not enclose the critical point (— 1J0).
Furthermore, if we are not interested in the number of roots of the char-
acteristicequation that are in the right-half plane, but only the closed-loop
only the G(s)H(s) plot that corresponds to the positive imaginary axis
stability,
of the j-plane is necessary. Figure 7-12 illustrates how the s /oo to s = =

/ Im GH
G(s)#(s>plane

ReGH

Fig. 7-12. Nyquist plot of G(s)H(s), which corresponds to s = jco to s 0,

to indicate whether the critical point is enclosed.
344 / Stability of Control Systems Chap. 7

portion of the Nyquist plot may be used to determine whether the critical point
at (— l.y'O) is enclosed.

7.7 Application of the Nyquist Criterion

The following examples serve to illustrate the practical application of the Nyquist
criterion to the stability of control systems.

Example 7-7 Consider a single-loop feedback control system with the loop transfer
function given by

W® = «rh) (7 - 64)

where K and a are positive constants. It is apparent that G{s)H{s) does not have any
pole in the right-half s-plane; thus, Pa = P_i = 0. To determine the stability of the
closed-loop system, it is necessary only to sketch the Nyquist plot of G(s)H(s) that
corresponds to j =/co to s = on the Nyquist path and see if it encloses the ( — 1,7*0)
point in the G(s)H(s)-p\ane. However, for the sake of illustration, we shall construct
the entire G(s)H(s) plot for this problem.
The Nyquist path necessary for the function of Eq. (7-64) is shown in Fig. 7-13.

s-plane

Fig. 7-13. Nyquist path for the system in Example 7-7.

Since G(s)H(s) has a pole at the origin, it is necessary that the Nyquist path includes a
small semicircle around s = 0. The entire Nyquist path is divided into four sections, as
shown in Fig. 7-13.
Section 2 of the Nyquist path is magnified as shown in Fig. 7-14(a). The points on
this section may be represented by the phasor

5 = €e"> (7-65)

where €{€ —
0) and 9 denote the magnitude and phase of the phasor, respectively.
As the Nyquist path is traversed from +j0 + to —jO* along section 2, the phasor of
Sec. 7.7 Application of the Nyquist Criterion / 345

i / Im GH GCs)i/(s)-plane

s-plane
80 counterclockwise
rotation

*~ a *- ReGH

(a)
co =

(b)

Fig. 7-14. (a) Section 2 of the Nyquist path of Fig. 7-13. (b) Nyquist plot of
G(s)H(s) that corresponds to section 2.

Eq. (7-65) rotates in the clockwise direction through 180°. Also, in going from +j0 +
to —y'0 +
, 9 varies from +90°
to —90° through
The corresponding Nyquist plot of
0°.

G(s)H(s) can be determined simply by substituting Eq. (7-65) into Eq. (7-64). Thus

G(s)H(s)
K (7-66)
s=ee )o ee">(ee"> + a)
Since e — 0, the last expression is simplified to

G(s)H{s)
K ooe -je (7-67)
aeeJ<>

which indicates that all points on the Nyquist plot of G(s)H(s) that correspond to
section 2 of the Nyquist path have an infinite magnitude, and the corresponding phase
is opposite that of the j-plane locus. Since the phase of the Nyquist path varies from

+90° to —90° in the clockwise direction, the minus sign in the phase relation of Eq.
(7-67) indicates that the corresponding G(s)H(s) plot should have a phase that varies
from —90° to +90° in the counterclockwise direction, as shown in Fig. 7-14(b).
In general, when one has acquired proficiency in the sketching of the Nyquist
plots, the determination of the behavior of G(s)H(s) at s = and s = oo may be
carried out by inspection. For instance, in the present problem the behavior of G(s)H(s)
corresponding to section 2 of the Nyquist path is determined from

lim G(s)H(s) lim ,

K— i-
lim —# (7-68)
1^0 s ^o s(s + a). .

From this equation it is clear that the behavior of G(s)H(s) at s = is inversely propor-
tional to s. As the Nyquist path is traversed by a phasor with infinitesimally small
magnitude, from +/0 + to —/0 + through a clockwise rotation of
1 80°, the corresponding

G(s)H(s) plot is traced out by a phasor with an infinite magnitude, 180° in the opposite
or counterclockwise direction. It can be concluded that, in general, if the limit of
346 / Stability of Control Systems Chap. 7

G(s)H(s) as * approaches zero assumes the form

lim G(s)H(s) = lim Jfo±" (7-69)

the Nyquist plot of G(s)H(s) that corresponds to section 2 of Fig. 7-14(a) is traced out
by a phasor of infinitesimally small magnitude n x 180 degrees in the clockwise
direction if the plus sign is used, and by a phasor of infinite magnitude n x 180° in the
counterclockwise direction if the negative sign is used.
The technique described above may also be used to determine the behavior of the
G(s)H(s) plot, which corresponds to the semicircle with infinite radius on the Nyquist
path. The large semicircle referred to as section 4 in Fig. 7-13 is isolated, as shown in
Fig. 7-1 5(a). The points on the semicircle may be represented by the phasor

5 = Rei* (7-70)

plane

Re/*

l / Im GH G(,s)i/(.s)-plane
180°
counterclockwise
rotation
360° clockwise
rotation

ReG

Radius -»

(b)

Fig. 7-15. (a) Section 4 of the Nyquist path of Fig. 7-13. (b) Nyquist plot
of G(s)H(s) that corresponds to section 4.

where R oo. Substituting Eq. (7-70) into Eq. (7-64) yields

K
G(s)H(s)
-_
Re j*
~ RW* = Oe-i 1*
(7-71)

which implies that the behavior of the G{s)H{s) plot at infinite frequency is described by
a phasor with infinitesimally small magnitude which rotates around the origin 2 x 1 80°
= 360° in the clockwise direction. Thus the G(s)H(s) plot that corresponds to section
4 of the Nyquist path is sketched as shown in Fig. 7-1 5(b).
Now to complete the Nyquist plot of the transfer function of Eq. (7-64) we must
consider sections 1 and 3. Section 1 is usually constructed by substituting s = jca into
Eq. (7-64) and solving for the possible crossing points on the real axis of the G(s)H(s)-
Sec. 7.7 Application of the Nyquist Criterion / 347

plane. Equation (7-64) becomes

G(jOi)H(jOi) = K (7-72)
jco(jco + a)
which is rationalized by multiplying the numerator and denominator by the complex
conjugate of the denominator. Thus

G(jco)HUco)
_ K(—co 2 — jaco)
= (7-73)
co* + a 2 co 2

The intersect of G(jco)HQco) on the real axis is determined by equating the imaginary
part of G(jCO)H(jco) to zero. Thus the frequency at which G(jCo)H(jco) intersects the
real axis is found from

= -Kaco -Ka
Im G(jCO)H(jco) (7-74)
co* + a 2 C0 2 co(cb 2 +a 2
)

which gives co = co. This means that the only intersect on the real axis in the G(s)H(s)-
plane is at the origin with co = co. Since the
Nyquist criterion is not concerned with
the exact shape of the G(s)H(s) locus but only the number of encirclements, it is not
necessary to obtain an exact plot of the locus. The complete Nyquist plot of the function
of Eq. (7-64) is now sketched in Fig. 7-16 by connecting the terminal points of the loci

i / Im GH
G(s)H(s)-p\ane

ReGH

Fig. 7-16. Complete Nyquist plot of G(s)H(s) = K/ls(s + a)].

348 / Stability of Control Systems Chap. 7

that correspond to sections 2 and 4, without intersecting any finite part of the real
axis.
It is of interest to check all the pertinent data that can be obtained from the

Nyquist plot of Fig. 7-16. First, N

= iV-i = 0. By inspection of Eq. (7-64), Z =
andP = 0, which satisfy Eq. (7-60). SinceP = P-u Eq. (7-61) leads to
Z_! = N-i +P_! =0 (7-75)

Therefore, the closed-loop system is stable. This solution should have been anticipated,
since for the second-order system, the characteristic equation is simply

s 1
+ as +K= (7-76)

whose roots will always lie in the left half of the i-plane for positive a and K.
Figure 7-16 also shows that for this problem it is necessary only to sketch the
portion of G(s)H(s) that corresponds to section 1 of the Nyquist path. It is apparent
that the (— 1, y'O) point will never be enclosed by the G(s)H(s) plot for all positive
values of K.

Example 7-8 Consider that a control system with single feedback loop has the loop
transfer function

<*»>"« = WTTY (7 " 77)

The characteristic equation of the system is

s 2
+ (1 + K)s - K = (7-78)

which has one root in the right half of the s-plane for all positive K. The Nyquist path
of Fig. 7-13 is applicable to this case.
Section 2. s = €e ie

—=
:

lim G(s)H(s) = lim ooe-^<» + ») (7-79)

This means that the Nyquist plot of G{s)H{s) which corresponds to section 2 of the
Nyquist path is traced by a phasor with infinite magnitude. This phasor starts at an
angle of +90° and ends at —90° and goes around the origin of the G(s)H(s)-p\ane
counterclockwise a total of 180°.
Section 4. s = Re'*

lim G{s)H{s) = lim— = 0e~'* (7-80)

Thus the Nyquist plot of G(s)H(s) corresponding to section 4 of the Nyquist path goes
around the origin 180° in the clockwise direction with zero magnitude.
Section 1. s =jCO:

G(j<d)H(jco) - - -A
JooUm + 1} £o4 +W 2
a>(co* + 1)
(7-81)
Setting the imaginary part of G(jOi)HU'CO) to zero, we have
CO = ±1 rad/sec (7-82)

which are frequencies at which the G(jCO)H{j03) locus crosses the real axis. Then
GUl)H(jl) =K (7-83)

Based on the information gathered in the last three steps, the complete Nyquist
Sec. 7.7 Application of the Nyquist Criterion / 349

i j lm GH

G(s)H(s) - plane

AL, = 1

K ReGH

Fig. 7-17. Complete Nyquist plot of G(s)H(s) = K(s - l)/[s(s + 1)].

plot of G(s)H(s) is sketched as shown in Fig. 7-17. We can conclude that, by inspection,
Z =1
P =p_, =o
Figure 7-17 indicates that Ao = 1, which is in agreement with the Nyquist criterion.
Figure 7-17 also gives A/_j = 1. Then
Z_, =AT_, +P_, = 1
( 7 _84)

which means that the closed-loop system is unstable, since the characteristic equation
has one root in the right half of the 5-plane. The Nyquist plot of Fig. 7-17 further
indicates that the system cannot be stabilized by changing only the value of A'.

Example 7-9 Consider the control system shown in Fig. 7-18. It is desired to deter-
mine the range of K
for which the system is stable. The open-loop
transfer function of the system is

C(s) _ Grrt _ 10*(j + 2)

(7-85)

Since this function does not have any pole or zero on the jco axis, the Nyquist path
 : :

350 / Stability of Control Systems Chap. 7

R(s) -N E( S )
K(s + 2) Y^ .S
2
10
(.5 + 3)
C
C(«)

_ >

Fig. 7-18. Block diagram of the control system for Example 7-9.

.s-plane

Fig. 7-19. Nyquist path for the transfer function of Eq. (7-85).

can consist of only three sections, as shown in Fig. 7-19. The construction of the
Nyquist plot of G(s) is outlined as follows
Section 4. s = Re 1 *
= 10K =
lim G(s) 0e -j2 * (7-86)

As the phasor for section 4 of the Nyquist path is traversed from —90° to +90°
through 180° counterclockwise, Eq. (7-86) indicates that the corresponding Nyquist
plot of G(s) is traced by a phasor of practically zero length from +180° to —180°
through a total of 360° in the clockwise sense.
Section 1. s =jco:
l0K(ja>
GUco) -(l0-3co*)-jcoi
+ 2)
(7-87)

Rationalizing, the last equation becomes

= 10 AT[2(10 - 3CQ 2 ) -co* +7CQ(10 - 3co 2 ) +./2CQ 3 ]

G(jco) (7-88)
(10 - 3co 2 ) 2 + co 6
Sec. 7.7 Application of the Nyquist Criterion / 351

Setting the imaginary part of G(jco) to zero gives

co = rad/sec
and
co = ± VlO rad/sec
which correspond to the frequencies at the intersects on the real axis of the G(s)-plane.
In this case it is necessary to determine the intersect of the G(s) plot on the imaginary
axis. Setting the real part of G(jG>) to zero in Eq. (7-88), we have

co 4 + 6co 2 - 20 = (7-89)
which gives
co = ±1.54 rad/sec
Therefore, the intersects on the real axis of the G(.s)-plane are at

C(/0) = 2 AT
and
C(yVlO) = -K
and the intersect on the imaginary axis is

G(j^/T) =j]0«/TK/3
With the information gathered in the preceding steps, the Nyquist plot for G(s)
of Eq. (7-85) is sketched as shown in Fig. 7-20. The information on the imaginary axis
is needed so that the direction of section 1 may be determined without actually plotting
the locus point by point.

../ImC
G(s)-plane

Fig. 7-20. Nyquist plot of G(s) = \QK(s + 2)/(s 3 + 3s 2 + 10).

352 / Stability of Control Systems Chap. 7

Inspection of the Nyquist plot of Fig. 7-20 reveals that A = —2. Since Eq.
(7-85) shows that G(s) has no zeros inside the right half of the .s-plane, Z = 0; this

means that P = 2. Thus P_ i

=2. Now, applying the Nyquist criterion, we have
JV_, = Z_ - P_t = Z_, -
t 2 (7-90)

Thus for the closed-loop system to be stable, Z_i = 0, which requires that JV_ = —2. t

With reference to Fig. 7-20, the stability criterion requires that the ( — 1,;0) point must
be encircled twice in the clockwise direction. In other words, the critical point should
be to the right of the crossover point at — K. Thus, for stability,
K> 1 (7-91)

The reader can easily verify this solution by applying the Routh-Hurwitz criterion
to the characteristic equation of the system.

It should be reiterated that although the Routh-Hurwitz criterion is much

simpler to use in stability problems such as the one stated in this illustrative
example, in general the Nyquist criterion leads to a more versatile solution,
which also includes information on the relative stability of the system.

7.8 Effects of Additional Poles and Zeros of G(s)H(s) on the

Shape of the Nyquist Locus

Since the performance and the stability of a feedback control system are often
influenced by adding and moving poles and zeros of the transfer functions, it

is informative to illustrate how the Nyquist locus is affected when poles and zeros
are added to a typical loop transfer function G{s)H{s). This investigation will also
be helpful to gain further insight on the quick sketch of the Nyquist locus of a
given function.
Let us begin with a first-order transfer function

=
G(s)H(s)
T1 ^ Ti
(7-92)

The Nyquist locus of G(jco)H(jco) for < ca < °° is a semicircle, as shown

in Fig. 7-21.

. / ImGH
G(s)H(s)-p\ane

CO = °° co =
ReGH
Ik

Fig. 7-21. Nyquist plot of G(s)H(s) = Kj(\ + T lS ).

Sec. 7.8 Effects of Additional Poles and Zeros of G{s)H(s) I 353

Addition of poles at s = 0. Consider that a pole at s = is added to the

transfer function of Eq. (7-92) ; then we have

G(s)H(s) = K (7-93)

The effect of adding this pole is that the phase of G(jco)H(ja>) is reduced by
—90° at both zero and infinite frequencies.
In other words, the Nyquist locus of G(j(o)H(joo) is rotated by —90° from
that of Fig. 7-21 at co = and a, = oo, as shown in Fig. 7-22. In addition, the

i I Im GH
G(s)H(s)-phne

•*-ReGH

Fig. 7-22. Nyquist locus of G(s)H(s) = K/[s(l + Tis)].

magnitude of G(jco)H(j(o) at co =
becomes infinite. In general, adding a pole
of multiplicity j at s = to the transfer function of Eq. (7-92) will give the fol-
lowing properties to the Nyquist locus of G(s)H(s)

lim IG(j<B)H(jco) = -(j + 1) ^ (7-94)

lim /GUco)H(jco) = -j* (7-95)

co~.0 Z
lim|GO'a>)#C/e>)| = (7-96)

\im\G{jm)H{j(o)\ = (7-97)

Figure 7-23 illustrates the Nyquist plots of

G(s)H(s) = K (7-98)

and
G(s)H{s)
K
(7-99)
s\\ +T t
s)

In view of these illustrations it is apparent that addition of poles at s =

will affect the stability adversely, and systems with a loop transfer function of
more than one pole at s = are likely to be unstable.
354 / Stability of Control Systems Chap. 7

G(s)#Cs)-plane

— ReGH

Fig. 7-23. Nyquist loci for (1) G(s)H(s) = K/[s 2 (l + TlS )]. (2) G(s)H(s) =
K/[sHl + Tis)].

Addition of finite poles. When a pole at s = —l/T2 is added to the G(s)

H(s) function of Eq. (7-92), we have

G{s)H(s)
K (7-100)
+ (1 7V)(1 +T 2 s)

The Nyquist locus of G(s)H(s) at co = is not affected by the addition of the

pole, since
lim GUco)H(joi) =K (7-101)

The Nyquist locus at co = °° is

lim G(jco)HUco) = lim

-K
~„" = /-180
3
(7-102)
2
<»-.«a m ->o» 1 \1 2.0)

i / Im GH G(s)H(s)-p\ane

*- Re Gtf

Fig. 7-24. Nyquist loci for (1) G(s)H(s) = Kj[{\ + TlS)(l + T2 s)]. (2)
G(s)H(s) = #/[(! + Tis)(l + T2 s)(\ + T 3 s)].
Sec. 7.8 Effects of Additional Poles and Zeros of G(s)H(s) / 355

Thus the effect of adding a pole at s = —\/T 2 to the transfer function of

Eq. (7-92) is to shift the phase of the Nyquist locus by —90° at infinite fre-
quency, as shown in Fig. 7-24. This figure also shows the Nyquist locus of

G(s)H(s)
K (7-103)
(1 + 7V)(1 + 7»(1 + T 3 s)

These examples show the adverse effects on stability that result from the
addition of poles to the loop transfer function.

Addition of zeros. It was pointed out in Chapter 6 that the effect of the
derivative control on a closed-loop control system is to make the system more
stable. In terms of the Nyquist plot, this stabilization effect is easily shown,
since the multiplication of the factor (1 + 7» to the loop transfer function
increases the phase of G(s)H(s) by 90° at co = oo.
Consider that the loop transfer function of a closed-loop system is given
by

^>^ )
-
,( i + ri(i + r 2 ,)
(7
- 104>

It can be shown that the closed-loop system is stable for < (7, + T2 )/ <K
T, . =
T2 Suppose that a zero at s — \\Td is added to the transfer function of Eq.
(7-104), such as with a derivative control. Then

G(s)H(s)
K{\ + 7» (7-105)
s(l T,5)(l + TV)
The Nyquist loci of the two transfer functions of Eqs. (7-104) and (7-105)
are sketched asshown in Fig. 7-25. The effect of the zero in Eq. (7-105) is to add

,
.
/ Im GH
G(s)H(s)-p\ane

T T2 t

K{T + T2 )
t

-*- KeGH

-<(&.- T)

(2)

co
I

Fig. 7-25. Nyquist loci for (1) G(s)H(s) = K/[s(l + T^Xl + T2 s)]. (2)
G(s)H(.s) = K(l + Tds)/[s(l + TisXl + T2 s)].
356 / Stability of Control Systems Chap. 7

90° to the phase of G(Jco)H(jco) at <y = oo while not affecting the locus at co = 0.
The crossover point on the real axis is moved from —K(Ti + T )/T T to
2 1 2

-K(T + T^IT^il + Td
t ), which is closer to the origin of the G(jco)H(jco)-
plane.

7.9 Stability of Multiloop Systems

The stability analyses conducted in the preceding sections are all centered
toward systems with a single feedback loop, with the exception of the Routh-
Hurwitz criterion, which apparently can be applied to systems of any con-
figuration, as long as the characteristic equation is known . We shall now illus-

trate how the Nyquist criterion is applied to a linear system with multiple feed-
back loops.
In principle, all feedback systems with single input and output can be
reduced to the basic system configuration of Fig. 7-4. Then it would seem that
the Nyquist criterion can be applied in a straightforward manner to the equiva-
lent loop transfer function G(s)H(s). However, owing to the multiloop nature
of the original system, the poles and/or the zeros of G(s)H(s) may be unknown,
and a systematic approach to the problem may be adopted.
Let us illustrate the procedure of applying Nyquist criterion to a multiloop
control system by means of a specific example. Figure 7-26 gives the block
diagram of a control system with two loops. The transfer function of each
block is indicated in the diagram. In this case it is simple to derive the open-
loop transfer function of the system as

C££) _ G(s) = - Gi(j)G»(j)

+ G (s)H(s)2
(7-106)
K(s + 2)
(s + 10)[j(j + 1)0 + 2) + 5]

The stability of the overall system can be investigated by sketching the Nyquist

H(s) = 5

Fig. 7-26. Multiloop feedback control system.

Sec. 7.9 Stability of Multiloop Systems / 357

locus of G(s), except that the poles of G(s) are not entirely known. To avoid
the construction of the entire Nyquist locus of G(s), we can attack the problem
in two stages, as there are two feedback loops. First, consider only the inner
loop, whose loop transfer function is G 2 (s)H(s). We shall first sketch the Nyquist
locus of G z (s)H(s) for <
co < oo. The property of the G 2 (s)H(s) plot with
respect to the (— l,jO) point gives an indication of the number of zeros of 1 +
G 2 (s)H(s) that are in the right half of the J-plane. Having found this information,
we then proceed to sketch the Nyquist locus of G(s) of Eq. (7-106) only for <
co < oo to determine the stability of the overall system.
Figure 7-27 shows the Nyquist locus of

5
G 2 (s)H(s) (7-107)
s{s + 1)(j + 2)

,. j\mG 2 H

G 2 (sW(s)-plane

ReG 2 //

Fig. 7-27. Nyquist plot of G 2 (s)H(s) = 5 Ms + i)(s + 2)].

Since the (— l,y'0) point is not enclosed by the locus, the inner loop is stable by
itself, and the zeros of 1 + G 2 is)His) are all in the left half of the j-plane. Next,
the Nyquist locus of C7(j) of Eq. (7-106) is sketched as shown in Fig. 7-28. Since
all the poles and zeros of G(s) are found to be in the left half of the j-plane, we
only have to investigate the crossover point of the G(s) locus with respect to the
(— l.v'O) point to determine the requirement on K for the overall system to be
stable. In this case the range of K for stability is < K < 50.
Now, let us consider a system that is a slightly modified version of Fig.
7-26. Use the same block diagram, but with

G,(j)
+2
__ s
(7-108)
s+ 1

Gl{s)
K
~ sis + IX* + 2)
(7-109)

His) =5 (7-110)
368 / Stability of Control Systems Chap. 7

jlmG
G(i)-plane

*- ReG

Fig. 7-28. Nyquist plot of G(s) [K(s + 2)]/{U + 10)[s(s +- l)(j + 2)

+ 5]}.

In this case we cannot use the method outline above, since the unknown gain
parameter is in the inner loop. However, we may still use the Nyquist criterion
to solve this problem. The open-loop transfer function of the system is

W ~_
G(S )
1
G.(j)Ga(j)
+G 2 (s)H(.
(7-111)
s + 2
(s + I0)[s(s + l)(j + 2) + 5tf]

Since the unknown parameter ^Tdoes not appear

as a gain factor of G(s),
it would be of no avail to sketch the
Nyquist locus of G(s)/K. However, we can
write the characteristic equation of the overall system as

s(s + 10)(j + 1)(* + 2) +s+2+ 5K(s + 10) = (7-112)

In order to create an equivalent open-loop transfer function with K as a

multiplying factor, we divide both sides of Eq. (7-112) by terms that do not
contain K. We have
SK(s + 10)
1 + s(s + + s + 2~° (7 " 113)
10Xs + l)(s + 2)
Since this equation is of the form 1 + G (s) = 0, the roots
3 of the charac-
teristic equation may be by sketching the Nyquist locus of G 3 (s).
investigated
However, the poles of G 3 (s) are not known, since the denominator of G 3 (s) is
not in factored form. The zeros of the polynominal s(s + 10)(s + l)(s + 2) -f
s + 2 may be studied by investigating the Nyquist plot of still another function
G 4 (j), which we create as follows

Figure 7-29 shows that the Nyquist locus of G 4 (s) intersects the real axis
to the right of the (— 1, y"0) point. Thus all the poles of G 3 (s) are in the left half
of the 5-plane. The Nyquist plot of G 3 (s) is sketched as shown in Fig. 7-30. Since
 Sec. 7.9
Stability of Multiloop Systems / 359

,.
/Im C 4
G 4 (s )-plane

0.009 /
ReC,

r
\

Fig. 7-29. Nyquist locus of G4 (s) = (s + 2)/[s(s + 10)(j + 1)(* + 2)].

,. / Im G 3 (s)

C 3 0?)-plane

*- ReG,

Fig. 7-30. Nyquist locus of G 3 (s) = [5K(s + W)]J[ s (s + I0){s + l)(s + 2)

+s+ 2].

the intersect of the locus on the real axis is at -O.IK, the range of K for stability
is <K< 10.
In this section we have investigated the
application of the Nyquist criterion
to multiloop control systems. For analysis
problems, the stability of the system
can be investigated by applying the Nyquist
criterion in a systematic fashion
from the inner loop toward the outer loops. For
design problems when a sys-
tem parameter Kh
to be determined for stability, it is
sometimes necessary to
start with the characteristic equation,
which may be written in the form

P{s) + KQ{s) = Q (7 _ 115)

where P(s) and Q(s) are polynomials. The stability
of the system is studied by
sketching the Nyquist plot of the equivalent
open-loop transfer function

<*)-3gj! (7-116)
360 / Stability of Control Systems Chap. 7

Thus we have also indicated a method of applying the Nyquist criterion

to design a stable system by using the characteristic equation. The Nyquist
locus represents a more convenient approach than the Routh-Hurwitz criterion,
since the latter often involves the solution of an inequality equation in the high
order of K, whereas in the Nyquist approach K can be determined from the
intersection of the G(s) locus with the real axis.
The stability analysis discussed thus far in this chapter can be applied to
linear multivariable systems. Since the characteristic equation of a multivariable
is still written
|*I-A| = (7-117)

the Routh-Hurwitz criterion is applied in the usual manner. We can also apply
the Nyquist criterion to Eq. (7-117) in the manner outlined in Eqs. (7-115) and
(7-116). On the other hand, since the stability of a linear system is independent
of the inputs and outputs of the system, we can apply the Nyquist criterion to
a multivariable system between any input-output pair while setting all other
inputs to zero, as long as the system is completely controllable and observable.

7.10 Stability of Linear Control Systems with Time Delays

Systems with time delays and their modeling have been discussed in Section
5.12. In general, closed-loop systems with time delays in the loops will be subject
to more stability problems than systems without delays. Since a pure time delay
Td is modeled by the transfer function relationship e~ TdS , the characteristic equa-
no longer have constant coefficients. Therefore, the Routh-
tion of the system will
Hurwitz criterion is not applicable. However, we shall show in the following
that the Nyquist criterion is readily applicable to a system with a pure time
delay.
Let us consider that the loop transfer function of a feedback control sys-
tem with pure time delay is represented by

G(s)H(s) ^e- T G 's

x
(s)H, (s) (7- 1 1 8)

where G^H^s) is a rational function with constant coefficients; Ta is the pure

time delay in seconds. Whether the time delay occurs in the forward path or the
feedback path of the system is immaterial from the stability standpoint.
In principle, the stability of the closed-loop system can be investigated by
sketching the Nyquist locus of G(s)H(s) and then observing its behavior with
reference to the (— l,j0) point of the complex function plane.
The effect of the exponential term in Eq. (7-118)
is that it rotates the phasor

GyfJo^Hiijco) at each co by an angle of coTd radians in the clockwise direction.

The amplitude of G^jcoJH^jco) is not affected by the time delay, since the
magnitude of e~' mT' is unity for all frequencies.
In control systems the magnitude of GiO'cw)^iO'co) usually approaches zero
as co approaches infinity. Thus the Nyquist locus of the transfer function of Eq.
(7-118) will usually spiral toward the origin as co approaches infinity, and there
are infinite number of intersects on the negative real axis of the G(s)H(s)-ip\a.ae.
For the closed-loop system to be stable, all the intersects of the G(jco)H(jco)
locus with the real axis must occur to the right of the (-IJ0) point.
 Sec. 7.10 Stability of Linear Control Systems with Time Delays / 361

/ Im GH
G(s)#(s)-plane

^ReGH

T,=4

0.16

Fig. 7-31. Nyquist plots of G(s)H(s) = e- T "/[s(s + l)(.s + 2)].

Figure 7-31 shows an example of the Nyquist plot of

G(s)H(s) = e-'-'G^H^s) = (7-119)

s(s + IX* + 2)

for several values of Td . It is observed from this diagram that the closed-loop
system is stable when the time delay Td is zero, but the stability condition dete-
riorates as Td increases. The system is on the verge of becoming unstable when
Td = 2 sec. This is shown with the Nyquist plot passing through the (— l,y0)
point.
Unlike the rational function case, the analytical solution of the crossover
points on the real axis of the G(y).ff(.s)-plane
is not trivial, since the equations

that govern the crossings are no longer algebraic. For instance, the loop transfer
function of Eq. (7-1 19)may be rationalized in the usual manner by multiplying
its numerator and denominator by the complex conjugate of the denominator.
The result is

< cos <>> T - /CQ(2 ~ G)2)1

GUcomjco) = <
ffi ff i - Lt <
7 - 120)
(
362 / Stability of Control Systems Chap. 7

The condition for crossings on the real axis of the G(s)H(s)-p\ane is

3co
2
sin coTd — co(2 — co 2
) cos coTd =
which is not easily solved, given Td .

Since the term e~ lmT " always has a magnitude of 1 for all frequencies, the
crossover problem is readily solved in the Bode plot domain (Appendix A).
Since the time-delay term affects only the phase but not the magnitude of G(jco)
H(jco), the phase of the latter is obtained in the Bode plot by adding a negative
angle of —coTd to the phase curve of GifJoc^H^jco). The frequency at which the
phase curve of G{ja>)H(jca) crosses the 180° axis is the place where the Nyquist
locus intersects the negative real axis. In general, analysis and design problems
involving pure time delays are more easily carried out graphically in the Bode
diagram.
If the time delay is small, it is possible to approximate the time delay
transfer function relation by a truncated power series; that is,

-Ts+ Tjs*
e~
T" = 1 d (7-121)
2! 3!

Figure 7-32 shows the Nyquist plots of the transfer function of Eq. (7-119)

/ Im GH

G(s)//(i)-plane

KeGH

Two-term
approximation

Fig. 7-32. Approximation of Nyquist plot of system with time delay by

truncated power-series expansion.
 Sec. 7.11 Stability of Nonlinear Systems— Popov's Criterion / 363

with Td = 0.8 sec, and

G(s)H(s) = l-7> (7-122)
s(s + l)(s + 2)

which is the result of truncating the series of Eq. (7-121) after two terms

7.11 Stability of Nonlinear Systems— Popov's Criterion 16

was mentioned in Section 7.1 that the stability analysis of nonlinear

It

systems is a complex subject, and unlike the linear time-invariant case, no single
method can be applied to all nonlinear systems. Although the major emphasis
of this book is on linear systems, we shall show in the following that a certain
class of nonlinear systems can be studied with stability criteria that are quite
similar to the Nyquist criterion.
When a system is nonlinear, it does not have a characteristic equation,
and therefore there are no eigenvalues to speak of. The Routh-Hurwitz criterion
becomes useless in this case. The kind of stability that we are concerned with
here is asymptotic stability, which means that the equilibrium state x e is as-
ymptotically stable if every motion starting at an initial state x will converge
to x e as time approaches infinity.

Popov's stability criterion applies to a closed-loop control system that con-

tains one nonlinear element, as shown in Fig. 7-33. The nonlinearity is described

Fig. 7-33. Nonlinear closed-loop control system.

by a functional relation that must lie in the first and third quadrants, as shown
in Fig. 7-34.
Many nonlinear control systems in practice can be modeled by the block
diagram and nonlinear characteristic of Figs. 7-33 and 7-34. For instance, Fig
7-35 illustrates several common-type nonlinearities, which are encountered
often in control systems that fit the characteristic specified by Fig. 7-34.
Popov's stability theorem is based on the following assumptions:

1. The is described by the transfer function

linear part of the system
G(s), which has more poles than zeros, and there are no cancel-
lations of poles and zeros.
2. The nonlinear characteristic is bound by k and k 2 as shown in Fig.x

7-34; or
k x
< N(e) < k z (7-123)
 Me)

Fig. 7-34. Nonlinear characteristics.

N(e) 1%)
/' 2

/
k, =0 / k x
=0
_z_
-D /

/ i

/
/
/
/
/
(b) Relay with deadzone

/
7 k, =0

(c) Saturation (d) Saturation with dead zone

Fig. 7-35. Several common nonlinearities encountered in control sys-

tems, (a) Ideal relay, (b) Relay with dead zone, (c) Saturation. Id) Satura-
tion with dead zone.

364
Sec. 7.11 Stability of Nonlinear Systems — Popov's Criterion / 365

The theorem is stated in the following without giving the proof: The closed-
loop system asymptotically stable if the Nyquist plot of G(jco) does not inter-
is

sect or enclose the circle that is described by

k2 *."
x i *i ~t~
y
2
= (7-124)
_ LK^Ki

where x and y denote the real and imaginary coordinates of the G(ja>)-
plane, respectively.
It should be noted that the Popov stability criterion is sufficient but is not
necessary. In other words, if the G(jco) locus does not intersect or enclose the
circle, the system is stable, although the criterion is usually overly conservative.
On the other hand, if the above condition is violated, it does not necessarily
mean that the system is unstable. The possible implications are illustrated in
Fig. 7-36.

i / Im G

G(/oj)-plane

ReG

Fig. 7-36. Interpretation of Popov's stability criterion.

It is interesting to observe that if the system is linear, k = k 2 = k, the s

circle degenerates to the (— 1 , y'O) point, and Popov's criterion becomes Nyquist's
criterion. Figure 7-35 shows that a great majority of the nonlinearities in practice
are with A:, = 0. In these cases, the circle of Eq. (7-124) becomes a straight line

x =— 1
(7-125)

For stability, the Nyquist locus of G(jco) must not intersect this line.
366 / Stability of Control Systems Chap. 7

The following example serves to illustrate the practical application of the

Popov stability criterion.

Example 7-10 Consider that the block diagram shown in Fig. 7-37 represents a
feedback control system that has a saturation element in the forward
path. The system could be a position control system using a motor
as an actuator. The saturation is in the power amplifier of the motor controller, with
K representing the gain of the linear portion of the amplifier characteristic.

-1 K

Fig. 7-37. Nonlinear system with a saturation characteristic.

The linear transfer function of the system is

1
G(s) = (7-126)
s(s + 1)0 + 2)

It can be shown that if the saturation characteristic were absent, the closed-loop

system would be stable for <K

< 6. It is desired to determine the value of so K
that the nonlinear system is assured of stability. Notice that the Popov criterion is of
such a nature that for the saturation nonlinearity, the result of the criterion is inde-
pendent of the level of saturation, which is in a way a qualitativeness of the test.
The Nyquist locus of GO' CO) is sketched as shown in Fig. 7-38. For stability of the
closed-loop system, the G(joi) locus must not intersect the vertical line, which passes
through the point (—1/K,j0).
The maximum value of K is determined by finding the maximum magnitude of the
real part of G(jco). Rationalizing G(yco), the real part is written

Re GU<o) = 9co2+{2 _ (0 2 2
(7-127)
)

The frequency at which Re G(Jco) is a maximum is determined from

d Re G(jco) _ <a[-18 + 4(2 - co 2

)] _ n (7-128)
dm ~~ [9co 2
+ (2 - o> 2 2 2
) ]

which gives (O = 0. Thus

Max. Re G(j<x>) = -f (7-129)

and the closed-loop system is stable for

K<§ (7-130)

which is considerably less than the critical value of K = 6 for the linear system.
Chap. 7 References / 367

/'
Im G

ReC

Fig. 7-38. Application of the Popov stability criterion to the nonlinear

system in Fig. 7-37.

REFERENCES

Routh-Hurwitz Criterion

1. E. J. Routh, Dynamics of a System of Rigid Bodies, Chap. 6, Part II, Macmillan

& Co. Ltd., London, 1905.

2. N. N. Puri and C. N. Weygandt, "Second Method of Liapunov and Routh's

Canonical Form,"/. Franklin Inst., Vol. 76, pp. 365-384, Nov. 1963.

3. G. V. S. S. Raju, "The Routh Canonical Form," IEEE Trans. Automatic Con-

trol, Vol. AC-12, pp. 463-464, Aug. 1967.

4. V. Krishnamurthi, "Correlation Between Routh's Stability Criterion and

Relative Stability of Linear Systems," IEEE Trans. Automatic Control, Vol.
AC-17, pp. 144-145, Feb. 1972.

5. V. Krishnamurthi, "Gain Margin of Conditionally Stable Systems from Routh's

Stability Criterion," IEEE Trans. Automatic Control, Vol. AC-17, pp. 551-552,
Aug. 1972.
 —

368 / Stability or Control Systems Chap. 7

General Linear Stability —Nyquist

6. H. Nyquist, "Regeneration Theory," Bell System Tech. J., Vol. 11, pp. 126-147,
Jan. 1932.

7. C. H. Hoffman, "How To Check Linear Systems Stability : I. Solving the Charac-

teristic Equation," Control Engineering, pp. 75-80, Aug. 1964.

8. C. H. Hoffman, "How To Check Linear Systems Stability: II. Locating the

Roots by Algebra," Control Engineering, pp. 84-88, Feb. 1965
9. C. H. Hoffman, "How To Check Linear Systems Stability: III. Locating the
Roots Graphically," Control Engineering, pp. 71-78, June 1965.
10. M. R. Stoji6 and D. D. Siuak, "Generalization of Hurwitz, Nyquist, and
Mikhailov Stability Criteria," IEEE Trans. Automatic Control, Vol. AC-10, pp.
250-254, July 1965.

11. R. W. Brockett and J. L. Willems, "Frequency Domain Stability Criteria

Part I," IEEE Trans. Automatic Control, Vol. AC-10, pp. 255-261, July 1965.
12. R. W. Brockett and J. L. Willems, "Frequency Domain Stability Criteria-
Part II,"IEEE Trans. Automatic Control, Vol. AC-10, pp. 407-413, Oct. 1965.
13. D. D. Siljak, "A Note on the Generalized Nyquist Criterion," IEEE Trans.
Automatic Control, Vol. AC-11, p. 317, April 1966.

14. L. Eisenberg, "Stability of Linear Systems with Transport Lag," IEEE Trans.
Automatic Control, Vol. AC-U, pp. 247-254, April 1966.

15. T. R. Natesan, "A Supplement to the Note on the Generalized Nyquist Crite-
rion," IEEE Trans. Automatic Control, Vol. AC-12, pp. 215-216, Apr. 1967.

Popov's Criterion

16. V. M. Popov, "Absolute Stability of Nonlinear Systems of Automatic Control,"

Automation and Remote Control, Vol. 22, pp. 961-978, Aug. 1961.

17. Z. V. Rekasuis, "A Stability Criterion for Feedback Systems with One Nonlinear
Element," IEEE Trans. Automatic Control, Vol. AC-9, pp. 46-50, Jan. 1964.
18. C. A. Desoer, "A Generalization of the Popov Criterion," IEEE Trans. Auto-
matic Control, Vol. AC-10, pp. 182-185, Apr. 1965.

19. S. C. Pincura, "On the Inapplicability of the Popov Stability Criterion in Certain
Classes of Control Systems," IEEE Trans. Automatic Control, Vol. AC-12, pp.
465-466, Aug. 1967.

20. Y. S. Cho and K. S. Narendia, "An Off-Axis Circle Criterion for the Stability of
Feedback Systems with a Monotonic Nonlinearity," IEEE Trans. Automatic
Control, Vol. AC-13, No. 4, pp. 413-416, Aug. 1968.

21. J. B. Moore, "A Circle Criterion Generalization for Relative Stability," IEEE
Trans. Automatic Control, Vol. AC-13, No. 1, pp. 127-128, Feb. 1968.

22. C. A. Desoer, "An Extension to the Circle Criterion," IEEE Trans. Automatic
Control, Vol. AC-13, pp. 587-588, Oct. 1968.

23. E. Y. Shapiro, "Circle Criterion for Nonlinear System Stability," Control

Engineering, Vol. 17, No. 3, pp. 81-83, Mar. 1970.
Chap. 7 Problems / 369

24. A. C. Tsoi and H. M. Power, "Equivalent Predictions of the Circle Criterion and
an Optimum Quadratic Form for a Second-Order System," IEEE Trans. Auto-
matic Control, Vol. AC-17, pp. 565-566, Aug. 1972.

25. C. E. Zimmerman and G. J. Thaler, "Application of the Popov Criterion to

Design of Nonlinear Systems," IEEE Trans. Automatic Control, Vol. AC-16,
pp. 76-79, Feb. 1971.

26. H. M. Power and A. C. Tsoi, "Improving the Predictions of the Circle Criterion
by Combining Quadratic Forms," IEEE Trans. Automatic Control, Vol. AC-18,
pp. 65-67, Feb. 1973.

PROBLEMS

7.1. By means of the Routh-Hurwitz criterion, determine the stability of systems

that have the following characteristic equations. In each case, determine the
number of roots of the equation which are in the right-half s-plane.
(a) s 3 + 20s + 9s + 100 =
2

(b) s + 20s + 9s + 200 =

3 1

(c) 3s* + 10s + 5s + s + 2 =

3 2

(d) s* + 2s + 6s 2 + 8s + 8 =
3

(e) s
6
+ 2s + 8s 4 + 12s + 20s 2 +
5 3
16s + 16 =
7.2. The characteristic equations for certain feedback control systems are given
below. In each case, determine the values of K that correspond to a stable
system.
(a) s* + + \0s 2 +2s + K =
22s 3
(b) s* + 20Ks 3+ 5s 2 + (10 + K)s + 15 =
(c) s3 + (K + 0.5)s 2 + AKs + 50 =
7.3. The conventional Routh-Hurwitz criterion gives only the location of the roots
of a polynomial with respect to the right half and the left half of the s-plane.
The open-loop transfer function of a unity feedback control system is given as

G(S) = s(l+7s)
It is desired that all the roots of the system's characteristic equation lie in the
region to the left of the line s = — a. This will assure not only that a stable
system is obtained, but also that the system has a minimum amount of damping.

Extend the Routh-Hurwitz criterion to this case, and determine the values of K
and T required so that there are no roots to the right of the line s = —a.
7.4. The loop transfer function of a feedback control system is given by

c ^^- s(i+ (

rsxi
1)
h2s)
The parameters K and Tmay be represented in a plane with K as the horizontal
axis and T as the vertical axis. Determine the region in which the closed-loop
system is stable.

7.5. The open-loop transfer function of a unity feedback control system is given by

K{s
GM =y '
s 3 {s
+ 5X* + 40)
+ 200)(s + 1000)
 : :

370 / Stability of Control Systems Chap. 7

Discuss the stability of the closed-loop system as a function of K. Determine

the values of K that will cause sustained oscillations in the closed-loop system.
What are the frequencies of oscillations ?

7.6. A controlled process is represented by the following state equations

X x
= Xi — 3X2
x2 = 5X\ + u
The control is obtained from state feedback such that

" = ^1*1 + glXl

where g t
and gz are real constants. Determine the region in the gz versus ^i
plane in which the overall system is stable.

7.7. Given a linear time-invariant system that is described by the following state
equations
±(t) = Ax(?) + B«(?)
where
1 (T -o
A= 1 B =
.0 -3 -2 _1

The closed-loop system is implemented by state feedback so that

u(t) = -Gx(t)
where G is the feedback matrix, G = [gi gz gi\ with g u g 2 and g 3 equal
, to
real constants. Determine the constraints on the elements of G so that the
overall system is asymptotically stable.

7.8. Given the system i = Ax + B«, where

1 o on 2
r
A= -2 B =
3_ _1

Consider that state feedback may be implemented. Is the system stabilizable by

state feedback?

7.9. For the following loop gain functions, sketch the Nyquist diagrams that
correspond to the entire Nyquist path. In each case check the values of N, P,
Z
and with respect to the origin in the Gfl-plane. Determine the values of N, P,
and Z with respect to the —1 point, and determine if the closed-loop system is
stable. Specify in which case it is necessary to sketch only the Nyquist plot for
Co = to oo (section 1) on the Nyquist path to investigate the stability of the
closed-loop system.

= s{l
(a) G(s)H(s)
+ Q J)(1 + Q2s)
(b) G(s)H(s) = j2(1
+ 25s){l + Q5s)
( \
<CJ r<\n<\
U(S)H(S) - 100(1 + s)
j(1 + Q ]i)(1 + a5j)(1 + Q8s)
(d) G(s)H(s) = 5(1 + oa7 5-
5(
1 0.25s)
)(
Chap. 7 Problems / 371

10
(e) G(s)H(s) =
j(1 + 02s)(s - 1)

(f) G(s)H(s) =
ffi+ffi
7.10. Sketch Nyquist diagrams for the following loop transfer functions. Sketch only
the portion that is necessary to determine the stability of the closed-loop system.
Determine the stability of the systems.

= 100
(a) G(s)H(s)
s(s 2 +2s + 2)0 1)
50
(b) G(s)H(s) =
s(s + 2)(s 2 4)
s
(c) G(s)H(s)
1 - 0.2s
7.11. Figure P7-1 1 shows the entire Nyquist plots of the loop gains G(s)H(s) of some
feedback control systems. It is known that in each case, the zeros of G(s)H(s)

, . / Im

oo = -0
cj = +

w Arrows on diagram
correspond to defined
sense of the Nyquist
path

"".^ G//-plane

j-«- Re
372 / Stability of Control Systems Chap. 7

i /Im
+
co = - """ ~~ "•» ^^

X
.
(///-plane

/?-*«» \

/
W *Ju
(
/ = -=»
\
|

Re
(-1./0) /**' co =+ °° 1

c /
/
/
'
+
CO= +
(c)

Figure P7-11 (Cont.).

are all located in the left half of the s-plane; that is, Z= with respect to the
origin in the GH-plane. Determine the number of poles of G(s)H(s) that are in
the right half of the s-plane. State the stability of the open-loop systems. State
whether the closed-loop system is stable ; if not, give the number of roots of the
characteristic equation that are in the right half of the s-plane.

7.12. The characteristic equation of a feedback control system is given by

s 3
+ 5Ks 2
+ (2K + 3)s + 10 =
Apply the Nyquist criterion to determine the values of K for a stable closed-
loop system. Check the answer by means of the Routh-Hurwitz criterion.
7.13. The Nyquist criterion was originally devised to investigate the absolute stability
of a closed-loop system. By
sketching the Nyquist plot of G(s)H(s) that corre-
sponds to the Nyquist path, it is possible to tell whether the system's charac-
teristic equation has roots in the right half of the i-plane.

(a) Define a new Nyquist path in the j-plane that may be used to ensure that

all the complex roots of the characteristic equation have damping ratios

greater than a value £i-

(b) Define a new Nyquist path in the .y-plane that may be used to ensure that
all the characteristic equation roots are in the left half of the i-plane with
real parts greater than a t .

7.14. The block diagram of a feedback control system is shown in Fig. P7-14.
(a) Determine the values of a so that the system is stable.

R(s) *- as)

Figure P7-14.
Chap. 7 Problems / 373

(b) Determine the values of a so that the eigenvalues of the system all lie to the
left of the Re(s) = — 1 line in the s-plane.

7.15. The block diagram of a multivariate control system is shown in Fig. P7-15.

*,(*) *-C. (s)

R 2 (s) *~C 2 (s)

Figure P7-15

The transfer function G(s) is given by

K
(j + IX* + 2)

and K is a positive constant. Determine the range of K so that the system is

asymptotically stable.
(a) Use the Routh-Hurwitz criterion.
(b) Use the Nyquist criterion.
7.16. Figure P7-1 6 shows the block diagram of a control system in which the amplifier

N
e u c
Amplifier G(s)

Amplifier
characteristic

G(s)-
s(l +0.1 ,s)(l +0.2s)

Figure P7-16.
374 / Stability of Control Systems Chap. 7

has a nonlinear saturation characteristic. Determine the maximum value of K

such that the overall system will be absolutely stable.

7.17. Figure P7-17 shows a control system that has a synchro control transformer as
an error transducer. The output of the synchro is proportional to the sine of the
input shaft position. Assuming that the synchro operation is limited to
—n/1 <9 < nil, the input-output characteristic of the device may be repre-
e

sented by the sine wave shown in Fig. P7-17. Determine the limiting value of
K so that the system is absolutely stable.

Synchro
Y%
J
control
transformer
u
G(s)

G(s) =
K
s(\ +0.5s)(l +.s)

Synchro
10 characteristic

Figure P7-17.
 8
Root Locus Techniques

8.1 Introduction

In the design of control systems it is often necessary to investigate the perfor-

mance of a system when one or more parameters of the system varies over
a given range. Since the characteristic equation plays an important role in the
dynamic behavior of linear systems, an important problem in linear control
systems theory is the investigation of the trajectories of the roots of the charac-
teristic equation, or simply, the root loci, when a certain system parameter
varies. In fact, the examples of Chapter 6 already have illustrated the importance
of the root loci in the study of linear control systems.
The root locus technique is not confined to the inclusive study of control
systems. The equation under investigation does not necessarily have to be the
characteristic equation of a linear system. The technique can also be used to
assist in the determination of roots of high-order algebraic equations.
In general, the root locus problem for one variable parameter can be defined
by referring to equations of the following form

F(s) = s" + ai s"-> +... + a n _ lS + an

where K is the parameter considered to vary between — oo and oo. The coeffi-
cients «!,..., a„, b b m are assumed to be fixed. These coefficients
x ,
... , bm _ t ,

can be real or complex, although our main interest here is in real coefficients.
The root locus problem of multiple-variable parameters will also be considered
in this chapter.
Although the loci of the roots of Eq. (8-1) when K varies between -co and

375
376 / Root Locus Techniques Chap. 8

oo are generally referred to as the root loci in control system literature, the follow-
ing categories are defined

1. Root loci: the portion of the root loci when A' assumes positive
values; that is, < K< oo.

2. Complementary root loci: the portion of the root loci when K

assumes negative values; that is, — oo < K< 0.

3. Root contours: loci of roots when more than one parameter varies.

The complete root loci refers to the combination of the root loci and the
complementary root loci.

8.2 Basic Conditions of the Root Loci

Since our main interest is in control systems, let us consider that Eq. (8-1) is

the characteristic equation of a linear control system that has the closed-loop
transfer function
C(s)
~~
_ G(s)
v
,„ ,
2>
R(s) 1 + G(s)H(s)

The characteristic equation is obtained by setting the numerator of 1 +

G(s)H{s) to zero. In other words, the left side of Eq. (8-1) should correspond to
the numerator of 1 + G(s)H(s). Or, the roots of the characteristic equation must
also satisfy
1 + G(s)H(s) = (8-3)

The reader should have a special understanding of the relationship between

the characteristic equation, the equation 1 G(s)H(s) = 0, and the function +
G(s)H(s). Although these relationships are very simple to understand, one can
easily overlook these important concepts when deeply entangled in the intri-

cacies of the construction of the root loci.

If we divide both sides of Eq. (8-1) by the terms that do not contain K, we
get
, ,
K(s m + 6, J""' + ... + b m . s + b m = i )
(S
»
^ s- + «,*»-» + ... + «„_,* + a. '

Comparing Eqs. (8-3) and (8-4) we see that the following relation can be estab-
lished :

w w = K(s» ++ b^+...
G(s)H(s)
+
+ b m + bJ
+ a„_i5 + a
s" aiS"
l
. .
- lS

where G(s)H(s) is known as the loop transfer function of the control system.
Since we have mentioned that the root locus technique is not limited to control
systems, in general, given Eq. (8-1), we can regard G(s)H(s) of Eq. (8-5) as the
loop transfer function of an equivalent control system. The control system
is equivalent in the sense that it has Eq. (8-1) as its characteristic equation.
Later we shall discover that the step in obtaining Eq. (8-4) from Eq. (8-1) is a

very useful one in many other analysis and design situations. For lack of a better
Sec. 8.2 Basic Conditions of the Root Loci / 377

name, we shall refer to the procedure of dividing both sides of the characteristic
equation by the terms that do not contain K as the Golden Rule.
We can now define the complete root loci as the loci of the points in the
s-plane that satisfy Eq. (8-3) as K is varied between and oo — oo
Now we are ready to establish the conditions under which Eq. (8-3) is
satisfied. Let us express G(s)H(s) as

G(s)H(s) = KG^H^s) (8-6)

where G^H^s) no longer contains the variable parameter A'. Then Eq. (8-3) is
written

G^H^s)^-^ (8-7)

To satisfy this equation, the following conditions must be met simultaneously

| (?,(*)#,(*) |
= pL -co < tf < oo (8-8)

= (2k +
/G,(j)ff,(j) l)w K> (8-9)

/GiWH^s) = 2kn K<0 (8-10)

where k = 0, ±1, ±2, . . . (all integers).

In practice, the complete root loci are constructed by finding all points in
the s-plane that satisfy Eqs. (8-9) and (8-10), and then the values of K along the
loci are determined using Eq. (8-8).
The construction of the root loci is basically a graphical problem, although
some of the rules of construction are arrived at analytically. The starting point
of the graphical construction of the root loci is based on knowledge of the poles
and zeros of the function G(s)H(s). In other words, Eq. (8-5) must first be
written*

W W -~ K(s++
G(*\H(*\
Pl
(s + z *)
Xs+p
J + z ")
Z 'X J
)...(s+ Pn 2
• • • (

) (8-11)
= KG^H^s)
where the poles and zeros of G(s)H(s) are real or complex-conjugate numbers.
Using Eq. (8-11), the conditions stated in Eqs. (8-8), (8-9), and (8-10)
become
m

\G (s)H {s)\=±i
l 1
= rlT -co<tf<co (8-12)
I*

and
/GMH^s) = H +z
'=i
ls i
-^ ls+
j=i
Pj
(8-13)
= (2k + l)7i <K< oo

We shall first consider that G(s)H(s) is a rational function of s. For systems with time
delays, G(s)H(s) will contain exponential terms such as e~ Ts .
378 / Root Locus Techniques Chap. 8

/G^HM = 2 + - S — i=i
/s z,
j-
/s
1
(8-14)
= 2kn -oo<K<0
for A: = 0, ±1, ±2,
It was mentioned earlier that Eqs. (8-13) and (8-14) may be used for the
construction of the complete root loci in the s-plane. In other words, Eq. (8-13)
implies that for any positive value of AT, a point (e.g., Si) in the s-plane is a point
on the root loci if the difference between the sums of the angles of the vectors
drawn from the zeros and the poles of G(s)H(s) to s, is an odd multiple of
180°. Similarly, for negative values of A", Eq. (8-14) shows that any point on the
complementary root loci must satisfy the condition that the difference between
the sums of the angles of the vectors drawn from the zeros and the poles to
the point is an even multiple of 180°, or 0°.

To illustrate the use of Eqs. (8-13) and (8-14) for the construction of the
root loci, let us consider

= K(s + z ) t
G(s)H(s) (8-15)
s(s + PiKs + Pi)
The locations of the poles and the zero of G(s)H(s) are arbitrarily assumed, as
shown in Fig. 8rl. Next, we select an arbitrary point, s in the .y-plane and draw t ,

vectors directing from the poles and the zero of G(s)H(s) to the point j,. If j,

s-plane

Fig. 8-1. Pole-zero configuration of G(s)H(s) = [K(s + zi)]/[s(s + p 2)

U+p 3 )].
 :

Sec. 8.2 Basic Conditions of the Root Loci / 379

is indeed a point on the root loci (0 <K< oo) [remember that the root loci
represent the loci of zeros of 1 + G(s)H(s)], it must satisfy the following two
conditions simultaneously: from Eq. (8-12),

.\_?j_±IiL L (8-16)
|jllkl+/»2ll*l+/»j| l*|
and from Eq. (8-13),

ISj + z - (/?, + /*, + p + M + Pz = (2fc +

t 2 ) 1)tt fc = 0, ±1,±2,...
(8-17)

Similarly, if s t is to be a point on the complementary root loci (— oo <K

< 0), it must satisfy Eq. (8-14); that is,

/j + zi - (/£i. + M+ J?2 + /Ji + ;>, ) = Ikn (8-18)

for& = 0, ±1, ±2,

As shown in Fig. 8-1, the angles ri , 0,,, d P2 and, 0,,, are the angles of the
vectors measured with the positive real axis as zero reference. Equations (8-17)
and (8-18) become

tl
- (0 Pl +e +d p! P3 )
= (2k + l)n 0<K<oo (8-19)
and
0.i
~ (K + 6„ + d pi ) = 2kn -co<K<0 (8-20)

respectively.
If s t is found to satisfy either Eq. (8-19) or Eq. (8-20), Eq. (8-16) is used to
determine the value of AT at the point. Rewriting Eq. (8-16), we have

|ff
I _ kilbi + Pi\\s, + p 3 \

(8
_ 21
)
\
s + zi i I

where the factor |

^x + zx |
is the length of the vector drawn from the zero z± to
the point s^ If, as in Fig. 8-1, the vector lengths are represented by A, B, C, and
D, Eq. (8-21) becomes

\K\ = B^- (8-22)

The depends on whether Sx is on the root loci or the

sign of K, of course,
complementary root Consequently, given the pole-zero configuration of
loci.

G(s)H(s), the construction of the complete root locus diagram involves the
following two steps

1. A search for all the *, points in the j-plane that satisfy Eqs. (8-9)
and (8-10).
2. The determination of the values of K at points on the root loci and
the complementary root loci by use of Eq. (8-8).

From the basic principles of the root locus plot discussed thus far, it may
seem that the search for all the s x points in the s-plane that satisfy Eqs. (8-9)
380 / Root Locus Techniques Chap. 8

and (8-10) is a very tedious task. However, aided with the properties of the root
loci that we are going to assemble in the next section, the actual sketching of
the root locus diagram in most cases is not so formidably complex. Normally,
with some experience on the part of the analyst, the root loci can be sketched by
following through the "rules" of construction. A special graphical aid, called
the Spirule, can also be used to help plot the root locus diagram. However,
since the Spirule is simply a device that assists in adding and subtracting angles
of vectors quickly, according to Eq. (8-13) or Eq. (8-14), it can be used effectively
only if we already know the general proximity of the roots. Indeed, when we set
out to find a point s on the root x
loci, we must first have some general knowledge
we select a trial point and test it in Eqs. (8-13)
of the location of this point; then
and not close to any point on the root loci, the search
(8-14). If the trial point is
procedure can be frustrating, even with the aid of a Spirule.
"
Digital and analog computer programs 32 34 can be prepared for the plotting
~ 29
of the root locus diagram. Analytical procedures 26 have also been developed
for obtaining the root loci. In fact, a computer or the analytical method must
be used if the poles and zeros of G(s)H(s) are not known a priori.

The material presented in this chapter will emphasize the principle of con-
struction of the root loci, since one must obtain a thorough understanding of
the basic principles before any engineering implementation or numerical
methods can be applied successfully.

8.3 Construction of the Complete Root Loci

The following rules of construction are developed from the relation between
the poles and zeros of G(s)H(s) and the zeros of 1 + G(s)H(s). These rules should
be regarded only as an aid to the construction of the root loci and the comple-
mentary root loci, as they do not give the exact plots.

K= Points

Theorem 8-1. The K= points on the complete root loci are at the poles
ofG(s)H(s).

Proof: From Eq. (8-12),

m

|
GMH^s) = I
^ = r^ (8-23)

J=i

As K approaches zero, the value of Eq. (8-23) approaches infinity, and,

correspondingly, s approaches the poles of G^H^s) or of G(s)H(s); that is,

s approaches —p s (j = 1, 2, . . . , n). It is apparent that this property applies

to both the root loci and the complementary root loci, since the sign of K has
no bearing in Eq. (8-23).
Sec. 8.3 Construction of the Complete Root Loci / 381

Example 8-1 Consider the following equation

s(s + 2)(j + 3) + K(s + 1) = (8-24)

When K = 0, the three roots of the equation are at s = 0, j = —2, and s = —3. These
three points are also the poles of the function G(s)H (s) if we divide both sides of Eq.
(8-24) by the terms that do not contain AT (the Golden Rule) and establish the relation-
ship

1 + G(s)H(s) = 1 +
K(s + 1)
= (8-25)
s(s + 2)C$ + 3)
Thus

G(s)H(s)
K(s + 1)
(8-26)
s(s + 2)(j + 3)

The three K— points on the complete root loci are as shown in Fig. 8-2.

.s-plane
;cj

K= Q K=Q K=

Fig. 8-2. Points at which K= on the complete root loci of s(s + 2)

(s + 3) + K(s + 1) = 0.

K = ±oo Points

Theorem 8-2. The K = ± oo points on the complete root loci are at the
zeros of G{s)H(s).

Proof: Referring again to Eq. (8-23), as K approaches ±co, the equation

approaches zero. This corresponds to i approaching the zeros of G{s)H{s); or
'»' —
s approaching — z
* {i=
t
"> ™^
1,2, ...,m). '

Example 8-2 Consider again the equation

+ 2)(s + 3) + K(s + 1) =
s(s (8-27)

It apparent that when K is very large, the equation can be approximated by

is

K(s+l) = (8-28)

which has the root s = -1. Notice that this is also the zero of G(s)H(s) in Eq. (8-26).
Therefore, Fig. 8-3 shows the point j = —1 at which K = ±oo. However, G(s)H(s) in
this case also has two other zeros located at infinity, because for a rational function,
 : ,

382 / Root Locus Techniques Chap. 8

/to
s-plane

A>±°°
-2

Two other K = ± °°
points at infinity

Fig. 8-3. Points at which K= ± <*> on the complete root loci of s(s + 2)
(.s + 3) + K(s + 1) = 0.

the total number of poles and zeros must be equal if the poles and zeros at infinity are

included. Therefore, for the equation in Eq. (8-27), the K = ± co points are at j = — 1

co, and co.

Number of Branches on the Complete Root Loci

A branch of the complete root loci is the locus of one root when K takes on
values between — oo and oo. Since the number of branches of the complete root
loci must equal the number of roots of the equation, the following theorem
results

Theorem 8-3. The number of branches of the root loci of Eq. (8-1) is equal to
the greater of n and m.

Example 8-3 The number of branches on the complete root loci of

+ 2)(s + 3) + Kis + 1) =
sis (8-29)

is three, since n = 3 and m = 1. Or, since the equation is of the third order in s, it

must have three roots, and therefore three root loci.

Symmetry of the Complete Root Loci

Theorem 8-4. The complete root loci are symmetrical with respect to the
real axis of the s-plane. In general, the complete root loci are symmetrical with
respect to the axes of symmetry of the poles and zeros ofG(s)H(s).

Proof: The proof of the first statement is self-evident, since, for real coeffi-
cients in Eq. (8-1), the roots must be real or in complex-conjugate pairs.
The reasoning behind the second statement is also simple, since if the poles
and zeros of G(s)H(s) are symmetrical to an axis other than the real axis in the
j-plane, we can regard this axis of symmetry as if it were the real axis of a new
complex plane obtained through a linear transformation.
Sec. 8.3 Construction of the Complete Root Loci / 383

Example 8-4 Let us consider the equation

s(s + 1)(* + 2) +K= (8-30)

Dividing both sides of the equation by the terms that do not contain K leads to

G(s)H(s) = K (8-31)
s(s + l)(.s + 2)

The complete root loci of Eq. (8-30) are sketched as shown in Fig. 8-4. Notice that
since the poles of G(s)H(s) are symmetrical with respect to the s = — 1 axis (in addition
to being always symmetrical with respect to the real axis), the complete root loci are
symmetrical to the j = —1 axis and the real axis.

.s-plane
\ \

\ K<0

* — < +-» x» \
* )!

2 1-1
/

I
I
I
I
/

/ K<0

A Axis of
symmetry

Fig. 8-4. Root loci of s(s + l)(s + 2) +K= 0, showing the properties
of symmetry.

Example 8-5 When the pole-zero configuration of G(s)H(s) is symmetrical with

respect to a point in the j-plane, the complete root loci will also be
symmetrical to that point. This is illustrated by the root locus plot of

s(s + 2)(.s + +/X* +

1 1 ~j) +K= (8-32)

as shown in Fig. 8-5.

384 / Root Locus Techniques Chap. 8

s-plane

Fig. 8-5. Root loci of s(s + 2)(s + 1 +y")0 + 1 -y) + A: == 0, showing

the properties of symmetry.

Asymptotes of the Complete Root Loci (Behavior of Root Loci at s = co)

The properties of the complete root loci near infinity in the j-plane are
important, since when n^tn, 2\n — m\ of the loci will approach infinity in

the j-plane.

Theorem 8-5. For large values ofs, the root loci for K>0 are asymptotic to
straight lines or asymptotes with angles given by

ek = (?K±Dz (8-33)
n — m

where k = 0, 1, 2, ,\n . . .
—
m\ — 1* and n and m are defined in Eq. (8-1).
For the complementary root loci, 0, the angles of the asymptotes are K<
2kn (8-34)
9,
n —m
where k = 0, 1, 2, . . . ,\n — m\ — 1.

According to the defining equations of the root loci, k = 0, ±1, ±2, However,
since there are only \n — m\ asymptotes for each type of root loci, we need to assign only

|
n —m |
values to k.
Sec. 8.3 Construction of the Complete Root Loci / 385

Equations (8-33) and (8-34) imply that the asymptotes of the complementary
root loci are linear extensions of the asymptotes of the root loci, and vice versa.
When the asymptotes of one type of root loci are determined, those of the other
type are found without further calculation.

Proof: Let us divide both sides of Eq. (8-1) by

s^ + b.s"- 1
+ ... + bm . 1
s + bm
We then have
S" + «!?" ... + a„-iJ + a.
=
j" + 6,5—' + ... + b m ^s + b n + K (8-35)

Carrying out the fraction of the left side of Eq. (8-35) by the process of
long division, and for large s neglecting all but the first two terms, we have

s"-
m
+ (a l - l
- -
b )s n m 1
^ -K (8-36)
or
a °'
1 + '

s j
= (-*)"<•- (8-37)

The factor [1 + (a, -b s

)/sy /l
"- m>
in Eq. (8-37) is expanded by binomial expan-
sion, and Eq. (8-37) becomes

1 + (« — m)s
= (-*)
t"vl/(n-m)
(8-38)

Again, if only the first two terms in the last series are retained, we get

n —m v v '

Now let s = a + jco, and, using DeMoivre's algebraic theorem, Eq. (8-39) is
written

o +jco + ai — bi K (2k + l)n (2k + \)n

—m +
.
l
cos
n —m n
sin
n —m (8-40)

for <K< oo, and

I K 1/( "~ m)
I
cos
n
2kn
—m
,

H /
.

sm
.
—2kn
-^—
n — m
(8-41)

for -oo < A:<0, and& = 0, ±1, ±2,

Equating the real and imaginary parts of both sides of Eq. (8-40), we have,
for 0< K< oo,

a + °J ^i ~ Ki/<»-m) cos-(2k + \)n

—m
,

(8-42)
n
and
<a^*' /(
-->sin
(2 * +1) *
n —m (8-43)

Solving for AT I/( "- m) from Eq. (8-43), we have

—b o + a t x

K a> —m n
l

sin
2k+\ —
— n cos 2k—+— n 1
(8-44)

n — m
!

n m
386 / Root Locus Techniques Chap. 8

or

~ tan ( 2k + V* + ZUZ±i) (8-45)

CO
n — m (g
\ n — m I

Equation (8-45) represents a straight line in the j-plane, and the equation is of
the form
co = M{a - a t ) (8-46)

where M represents the slope of the straight line or the asymptote, and a x
is

the intersect with the o axis.

From Eqs. (8-45) and (8-46) we have

M = tan?^iirc
n — m
(8-47)

0, 1, 2, . . .
,
\n — m\ — 1, and
a,
M — b,
(8-48)
m
Note that these properties of the asymptotes are for the root loci (0 < K < oo)
only.
Similarly, from Eq. (8-41) we can show that for the complementary root
loci(-oo < K<G),
M = tan -^-
n — m
(8-49)

k = 0, 1, 2, . . . ,
\n — m\ — 1, and the same expression as in Eq. (8-48) is

obtained for a x
. Therefore, the angular relations for the asymptotes, given by
Eqs. (8-33) and (8-34), have been proved. This proof also provided a by-
product, which is the intersect of the asymptotes with the real axis of the
j-plane, and therefore resulted in the following theorem.

Intersection of the Asymptotes (Centroid)

Theorem 8-6. (a) The intersection of the 2\n m\ asymptotes of the — com-
plete root loci lies on the real axis of the s-plane.
(b) The intersection of the asymptotes is given by

~ = fc
-g '
'
(8-50)
n —m
where a u b u n, and m are defined in Eq. (8-1).

Proof: The proof of (a) is straightforward, since it has been established

that the complete root loci are symmetrical to the real axis.
The proof of (b) is a consequence of Eq. (8-48). Furthermore, if we define
a function G(s)H(s) as in Eq. (8-5), Eq. (8-50) may be written as

b_
xZL aA
n —m
_ £ finite poles of G(s)H(s) - £ finite zeros of G(s)H(s)
~ number of finite poles of G(s)H(s)
(8
' 51
)

— number of finite zeros of G(s)H(s)

 . = : :

Sec. 8.3 Construction of the Complete Root Loci / 387

since from the law of algebra,

—a,= sum of the roots of s" + a,s"' x + . . . + a„_,s + a. =

(8-52)
= sum of the finite poles of G(s)H(s)

—b — t
sum of the roots of s m + b,s m
~ l
+ . . . + b^^s + bm =
(8-53)
= sum of the finite zeros of G(s)H(s)

Since the poles and zeros are either real or complex-conjugate pairs, the imagi-
nary parts always cancel each other. Thus in Eq. (8-51) the terms in the summa-
tions may be replaced by the real parts of the poles and zeros of G(s)H(s),
respectively.
It should be noted that Eq. (8-51) is valid for the root loci as well as the
complementary root loci.

Example 8-6 Consider the equation

s(s + 4)(s 2 + 2s + 2) + K(s + 1) = (8-54)

This equation corresponds to the characteristic equation of a feedback control system

with the loop transfer function

<*'>*«
«+ w VL + 2) <8
- 55 >

The pole-zero configuration of G(s)H(s) is shown in Fig. 8-6. From the six theorems
on the construction of the complete root loci described so far, the following information
concerning the root loci and the complementary root loci of Eq. (8-54) is obtained

1 K= : The K= points on the complete root loci are at the poles of

G(s)H(s): s = -4, s = -1 +j\, and s = -1 -yl.
= 0, s
2. K = ±co The K = ±co points on the complete root loci are at the zeros
:

of G(s)H(s): s = — 1, s = co, j = co, and s = <x>.

3. Since Eq. (8-54) of the fourth order, there are four complete root loci.
is

4. The complete root loci are symmetrical to the real axis.

5. For large values of s, the root loci are asymptotic to straight lines with
angles measured from the real axis Root loci {K 0), Eq. (8-33)
: >
k = 0„ = ^ = 60°

k = l d,=~ = 180°

900°
k = 2 92 =^¥- = 20O°
3

The angles of the asymptotes of the complementary root loci are given
byEq. (8-34):
A: =0 e =^ = o°
k = 1 6^^ = 120

k = 2 =^ =2 24O
388 / Root Locus Techniques Chap. 8

s-plane

Asymptotes of \
complementary root loci \

Fig. 8-6. Asymptotes of the complete root loci of s(s 4)0 2 -r Is + 2)

+ K(s + 1) = 0.

The asymptotes of the complementary root loci may be obtained by extend-

ing the asymptotes of the root loci.

6. The six asymptotes of the complete root loci intersect at

= S finite poles of G(s)H Q) — S finite zeros of G(s)H(s)

<r,

(8-56)
- (o-4-i +yi -l -yi)-(-i) _ _a
4-1 3

The asymptotes are sketched as shown in Fig. 8-6.

Example 8-7 The asymptotes of the complete root loci for several different equations
areshown in Fig. 8-7.

Root Loci on the Real Axis

Theorem 8-7. (d) Root loci: On a given section of the real axis, root loci
{K > 0) may
be found in the section only if the total number of real poles and
zeros of G(s)H(s) to the right of the section is odd.
(b) Complementary root loci: On a given section of the real axis, comple-
mentary root loci (K <
0) may be found in the section only if the total number of
real poles and zeros of G(s)H(s) to the right of the section is even. Alternatively,
 Asymptotes of
/CO
i-plane
s-plane

Asymptote of
root locus

Asymptote of
complementary root locus

"l ="Pi/2 °'l

_~(Pi + P 2 +P3) ,
4
/
/|V I \
\
45

/ \
/ \
/ \
/ \
/
\
I

G(s)H(s) =
K
G(s)H(s) =
s(s +p x
) s(s+p l )(s+p 2 )(s + p i )

Asymptotes of /CO . /"

s-plane \ / s-plane I

'
complementary '

\ root loci /
\ I

\ /
\ |
Asymptotes of Asymptotes of
/
/ \ complementary! root loci
/
/ \ root loci | / \ y
/ \ /
^ Asymptotes y /
of
root loci /
/
i\
\ vX/\ '
/
/\
/ \| / 45°

'(Pi +P2+P3)^*7V^ X o, = -(pi+P2+p 3 )-(-^) ^n\

-. / N 1
45
rc
/
/
/ i \
/ \
/ \
\ / \
/ \
/ \
/
\
I
\

K K(s + zi)
G(s)H(s)= -v— ->
G(s)H(s)
s
2
(s+p )(s+p 2 )(s+ p 3 )
l
i 2(i +p 1 )( J + p 2 )( s + p 3 )

Fig. 8-7. Examples of the asymptotes of root loci.

389
 .

390 / Root Locus Techniques Chap. 8

we can state that complementary root loci will be found in sections on the real axis
not occupied by the root loci.
In all cases the complex poles and zeros ofG{s)H{s) do not affect the existence
properties of the root loci on the real axis.

Proof: The proof of the theorem is based on the following observations:

1 At any point (e.g., s ) on the real axis, the angles of the vectors drawn
t

from the complex-conjugate poles and zeros of G(s)H(s) add up to

be zero. Therefore, the only contribution to the angular relations in
Eqs. (8-13) and (8-14) is from the real poles and zeros of G(s)H(s).

2. Only the and zeros of G(s)H(s) that lie to the right of the
real poles
point St may contribute to Eqs. (8-13) and (8-14), since real poles
and zeros that lie to the left of the point contribute zero degrees.
3. Each real pole of G(s)H(s) to the right of the point j, contributes
— 180° and each zero to the right of the point contributes 180° to
Eqs. (8-13) and (8-14).

The last observation shows that for s to be a point on the root loci, there
t

must be an odd number of poles and zeros of G(s)H(s) to the right of s u and for
Si to be a point on the complementary root loci the total number of poles and

s-plane

Complementary
root loci

Root loci

Fig. 8-8. Properties of root loci on the real axis.

Sec. 8.3 Construction of the Complete Root Loci / 391

zeros of G(s)H(s) to the right of s t must be even. The following example illus-
trates the properties of the complete root loci on the real axis of the s-plane.

Example 8-8 The complete root loci on the real axis in the s-plane are shown in
Fig. 8-8 for two different pole-zero configuraions of G(s)H(s). No-
tice that the entire real axis is occupied by either the root loci or the
complementary root loci.

Angles of Departure (from Poles) and the Angles of Arrival (at Zeros)
of the Complete Root Loci

The angle of departure {arrival) of the complete root locus at a pole (zero)
ofG(s)H(s) denotes the behavior of the root loci near that pole (zero). For the root
loci (K > 0) these angles can be determined by use of Eq. (8-13). For instance, in
the pole-zero configuration of G(s)H(s) given in Fig. 8-9, it is desired to deter-

-<*><- AT

Fig. 8-9. Complete root loci of s(s + 3)0 2 + 2s + 2) + K= to illus-

trate the angles of departure or arrival.
392 / Root Locus Techniques Chap. 8

mine the angle which the root locus leaves the pole at — 1 +jl. Notice that
at
the unknown is measured with respect to the real axis. Let us assume
angle 9 2
that s is a point on the root locus leaving the pole at — 1 + j 1 and is very near
l

the pole. Then s must satisfy Eq. (8-13). Thus

t

/GjsjHjSj) = -(0, + 62+ + 3 0«) = (2k + 1)180° (8-57)

Since s, is very close to the pole at — + j\,1 the angles of the vectors drawn
from the other three poles are determined from Fig. 8-9, and Eq. (8-57) becomes
-(135° + 92 + 90° + 26.6°) = (2k + 1)180° (8-58)

We can simply set k equal to zero, since the same result is obtained for all
other values. Therefore,
92 = -431.6°
which is the same as —71.6°.
When the angle of the root locus at a pole or zero of G(s)H(s) is determined,
the angle of the complementary root loci at the same point differs from this
angle by 180°, since Eq. (8-14) must now be used.

Intersection of the Root Loci with the Imaginary Axis

The points where the complete root loci intersect the imaginary axis of the
s-plane,and the corresponding values of K, may be determined by means of the
Routh-Hurwitz criterion.
For complex situations with multiple intersections, the critical values of
K and co may be more easily determined approximately from the Bode diagram.
Example 8-9 The complete root loci of the equation

s(s + 3)(s 2 +2s + 2) +K= (8-59)

are drawn The root loci intersect the /(W-axis at two conjugate points.
in Fig. 8-9.
Applying the Routh-Hurwitz criterion to Eq. (8-59), we have, by solving the auxiliary
equation, Kc = 8.16 and co c = ±1.095 rad/sec.

Breakaway Points (Saddle Points) on the Complete Root Loci

Breakaway points or saddle points on the root loci of an equation corre-

spond to multiple-order roots of the equation. Figure 8- 10(a) illustrates a case
in which two branches of the root loci meet at the breakaway point on the real
axis and then depart from the axis in opposite directions. In this case the break-
away point represents a double root of the equation to which the root loci
belong. Figure 8- 10(b) shows another common situation where a breakaway
point may occur.
In general a breakaway point may involve more than two root loci. Figure
8-11 illustrates a situation where the breakaway point represents a fourth-order
root.
A root locus diagram can, of course, have more than one breakaway point.
Moreover, the breakaway points need not always be on the real axis. However,
because of the conjugate symmetry of the root loci, the breakaway points must
either be real or in complex-conjugate pairs.
 Sec. 8.3 Construction of the Complete Root Loci / 393

s-plane s-plane

—K K=Q
=
* K O < —e
Breakaway Breakaway
point point

(a) (b)

Fig. 8-10. Examples of breakaway points on the real axis in the s-plane.

/'co

s-plane

Fig. 8-11. Fourth-order breakaway point.

Because of the symmetry of the root loci, it is easy to see that the root loci
in Fig. 8-10(a) and (b) break away at 180° apart, whereas in Fig. 8-11 the four
root loci depart with angles 90° apart. In general, if n root loci (— oo < < oo) K
approach or leave a breakaway point, they must be 180/« degrees apart.
394 / Root Locus Techniques Chap. 8

Several graphical and analytical methods are available for the determination
of the location of the breakaway points. Two analytical methods that seem to
be the most general are presented below.

Method 1.

Theorem 8-8. The breakaway points on the complete root loci of 1 +

KG x
(s)H (s) =x must satisfy

dG^sJH^s) = Q (86Q)

Proof: Let Eq. (8-1) be written

Q(s) + KP(s) = (8-61)

Then Eq. (8-5) may be written

If we consider that A" is

GW" =
varied by an increment
(5)
w AK, Eq. (8-61) becomes
*
(8 62)

Q(s) + (K +AK)P(s) = (8-63)

Dividing both sides of Eq. (8-63) by Q(s) + KP(s), we have

which can be written

1 + AKF(s) = (8-65)
where
S)
F(S) = Q(s)+ KP(s) (*-<>V

Since the denominator of F(s) is the same as the left-hand side of Eq. (8-61), at
points very close to an wth-order root s = s t of Eq. (8-61), which corresponds to
a breakaway point of n loci, F(s) can be approximated by

where A t
is a constant.
™= (F*i? = (0r <
8 - 67 >

Substituting Eq. (8-67) into Eq. (8-65) gives

1+^ = (8-68)

from which we obtain

Taking the limit on both sides of the last equation as AK approaches zero, we
have
lim
Ajc-o
*£ =
As ^=
as
(8-70)
v /

We have shown that at a breakaway point on the root loci, dKjds is zero.*
*The quantity (dsjs)l(dKIK) is defined as the root sensitivity 44 " 6 of an equation with
respect to incremental variation of the parameter K. In this case it is proved that at the break-
away points of the root loci, the roots have infinite sensitivity.
Sec. 8.3 Construction of the Complete Root Loci / 395

Now, since the roots of Eq. (8-61) must also satisfy

1 + KG^H^s) = (8-71)
or

it is simple to see that dK/ds = is equivalent to

ds

It is important to point out that the condition for the breakaway point given

by Eq. (8-73) is necessary but not sufficient. In other words, all breakaway points
must satisfy Eq. (8-73), but not all solutions of Eq. (8-73) are breakaway points.
To be a breakaway point, the solution of Eq. (8-73) must also satisfy Eq. (8-71);
or, Eq. (8-73) must be a factor ofEq. (8-71) for some real K.
In general, the following conclusions can be made with regard to the solu-
tions of Eq. (8-73):

1. All real solutions of Eq. (8-73) are breakaway points on the root
loci (— oo <K< oo), since the entire real axis of the s-plane is

occupied by the complete root loci.

2. The complex-conjugate solutions of Eq. (8-73) are breakaway

points only if they also satisfy Eq. (8-1). This uncertainty does not
cause difficulty in the effective use of Eq. (8-73), since the other
properties of the root loci are usually sufficient to provide informa-
tion on the general proximity of the breakaway points. This infor-
mation may also be helpful in solving for the roots of a high-order
equation, as a result of Eq. (8-60), by trial and error.

The following examples are devised to illustrate the application of Eq.

(8-60) for the determination of breakaway points on the root loci.

Example 8-10 Consider that it is desired to sketch the root loci of the second-order
equation
s(s + 2) + K(s + 4) = (8-74)

Based on some of the theorems on root loci, the root loci of Eq. (8-74) are easily
sketched, asshown in Fig. 8-12. It can be proven that the complex part of the loci is
described by a circle. The two breakaway points are all on the real axis, one between
and —2 and the other between —4 and -co.
When we divide both sides of Eq. (8-74) by s(s + 2) (the Golden Rule), we obtain
the identity

C,(5)ff,(5) = g±^ (8-75)

Therefore, from Eq. (8-60), the breakaway points must satisfy

dG&Wiis) = *fr + 2) - 2(s + l)(s + 4) =

ds s 2 (s + 2) 2 * ;

or
s2 + Ss + 8 = (8-77)

Solving Eq. (8-77), we find that the two breakaway points of the root loci are at s =
396 / Root Locus Techniques Chap. 8

s-plane

Fig. 8-12. Root loci of s(s + 2) + K(s + 4) = 0.

— 1.172 and s = —6.828. Note also that the breakaway points happen to occur all on
the root loci(K > 0).

Example 8-11 Consider the equation

s2 + 2s + 2 + K(s + 2) = (8-78)

The equivalent G(s)H(s) is obtained by dividing both sides of Eq. (8-78) by s 2

+ 2s + 2,

G(s)H(s)
K(s + 2)
(8-79)
s2 +2s + 2

Based on the poles and zeros of G(s)H(s), the complete root loci of Eq. (8-78)
shown in Fig. 8-13. The diagram shows that both the root loci and the
are sketched as
complementary root loci possess a breakaway point. These breakaway points are
determined from
<«iWgi(») ... d (s + 2)
ds ds s 2 + 2s + 2
(8-80)
s 2
+2s + 2-2(s + !)(.? + 2)
=
(s +2s + 2)
2 2
Sec. 8.3 Construction of the Complete Root Loci / 397

, . /CO

x-plane

Fig. 8-13. Root loci of s 2 + 2s + 2 + K(s + 2) = 0.

or
s2 + 4s + 2 = (8-81)

Upon solving Eq. (8-81), the breakaway points are found to be at j = —0.586
and s = —3.414. Notice that in this case s = —3.414 is a breakaway point on the root
loci, whereas s = —0.586 is a breakaway point on the complementary root loci.

Example 8-12 Figure 8-14 shows the complete root loci of the equation

s(s + 4)(s 2 + 4s + 20) +K= (8-82)

Dividing both sides of Eq. (8-82) by the terms that do not contain K, we have

1 + G(s)H(s) = 1 + K = (8-83)
s(s + 4)(s 2 +4s + 20)

Since the poles of G(s)H(s) are symmetrical about the axes a 2 and ca in the =— =
i-plane, the complete root loci of the equation are also symmetrical with respect to
these two axes.
Taking the derivative of Gi(s)Hi(s) with respect to s, we get

ddJsW^s) = 4s 3 + 24s 2 + 12s + 80 _

ds [s(s + 4)(s +4s + 20)] ~ u
2 2 (8 " 84)

or
s3 + 6s 2 + 18s + 20 = (8-85)
398 / Root Locus Techniques Chap. 8

x-plane

Fig. 8-14. Complete root loci of s(s + 4)0 2 + 4s + 20) +K =-- 0.

Because of the symmetry of the poles of G(s)H(s), one of the breakaway points is
easily determined to be at s = —2. The other two breakaway points are found by
solving Eq. (8-85) using this information; they are s = —2 +j2.45 and s = —2 —
j2A5.

Example 8-13 In this example we shall show that the solutions of Eq. (8-60) do not
necessarily represent breakaway points on the root loci.

The complete root loci of the equation

s(s* 2s + 2) + K = (8-86)

are shown in Fig. 8-15; neither the root loci nor the complementary root loci have any
breakaway point in this case. However, writing Eq. (8-86) as

+ KGMH^s) = K
1 1 + s(s 2 + 2s + 2) (8-87)
Sec. 8.3 Construction of the Complete Root Loci / 399

Fig. 8-15. Complete root loci of s(s* + 2s + 2) +K= 0.

and applying Eq. (8-60), we have

1
= (8-88)
ds ds s{s 2 +25 + 2)
which gives
3s 2 + 4s + 2 = (8-89)

The roots of Eq. (8-89) are s = -0.677 +./0.471 and* = -0.677 -yO.471. These two
roots do not represent breakaway points on the root loci, since they do not satisfy Eq.
(8-86) for any real values of K. Another way of stating this is that Eq. (8-89) is not a
factor of Eq. (8-86) for any real K.

Method 2. An algorithm for finding the breakaway points on the root

loci (— co <K< was introduced by Remec. 17 The method is derived from
oo)
the theory of equations, 11 and the proofs of its necessary and sufficient condi-
tions are given in the literature.
 . :

400 / Root Locus Techniques Chap. 8

The breakaway-point algorithm using a tabulation that resembles the

Routh tabulation for stability study is described below:

1. Let the equation for which the root loci are desired be written as

F(s) =A s" + A,s- 1

+ . . . + A n _ lS -+ Am = (8-90)

where A ,A A n _ A„ are constant coefficients and the variable

t ,
. . . , i ,

parameter K is considered to be contained in the coefficients.

2. Obtain the function F'(s) which is the derivative of F(s) with respect
to s. Let F'(s) be of the form

= Bvs"- + B,5"- +
F'(s) + 1 2
. . . B„. 2 s + B„_, = (8-91)

where B = nA B = (n — 1)A U , l
. . , etc.

3. Arrange the coefficients of F(s) and F'(s) in two rows as follows

Ap Al A2 ... An- 1 An
Bo Si B2 ... U„_i

4. Form numbers obtained by the indicated

the following array of
operations. Notice that this not a Routh tabulation, although is

the cross-multiplication operation is the same as in the Routh

tabulation. The illustrated tabulation is shown for a fourth-order
equation, that is, for n = 4. Higher-order cases can be easily
extended.

s* Ap Ai A2 a 3 At
s* B Bi B2 B,
, „ BoAi — BiAo n _ BpAi — BjAg r _ B A 3 — B 3 Ap r _ Bq A* —A _ Am n
Bo Bo Bo Ho
j3 Bo Bi B2 Bi

„2 n _ BqC\ — BiCo n — BqCi — B 2 Cp n _ BpC — B 3 3 Cp n3 _ B a — OCp _ u

s Do Ul ~~
ft u
ft

To '
To To Bo
DqBi - DiBp = D B2 - D 2 Bq ^ = D B3 - D 3B =0
S2 E = Dp
£i
Dp Dp
Ei

s2 Dp Di D2
,i F = D ° Ei ~ DjE° Fi = DoEi ~ DlE° F2 =0
-
,i G = F° Di D Fl
Fo
<>
Gi= F° D *-
'o
F* D °
j> F Ft

sp ff0==
^oG l -F Go 1

Several important features of this tabulation must be mentioned at this

point.

1. The s 1 (J =
0, 1, 2, , ri) terms assigned to each
. . . row of the tabula-
tion are used for reference purposes.
Sec. 8.3 Construction of the Complete Root Loci / 401

2. The s' terms repeat for three consecutive rows for/ = (n — 1), . . .
,

1. There is only one s" row and one s" row.

3. If we regard the rows of coefficients in the tabulation that have the
same s J as a group, then it is noticed that the coefficients in the last
row of a group are always a repeat of those in the first row of the
same group.

If F(s) has multiple-order roots, which means that the root loci will have
breakaway points, a row of the tabulation shown above will contain all zero
elements. In other words, the tabulation process will terminate prematurely.
The multiple-order roots, which are the breakaway points, are obtained by
solving the equation formed by using the row of coefficients just preceding the
row of zeros.
Let us use the following numerical examples to illustrate the application
of the tabulation method of finding the breakaway points.

Example 8-14 Consider the equation

F(s) = (s + 2
1) C$ + 2)
2
= s* + 6i 3 + 13s 2 + 12s + 4 = (8-92)

We have stated the problem in such a way that the equation is known to have two
double roots at s = —1 and s = — 2. We are merely going to demonstrate the prop-
erties of the tabulation when the multiple-order-root situation occurs.
Taking the derivative on both sides of Eq. (8-92) with respect to s, we have
F'( s ) = 4s +3
l&s 2
+ 26s + 12 = (8-93)

The tabulation using the coefficients of F(s) and F'(s) is made in the following manner:

s* 1 6 13 12 4
3
S 4 18 26 12
(4)(6) - (D(18) _ 3 (4X13) - (D(26) _ 13 (4X12)
- 0X12)
s3 4
4 2 4 2 4
S3 4 18 26 12

S2 _i _3
4 -i
S2 6 18 12
S2 _i _3
"i
jl

Since therea row of zero elements in the tabulation before the tabulation process
is

is completed, equation F(s) has multiple-order roots. The equa-

this indicates that the
tion that needs to be solved is formed with the coefficients taken from the row just
above the row of zeros. The order of the equation is given by the power of s in the
reference column. Therefore, we have

-\s 2 - Is - \ = (8-94)
or
s2 + 3s +2= (8-95)

The roots of Eq. (8-95) are s = —1 and s = —2, which are the roots of F(s) with
multiple order.
 :

402 / Root Locus Techniques Chap. 8

Example 8-15 Let us consider a root locus problem, that is, the determination of the
breakaway points on the root loci. The equation, Eq. (8-74), con-
sidered in Example 8-10 will be used here. The equation is rewritten

F(s) =j + 2
(2 + K)s + 4K = (8-96)
Then
F'C?) = 2s (2 + K) = (8-97)

The following tabulation is made

J2 1 2 +K 4K
2 2 +K
sl
2 +K 4K
2
si 2 2 +K
s° «-< 2 +4 *> 2

None of the rows can be made to contain all zeros except the last one. Then, we
set
4- JQ 2
4K- (2 = (8-98)

or
-K + 2
12^: —4= (8-99)

Solving for K from the last equation, we have

K = 0.344
and
K= 11.656
When K equals either one of these two values, the s° row of the tabulation will
contain all zeros, thus signifying a multiple-order root for F(s). The equation from
which the roots or breakaway points can be obtained is formed by using the coefficients
in the row preceding the s° row. The equation thus formed is

2s + (2 + K) = (8-100)

Now substituting K = 0.344 and K = 11.656 into Eq. (8-100) in turn, we find the
two breakaway points on the root loci at

s = -1.172 K = 0.344
and
s = -6.828 K= 11.656
apparent that these answers agree with those obtained in Example 8-10. Further-
It is

more, a by-product of the tabulation method is that the values of K at the breakaway
points are also determined. In fact, the values of K are always determined first before
the breakaway points are found.

Example 8-16 Now consider the root locus problem of Example 8-12. Equation
(8-82) is written as

F(s) = s* + K= Ss 3 +36s 2 +80s +

(8-101)

The root loci of Eq. (8-101) have three breakaway points at s = — 2, — 2 + J2.45, and
—2 — y'2.45. Let us now determine the breakaway points by the tabulation method.
 : :

Sec. 8.3 Construction of the Complete Root Loci / 403

We have
F'( s ) = 4s + 24s + 3 2
12s + 80 = (8-102)

s 3
+ 6s 1 + 18s + 20 = (8-103)

The following tabulation is made

36 80 #
1 6 18 20
2 18 60 #
1 6 18 20

6 24 isT -40
2 18 - K-40 20
6
6 24 #-40
10- #-40 20
#-40

We would like to interrupt the tabulation at this point to remark that the elements
in the first group can be made zero by setting K = 100. Therefore, F(s)
row of the s1
has multiple-order roots when K
= 100. The breakaway points are found from the
equation
6s 2 + 24s + (K - 40) = K= 100 (8-104)
or
s 2
+ 4s + 10 = (8-105)

Therefore, the two breakaway points occur at s = — 2 +y'2.45 and s = — 2 — y'2.45,

which are the solutions of Eq. (8-105).
In order to complete the tabulation, we must now consider that K^ 100, so that
the coefficients in the subsequent row (the second
finite. In row of the s l group) are
fact, since the multiple-order roots at K=
100 are already determined, should there
be any additional multiple-order roots, they would have to occur at different values of
K than 100. Therefore, in the remaining three rows of tabulation it is implied that
K ^£ 100. Resuming the tabulation, we have

12 #-40
10 _ #-40 20
#-40 K^ 100
6

K- 64

Now the only row that can be all zero is the s° row, and only when K= 64. Therefore,
the breakaway point is found by substituting K= 64 into

40
10 ^)-+(» ')=» (8-106)

which gives
s = -2
An alternative way of completing the tabulation is to factor out the common
factor 10 - (# — row of the s group, and
40)/6 in the last 1
the tabulation is as follows
404 / Root Locus Techniques Chap. 8

12 K-40
1 2

s° K-64
Therefore, the same results, K= 64 and s = — 2, are obtained.

Example 8-17 In this example we shall show that the tabulation method actually
indicates explicitly that the root locus diagram Example
in Fig. 8-15,
8-13, does not have any breakaway points.
Equation (8-86) is written

F(s) = s3 + 2s + 2
2s +K= (8-107)
Then
F'(s) = 3s 2 + 4s + 2 = (8-108)

The following tabulation is made

K
2
K
2

K-%
1
- 21K
K-

*-£(*--£)('-¥) °

It is apparent that only the elements of the last row of the tabulation can be made
zero for any K. Therefore, we set

2-\(k-±)(i-^)=0 (8-109)

which is simplified to
811s: 2 - 4SK + 16 = (8-110)

However, there are no real values of K that will satisfy this equation. Therefore, the
conclusionis that F{s) does not have any multiple-order roots, or the root loci do not
have any breakaway points.

Comparison of the two methods of calculation of breakaway points.

1 The condition stated in Eq. (8-60) for the breakaway point is neces-
sary but not sufficient. The method involves solving for the possible
breakaway points as roots of Eq. (8-60). For higher-order systems
with a large number of breakaway points, the amount of work
involved in solving a higher-order equation in Eq. (8-60) may be
excessive.
2. The tabulation method gives a necessary and sufficient condition
for the breakaway points. In general, the procedure still involves
the solving of the roots of an equation, but the order of the equation
may be lower than that involved in the first method. The tabulation
method also gives the values of K
at the breakaway points.
Sec. 8.3 Construction of the Complete Root Loci / 405

These methods also represent ways of solving for the multiple-order root
of an equation.

Calculation of K on the Root Loci

Once the root loci have been constructed, the values of K at any point s^
on the loci can be determined by use of the denning equation of Eq. (8-8); that
is,

-
|r|
iwwi <!M1,)

If G,(j)#i(j) is of the form shown in Eq. (8-1 1), Eq. (8-12) is written

ni*i+/»j
1*1 = ^ (8-112)

or

l£-i _ product of lengths of vectors drawn from the poles of G^H^s) to s x

product of lengths of vectors drawn from the zeros of dls^H^s) to s t

(8-113)

Equations (8-112) and (8-113) can be evaluated either graphically or analyt-

ically. Usually, if the root loci are already drawn accurately, the graphical

method is more convenient. For example, the root loci of the equation
s2+ 2s + 2 + K(s + 2) = (8-114)

are shown in Fig. 8-16. The value of K at the point s, is given by

*=^ (8-115)

where A and B are the lengths of the vectors drawn from the poles of G(s)H(s)
= K{s + 2)/(s 2 + 2s + 2) to the point j, and C is the length of the vector
drawn from the zero of G{s)H(s) to s In the illustrated case s is on the root
t
.
t

loci, so K is positive. If j, is a point on the complementary root loci, K should

have a negative sign.
K
The value of at the point where the root loci intersect the imaginary axis
can also be found by the method just described. However, the Routh-Hurwitz
criterion usually represents a more direct method of computing this critical
value of K.
The eleven rules on the construction of root locus diagrams described
above should be regarded only as important properties of the root loci. Remem-
ber that earlier it was pointed out [Eq. (8-11)] that the usefulness of most of
these rules of construction depends on first writing Eq. (8-1) in the form

(s + Pl )(s +p 2 )...(s+ p„) + K(s + z t )(s + z2 ) (s + zm ) = (8-116)

Then, except for extremely complex cases, these rules are usually adequate for
the analyst to make a reasonably accurate sketch of the root loci just short of
plotting them point by point. In complicated situations, one has to rely on
a computer as a more practical means of constructing the root loci.
406 / Root Locus Techniques Chap. 8

.s-plane

Fig. 8-16. Graphical method of finding the values of K on the root loci.
The following example serves as an illustration on the application of the
rules of construction of the root loci.

Example 8-18 Consider the equation

s(s + + 6)(s 2 + 2s + 2) + K(s + 3) =

5)(s (8-117)

The complete root loci (— oo < K < co) of the system are to be constructed. Using
the rules of construction, the following properties of the root loci are determined:

1. The K = points on the complete root loci are at s =- 0, —5, —6, —1 +

j1,and — 1 — j 1 Notice that these points are the poles of G(s)H(s), where
.

G(s)H(s) = K(s + 3)
(8-118)
s(s + 5)(s + 6)(s 2 +2s + 2)

2. The K= ±co points on the complete root loci are at s -3, oo, oo,
oo, oo,which are the zeros of G(s)H(s).
3. There are five separate branches on the complete root loci.
4. The complete root loci are symmetrical with respect to the real axis of
the s-plane.
5. The angles of the asymptotes of the root loci at infinity are given by [Eq.
(8-33)]

6k = S£±J3«
n — m
= Q!L±TE
5 -1
o<*<co (8-119)

for k = 0, 1, 2, 3. Thus the four root loci that approach infinity in the
j-plane as K approaches +oo should approach asymptotes with angles
Sec. 8.3 Construction of the Complete Root Loci / 407

of 45°, -45°, 135°, and -135°, respectively. The angles of the asymptotes
of the complementary root loci at infinity are given by [Eq. (8-34)]

2kn 2kn
ek = 1
-co <K<0 (8-120)

Therefore, as K approaches -co, four complementary root loci should

approach infinity along asymptotes with angles of 0°, 90°, 180°, and 270°.
6. The intersection of the asymptotes is given by [Eq. (8-51)]

b - «i S Poles of G(s)H(s) - £ zeros of G(s)H(s)

—m =
\

n n — m
(8-121)
(0-5 -6 1 +/1 -1 -jp-(-3) _ = -2.5
The results from these six steps are illustrated in Fig. 8-17.

;w

s-plane

-n

°°+-K K= K= AT= + oo\ y'45A K= AT->--°°

-6 4 -3' \~ 2 _1 °

\\ K
-2.5 \
=
X --/l

Fig. 8-17. Preliminary calculations of the root loci of s(s + 5)(s + 6)(s 2
+ 2s + 2) + K(s + 3) = 0.
408 / Root Locus Techniques Chap. 8

In general, the rules on the asymptotes do not indicate on which side

of the asymptote the root locus will lie. Therefore, the asymptotes indi-
cate nothing more than the behavior of the root loci as s < oo. The —
complete root loci portions can be correctly sketched as shown in Fig.
8-17 only if the entire solution to the root loci problem is known.
7. Complete root loci on the real axis There are root loci (0 : K < oo) <
on the between s =
real axis and s = —3, s ==—5, and s = —6.
There are complementary root loci (— oo < K< 0) on the remaining
portion of the real axis, that is, between s = —3 and s = — 5, and * =
—6 and s = -co (see Fig. 8-18).

s-plane K=
Complementary
x
root loci

Fig. 8-18. Complete root loci on the real axis of s(s + 5)0 — 6)0 2 + 2s
+ 2) + K(s + 3) = 0.

8. Angles of departure: The angle of departure, 9, of the root locus leaving

the pole at—1 + / 1 is determined using Eq. (8-13). If s is a point on the t

root locus leaving —1 +jl, and St is very close to - -1 +jl as shown in

Fig. 8-19, Eq. (8-13) gives

1st +3 - (/j, + Is, + 1 +yl + /s t +5 + sv

/.

+ M + 1 -;1) = (Ik + 1)180° (8-122)

i-plane

Fig. 8-19. Computation of angle of departure of the root loci of s(s + 5)

(s + 6)0 2 + 2s + 2) + K(s + 3) = 0.
Sec. 8.3 Construction of the Complete Root Loci / 409

or
26.6° - (135° + 90° + 14° + 11.4° + 9) s; (2k + 1)180° (8-123)

for A: = 0, ±1, ±2,

Therefore,
9 =* -43.8° (8-124)

Similarly, Eq. (8-14) is used to determine the angle of arrival of the com-
plementary root locus arriving at the point —1 +jl. If this angle is
designated as 9', it is easy to see that 9' differs from 9 by 180°; that is,

9' = 180° - 43.8° = 136.2° (8-125)

9. The intersection of the root loci with the imaginary axis is determined by
the Routh-Hurwitz criterion. Equation (8-117) is rewritten

ss + 13s* + 54s 3 + 82s 2 + (60 + K)s + 3K = (8-126)

The Routh tabulation is

1 54 60 +K
13 82 3K
47.7 60 + 0.769-ST
65.6 - 0.212A: 3K
3940- 105A:-0.163*:2
65.6 - 0.212K
.s° 3K

For Eq. no roots in the right half of the s-plane, the

(8-126) to have
quantities in the first column of the Routh tabulation should be of the
same sign. Therefore, the following inequalities must be satisfied:

- 0.212#>
65.6 or K< 309 (8-127)

3940 - 105K - 0.1 63^ > or K< 35 (8-128)

K>0 (8-129)

Hence all of Eq. (8-126) will stay in the left half of the j-plane
the roots
if K lies betweenand 35, which means that the root loci of Eq. (8-126)
cross the imaginary axis when K = 35 and K = 0. The coordinate at the
crossover point on the imaginary axis that corresponds to K = 35 is
determined from the auxiliary equation

A(s) = (65.6 - 0.212A> + 3K = 2

(8-130)

Substituting K= 35 into Eq. (8-130), we have

58.2s 2
+ 105 = (8-131)
which yields
s = ±/1.34
10. Breakaway points Based on the information obtained from the preced-
:

ing nine steps, a trial sketch of the root loci indicates that there can be
only one breakaway point on the entire root loci, and the point lies

between the two poles of G(s)H(s) at s = —5 and —6. In this case, since
there is only one breakaway point for this fifth-order system, the value of
K at the point is obtained from the s° row of the breakaway-point tabula-
tion, which would contain a total of 14 rows if carried out. In this case
410 / Root Locus Techniques Chap. 8

it is actually easier to solve for the breakaway point by trial and error,
since we know that the desired root is between - 5 and —6.

Applying dK/ds = to Eq. (8-126) gives

s3 + 13.55* + 66.s
3
+ 142s 1 + 123s + 45 =
After a few trial-and-error calculations, the root of the last equation that
corresponds to the breakaway point is found to be s = —5.53.

From the information obtained in these 10 steps, the complete root locus diagram
is sketched as shown in Fig. 8-20.

In this section we have described 1 1 important properties of the root loci.

These properties have been regarded as rules when they are used in aiding the

Fig. 8-20. Complete root loci of s(s + 5)(s + 6)0 2 + 2s + 2) + K(s

+ 3) = 0.
Sec. 8.3 Construction of the Complete Root Loci / 411

construction of the root locus diagram. Of course, there are other minor prop-
erties of the root loci which are not mentioned here. However, in general, it is

found that these 1 1 rules are adequate in helping to obtain a reasonably accurate
sketch of the complete root loci just short of actually plotting them.
For easy reference, the 11 rules of construction are tabulated in Table 8-1.

Table 8-1 Rules of Construction of Root Loci

1 . K= points The K= pointson the complete root loci are at

the poles of G(s)H (s). (The poles include those at
infinity.)

2. K= ±oo points The K = ± oo points on the complete root loci are

at the zeros of G(s)H(s). (The zeros include those at
infinity.)

3. Number of separate The total number of root loci is equal to the order
root loci of the equation F(s) = 0.

4. Symmetry of root The complete root loci of systems with rational

loci transfer functions with constant coefficients are
symmetrical with respect to the real axis of the s-
plane.

5. Asymptotes of root For large values of s, the root loci (K > 0) are
loci as i — > oo asymptotic to straight lines with angles given by

and for the complementary root loci (K < 0)

a = 2kn
Ok

where A: = 0,\,2,...,\n — m\ — \.

6. Intersection of the (a) The intersection of the asymptotes lies only on

asymptotes the real axis in the j-plane.
(centroids) (b) The point of intersection of the asymptotes on
the real axis is given by (for all values of K)
2 real parts of _ 2 real parts of

a = poles of G(s)H(s) zeros of G(s)H(s)

7. Root loci on the On a given section on the real axis in the s-plane,
real axis root loci are found for K>
in the section only if
the total number of real poles and real zeros of
G(s)H(s) to the right of the section is odd. If the
totalnumber of real poles and zeros to the right of
a given section is even, complementary root loci
(^< 0) are found in the section.

8. Angles of departure The angle of departure of the root locus (K 0) >

and arrival from a pole or the angle of arrival at a zero of
G(s)H(s) can be determined by assuming a point si
that is on the root locus associated with the pole, or
zero, and which is very close to the pole, or zero,
 .

412 / Root Locus Techniques Chap. 8

Table 8-1 (Cont.)

and applying the equation

m n

/GfriV/fri) -£ Ai + z, - 2 Ai + P;
1=1 J= l

= (2k + \)n k = 0, ±1, ±2, ...

The angle of departure or arrival of a complemen-
tary root locus is determined from

/G(ii)H(si) = S Ai + - £ +/>,- z,- ,/s ,

;=i j=i
= 2A:ji = 0, ±1, ±2, fc . . .

9. Intersection of the The values of co and K at the crossing points of the

root loci with the root loci on the imaginary axis of the j-plane may
imaginary axis be obtained by use of the Routh-Hurwitz criterion.
The Bode plot of G(s)H(s) may also be used.

10. Breakaway points The breakaway points on the complete root loci
(saddle points) are determined by finding the roots of dK/ds = 0,
or dG(s)H(s)/ds —
These are necessary condi-
0.
tions only. Alternatively, the breakaway points are
determined from a tabulation using the coefficients
of the characteristic equations F(s) --= and F'(s)
= 0. The conditions are necessary and sufficient.
1 1 Calculation of the The absolute value of K at any point s i on the corn-
values of K on the plete root loci is determined from the equation
root loci 1
A
\G(si)H{ Sl )\
product of lengths of vectors drawn
_ from the poles of G{s)H(s) to s\
product of lengths of vectors drawn
from the zeros of G(s)H(s) to s\

8.4 Application of the Root Locus Technique to the Solution

of Roots of a Polynomial

Although the root locus diagram is primarily intended for the portrayal of
the trajectory of roots of a polynomial when a parameter, K, varies, the tech-
nique may also be used for the solution of the roots of a polynomial when all

the coefficients are known. The principle is best illustrated by an example. Con-
sider that it is desired to find the roots of the polynomial

F(s) = s3 + 3s 2 + 45 + 20 = (8-132)

To formulate this as a root locus problem, we should first convert Eq.

(8-132) into the form of Eq.
Since Eq. (8-132) does not have a variable
(8-4).
parameter K, the step of conversion is generally not unique. In other words,
we can regard any one of the four coefficients of the polynomial as the parameter
K. However, in general, the selection of a number of these coefficients as K will
result in simpler construction of the root loci. In this case it is not difficult to

see that it is more preferable to divide both sides of Eq. (8-132) by s 3 + 3s 2 .

Thus Eq. (8-132) leads to

 — —

Sec. 8.4 Application of the Root Locus Technique to the Solution / 413

+ (8-133)
A* + 3)
which is of the form of Eq. (8-4), with K = 4. Furthermore, it is apparent that
must also satisfy Eq. (8-133). Now the problem of solv-
the roots of Eq. (8-132)
ing Eq. (8-132) becomes that of a root locus problem, based on the pole-zero
configuration of

bWW _- Ks\s ++
G(s)H(s) (s 5)
3)
(8-134)

The desired roots are then found by setting K = 4. From a logistic standpoint,
we have embedded a specific problem in a more general one by first solving for
the solution to the general problem.
The root locus diagram based on the function G(s)H(s) of Eq. (8-134) for
K is shown in Fig. 8-21. When K = 4, the real root of Eq. (8-132) lies
positive
between —3 and —5, while the other two roots are complex with positive real

s-plane

—K-5=
e
°°
-*
K=
X
K=
K=

Fig. 8-21. Root locus diagram for the solution of the roots of s 3 + 3s 2
+ 4s + 20 = 0.
414 / Root Locus Techniques Chap. 8

parts.A few trial-and-error steps easily established the real root to be at s =

— 3.495. Thus the complex roots are found to be at s = 0.2475 + y2.381 and
s = 0.2475 -./2.381.

Example 8-19 Consider the problem of finding the roots of

s* + 5s 3 + 2s 2 +s+ 10 = (8-135)

Since this is a fourth-order polynomial, it is expected that the problem of solving for
its roots will be more difficult than that of the last example. Let us first divide both
sides of Eq. (8-135) by the first two terms of the equation; we have
2(s 2 + 0.5j + 5)
" , „„
1 + s*(s + 5) ° (8_136)

or
G(s)H(s) = ^afr+y
5>
K=2 (8-137)

The root locus diagram based on the poles and zeros of G(s)H(s) is constructed as
shown in Fig. 8-22 for oo > K >
0. However, from this root locus diagram it is not

s-plane

Fig. 8-22. Root locus diagram for the solution of the roots of s* + 5s 3
+ 2s 2 +s+ 10 = 0.
Sec. 8.4 Application of the Root Locus Technique to the Solution / 415

clear where the roots are when K = 2. Next let us divide both sides of Eq. (8-135) by
the first three terms of the equation ; we have

1
s + 10 = (8-138)
s 2 (s 2 + 5s + 2)
The root locus plot of

G(s)H(s) = K(s + 10) K= 1 (8-139)

s 2 (s 2 + 5s + 2)

with co > K> is shown The purpose of constructing the second root
in Fig. 8-23.
locus diagram is to establish a cross reference between the
two root locus diagrams,
since the roots of Eq. (8-135) must be found at the same points on both diagrams. Com-
paring the two root locus diagrams, we may conclude that if Eq. (8-135) has real roots,
there must be two, and they must lie between —0.44 and —4.56 on the real axis. Once
this range has been established, the real roots can be found by trial and error from
Eq. (8-135). One real root is found to be at s = —1.575. Dividing both sides of Eq.
(8-135) by (s + 1.575), we have the remainder,

s 3
+ 3.425.S 2 - 3.394.T + 6.346 = (8-140)
Now we repeat the procedure to find the roots of Eq. (8-140). Since one of the
three roots must be real, we can reason that it must be to the right of s = —4.56 on

/CJ

s-plane

Fig. 8-23. Root locus diagram for the solution of the roots of s* + 5s 3
+ 2s 2 +s+ 10 = 0.
416 / Root Locus Techniques Chap. 8

the real axis. However, we may acquire more information on this root by dividing both
sides of the equation by the first two terms. We have
-3.3940? - 1.87)
1 + s 2 (s + 3.425)

= K(s - 1.87)
G(s)H(s)
s 2 (s + 3.425) K -3.394 (8-141)

The complementary root loci of Eq. (8-140) with the G(s)H(s) of Eq. (8-141) are
sketched in Fig. 8-24 for -co < K< 0. Investigation of this complementary root locus

'
/w
s-plane

K<0

K<0

Fig. 8-24. Root locus diagram for the solution of the roots of s 3.425.S 2
- 3.3945 + 6.346 = 0.

diagram and the root loci of Fig. 8-24 reveals that the real root must lie between —3.425
and —4.56. To find the real root of Eq. (8-140), which is at s = —4.493, is a simple
matter. The two complex roots are subsequently found to be at 5 = 0.534 + j 1.057
and s = 0.534 — / 1.057. Thus the roots of Eq. (8-135) are now all determined:
5 = -1.575
s = -4.493
s = 0.534 +/ 1.057
.y = 0.534 -y 1.057
In summarizing, we have used two examples to illustrate the application
of the root locus method to the solution of the roots of a high-order equation,
F(s) = (8-142)
 Sec. 8.5 Some Important Aspects of the Construction of the Root Loci / 417

where F(s) is a polynomial in s. The general procedure is to convert the equation

to be solved into the form 1 + G(s)H(s) = by dividing both sides of the equa-
tion by a certain appropriate portion of the equation.
In other words, we should express Eq. (8-142) as

P(s) + GO) - (8-143)

Then, G(s)H(s) = Q(s)/P(s), or G(s)H(s) = P(s)/Q(s). The selection of the poly-

nomials P(s) and Q{s) from F(s) is more or less arbitrary, although in general
the orders of P(s) and Q{s) should be close so that the root locus problem is made
as simple as possible. In many circumstances it may be desirable to use more
than one choice of division of F{s) into P{s) and Q(s). This will usually provide
additional information on the location of some of the roots of the equation, so
that the trial-and-error procedure can be simplified. Example 8-19 illustrates
how this is done.
In essence, the method of root finding presented here is that of utilizing
the root locus technique to give an indication of the general location of some
of the roots. The basic method of finding the exact roots is still cut-and-try.
For high-order equations, and for equations only with complex roots, the root
locus method of root finding may not be effective, and a computer solution is

still the most preferable one.

8.5 Some Important Aspects of the Construction of the Root Loci

One of the important aspects of the root locus techniques is that for most con-
trol systems with moderate complexity, the analyst or designer may conduct
a quick study of the system in the s-plane by making a sketch of the root loci
using some or all of the rules of construction. In general, it is not necessary to
make an exact plot of the root loci. Therefore, time may be saved by skipping
some of the rules, and the sketching of the root locus diagram becomes an art
that depends to some extent on the experience of the analyst.
In this section we shall present some of the important properties of the root
loci which may be helpful in the construction of the root locus diagram.

Effect of Adding Poles and Zeros to G(s)H(s)

In Chapter 6 the effects of the derivative and integral control were illustrated
by means of the root locus diagram. From the fundamental viewpoint we may
investigate the effects to the root loci when poles and zeros are added to G(s)H(s).

Addition of poles. In general we may state that adding a pole to the func-
tion G(s)H(s) in the left half of the s-plane has the effect of pushing the original
root loci toward the right-half plane. Although it is difficult to make a precise
statement and provide the necessary proof, we can illustrate the point by several
examples.
Let us consider the function

G(s)H(s) = K a >
, , (8-144)

The zeros of 1 + G(s)H(s) are represented by the root locus diagram of Fig.
8-25(a). These root loci are constructed based on the poles of G(s)H(s) at s =
 OO

t
s-plane s-plane
K

K=Q K= a
l

—a —a o
2

<

K
i

(a) (b)

(c) (d)

Fig. 8-25. Root locus diagrams that show the effects of adding poles to
G(s)H(s).

418
 Sec. 8.5 Some Important Aspects of the Construction of the Root Loci / 419

and s = — a. Now let us introduce a pole at s -b so that

G(s)H(s) = K b> (8- 1 45)

s(s ~|- a)(s + b)

Figure 8-25(b) shows that the additional pole causes the complex part of the
The angles of the asymp-
root loci to bend toward the right half of the s-plane.
totes are changed from ±90° to ±60°. The breakaway point is also moved to
the right. For instance, if a = and b = 2, the breakaway point is moved from
l

—0.5 to —0.422 on the real axis. If G(s)H(s) represents the loop transfer func-
tion of a feedback control system, the system with the root loci in Fig. 8-25(b)
may become unstable if the value of K exceeds the critical value, whereas the
system represented by the root of Fig. 8-25(a) is always stable. Figure
loci
8-25(c) shows the root loci when another pole is added to G(s)H{s) at s = — c.
The system is now of the fourth order, and the two complex root loci are moved
farther to the right. The angles of the asymptotes of these two loci are ±45°.
For a feedback control system, the stability condition of the system becomes
even more restricted. Figure 8-25(d) illustrates that the addition of a pair of
complex-conjugate poles to the original two-pole configuration will result in
a similar effect. Therefore, we may draw a general conclusion that the addition
of poles to the function G(s)H(s) has the effect of moving the root loci toward
the right half of the .y-plane.

Addition of zeros. Adding zeros to the function G(s)H(s) has the effect of
moving the root loci toward the left half of the j-plane. For instance, Fig. 8-26(a)

OO
/CJ i

/CO
t

6 = < is
s-plane 5-plane
K = °°
®--^
I

/ X \
\
/ \
\ \
/
— -o — a:
-*»
= o

T
\

)
*
-a/2
4— tf =
»- o
K=0
~b
i
K=
-a
\

-a/2
K=
o
1
I 1
\ /
\ /
\ / '
'

e
K = °°
b = <*>^ %
i
(a) (b)

Fig. 8-26. Root locus diagrams that show the effects of adding a zero to
G(.s)H(s).
420 / Root Locus Techniques Chap. 8

K/7
f
71 JCO i

/
<
s-plane

J I

u
\
AT = °° K=Q K = A' =

- c - b — a

\
\
\
w \\
A -\
(c)

Fig. 8-26 (Cont.). Root locus diagrams that show the effects of adding
azerotoG(.s)//(.f).

shows the root locus diagram when a zero at 5 = —b is added to the function
G(s)H(s) of Eq. (8-144), with b > a; the resultant root loci are bent toward the
left and form a circle. Therefore, if G(s)H(s) represents the loop transfer function
of a feedback control system, the relative stability of the system is improved

by the addition of the zero. Figure 8-26(b) illustrates that a similar effect will
result if a pair of complex-conjugate zeros is added to the function of Eq. (8-144).
Figure 8-26(c) shows the root locus diagram when a zero at v = — c is added to
the transfer function of Eq. (8-145).

Effects of Movements of Poles and Zeros

It was mentioned earlier that the construction of the root locus diagram

depends greatly on the understanding of the principle of the technique rather

than just the rigid rules of construction. In this section we show that in all
cases the study of the effects of the movement of the poles and zeros of G{s)H(s)
on the root loci is an important and useful subject. Again, the best way to
illustrate the subject is to use a few examples.
 II

o u,

c
"E.

o
II : :

e
c :=;

E +
.a ^
" II

•S «
"> „
3 II
"

O £,
f}
-o ^
S3
o £, II

aS O Q
o
N ^
as a> 00
M «*, II

.fa tt Q

421
 o
II

422
Sec. 8.5 Some Important Aspects of the Construction of the Root Loci / 423

/CJ

5-plane

K=
* * a
K=

(e)a= 1

Fig. 8-27 (Cont.).

Example 8-20 Consider the equation

^(s + a) + K(s + b) = (8-146)

which is easily converted to the form of 1 + G(s)H(s) = 0, with

G(s)H(s) --=
*£±^ (8-147)

Let us set b = 1 and investigate the root loci of Eq. (8-146) for several values of a.
Figure 8-27(a) illustrates the root loci of Eq. (8-146) with a = 10 and b 1. The =
two breakaway points are found at s = —2.5 and —6.5. It can be shown that for arbi-
trary a the nonzero breakaway points are given by

a -1- 3
^Vo - 2
10a (8-148)

When = 9, Eq. (8-148) indicates that the breakaway points converge to one point
a
at j = —3, and the root locus diagram becomes that of Fig. 8-27(b). It is interesting
to note that a change of the pole from — 10 to —9 equals a considerable change to the
root loci. For values of a less than 9, the values of s as given by Eq. (8-148) no longer
satisfy the equation in Eq. (8-146), which means that there are no finite, nonzero, break-
away points. Figure 8-27(c) illustrates this case with a = 8. As the pole at s —a is ------
424 / Root Locus Techniques Chap. 8

moved farther to the right, the complex portion of the root loci is pushed farther
toward the right-half plane. When a = b, the pole ats = —a and the zero at —b cancel
each other, and the root loci degenerate into a second-order one and lie on the imagi-
nary axis. These two cases are shown in Fig. 8-27(d) and (e), respectively.

Example 8-21 Consider the equation

s(s 2 + 2s + a) + K(s + 2) = (8-149)

which is converted to the form of 1 + G(s)H(s) = 0, with

+ 2
G(s)H(s) = (8-150)
l a) s{ «l
The objective is to study the complete root loci (-co < K < co) for various values of
a{> 0). As a start, let a = so that the poles of G(s)H(s) are at s = 0, — 1, and —1.
1

The complete root loci for this case are sketched in Fig. 8-28(a). By setting dG(s)H(s),lds
to zero, thebreakaway points are found at * = —0.38, —1, and - 2.618.
As the value of a is increased from unity, the two double poles of G(s)H{s) at
s = — 1 will move vertically up and down. The sketch of the root loci is governed
mainly by the knowledge of the breakaway points. We can show that dG(s)H(s)lds
leads to
.y
3
+ 4s 2 + As + a = (8-151)

As the value of a increases, the breakaway points at s = —0.38 and s = —2.618

move to the left, whereas the breakaway point at s = —1 moves toward the right.
Figure 8-28(b) shows the complete root loci with a = 1.12; that is,

Since the real parts of the poles and zeros of G(s)H(s) are not affected by the value
of a, the intersect of the asymptotes is always at the origin of the .j-plane. The break-
away points when a = 1.12 are at s = —0.493, —0.857, and —2.65. These are obtained
by solving Eq. (8-151).
By solving for a double-root condition in Eq. (8-151) when a = 1.185, it can be
shown that the two breakaway points that and s = — 1 converge to
lie between s =
a point. The root loci for this situation are sketched as shown in Fig. 8-28(c).
When a is greater than 1.185, Eq. (8-151) yields one real root and two complex-
conjugate roots. Although complex breakaway points do occur quite often in root loci,
we can easily show in the present case that these complex roots do not satisfy the origi-
nal equation of Eq. (8-149) for any real K. Thus the root loci have only one breakaway
point, as shown in Fig. 8-28(d) for a = 3. The transition between the cases in Fig.
8-28(c) and (d) should be apparent.

8.6 Root Contour— Multiple-Parameter Variation

The root locus technique discussed thus far is restricted to only one variable
parameter in K. However, in many control systems problems, the effects of
varying several parameters must be studied. For example, when designing
a controller that is represented by a transfer function with poles and zeros, it is

necessary to investigate the effects on the performance of the overall system

when these poles and zeros take on various values.
 \

a o ^
-'
&
""
—
ft I!

C XI

o _
£ll

"3
£
o
-C +
r»
(1)

J3
**

fc
+
3 *
.S r*i

3 II II

O '4 <3
o
&5
o
& ^ 00
1—1

90 .— 1—1
n v^2-
II
as a;
DC
H. "b

425
 b

>t
t

-* if 5
3

""
8 *- k ^ -*
/^"^ 8

2
o
II ./
H^ * (

x
o
t^r + 7 T ^
7 I

l
8
\ /
1

\ 2' /
a)

to v.: \^y en
1

e
8
1

426
Sec. 8.6 Root Contour— Multiple-Parameter Variation / 427

In Section 8.5 the root locus diagrams of equations with two variable
parameters are studied by assigning different values to one of the parameters.
In this section the multiparameter problem is investigated through a more
systematic method of embedding. When more than one parameter varies con-
tinuously from —
oo to oo, the loci of the root are referred to as the root contours.
It willbe shown that the same conditions and rules of the root loci are still
applicable to the construction of the root contours.
The principle of the root contours can be illustrated by considering the
equation
Q(s) + K.PAs) +KP 2 2 (s)
= (8-153)

where K and K2 are the variable parameters and Q(s), P^s), and P2 (s) are
t

polynomials of s. The first step involves the setting of one of the parameters
equal to zero. Let us set K2 equal to zero. Then Eq. (8-153) becomes

Q(s) + K^is) = (8-154)

The root loci of this equation may be obtained by dividing both sides of the
equation (the Golden Rule) by Q(s). Thus

Kx
1 + Z^= 1
(8-155)

or
1 + G,(j)ff,(s) = (8-156)

The construction of the root loci depends on the pole-zero configuration of

G^H^s) =
^^ (8-157)

Next, we restore the value of K 2, and again apply the Golden Rule to Eq.
(8-153), with K 2 as the variable parameter. We have

or
1 +G a (s)H2 (s) = (8-159)

Now the root contours of Eq. (8-153) are constructed based upon the poles and
zeros of

^'^ ^HW) (8 - 160)

However, one important feature is that the poles of G 2 (s)H2 (s) are identical to
Thus the root contours of the original
the roots of Eq. (8-156) or of Eq. (8-154).
equation must all start {K 2 = 0) at the points that lie on the root loci of Eq.
(8-156). This is the reason why one root contour problem is considered to be
embedded in another. The same procedure may be extended to more than two
variable parameters.

Example 8-22 Consider the equation

i 3
+ K s 2 + K iS + K =0
2 x (8-161)

where KA and K
2 are the variable parameters and with values that lie between and oo.
428 / Root Locus Techniques Chap. 8

As a first step we let K =

2 0; Eq. (8-161) becomes
.y
3
+ K^s + Ki =0 (8-162)
which is converted to

(8-163)

The root loci of Eq. (8-162) are drawn from the poles and zeros of

(8-164)

as shown in Fig. 8-29(a).

Next, we let K
2 vary between zero and infinity while holding Ki at a constant

nonzero value. Dividing both sides of Eq. (8-161) by the terms that do not contain K2 ,

we have
, ,
K s*
2
(8-165)
j 3
+ K,s +K t

Thus the root contours of Eq. (8-161) when K 2 varies may be drawn from the pole-zero
configuration of

G 2 (s)H2 (s) = (8-166)

s 1
+ KiS + K,

iu f
AT,

s-plane

K, = °° =0
AT,

0.5

(a)

Fig. 8-29. (a) Root contours for s 3 + K2 s 2 + K t s +K = t 0, K2 =

Sec. 8.6 Root Contour — Multiple-Parameter Variation / 429

s-plane

K2 =0(0.4+/1.74)
/ K, = 2.56

(b)

Fig. 8-29 (Cont.). (b) Root contours for s 3 -\ K2S i + K iS + K t

K2 varies, K =
t constant.

The zeros of G 2 (s)H2 (s) are at s = 0, 0; but the poles are the zeros of 1 +
G^H^s) which have been found on the contours of Fig. 8-29(a). Thus for fixed A"i
the root contours when K2 varies must all emanate from the root contours of Fig.
8-29(a).

Example 8-23 Consider the loop transfer function

G{s)H{s)
K
(8-167)
s{[ + Ts)(s z + 2s + 1)

of a closed-loop control system. It is desired to construct the root contours of the

characteristic equation with K and T as variable parameters.
The characteristic equation of the system is written

s(l + Ts)(s 2
+ 2s + 2) +K= (8-168)
430 / Root Locus Techniques Chap. 8

First, we shall set T equal to zero. The characteristic equation becomes

s(s 2
+ 2s + 2) + K = (8-169)

The root contours of this equation when K varies are drawn based on the poles and
zeros of

Gi(s)Hds) = rf . 2
s(s A,
+ 2s + ^ .l.
2)
(8
" 17 °)
as shown in Fig. 8-30(a).

,
fU) .. /CO

s-plane •s-plane
/
/t = o
- 1 +/1 / K=\0
K= *>%_ K= 4

°°^K K= T= oo
<
]

o k=io o"t = ~
T=
K=4
K= K= ^ a:=io
\\ T= oo \ T=
\
(a) (b)

Fig. 8-30. (a) Root loci for s(s 2 + 2s + 2) +K= 0. (b) Pole -zero con-
figuration of G 2 (s)H2 (s) = [Ts 2 (s 2 + 2s + 2)]/s(s 2 + 2s + 2) + K].

For the root contours with T varying and K held constant, we write Eq. (8-168) as

1 4- G^tfaCs) = 1 +
Ts 2
(s
2
+ 2s + 2) (8-171)
s(s 2 + 2s + 2) + K
Therefore, the root contours when T
from the pole-zero
varies are constructed
configuration of G 2 (s)H2 (s). When T = on the root contours are at the
0, the points
poles of G 2 (s)H 2 (s), which are the points on the root loci of Eq. (8-169), as shown in
Fig. 8-30(b) for K = 10. The T = oo points on the root contours are at the zeros of
G 2 (s)H 2 (s), and these are at s = 0, 0, — 1 +j\, and —1 — y'l. The root contours for
the system are sketched in Figs. 8-31, 8-32, and 8-33 for three different values of K;
when K — 0.5 and T = 0.5, the characteristic equation has a quadruple root at s —
-1.

Example 8-24 As an example illustrating the effect of the variation of a zero of

G(s)H(s), consider

G(s)H(s)
K(l + Ts)
(8-172)
s(s + 1)(* + 2)
The problem may be regarded as a study of the effect of derivative control, as discussed
in Section 6.7, on the root locations of the characteristic equation.
The characteristic equation of the system is
s(s + l)(s + 2) + K(l +Ts)=0 (8-173)

Let us first consider the effect of varying the parameter K. Setting T= in Eq. (8-173)
 Sec. 8.6 Root Contour — Multiple-Parameter Variation / 431

s-plane

Fig. 8-31. Root contours for s(l + sT)(s 2 + 2s + 2) +K= 0; K> 4.

i
i u A /
s-plane i u /
s-plane
/
/ /
/ r-^-oo /
v" r =
^
r=i 1
/ T=0 o

o<-r ^\. O^T T'-* oo

T=0
'o
T^KX>
r-K»
7^0^
MJ r=0
f
o r->-=c

r-> CO ^ jT-^-oo X
\ \
\ \
\ \
\
Fig. 8-32. Root contours for s(l + sT){s 2 + 2s Fig. 8-33. Root contours for ,j(l + sT)(s 2 + 2s
+ 2) + K = 0; K = 0.5. + 2) +K= 0; K< 0.5.

yields
*(j + l)(s + 2) +K= (8-174)
which leads to
= (8-175)
s(s + l)(s + 2)

The root loci of Eq. (8-174) are sketched in Fig. 8-34, based on the pole-zero configu-
ration of

When T varies between zero and infinity, we write Eq. (8-173) as

TKs
1 + G 2 (s)H2 (s) = 1 + s(s + (8-177)
l)(s + 2)+K
432 / Root Locus Techniques Chap. 8

/CO ;

s-plane

K= 6

°°*-AT K=0 K=0 K=

-K- -*•»

\',K = 6

Fig. 8-34. Root loci for s(s + 1)0 + 2) +K= 0.

/OJ /
/
/
/
/
,*K =20
5-plane
/ 0.425 +/2. 235

AT = 20
iHt- +H^
-3.85 1
I

\ 0.425-/2.235
X K= 20
\
\
\\
Fig. 8-35. Pole-zero configuration of G 2 (s)H2 (s) = TKs/[s(s + 1)0 + 2)
+ K], K= 20.

The points that correspond to T = on the root contours are at the roots of
s(s+ 1)0 + 2) + K = 0, whose loci are sketched as shown in Fig. 8-34. If we choose
K = 20, the pole-zero configuration of G (s)H2 (s) is shown in Fig. 8-35. The root
2

contours of Eq. (8-173) for < T < co are sketched in Fig. 8-36 for three values of
Sec. 8.6 Root Contour—Multiple Parameter-Variation / 433

s-plane

\a:=20, r=o

Fig. 8-36. Root contours of s(s + l)(s + 2) +K+ KTs = 0.

K. The intersection of the asymptotes of the root contours is obtained from Eq. (8-51);
that is,

-3.85 + 0.425 + 0.425 = -1.5 (8-178)

1

Therefore, the intersection of the asymptotes is always at s = —1.5 because the sum

of the poles of Gt{s)Hi{s) is always equal to —3, regardless of the value of K, and the
sum of the zeros of G2(s)H2(s) is zero.
434 Root Locus Techniques
/
Cna _ g

The root contours shown in Fig. 8-36 verify the well-known fact that the derivative
control generally improves the relative stability of the closed-loop system by moving
the characteristic equation roots toward the left in the s-plane. The root contours also
clearly indicate an important characteristic of the derivative control in that the band-
width of the system is increased. In certain cases the contribution to the increase in
bandwidth by increasing the value of T far exceeds the improvement made on the
relative stability of the system. As shown in Fig. 8-36, for K = 20, the system is sta-
bilized for all values of r greater than 0.2333. However, the largest damping ratio that
the system can have by increasing Tis approximately 30 per cent.

8.7 Root Loci of Systems with Pure Time Delay 51

In Chapter 5 we investigated the modeling of systems with pure time delays and
pointed out that the time delay between signals at various points of a system
can be represented by the exponential term e~ Ts as its transfer function, where
T is the delay in seconds. Therefore, we shall assume that the characteristic
equation of a typical closed-loop system with pure time delay may be written

Q{s) + KP{s)e' T ' = (8-179)

where Q(s) and P(s) are polynomials of s. An alternative condition of Eq.

(8-179) is

1 + JK? (j)tf,(»e- r '

1
= (8-180)
where

G.fr)^) = (8-181)
gg
Thus, similar to the development in Section 8.2, in order to satisfy Eq. (8-180),
the following conditions must be met simultaneously:

|
= j~
e~ r ' G,(5)ff ,(j) |
- oo < K < co (8-1 82)

IG^Hjjs) = (2k +1)ti + coT K>0 (8-183)

/G^H^s) = 2kn + (oT K<0 (8-184)

where s =a + joo and k = 0, ±1, ±2, Note that the condition for any
point s = 5] in on the complete root loci is given in
the 5-plane to be a point
Eqs. (8-183) and (8-184), which differ from the conditions of Eqs. (8-9) and
(8-10) by the term coT. When T=
0, Eqs. (8-183) and (8-184) revert to Eqs.
(8-9) and (8-10). Since co is a variable in the j-plane, the angular conditions of
Eqs. (8-183) and (8-184) are no longer constant in the s-plane but depend upon
the point at which a root of Eq. (8-179) may lie. Viewing the problem from
another standpoint, it is recognized that if T =
0, given a value of K, there are
only n points in the s-plane that will satisfy either Eq. (8-183) or Eq. (8-184),
for all possible values of k, where n is the highest order of P(s) and Q{s). How-
ever, for I?t0, the angular conditions in Eqs. (8-183) and (8-184) depend on co,
which varies along the vertical axis in the .s-plane. Thus, for a given K, there may
Sec. 8.7 Root Loci of Systems with Pure Time Delay / 435

be more than n points which satisfy the angular conditions in the s-plane, as k
takeson all possible integral values. In fact, there are an infinite number of
these points, since Eq. (8-179), which is transcendental, is known to have an
infinitenumber of roots.
The difficulty with the construction of the root loci of Eq. (8-179) is that
many of the rules of construction developed originally for systems without time
delay areno longer valid for the present case. It is of interest to investigate how
some of the rules of construction given in Section 8.3 may be modified to apply
to the time-delay case.

K= Points

Theorem 8-9. The K= points on the complete root loci of Eq. (8-180) are
at the poles ofGi(s)Hi(s) and a = — °°.
Proof: Equation (8-182) is repeated,

e-^IG^)/^)^^ (8-185)

Thus, if K equals zero, s approaches the poles of G^H^s), or a, which is

the real part of s, approaches -co.

The K — ± co Points

Theorem 8-10. The K = ± °° points on the complete root loci of Eq. (8-180)
are at the zeros of Gi(s)H (s) and a = co.
t

Proof: Referring again to Eq. (8-185), the proof of this theorem becomes
evident.

Number of Branches on the Complete Root Loci

The number of branches on the root loci of Eq. (8-179) is infinite, since
number of roots.
the equation has an infinite

Symmetry of the Complete Root Loci

The complete root loci are symmetrical with respect to the real axis of
the i-plane. This is explained by expanding e~ Ts into an infinite series; then Eq.
(8-179) again becomes a polynomial with a real coefficient but with infinite
order.

Asymptotes of the Complete Root Loci

Theorem 8-11. The asymptotes of the root loci of Eq. (8-179) are infinite in
number and all are parallel to the real axis of the s-plane. The intersects of the
asymptotes with the imaginary axis are given by

o> = ~ (8-186)

where N is tabulated in Table 8-2 for the various conditions indicated.

 .

436 / Root Locus Techniques Chap. 8

n= number of finite poles of G^H^s)

m= number of finite zeros of G^H^s)

Table 8-2

K n —m K= Asymptotes K = ± °° Asymptotes

>o Odd N = even integers N — odd integers

= 0,±2, ±4, . . . = ±1, :t3, ±5,...
Even N= odd integers iV= odd integers
= ±1, ±3, ±5,... = ±1, .b3, ±5,...
<o Odd N= odd integers N = even integers
= ±1, ±3, ±5, ... = 0, ±2, ±4, . .

Even N= even integers N = even integers

= 0, ±2, ±4, . . . = 0, ±2, t4,..
: .

Proof: Since as s —
> oo on the root loci, K either approaches zero or ± oo,

Theorems 8-9 and 8-10 show that the asymptotes are at a = oo (K = ± oo)
and cr = — oo (K = 0). The intersections of the asymptotes with the ./co-axis and
the conditions given in Table 8-2 are arrived at by use of Eqs. (8-183) and
(8-184).

Root Loci on the Real Axis

The property of the root loci of Eq. (8-179) on the real axis is the same as
stated in Theorem 8-7, because on the real axis, co = 0, the angular conditions
of Eqs. (8-183) and (8-184) revert to those of Eqs. (8-9) and (8-10), respectively.

Angles of Departure and Arrival

Angles of departure and arrival are determined by use of Eqs. (8-183) and
(8-184).

Intersection of the Root Loci with the Imaginary Axis

Since Eq. (8-179) is not an algebraic equation of s, the intersection of its

loci with the imaginary axis cannot be determined by use of the Routh-Hurwitz
criterion. The determination of all the points of intersection of the root loci
with the^'cD-axis is a difficult task, since the root loci have an infinite number of
branches. However, we shall show in the following section that only the inter-
sections nearest the real axis are of interest for stability studies.

Breakaway Points

Theorem 8-12. The breakaway points on the complete root loci ofEq. (8-179)
must satisfy
Sec. 8.7 Root Loci of Systems with Pure Time Delay / 437

T'
dG 1
(.s)H 1 (s)e- _ (8-187)
ds

Proof: The proof of this theorem is similar to that of Theorem 8-8.

Determination of the Values of K on the Root Loci

The value of K at any point s = st on the root loci is determined from Eq.
(8-182); that is,

l*l = Tr77^T771T
|Gl(Sl)«lC5l)l
(8
"
188)

where <r t
is the real part of s t
.

Example 8-25 Consider the equation

s + Ke- T *=0 (8-189)

It is desired to construct the complete root loci of this equation for a fixed value of T.
Dividing both sides of Eq. (8-189) by s, we get

+ —s —=
Kp~ Ts
1 (8-190)

which is of the form of Eq. (8-180) with

CPifj) = y (8-191)

The following properties of the root loci of Eq. (8-189) are determined:

1. The K= points: From Theorem 8-9, K= at s = and at a = -co.

Using Theorem 8-11 and Table 8-2, we have
K > 0: K approaches zero as a approaches -co at co = 0, ±2a/T,
±47t/r, . .

K < 0: K approaches zero as a approaches -co at co = ±n/T, ±371/7",

±571 IT, . .

2. The K = ±°o points: From Theorem 8-10, K= ±oo at a = co. Using

Theorem 8-11 and Table 8-2, we have
K>0: K approaches +oo as a approaches +co at ft) = ±nlT,
±37t/r, . .

K< 0: A" approaches — co as a approaches +oo at co = 0, ±.1njT,

±47r/r; . .

The K = 0, K= ±oo points, and the asymptotes of the root loci are
+
shown in Fig. 8-37. The notation of is used to indicate the asymptotes
of the root loci, and 0~ is for the complementary root loci.

3. The root loci (K > 0) occupy the negative real axis, and the complemen-
tary root loci (K <
0) occupy the positive real axis.
4. The intersections of the root loci with the jco axis are relatively easy to
determine for this simple problem.
Since G i(s)Hi(s) has only a simple pole at j = 0, for any point si on
438 / Root Locus Techniques Chap. 8

* /CO
s-plane

• •
• •
• •
•
+ ^K •
j4ir/T K^-oo

0-^K Pn/T K~>+ oo

+ ^K I2v/T K^ -oo

0-+-K MT K^+ <*>

+ ^K dzi^. K^ -oo

Q-^K -jn/T £->• + °°

+ +- K -J2*IT a:-* -oo

0-^K -miT K-* +oo

+ <-K -j4*/T £- -°o

• •
• •

• •

• •

Fig. 8-37. Asymptotes of the complete root loci of the equation s + Ke~ Ts
= 0.

the positive jco axis,

/GifoWifri) = -- (8-192)

and for any point s^ on the negative j CO axis,

/gi(Ji)tfifri) =4r (8-193)

Thus, for K> 0, Eq. (8-183) gives the condition of root loci on they CO
axis (co > 0),

-~ = (2k + l)7t +coT (8-194)

k = 0, ± 1, ±2, . . . The values of co that correspond to the points at which

the root loci cross the positive jco axis are

n_ 5n_ 9n_
f
co (8-195)
27" 27" 27"
Sec. 8.7 Root Loci of Systems with Pure Time Delay / 439

For K > 0, and <a < 0,

~- = (2k + l)w + coT (8-196)

and for k = 0, ±1, ±2, . .

.
, the crossover points are found to be at

CO = -jj, ~2j, ~2j;, ... (8-197)

Similarly, for K < 0, the conditions for the complementary root loci
to cross the yco axis are

-y = 2ta +cor co>0 (8-198)

y= 2A:7t + cor co < (8-199)

The crossover points are found by substituting k = 0, ±1, ±2, . . . into

the last two equations. We have

G>=±f|,±2y,±^,..- (8-200)

5. Breakaway points: The breakaway points on the complete root loci are

determined by the use of Eq. (8-187). Thus

dG (s)H i 1
(s)e-Ts
= d^\ (8 . 201)
ds ds\ s }

or
—Te~ Ts s — e~ T *
n (8-202)

from which we have

e~ T '(Ts + 1) = (8-203)

Therefore, the finite breakaway point is at s = —IjT.

6. The values of K at the crossover point on they'd) axis are found by using
Eq. (8-188). Since = 0on they co axis, we have
<7

where COc is a crossover point.

Based on the properties accumulated above, the complete root loci of Eq. (8-189)
are sketched as shown in Fig. 8-38.
Although the equation of Eq. (8-179) has aninfinite number of roots, and there-
fore the root loci have an infinitenumber of branches, from the system analysis stand-
point, only the branches that lie between — njT < co < n\T are of interest. We shall
refer to these as the primary branches. One reason is that the critical value of K at the
crossover point on this portion of the root loci is equal to 71/27', whereas the critical
value ofK at the next branch at co = ±5nj2T 571/2T, which much greater. There- is is

fore, K = 7C/Tis the critical value for stability. Another reason for labeling the primary
branches as the dominant loci that for any value of K less than the critical value of
is

7t/2r, the corresponding roots on the other branches are all far to the left in the j-plane.
440 / Root Locus Techniques
Chap. 8

s-plane

+ <-K

+ <-K

Fig. 8-38. Complete root loci for j + Ke~ Ts = 0.

Therefore, the transient response of the system, which has Eq. (8-189) as its character-
istic equation, is predominantly controlled by the roots on the primary branches.

Example 8-26 As a slightly more complex problem in the construction of the root
loci of a system with pure time delay, let us consider the control sys-
tem shown in Fig. 8-39. The loop transfer function of the system is

= Ke~ T °

G(s)H(s)
s(s + 1)

The characteristic equation is

Sec. 8.7 Root Loci of Systems with Pure Time Delay / 441

r(t)
^ .
s{s+
K
l)
cU)

9.

e -Ts

Fig. 8-39. Feedback control system with a pure time delay in the feedback
path.

s2 +s+ Ke~ Ts =
In order to construct the complete root loci, the following properties are assembled
by using the rules given in Theorems 8-9 through 8-12.

1. The K = points: From Theorem 8-9, K = at s = 0, s = —\, and

a = — co. Using Theorem 8-11 and Table 8-2, we have
K > 0: K approaches zero as a approaches -co at CO = ±7t/T, ±3/nT,
±5n /T, . . .

K<0: K approaches zero as a approaches — co at co = 0, ±2tc/T,

±4n/r, . .

2. The K = ±c° points: From Theorem 8-10, K = ±oo at a = co. Using

Theorem 8-11 and Table 8-2, we have
K>0: K approaches +co as a approaches +co at co = ±njT,
±37t/T, . .

K<0: A' approaches — co as a approaches +co at co =0, ±27t/r,

±4a/T, . .

Notice that the asymptotes depend upon n — m, which is

K=
even in this case, but the K = ±<x> asymptotes depend only on the sign
of Kand not on n — m.
3. The root loci (K> 0) occupy the region between s = and s —1 on =
the real axis. The rest of the real axis is occupied by the complementary
root loci (AT < 0).
4. Breakaway points: The breakaway points of the complete root loci are
found from Eq. (8-187); that is,

from which we have the two breakaway points at

s=jf [-(T+2)±*/T*
1
+4]

For T= 1 sec the two breakaway points are at

j = -0.382 ^ = -2.618
442 / Root Locus Techniques Chap. 8

s-plane

+ ^K

+ <-K

Fig. 8-40. Complete root loci for s 2 +s+ Ke~ T ° = 0, T= 1.

where it is easily verified that one belongs to the root loci, and the other
is on the complementary root loci.

The complete root loci of the system are sketched as shown in Fig. 8-40 for T 1 =
sec.Notice that from the system analysis standpoint, only the portion of the root loci
that lies between CO =n
and CO = — n is of significance. The closed-loop system of
Fig. 8-39 is stable for <: K< 1.157. Therefore, the other root loci branches, including
the complementary root loci, are perhaps only of academic interest.

Approximation of Systems with Pure Time Delay

Although the discussions given in the preceding section point to a system-

atic way of constructing the root loci of a closed-loop system with pure time
delay, in general, for complex systems, the problem can still be quite difficult.
 :

Root Loci of Systems with Pure Time Delay / 443

Sec. 8.7

Ts
We shall investigate ways of approximating the time delay term, e~ , by a poly-
Ts
nomial or a rational function of 5. One method is to approximate e' ,
as

follows

e " = [1 +
1

(Tsln)Y
(8-205)

Since e' Ts has an infinite number of poles, the approximation is perfect when n
becomes infinite. Figure 8-41 illustrates the effect of the approximation when
the input to the pure time delay is a unit step function.

(b)

Fig. 8-41. Waveforms that illustrate the effect of approximating a pure

time delay by finite number of poles.

If Eq. (8-205) is used as the approximation for the root locus problem, only
the primary branches of the root loci will be realized. However, this will be
adequate for the great majority of practical problems, since only the primary
branches will contain the dominant eigenvalues of the system.
Let us approximate the exponential term of Eq. (8-189) by the right side of
Eq. (8-205). Figure 8-42 illustrates the dominant root loci for n 2, 3, and =
4; T =1, together with the primary branch of the exact root loci. The approxi-
mating root loci approach the exact ones as n becomes large.
Another way of approximating the pure time delay transfer relation is to
use a power series; that is,

= - 1 Ts + T2!s
2 2
7V + • (8-206)
3!

The and that of Eq. (8-205) is that

difference between this approximation
in the former, the accuracy improves as the order n becomes larger, whereas in
the present case, the validity of the approximation depends on the smallness of
T. It is apparent that Eq. (8-206) can be conveniently applied if only a few
terms of the series are used.
 / Root Locus Techniques
Chap. 8

/oo .

s-plane

Fig. 8-42. Approximation of the root loci of 5 + Ke~ s = by those of

(1 + s/n)" +K= 0.

8.8 Relationship Between Root Loci and the Polar Plot

In Chapter 7 the polar plot of a transfer function G(s)H(s) is shown to be a map-

ping of the imaginary axis of the s-plane onto the G(s)H(s) plane. In general,
other trajectories in the s-plane can be mapped onto the G(s)H(s)-plane through
the function G(s)H(s), such as in the case of a system with pure time delay.
upon the function G(s)H(s), we can
Since the properties of the root loci depend
regard the root loci as a mapping from the G(s)H(s) plane to the .y-plane. Since
the points on the complete root loci of the equation
Sec. 8.i Relationship Between Root Loci and the Polar Plot / 445

1 + G(s)H(s) = + KG ^H i{s) =
1 (8-207)

satisfy the conditions

KGi(s)Hi(s) = (2k +l)n K>0 (8-208)

KG 1 (s)Hi(s) = 2kn K<0 (8-209)

the root loci simply represent a mapping of the real axis of the G(s)H(s) plane
onto the j-plane. In fact, for the root loci, K > 0, the mapping points are on
the negative real axis of the j-plane; and for the complementary root loci,
K < 0, the mapping points are on the positive real axis of the j-plane.
It was shown in Chapter 7 that the mapping from the j-plane to the G(s)H(s)
plane via the polar plot or the Nyquist plot is single-valued. In other words,
a point cw = 0Ji on the imaginary axis of the s-plane is mapped onto only one
point, G(ja>i)H(jcoi), in the G(s)H(s)-plane. However, in the root locus case,
which is a reverse process, the mapping is multivalued. As an illustration, the
polar plot of a type 1 transfer function of the third order is shown in Fig. 8-43.

/ Im GH j , /CO
s-plane

Mapping from s to GH-pXsme

ReGH

Fig. 8-43. Polax plot of G(s)H(s) = K/[s(s + a)(s + b)] interpreted as a

mapping of the /to axis of the s-plane onto the G(j)i/(i)-plane.

The entire polar plot in the G(s)H(s)

is interpreted as the mapping of the entire

ja> axis of the j-plane.The complete root locus diagram for the same system is
shown in Fig. 8-44 as a mapping of the real axis of the G(s)H(s)-plam onto
the j-plane. Note that in this case each point of the G(s)H(s)-plans corresponds
to three points in the -y-plane. For instance, the (— 1,/0) point of the G(s)H(s)-
plane corresponds to the two points where the root loci intersect the jco axis
and a point on the real axis.
The polar plot and the root loci each represents the mapping of only a very
limited portion of one domain to the other. In general, it would be useful to
consider the mapping of points other than those on the jco axis of the s-plane and
on the real axis of the G(s)H(s)-plam. For instance, we may use the mapping
446 / Root Locus Techniques Chap. 8

.
/'
Im GH
G(s)H(s)-p\ane

/ ReGH
lGH=2kv^ Rp (
lGH=(2k+ 1)tt

Fig. 8-44. Root locus diagram of G(s)H(s) = Kj[s(s + a)(s + b)] inter-
preted as a mapping of the real axis of the G(s)H(s)-p\ane onto the j-plane.

of the constant-damping-ratio lines onto the G(s)H(s)-plane for the purpose of

determining relative stability of the closed-loop system. Figure 8-45 illustrates
the G(s)H(s) plots that correspond to different constant-damping-ratio lines in
the j-plane. As shown by curve (3) in Fig. 8-45, when the G(s)H(s) curve passes

/ Im GH /CO

G(s)H(s)-pUne s-plane

A root of the
characteristic equation

ReGH

Fig. 8-45. G(s)H(s) plots that correspond to constant-damping ratio lines

in the s-plane.
 Sec. 8.9 Root Loci of Discrete-Data Control System / 447

through the (— l,y"0) point, it means that Eq. (8-207) is satisfied, and the corre-
sponding trajectory in the j-plane passes through a root of the characteristic
equation. Similarly, we may construct root loci that correspond to straight
lines rotated at various angles from the real axis in the G(s)H(s)-p\ane, as shown
by Fig. 8-46. Notice that these root loci now satisfy the condition of

/KG^H^s) = (2k +1)ti -6 K> 0. (8-210)

Or, the root loci of Fig. 8-46 satisfy the equation

1 + G(s)H(s)e je = (8-211)

for the various values of 8 indicated.

/ Im GH
s-plane

Fig. 8-46. Root loci that correspond to different phase-angle loci in the
G(s)H(s)-p\a.ne. (The complementary root loci are not shown.)

It is of interest to note that Eq. (8-210) is very similar to the condition

of Eq. (8-183), which is for the root loci of systems with pure time delays.
In Eq. (8-183), the angle that is added to (2k +
\)n is a function of the fre-
quency co.

8.9 Root Loci of Discrete- Data Control System 4

The root locus technique for continuous-data systems can be readily applied
to discrete-data systems without requiring any modifications. The characteristic
equation roots of the discrete-data system having the block diagram of Fig. 8-47
448 / Root Locus Techniques Chap. 8

Fig. 8-47. Discrete-data control system.

must satisfy

1 + GH*(s) = (8-212)

if the roots are defined in the s-plane, or

1 + GH{z) = (8-213)

if the z-plane is referred to.

Since

GH*(s)^-L
J n=-°o
2 G(s + jnca )H(s + jnto.)
s
(8-214)

which an infinite series, the poles and zeros of GH*(s) in the j-plane will be
is

number. This evidently makes the construction of the root loci of Eq.
infinite in

(8-212) more difficult. However, as an illustrative example of the difference

between the characteristics of the root loci of continuous-data and discrete-
data systems, let us consider that for the system of Fig. 8-47,

G(s)H(s) (8-215)
s(s + 1)

Using Eq. (8-214), we have

GH*(s) = -L±^ A (8-216)

+ jncOs)(s + jna> + 1)
(s s

which has poles at s = —jnco s 1and s = — — jnco where n takes on all integers
s ,

between -oo and oo. The pole configuration of GH*(s) is shown in Fig. 8-48(a).
Using the rules of construction outlined earlier, the root loci of + GH*(s) = 1

for positive K are drawn as shown in Fig. 8-48(b) for T= 1 . The root loci con-
tain an infinite number of branches, and these clearly indicate that the closed-
loop system is unstable for all values of K greater than 4.32. In contrast, it is

well known that the same system without sampling is stable for all positive
values of K.
The root locus problem for discrete-data systems is simplified if the root
Sec. 8.9 Root Loci of Discrete-Data Control System / 449

.s-plane K ,/co s-plane

X-- — :;-

'_JL °°^K ^K = 4.32 a:-*°°

/2co s
--)r
3co s K- X

2
oo<-£ K-too

/" s I

x- it

oo<-a: ^K = 4.32 K-k*>

K= ,K =
x, a
7
T <~*-K ^-K = 4.32
.
L

--X it ,K =
3 co t

'V oo^a: ^-K = 432 *:->«>

/2cos
--* it K=
5co.
-/-
°°<-a: ^K = 4.32 K^-<*>

-;3cos .

(a)
(b)

Fig. 8-48. Pole configuration of GH*(s) and the root locus diagram in the
.s-plane for the discrete-data system in Fig. 8-47 with G(s)H (s) =
K/[s(s + 1)], T= 1 sec.

loci are constructed in the z-plane using Eq. (8-213). Since Eq. (8-213) is, in
general, a polynomial in z with constant coefficients, the number of root loci is
finite, and the same rules of construction for continuous-data systems are directly
applicable.
As an illustrative example of the construction of root loci for discrete-data
systems in the z-plane, us consider the system of Fig. 8-47 with T
let 1 sec, =
and G(s)H(s) as given by Eq. (8-215). Taking the z-transform of Eq. (8-215),
we have

which has a zero at z and poles at z = 1 and z =

0.368. The root loci for =
the closed-loop system are constructed based on the pole-zero configuration of
Eq. (8-217) and are shown in Fig. 8-49. Notice that when the value of AT exceeds
4.32, one of the roots of the characteristic equation moves outside the unit
circle in the z-plane, and the system becomes unstable.
450 / Root Locus Techniques Chap. 8

/ Imz

z-plane

f-path (f = 50%)

Rez

Unit
circle

Fig. 8-49. Root locus diagram of discrete-data control system without

zero-order hold. G(s)H(s) = K/[s(s + 1)], T= 1 sec.

, , / Im z

z-plane

Unit
circle

Rez

Fig. 8-50. Root locus diagram of discrete-data control system without

zero-order hold. G(s)H(s) = Kl[s(s + 1)], T= 5 sec.

For the same system, if the sampling period is changed to T= 5 sec, the z-
transform of G(s)H(s) becomes

= 0.993Kz
GH(z) (8-218)
(z - l)(z - 0.0067)
Sec. 8.9 Root Loci of Discrete-Data Control System / 451

The root loci for this case are drawn as shown in Fig. 8-50. It should be noted
that although the complex part of the root loci for T= 5 sec takes the form of

a smaller circle than that when T = 1 sec, the system is actually less stable.
The marginal value of K for stability for T= 5 sec is 2.02 as compared to the
marginal K of 4.32 for T = 1 sec.

The constant-damping-ratio path 48 may be superimposed on the root

loci to determine the required value of K for a specified damping ratio. In
Fig. 8-49 the constant-damping-ratio path for £ = 0.5 is drawn, and the
intersection with the root loci gives the desired value of K = 1 Thus for all .

values of K less than 1 the damping ratio of the system will be greater than 50
,

per cent.
As another example, let us consider that a zero-order hold is inserted
between the sampler and the controlled process G(s) in the system of Fig. 8-47.
For the loop transfer function of Eq. (8-2 1 5), the z-transform with the zero-order
hold is

GM GH{z) _ K[(T
- 1 + e~ T )z - Te~ T + -
1 e~ T ]
(8-219)
(z - l)(z - e-*)

The root loci of the system with sample-and-hold for T = 1 sec and T = 5
sec are shown in Fig. 8-5 1(a) and (b), respectively. In this case the marginal

• . /Iraz

z-plane

Rez

(a) Root loci for T= 1 sec

Fig. 8-51. Root locus diagrams of discrete-data control system with sam-
ple-and-hold. G(s)H(s) = K/ls(s + 1)]. (a) Root loci for T= 1 sec. (b)
Root loci for T= 5 sec.
452 / Root Locus Techniques
Chap. 8

z-plaue

(b) Root loci for T= 5 sec

Fig. 8-51 (Cont.).

value of stability for K is 2.3 for T= 1 sec and 0.66 for T= 5 sec. This illus-

trates the well-established fact that the zero-order hold reduces the stability
margin of a discrete-data system.

REFERENCES

General Subjects

1. W. R. Evans, "Graphical Analysis of Control Sytems," Trans. AIEE, Vol. 67,

pp. 547-551, 1948.

2. W. R. Evans, "Control System Synthesis by Root Locus Method," Trans. AIEE,

Vol. 69, pp. 66-69, 1950.

3. W. R. Evans, Control System Dynamics, McGraw-Hill Book Company, New

York, 1954.

4. Y. Chu and V. C. M. Yeh, "Study of the Cubic Characteristic Equation by the

Root Locus Method," Trans. ASMS, Vol. 76, pp. 343-348, Apr. 1954.

5. J. G. Truxal, Automatic Feedback Systems Synthesis, McGraw-Hill Book Com-

pany, New York, 1955.
6. K.S. Narendra, "Inverse Root Locus, Reversed Root Locus, or Complementary
Root Locus," IRE Trans. Automatic Control, pp. 359-360, Sept. 1961.
Chap. 8 References / 453

7. P. Mosner, J. Vullo, and K. Bley, "A Possible Source of Error in Root-Locus

Analysis of Nonminimum Phase Systems," IRE Trans. Automatic Control, Vol.
AC-7, pp. 87-88, Apr. 1962.
8. B. J. Matkowsky and A. H. Zemanian, "A Contribution to Root-Locus Tech-
niques for the Analysis of Transient Responses," IRE Trans. Automatic Control,
Vol. AC-7, pp. 69-73, Apr. 1962.

9. J. D. S. Muhlenberg, "Quartic Root Locus Types," IEEE Trans. Automatic

Control, Vol. AC-12, pp. 228-229, Apr. 1967.

10. H. M. Power, "Root Loci Having a Total of Four Poles and Zeros," IEEE
Trans. Automatic Control, Vol. AC-16, pp. 484-486, Oct. 1971.

Construction and Properties of The Root Locus

11. C. C. MacDuffe, Theory of Equations, John Wiley & Sons, Inc., New York, pp.
29-104, 1954.

12. G. A. Bendrikov and K. F. Teodorchik, "The Laws of Root Migration for

Third and Fourth Power Linear Algebraic Equations as the Free Terms Change
Continuously," Automation and Remote Control, Vol. 16, No. 3, 1955.

13. J. E. Gibson, "Build a Dynamic Root-Locus Plotter," Control Eng., Feb. 1956.

14. C. S. Lorens and R. C. Titsworth, "Properties of Root Locus Asymptotes,"

IRE Trans. Automatic Control, AC-5, pp. 71-72, Jan. 1960.

15. K. F. Teodorchik and G. A. Bendrikov, "The Methods of Plotting Root Paths

of Linear Systems and for Qualitative Determination of Path Type," Proc. IFAC
Cong., Vol. 1, pp. 8-12, 1960.

16. C. A. Stapleton, "On Root Locus Breakaway Points," IRE Trans. Automatic
Control, Vol. AC-7, pp. 88-89, Apr. 1962.
17. M. J. Remec, "Saddle-Points of a Complete Root Locus and an Algorithm for
Their Easy Location in the Complex Frequency Plane," Proc. Natl. Electronics
Conf, Vol. 21, pp. 605-608, 1965.

18. C. F. Chen, "A New Rule for Finding Breaking Points of Root Loci Involving
Complex Roots," IEEE Trans. Automatic Control, Vol. AC-10, pp. 373-374,
July 1965.

19. V. Krishnan, "Semi-analytic Approach to Root Locus," IEEE Trans. Automatic

Control, Vol. AC-U, pp. 102-108, Jan. 1966.

20. R. H. LaBounty and C. H. Houpis, "Root Locus Analysis of a High-Gain

Linear System with Variable Coefficients; Application of Horowitz's Method,"
IEEE Trans. Automatic Control, Vol. AC-11, pp. 255-263, Apr. 1966.

21. R. J. Fitzgerald, "Finding Root-Locus Breakaway Points with the Spirule,"

IEEE Trans. Automatic Control, Vol. AC-11, pp. 317-318, Apr. 1966.

22. J. D. S. Muhlenberg, "Synthetic Division for Gain Calibration on the Complex

Root Locus," IEEE Trans. Automatic Control, Vol. AC-11, pp. 628-629, July
1966.

23. J. Feinstein and A. Fregosi, "Some Invariances of the Root Locus," IEEE
Trans. Automatic Control, Vol. AC-14, pp. 102-103, Feb. 1969.
454 / Root Locus Techniques Chap 8

24. A. Fregosi and J. Feinstein, "Some Exclusive Properties of the Negative Root
Locus," IEEE Trans. Automatic Control, Vol. AC-14, pp. 304-305, June 1969.

25. H. M. Power, "Application of Bilinear Transformation to Root Locus Plotting,"

IEEE Trans. Automatic Control, Vol. AC-15, pp. 693-694, Dec. 1970.

Analytical Representation of Root Loci

26. G. A. Bendrikov and K. F. Teodorchik, "The Analytic Theory of Constructing

Root Loci," Automation and Remote Control, pp. 340-344, Mar. 1959.

27. K. Steiglitz, "An Analytical Approach to Root Loci," IRE Trans. Automatic
Control, Vol. AC-6, pp. 326-332, Sept. 1961.

28. C. Wojcik, "Analytical Representation of Root Locus," Trans. ASME, J. Basic

Engineering, Ser. D, Vol. 86, Mar. 1964.
29. C. S. Chang, "An Analytical Method for Obtaining the Root Locus with Positive
and Negative Gain," IEEE Trans. Automatic Control, Vol. AC-10, pp. 92-94,
Jan. 1965.

30. B. P. Bhattacharyya, "Root Locus Equations of the Fourth Degree," Internat.

J. Control, Vol. 1, No. 6, pp. 533-556, 1965.

31. J. D. S. Muhlenberg, "Some Expressions for Root Locus Gain Calibration,"

IEEE Trans. Automatic Control, Vol. AC-12, pp. 796-797, Dec. 1967.

Computer-Aided Plotting of Root Loci

32. D. J. Doda, "The Digital Computer Makes Root Locus Easy," Control Eng.,
May 1958.

33. Z. Klagsbrunn and Y. Wallach, "On Computer Implementation of Analytic

Root-Locus Plotting," IEEE Trans. Automatic Control, Vol. AC-13, pp. 744-745,
Dec. 1968.

34. R. H. Ash and G. R. Ash, "Numerical Computation of Root Loci Using the
Newton-Raphson Technique," IEEE Trans. Automatic Control, Vol. AC-13, pp.
576-582, Oct. 1968.

Root Locus Diagram for Design

35. Y.Chu, "Synthesis of Feedback Control Systems by Phase Angle Loci," Trans.
AIEE, Vol. 71, Part II, 1952.
36. F. M. Reza, "Some Mathematical Properties of Root Loci for Control Systems
Design," Trans. AIEE Commun. Electronics, Vol. 75, Part I, pp. 103-108, Mar.
1956.

37. J. A. Aseltine, "Feedback System Synthesis by the Inverse Root-Locus Method,"

IRE Natl. Convention Record, Part 2, pp. 13-17, 1956.

38. G. A. Bendrikov and K. F. Teodorchik, "The Theory of Derivative Control in

Third Order Linear Systems," Automation and Remote Control, pp. 516-517, May
1957.

39. R. J. Hruby, "Design of Optimum Beam Flexural Damping in a Missile by

Application of Root-Locus Techniques," IRE Trans. Automatic Control, Vol.
AC-5, pp. 237-246, Aug. 1960.
Chap. 8 Problems / 455

40. A. Katayama, "Semi-graphical Design of Servomechanisms Using Inverse Root-

Locus," /. IEE, Vol. 80, pp. 1140-1149, Aug. 1960.

41. D. A. Calahan, "A Note on Root-Locus Synthesis," IRE Trans. Automatic

Control, Vol. AC-7, p. 84, Jan. 1962.

42. A. Katayama, "Adaptive Control System Design Using Root-Locus," IRE

Trans. Automatic Control, Vol. AC-7, pp. 81-83, Jan. 1962.

43. J. R. Mitchell and W. L. McDaniel, Jr., "A Generalized Root Locus Follow-
ing Technique," IEEE Trans. Automatic Control, Vol. AC-15, pp. 483-485, Aug.
1970.

Root Sensitivity

44. J. G. Truxal and I. M. Horowitz, "Sensitivity Considerations in Active Network

Synthesis," Proc. Second Midwest Symposium on Circuit Theory, East Lansing,
Mich., Dec. 1956.

45. R. Y. Huang, "The Sensitivity of the Poles of Linear Closed-Loop Systems,"

Trans. AIEE Appl. Ind., Vol. 77, Part 2, pp. 182-187, Sept. 1958.

46. H. Ur, "Root Locus Properties and Sensitivity Relations in Control Systems,"
IRE Trans. Automatic Control, Vol. AC-5, pp. 57-65, Jan. 1960.

Root Locus for Discrete- Data Systems

47. M. Mori, "Root Locus Method of Pulse Transfer Function for Sampled-Data
Control Systems," IRE Trans. Automatic Control, Vol. AC-3, pp. 13-20, Nov.
1963.

48. B. C. Kuo, Analysis and Synthesis of Sampled-Data Control Systems, Prentice-

Hall, Inc., Englewood Cliffs, N.J., 1963.
49. B. C. Kuo, Discrete-Data Control Systems, Science-Tech, P.O. Box 2277, Station
A, Champaign, 111. 1974.

Root Locus for Nonlinear Systems

50. M. J. Abzug, "A Root-Locus Method for the Analysis of Nonlinear Servomech-
anisms," IRE Trans. Automatic Control, Vol. AC-4, No. 3, pp. 38-44, Dec.
1959.

Root Locus for Systems with Time Delays

51. Y. Chu, "Feedback Control System with Dead-Time Lag or Distributed Lag by
Root-Locus Method," Trans. AIEE, Vol. 70, Part II, p. 291, 1951.

PROBLEMS

8.1. Sketch the root locus diagram for each of the following feedback control sys-
tems. In each case determine everything about the locus of roots for — oo <
K< co and sketch the root loci. Indicate on each locus the starting point, the
ending point, and the direction of increasing value of K. The poles and zeros
of G(s)H(s) of the systems are given as follows:
456 / Root Locus Techniques
chap 8

(a) Poles at 0, -2, and —3; zero at -5.

(b) Poles at 0, 0, -2, and -3; zero at -5.
(c) Poles at -2 +j2, and -2 -j2; zero at -3.
(d) Poles at 0, -10 +7IO, and -10 -ylO; zero at -20.
(e) Poles at 0, -20, -10 + /10, and -10 -yiO; no finite zeros.
(f) Poles at -20, -10 +yi0, and -10 -ylO; zero at -30.
(g) Poles at 0, 0, -12, and -12; zeros at -4 and -8.
8.2. The open-loop transfer function of a unity-feedback control system is given by

K '
s(s 2 + 2s + 2)(s + 5)(s + 6)

(a) Sketch the root locus diagram as a function of K(— 00 < K < 00).
(b) Determine the value of K
that makes the relative damping ratio of the
closed-loop complex poles equal to 0.4.

8.3. A unity feedback control system has an open-loop transfer function

K + °- 2s)( + 0-0255)
C( w\- s\l <1
+ 0.001.0(1 + 0.0050
1

Sketch the complete (—00 < K < co) root locus diagram for the system. Indi-
cate the crossing points of the loci on the j CO axis, and the corresponding values
of K and co at these points.

8.4. A unity feedback control system has an open-loop transfer function

G(i) =
s(\ + 0.020(1 + 0.010
(a) Sketch the root locus diagram of the system (0 < K < co).
(b) Determine the marginal value of K that will cause instability.
(c) Determine the value of K when the system is critically damped.
8.5. The transfer functions of a feedback control system are given as

G(S) = s\s + 2)(s + 5)

and "W = '

(a) Sketch the root locus diagram for the system. Indicate the crossing points
of the loci on they© axis and the corresponding values of and co at these K
points (positive values of K only).
(b) The transfer function of the feedback loop element is now changed to
H(s) = 1+2^. Determine the stability of the modified system as a function
of K. Investigate the effect on the root locus diagram due to this change in
H{s).

8.6. The characteristic equation of a feedback control system is given by

s 3
+ 3.s
2
+(K + 2)s + 10K =
Sketch the root locus diagram (positive K only) for this system.

8.7. For the following loop transfer function, sketch the root locus diagram as a
function of T (T varies from to 00). Determine the value of T so that the
damping ratio of the complex roots of the characteristic equation is 0.2.

ww =
G(s)H(s)
s(l
1000(1 + Ts)
+ 0.10(1 + 0.0010
Chap. 8 Problems / 457

8.8. For the following loop transfer function, sketch the root locus diagram as a
function of T. Determine the value of Tso that the damping ratio of the complex
roots of the characteristic equation is 0.2.

30
G(s)H(s) = j(1 + Q U)(1 + a2j)(1 + Ts)

8.9. For the bridged-T network of Fig. P8-9,

C2

+ o- -VW\r -VW\r- -o +
R R
±C,

-o-
Figure P8-9.

(a) Sketch the root locus diagrams of the zeros and poles of E1 jE l as a function
of d
(C, varies from to oo).
(b) Sketch the root locus diagrams of the zeros and poles of £2 /£i as a func-
tion of C 2 .

8.10. The open-loop transfer function of a control system with positive feedback is

given by

G(s) =
s(s 2 + 4s + 4)
Sketch the root locus diagram of the system as a function of K (0 < K < oo).

8.11. It is desired that the closed-loop transfer function of a control system be

C(s) 1
R(s) ~ (1 + 0.03*)(1 + 0.2s + 0.02s 2
)

Determine the open-loop transfer function G(s) of the system. Assume that the
system has unity feedback.

8.12. Given the equation

s3 + as 1 + Ks +K=
It is desired to investigate the root loci of this equation for — oo < K < co and
for several values of a.
(a) loci (-co < K < oo) for a = 10.
Sketch the root
(b) Repeat for a = 3.
(c) Determine the value of a so that there is only one nonzero breakaway point

on the complete root loci. Sketch the loci.

In sketching the root loci you should apply all known rules whenever they are
applicable.

8.13. The open-loop transfer function of a control system with unity feedback is given
by

G(s) = K{s + a)
s 2 (s + 1)
 )

458 / Root Locus Techniques

Chap. 8

Determine the values of a so that the root locus diagram will have zero, one,
and two breakaway points, respectively, not counting the one at s = 0. Sketch
the root loci for — co < K < co for all three cases.
8.14. For the sampled-data control system shown in Fig. P8-14,

G(s) = is:

s(l + 0.2s)

r(t)
z.o.h. G(s) + c(t)
T

Figure P8-14.

(a) Sketch the root loci for the system (0 <K< co) without the zero-order
hold, for T= 0.1 sec and T= 1 sec. Determine the marginal value of K
for stability in each case.
(b) Repeat part (a) when the system has a zero-order hold.
8.15. The following polynomial in z represents the characteristic equation of a certain
discrete-data control system. Sketch the root loci ( — oo < K < co) for the
system. Determine the marginal value of K for stability.

z3 + Kz 2 + \.5Kz -(K+\)=0
8.16. Sketch the root loci (0 <K< co) in the z-plane for the discrete-data control
system shown in Fig. P8-16.

r(t)
^ ^ z.o.h.
s(s
Ke -0.is
+ )(s + 3
c(t)
j-/ T= 0. 1 sec
1

Figure P8-16.
 9
Frequency-Domain Analysis
of Control Systems

9.1 Introduction

It was pointed out earlier that in practice the performance of a feedback control
system is more preferably measured by its time-domain response characteristics.

This is in contrast to the analysis and design of systems in the communication

field, where the frequency response is of more importance, since in this case

most of the signals to be processed are either sinusoidal or periodic in nature.

However, analytically, the time response of a control system is usually difficult
to determine, especially in the case of high-order systems. In the design aspects,
there are no unified ways of arriving at a designed system given the time-domain
specifications, such as peak overshoot, rise time, delay time, and settling time.
On the other hand, there a wealth of graphical methods available in the
is

frequency-domain analysis, all suitable for the analysis and design of linear
feedback control systems. Once the analysis and design are carried out in the
frequency domain, the time-domain behavior of the system can be interpreted
based on the relationships that exist between the time-domain and the fre-
quency-domain properties. Therefore, we may consider that the main purpose
of conducting control systems analysis and design in the frequency domain is
merely to use the techniques as a convenient vehicle toward the same objectives
as with time-domain methods.
The starting point in frequency-domain analysis is the transfer function.
We shall first discuss transfer function relations based on, first, the state variable
representation and, second, the classical approach.
In Section 3.3 the transfer function of a multivariable closed-loop control
system is derived. Referring to Fig. 3-9, the closed-loop transfer function matrix

459
460 / Frequency-Domain Analysis of Control Systems Chap. 9

relation is written [Eq. (3-44)]

CO) = [I + G(j)H(j)]- »G(j)R(j) (9-1)

where CO) is a9 x 1 vector and R0) is ap X 1 vector. The closed-loop transfer

function matrix is defined as

MO) = [I + G(j)H(*)]" »G(*) (9-2)

which is a q x p matrix.
Under the sinusoidal steady state, we set s = joa; then Eq. (9-2) becomes
M(./ca) = [I + GUaWVmfi-iGUm) (9-3)

The yth element of M(jca) is defined as

where 1 represents the

MM =
row and j the
^
1X J\JU') all

column of M(jco).
other inputs=0
(9-4)

State-Variable Representation

For a system that is represented by state equations,

x(r) = Ax(f) + Bu(r) (9-5)

c(0 = Dx(?) + Eu(0 (9-6)

For the present case we assume that the feedback is described by

u(?) = r(0 - Hc(0 (9-7)

where
x(?) = nxl state vector

UW = P X 1 control vector

c(0 = 9 X 1 output vector

r 0) = x /> 1 input vector

A, B, D, and E are constant matrices of appropriate dimensions, and H is the

p X g feedback matrix. The transfer function relation of the system is

CO) = [DOI - A)"»B + E]U0) (9-8)

The open-loop transfer function matrix is defined as

GO) = DOI - A)-«B + E (9-9)

The closed-loop transfer function relation is described by the equation

CO) = [I + G0)H]" G0)R0) 1

(9-10)
Thus
MO) = [I + G0)H]-'G0) (9-11)

It should be noted that the matrix H in Eq. (9-1 1) has only constant elements.
Sec. 9.1 Introduction / 461

In general, the elements of the transfer function matrices are rational func-
tions of s. In Chapter 4 it is proved that if the system is completely controllable

and observable, there will be no pole-zero cancellations in the transfer func-

tions. Under this condition the poles of the transfer function will also be the
eigenvalues of the system.
The analysis techniques in the frequency domain discussed in the following

sections are conducted with the single-variable notation. Because linear systems
satisfy the principle of superposition, these basic techniques can all be applied
to multivariable systems.
For a single-loop feedback system, the closed-loop transfer function is

written

^>=§)= r+wm (9 " 12)

Under the sinusoidal steady state, we set s = jco; then Eq. (9-12) becomes
M( fm) - C^°^ - G Jc°) (
U)
MUC0) ~ RUa>) ~ + GUcoWJco)
1
(9-13)
{ *

The sinusoidal steady-state transfer relation M(jca), which is a complex

function of co, may be expressed in terms of a real and an imaginary part; that is,

M(jco) = Re [M(jco)] + j Im [M(jco)] (9-14)

Or, M(jco) can be expressed in terms of its magnitude and phase as

M(jco) = M{Q>) l<t> m {(o) (9-15)

where
G(jco)
(9-16)
1 + GUco)HUa>)
and

<l>m(co)
= Gjjco)
+ G(j(o)HUco)
1 (9-17)

= IG{jg>) - /l + GUa>)H{j<o)
Since the analysis is now in the frequency domain, some of the terminology
used in communication systems may be applied to the present control system
characterization. For instance, M(co) of Eq. (9-16) may be regarded as the
magnification of the feedback control system. The significance of M(a>) to a
control system is similar to the gain or amplification of an electronic amplifier.
In an audio amplifier, for instance, an ideal design criterion is that the amplifier
must have a flat gain for all frequencies. Of course, realistically, the design
criterion becomes that of having a flat gain in the audio frequency range. In
control systems the ideal design criterion is similar. If it is desirable to keep
the output C(jco) identical to the input R(jco) at all frequencies, M(ja>) must be
unity for all frequencies. However, from Eq. (9-13) it is apparent that M(jco)
 462 / Frequency-Domain Analysis of Control Systems Chap. 9

can be unity only when G(joi) is infinite, while H(jco) is finite and nonzero. An
infinite magnitude for G(jco) is, of course, impossible to achieve in practice, nor
would it be desirable, since most control systems become unstable when its
loop gain becomes very high. Furthermore, all control systems are subjected to
noise. Thus, in addition to responding to the input signal, the system should be
able to rejectand suppress noise and unwanted signals. This means that the
frequency response of a control system should have a cutoff characteristic in
general, and sometimes even a band-pass characteristic.
The phase characteristics of the frequency response are also of importance.
The ideal situation is that the phase must be a linear function of frequency
within the frequency range of interest. Figure 9-1 shows the gain and phase
characteristics ofan ideal low-pass filter, which is impossible to realize phys-
Typical gain and phase characteristics of a feedback control system are
ically.

shown in Fig. 9-2. The fact is that the great majority of control systems have
the characteristics of a low-pass filter, so the gain decreases as the frequency
increases.

M(a>) .

*m(")

Fig. 9-1. Gain-phase characteristics of an ideal low-pass filter.

M(u)

Fig. 9-2. Typical gain and phase characteristics of a feedback control system.

9.2 Frequency-Domain Characteristics

If a control system is to be designed or analyzed using frequency-domain

techniques, we need a
of specifications to describe the system performance.
set
The following frequency-domain specifications are often used in practice.
Sec. 9.2 Frequency-Domain Characteristics / 463

Peak resonance M p. The peak resonance M p is defined

as the maximum
value of M(co) that is given in Eq. (9-16). In general, the magnitude of M p
gives

an indication of the relative stability of a feedback control system. Normally,

a large M
p corresponds to a
large peak overshoot in the step response. For
most design problems it is generally accepted that an optimum value of M p
should be somewhere between 1.1 and 1.5.

Resonant frequency <a r The resonant frequency

. co p is defined as the fre-
quency at which the peak resonance p occurs. M
Bandwidth. The bandwidth, BW, is defined as the frequency at which the
magnitude of M(jco), M(co), drops to 70.7 per cent of its zero-frequency level,
or 3 dB down from the zero-frequency gain. In general, the bandwidth of a
control system indicates the noise-filtering characteristics of the system. Also,
bandwidth gives a measure of the transient response properties, in that a large
bandwidth corresponds to a faster rise time, since higher-frequency signals are
passed on to the outputs. Conversely, if the bandwidth is small, only signals
of relatively low frequencies are passed, and the time response will generally be
slow and sluggish.

Cutoff rate. Often, bandwidth alone is inadequate in the indication of the

characteristics of the system in distinguishing signals from noise. Sometimes it

may be necessary to specify the cutoff rate of the frequency response at the high
frequencies. However, in general, a steep cutoff characteristic may be accom-
panied by a large M
p which, corresponds to a system with a low stability margin.
The performance criteria defined above for the frequency-domain analysis
are illustrated on the closed-loop frequency response, as shown in Fig. 9-3.
There are other criteria that may be used to specify the relative stability
and performance of a feedback control system. These are defined in the ensuing
sections of this chapter.

M(cj)

0.707

Fig. 9-3. Typical magnification curve of a feedback control system.

464/ Frequency- Domain Analysis of Control Systems Chap. 9

9.3 M p . CO,, and the Bandwidth of a Second-Order System

For a second-order feedback control system, the peak resonance p the reso- M ,

nant frequency co p and the bandwidth are all uniquely related to the damping
,

ratio £ and the natural undamped frequency co„ of the system. Consider the
second-order sinusoidal steady-state transfer function of a closed-loop system,

M(jco) = C Jco) =
( ^
1 + Aeo/aOC - (a/a*) 2

We may simplify the last expression by letting u = cojco n - Then, Eq. (9-18)
becomes

The magnitude and phase of M(jco) are

M(yM) = M(M) = _ (9-20)

| |

[(1 m2)2 ^ (2Cm)2]1/2

and

MM = 0„(«) = -tan- '

^Lj (9-21)

The resonant frequency is determined first by taking the derivative of M(u)

with respect to u and setting it equal to zero. Thus

^^ = -i-[(l - u 2) 2 + (2£u) 2 ]- 3/2 (4u 3 - 4« + 8«C

2
) = (9-22)

from which
4m 3 - 4m + SuC 2 = (9-23)

The roots of Eq. (9-23) are

u, = (9-24)
and
u, = J\- 2£ 2 (9-25)

The solution in Eq. (9-24) merely indicates that the slope of the M(co) versus co
curve is zero at co =
0; it is not a true maximum. The solution of Eq. (9-25)
gives the resonant frequency,

co, = co.Vl - 2£ 2
(9-26)

Since frequency a real quantity, Eq. (9-26) is valid only for 1

is 2
2£ or >
£ <; 0.707. This means simply that for all values of £ greater than 0.707, the
solution of (o p= becomes the valid one, and p 1 M =
Sec. 9.3 MB , a>„, and the Bandwidth of a Second-Order System / 465

Substituting Eq. (9-25) into Eq. (9-20) and simplifying, we get

M„ = 1
(9-27)
2CV1 - C
2

It is important to note that M„ is a function of £ only, whereas co p is a function

of £ and a>„.
It should be noted that information on the magnitude and phase of M(jco)

of Eq. (9-18) may readily be derived from the Bode plot of Eq. (A-51), Figs.
A-10 and A-12. In other words, Fig. A-12 is an exact representation of Eq.
(9-21). The magnitude of M(jco), however, may be represented in decibels versus
frequency, which is the Bode plot in Fig. A-10, or it may be plotted in absolute
magnitude, M(co), versus co. Figure 9-4 illustrates the plots of M(u) of Eq. (9-20)

u = to/co„

Fig. 9-4. Magnification-versus-normalized frequency of a second-order

closed-loop control system.
466 / Frequency-Domain Analysis of Control Systems Chap. 9

M„

Damping ratio f

Fig. 9- 5. M p -v ersus-damping ratio for a second-order system, Mp

1/(2{V1 -£ 2 ).

versus u for various values of f . Notice that if the fre-

quency scale were unnormalized, the value of co p would
increase when £ decreases, as indicated by Eq. (9-26).
When £ = 0, co p and co„ become identical. Figures 9-5
and 9-6 illustrate the relationship between M p
and £, and
u = co p /co„ and f , respectively.
As defined in Section 9.2, the bandwidth, BW, of a
system is the frequency at which M (<y) drops to 70.7 per
cent of its zero-frequency value, or 3 d B down from the
zero-frequency gain. Equating Eq. (9-20) to 0.707, we
have
1
M(u) ~ 0.707 (9-28)
0.5 0.707 1.0
[(i - u*y + (2t;uyy

Damping ratio J Thus

[(i - u*y + (2{k) 2 F 2
(9-29)
Normalized resonant frequency-
Fig. 9-6.
versus-damping ratio for a second-order This equation leads to
system, u p = */\ — 2£ 2
=
.

w2 (l 2C
2
) ± V4C - 4
4£ 2 + 2 (9-30)

In the last expression the plus sign should be chosen, since u must be a positive
real quantity for any f. Therefore, from Eq. (9-30), the bandwidth of the
second-order system is determined as

BW = a>.[(l - 2C + V4C - 4£ + 2
)
4 2
2]>
2
(9-31)

Figure 9-7 gives a plot of BW/co„ as a function of f It is of interest to note that .

for a fixed co„, as the damping ratio £ decreases from unity, the bandwidth
increasesand the resonance peak p also increases. M
For the second-order system under consideration, we easily establish some
simple relationships between the time-domain response and the frequency-
domain response of the system.
 .

Sec. 9.4 Effects of Adding a Zero to the Open-Loop Transfer Function / 467

2.0

0.4 0.6
Damping ratio f

Fig. 9-7. Bandwidth/evversus-damping ratio for a second-order system,

BW = <»„[(1 - 2C + </4f-4f 2 + 2]i/2.
2
)

1 The maximum overshoot of the unit step response in the time domain
depends upon £ only [Eq. (6-96)].
2. The resonance peak of the closed-loop frequency response M„
depends upon £ only [Eq. (9-27)].
3. The rise time increases with £, and the bandwidth decreases with
the increase of £, for a fixed con , Eq. (6-102), Eq. (9-31), Fig. 6-14,
and Fig. 9-7. Therefore, bandwidth and rise time are inversely
proportional to each other.
4. Bandwidth is directly proportional to co„.
5. Higher bandwidth corresponds to larger M p .

9.4 Effects of Adding a Zero to the Open-Loop Transfer Function

The relationships established between the time-domain and the frequency-

domain responses arrived at in Section 9.3 are valid only for the second-order
closed-loop transfer function of Eq. (9-18). When other transfer functions or
more parameters are involved, the relationships between the time-domain and
the frequency-domain responses are altered and may be more complex. It is of
interest to consider the effects on the frequency-domain characteristics of a
feedback control system when poles and zeros are added to the open-loop
transfer function. It would be a simpler procedure to study the effects of adding
 BW 3

Fig. 9-8. Bandwidth of a second-order system with open-loop transfer

function G(s) = (1 + Ts)l[s(s + 1.414)].

1.2
1 1
1 1 | 1 1 1 | 1 1 i
1 1 1

. (1 + Ts)
G(s)'
1.0
s(s+ 1.414) —
-
X vv

0.8 — ^\5 —
-£~L_
0.707

0.6 — —
3
5=
- -

0.4 £^£^4
^^
-
r=o.i
^. T=0.0l ^2rr^"
0.2 — ~

i i 1 1 1 1 i i i i
1 1 I 1 i

co rad/sec

Fig. 9-9. Magnification curves for a second-order system with an open-

loop transfer function G(s).

468
Sec. 9.4 Effects of Adding a Zero to the Open-Loop Transfer Function / 469

cj rad/sec

Fig. 9-10. Magnification curves for a second-order system with an open-

loop transfer function G(s) = (1 + Ts)/[s(s + 0.4)].

poles and zeros to the closed-loop transfer function. However, it is more realistic
to consider modifying the open-loop transfer function directly.
The closed-loop transfer function of Eq. (9-18) may be considered as that
of a unity feedback control system with an open-loop transfer function of

G(s)
ml (9-32)
s{s + 2£co„)

Let us add a zero at s = —1/Tto the last transfer function so that Eq. (9-32)
becomes

<?(*) =
2
ft)„ (l + Ts)
(9-33)
s(s + 2£c0„)

This corresponds to the second-order system with derivative control studied in

Section 6.7. The closed-loop transfer function of the system is given by Eq.
(6-156), and is repeated here:

M(s)
_Qs) coljl + Ts) (9-34)
R(s) s
2
+ (2£c»„ + TcoDs + ml
In principle, M p , cop , and BW of the system can all be derived using the
 470 / Frequency-Domain Analysis of Control Systems Chap. 9

1.2
i |
r
r=o

-
7^2—-^^^=sa^^
0.8
" y^T = L414
^^r^

S 0.6

0.4 /T=0A /y\

Iff r= o.oi

0.2

I 1 i 1 i 1 i

Fig. 9-11. Unit step responses of a second-order system with an open-loop

transfer function G(s) = (1 + Ts)/[s(s + 1.414)].

same However, since there are now three

steps as illustrated in Section 9.3.
parameters in and T, the exact expressions for
£> ©„, p co p and BW are M , ,

difficult to obtain even though the system is still of the second order. For
instance, the bandwidth of the system is

BW = (^ + {^ + 46,:)' (9-35)

where
b = 4C
2 2
co„ + 3
4Cco „T - 2col - 4
o>„ r (9-36)

to see how each of the parameters in Eq. (9-35) affects the

It is difficult

bandwidth. Figure 9-8 shows the relationship between and T for £ = 0.707 BW
and o>„ =
1. Notice that the general effect of adding a zero to the open-loop

transfer function is to increase the bandwidth of the closed-loop system. How-

ever, for small values of T over a certain range the bandwidth is actually
decreased. Figures 9-9 and 9-10 give the plots for M(co) for the closed-loop
system that has the G(s) of Eq. (9-33) as its open-loop transfer function; co
n 1, =
Tis given various values, and £ 0.707 and £ =
0.2, respectively. These curves =
show that for large values of T the bandwidth of the closed-loop system is
 Sec. 9.5 Effects of Adding a Pole to the Open-Loop Transfer Function / 471

"i I
i
I
r
r= o.2

^ 0.6

Time (second)

Fig. 9-12. Unit step responses of a second-order system with an open-loop

transfer function G{s) = (1 + Ts)j[s(s + 0.4)].

increased, whereas there exists a range of smaller values of T in which the BW

is decreased by the addition of the zero to G(s). Figures 9-11 and 9-12 show the
corresponding unit step responses of the closed-loop system. The time-domain
responses indicate that a high bandwidth corresponds to a faster rise time.
However, as T becomes very large, the zero of the closed-loop transfer function,
which is at s = — l/T, moves very close to the origin, causing the system to have
a large time constant. Thus Fig. 9-11 illustrates the situation that the rise time
is fast but the large time constant of the zero near the origin of the s-plane
causes the time response to drag out in reaching the final steady state.

9.5 Effects of Adding a Pole to the Open-Loop Transfer Function

The addition of a pole to the open-loop transfer function generally has the
effect of decreasing the bandwidth of the closed-loop system. The following
transfer function is arrived at by adding (1 + Ts) to the denominator of Eq.
(9-32):

G(s) = s(s co„

(9-37)
+ 2Cco„)(l + Ts)
472 / Frequency-Domain Analysis of Control Systems Chap. 9

3.0
i r i r 1 i r "i r

s(s+ I.414)(1 + Ts)

Fig. 9-13. Magnification curves for a third-order system with an open-

loop transfer function G(s)= \j[s(s + 1.414)(1 + Ts)].

The derivation of the bandwidth of the closed-loop system which has the
G(s) of Eq. (9-37) as its open-loop transfer function is quite difficult. It can be
shown that the BW is the real solution of the following equation:

T 2 co s + (1 + 4£
2
co
2
r 2 )co* + (4£
2 2
a>„ - 2<a„
2
- 4£cw„
3
2> - co„ =2 4
(9-38)

We may obtain a qualitative indication on the banciwidth properties by

referring to Fig. 9-13, which shows the plots for M(co) for co„ 1, £ 0.707, = =
and various values of T. Since the system is now of the third order, it can be
unstable for a certain set of system parameters. However, it can be easily shown
by use of the Routh-Hurwitz criterion that for co„ 1 and £ 0.707, the system = =
is stable for all positive values of T. The M(co) versus co curves of Fig. 9-13 show
that for small values of T, the bandwidth of the system is slightly increased but
M„ is also increased. When T becomes large, the pole added to G{s) has the
effect of decreasing the bandwidth but increasing M p . Therefore, we may
conclude that, in general, the of adding a pole to the open-loop transfer
effect
function is to make the closed-loop system less stable. The unit step responses
of Fig. 9-14 clearly show that for the larger values of T,T=\ and T= 5, the
 Sec. 9.6 Relative Stability — Gain Margin, Phase Margin, and Mp / 473

1 1 1 1 1 1 1 i

1.4 -

1.0
T=0.s// J

0.6

-**^ l 1
1 1
i i i
i

8 10 12 14 16

Time (seconds)

Fig. 9-14. Unit step responses of a third-order system with an open-loop

transfer function G{s) = \\\s(s + 1.414)(1 + Ts)].

rise time increases with the decrease of the bandwidth, and the larger values of
M B also correspond to greater peak overshoots in the step responses. However,
important to point out that the correlation between
it is Mp and the peak over-
shoot is meaningful only when the closed-loop system is stable. When the
magnitude of G(jco) equals unity, M (co) is infinite, but if the closed-loop system
is unstable with G(jco)
| |
> 1 at IG(jco) = 180°, M(a>) is finite and can assume an
arbitrarily small number.
The two sections is to demonstrate the simple rela-
objective of these last
tionships between the bandwidth and M
p and the characteristics of
the time-
,

domain response. The effects of adding a pole and a zero to the open-loop
transfer function are discussed. However, no attempt is made to include all
general cases.

9.6 Relative Stability— Gain Margin. Phase Margin, and Mp

We have demonstrated in the last three sections the general relationship between
the resonance peak M„ of the frequency response and the peak overshoot of
the step response. Comparisons and correlations between frequency-domain
and time-domain parameters such as these are useful in the prediction of the
performance of a feedback control system. In general, we are interested not
only in systems that are stable, but also in systems that have a certain degree of
stability. The latter is often termed relative stability. In many situations we may
use M p to indicate the
relative stability of a feedback control system. Another
474 / Frequency-Domain Analysis of Control Systems Chap. 9

way of measuring relative stability of a closed-loop system is by means of the

Nyquist plot of the loop transfer function, G(s)H(s). The closeness of the
G(j(o)H{joi) plot in the polar coordinates to the (— 1, y'O) point gives an indica-
tion of how stable or unstable the closed-loop system is.
To demonstrate the concept of relative stability, the Nyquist plots and the
corresponding step responses and frequency responses of a typical third-order

G(/co)//0'cj)-plane ) Im GH

RtGH

K=K

(a) Stable and well-damped system.

G(/co)//(;co)-plane I / Im GH

(b) Stable but oscillatory system.

Fig. 9-15. Correlation among Nyquist plots, step responses, and fre-
quency responses.
 Sec. 9.6 Relative Stability— Gain Margin, Phase Margin, and Mp / 475

C(/'cjM/w)-plane i I Im GH

(c) Marginally unstable system.

G(/co)#(/a))-plane / Im GH

c(/)l

ReGH 1

(d) Unstable system.

Fig. 9-15 (Cont.). Correlation among Nyquist plots, step responses, and
frequency responses.

system are shown in Fig. 9-15 for four different values of loop gain K. Let us
consider the caseshown in Fig. 9- 15(a), in which the loop gain K is low, so the
Nyquist plot of G(s)H(s) intersects the negative real axis at a point (the phase-
crossover point) quite far away from the ( — 1, JO) point. The corresponding step
response is shown to be quite well behaved, and Mp is low. As K is increased,
Fig. 9-1 5(b) shows that the phase-crossover point is moved closer to the (— 1 , jO)
point ; the system is still stable, but the step response has a higher peak over-
476 / Frequency-Domain Analysis of Control Systems Chap. 9

shoot, and M p
is larger. The phase curve
for <j> m does not give as good an indica-
tion of relative stability as
p
except that M
one should note the slope of the <f> m
,

curve which gets steeper as the relative stability decreases. The Nyquist plot of
Fig. 9-15(c) intersects the (— 1, jO) point, and the system is unstable with con-
stant-amplitude oscillation, as shown by the step response; M p becomes infinite.
If AT is increased still further, the Nyquist plot will enclose the (— l,/0) point,
and the system is unstable with unbounded response, as shown in Fig. 9-1 5(d).
In this case the magnitude curve M(co) ceases to have any significance, and the
only symptom of instability from the closed-loop frequency response is that the
phase curve now has a positive slope at the resonant frequency.

Gain Margin

To give a quantitative way of measuring the relative distance between the

G(s)H(s) plot and the (— \,j0) point, we define a quantity that is called the
gain margin.
Specifically, the gain margin is a measure of the closeness of the phase-
crossover point to the (— l,yO) point in the G(s)H(s)-p\ane, With reference to
Fig. 9-16, the phase-crossover frequency is denoted by co c , and the magnitude of
G(joo)H(j(o) at co = co c is designated by G(jco c )H(jco c )
| |. Then, the gain margin

/ Im GH

G(/co)//(/co) -plane

Phase crossover

CO = CO
ReGH

Fig. 9-16. Definition of the gain margin in the polar coordinates.

Sec. 9.6 Relative Stability — Gain Margin, Phase Margin, and Mp I 477

of the closed-loop system that has G(s)H(s) as its loop transfer function is

defined as

gain margin = G.M. = 20 log,

^^j, dB (9-39)

On the basis of this definition, it is noticed that in the G(jco)H(jco) plot of

Fig. 9-16, if the loop gain is increased to the extent that the plot passes through
the (— 1, y'0) point, so that G(ja> c )H(jco c ) equals unity, the gain margin
| |
becomes
dB. On the other hand, if the G(jco)H(jcoi) plot of a given system does not
intersect the negative real axis, | G(jco c )H(ja> c ) equals zero, and the gain margin
|

denned by Eq. (9-39) Based on the above evaluation, the

is infinite in decibels.
physical significance of gain margin can be stated as follows Gain margin is :

the amount of gain in decibels that can be allowed to increase in the loop before
the closed-loop system reaches instability.
When theG(jco)H(jco) plot goes through the (— 1 jO) point, the gain margin ,

is dB, which implies that the loop gain can no longer be increased as the
system is already on the margin of instability. When the G(jco)H(jco) plot does
not intersect the negative real axis at any finite nonzero frequency, and the
Nyquist stability criterion indicates that the (— 1 jO) point must not be enclosed ,

for system stability, the gain margin is infinite in decibels; this means that,
theoretically, the value of the loop gain can be increased to infinity before
instability occurs.
When the (— 1 , jO) point is to the right of the phase-crossover point, the
magnitude of G{jooc)H(jco c) is greater than unity, and the gain margin as given
by Eq. (9-39) in decibels is negative. In the general case, when the above men-
tioned condition implies an unstable system, the gain margin is negative in
decibels. It was pointed out in Chapter 7 that if G(s)H(s) has poles or zeros in
the right half of the j-plane, the (— 1, jO) point may have to be encircled by the
G(jco)H(jco) plot in order for the closed-loop system to be stable. Under this
condition, a stable system yields a negative gain margin. In practice, we must
first determine the stability of the system (i.e., stable or unstable), and then the

magnitude of the gain margin is evaluated. Once the stability or instability

condition is ascertained, the magnitude of the gain margin simply denotes the
margin of stability or instability, and the sign of the gain margin becomes
insignificant.

Phase Margin

The gain marginis merely one of many ways of representing the relative

of a feedback control system. In principle, a system with a large gain

stability
margin should be relatively more stable than one that has a smaller gain margin.
Unfortunately, gain margin alone does not sufficiently indicate the relative
stability of all systems, especially if parameters other than the loop gain are
variable. For instance, the two systems represented by the G(jco)H(jco) plots of
Fig. 9-17 apparently have the same gain margin. However, locus A actually
478 / Frequency- Domain Analysis of Control Systems Chap. 9

i / Im GH
G(/cj)//(/to) -plane

*- ReGH

Fig. 9-17. Nyquist plots showing systems with same gain margin but dif-
ferent amount of relative stability.

corresponds to a more stable system than locus B. The reason is that with any
change in a system parameter (or parameters) other than the loop gain, it is
easier for locus B to pass through or even enclose the (— 1, jO) point. Fur-
thermore, system B has a much larger p than system A.M
In order to strengthen the representation of relative stability of a feedback
control system, we define the phase margin as a supplement to gain margin.
Phase margin is defined as the angle in degrees through which the G(jco)H(jco)
plot must be rotated about the origin in order that the gain-crossover point on the
locus passes through the (—/,/)) point. Figure 9-18 shows the phase margin as
the angle between the phasor that passes through the gain-crossover point and
the negative real axis of the G(jco)H(jco)-plane. In contrast to the gain margin,
which gives a measure of the effect of the loop gain on the stability of the closed-
loop system, the phase margin indicates the effect on stability due to changes of
system parameters, which theoretically alter the phase of G(joS)H(jco) only.
The analytical procedure of computing the phase margin involves first

the calculation of the phase of G(jco)H(jco) at the gain-crossover frequency, and

then subtracting 180° from this phase; that is,

phase margin = 0.M. = IGUco )H{j(o

g g)
180° (9-40)

where co g denotes the gain-crossover frequency. If the (— 1 , ;0) point is encircled

Sec. 9.1 Relative Stability — Gain Margin, Phase Margin, and Mp / 479

i/ImGff

G(/'co)//(/a>) -plane

*- ReGH

Fig. 9-18. Phase margin defined in the GO'ct>)//(y'co)-plane.

by the G(jco)H(j(o) plot, the gain-crossover point would be found in the second
quadrant of the G(y'co)//(jca)-plane, and Eq. (9-40) would give a negative phase
margin.

Graphic Methods of Determining Gain Margin and Phase Margin

Although the formulas for the gain margin and the phase margin are
simple to understand, in practice it is more convenient to evaluate these quan-

tities graphically from the Bode plot or the magnitude-versus-phase plot. As an

illustrative example, consider that the open-loop transfer function of a control

system with unity feedback is given by

10
G(s) = (9-41)
s(l + 0.02j)(1 + 0.2s)

The Bode plot of G(j(o) is shown in Fig. 9-19. Using the asymptotic approxima-

tion of | G(jco) |,
the gain-crossover and the phase-crossover points are deter-
mined as shown in the figure. The phase-crossover frequency is approximately
16 rad/sec, and the magnitude of G(jco) at this frequency is about — 1 5 dB. This
means that if the loop gain of the system is increased by 15 dB, the magnitude
curve will cross the 0-dB axis at the phase-crossover frequency. This condition
480 / Frequency-Domain Analysis of Control Systems Chap. 9

20 Gain crossover
-
.^^

1
^v^ Gain margin

20
^>^- 15dB -

3 40 -

60 \.
80 ^^
00 ^^ -

in i i
!
i

10 ., 50 100 1000
w rad/sec
1 1
i
i

-90°

Phase margin ^-s^f Phase crossover

180°
T
270°

i I
1

10 16 50 100 1000

co rad/sec

Fig. 9-19. Bode plot of G(s) = 10/[j(l + 0.2s)(l + 0.02s)].

corresponds to the Nyquist plot of G(jco) passing through the (— 1, jO) point,
and the system becomes marginally unstable. Therefore, from the definition of
the gain margin, the gain margin of the system is 15 dB.
To determine the phase margin, we note that the gain-crossover frequency
isat co = 7 rad/sec. The phase of Gfjco) at this frequency approximately is

— 125°. The phase margin is the angle the phase curve must be shifted so that it

will pass through the —180° axis at the gain-crossover frequency. In this case,

0.M. = 180° - 125° = 55° (9-42)

In general, the procedure of determining the gain margin and the phase
margin from the Bode plot may be outlined as follows

1. The gain margin is measured at the phase-crossover frequency co c :

G.M. = -|GCA» )#(M)l e

dB (9-43)
Sec. 9.6 Relative Stability— Gain Margin. Phase Margin, and M„ / 481

I I
1

u=l7
Gain crossover /
\ 5 /
Phase 7 ^^
v- margin
XQ
yr*
Gain jT
margin

20
^r 16

>^30
<^40 / cj rad/sec
•o

40
A60

fioo

60 -

T300

80

OO

00 1 1
1

-270 225 180 135 90

Phase (degrees)

Fig. 9-20. Magnitude-versus-phase plot of G(s) = 10/[i(l + 0.2s)(l +

0.02s)].

2. The phase margin is measured at the gain-crossover frequency co g :

0.M. = 180° + IG(jcQ g )H(jg} g ) (9-44)

The gain and phase margins are even better illustrated on the magnitude-
versus-phase plot. For the transfer function of Eq. (9-41), the magnitude-
versus-phase plot is shown in Fig. 9-20, which is constructed by use of the data
from the Bode plot of Fig. 9-19. On the magnitude-versus-phase plot of G(jco),
the phase crossover point is where the locus intersects the —180° axis, and the
gain crossover is where the locus intersects the 0-dB axis. Therefore, the gain

margin is simply the distance in decibels measured from the phase crossover to
 •a
o
CO

o
"3,

a
D.

4>
a3 .
fa- t/i

c .S
so as

fa o

<s
83
. J3
E ^c
IS
J--
—
2? >»
TO C
*o '5b

"8 e

. cS

°"o
1-. c
cs cs

P c
CS
BO
00
c
. %
M g
_2 S "5

.a
&

482
 Sec. 9.7 Relative Stability as Related to the Slope / 483

- 180°
Phase
(c) Magnitude-versus-phase plot.

Fig. 9-21 (Cont.). Polar plot, Bode diagram, and magnitude-versus-phase

plot, showing gain and phase margins in these domains.

the critical point at dB and — 1 80°, and the phase margin is the horizontal
distance in degrees measured from the gain crossover to the critical point.

In summarizing, the relations between the measurements of the gain margin

and the phase margin in the polar plot, the Bode plot, and the magnitude-
versus-phase plot are illustrated in Fig. 9-21.

9.7 Relative Stability As Related to the Slope of the Magnitude

Curve of the Bode Plot

In general, a definite relation between the relative stability of a closed-loop

system and the slope of the magnitude curve of the Bode plot of G(jco)H(ja>) at
if the loop gain
the gain crossover can be established. For example, in Fig. 9-19,
of the system is decreased from the nominal value, the gain crossover may be
moved which the slope of the magnitude curve is only —20
to the region in
dB/decade the corresponding phase margin of the system would be increased.
;

On the other hand, if the loop gain is increased, the relative stability of the sys-
tem will deteriorate, and if the gain is increased to the extent that the gain
crossover occurs in the region where the slope of the magnitude curve is —60
dB/decade, the system will definitely be unstable. The example
is a cited above
simple one, since the slope of the magnitude curve decreases monotonically as
co increases. Let us consider a conditionally stable system for the purpose of
484 / Frequency-Domain Analysis of Control Systems Chap. 9

-20

-40

40 dB/decade

V~ 60 dB/decade

1000

1000

Fig. 9-22. Bode plot of GO) = [K(l + 0.2s)(l + 0.025s)]/[s 3 (l + 0.01s)

(1 + 0.005.S)].

illustrating relative stability. Consider that a control system with unity feedback
has the open-loop transfer function

~ M _ s*(1 ++ 0.01j)(1 ++ 0.025s)

yj ~~ 3
(l
0.2j)(1
0.005s)
(9-45)

The Bode plot of G(s) is shown in Fig. 9-22 for K= 1 The gain-crossover
.

frequency is 1 rad/sec, and the phase margin is negative. The closed-loop system
is unstable even for a very small value of K. There are 1 wo phase crossover
points: one at co = 15 rad/sec and the other at a> = 12C rad/sec. The phase
characteristicsbetween these two frequencies indicate that if the gain crossover
lies in this is stable. From the magnitude curve of the Bode
range, the system
plot, the range of K for stable operation is found to be between 60 and 90 dB.
For values of K above and below this range, the phase lag of G(jco) exceeds
— 180° and the system is unstable. This system serves as a good example of the
relation between relative stability and the slope of the ma gnitude curve at the
gain crossover. As observed from Fig. 9-22, at both very low and very high
 Sec. 9.8 Constant M Loci in the G(/»)-Plane / 485

frequencies, the slope of the magnitude curve is — 60dB/decade; if the gain

crossover falls in either one of these two regions, the phase margin becomes
negative and the system is unstable. In the two sections of the magnitude curve
that have a slope of —40 dB/decade, the system is stable only if the gain cross-
over about half of these regions, but even then the resultant phase margin
falls in

is small. However, if the gain crossover falls in the region in which the magnitude

curve has a slope of —20 dB/decade, the system is stable.

The correlation between the slope of the magnitude curve of the Bode plot
at the gain crossover and the relative stability can be used in a qualitative way
for design purposes.

9.8 Constant M Loci in the G(/CO)-Plane

In previous sections it was shown that the resonance peak M p of the closed-loop
frequency response is directly related to the maximum overshoot of the transient
response. Normally, the magnification curve of M(co) versus co may be con-
structed by the method if the closed-loop transfer function M(s) is
Bode plot
given and if its numerator and denominator are in factored form. Unfortunately,
this isnot the usual case, as the open-loop transfer function G(s) is normally
given. For the purpose of analysis we can always obtain the magnification curve
by digital computation on a computer. However, our motivation is to be able
to predictM p from the plots of G(jco), and eventually to design a system with a
specified M p
.

Consider that the closed-loop transfer function of a feedback control

system with unity feedback is given by

M(s) - ^- G^ (9-46)

For sinusoidal steady state, G(s) = G(jco) is written

G(jai) = Re G(jco) + j Im G(jco)

(9-47)
= * +jy
Then
G(jco)
M(fi)) = \MUco)\
1 + G(jco)
(9-48)
Jx* + y
2

V(i + x? + y
1

For simplicity, let M = M(co); then Eq. (9-48) leads to

MV(1 + x) 2 + y
2
= V* + 2
y
1
(9-49)

Squaring both sides of the last equation gives

M 2
[(l + xY + y
2
] = x2 + y
2
(9-50)
486 / Frequency-Domain Analysis of Control Systems Chap. 9

Rearranging this equation yields

(1 - M )x + - M )y ~ 2M x = M
2 2
(1
2 2 2 2
(9-51)

This equation is conditioned by dividing through by (1 — M 1

) and adding the
term [M 2 /(l - M 2
)]
2
on both sides. We have

x2 + yv -
+ 2
2Afl
i-m 2Xx + ({t=W )V = r=m ±(
+ d^ip)V
M1
2
M1 M2 ,o
(
™
9 -52)

which is finally simplified to

For a given M, Eq. (9-53) represents a circle with the center at x =

M 2
/(l M
- 2 ), y = 0. The radius of the circle is r = M/( - 2 ) Equation | M |.

(9-53) is invalid for M

= 1. For = 1, Eq. (9-50) gives M
x =~j (9-54)

which is the equation of a straight line parallel to they Im G(jco) axis and passing
through the (— £, JO) point in the G(;co)-plane.
When M
takes on different values, Eq. (9-53) describes in the G0'co)-plane
a family of circles that are called the constant loci or the constant circles. M M
The coordinates of the centers and the radii of the constant loci for various M
values of Mare given in Table 9-1, and some of the loci are shown in Fig. 9-23.

Table 9-1 Constant M Circles

M „
Center x — M 2
— M
_ M v 1

j 2
-M 2
\

0.3 0.01 0.33

0.5 0.33 0.67
0.7 0.96 1.37
1.0 oo CO
1.1 -5.76 5.24
1.2 -3.27 2.73
1.3 -2.45 1.88
1.4 -2.04 1.46
1.5 -1.80 1.20
1.6 -1.64 1.03
1.7 -1.53 0.90
1.8 -1.46 0.80
1.9 -1.38 0.73
2.0 -1.33 0.67
2.5 -1.19 0.48
3.0 -1.13 0.38
4.0 -1.07 0.27
5.0 -1.04 0.21
6.0 -1.03 0.17
Sec. 9.8 Constant M Loci in the G(y'a>)-Plane / 487

, . / Im G

G(/co)-plane

M = 0.833

ReG

Fig. 9-23. Constant M circles in the polar coordinates.

Note that when M becomes infinite, the circle degenerates into a point at the
critical point, This agrees with the well-known fact that when the
(— l,y'0).
Nyquist plot of G(jco) passes through the (— 1 jO) point, the system is marginally
,

unstable and M
p is infinite.
Figure 9-23 shows that the constant M loci in the

G(7'co)-plane are symmetrical with respect to the M= and the 1 line real axis.

The to the
circles of the M = locus correspond to values of M greater than
left 1

and those to the

1, of the M =
right are for M1 than
line less 1.

Graphically, the intersections of the G(jco) plot and the constant M give loci

the value of M at the frequency denoted on the G(jco) curve. desired to If it is

keep the value of M than a certain value, the G(jco) curve must not
p less
intersect

the corresponding M any point, and

circle at the same time must not enclose
at

the (— point. The constant M

1, j'0) with the smallest radius that tangent
circle is

to the G(jco) curve gives the value of M and the resonant frequency
p,
read co p is

off at the tangent point on the G(ja>) curve.

Figure 9-24(a) illustrates the Nyquist plot of G(jco) for a unity feedback
control system, together with several constant loci. For a given loop gain M
K= K u the intersects between the G(jco) curve and the constant loci give M
the points on the M(co)-versus-co The peak resonance
curve. p is found by M
locating the smallest circle that is tangent to the G(jco) plot. The resonant fre-
quency is found at the point of tangency and is designated as co pl If the loop .

gain is increased to K 2 and if the system is still stable, a constant

,
circle with M
a smaller radius that corresponds to a larger M is found tangent to the G(jco)
488 / Frequency-Domain Analysis of Control Systems Chap. 9

1/ Im G

G(/'cj) -plane

M=M l

M >M2

ReG

Fig. 9-24. Polar plots of G(s) and constant M

loci showing the procedure
of determining M
p and the magnification curves.

curve, and thus the peak resonance M

p is larger. The resonant frequency is
shown which is closer to the phase-crossover frequency co than co
to be co p2 ,
c pl .

When K is increased to K
so that the G(jco) curve passes through the (-1,
3
JO)
point, the system is marginally unstable, M
p is infinite, and co pl co c In all = .

cases the bandwidth of the closed-loop system is found at the

intersect of the
G(jco) curve and the M=
0.707 locus. For values of K
beyond 3 the system is K ,

unstable, and the magnification curve and M

p no longer have any meaning.
When enough points of intersections between the G(Jco) curve and the constant
M loci are obtained, the magnification curves are plotted
as shown in Fis s '

9-24(b).
 Sec. 9.9 Constant Phase Loci in the G(/»)-Plane / 489

9.9 Constant Phase Loci in the G(/(0)-Plane

The of constant phase of the closed-loop system may also be determined in

loci
the G(jco)-p\ane by a method similar to that used to secure the constant loci. M
With reference to Eqs. (9-46) and (9-47), the phase of the closed-loop system
is written as

m (co) = /Mijco) = tan" «

(X) - tan" ' (9-55)
<t>
(j-f^)
Taking the tangent on both sides of the last equation and letting </> m = <f> m (co),
we have

W- ^ + S + j,
(^6)

Let N= tan (f> m ; then Eq. (9-56) becomes

xi + x + y*--£ = (9-57)

Adding the term (1/4) + (1/4N 1 ) to both sides of Eq. (9-57) yields

l ^y --l+-An=^ + Ai
2
(9-58)

which is regrouped to give

2

(* + ±) + {y -2^ = 4 + 4^ <9
- 59 >

When N assumes various values, this equation represents a family of circles

with centers at (x, y) = (—1/2, 1/2AT)- The radii are given by

(N 2 + iy /2
(*&r <»°>

The centers and the radii of the constant N circles for various values of N
are tabulated in Table 9-2, and the loci are shown in Fig. 9-25.

Table 9-2 Constant N Circles

180° n
=
AT = tan rm
m Center x
2"y =
= — -^-, jtt Radius r = J-
n 0,1,2... 2N \ AN 2
-90 — oo 0.500
-60 -1.732 -0.289 0.577
-45 -1.000 -0.500 0.707
-30 -0.577 -0.866 1.000
-15 -0.268 -1.866 1.931
OO OO
15 0.268 1.866 1.931
30 0.577 0.866 1.000
45 1.000 0.500 0.707
60 1.732 0.289 0.577
90 OO 0.500
490 / Frequency-Domain Analysis of Control Systems Chap. 9

,,/ImG
<t>= 15° (- 165°) G(/to)-plane

*- ReG

Fig. 9-25. Constant N circles in the polar coordinates.

9.10 Constant M and N Loci in the Magnitude-Versus-Phase

Plane —The Nichols Chart
In principle we need both the magnitude and the phase of the closed-loop
frequency response to analyze the performance of the system. However, we have
shown that the magnitude curve, which includes such information as
p co p M , ,

and BW, normally is more useful for relative stability studies.

A major disadvantage in working with the polar coordinates for the G(jco)
plot is that the curve no longer retains its original shape when a simple modifica-
tion such as the change of the loop gain is made to the system. In design prob-
lems, frequently not only the loop gain must be altered, but series or feedback
controllers are to be added to the original system which require the complete
reconstruction of the resulting G(jco). For design purposes it is far more con-
venient to work in the Bode diagram or the magnitude-versus-phase domain.
In the Bode diagram the magnitude curve is shifted up and down without dis-
tortion when the loop gain is varied; in the magnitude-versus-phase plot, the
Sec. 9.10 Constant M and N Loci in the Magnitude-Versus-Phase Plane / 491

.,/Im G
M= 1

G0'co)-plane

*-ReG

20 log 10 C

Phase
(b)

Fig. 9-26. (a) Constant M circles in the G(/a>)-plane. (b) Nichols chart in
the magnitude-versus-phase coordinates.

entire G(jcai) curve is shifted up or down vertically when the gain is altered. In
addition, the Bode plot can be easily modified to accommodate any modifica-
tions made to G(jco) in the form of added poles and zeros.
The constant M and constant N loci in the polar coordinates may be
transferred to the magnitude-versus-phase coordinates without difficulty.

Figure 9-26 illustrates how this is done. Given a point on a constant M circle

in the G(ya))-plane, the corresponding point in the magnitude-versus-phase

plane may be determined by drawing a vector directly from the origin of the
 Phase

Fig. 9-27. Nichols chart (for phase from -180° to 0°).

G(/a>)-plane to the particular point on the constant M

circle; the length of the
vector in decibels and the phase angle in degrees give the corresponding point
in the magnitude- versus-phase plane. Figure 9-26 illustrates the process of locat-
ing three arbitrary corresponding points on the constant M loci in the magni-
tude-versus-phase plane. The critical point, (— 1,/)), in the G(y'o))-plane corre-
sponds to the point with dB and —180° in the magnitude- versus-phase plane.
492
dB 12

270° - 260° - 240° - 220' 140° - 120° - 100° -90°

Phase

Fig. 9-28. Nichols chart (for phase from -270° to -90°).

Using the same procedure as described above, the constant loci are also N
transferred into the magnitude-versus-phase plane. These constant and M N
2
loci in the magnitude-versus-phase coordinates were originated by Nichols and
called the Nichols chart. A typical Nichols chart is constructed in Fig. 9-27 for
the phase angle that extends from —180° to 0°. The chart that corresponds to
the phase from —360° to —180° is a mirror image of that in Fig. 9-27 with
493
 494 / Frequency-Domain Analysis of Control Systems Chap. 9

respect to the —180° axis. In Fig. 9-28 the Nichols chart is shown for —90° to
— 270°, which is the useful region of phase for many practical control systems.
The values of M on these constant M loci are given in decibels. To determine
bandwidth, the — 3-dB locus should be used.
The following example will illustrate the relationships among the analysis
methods using the Bode plot, the magnitude- versus-phase plot, and the Nichols
chart.

Example 9-1 Let us consider the positional control system discussed in Section 6.6.
When the inductance of the dc motor is 0.1 H, the open-loop transfer
function of the system is given by Eq. (6-149) and is repeated here

25QA
G{s) (9-61)
s(s 2 + 50.5s + 1725)

The Bode plot for G(s) is drawn as shown in Fig. 9-29 for A 200. The gain

dB

41.5 100 463

co rad/sec

Fig. 9-29. Bode plots of the system in Example 9-1

margin and the phase margin are determined from the Bode plot to be 5 dB and 40°,
The data on the magnitude and phase of G(jco) are transferred to the
respectively.
magnitude-versus-phase plot of Fig. 9-30. From Fig. 9-30, the peak resonance
p is M
found to be approximately 5 dB (or 1.78), the resonant frequency is 40 rad/sec, the
bandwidth of the system is 48 rad/sec, and the results for the gain and phase margins
are as given above.
When the motor inductance is reduced to 0.01 henry, the open-loop transfer
 .

Sec. 9.10 Constant M and N Loci in the Magnitude-Versus-Phase Plane / 495

dB -4

-180 -160 -140 -120 -100-90

Phase (degrees)

Fig. 9-30. Magnitude-versus-phase plots for the system in Example 9-1

function of the system becomes

2500,4
G(s) (9-62)
s(.s
2
+ 500s + 17270)

With A = 200, and for the purpose of constructing the Bode plot, G(s) is written

29
G{s) = (9-63)
*(1 + 0.00216s)(l + 0.0268j)
496 / Frequency-Domain Analysis of Control Systems Chap. 9

The Bode plot of G(s) shown in Fig. 9-29. It is shown that by decreasing the
is

inductance, the gain margin improved to approximately 24 dB, and the phase margin
is

is about 50°. The magnitude-versus-phase plot in Fig. 9-30 indicates that the bandwidth
of the system is not noticeably changed, but M
p is reduced to approximately 1 .4 dB,
or 1.18. The frequency-domain results obtained in this example for the two values of
inductance correlate quite well with the time-domain analysis that resulted in the time
responses of Fig. 6-21 Table 9-3 gives a comparison of the time-domain and frequency-
.

domain parameters.

Table 9-3 Comparison of the Time-Domain and Frequency-Domain Performances

of the System in Example 9-1

Peak Rise Delay Gain Phase

L Overshoot Time Time BW Margin Margin
(henrys) (%) (sec) (sec) (radjsec) M„ <o p (dB) (deg)

0.1 41 0.4 0.5 48 1.78 40 5 4p

0.01 11.5 0.6 0.4 50 1.18 35 24 50

9.11 Closed-Loop Frequency Response Analysis of Nonunity

Feedback Systems

The constant M and N loci and the Nichols chart analysis discussed in preceding
sections are limited to closed-loop systems with unity feedback, whose transfer
function is given by Eq. (9-46). When a system has nonunity feedback, the closed-
loop transfer function is

M(s) = Q°) = <M f9-64)

the constant M
and TV loci derived earlier and the Nichols charts of Figs. 9-27
and 9-28 cannot be applied directly. However, we can show that with a slight
modification these loci can still be applied to systems with nonunity feedback.
Let us consider the function

P( S ) - G (s)H(s)
(965)

Comparing Eq. (9-64) with Eq. (9-65), we have

P(s) = M(s)H(s) (9-66)

Information on the gain margin and the phase margin of the system of Eq.
(9-64)can be obtained in the usual fashion by constructing the Bode plot of
G(s)H(s). However, the G(jca)H(jco) curve and the Nichols chart together do not
give the magnitude and phase plots for M(jco), but forP(jco). Since M(jco) and
P(jco) are related through Eq. (9-66), once the plots for P(co) versus co and
/Pijoo) versus co are obtained, the curves for M(co) and <f>„,(co) versus co are
determined from the following relationships:
 Sec. 9.12 Sensitivity Studies in the Frequency Domain / 497

m (co) = IPjjco) - /H(jco) (9-68)

9.12 Sensitivity Studies in the Frequency Domain 4

The frequency-domain study of feedback control systems has an advantage in

that the sensitivity of a transfer function with respect to a given parameter can
be clearly interpreted. We shall show how the Nyquist plot and the Nichols
chart can be utilized for analysis and design of a control system based on sensi-
tivity considerations.

Consider that a control system with unity feedback has the transfer function

M^ = W)=TTW) (9 ' 69)

The sensitivity of M(s) with respect to G(s) is defined as

om,"^
{S)
_
~
dM{s)jM(s) , _
q /U)
(y m
*" dG(s)IG(s)
or
qm(„\ _ dM(s) G(s) (Q-iu

Substituting Eq. (9-69) into Eq. (9-71) and simplifying, we have

S
^ = r+W) (9 " 72)

Clearly, the sensitivity function is a function of the complex variable s.

In general, it is desirable to keep the sensitivity to a small magnitude. From

a design standpoint, it is possible to formulate a design criterion in the following
form:

5*<'>l = (9 " 73)

'

iT+Wl^*
In the sinusoidal steady state, Eq. (9-73) is easily interpreted in the polar
coordinate by a Nyquist plot. Equation (9-73) is written

\l + G(jco)\>l (9 . 74)

Figure 9-31 illustrates the Nyquist plot of a stable closed-loop system.

The constraint on sensitivity given in Eq. (9-74) is interpreted as the condition
that the G(jco) locus must not enter the circle with radius k. It is interesting to
note that the sensitivity criterion is somewhat similar to the relative stability
specifications of gain and phase margins. When the value of k is unity, the
G(jco) locus must be tangent or outside the circle with a unity radius and centered
at the (— 1, jO) point. This corresponds to a very stringent stability requirement,
margin is infinite. On the other hand, if the Nyquist plot of G(jco)
since the gain
passes through the ( — 1, yO) point, the system is unstable, and the sensitivity is
infinite.
498 / Frequency-Domain Analysis of Control Systems Chap. 9

/ImC
G(.s)-plane

Fig. 9-31. Interpretation of sensitivity criterion with the Nyquist plot.

Equation (9-74) and Fig. 9-31 also indicate clearly that for low sensitivity,
the magnitude of G(jco) should be high, which reduces the stability margin.
This again points to the need of compromise among the criteria in designing
control systems.
Although Fig. 9-31 gives a clear interpretation of the sensitivity function
in the frequency domain, in general, the Nyquist plot is awkward to use for
design purposes. In this case the Nichols chart is again more convenient for
the purpose of analysis and design of a feedback control system with a prescribed
sensitivity. Equation (9-72) is written

g-'(Jto)
TO») 1 + G-^ja)
(9-75)

which clearly indicates that the magnitude and phase of Sffijco) can be obtained
by plotting G'^jco) in the Nichols chart and making use of the constant loci M
for constant sensitivity function. Since the vertical coordinate of the Nichols
chart is in decibels, the G'^jco) curve in the magnitude-versus-phase coordinates
can be easily obtained if the G(ja>) is already available, since

|
G" '
C/co) |
dB = - G(jco)
1
|
dB (9-76)

/g-'(ja>) = - /G(jco) (9-77)

As an example, the function G'^jco) for Eq. (9-61), Example

illustrative
9-1, is plotted in the Nichols chart as shown in Fig. 9-32, using the G(jco) plot
in Fig. 9-30. The intersects of the G _1 (yto) curve in the Nichols chart with the
constant M
loci give the magnitudes of Sg(jco) at the corresponding frequencies.
 Sec. 9.12 Sensitivity Studies in the Frequency Domain / 499

dB

Fig. 9-32. Determination of the sensitivity function Sq in the Nichols chart.

Figure 9-32 indicates several interesting points with regard to the sensitivity
function of the feedback control system. The sensitivity function Sg approaches
dB or unity as co approaches infinity. Sg becomes zero as co approaches zero.
A peak value of 8 dB is reached by S% at co = 42 rad/sec. This means that
the closed-loop transfer function is most sensitive to the change of G(jeo) at this
frequency and more generally in this frequency range. This result is not difficult
to comprehend, since from the Nichols chart of Fig. 9-30 it is observed that
the stability and dynamic behavior of the closed-loop system is more directly
governed by the G(jco) curve near co p , which 40 rad/sec. Changes made to
is

portions of G(jco) at frequencies much higher and much lower than 40 rad/sec
are not going to have a great effect on the relative stability of the system directly.
When the loop gain of the system increases, the G(jco) curve is raised in the
Nichols chart domain, and the G _1 (yco) curve must be lowered. If the G(ja>) curve
passes through the critical point at dB and — 1 80°, the system becomes mar-
500 / Frequency-Domain Analysis of Control Systems Chap. 9

ginally unstable; the G' i

{ja>) also passes through the same point, and the sensi-
tivity is infinite.

In this section we have simply demonstrated the use of the Nichols chart
for the analysis of the sensitivity function of a closed-loop system. In a design
problem, the objective may be to find a controller such that the sensitivity due
to certain system parameters is small.

REFERENCES

1. Y. Chu, "Correlation Between Frequency and Transient Response of Feedback

Control Systems," AIEE Trans. Application and Industry, Part II, Vol. 72, p. 82,
1953.

2. H. M. James, N. B. Nichols, and R. S. Phillips, Theory of Servomechanisms,

McGraw-Hill Book Company, New York, 1947.
3. B. H. Willis and R. W. Brockett, "The Frequency Domain Solution of Regu-
lator Problems," IEEE Trans. Automatic Control, Vol. AC-10, pp. 262-267,
July 1965.

4. A. Gelb, "Graphical Evaluation of the Sensitivity Function Using the Nichols

Chart," IRE Trans. Automatic Control, Vol. AC-7, pp. 57-58, July 1962.

PROBLEMS

9.1. The pole-zero configuration of a closed-loop transfer function is shown in

Fig. P9-l(a).
(a) Compute the bandwidth of the system.
(b) A zero is added to the closed-loop system, as shown in Fig. P9-l(b); how
is the bandwidth affected ?

../« ,
.)">

x-plane s-plane

\ s*
•

45° 4?
\
\
-O-
-2

(a) (b)

Figure P9-1.
Chap. 9 Problems / 501

(c) Another pole is inserted on the negative real axis in Fig. P9-l(b), but at a
distance 10 times farther from the origin than the zero; how is the band-
width affected ?

9.2. The specification given on a certain second-order feedback control system is

that the overshoot of the step response should not exceed 25 per cent.
(a) What are the corresponding limiting values of the damping ratio and peak
resonance M p?
(b) Determine the corresponding values for (o p and / max .

9.3. Sketch the closed-loop frequency response M(jco) as a function of frequency

| |

for the systems and responses shown in Fig. P9-3.

xT~~

-<><H 10 C
o-As
s(s + 2)
X^_
i i

(b)

~~x
A
\r* jy
, 1 +2s
s(s
10
+ 2)
C

(c)

1 +0.5x 10
C
s(s + 2)

(d)

Figure P9-3.
502 / Frequency-Domain Analysis of Control Systems Chap. 9

(e) Sketch the unit step response for the system whose |
M |-versus-co curve is as
shown. Assume that the system is of second order.

\M\ |1.5

0.8

i ^*~~~—
2 a.

rad/sec

(e)

Figure P9-3 (Cont.).

9.4. The closed-loop transfer function of a feedback control system is given by

(s) ~ R(s) ~ (1 + 0.01s)(l + 0.055 + 0.0ls T )

(a) Plot the frequency response curve for the closed-loop system.
(b) Determine the peak resonance peak M p and the resonant frequency a>„ of
the system.
(c) Determine the damping ratio £ and the natural undamped frequency co„ of
the second-order system that will produce the same M„ and (O p determined
for the original system.

9.5. The open-loop transfer function of a unity feedback control system is

it
G(i) =
s(\ + 0As)(l + s)
(a) Determine the value of K so that the resonance peak M p of the system is

equal to 1.4.

(b) K
Determine the value of so that the gain margin of the system is 20 dB.
(c) Determine the value of K so that the pliase margin of the system is 60°.
9.6. The open-loop transfer function of a unity feedback control system is

m
uw = s(l + s)(l + 0.01s)
G(s)
+ tS )

Determine the smallest possible value of T so that the system has an infinite
gain margin.

9.7. The open-loop transfer function of a unity feedback control system is

G(i) =
s(l + 0.ls)(.l + 0.0015)
Determine the value of K if the steady-state error of the output position must
be less than or equal to 0.1 per cent for a ramp function input With this value of
K, what are the gain margin and the phase margin of the system? Plot G(s) in
the gain-phase plot and determine the resonance peak p and the resonant M
frequency a>„.

9.8. A random compensation network is added to the forward path of the system
in Problem 9.7, so that now the open -loop transfer function reads
Chap. 9 Problems / 503

K(l + 0.01675)
G(s) =
5(1 + 0.00222j)(1 + 0.1s)(l + 0.0015)
where K is determined in part (a) of Problem 9.7. Plot the gain-phase diagram
of G(s). Evaluate M
p , 0i p , the gain margin, the phase margin, and the band-
width of the compensated system.
9.9. The Bode diagram of the open-loop transfer function G(s) of a unity feedback
control system is shown in Fig. P9-9.

co(rad/sec
0.1
10 3 10 4
Same slope
to cj -»0

/G(/co)

Figure P9-9.

(a) Find the gain margin and the phase margin of the system.
(b) If the open-loop transfer function is changed to e~ Ts G(s), find the value of T
so that the phase margin of the system is 45°. Then find the value of T so
that the gain margin is 20 dB.
(c) What is the velocity error constant of the system in part (a) ? in part (b) ?
 10
Introduction to Control

Systems Design

10.1 Introduction

Design of control systems represents an interesting and complex subject in con-

trol systems studies. In a simplified manner the design problem of control sys-
tems can be described with the aid of the block diagram of Fig. 10-1. The figure

u(0 Controlled c(f)

process G
—»-
Control p Controlled variable
vector vector (output vector)

Fig. 10-1. Block diagram of a controlled process.

shows a controlled process whose output vector c(f) represents q controlled

and the control vector u(f) represents p control signals. The problem
variables,
is to find a set of "appropriate" signals, u(7), so that the controlled variable
vector c(f) behaves as desired. The description of the basic design problem is
simplified by overlooking the possible existence of external disturbances.
Once the desired control vector u(f) for satisfactory controlis determined,

a controller usually needed to generate this control from the reference inputs
is

and the state vector x(r) or output c{i). Figure 10-2 illustrates the block diagram
of a control system whose control vector is derived from the input vector and
the state vector. This type of system is also referred to as one with state feed-
back. The block diagram of Fig. 10-2 is intended only for the purpose of illus-
trating the philosophy of designing control systems, and no attempt is made for
it to include all possible configurations.

504
Sec. 10.1 Introduction / 505

lit) u(0 c(0

Controlled
Controller
process

x(f)

Fig. 10-2. Block diagram of a control system with state feedback.

It is interesting to give a brief review on the history of development of con-

trol system design theory, which may help gain perspective in the understanding
of the subject.
The early stage of the theoretical development of the design of control sys-
tems was characterized by the works of Nyquist, Hall, Nichols, 2 and Bode, 1

who developed such classical methods as the Nyquist plot, Bode diagram, and
Nichols chart. A unique feature of these methods is that they are all graphical
techniques which are conducted in the frequency domain. As was pointed out
earlier that in the design of control systems, it is the time response that is of
importance, rather than the frequency response. The use of the frequency-
domain techniques is simply due to the fact that the graphical techniques are
convenient to apply.
The classical design methods are characterized by first fixing the configura-
tion of the system to be designed. In other words, the designer must first choose
a suitable system configuration. For instance, Fig. 10-3(a) shows the block
diagram of a system with the controller located in the forward path of the sys-
tem. This is a very common practice because of the versatility of the scheme, and
the system is said to have a cascade or series compensation. Figure 10-3(b)

shows another scheme of compensation by having a controller in the feedback

path, and this is often referred to as the feedback compensation. In general,
other configurations, such as having controllers in both the forward path and
the feedback path, may be used if desired. In practice, the controllers or com-
pensators used in control systems may assume a great variety of forms. In the
simple cases, the controller may be passive networks in the form of low-pass,
high-pass, or band-pass filters, or networks with active elements. In elaborate
cases, the controller may even be a mini-computer.
The proper selection of the system configuration as well as the contents of
the controller depend to a great extent on the experience and ingenuity on the
part of the designer. In the frequency-domain design, the design specifications
usually are given in terms of such criteria as gain margin, phase margin, peak
resonance, and bandwidth. These criteria, however, should be related to the
time-domain specifications, such as rise time, overshoot, and settling time,
which are more direct measurements of the system's performance.
The of control systems is very much a trial-and-error pro-
classical design
position. This a distinct disadvantage of the method, since it does not indicate
is

whether a solution even actually exists for the design problem at the outset. It
506 / Introduction to Control Systems Design Chap. 10

r(f)

-6/\ e(0
Controller u(t)
Controlled
process G
c(r)

(a)

r(0 e(f)
/\ u(f) Controlled c(0
process Gp

Controller
Gc

(b)

Fig. 10-3. Block diagrams of control systems with two different schemes
of compensation, (a) Series compensation, (b) Feedback compensation.

is entirely possible that the design requirements are so stringent or may even
be
contradictory so that they cannot be satisfied by any system configuration or
controllers that are physically realizable. Even when a solution does exist, the
classical design yields a
system that is very seldom the best by any standards.
For margin and phase margin are measures of the relative stability
instance, gain
of a control system. A system having a gain margin of, say, 20 dB or a phase
margin of 45° does not imply that it is optimal in any sense.
The introduction of the root locus technique by Evans 4
in 1950 made pos-
sible the designof control systems to be carried out in the ,?-plane. The main
advantage of the root locus method is that information on frequency-domain as
well as time-domain characteristics can be derived directly from the pole-zero
With the knowledge of the closed-loop transfer
configuration in the j-plane.
function poles and zeros, the time-domain response is determined readily by
means of inverse Laplace transform, and the frequency response is obtained
from the Bode plot. However, the root locus design is still basically a trial-and-
error procedure, and it relies on the reshaping of the root loci to obtain a satis-
factory pole-zero configuration for the closed-loop transfer function.
The work by Norbert Wiener 3 in the late 1940s opened a new horizon to
the design of control systems. Wiener introduced not only the statistical con-
siderations of control systems but also the idea of the performance index. For
the first time, the design engineer was able to start from a set of design criteria
 Sec. 10.1 Introduction / 507

and carry out the design by means of a completely analytical procedure. He is

able to design a control system that is optimum or the best possible with respect
to a given performance criterion.
In many practical applications of control systems, the actual signals and
disturbances subjected by a control system may be random in nature. Unlike
the deterministic signals, such as the step function and the sinusoidal function
considered in the preceding chapters, random signals can be adequately described
only by their statistical properties. For instance, in the problem of controlling
the antenna of a radar system, the wind force acting on the antenna is best
described by some probabilistic function rather than by a sine wave or any other
deterministic signals. The main difference between a deterministic signal and
a random signal is that the magnitude of the latter can only be described as what
is the probability that it a given time.
will lie in a certain range at
The principle of Wiener's optimization technique is demonstrated by the
block diagram shown in Fig. 10-4. The design objective is to determine the
closed-loop transfer function C(s)/R(s) of the system such that the error between
the desired output and the actual output is minimized. In Wiener's statistical
design technique, the mean-square value of the error e(t) is used as the perfor-
mance index, /; that is,

J = lim
M \t) dt (10-1)

r„(0

Noise
7\ Controller Process
Gp c(0

r '

/x +

rs (0
Signal

Desired system
(model)

Fig. 10-4. Block diagram of control systems designed by Wiener's optimi-

zation technique.

The reason for using the mean-square error as the performance index is that the
minimization of this particular performance index induces an analytical design
procedure which makes use of the mathematical functions already defined in the
theory of probability and statistics. However, in practice, a system that is
optimum in the sense of minimum mean-square error may be ideal only for

certain situations but not for others. In fact, it is not difficult to see from Eq.
(10-1) that the mean-square-error criterion places heavier emphasis on large
errors than on smaller ones.
508/ Introduction to Control Systems Design Chap. 10

In reality, the configuration shown in Fig. 10-4 can be used for the analyti-
cal design of systems with random inputs as well as systems with deterministic
inputs. When the input signal is considered to be deterministic, other perfor-
mance indices such as those listed below can be treated mathematically:

2
f"«
(0<ft f\<t)\dt [°t\e{t)\dt
J Jo Jo
first of these criteria is known as the integral-square error (ISE) and
In fact, the
is most popular one used for the analytical design of control systems. The
the
reason for this popularity is due to the fact that the integral is directly related
to the Laplace transform domain through Parseval's theorem.
The importance of Wiener's work and the analytical design is not so much
because the techniques have found significant applications in control systems
practice, but because these represent a revolution in design principle from the
conventional trial-and-error methods.
At approximately the same time the analytical design principle and tech-
niques were being developed, Truxal 5 proposed a synthesis procedure through
pole-zero configuration in the j-plane. The synthesis still makes use of the con-
ventional design specifications, such as the relative damping ratio , error constants,
bandwidth, and rise time, and the input Based on the
signals are deterministic.
design specifications, the closed-loop transfer function of the control system is
first determined, and then the corresponding open-loop transfer function is

found. The advantage of this synthesis method over the classical frequency-
domain design is that the designer is able to determine if the given set of specifi-
cations are consistent at the beginning of the design, so the amount of guesswork
and trial and error are cut to a minimum. Furthermore, Truxal' s synthesis starts

with the closed-loop transfer function and then works toward the transfer func-
tion of the controller, whereas the frequency-domain design starts out with the
controller and then works toward the closed-loop transfer function to see if the
design specifications are satisfied.
It is very difficult to pinpoint exactly when the modern control systems
theory was inaugurated. In fact, the mathematical foundation of certain aspects
of modern control theory can be traced far back to works that were completed
some seventy years ago. For instance, the state-variable approach to linear sys-
tems is well known to mathematicians as the theory of solutions of first-order
differential equations.Liapunov's method on stability was based on his Ph.D
thesis,which was completed in 1892. The linear programming technique, which
has significant impact on modern control theory and practice, was developed
about 1939. These significant contributions, among many others, did not become
widely known until recent years because they were much too far ahead of their
time.
The launch of the space age has placed a challenge to the control engineer
to find new methods of design of more complex control systems and to meet
more rigid requirements. The control engineer soon discovered that the conven-
tional design was no longer adequate and rigorous enough to handle the com-
plicated problems encountered in modern fire control systems, autopilot sys-
 :

Sec. 10.1 Introduction / 509

terns, missile guidance systems, spacecraft rendezvous control systems, and

many others. Modern societal systems are modeled by complex models with
multiple inputs and outputs which are difficult to handle even with optimal con-
trol theory.Consequently, not only new design principles have been developed,
but many of the mathematical contributions that have long been neglected were
rediscovered and made applicable to the practical control problems. In a sense,
the classical control system design is truly an engineering endeavor, and the
modern control design contains, first, the development and formulation of the
mathematical theory, and second, the application of the mathematical princi-
ples to practical design problems. Indeed, at present, many areas of the advanced
control theory are still at the theoretical stage, and it is generally recognized that
a gap still between theory and practice in the design of control systems.
exists
The objective of modern control design can be described by two words :

optimal control. In other words, a system is to be designed so that it is optimum

in a prescribed sense. For instance, with reference to the block diagram of the
multivariable controlled process shown in Fig. 10-2, one of the common prob-
lems in optimal control is to determine the control vector u(r) over the time
interval t < <t
f so that the state vector x(t) is brought from the initial state
t

x(/ ) to the desired final state x(^) in the shortest possible time, subject to the
given controlled process and possibly other constraints. The problem is usually
referred to as the minimal-time problem or time-optimal problem.
In general, the performance index / can be represented by the following
integral

/= CF[x(t),u(t),t]dt (10-2)

where Fis a scalar function. For the minimal-time problem, Fis set at unity, so

r dt = t
f
— ta (10-3)

Minimizing / is therefore equivalent to minimizing the time interval between the

initial and the final times.
As another example, if it is desired to drive the state vector x(t) as close as
possible to the desired state x,, over the time interval
(t t ), while keeping the
f ,

magnitudes of the controls within reason, the following performance index may
be used
C

f
" [x(0
f - xJ'Q[x(/) - xj + u'(0Ru(0} dt (10-4)

where Q and R are symmetric matrices.

The extensive use of applied mathematics in modern control theory has
made it difficult for one to make a quick transition from the classical design to
the modern. The classical design is characterized by such terminology and tools
as transfer function, poles and zeros, frequency response, root loci, Bode plot,
and Nyquist plot. In optimal control studies we shall find a set of new terms such
as state variables, state equations, state transition matrix, maximum or minimum
principle, Liapunov's methods, gradient technique, linear programming, dyna-
mic programming, controllability, and observability.
 : .

510/ Introduction to Control Systems Design Chap. 10

In this section we have given a brief discussion of the historical development

of control systems theory. The discussions and reference made are by no means
exhaustive. It is hoped that these introductory remarks will give the reader a
general idea of the basic problems involved in the design of control systems be-
fore entering the details on the subject of design. The remaining part of this
chapter will contain subjects on the classical design methods. The analytical
design method and certain aspects of optimal control are covered in Chapter 1 1

10.2 Classical Design of Control Systems

In this section the classical design of control systems will be carried out in the
frequency domain and the .-domain. The designs are mainly affected by means
of Bode plot, Nichols chart, and the root locus.
To illustrate the basic principle of the classical design, let us consider the
following example. Let us begin by considering the transfer function of a con-
trolled process

G »= ,(1 + ,X *+ 0-0125.)
(
10 " 5 >

The closed-loop system is considered to have a unity feedback. It is required

that when a unit ramp input is applied to the closed-loop system, the steady-state
error of the system does not exceed 1 per cent of the amplitude of the input ramp,
which is unity. Thus when we use the approach discussed in Chapter 6 on steady-
state error, we can find the minimum value of K in order to fulfill this error
requirement

steady-state error = e ss — lim = _ < 0.01 (10-6)

Therefore, K must be greater than 100. However, applying the Routh-Hurwitz

criterion to the characteristic equation of the closed-loop system it is easy to
show that the system is unstable for all values of A" greater than 81. This means
that some kind of compensation scheme or controller should be applied to the
system so that the steady-state error and the relative stability requirements can
be satisfied simultaneously. It is apparent that this controller must be able to
keep the zero-frequency gain of s times the open-loop transfer function of the
compensated system effectively at 100 while maintaining a prescribed degree of
relative stability. The principle of the design in the frequency domain is best
illustrated by the Nyquist plot of Gp (s) shown in Fig. 10-5. In practice, we sel-
dom use the Nyquist plot but rather the Bode plot for design, because the latter
is easier to construct. When K = 100, the system is unstable, and the Nyquist

plot of G p (s) is shown to enclose the ( — 1,7*0) point. Let us assume that we desire
to realize a resonance peak of M
p = 1.25. This means that the Nyquist plot of
Gp(s) must be tangent to the constant-M circle for M
= 1 .25 from below. If K
is the only parameter that we can adjust to achieve the objective of p = 1.25, M
Fig. 10-5 shows that the desired value of K is 1. However, with this value of K,
 :

Sec. 10.2 Classical Design of Control Systems / 511

G-plane i/ImC

Constant M locus, M = 1 .25

ReG

Fig. 10-5. Nyquist plot for Gp{s) = K/[s(l + s)(l + 0.01255)].

the velocity error constant is and the steady-state error requirement

only 1 sec" 1 ,
is not Since the steady-state performance of the system is governed
satisfied.

by the characteristics of the transfer function at the low frequency, Fig. 10-5
shows that in order to simultaneously satisfy the transient and the steady-state
requirements, the Nyquist plot of Gp (s) has to be reshaped so that the high-
frequency portion of the locus follows the K=
1 plot and the low-frequency

portion follows the K=

100 locus. The significance of this locus reshaping is
that the compensated locus shown in Fig. 10-5 will be tangent to the 1.25 M=
circle ata relatively high frequency, while the zero-frequency gain is maintained
at 100 to satisfy the steady-state requirement. When we inspect the loci of Fig.
10-5, it is clear that there are two alternative approaches in arriving at the com-
pensated locus

1. Starting from the K=

100 locus and reshaping the locus in the
region near the resonant frequency ca while keeping the low-fre-
p ,

quency region of Gp(s) relatively unaltered.

2. Starting from the K= 1 locus and reshaping the low-frequency
portion of Gp (s) to obtain a velocity error constant of K =
v 100
while keeping the locus near co = co p relatively unchanged.
512 / Introduction to Control Systems Design Chap. 10

In the first approach, the high-frequency portion ofis pushed in the G p (s)
counterclockwise direction, which means that more phase added to the sys- is

tem in the positive direction in the proper frequency range. This scheme is
basically referred to as phase-lead compensation, and controllers used for this
purpose are often of the high-pass-filter type. The second approach apparently
involves the shifting of the low-frequency part of the K= 1 trajectory in the
clockwise direction, or alternatively, reducing the magnitude of G p (s) with K=
100 at the high-frequency range. This scheme is often referred to as phase-lag
compensation, since more phase lag is introduced to the system in the low-
frequency range. The type of network that is used for phase-lag compensation
is often referred to as low-pass filters.

Figures 10-6 and 10-7 further illustrate the philosophy of dssign in the fre-
quency domain using the Bode diagram. In this case the relative stability of the
system is more conveniently represented by the gain margin and the phase
margin. In Fig. 10-6 the Bode plots ofG p (jco) show that when K = 100, the gain
and phase margins are both negative, and the system is unstable. When K = 1,
the gain and phase margins are both positive, and the system has quite a com-
fortable safety margin. Using the first approach, the phase-lead compensation,
as described earlier, we add more phase lead to G p {jcS) so as to improve the phase
margin. However, in attempting to reshape the phase curve by use of a high-
pass filter, the magnitude curve of Gp {jco) is unavoidably altered as shown in
Fig. 10-6. If the design is carried out properly, it is possible to obtain a net gain
in relative stability using this approach. The Bode diagram of Fig. 10-7 serves

aa
a 80
c
^ 60
'A

40
^>

o 20
au3
a.
-20
<
-40
-90°

^
- 180°

ft.
-270°
<*—
O
<o
rt
J=
.0*
0.01 10 100 1000

co rad/sec

Fig. 10-6. Bode plot of G p (s) = K/[s(l + ?)(! + 0.0125.$)] with phase-lead
compensation.
 Sec. 10.2 Classical Design of Control Systems / 513

CQ 80
c
^ 60
3
40
<1

o 20
<D
T)
3
a
H - 20
<
-40
-90°
^-v
J - 180°
d.
IS
-270°
o
«5 -360°
43
a.

0.01 1000

cj rad/sec

Fig. 10-7. Bode plot of Gp (s) = #/[,$( I + s)(\ + 0.0125.*)] with phase-lag
compensation.

to illustrate the principle of phase-lag compensation. If, instead of adding more

positive phase to G p (jco) at the high-frequency range as in Fig. 10-6, we attenuate
the amplitude of Gp {jco) at the low frequency by means of a low-pass filter, a
similar stabilization effect can be achieved. The Bode diagram of Fig. 10-7 shows
that if the attenuation is affected at a sufficiently low frequency range, the effect
on the phase shift due to the phase-lag compensation is negligible at the phase-
crossover frequency. Thus the net effect of the compensating scheme is the
improvement on the relative stability of the system.
It will be shown later that the transfer function of a simple passive network
controller can be represented in the form of

G {s)
s + z, (10-7)
c
s + Pi
where for a high-pass filter, p^ > z,, and for a low-pass filter, p t
< z,. Using
this transfer function, it is now possible to explain the two basic approaches of
compensation using the root locus diagram.
The root locus diagram of the closed-loop control system which has the
Gp (s) of Eq. (10-5) as its open-loop transfer function is shown in Fig. 10-8.
The root loci clearly indicate the instability condition when K = 100. Using the
phase-lead compensation, with the transfer function of Eq. (10-7), p > z,, x

the resulting root loci are shown in Fig. 10-8. If the values of z and;?, are chosen x

appropriately, the complex roots of the closed-loop system with K= 100 may
be moved into the left half of the j-plane.
514 / Introduction to Control Systems Design Chap. 10

i-plane

No compensation

K K=0
——* K —
-80

Fig. 10-8. Root locus diagrams of Gp (s) = K/[s(l + s)(l + 0.0125,?)] and
GMGp = [K(s + Zi)V[s(s +/»iXl
(s) + s)0 + 0.0125.*)] (/>, > z,).

The root locus diagram of Fig. 10-9 illustrates the principle of phase-lag
compensation in the s-domain. In this case /», is chosen to be less than z u but
for effective control, the two values are chosen to be very close to each other and
are relatively small. Because of the closeness of z, and p u the portion of the root
loci that represents thedominant eigenvalues of the system are not much affected
by the compensation. However, the stability of the compensated system is im-
proved since the points at which = 100 are shifted into the left half of the
K
j-plane. Details of the design using the above mentioned techniques are given in
the following sections. One should not be misled that any given control system
can always be compensated satisfactorily by either of the two schemes discussed
here. It will be shown that, for systems with certain characteristics, satisfactory
 —

Sec. 10.3 Phase-Lead Compensation / 515

, . 700

s-plane

No With
compensation // 'compensation
/(

K= 100

K K= K= K=
-« X * *
-80

Fig. 10-9. Root locus diagrams of G p (s) = Kl[s(\ + s)(l + 0.0125.$)] and
Gc(s)G„(s) = [K(s + ziflMs +pi)(l + 5)(1 + 0.0125*)] (Pi < r t ).

compensation cannot be accomplished by phase-lead networks alone. However,

this does not mean that proper compensation may then be achieved by using
phase-lag networks, for it is quite common that neither scheme is feasible, and
some combination of lead and lag characteristics is needed.

10.3 Phase-Lead Compensation

In Chapter 6 a simple phase-lead compensation scheme using the transfer func-

tion (1 +
Ts) was described under the name "derivative control." Although this
transfer function represents the simplest configuration for a phase-lead scheme,
in practice, (1 + Ts) cannot be realized as a transfer function by any passive
network. A and simple phase-lead network is shown in Fig. 10-10.
practical
Although the network may be simplified still further by eliminating R such a l ,
516 / Introduction to Control Systems Design Chap. 10

Sees
-WW Sccs
zero infinite
impedance impeda ice

Fig. 10-10. Passive phase-lead network.

network would block dc signals completely and cannot be used as a series com-
pensator for a control system.
The transfer function of the network is derived as follows by assuming that
the source impedance which the lead network sees is zero, and the output load
impedance is infinite. This assumption is necessary in the derivation of the
transfer function of any four-terminal network.

E 2 {s) R 2 + R,R 2 Cs (10-8)

E,{s) Ri +R R R 2 Cs
2 t

or
E2 (s) 1 + R,Cs
(10-9)
E^s) Ri +R 2 ]
+ r7Tr~
_i_
RjR 2 r„
Cs
2

Let
/?, +R 2
a> 1 (10-10)
R2
and
R^R 2 c
(10-11)
R, +R 2

then Eq. (10-9) becomes

1 1 + aTs a > 1 (10-12)

E 1
(s) a 1 + Ts
In the following sections, the characteristics of the RC phase-lead network
and various design considerations are considered.

Characteristics of the RC Phase-Lead Network

Pole-zero configuration of the RC
phase-lead network. As seen from Eq.
(10-12), the transfer function of the phase-lead network has a real zero at
s =— \/aT and a real pole at s =
—l/T. These are represented in the s-plane as
shown in Fig. 10-1 1. By varying the values of a and T, the pole and zero may be
located at any point on the negative real axis in the s-plane. Since a > 1, the
zero always located to the right of the pole, and the distance between them
is is

determined by the constant a.

Polar plot of the RC phase-lead network. When using the RC phase-lead

network of Fig. 10-10 as a compensator for control systems, the attenuation,
l/a, of Eq. (10-12) is overcome by the amplifier gain of the system, so it is neces-
Sec. 10.3 Phase-Lead Compensation / 617

/"

s-plane

Pole Zero

- i/r - \/aT

Fig. 10-11. Pole-zero configuration of a passive phase-lead network.

E2 (j)/£,(s) =(l/a)(l + a7S)/(l + Ts).

sary only to investigate the transfer function

a
E 2 (s) _ 1 + aTs (10-13)
E x
{s) 1 + T5
The polar plot of Eq. (10-13) is shown in Fig. 10-12 for several different
values of a. For any particular value of a, the angle between the tangent line
drawn from the origin to the semicircle and the real axis gives the maximum
phase lead m which the network can provide. The frequency at the tangent
<f>

point, co m represents the frequency at which <f> m occurs. It is seen that, as a in-
,

creases, the maximum phase lead, <j> m> also increases, approaching a limit of 90°
as a approaches infinity. The frequency co m decreases with the increase in a.

0\ <aj< "3
J Im
E 2 (fu) ,
a plane
E (/co)
x

03
\

\
CO

s\
\ \
\

/^^^l/^
| co =
mX

co -» °°\
CO \
<t>m\

\
CO -V ool
\
CO -*oo| \
\
Re
1 a, a2 a3

Fig. 10-12. Polar plot of a[E 2 (.s)IE i (s)\ = (1 + aTs)l(l + Ts).

518 / Introduction to Control Systems Design Chap. 10

Bode plot of the RC phase-lead network. In terms of the Bode plot, the
RC network of Fig. 10-10 has two corner frequencies: a positive corner fre-
quency at co = 1/aT and a negative corner frequency at at == \\T. The Bode
diagram of aE2 (jco)/E (jco) is shown in Fig. 10-13.
1

aE-,
20 dB/decade
20 log 10 a
slope

dB

J_
aT

aE-,

Fig. 10-13. Bode plot of the phase-lead network aE2 (s)/Ei(s) =- (I + aTs)/
(1 + Ts) (a > 1).

Analytically, 0„ and co m may be related to the parameters a and T. Since co m

is the geometric mean of the two corner frequencies, we can write

log io co m = yflogio ^j, + log 10

y) (10-14)

Thus

(10-15)

To determine the maximum phase lead, <f> m , we write the phase of aE 2 (jco)/
E^jco) as

= E 2 (jcoT
-
^ Arg tan' 1
aTco tan" 1
To (10-16)
- EiUco)-
from which we have

tan
aTco — Tco
(10-17)
1 + (aTco)(Tco)
Sec. 10.3 Phase-Lead Compensation / 519

When = m ,

q> = a),, = -JL- (10-18)

VaT
Therefore, Eq. (10-17) gives

tan m
= ia-mUa) = a_^
1 + 1 2v a
or

sin
*» = JTI (10 " 20)

This last expression is a very useful relationship in the proper selection of the
value of a for compensation design.

Design of Phase-Lead Compensation by the Bode Plot Method

Design of linear control systems in the frequency domain is more pref-

erably carried out with the aid of theBode plot. The reason is simply because
the effect of the compensation network is easily obtained by adding its magnitude
and phase curves, respectively, to that of the original process. For the phase-
lead compensation employing the RC network of Fig. 10-10, the general outline
of the design procedure is as follows. It is assumed that the design specifications
simply include a steady-state error and phase margin-versus-gain margin re-
quirements.

1. The magnitude and phase-versus-frequency curves for the uncom-

pensated process Gp (s) are plotted with the gain constant K set
according to the steady-state error requirement.
2. The phase margin and the gain margin of the original system are
read from the Bode plot, and the additional amount of phase lead
needed to provide the required degree of relative stability is deter-
mined. From the additional phase lead required, the desired
value of m is estimated accordingly, and a is calculated from Eq.
<j)

(10-20).
3. Once a is determined, it is necessary only to obtain the proper value
of T, and the design is in principle completed. The important step is
to place the corner frequencies of the phase-lead network, 1/aT,
and \/T such that m is located at the new gain-crossover frequency.
(/>

4. The Bode plot of the compensated system is investigated to check

that all performance specifications are met; if not, a new value of <f> m
must be chosen and the steps repeated.
5. If the specifications are all satisfied, the transfer function of the
phase-lead network is established from the values of a and T.

The following numerical example will illustrate the steps involved in the
phase-lead design.
 , 1

«.

a .5
— Ic
a s
, 1

OS

— ;
^i

'

5" ^

+ E
xf * u
CQ

e
o
o
«
0/ u
<u

o
5
O o
o
M

520
 . :

Sec. 10.3 Phase- Lead Compensation / 521

Example 10-1 Consider the sun-seeker control system described in Section 5.13.
The block diagram of the system is shown in Fig. 10-14. It is assumed
that for small signal, the error discriminator can be approximated
by a linear gain ; that is,

7° ~7 '
=K = s constant (10-21)
(X

The tachometer in the original system has also been eliminated for the present design
problem. Ablock representing the controller, in case it is needed, is inserted
between the two amplifiers. The parameters of the system are given as
RF = 10,000 Q.

K = 0.0125 volt/rad/sec
b

Km = 0.0125 newton-m/amp
Ra = 6.25 Q
/ = 10" 6 kg-m 2
K =0.1 amp/rad
s

K = variable
B=
n = 800

The open-loop transfer function of the uncompensated system is written

Oq(s) _ K RF KKm jn
s ,.„ -.,
Kl "~^>
~~
a(s) RaJs 2 + Km K s b

Substituting the numerical values of the system parameters, Eq. (10-22) gives

6 (s) _ 2500* no ,„
KW' Li)
Ws) ~ s(s + 25)

The specifications of the system are given as

1 The phase margin of the system should be greater than 45°.

2. The steady-state error of a(/) due to a unit ramp function input should be
less than or equal to 0.01 rad per rad/sec of the final steady-state output
velocity. In other words, the steady-state error due to a ramp input should
be less than or equal to 1 per cent.

The following steps are carried out in the design of the phase-lead compensation

1 . Applying the final-value theorem, we have

d
lim ait) = lim s*(s) = lim s /*\ (10-24)
l^~ j-,0 s-0 1 + [Po(s)/«(j)]
Since 9 r (s) = 1/s 1 , using Eq. (10-23), Eq. (10-24) becomes

lim a(0 = 0.01/* (10-25)

Thus we see that if K = 1, we have the steady-state error equal to 0.01.

However, for this amplifier gain, the damping ratio of the closed-loop
system is merely 25 per cent, which corresponds to an overshoot of over
522 / Introduction to Control Systems Design Chap. 10

1.5

Without compensation

+
With phase-lead compensation, GJs) = ', "'2?^ 2s
1 + 0.0106s

MO
With phase-lead compensation, Gc (s) = ! tn^tlir

0.2 0.3 0.4

Time (seconds)

Fig. 10-15. Step responses of the sun-seeker system in Example 10-1.

44.4 per cent. Figure 10-15 shows the unit step response of the closed-loop
system with K= 1. It is seen that the step response is quite oscillatory.
2. The Bode 6 (s)loi,(s) of the uncompensated system, with K = 1, is
plot of
sketched as shown in Fig. 10-16.
3. The phase margin of the uncompensated system, read at the gain-crossover
frequency, <a c = 50 rad/sec, is 25°. Since the phase margin is less than the
desired value of 45°, more phase lead should be added to the open-loop
system.
4. Let us choose to use the phase-lead network of Fig. 10-10 as the controller.
Then, the transfer function for the controller of Fig. 10-14 is

= G (s) 1 +aTs
(10-26)
E (.s)
c
1 + Ts

As mentioned accompanied by the phase-

earlier, the attenuation, 1/a,
lead network is assumed to be absorbed
by the other amplifier gains.
Since the desired phase margin is 45° and the uncompensated system has
a phase margin of 25°, the phase-lead network must provide the additional
20° in the vicinity of the gain-crossover frequency. However, by inserting the
phase-lead network, the magnitude curve of the Bode plot is also affected in
 Sec. 10.3 Phase-Lead Compensation / S23

40
I I I I I I I I I I I 1 1 I I II I I I I I I

0) _ 100
a(s) s(l+ 0.04s)
20

dB
„.,
G
(1+ 0.0262s)
cW-(i+ 0.0106s)

100(1 + 0.0262s)
20
s(l + 0.04s)(l + 0.0106s)
I

Phase of lead network ^^

-90°
Phase of compensated system
Phase of original system
Phase margin of compensated system
Phase margin of original system
180° _l I I I I I I I I M I I
J t I Itl I -' i '''I L_
10 50 60 100 1000

co rad/sec

Fig. 10-16. Bode plots of compensated and uncompensated systems in

Example 10-1.

such a way that the gain-crossover frequency is shifted to a higher frequency.

Although it is a simple matter to adjust the corner frequencies, IjaT&nd \jT,
so that the maximum phase of the network, m falls exactly at the new gain- ,

crossover frequency, the original phase curve at this point is no longer 25°,
and could be considerably less. This represents one of the main difficulties in

the phase-lead design. In fact, if the phase of the uncompensated system

decreases rapidly with increasing frequency near the gain-crossover frequency,
phase-lead compensation may become ineffective.
In view of the above mentioned difficulty, when estimating the necessary
amount of phase lead, it is essential to include some safety margin to account
for the inevitable phase dropoff. Therefore, in the present design, instead of
selecting a m of 20°, we let m be 25°. Using Eq. (10-20), we have

sin0„ sin 25° = 0.422 = a+ 1

(10-27)

from which we get

2.46 (10-28)

To determine the proper location of the two corner frequencies, 1/aTa.nd

ljT, it is known from Eq. (10-18) that the maximum phase lead m occurs <j>

at the geometrical mean of the corners. To achieve the maximum phase

margin with the value of a already determined, m should occur at the new <f>

gain-crossover frequency a>'c which is not known. Thus the problem now
,
524 / Introduction to Control Systems Design Chap. 10

is to locate the two corner frequencies so that may be

(f> m occurs at co'c . This
accomplished graphically as follows:
(a) The zero-frequency attenuation of the phase-lead network is calcu-
lated:
20 log 10 a = 20 logi 2.46 = 7.82 dB (10-29)

(b) The geometric mean co m of the two corner frequencies \/aT and \\T
should be located at the frequency at which the magnitude of the
uncompensated transfer function 9 (jco)la(jco) in decibels is equal to
the negative value in decibels of one half of this attenuation. This way,
the magnitude plot of the compensated transfer function will pass
through the 0-dB axis at co = co m Thus CO m should be located at the
.

frequency where

floC/Q>)
~± = 7 oo
-3.91 dB (10-30)

From Fig. 10-16, this frequency is found to be co m --= 60 rad/sec. Now

using Eq. (10-15), we have

-j = VaOJm = \/2A6 x 60 = 94 rad/sec (10-31)

Then

^j, = 38.2 rad/sec (10-32)

The parameters of the phase-lead controller are now determined. Figure 10-16
shows that the phase margin of the compensated system is approximately 48°. The
transfer function of the phase-lead network is

17(f) _ I 1 +aT,9 11+ 0.02625 (10-33)

E (s) a 1 + Ts 2.46 1 + 0.0106*
Since it is assumed that the amplifier gains are increased by a factor of 2.46, the
open-loop transfer function of the compensated sun-seeker system becomes

flofr) 6150(5 + 38.2)

*{s) s(s + 25){s + 94) u "-34)
In Fig. 10-17 the magnitude and phase of the original and the compensated sys-
tems are plotted on the Nichols chart. These plots are obtained by taking the data
directly from the Bode plot of Fig. 10-16. From the Nichols chart, the resonance peak,
M p , of the uncompensated system is found to be 1 .88, or 5.5 dB. The value of
p with M
compensation is 1.175, or 1.4 dB. One more important point is that the resonant
frequency of the system is decreased from 50 rad/sec to approximately 37 rad/sec, but
the bandwidth is increased from 70 rad/sec to 110 rad/sec.
The unit step response of the compensated system is shown in Fig. 10-15. Note that
the response of the system with the phase-lead controller is far less oscillatory than
that of the original system. The overshoot is reduced from 44.4 per cent to 24.5 per
cent, and the rise time is also reduced.The reduction of the rise time is due to the
increase of the bandwidth by the phase-lead controller. On the other hand, excessive
bandwidth may be objectionable in certain systems where noise and disturbance
signals may be critical.

In the present design problem, we notice that a specification of 45° for the phase
margin yields an overshoot of 24.5 per cent in the step response. To demonstrate the
capability of the phase-lead controller, we select a to be 5.828. The resulting controller
 -160 -140 -120
Phase (degrees)

Fig. 10-17. Plots of GO) in Nichols chart for the system in Example 10-1.

52S
 M 1.0

300

Fig. 10-18. Closed-loop frequency responses of the sun-seeker system in

Example 10-1.

oo
'/CO
t
K
5-plane , 1

/sT=2500- - /48.3

K= a: = o

-25

K =2500- — /48.3

1 '

K
oo

(a) G(s) =

s(s + 25)

Fig. 10-19. Root locus diagrams of the sun-seeker system in Example 10-1.
(a) Uncompensated system.

526
Sec. 10.3 Phase-Lead Compensation / 527

.
.
/to

s-plane
A: = 250oY (-36.7+/61.7)
I

K= AT =2500 |
K=
I

94 14.7

AT =2500 / (-36.7-/61.7)
I

K{s + 38.2)
(b) G(s)
s(s + 25)(s + 94)

Fig. 10-19 (Cont.). (b) Compensated system with phase-lead controller.

has the transfer function

1 1 + 0.03427*
Gc ^ " 5.828 1 + 0.00588J (10-35)

The unit step response of the compensated system is plotted as shown in Fig. 10-15.
In this case the rise time is shorter still, and the peak overshoot is reduced to 7.7 per

cent.
Using the magnitude-versus-phase plots and the Nichols chart of Fig. 10-17, the
closed-loop frequency responses for the sun-seeker system, before and after compen-
sation, are plotted as shown in Fig. 10-18.
To show the effects of the phase-lead compensation, the root locus diagrams of
the system, before and after compensation, are shown in Fig. 10-19. It is clear from
these diagrams that the phase-lead compensation has the effect of bending the complex
root loci toward the left, thus improving the stability of the system. The eigenvalues of
the compensated closed-loop system are at s —45.5, —36.7 =
j61.7, and —36.7 +
— y"61.7. Therefore, the damping ratio of the complex eigenvalues of the compensated
system is approximately 50 per cent.
it is difficult to visualize how the phase-lead design can be carried out
In general,
with the root locus method independently. The root loci of Fig. 10-19 are drawn based
on the results of the Bode plot design. In the following section, we shall illustrate a
procedure of designing phase-lead compensation using the root contour method.
528 / Introduction to Control Systems Design Chap. 10

Design of Phase- Lead Compensation by the Root Locus Method

The root contour method can be used for the design of control systems.
Let us use the sun-seeker system studied in Example 10-1 to illustrate the design
procedure.
The open-loop transfer function of the uncompensated system is

0M = G(s)
{} =
250 °
HO-36^
(1UJ&)
a(j) s(s + 25)

The characteristic equation of the closed-loop system is

s
2
+ 25s + 2500 = (10-37)

and the eigenvalues are Si = —12.5 + j'48.4 and s 2 = —12.5 — y'48.4.

The open-loop transfer function of the system with phase-lead compensa-
tion is written

^ --
G(s)
2500(1
s(s +
+
25)(1
aTs)
+ Ts)
.
Q 38)
(1 °"
.

The problem is to determine the values of a and Tso that the system will perform
as desired. In this case, rather than using the frequency-domain specifications,
such as the phase margin and gain margin, it is more convenient to specify the
relative positions of the complex eigenvalues.
To begin with the root contour design, we first set a to zero in Eq. (10-38).
Then, the characteristic equation of the compensated system becomes

s(s + 25)(1 + Ts) + 2500 = (10-39)

Since Tis the variable parameter, we divide both sides of Eq. (10-39) by the terms
that do not contain T. We have

+ 25
1+ ^+lL + 2 500 -° (
10 " 4 °)

This equation is of the form of 1 + G^s) = 0, where G^s) is an equivalent

transfer function that can be used to study the root loci of Eq. (10-39). The root
contour of Eq. (10-39) is drawn as shown in Fig. 10-20, starting with the poles

and zeros of G^s). Of significance is that the poles of G (s) are the eigenvalues t

of the system when a = and T = 0. As can be seen from the figure, the factor
1 + Ts in the denominator of Eq. (10-38) alone would not improve the system

performance at all. In fact, the eigenvalues of the system are moved toward the
right half of the s-plane, and the system becomes unstable when T is greater than
0.0133. To achieve the full effect of the phase-lead compensation, we must re-
store the value of a in Eq. (10-38). The characteristic equation of the compen-
sated system now becomes

s(s + 25)(1 + Ts) + 2500(1 + aTs) = (10-41)

Now we must consider the effect of varying a while keeping T constant.

This is accomplished by dividing both sides of Eq. (10-41) by the terms that do
not contain a. We have
2500aTs
1+
+ s(s + 25)(1 .

+ Ts) + 2500 ~
Q
U 1Q 4Z)
(iU
Sec. 10.3 Phase-Lead Compensation / 529

ico

i-plane

F=0* -/48.4
/43.3(r = 0.0133)

o<-r T=cc r=oo

7"=oo
25

-/43.3(r= 0.0133)
^=0*- /48.4

Fig. 10-20. Root contours of the sun-seeker system of Example 10-1 with
a = 0, and T varies from 0to<».

This equation is again of the form of 1 +G 2 {s)

— 0, and for a given T, the root
loci of Eq. (10-41) can be obtained based upon the poles and zeros of

2500aTs
A ' ~ s(s + 25X1 + Ts) + 2500
(10-43)

Notice also that the denominator of G 2 (s) is identical to the left side of Eq.
(10-39), which means that the poles of G 2 (s) must lie on the root contours of Fig.
10-20, for a given T. In other words, the root contours of Eq. (10-41) as a varies
must start from the points on the trajectories of Fig. 10-20. These root contours
end at ^ = 0, oo, oo, which are the zeros of G 2 (s). The complete root contours
of the system with phase-lead compensation are now sketched in Fig. 10-21.
From the root contours of Fig. 10-21, we can see that for effective phase-
lead compensation, the value of T should be small. For large values of T, the
bandwidth of the system increases very rapidly as a increases, while very little
 :

530 / Introduction to Control Systems Design Chap. 10

* oo » oo . 00 /'<0

\\ A a
\ a
\
t

s-plane

Nt 2.46\ \
a =
- 36.7 +/61.7_S*_J^»^v\

^
- 12.5+/48.4^5?
a = 1

ly\
'
,T=0.0106
/\--T = 0.0133

\
J 7"= 0.05
/^</
T=0 / =

/ a<\
(t=2.5
1 <
7 = 0=
/
Root contour on
-25 o
5
1 r=«=
\
real axis

/
'*.
— /48.4

Fig. ip-21. Root contours of the sun-seeker system with a phase-lead con-
troller, G c (s)= (1 + aTs)l(l + Ts), Example 10-1.

improvement is made on the damping of the system. We must remember that

these remarks are made with respect to the phase-lead design that corresponds
to values of a greater than unity.

Effects and Limitations of Phase-Lead Compensation

From the results of the last illustrative example, we may summarize the
general effects of phase-lead compensation on the performance of control sys-
tems as follows

1. The phase of the open-loop transfer function in the vicinity of the

gain-crossover frequency is increased. Thus the phase margin is

usually improved.
2. The slope of the magnitude curve representing the magnitude of the
open-loop transfer function is reduced at the gain-crossover fre-
 . : :

Sec. 10.3 Phase- Lead Compensation / 531

quency. This usually corresponds to an improvement in the relative

stability of the system. In other words, the phase and gain margins
are improved.
3. The bandwidths of the open-loop system and the closed-loop system
are increased.
4. The overshoot of the step response is reduced.
5. The steady-state error of the system is not affected.

It was mentioned earlier that certain types of systems cannot be effectively

compensated satisfactorily by phase-lead compensation. The sun-seeker system

studied in Example 10-1 happens to be one for which phase-lead control is
effective and practical. In general, successful application of phase-lead compen-
sation is hinged upon the following considerations

1. Bandwidth considerations: If the original system is unstable, the

additional phase lead required to obtain a certain desirable phase
margin may be excessive. This requires a relatively large value of a
in Eq. (10-12), which, as a result, will give rise to a large bandwidth
for the compensated system, and the transmission of noise may
become objectionable. Also, if the value of a becomes too large, the
values of the network elements of the phase-lead network may be-
come disproportionate, such as a very large capacitor. Therefore, in
practice, the value of ais seldom chosen greater than 15. If a larger

value of a is sometimes two or more phase-lead controllers

justified,
are connected in cascade to achieve the large phase lead.
2. If the original system is unstable or has low stability margin, the
phase plot of the open-loop transfer function has a steep negative
slope near the gain-crossover frequency. In other words, the phase
decreases rapidly near the gain crossover. Under this condition,
phase-lead compensation usually becomes ineffective because the
additional phase lead at the new gain crossover is added to a much
smaller phase angle than that at the old gain crossover. The desired
phase margin may be realized only by using a very large value of a.
However, the resulting system may still be unsatisfactory because a
portion of the phase curve may still be below the 180° axis, which
corresponds to a conditionally stable system.

In general, the following situations may also cause the phase to change
rapidly near the gain-crossover frequency

1 The open-loop transfer function has two or more poles that are close
to each other and are close to the gain-crossover frequency.
2. The open-loop transfer function has one or more pairs of complex
conjugate poles near the gain-crossover frequency.

The following example will illustrate a typical situation under which phase-
lead compensation is ineffective.
532 / Introduction to Control Systems Design Chap. 10

Example 10-2 Let the open-loop transfer function of a control system with unity
feedback be

G> (s) = (1(M4)

,(1 + .1*X1 + 0.25)
It is desired that the system satisfies the following performance specifications:

1. K„ = 100; or the steady-state error of the system due to a unit ramp

function input is 0.01 in magnitude.
2. Phase margin > 40°.

dB

1000

Phase curves of
phase-lead
compensated system

J I I I M II

000

Fig. 10-22. Bode plots of G p {s) = 100/[i(l + 0.l5)(l + 0.2j)] and the
effects of using phase-lead compensation.
Sec. 10.3 Phase- Lead Compensation / 533

From the steady-state error requirement, we set K = 100. The Bode plot of
G p (s) when K — 100 is shown in Fig. 10-22. As observed from this Bode plot, the phase
margin of the system is approximately —40°, which means that the system is unstable.
In fact, the system is unstable for all values of K greater than 15. The rapid decrease of
phase at the gain-crossover frequency, co c = 17 rad/sec, implies that the phase-lead
compensation may be ineffective for this case. To illustrate the point, the phase-lead
network of Fig. 10-10 and Eq. (10-12) with a = 4.6, 14, and 100, respectively, is used
to compensate the system. Figure 10-22 illustrates the effects of phase-lead compen-
sation when the values of Tare chosen according to the procedure described in Example
10-1.
It is clearly shown more phase lead is being added, the gain-
in Fig. 10-22 that as
crossover frequency pushed to a higher value. Therefore, for this case, in
is also being
which the uncompensated system is very unstable at the outset, it may be impossible to
realize a phase margin of 40° by the phase-lead network of Fig. 10-10.
In a similar fashion, we may use the root contour method to illustrate why the
phase-lead compensation is ineffective in this case of a highly unstable system. Figure
10-23 shows the root locus diagram of the uncompensated system with the process
described by the transfer function of Eq. (10-44). When A^ = 100, the two complex
eigenvalues of the closed-loop system are at s = 3.8 +/14.4 and 3.8 — j 14.4.

s-plane

/15
K= 100

'/7.07 (K= 15)

/5

K=100 a: = o K= K=
1 1
rf

-22.6 -20 - 10 -5

-/5

-/7.07(AT = 15)

\ K= 100
-/is \ K
\

Fig. 10-23. Root loci of s(s + 5)(s + 10) + 50AT = 0.

534 / Introduction to Control Systems Design Chap. 10

Let the controller be represented by

„ ,
%
1 + aTs a> 1 (10-45)
1 + Ts
Then, the open-loop transfer function becomes

Gc (s)Gp (s) 5000(1 + aTs) (10-46)

s(s + 5)(s + 10)(1 + Ts)

s-plane

r=o

Fig. 10-24. Root contours of s(s + 5)(s + 10)(1 + Ts) + 5000 = 0, and
of s(s + 5)(s + 10)(1 + Ts) + 5000(1 + aTs) = 0.
 Sec. 10.4 Phase-Lag Compensation / 535

First, we set a = while we vary rfrom zero to infinity. The characteristic equa-
tion becomes
s(s + 5)(s + 10)(1 + Ts) + 5000 = (10-47)

The root contours of Eq. (10-47) are constructed from the pole-zero configuration of

° l{s) ~ (1(M8)
s(s + 5)(s + 10) + 5000

Thus, in Fig. 10-24 the poles of Gt(s) are labeled as T= points, and the zeros of
Gi(s) are points at which T= oo.
Next, we restore the value of a in Eq. (10-46), and the root contours of Eq. (10-47)
become the trajectories on which a = 0. In other words, the characteristic equation of
the overall system is written

s(s + 5)(.s + 10)(1 + Ts) + 5000(1 + aTs) = (10-49)

When a is considered the variable parameter, we divide both sides of Eq. (10-49) by the
terms that do not contain a; we have
5000aTs
1 +
^ s(s + = Vo (iv-W)
(lO-SO)
5)(s + 10)(1 + Ts) + 5000
Thus, as we have stated, the root contours with a varying, start (a = 0) at the poles of
r /x_
U2W
5000ar,r
uuol)
s(s + 5)(.s + 10)(1 + Ts) + 5000

The dominant part of the rootloci of Eq. (10-49) is sketched in Fig. 10-24. Notice
that, since for phase-leadcompensation the value of a is limited to greater than 1, the
root contours that correspond to this range are mostly in the right-half plane. It is
apparent from this root contour plot that the ineffectiveness of phase-lead compensa-
tion, in this case, may be attributed to the eigenvalues of the uncompensated system
being in the right-half plane.
We shall show in the following that for the system considered in this example, it
is far more effective to compensate it by a phase-lag network, or by means of an auxil-
iary feedback loop.

10.4 Phase-Lag Compensation

In contrast to using a high-pass filter as a compensator for control systems, we

may use a low-pass filter for similar purposes. Figure 10-25 shows a simple RC
network that may be used for the low-pass or phase-lag compensation of con-
trol systems. If we assume that the input impedance of
the network is zero and that the output impedance which

—^ ^i
1
+
the network sees is infinite, the transfer function of the
network is
E^> - 1 + R t Cs
nf „,
£,(*)
- 1 + (*, + R 2 )Cs llU
.

°^
b2
Let
C^ aT =RC 2 (10-53)
and
a = R* a < 1 (10-54)
Fig. 10-25. RC phase-lag network. Rp + RD 2
t
K '
536 / Introduction to Control Systems Design Chap. 10

Equation (10-52) becomes

aTs
a < 1 (10-55)
1 + Ts

Characteristics of the RC Phase-Lag Network

Pole-zero configuration of the RC phase-lag network. The transfer function
of the phase-lag network of Fig. 10-25 has a real zero at s == — 1/aT and a real
pole at s = —l/T. As shown in Fig. 10-26, since a is less than unity, the pole is

always located to the right of the zero, and the distance between them is deter-
mined by a.

, iu>

5-plane

i 1

aT T

Fig. 10-26. Pole-zero configuration of the transfer function (1 + aTs)j(l

+ Ts), a < 1 , of a phase-lag network.

Polar plot of the RC phase-lag network. As seen from Eq. (10-52), the
transfer function of the phase-lag network does not have any attenuation at
zero frequency, contrary to the case of the phase-lead network.
When we let s = jco, Eq. (10-55) becomes

E2 (jco) = 1 + jooaT < 1 (10-56)

EiUm) l+ja>T
The polar plot of this transfer function is shown in Fig. 10-27 for three
values of a, 1 > a x
> a2 >a 3 . Just as in the case of the phase-lead network, for
any value of a (a < 1), the angle between the tangent line drawn from the origin
to the semicircle and the real axis gives the maximum phase lag m m < 0°) <f> (<f>

of the network. As the value of a decreases, the maximum phase lag m becomes <f>

more negative, approaching the limit of —90° as a approaches zero. As the

value of a decreases, the frequency at which m occurs, co m , increases; that is,

in Fig. 10-26, co m , > co ml > oo ml .

Sec. 10.4 Phase-Lag Compensation / 537

*- Re

-/ Im

Fig. 10-27. Polar plots of E2 (s)IE 1 (s) = (1 + aTs)/(l + Ts)(a < 1).

Bode plot of the RC phase-lag network. The Bode plot of the transfer
function of Eq. (10-56) is shown in Fig. 10-28. The magnitude plot has a posi-
tive corner frequency at and a negative corner frequency at co = l/T.
<a = 1/aT,
Since the transfer functions of the phase-lead and phase-lag networks are identi-
cal in form except for the zero-frequency attenuation and the value of a, it can

1 1

T aT

90° i

^ ~ "^
*• ^ 1

'i * '
u «**?
90°
"~ —— K

Fig. 10-28. Bode plot of the phase-lag network E 2 (s)/Ei(s) = (1 + aTs)l(\

+ Ts) (a < 1).
538 / Introduction to Control Systems Design Chap. 10

readily be shown that the maximum phase lag <f> m of the phase curve of Fig. 10-28
satisfies the following relation

Design of Phase-Lag Compensation by Bode Plot

Unlike the design of phase-lead compensation, which utilizes the maximum

phase lead of the network, the design of phase-lag compensation utilizes the
attenuation of the network at the high frequencies. It was pointed out earlier
that, for phase-lead compensation, the function of the network is to increase
the phase in the vicinity of the gain-crossover frequency while keeping the
magnitude curve of the Bode plot relatively unchanged near that frequency.
However, usually in phase-lead design, the gain crossover frequency is increased
because of the phase-lead network, and the design is essentially the finding of a
compromise between the increase in bandwidth and the desired amount of
relative stability (phase margin or gain margin). In phase-lag compensation,
however, the objective is to move the gain-crossover frequency to a lower fre-
quency while keeping the phase curve of the Bode plot relatively unchanged at
the gain-crossover frequency.
The design procedure for phase-lag compensation using the Bode plot is

outlined as follows:

1. The Bode plot of the open-loop transfer function of the uncom-

pensated system is made. The open-loop gain of the system is set
according to the steady-state error requirement.
2. The phase margin and the gain margin of the uncompensated system
are determined from the Bode plot. For a certain specified phase
margin, the frequency corresponding to this phase margin is found
on the Bode plot. The magnitude plot must pass through the 0-dB
axis at this frequency in order to realize the desired phase margin.
In other words, the gain-crossover frequency of the compensated
system must be located at the point where the specified phase margin
is found.
3. To bring the magnitude curve down to dB at the new prescribed
gain-crossover frequency, co'c , the phase-lag network must provide
the amount of attenuation equal to the gain of the magnitude curve
at the new gain-crossover frequency. In other words, let the open-
loop transfer function of the uncompensated system be Gp (s); then

\G P Uo>'c)\ = -20 log, o a dB a < 1 (10-58)

from which
a = io-io>.ow>i/2o a < i (10-59)

4. Once the value of a determined, it is necessary only to select the

is

proper value of T to
complete the design. Up to this point, we have
assumed that although the gain-crossover frequency is altered by
attenuating the gain at m c, the original phase curve is not affected.
 : .

Sec. 10.4 Phase-Lag Compensation / 539

This is not possible, however, since any modification of the magni-

tude curve will bring change to the phase curve, and vice versa.
With reference to the phase characteristics of the phase-lag network
shown in Fig. 10-28, it is apparent that if the positive corner fre-
quency, 1/aT, is placed far below the new gain-crossover frequency,
cd'c , the phase characteristics of the compensated system will not be
appreciably affected near co'c by phase-lag compensation. On the
other hand, the value of 1/aT should not be too much less than co'c or ,

the bandwidth of the system will be too low, causing the system to
be too sluggish. Usually, as a general guideline, it is recommended
that the corner frequency 1/aT be placed at a frequency that is
approximately 1 decade below the new gain-crossover frequency,
co'c ; that is,

rad/sec (10-60)
aT 10
Therefore,

a rad/sec (10-61)
T 10

5. The Bode plot of the phase-lag compensated system is investigated

to see if the performance specifications are met.
6. If all the design specifications are met, the values of a and T are sub-
stituted in Eq. (10-55) to give the desired transfer function of the
phase-lag network.

Example 10-3 In this example we a phase-lag controller for the sun-

shall design
10-1. The open-loop transfer
Example

S
seeker system considered in
function of the sun-seeker system is given by Eq. (10-23),

Sift
flC(s)
= s(s + 25)
dO-62)'

The specifications of the system are repeated as follows

1. The phase margin of the system should be greater than 45°.

2. The steady-state error of 0C(/) due to a unit ramp function input should be
less than or equal to 0.01 rad/rad/sec of the final steady-state output
velocity. This is translated into the requirement of K> 1

The Bode plot of d a (s)/a,(s) with K= 1 is shown in Fig. 10-29. As seen, the phase
margin is only 25°.
For a phase-lag compensation, let us choose the network of Fig. 10-25, whose
transfer function is

E^7) = G^S) = T + W a < 1 (10 " 63)

From Fig. 10-29 it is observed that the desired 45° phase margin can be obtained if the
gain-crossover frequency co'c is at 25 rad/sec. This means that the phase-lag controller

must reduce the magnitude of 6 Q {j(0)l<l{j<a) to dB at co = 25 rad/sec, while it does

not appreciably affect the phase curve in the vicinity of this frequency. Since, actually,
a small negative phase shift is still accompanied by the phase-lag network at the new
 1

540 / Introduction to Control Systems Design Chap. 10

60 i i
M i
1 1 i
i i i I i i m ii i i 1 1 —— —
I
i i

i i i 1 1
n — i
i i i i

a(s) s(l+0.Q4s)

dB

Phase margin of
compensated system

|
Phase margin of
]
uncompensated system

180
c

0.1
J 1 I

0.4
i

1.0
—
J I

2.0
I
compensated system
i i i_i i_i_ J-U
100

co rad/sec

Fig. 10-29. Bode plots of compensated and uncompensated systems in

Example 10-3.

gain-crossover frequency, it is a safe measure to choose this new gain-crossover fre-

quency somewhat less than 25 rad/sec, say, 20 rad/sec.

From the magnitude plot of {jco)l«,{jco), the value of 6 (jco)/tx,(jco) at Co'c = 20 1 |

rad/sec is 14 dB. This means that the phase-lag controller must provide an attenuation
of 14 dB at this frequency, in order to bring the magnitude curve down to dB at
CO'c = 20 rad/sec. Thus, using Eq. (10-59), we have
a = 1()-1»oOW)/«(./£i>«'>1/20
(10-64)
= lO" =0.2 - 7

The value of a indicates the required distance between the two corner frequencies of the
phase-lag controller, in order that the attenuation of 14 dB is realized.
In order that the phase lag of the controller does not appreciably affect the phase
at the new gain-crossover frequency, we choose the corner frequency 1/aT to be at 1
decade below co'c 20 rad/sec. Thus =
1 CO'c 20 = „ ,,
== 2rad/sec (10-65)
5r=To Io
which gives
-= = 0.4 rad/sec (10-66)
Sec. 10.4 Phase-Lag Compensation / 541

The transfer function of the phase-lag controller is

U(s)
= + 0.5^
1
(10-67)
E (s) + 2.5s
1

and the open-loop transfer function of the compensated system is

0q(s) _ 5(M.s + 2) -_.

nft
(iv-ob)
a(s) s(s + 0.4)(j + 25)
The Bode plot of the open-loop transfer function of the compensated system is
shown in Fig. 10-29. We see that the magnitude curve beyond CO = 0.4 rad/sec is attenu-
ated by the phase-lag controller while the low-frequency portion is not affected. In the
meantime, the phase curve is not much affected by the phase-lag characteristic near the
new gain-crossover frequency, which is at 25 rad/sec. The phase margin of the com-

pensated system is determined from Fig. 10-29 to be about 50°.

The magnitude-versus-phase curves of the uncompensated and the compensated
systems are plotted on the Nichols chart, as shown in Fig. 10-30. It is seen that the
resonant peak, M„ of the compensated system is approximately 1.4 dB. The bandwidth
of the system is reduced from 70 rad/sec to 32 rad/sec.
The unit step responses of the uncompensated and the compensated systems are
shown in Fig. 10-31. The effects of the phase-lag controller are that the overshoot is
reduced from 44.4 per cent to 22 per cent, but the rise time is increased considerably.
The latter effect is apparently due to the reduction of the bandwidth by the phase-lag
controller. Figure 10-31 also gives the step responses of the system when the value of
7* of the phase-lag controller changed to 5 and then to 10. It is seen that larger values
is

of jfgive only slight improvements on the overshoot of the step response. Earlier it was
pointed out that the value of T is not critical when T = 5, it is equivalent to setting
;

\jaT at 20 times below the gain-crossover frequency of co'c = 20 rad/sec. Similarly,

T= 10 corresponds to placing 1/aTa.t 40 times below CO'c .

Phase-Lag Compensation by the Root Locus Method

The design of phase-lag compensation is best illustrated by the root locus

diagram of the system considered in Example 10-3, which is shown in Fig. 10-32.
In this figure the root loci of the uncompensated system and the compensated
system are shown. It is important to note that since the simple pole and zero of
the phase-lag controller are placed close together and are very near the origin,
they have very little effect on the shape of the original root loci, especially near
the region where the eigenvalues should be located to achieve the desired per-
formance. However, the values of K that correspond to similar points on the
two root For example, when K = 1, which gives the desired
loci are different.
steady-state response, the eigenvalues of the uncompensated system are s =
— 12.5 +y48.4 and s = —12.5 — y'48.4, which correspond to a damping ratio
of 25 per cent. The comparable points to these eigenvalues on the compensated
root loci correspond to K = 5, which is 5 times greater. In fact, when K = 1
on the root loci of the compensated system, the eigenvalues are at s = —11.6 +
j 18 and s =
— 11.6 —j 18, which correspond to a damping ratio (for these com-
plex roots) of 56.2 per cent.
 dB 8

-160 -140 -120

Phase (degrees)

Fig. 10-30. Plots of G(s) in Nichols chart for the system in Example 10-3.

542
Sec. 10.4 Phase- Lag Compensation / 543

1.5

^ Without compensation
+ Ts
With phase-lag controller, GJs) = '
, °I,
I + Ts

0.2 0.3 0.4

Time (seconds)

Fig. 10-31. Step responses of the sun-seeker system in Example 10-3.

It is simple to show that the values of K

on the root loci of the compensated
system at points relatively far away from the origin are 5 times greater than those
values of K
at similar points on the root loci of the uncompensated system. For
instance, at the root s = —11. 6+ j 18 on the compensated root loci of Fig.
10-32, the value of K is 1 ; the comparable point on the uncompensated root loci
is s = — 12.5 + j'18.4, and the value of KaX that point is 0.2. The Bode diagram

result of Fig. 10-29 already reflects this relation on the values of K, since the
phase-lag network hardly altered the phase plot near the gain-crossover fre-
quency, but reduced the gain by 14 dB, which is a factor of 5. Another explana-
tion of the relation between the values of K of the uncompensated loci and those
of the compensated may he obtained by referring to the open-loop transfer
functions of the two systems. From Eq. (10-62), the value of at a point s on K j

the root loci for the uncompensated system is written

= I*, +25| (10-69)

1*1
2500

Assuming that a point s, is on the compensated root loci and is far from the
pole and zero of the phase-lag network at s = —0.4 and s = — 2, respectively,
 :

544 / Introduction to Control Systems Design Chap. 10

the value of K at s t
is given by

,„, ._ lg,|lJ 1 +0.4|lj +25| 1

(10-70)
1 l_ 500|j,+2|
or

1*1
|J,||J, +25|
500
since the distance from s t to —0.4 will be approximately the same as that from
Si to — 2. This argument also points to the fact that the exact location of the pole
and the zero of the phase-lag network is not significant as long as they are close
to the origin, and that the distance between the pole and the zero is a fixed
desired quantity. In the case of the last example, the ratio between 1 jaT and l/T
is 5.

Based on the discussions given above, we may outline a root locus design
procedure for the phase-lag design of control systems as follows
Since the design will be carried out in the s-plane, the specifications on the
damp-
transient response or the relative stability should be given in terms of the
and other quantities such as rise time, band-
ing ratio of the dominant roots,
width, and maximum overshoot, which can be correlated with the location of
the eigenvalues.

/CO
.s-plane

? = 0.562 \
K
t ,

f
\\
= 0.707
^
a: = 0.2 ;18
\ \
\ \
K = 0.125X \ /12.5

K = 0.0625
K = \ 4e
a: = o

25 12.5

(a)

Fig. 10-32. (a) Root loci of G(s) = 250OK/[s(s + 25)] of the sun-seeker
system.
 .

Sec. 1 0.4 Phase-Lag Compensation / 545

s-plane

r = 0.562\ ^-|

f = 0.707^

(b)

Fig. 10-32 (Cont.). (b) Root loci of G(s) = 500(5 + 2)/[s(s + 25)(s + 0.4)]
of the sun-seeker system.

1 Sketch the root loci of the characteristic equation of the uncompen-

sated system.
2. Determine on these root loci where the desired eigenvalues should be
located to achieve the desired relative stability of the system. Find
the value ofK that corresponds to these eigenvalues.
3. Compare the value of K required for steady-state performance and
the K found in the last step. The ratio of these two Ks a (a < 1), is

which is the desired ratio between the pole and the zero of the phase-
lag controller.
4. The exact value of T is not critical as long as it is relatively large.
We may choose the value of 1/aTto be many orders of magnitudes
smaller than the smallest pole of the process transfer function.

Let us repeat the design of the system in Example 10-3 by means of the
root locus method just outlined. Instead of using the phase-margin specification,
we require that the damping ratio of the dominant eigenvalues of the closed-
loop system be approximately 56.2 per cent. This value of the damping ratio is
chosen so that we can compare the result of the root locus design with that
which was obtained independently by the Bode diagram method. The root locus
diagram of the original uncompensated system is drawn as shown in Fig.
10-32(a), based on the pole-zero configuration of Eq. (10-62). The steady-state
performance specification requires that the value of K should be equal to 1 or
 :

546 / Introduction to Control Systems Design _ Chap. 10

greater. From the root loci of Fig. 10-32(a), it is found that a damping ratio of
56.2 per cent may be attained by setting K of the original system equal to 0.2.
Therefore, the ratio of the two values of K is

= 02
T=T
1
a (10 " 71)

The desired relative stability is attained by setting K to 0.2 in the open-loop

transfer function of Eq. (10-62) to give

$ (s) _ 500
{W ' U)
a(s)
~ s(s + 25)

The open-loop transfer function of the compensated system is given by

fl Q) _ 2500^(1 + aTs)
a(s) s{s + 25)(I + Ts)
(10-73)
_ 2500a^(j+ 1/aD
s(s + 25)(s + 1/T)

If the values of aT and T are chosen to be large, Eq. (10-73) is approximately

(Us)^2500aK_
= „
a(j) j(j + 25) < ;

from the transient standpoint. Since K is necessarily equal to unity, to have the
right sides of Eqs. (10-72) and (10-74) equal to each other, a = 1/5, as is already
determined in Eq. (10-71). Theoretically, the value of T can be arbitrarily large.
However, in order that the bandwidth of the closed-loop system is not too small,
we must not make Ttoo large. By setting T =
2.5, the root loci of the compen-
sated system will be identical to that of Fig. 10-32(b).
As an alternative, the root locus design can also be carried out by means
of the root contour method. The root contour method was applied to the sun-
seeker system earlier in the phase-lead design. The design carried out by Eqs.
(10-38) through (10-43) and and 10-21 is still valid for phase-lag
Figs. 10-20
control, except that in the present case a <
I. Thus in Fig. 10-21 only the por-

tions of the root contours that correspond to a < 1 are applicable for phase-lag
compensation. These root contours show that for effective phase-lag control,
the value ofT should be relatively large. In Fig. 10-33 we illustrate further that
the complex eigenvalues of the closed-loop system are rather insensitive to the
value of T when the latter is relatively large.

Example 10-4 Consider the system given in Example 10-2 for which the phase-lead
compensation is ineffective. The open-loop transfer function of the
original system and the performance specifications are repeated as
follows

The performance specifications are

1. K =
v 100 sec" 1
.

2. Phase margin > 40°.

Sec. 10.4 Phase-Lag Compensation / 547

i-plane

(- 12.3 +/1 8.4)

;Q.2 (- 12.08 +/18.3)
(- 11.6+/18) r=2.5 (0.525 +/6. 15)
T = 5.0 (0.28+/4.4)
r=10 (0.145 +/3. 135)

T= 10 (-0.51)
T=2.S T=5 T=5 (-1.04)
- 26.45 - 25.76 T= 2.5 (-2.2)

Fig. 10-33. Root contours of s(s + 25)(1 + Ts) + 2500A:(1 + aTs) = 0;

K= 1.

The phase-lag design procedure is as follows

1. The Bode plot of Gp(s) is made as shown in Fig. 10-34 for K = 100.
2. The phase margin at the gain-crossover frequency, CO c = 17 rad/sec, is
approximately —45°, and the closed-loop system is unstable.
3. The desired phase margin is 40°; and from Fig. 10-34 this can be realized
if the gain-crossover frequency is moved to approximately 4 rad/sec. This

means that the phase-lag controller must reduce the magnitude of G p (jco)
to dB, while it does not affect the phase curve at this new gain crossover
frequency, G>'c . Since actually a small negative phase is still introduced by

the phase-lag controller at co'c , it is a safe measure to choose the new gain-
crossover frequency somewhat less then 4 rad/sec, say at 3.5 rad/sec. As an
alternative,we may select a larger phase margin of 45°.
4. From the Bode plot, the magnitude of Gp (ja>) at (o'c = 3.5 rad/sec is 30 dB,
which means that the controller must introduce 30 dB of attenuation at
this frequency, in order to bring down the magnitude curve of Gp (jCo) to
dB. Thus, from Eq. (10-59),

a = lQ-\oMo>c')\no = 10 -i.5 = o.032 (10-76)

This equation implies that the two corners of the phase-lag controller
 548 / Introduction to Control Systems Design Chap. 10

80

G" (yS )' = 10Q

s{\ + 0.1 s)(l + 0.2s)

cj c = 1 7 rad/sec

-180°- Phase margin of

compensated system Phase margin of
270' J_ uncompensated system
0.01 0.0112 0.1 0.35 1.0 2.5 cjI 100

w rad/sec
Fig. 10-34. Bode plots of compensated and uncompensated systems in
Example 10-4.

must be placed 1.5 decades apart, in order to produce the required 30 dB

of attenuation.
5. Using the guideline that the upper corner frequency of the controller,
1/aT, is placed at the frequency of 1 decade below the new gain-crossover
frequency, we have

1 (o'c 3.5 „,„ ,,

^ = To=To =a35rad/sec (10-77)

which gives
T= 89.3

6. The Bode plot of the compensated system, with the phase-lag controller
transfer function given by

Gc (s) = 1 + aTs 1 + 2.86j

(10-78)
1 +Ts 1 + 89.3s

is sketched in Fig. 10-34. It is seen that the phase margin of the compen-
sated system is approximately 40°.
7. The open-loop transfer function of the compensated system is

G(s) = G (s)Gp (s) 100(1 + 2.865) (10-79)

c
s(l + 0.ls)(l + 0.2^(1 + 89.3*)

The magnitude-versus-phase curves of the compensated and the uncompensated

systems are plotted on the Nichols chart, as shown in Fig. 10-35, These curves show
260 240 220 180 - 160 140 120 100

Phase (degrees)

Fig. 10-35. Plots of G(i) of the compensated and the uncompensated sys-
tems in the Nichols chart for Example 10-4.

549
550 / Introduction to Control Systems Design
Chap. 10

that the uncompensated system is unstable, but the compensated system has the
following performance data as measured by the frequency-domain criteria:

resonant peak M p = 3 dB (1.41)

phase margin = 40 deg

gain margin = 10 dB
bandwidth = 6 rad/sec
When K=100, the open-loop transfer function of the uncompensated system and
the compensated system may be written

5000
G{s) (10-80)
s(s + 5)0 + 10)

G0) = 160Q + 0.35) (10-81)

s(s + 5)0 + 10)0 + 0.0112)

s-plane

/7.07(A = 15)
/6.73 (A = 420)

(AT = 3.2)
- 1.53+/3.335
(K= 100)- 1.44 +/2
K=0 K=0
10

G(s)
50K
s(s + 5)(s+ 10)

G(s)-
1.6A"(s + 0.35)
s(s + 5)(s+ 10)(s + 0.0112)

Fig. 10-36. Root loci of compensated and uncompensated systems in

Example 10-5.
 .

Sec. 10.4 Phase-Lag Compensation / 551

respectively. These transfer functions give indication of the improvements made on the
system's performance from the root locus point of view. The gain constant of G(s) in
Eq. (10-80) is 5000, whereas that of G(s) in Eq. (10-81) only 160; the ratio of 160 to
is

5000 is the constant, a, which is determined earlier to be 0.032. This means that since
the pole and zero of the phase-lag controller are very close to the origin, as compared
with the poles at s =
—5 and —10, the controller effectively reduced the loop gain of
the system by a factor of 0.032. Figure 10-36 shows the root loci of the uncompensated
and the compensated systems. Again, the loci of the complex roots are very close to
each other for the two cases. The critical value of K
for the uncompensated system is
15, whereas for the compensated system it is 420. When = 100, the compensated
K
system has eigenvalues at s = —11.33, s = —0.8, s = —1.436 +j2, and j = —1.436
— j2. The unit step response of the system with the phase-lag controller is shown in
Fig. 10-37. The peak overshoot of the system is approximately 35 per cent.

3 4
Time (seconds)

Fig. 10-37. Unit step response of the system with phase-lag compensation
in Example 10-4.

Effects and Limitations of Phase-Lag Compensation

From the and limi-

results of the preceding illustrative examples, the effects
tations of phase-lag compensation on the performance of a control system may
be summarized as follows:

1 For a given relative stability, the velocity error constant is increased.

2. The gain-crossover frequency is decreased; thus the bandwidth of
the closed-loop system is decreased.
3. For a given loop gain, K, the magnitude of the open-loop transfer
552/ Introduction to Control Systems Design Chap. 10

function is attenuated at the gain-crossover frequency, thus allowing

improvement in the phase margin, gain margin, and resonance peak
of the system.
4. The rise time of the system with phase-lag compensation is usually
slower, since the bandwidth is usually decreased. For additional
reduction in the overshoot, the rise time is further increased.

10.5 Lag-Lead Compensation

We have learned from the preceding sections that phase-lead compensation

usually improves the rise time and the overshoot but increases the bandwidth
of a feedback control system, and it is effective for systems without too severe

a stability problem. On the other hand, phase-lag compensation improves the

overshoot and relative stability but often results in longer rise time because of
reduced bandwidth. Therefore, we can say that each of these two types of con-
trol has its advantages and disadvantages. However, there are many systems
that cannot be satisfactorily improved by the use of either scheme. Therefore,
it is natural to consider the use of a combination of the lead and the lag con-
trollers, so that the advantages of both schemes may be utilized; at the same
time, some of the undesirable features in each are eliminated in the combined
structure.
The transfer function of such a lag-lead (or lead-lag) controller may be
written

g 1M2 >
-«-(t^)(ttt#) <

I
<- lead -> 1
1
<- lag - |

where a > 1 and b < 1, and the attenuation factor 1 /a is not shown in the equa-
tion if we assume that adequate loop gain is available in the system to compen-
sate for this loss.
Usually it is not necessary to cascade the lead and the lag networks of Figs.
10-10 and 10-25 for the realization of Eq. (10-82) if a and b need not be specified
independently. A network that has lag-lead characteristics, but with fewer num-
ber of elements, is shown in Fig. 10-38. The transfer function of the network is

-vwv-

R2

C2 i

Fig. 10-38. Lag-lead network.

Sec. 10.5 Lag-Lead Compensation / 553

M _~ e£,(j)
G A) 2 (s)
(i + *,c,jX1 + R c s) 1 2
(10 .
83)
~ 1 + (J?,C, + * ,C + tf C > + R^C^s
2 2 2
2 v

Comparing Eq. (10-82) with Eq. (10-83), we have

«7\= ^C, (10-84)

AT = * C 2 2 2
(10-85)

T T =R R CC
l 2 l 2 1 2
(10-86)

From Eqs. (10-84) and (10-85), we have

abT T 2 1
= RiRiCiCt (10-87)

Thus
ab = 1 (10-88)

which means that a and A cannot be specified independently.

Example 10-5 In this example we shall design a lag-lead controller for the control
system considered in Examples 10-2 and 10-4. The open-loop transfer
function of the original system is repeated as

W = ,(1 + 0.1 *(1 + 0.2.) < 10

- 89
>

The performance specifications are

1. K =
v 100 sec" 1
.

2. Phase margin > 40°.

These requirements have been satisfied by the phase-lag controller designed in
Example 10-4. However, it is noted that the phase-lag controller yielded a step response
that has a relatively large rise time. Tn this example we shall design a lag-lead controller
so that the rise time is reduced.

Let the series controller be represented by the transfer function

GM
r<\ = (1 + aT,s)(l + bT s) 2
no9m
(10 - 9 °)
(1 + ri ,xl + TlS)
For the first part we consider that the lag-lead controller is realized by the network of
Fig. 10-38, so that the coefficients a and b are related through Eq. (10-88).

In general, there is no fixed procedure available for the design of the lag-lead
controller. Usually, a trial-and-error procedure, using the design techniques outlined
for the phase-lag and the phase-lead controllers, may provide a satisfactory design
arrangement.
Let us first determine the phase-lag portion of the compensation by selecting the
proper values of T2 and b of Eq. (10-90). The Bode plot of Gp (s) of Eq. (10-89) is
sketched in Fig. 10-39 for K= 100. We arbitrarily choose to move the gain-crossover
frequency of Gp (ja>) from 17 rad/sec to 6 rad/sec. Recall that in Example 10-4, the

desired new gain-crossover frequency is 3.5 rad/sec when phase-lag compensation

alone is used. With the gain-crossover frequency at 6 rad/sec, the phase margin of the
system should be improved to approximately 10°. Using the phase-lag design technique,
we notice that the attenuation needed to bring the magnitude of G„{jco) down to dB
at co'c = 6 rad/sec is —22 dB. Thus using Eq. (10-59), we have

b = io- 22 20
'
= 10- 1 ' =0.08 (10-91)
554 / Introduction to Control Systems Design Chap. 10

80

60
gpX(S )' = m
s(l+0.\s)(l+0.2s
40

20

dB
-20
100(1 + 0.32s)(l +1.667^)
- 40 - s(\ +0.1s)(l + 0.2s)(l +0.0257i)(l + 20.83*)

-60

Phase margin of
compensated system
~ 90
°h- Phase margin of
uncompensated system
180° -

-270° J_
«">,.
0.048 0.1 0.6 1.0 3.11 38.9 100

tj rad/sec

Fig. 10-39. Bode plots of uncompensated system and compensated system

with lag-lead controller in Example 10-5.

Placing \jbT2 at 1 decade of frequency below co'c = 6 rad/sec, we have

j-jt = -j-jj
= 0.6 rad/sec (10-92)
Thus
20.83 (10-93)
and

0.048 (10-94)
T2
The phase-lag portion of the controller is described by

1 + 1.667j
1 + 20.83.?
Now we turn to the phase-lead part of the controller. Since a is equal to the
inverse of b, the value of found to be 12.5. From Eq.
a is (10-20), the maximum phase
lead that corresponds to this value of a is

a- 1
sin0„
a+ = 0.8518
1
(10-95)
or

(f> m = 58.3 deg

The zero-frequency attenuation of the phase-lead controller is

Sec. 10.5 Lag-Lead Compensation / 555

20 Iog, a = 20 log, 12.5 = 21.9 dB (10-96)

Using the procedure outlined earlier, the new gain-crossover frequency is found to be
at co„= 11 rad/sec. Then, using Eq. (10-15), we get

r,
V« a>m = 38.: (10-97)

and
3.11
aT y

The transfer function of the lag -lead controller is determined as

G (1 +0.32s)(l + 1.6675) (10-98)

c (s)
(1 + 0.0257j)(1 + 20.83*)
where as a usual practice, the attenuation factor in the phase-lead portion has been
dropped.
Figure 10-39 shows that the phase margin of the compensated system is approxi-
mately 40°. The open-loop transfer function of the compensated system with K = 100
is

G(s) = G (s)Gp (s) 4981(* + 3.11)(j + 0-6) (10-99)

c
s(s + 5)(s + 10)(.s + 38.88) (s+ 0.048)

The unit step response of the compensated system is shown in Fig. 10-40. It is apparent

1.5

System with phase-lag

controller

1.0 ^._-»"

System with lag-lead

controller

0.5

I 1 2 3 4 5 6

Time (seconds)

Fig. 10-40. Unit step response of the system in Example 10-5 with lag-lead
controller.
 566 / Introduction to Control Systems Design
Chap. 10

5-plane

(- 4.159 +/1 0.556) /K= 4981

AT = 4981 K=
-
42.627 X 38.88

Fig. 10-41. Root locus diagram of the system in Example 10-5 with lag-
lead controller.

that the step response of the system with the lag-lead controller is much improved
over that of the phase-lag compensated system. Not only the overshoot is smaller, but
the rise time is also greatly reduced. We may attribute these improvements
to the
addition of the phase-lead portion of the controller. It can be easily verified that the
bandwidth of the closed-loop system is now approximately 18 rad/sec, which is 3
times that of the system with the phase-lag controller alone.
The root locus diagram of Fig. 10-41 shows that the dominant complex roots of
the lag-lead compensated system correspond to a much larger natural undamped
frequency than that of the system with only the phase-lag controller designed in Exam-
ple 10-4, Fig. 10-36. This accounts for the fact that the lag-lead compensation gives a
wider bandwidth and thus a shorter rise time. The root locus diagram of Fig. 10-41
also shows that for all practical purposes, the transfer function of the closed-loop
system with the lag-lead controller can be approximated by

C(s)
= 5487
10 - 10°)
R(s) (* + 42.627X* + 4.159 +/10.556)(j + 4.159 -7T0556) (

where the closed-loop poles at —0.616 and -2.76 are effectively canceled by the zeros
at —0.6 and —3.11, respectively.
It should be pointed out that we have been fortunate to have found a solution to
the design problem on the first trial. In general, any trial-and-error design scheme may
 Sec. 10.6 Bridged-T Network Compensation / 557

not be as straightforward, especially since there is a constraint between a and b of Eq.

(10-90). In other words, once we selected a value for b, a is also determined. It is
entirely possible that the combination does not satisfy the design specifications.

10.6 Bridged-T Network Compensation

Many controlled processes have transfer functions that contain one or more
pairs of complex-conjugate poles. One of the distinct features of a pair of com-
plex poles is that the gradient of the phase with respect to frequency near the
gain-crossover frequency is large. This is similar to the situation when two or
more simple poles of the transfer function are placed close to each other. With
reference to the system considered in Examples 10-2 and 10-4, the reason the
phase-lead controller is ineffective in improving the relative stability of the sys-
tem is because the phase of the process has a steep negative slope near the gain-
crossover frequency (see Fig. 10-22). When a process has complex poles, especially
if the poles are close to the imaginary axis, the design problem may become

more acute. We may suggest the use of a controller that has a transfer function
with zeros so selected as to cancel the undesired complex poles of the origi-
nal process, and the poles of the controller are placed at the desired locations
in the s-plane. For instance, if the transfer function of a process is

G » = ^ +* + io)
(
1(M01)

in which the complex-conjugate poles are causing stability problems in the

closed-loop system, the suggested series controller may have the transfer func-
tion

Gc{s) = /'+/+]° (1(W 02)

The constants £ and ©„ are determined according to the performance

specifications of the system.Although the transfer function of Eq. (10-102) can
be realized by various types of passive networks, the bridged-T networks have
the advantage of containing only RC elements. Figure 10-42 illustrates two basic
types of the bridged-T RC networks. In the following discussion, the network
shown in Fig. 10-42(a) is referred to as the bridged-T type 1, and that of Fig.
10-42(b) is referred to as type 2.
With the assumption of zero input source impedance and infinite output
impedance, the transfer function of the bridged-T type 1 network is given by

E 2 (s) _ + 2RC s + R C,C s

1 2
2
2
2
(10-103)
E^s) 1 + £(C, + 2C )s + R C C s
2
1
l z
1 v

and that of the bridged-T type 2 network is

E 2 (s) _ 1 + 2R.Cs + C2 R,R 2 s 2 (10-104)

v '
Ei(s) 1 + C(R 2 + 2R,)s + C 2R,R 2 s 2

When these two equations are compared, it is apparent that the two net-
568 / Introduction to Control Systems Design
Chap. 10

H(-
R R
-\MAr-

£l ^c,

(a)

R2
-WAr-

C C
-K-

(b)

Fig. 10-42. Two basic types of bridged-T network, (a) Type 1 network, (b)
Type 2 network.

works have similar transfer characteristics. In fact, if R, C,, and C in Eq.

2
(10-103) are replaced by C, R 2 and R u respectively, Eq. (10-103) becomes the
,

transfer function of the type 2 network given by Eq. (10-104).

It is useful to study the behavior of the poles and zeros of the transfer func-

tions of the bridged-T networks of Eqs. (10-103) and (10-104) when the net-
work parameters are varied. Owing to the similarity of the two networks, only
type 1 will be studied here.

Equation (10-103) is written as

1
*
2
T^S ^ R C^C
+ RCi %
2
(10-105)
Ms) Cj+2C s 2 1
'
RC^ + RtCtCz '

If both the numerator and the denominator polynomials of Eq. (10-105) are
written in the standard form

2Cco„s + col = (10-106)

we have, for the numerator,

co„ 2 =+ -
(10-107)
R*JC C2 X

(10-108)
 L

Sec. 10.6 Bridged-T Network Compensation / 559

and fdr the denominator,

<o„„ = ± 1
= co„ (10-109)
RjC.C,
r -C + '
2C2 _ 1 + 2C 2 /C, _ 1 + 2tf
(10-110)
*"
2yc,C 2 2VQ/C, 2£,
The loci of the poles and zeros of Ez (s)IE {s) of Eq. (10-105) when C
u Ci, l

and R vary individually are sketched in Fig. 10-43. When R varies, the numerator
and the denominator of Eq. (10-105) contain R in the form of R\ and the root
locus method cannot be applied directly to this nonlinear problem.
Fortunately,

Root locus of zeros Root locus of poles

s-plane s-plane /CO

0-C, *- c i
C, -> oo c, =
» O X > < >

/?C2 2/?C2

/CO
s-plane o /CO
ii
s-plane

'

C2 ^°° C2 -*°o "* c2 C2 -» °° C2 =

2 10
e— **-<>
2 __J_
RC l
' I

RCy RCy

II

s = -_ ( i
+ /^Z^ ,.

i_A /?-»oo
s-plane
/CO

C2 >C 1
(C*< 1)
/?C,
0*-/?\ /?-*oo

RC

RC,

C2 < C, (C > 1)

Fig. 10-43. Loci of poles and zeros of the bridged-T type 1 network.
560 / Introduction to Control Systems Desiign
Chap. 10

equations that are of the form of Eq. (10-106) are of the second
in this case, the
order and can be solved easily. Therefore, the pole and zero loci of Fig. 10-43
show that the two zeros of the bridged-T network type can be either real or 1

complex; for complex zeros, C 2 must be less than C The

y
. poles of the transfer
function always lie on the negative real axis.
The £ and parameters of the denominator and the numerator of the
co„
transfer function of the type 2 network may be obtained by replacing R, C,, and
C 2 in Eqs. (10-107) through (10-1 10) by C, R 2 and R u respectively. Thus ,

COnz = ±: (10-111)
C*/R R 2 1

(10-112)

(o„ (o„ 2 (10-113)

2*.
/
(10-114)
2 v /?i/?2
The root loci shown in Fig. 10-43 can still be used for the type 2 bridged-T
network if the corresponding symbols are altered.

Example 10-6 The control system shown in Fig. 10-44 is selected to demonstrate
the use of the bridged-T network as a controller
and the principle of
improving system performance by pole-zero cancellation.

R(s) 3>\ *M , Controller

K(\ + lO.s) C(s)

5(1 + 0.2s + 0.25s 2 )

Fig. 10-44. Feedback control system for Example 10-6.

The controlled process has the transfer function

G,(s) = ATO+JOs)
(10-115)
.5(1 + 0.2* -f 0.25.S 2 )

The root locus diagram of the uncompensated closed-loop system is shown in Fig.
10-45. It is shown that although the closed-loop system
always stable, the damping is

of the system is very low, and the step response of the system will be quite oscillatory
for any positive K. Figure 10-47 illustrates the unit step response of the system when
K = 1. Notice that the zero &\ s = —0.1 of the open-loop transfer function causes the
closed-loop system to have an eigenvalue at * =
—0.091, which corresponds to a time
constant of 1 1 sec. Thus this small eigenvalue causes the step response of Fig. 10-47 to
oscillate about a level that is below unity, and it would take a long time for the response
to reach the desired steady state.
 Sec. 10.6 Bridged-T Network Compensation / 561

/CJ Let us select the type 1 bridged-T network as series com-

t pensation to improve the relative stability of the system. The
K complex-conjugate zeros of the network should be so placed
K= 1 - 0.354 +/6.62 that they will cancel the undesirable poles of the controlled
process. Therefore, the transfer function of the bridged-T net-
work should be
i-plane
GAs)
j' 2
+ 0.8s + 4
(10-116)
S 2
+ + CO
2t P co op s
Using Eqs. (10-107) and (10-108), we have
/3
(10-117)
/?V C\Cz
-0.4+ /1. 96
(10-118)

K =
From Eqs. (10-109) and (10-110),

One root at
W„„ = co„, = 2 (10-119)
- 0.093 (K= and
1)

K = o: -0.4-/1.96 c,=
1 + 2E = 2.7 (10-120)

;3 The transfer function of the type 1 bridged-T network is

Gc (s) = + 0.8s + 4
s1
(10-121)
(J + 0.384)(j + 10.42)
-/5

0.354 - /6.62
The open-loop

^ BC ^ > XU
transfer function of the

+
The root loci of the compensated system are sketched as
compensated system

+
is

(10 " I22)

shown in Fig. 10-46. The root loci show that for K greater
than 0.64, two of the eigenvalues of the system are complex.
Root loci of the feedback control
Fig. 10-45. However, the dynamic response of the system is still domi-
system in Example 10-6 with G (s) = nated by the eigenvalue that lies near the origin of the s-plane.
p
[AC(1 + 10s)]/[.s(l + 0.2s + 0.25^)]. Figure 10-47 illustrates the unit step response of the compen-
when
sated system 1 K=
In this case the complex eigenvalues
.

of the compensated system cause the response to oscillate only slightly,

but the eigen-
value at s =
-0.093 causes the system response to take a very long time in reaching
its steady state of unity. It is apparent that the
relative stability of the system is greatly
improved by the bridged-T controller through pole-zero cancellation.

Although the preceding design is carried out on the basis that the undesir-
able poles of the controlled process are exactly canceled by the zeros of the
con-
troller, in practice this ideal situation is difficult to achieve.
One of the difficulties
lies in the fact that we cannot design the controller so
that its poles and zeros
will fall exactly at the specified locations. Another problem
is lack of knowledge
as to exactly where the poles of the process are. Usually, in the determination
of
the transfer function of a process, assumptions and approximations
have to be
made, so that the mathematical description of the system is never precise. In
562 / Introduction to Control Systems Design
Chap. 10

. /cj

s-plane

K=\ S.33+/3.8

K=
-»-o
10.4

One root at
- 0.093 when K= 1

K= 1 5.33-/3.8

Fig. 10-46. Root loci of feedback control system in Example 1 0-6 with
bridged-T controller.

view of these practical considerations, in practice, a more realistic situation in

the attempt of the pole-zero cancellation design is at best a "near cancellation."
Let us consider that the design of the bridged-T compensation of the system
in Example 10-6 resulted in an inexact pole-zero cancellation, so that the trans-
fer function of the controller is

G c (s)
= s1+ 0.76s + 3.8 (10-123)
(s + 0.366)0 + 10-4)
Since the complex poles of the controlled process are not canceled by the zeros
of G c (s), the open-loop transfer function of the system becomes

w=
G(s )
40K(s + Q.1Xj» +
0J6s 3.8) + (10-124)
s(s + 0.366)0 +
10.4)0 2 0.8j +
4) +
For all practical purposes, the complex poles and zeros of G(s) of Eq. (10-124)
can still be considered as close enough for cancellation. For K= 1, the closed-
loop transfer function is
 Sec. 10.6 Bridged-T Network Compensation / 563

i T n 1

Uncompensated system

c(t)

5 10

Seconds

Fig. 10-47. Unit step responses of the compensated and the uncompen-
sated systems in Example 10-6.

C(£) 40(,y + 0-l)Cr

2
+ 0-76a + 3.8) (10-125)
R(s) (s + 0.093)O
2
+ 0.81 ^ + 3.805)O
2
+ 10.66.S + 42.88)

Since the zeros of the open-loop transfer function are retained as zeros of the
closed-loop transfer function, Eq. (10-125) verifies that near cancellation of
poles and zeros in the open-loop transfer function will result in the same situa-
tion for the closed-loop transfer function. Therefore, we may conclude that from
the transient response standpoint, the effectiveness of the cancellation com-
pensation not diminished even if the pole-zero cancellation is not exact.
is

Alternatively, we can show that if the partial-fraction expansion of Eq. (10-125)

is carried out, the coefficients that correspond to the roots of s
2
0.8 Is + +
3.805 = will be very small, so that the contribution to the time response from
these roots will be negligible.
If the undesirable poles of the open-loop transfer function are very close to
or right on the imaginary axis of the .s-plane, inexact cancellation may result in a
conditionally stable or an unstable system. Figure 10-48(a) illustrates a situation
in which the relative positions of the poles and zeros intended for cancellation
 .

564 / Introduction to Control Systems Design

Chap. 10

/CO

X?fC=°o
s-plane

K=

(a)

' /'co

>
5-plane K=

K=

K=Q
(b)

P.K = °°

Fig. 10-48. Root locus diagrams showing the effects of inexact pole-zero
cancellations.

result in a stable system,whereas in Fig. 10-48(b), the inexact cancellation is

unacceptable, although the relative distance between the poles and zeros
intended for cancellation is small. Normally, the two situations may be dis-
tinguished simply by the use of the angles of departure properties of the root loci.

REFERENCES

1 H. W. Bode, Network Analysis and Feedback Amplifier Design, D. Van Nostrand

Reinhold Company, New York, 1945.
2. H. M. James, N. B. Nichols, and R. S. Phillips, Theory of Servomechanisms,
McGraw-Hill Book Company, New York, 1947.
Chap. 10 Problems / 565

3. N. Wiener, Extrapolation, Interpolation, and Smoothing of Stationary Time

Technology Press, MIT, Cambridge, Mass., 1949.
Series,

4. W. R. Evans, "Control System Synthesis by Root Locus Method," Trans. AIEE,

Vol. 69, pp. 66-69, 1950.

5. J. G. Truxal, Automatic Feedback Control System Synthesis, McGraw-Hill

Book Company, New York, 1955.

PROBLEMS

10.1. The open-loop transfer function of a feedback control system is given by

jr
G(s) =
s(l + 0.2i)(l + 0.5s)
The feedback is unity. The output of the system is to follow a reference ramp
input to yield a velocity of 2 rpm with a maximum steady-state error of 2°.
(a) Determine the smallest value of K that will satisfy the specification given
above. With this value of K, analyze the system performance by evaluating
the system gain margin, phase margin, M„, and bandwidth.
(b) A lead compensation with the transfer function (1 + 0As){\ + 0.08*) is

inserted in the forward path of the system. Evaluate the values of the gain
margin, phase margin, M p, and bandwidth of the compensated system.
Comment on the effects of the lead compensation of the system perfor-
mance.
(c) Sketch the root loci of the compensated and the uncompensated systems.
10.2. A type-2 control system is shown in Fig. P10-2. The system must meet the
following performance specifications:
(a) Acceleration constant K = 5 sec -2
a .

(b) The resonance peak M p < 1.5.

*-C(s)

Figure P10-2.

Design a series phase-lead controller to satisfy these requirements. Sketch the

root loci for the uncompensated and compensated systems. What are the values
of the damping ratio (of the complex roots) and the bandwidth of the compen-
sated system ?

10.3. Figure P10-3 illustrates the block diagram of a positioning control system.
The system may be used for the positioning of a shaft by a command from a
remote location. The parameters of the system are given as follows:

ze = 0.01 sec

/ = 0.05 oz-in-sec 2
566 / Introduction to Control Systems Design Chap. 10

Motor
electric Motor and
circuit load

Km
Ka Gc (s)
1 +Te S

Figure P10-3.

= 0.5 oz-in-sec
B
Km = 10 oz-in/volt
TL = disturbance torque

(a) Determine the minimum value of the error sensor and amplifier gain Ka so
that the steady-state error of 9 due to a unit step torque disturbance is
less than or equal to 0.01 (1 per cent).
(b) Determine the stability of the system when Ka is set at the value deter-
mined in part (a).
Design a phase-lead controller in the form

= 1 +aTs
G c {s)
1 Ts
a> 1

so that the closed-loop system has a phase margin of approximately 40°.

Determine the bandwidth of the compensated system. Plot the output
response O (O when the input r (t) is a unit step function. (TL = 0.)
10.4. Repeat the design problem in Problem 10-3 with a phase-lag controller

1 +aTs
G c (s)
1 +Ts a < 1

10.5. Human beings breathe in order to provide the means for gas exchange for the
entire body. A respiratory control system is needed to ensure that the body's
needs for this gas exchange are adequately met. The criterion of control is
adequate ventilation, which ensures satisfactory levels of both oxygen and
carbon dioxide in the arterial blood. Respiration is controlled by neural
impulses that originate within the lower brain and are transmitted to the chest
cavity and diaphragm to govern the rate andtidal volume. One source of the
signals the chemoreceptors located near the respiratory center, which are
is

sensitive to carbon dioxide and oxygen concentrations. Figure P10-5 shows the
block diagram of a simplified model of the human respiratory control system.
The objective is to control the effective ventilation of the lungs so that satis-
factory balance of concentrations of carbon dioxide and oxygen is maintained
in the blood circulated at the chemoreceptor.
(a) A normal value of the chemoreceptor gain Kf is 0. 1 . Determine the gain
margin and the phase margin of the system.
(b) Assume that the chemoreceptor is defective so that its gain K is increased
f
to 1.Design a controller in the forward path (to be inserted in front of the
block representing the lungs) so that a phase margin of 45° is maintained.
 Chap. 10
Problems / 567

Lungs Circulatory system

Concentration
0.1 CO)
Desired
+ 0.5)0 + 0.1)0 + 0.2)
ventilation

V
Chemoreceptor

Figure PI 0-5.

10.6. This problem deals with the cable-reel unwind

process described in Problem
5-16 and shown in Fig. P5-16. The inertia of
the reel is

JR = 187?* — 200 ft-lb-sec 2

where R is
the effective radius of the cable reel.
(a) Assume that R
and JR are constant between layers of the cable. Determine
the maximum value of the amplifier gain
K
so that the entire unwinding
process is stable from beginning until end.
(b) Let K= 10. Design a series controller so that the system has a phase
margin of 45° at the end of the unwind
process {R = 2 ft). With this
controller, what are the phase and gain
margins of the system at the begin-
ning of the unwind process? Sketch the
root loci of the compensated
process with K = 10 and indicate the variation of the roots on the loci as
the unwind process progresses. It should be noted
that the treatment of
this problem by a transfer function
is only an approximation; strictly, the
process is nonlinear as well as time varying.
10.7. Figure P10-7 shows the block diagram of
the speed control system for an
electrical power generating system. The speed governor
valve controls the
steam flow input to the turbine. The turbine
drives the generator, which puts
out electric power at a frequency
proportional to the generator speed co The
g

Speed
governor
valve
Power system
generator

Reference
input
voltage

Tachometer

Figure PI0-7.
568 / Introduction to Control Systems Design Chap. 10

desired steady-state speed of the generator is 1200 rpm, at which the generated

output voltage 60 Hz. / = 100.

is

(a) Let the speed governor valve gain K be set at 10 rad/volt. Determine the
tachometer gain so that the complex eigenvalues of the closed-loop system
correspond to a damping ratio of 0.707. Sketch the root loci as a function
of KT and indicate the location of the roots with the desired damping.
(b) Determine the desired reference input voltage, with the value of KT set at
the value determined in (a), so that the generator speed is 1200 rpm
(TL = 0).
(c) In Fig. P10-7, TL denotes a load change. Determine the per cent change in
K
the steady-state speed due to a constant load change when T is as deter-
mined in (a).
(d) It is desired to keep the frequency variation due to load change within
±0.1 per At the same time, the relative damping of the complex
cent.
eigenvalues of the overall system must be approximately 0.707. Can both
requirements be satisfied by only changing the values of A" and KT 1 If not,
design a series controller in the forward path for this purpose. Sketch the
root locus for the compensated system with K = 10, with KT as the
variable parameter.

10.8. The phase-lock loop technique is a popular method for dc motor speed control.
A basic phase-lock loop motor speed control system is shown in Fig. P10-8.

Reference
speed

Eo Motor ym speed
control Motor
circuit <0

Counter Encoder

*2 C
I— MM 1(-

l
i

o vwv-
Em t>
Filter

Figure P10-8.

The encoder produces digital pulses that represent motor speed. The pulse
from the encoder is compared with the reference frequency by a phase
train
comparator or detector. The output of the phase detector is a voltage that is
proportional to the phase difference between the reference speed and the
actual motor speed. This error voltage upon filtering is used as the control
signal for the motor. The system parameters and transfer functions are as
follows:
 Chap. 10
Problems / 569

Phase detector Kp = 0.0609 volt/rad

Motor control circuit gain Ka = 1

Motor transfer function QM

Vm (s)
= K"
s(\ + Tm s)
(K -tor
{Km - I0 T <
"> 0.04)
Encoder gain K = 5.73 pulses/rad
e

Ea_ __ R Cs + 2 1
Filter transfer function
E in RiCs
*i = 1.745 x 10 Q, C, = fiF 6
1

Counter 1_
1
N
Determine the value of R 2 so that the complex roots of the closed-loop
system
have a damping ratio of approximately 0.707. If the problem
has more than
one solution, use the one that corresponds to the smallest
system bandwidth.
Sketch the root locus diagram for the characteristic
equation roots as R 2
varies.

10.9. Tachometer feedback is employed frequently in feedback control systems

for
the purpose of stabilization. Figure P10-9 shows the
block diagram ofa system
with tachometer feedback. Choose the tachometer gain
constant K, so that the
relative damping ratio of the system is 50 per cent.
How does the tachometer
feedback affect the bandwidth of the system ?

R(s)

X^
— , *
10
s(s+ I)
a*)

K,s

Tachometer

F gure P10-9.

10.10. The block diagram of a control system is shown in Fig. P10-10.

By means of
the root contour method, show the effect of variation
in the value of T on the
location of the closed-loop poles of the system.

R(s)

KX^
E(s)
10 C(s)

s(l +0.1.S) 2

-T

1 + Ts

Figure P10-10.
570 / Introduction to Control Systems Design Chap. 10

10.11. A computer-tape-drive system utilizing a permanent-magnet dc motor is

shown in Fig. P10-ll(a). The system is modeled by the diagram shown in Fig.
P10-ll(b). K
The constant L represents the spring constant of the elastic tape,
and BLdenotes the viscous frictional coefficient between the tape and the
capstans. The system parameters are as follows

36 oz-in/volt 0.023 oz-in-sec 2

Ra
K = e 6.92 oz-in/rad/sec KL = 2857.6 oz-in/rad

BL = 10 oz-in-sec JL = 7.24 oz-in-sec 2

where K = b back emf constant, K, = K K IR + Bm Ka = torque

b a a , constant
in oz-in/amp, and B, motor viscous friction constant.
(a) Write the state equations of the system shown in Fig. P10-ll(b) with 9 L ,
Q>l, m and
, co m as the state variables in the prescribed order. Derive the
transfer functions

Tm'U m e ,

+ o WW

(b)

Amplifier

Controller

Motor
9m
rr, i> '
*o,

Speed
transducer

(c)

Figure P10-11.
 :

Chap. 10 Problems / 571

a>m(s) co L(s)
and
Ea {s)
(b) The objective of the system is to control the speed of the load, co L accu- ,

rately.Figure P10-ll(c) shows a closed-loop system in which the load

speed is fed back through a speed transducer and compared with the
reference input, with f = 0.01. Design a controller and select the ampli-
K
fier gain so that the following specifications are satisfied: (1) no steady-state
speed error when the input e r is a step function; (2) the dominant roots of
the characteristic equation correspond to a ratio of approxi- damping
mately 0.707 and (3) what should be the value of the input e r if the steady-
;

state speed is to be 100 rpm? Sketch the root loci of the designed system
with K
as the variable parameter.

10.12. The computer-tape-drive system described in Problem 10-1 1 has the following
system parameters

= 36 oz-in/volt Jm = 0.023 oz-in-sec 2

Ra
_ K K + Bm =
b c
6.92 oz-in/rad/sec BL =
Ra
KL = 28,576 oz-in/rad Kf = 0.01

JL = 7.24 oz-in-sec 2

(a) Show that the closed-loop system without compensation has an oscillatory
response in co L for any positive K.
(b) In order to reduce the speed oscillations, a bridged-T network is proposed
to cancel the undesirable poles of the open-l6op transfer function. Figure
P10-12 shows the block diagram of the overall system. The phase-lag
controller preceding the bridged-T controlleris for the purpose of ensuring

zero steady-state speed error when a step input is applied. Determine the
transfer function of the bridged-T network and the amplifier gain so that
the dominant roots of the characteristic equation correspond to a damping
ratio of approximately 70.7 per cent.

Phase-lag
Amplifier network
-j-VVW-n
K > s + 0.953
5
— *-
Bridged-T
network

Speed
transducer

Figure P10-12.
 :

11
Introduction to Optimal

Control

11.1 Introduction

From the discussions given in Chapter 10 we realize that the conventional design
of feedback control systems has many limitations. The most important disad-
vantage of the conventional design approach is that these methods are not

rigorous and rely heavily on trial and

For a system with multiple inputs
error.
and outputs and a high degree of complexity, the trial-and-error approach may
not lead to a satisfactory design.
The optimal control design is aimed at obtaining a system that is the best
possible with respect to a certain performance index or design criterion. The
state-variable formulation is often used for the representation of the dynamic
process. Since the optimal control problem is more of a mathematical develop-
ment, the problem is stated as
Find the control u(t) that minimizes the performance index,

J =['' F[x{t),n{t),t]dt (11-1)

J to

given the initial state x(/ ) = x and subject to the constraint that the process is
described by
i(0 = f[x(r), u(r), t] (11-2)

The performance index is a scalar quantity that consists of an integral of

a scalar function of the state vector and the control vector. Notice that the
state equationswhich describe the dynamics of the system to be controlled are
interpreted as constraints to the design problem. In general, additional con-
straints on the state variables and/or the control are often used.

572
Sec. 11.1 Introduction / 573

In the optimal control design, the performance index J replaces the conven-
tional design criteria such aspeak overshoot, damping ratio, gain margin, and
phase margin. Of course, the designer must be able to select the performance
index properly so that the resulting system from the design will perform satis-
factorily according to physical standards which are more easily interpreted by
conventional performance criteria.
It must be noted that the control u(r) is usually not what has been referred

to as the reference input in conventional design terminology. Since the state

equation of Eq. (11-2) describes the controlled process only, the statement of the
optimal control problem may give one the misconception that the problem deals
with the design of an open-loop system. However, when the design is accom-

plished, usually, the optimal control will be of such a nature that it depends on
the output or the state variables, so that a closed-loop control system is naturally
formed.
To illustrate the various possible formulations of the performance index /,
the following examples are given.

Minimum-Time Problem
One of the most frequently encountered problems in optimal control is the
design of a system that will be able to drive its states from some initial condition
to the final state in minimum time. In terms of the performance index of Eq.
(1 1-1), the minimum-time problem may be stated so as to minimize

f
•> to
dt (11-3)

Thus, in Eq. (1 1-1), F[x(t), u(r), t]=\.

Linear Regulator Problem

Given a linear system, the design objective may be to keep the states at the
equilibrium state, and the system should be able to return to the equilibrium
statefrom any initial state. The performance index for the linear regulator
problem is
J =$ [" x'Qxdt (11-4)

where Q is a symmetric positive-definite matrix. Or,

/ = £[" [x'Qx + u'Ru] dt (1 1-5)

•* to

where a constraint is placed on the control vector u. The matrices Q and R are

called the weighting matrices, as their components put various weights on the
elements of x and u in the optimization process.

Tracking Problem

If it is desired that the state x of a given system track, or be as close as possible

to a certain target trajectory x d the problem may be formulated by minimizing
,

/ = f" (x - x„)'Q(x -x )dt d (11 -6)

•I <<,
574 / Introduction to Optimal Control Chap. 11

In addition to these examples, there are the minimum-energy criterion,

minimum-fuel criterion, and many others.

11.2 Analytical Design 1

One of the first design methods that eliminates the trial-and-error uncertainties
is analytical design, sometimes known as minimum integral-square-error design.
The design is forms a bridge between the conventional
significant in that it

design philosophy and the modern optimal control principle. For the first time,
the designer can simply set up the performance index and then carry out the
complete design in a mathematical and analytical way. The only problems the
designer has to be concerned with are: (1) if the design has a solution, and (2) if
the solution can be physically implemented. The similarity of analytical design
to the conventional approach is that the design is carried out in the frequency
domain or, more appropriately stated, the ^-domain. However, none of the
frequency-domain graphical techniques is necessary.
We first begin with the system configuration of Fig. 11-1. The assumptions
are that the system is linear, time invariant, and that there is only one input and
one output. Systems with multivariables may be designed by the analytical
method, but the procedure is much more involved. Therefore, in the current
discussion we are restricted to single-variable systems.

KO ~\ eit) Linear uit) Linear

*- cit)

J controller

, i
process G(s)

Fig. 11-1. Block diagram of a linear system with a controller.

The linear controlled process of the system in Fig. 11-1 is described by the

transfer function G(s). The control signal and the output is c(t). Notice
is u{i),

that the block diagram of Fig. 1 1-1 is not an algebraic one in that the controller
isportrayed in such a way that its specific input-output configuration is not yet
defined. This allows the flexibility of selecting the final controller configuration,
such as a series controller or a feedback controller, once the design is completed.
The performance index J is composed of the following two integrals:

ye = C e\t)dt (11-7)
J

Ju = Cu\t)dt (11-8)
J

where e(t) is the error signal and u(t) is the control signal. J„ is referred to as the
integral square error (ISE) and /„ is the integral square control (ISC).
 :

Sec. 11.2 Analytical Design / 575

The Je is that this performance index gives a flexible

significance of using
restriction on such peak overshoot, rise time, and relative stability. If
criteria as
the error between the output and the reference input is large, the value of Je will
be large. The ISC index Ju may be regarded as a measure of the energy required
by the control. Since u 2 (t) is proportional to the power consumed for control,
the time integral of u 2 (t) is a measure of the energy consumed by the system.
Ordinarily, the analytical design problem may be formulated as one that desires
the minimization of Je while keeping Ju within a certain limit, or minimizing /„
while keeping Je not greater than a certain value. Using the calculus-of-variation
method, the problem of minimizing one function while constraining another
function to a constant value can be solved by adjoining the constraint to
the function to be minimized (see Appendix C). For instance, if the optimiz-
ation problem is
minimize Je
and
Ju — K= constant
the problem is equivalent to

minimize J =J + e k 2 Ju (11-9)

where k z
is a parameter yet to be determined and is called the Lagrange multiplier.
On the other hand, if we wish to

minimize Ju
and
Je =K— constant

then the problem is equivalent to minimizing

J = Ju + k 2Jt (11-10)

The major development of the analytical design is the use of an equivalent

frequency-domain expression for Eq. (11-9) or Eq. (1 1-10) in terms of the system
transfer function through the use of Parseval's theorem. The version of the
Parseval's theorem that is applied to the present problem is

[ e 2 (t)dt = ~. f E(s)E(-s)ds (H-H)

where E(s) is the Laplace transform of e{t).

Let us formulate the problem as being that of finding the optimal control
for the system of Fig. 11-1 so that

J =J + e
k 2 Ju = P [e 2
(t) + k 2 u\t)] dt (H-12)
J

is minimized.
Applying Parseval's theorem to Eq. (11-12), we have

J = ^p f
'
\E{s)E{-s) + k 2 U(s) U(-s)] ds (1 1-13)
576 / Introduction to Optimal Control Chap. 11

With reference to Fig. 11-1,

E(s) = R(s) - G(s)U(s) (1 1-14)

Substituting Eq. (11-14) into Eq. (11-13), we get

[R(s) - G(s)U(s)][R(-s) - G(-s)U(-s)]ds

(11-15)

+ 2nj)_ U(s)U(-s) ds
J ,

Let Uok (s) be the optimal control, subject to the determination of k 2 . We

can write any arbitrary U{s) as

U(s)=UJs) + W 1
{s) (11-16)

where A is a constant and U^s) is any arbitrary function thai: has only poles in
the left half of the s-plane. The last condition is due to the requirement that u(t)
must be bounded. Substitution of Eq. (11-16) into Eq. (11-15) yields

J = J + k Ju = /, + X(J + /,) + A V
e
2
2 4 (11-17)
where

Jl = (** ~ R °°°k ~ RGU k + ° k * U*V* + G GVBk Uok ) ds (11-18)

db |
Jl = (<kl ^°
k ~ G* + GGVtWi ds t1 M9 >
5T I
J% = {- klUok -GR + GGUokWi ds (1 1-20)
2^7 /

/4 = _L (k* + GCftUiUi ds (11-21)

For simplicity, we have used G for G(s), Uok for Uok (s), Uok for Uok {—s), and
so on.
In Eq. (11-17), J t is the optimum performance index, that is, the minimum
/. Equations (11-19) and (11-20) indicate that J2 J3 since the integrand of J2 = ,

is identical to that of J3 with s replaced by —s. Also, J4 is always positive, since

its integrand is symmetrical with respect to the imaginary axis of the j-plane.
The necessary and sufficient conditions for /in Eq. (1 1-17) to be a minimum
are
dJ = (11-22)
A=0
and
d 2J
dk 2
>o (11-23)
-1=0

Applying Eq. (11-22) to Eq. (11-17) leads to

J2 +J = 2J =
} 3 (11-24)
and Eq. (11-23) gives
J* >o (11-25)
Sec. 11.2 Analytical Design / 577

which is always satisfied. Therefore, the necessary and sufficient conditions for
minimum J are obtained from Eqs. (1 1-24) and (1 1-20),

(k 2 Uok -GR + GGU ok )U ds

1
= (11-26)

Since we have assumed that all the poles of U %

are inside the left half of the
j-plane, the poles of C7, must all lie in the right half of the s-plane. One solution
of Eq. (1 1-26) is that all the poles of the integrand are in the right half of the
j-plane, and the line integral is evaluated as a contour integral around the left

half of the s-plane. Since the contour does not enclose any singularity of the
integrand, the contour integration is zero. Thus J3 of Eq. (11-26) is identically
zero if all the poles of

X(s) = GGUok -GR + k Uok 2

(1 1-27)

are located in the right half of the j-plane. Equation (11-27) is written

X(s) = (k 2 + GG)Uok - GR (1 1-28)

The function {k 2 + GG) is symmetrical with respect to the imaginary axis in the
j-plane. We let
Y(s) Y(-s) = YY=k 2 + GG (1 1-29)

where Y is a function that has only poles and zeros in the left half of the j-plane,
and Fis Y with s replaced by — s and has only right-half-plane poles and zeros.
The following notation is defined:

Y={k 2 + GG} + (11-30)

Y={k 2 + GG}~ (11-31)

The process of finding Y and Y is called spectral factorization; this is relatively

straightforward for scalar functions but is quite complex for multivariable
systems since matrix functions are involved.
Equation (11-28) is now written

X=YYU -GR ok (11-32)

We must now separate the right side of Eq. (1 1-32) into two parts, one with
poles only in the left and one with poles only in the right-half
half of the s-plane,
plane. This is accomplished by dividing both sides of the equation by Y, yielding

4 = YU ak -^ (11-33)

For convenience, we assume that R has no poles in the right-half plane.

Performing partial fraction of GRjY in Eq. (1 1-33), the result is written
GR CGR~ rGRT
_Y + + _ Y _
(11-34)
Y .

where
GR
expansion with poles in left-half plane (11-35)
_Y .

GK expansion with poles in right-half plane (11-36)

Y .
578 / Introduction to Optimal Control Chap. 11

Then, Eq. (11-33) becomes

X_ GR GR
YUok - (11-37)
Y
In this equation, the left-side terms have only poles in the right-half plane, and
the right-side terms have only poles in the left-half plane. Therefore, the left-side
terms must be independent of the right-side terms, or

Uak =T7 GR (11-38)

Note that Uok is still a function of

and it is the optimal control that minimizes
A:
2

the performance index of Eq. (11-12) for a given k 1 To find the absolute .

optimum value for Uok , which is U , we must use the constraint condition

= j_ r U(s)U(-s) ds =K (11-39)

to determine k 2 The
. integral of Eq. (11-39) may be evaluated with the aid of
Table 11-1.

Table 11-1 Tabulation of the Definite Integral

N(s)N(-s)
ds
,
D(s)D{-s)
N{s) = N„-is»-i + N„- 2 s"- 2 + + Nts + N . . .

D(s) = Dn s" + Dn-is"- + ... + DiS + D

1

Jl
IDoDt
j
2
_ N\D a + NlD 1
2D D D 1 1

j _ NjDoDj + (N - 2NoN1 )D D + NID 2 D

2
i 3
2DoD (-DoD + O1/J2)
3 3

/ = Nl(-DjD + DoD,D 2 + (N - 2N N )DqD D 4 + (N\ - 2NqN 2 )DoD + Nl(

2
. 3 ) 1 3 1 3 D4 -D\D\ + -O2O3P4)
2DoD^-DoD\ - D\D* + DiD 2 D 3)

Once Uok (s) is determined, if desired, the optimal closed-loop transfer

function can be easily determined. Let

M ok {s)
= C(s)
R(s)
(11-40)

which is the optimal closed-loop transfer function subject to finding k 2 Since .

C(s) = G(s)Uok (s) (11-41)

M ° k{s)
= u k{s) (11 " 42)
W) °

or

M ok = R-y ~GRT (11-43)

Sec. 11.2 Analytical Design / 579

Equations (1 1-38) and (1 1-43) are made more useful by defining

G(s) (11-44)

where N(s) and D(s) are polynomials of s.

Then,
Y= {k 2 + GG} +
= \
k2 DD + NN )
+
_ {k DD + NN} + 2 (11-45)

1 DD J D
and
-
_
~~
{k 2 DD + NN}- (11-46)
D
Equation (1 1-38) now gives

U„ k = {k 2
D NR
DD + NN} + L{k DD + NN}-J + 2
(11-47)

Similarly, Eq. (11-43) leads to

M = N NR (11-48)
nk
R{k 2 DD + NN} + l{k DD + NN} 2

which is the optimal transfer function of the closed-loop system, subject to

finding k1 .

Example 11-1 Referring to the block diagram of Fig. 11-1, consider that

n \
10
(11-49)

and the input transform is

«•>-¥ (11-50)

Determine the optimal control U (s) and the optimal closed-loop transfer function
M (s) so that

(1) Je = e 2 (t) dt = minimum (11-51)

J

(2) /„ = f~ u\t) dt
J o
< 2.5 (11-52)

Since N— 10 and D = s 2 we have ,

{k 2 DD + NN}* = [k 2
s* + 100} + (11-53)

To carry out the spectral factorization of the last equation, we write

k 2 s* + 100 = {a t s 2 + a,5 + Oo)(«2* 2 - a lS + a )

(11-54)
= als* - (a\ - 2a a )s 2 + a 2 2

Equating like coefficients on both sides of Eq. (11-54),

a\=k 2
(11-55)
 1

$80 / Introduction to Optimal Control Chap. 11

a2 =k (11-56)

a\ — 2a «o 2 = (11-57)

a\ = 100 (11-58)
which gives
aQ = 10 (11-59)

Substituting the results of a and a 2 into Eq. (11-57), we get

tf t = v 20fc /
(11-60)
Therefore,
[k 2 DD + NN} + = (a 2s
2
+as+a t )
(11-61)
= ks 2 + */20ks + 10
and
{k 2 DD + NN}~ = to - */Wcs + 2 10 (11-62)

The second term on the right side of Eq. (11-47) is

NR 10(0.5/.?)

.{k 2 DD + NN}-} + Iks 2 - */20ks + 10_

(11-63)
0.5

where we have considered that the pole at the origin is slightly over in the left-half
plane. Equation (11-47) now gives

- 2

ok ~ ks 2 + +/20ks + 10
~=
0.5
ks 2
0.5.$

+ */2Qk7+ 10
(11-64)

Substituting Uok into Eq. (11-39) for U, we have

s
0.5s 0.5s ,
^^=f rr«y (11-65)
i-L. ks o
2
+ >j20ks + \0ks 2 - */20ks + 10

This integral is evaluated by use of the formula in Table 11-1. From Table 1 -1,

/.= N D + N D 2
2 2
(11-66)
2D D D 2 1

where N = 0.5, N = 0, D =
t Q 10, Z>, = „/20£, D 2 = k. Thus
0.125
S= = 2.5 (11-67)
W20k
from which
k = 0.05
The optimal control is given by

tt s \ 0.5 s
UoKS> - 0.05s 2 + s + 10
(11-68)

and the optimal closed-loop transfer function is

M^=W)
'His)
Uo{s) =
G(s)
0.05s 2
+S+ 10
(11-69)
;
200
s 2
+ 20s + 200
 ,. ' :

Sec. 11.2 Analytical Design / 581

The following features of the design should be noted

1 The performance index is the integral square error with constraint on the
control signal in the form of integral square control. No specifications are
given in the form of overshoot, bandwidth, rise time, gain margin, and
so on.
2. The optimal control and the optimal closed-loop transfer function are
determined without specifying the final system configuration.

It is interesting to note that the optimal system represented by the transfer function

of Eq. (11-69) has a damping ratio of 0.707 and a bandwidth of 14.14 rad/sec.
There is a great deal of flexibility in arriving at the final system configuration and
the transfer function of the controller. Figure 11 -2(a) shows the system with a series
controller. The transfer function of the series controller is determined as

G = 1 M (s)
c (s)
G(s) 1 - AfoCs)
(11-70)
20s
s +20

As an alternative, Fig. 11 -2(b) shows the system with a feedback controller which is a
tachometer, and series controller, with

G c (s)
= 20
and
H(s) = 0.1s

lit) ~\ e(0 u(t) 10

f- 20s *- c(t)
•A
—
J .s + 20 s2

(a)

iiO ~\ e«) rA ,
"(O 10
+- c(t)

J +L
—
S^

0.1s

(b)

Fig. 11-2. (a) Analytically designed system in Example 11-1 with a series
controller, (b) Analytically designed system in Example 11-1 with a feed-
back controller.
 .

582 / Introduction to Optimal Control Chap. 11

Example 11-2 Referring to the system of Fig. 11-1, consider that

^)= £^i (11-71)

The input is a unit step function. It is desired to determine the optimal closed-loop
transfer function M 0) so that Je= minimum and /„ < 1

Using the notation of Eq. (11-44), N = s — 1, D = s 2 . Then, N = —s — 1 and

D = s2 .

{k 2 DD + NN} + = [k 2
s* - s2 + 1}
+
(11-72)

The spectral factorization of Eq. (11-72) is carried out by writing

k 2
s* - s2 + 1 = (a2 s + a^ + a )(a 2 s 2 - a s +
2
t
aQ )
(11-73)
= als* - (a\ - 2a 2 a )s + a 2 2

Equating like terms on both sides of Eq. (11-73) yields

a2 =k (11-74)

d! = */2k + 1 (11-75)

a = 1 (11-76)
Thus
{k 2 DD + NN}+ = ks 2 + «Jlk + 1,5 + 1 (11-77)
and
[k 2 DD + NN}- = ks ~ */2k 2
1,5+ 1 (11-78)

The second term on the right side of Eq. (11-47) is

NR _ -Or + 1)
=: (11 " 79)
\_{k
2
DD + NNj'X LsVcs - ^/W+ls + 2
1)J + ~T
where we have considered that the pole at the origin is slightly to the left of the imagi-
nary axis. The optimal control is now found from Eq. (1 1-47),

Uok = —s
(11-80)
ks 2 + V2A: + Is + 1

subject to the determination of k.

Using Eq. (11-39), and evaluating the integral with the help of Table 11-1, we have
k = 0.378
The optimal closed-loop transfer function is found by use of Eq. (1 1-48),

It is worth noting that the closed-loop transfer function contains the right-half-

plane zero of G(s). This is a necessary condition for u(t) to be bounded. In other words,
if M (s) does not contain the term (s - 1) in its numerator in this problem, it
would
not be possible to satisfy the Ju <\
requirement. However, the analytical design
formulation derived in this section naturally leads to the proper solution for this case.

It should be pointed out that the analytical design can yield a physically
realizable solution only if Je and Ju are finite. Given an arbitrary input and
process transfer function G(s), a minimum integral-square-error solution may
 : ,

Sec. 11.3 Parameter Optimization / 583

not exist. For instance, it is well known that the steady-state error of a type-0
system is nonzero for a step input. Thus J„ would be infinite if G(s) is of the
form

<** = (s + a)(s + t)(s + c)

a> b C ' *° (I1 " 82)

11 .3 Parameter Optimization

The analytical design method introduced in Section 1 1 .2 is restricted to single-

input systems only. For multivariable systems, the spectral factorization problem
causes the solution to be a great deal more complex. The advantage with the
analytical design is that the configuration of the system to be designed is semi-
fixed, so that the end result of the design is the optimal closed-loop transfer
function.
As a variation to the analytical design, we may consider a fixed-configura-
tion system but with free parameters that are to be optimized so that the fol-
lowing specifications are satisfied:

Je = C e \i)dt (11-83)
J

Ju = Cu\i)dt<K (11-84)
J

As in the analytical design, we consider the case when /„ == K, and the

problem becomes that of minimizing the following performance index

J =J + e k 2Ja (11-85)

2
where k is the Lagrange multiplier whose value is to be determined. The
system is optimized with respect to a given reference input.
The advantage with the parameter optimization method is that it can be
applied to systems with multiple inputs. The disadvantage is that the actual
implementation of the design is quite tedious for complex systems without the
use of a digital computer.
Consider that the control system to be designed has a fixed configuration
for the process and the controllers that contain n free parameters, u Kz K , . . .

K„. These parameters are to be determined so that J of Eq. (1 1-85) is minimized,

subject to the constraint
JU =K (11-86)

Applying Parseval's theorem and Table 11-1, the performance index / is

expressed as a function of the n free parameters and k 2 ; that is,

J = J(Ku K ,...,K ,k 2 n
2
) (11-87)

The necessary condition for / to be a minimum is

^= i=l, 2,..., ii (11-88)

 :

584 / Introduction to Optimal Control Chap. 11

and the sufficient condition is that the following Hessian matrix is positive
definite

d 2J d 2J d 2J
" '

3K dK2
x
' dK s
dK„
2
d J d J
2
d 2
J
'

V2/ dK2 6K i
6K\ ' dK 2 dK„ (11-89)

d 2J d 2J d 2J
_dKn dK, dKMi '
"
dKl .

Since d 2 JldK, dKj = d 2 J/dKj dK„ the Hessian matrix is always symmetric. In
addition to the necessary and sufficient conditions of Eqs. (11-88) and (11-89),
the equality constraint of Eq. (1 1-86) must be satisfied.
The following examples illustrate how the parameter optimization design
is carried out.

Example Consider the control system of Example 11-1. Let us use the cascade
11-3
controller configuration of Fig. 11-2, with the controller parameters
set as Ki and K
2 The block diagram of the system is shown in Fig.
.

11-3. The problem is to determine the optimal values of and K2 such that K x

minimum (11-90)
J o

J»= {" u 2 (t)dt = 2.5 (11-91)

J o

The reference input to the system is a step function with an amplitude of 0.5.

Kt)
.
+
r
v s
K lS
+ K2
u(t)
10
S2
c(t)

Gc (s) Gp (s)

Fig. 11-3. Control system for parameter optimization.

The problem is equivalent to that of minimizing

J =J +k
e
2
Ju = 2^7 f [E(s)E(-s) + k 2 U(s)U{-s)] ds (11-92)

From Fig. 11-3 we have

R(s) 0.5(s + K
E(S) =1 2)
(11-93)
:

+ G {s)Gp {s)
c j2 + K2 s + 10Ki
and
= = OSKiS
U(s) E(s)G c (s) (11-94)
s2 + K2 s + 10Ki

Substituting Eqs. (1 1-93) and (1 1-94) in Eq. (1 1-92), and using Parseval's theorem and
Table 11-1, we have
 Sec. 11.4 Design of System with Specific Eigenvalues / 585

j = 0.25(10*, + Kl) j_
k2
,v
2.5K1
(U "95)
lOKi^ •
lOK^z
Applying the necessary condition for J to be a minimum, we get from setting the
necessary conditions and the equality constraint,

dJ =
-0.25KJ + 5k 2 Kj (11-96)
dK x

dJ
-£L =o - - lOA: 2 =
Kl 10AT, *? (11-97)
dK2
Ju = 2.5 K\ = 20K2 (11-98)

The last three simultaneous equations give the solutions

Ki =20 K =
2 20 A:
2 = 0.05 (11-99)

Example can be shown that the

which are
Hessian matrix
The fact that
W
identical to the results
is

Ki equals K2
found

in this
in
positive definite, so that the sufficient condition
problem is
11-1. It

purely coincidental.
is

We
also satisfied.
could have
formulated a less flexible problem by assigning K =K x 2 . Then there would be only
one free parameter to optimize. The solution for such a situation would be quite
simple for the design criteria given in Eqs. (1 1-90) and (11-91), since the optimal value
of the parameter can be obtained by using only the constraint requirement of Eq.
(11-91).

11 .4 Design of System with Specific Eigenvalues— An Application

-
of Controllability 3 9

An interesting and useful application of state controllability is the design of a

linear feedback control system with specific eigenvalues. The system under
consideration may be represented by the block diagram of Fig. 11-4. The linear
process is described by the state equation

i = Ax + B« (11-100)

where the scalar control is

M =-Gx + r (11-101)

The matrix G is an 1 X n feedback matrix. Combining Eqs. (11-100) and

(11-101), the closed-loop system is represented by the state equation

i = (A - BG)x + Br (11-102)

x = Ax + Bu

Fig. 11-4. Linear system with state feedback.

 :

586 / Introduction to Optimal Control

Chap. 11

It can be shown that if [A, B] is a controllable pair, then a matrix exists G

that will give an arbitrary set of eigenvalues of (A BG). In other words, the —
roots of the characteristic equation

|AI-(A-BG)| = (11-103)
can be arbitrarily placed.
It has been shown in Section 4.7 that if a system is state controllable, it can

always be represented in the phase- variable canonical form; that is, in Eq.
(11-100),
1 ~ ~0
1

A= B =
1

—a n -«*-! — tf»-2 • . . —a. 1

The reverse is also true in that if the system is represented in the phase-
variable canonical form, it is always state controllable. To show this, we form
the following matrices

"

AB = AB= 2
AB=
3

— a.
_a\ - «2_
- a\ + a2 a x — a 3j
-
Continuing with the matrix product through A" 'B, it will become apparent that
regardless of what the values of a,, a 2 a„, are, the determinant of S =
, . . . ,

[B AB A 2 B A" _1 B] will always be equal to — 1, since S is a triangular

. . .

matrix with Is on the main diagonal. Therefore, we have proved that if the
system is represented by the phase-variable canonical form, it is always state
controllable.
The feedback matrix G can be written
G = fci g 2 gn] (11-104)
Then
1

1 .. .

..
BG .

(11-105)

gi —a„-i — g2 gn.
Sec. 11.4 Design of System with Specific Eigenvalues / 587

The eigenvalues of A — BG are then found from the characteristic equation

|AI - (A - BG)| = A" + (a, + gJA"" + (a + g^M"' +...+(«„ + Si)
1
2
2

= (11-106)

Clearly, the eigenvalues can be arbitrarily chosen by the choice of g u g 2 , . .

. , g„.

Example 11-4 Consider that a linear process is described by the transfer function

£(£)_. 10
U(s)
~ s(s + l)(s + 2)
(11-107)

It is desired to design a feedback controller with state feedback so that the eigenvalues
of the closed-loop system are at —2,-1 +;'l,and —1 —jl. Since the transfer function
does not have common poles and zeros, the process is completely state controllable.
By direct decomposition, the state equations for Eq. (11-107) are written

"o i °T*ii r°~

x2 1 x2 + u (11-108)

L*3. _0 -2 -3J|_* 3 J |_1_

Since this is a third-order system, the feedback matrix G is of the form
G=te, g 2 g 3] (11-109)

The characteristic equation of the closed-loop system is

A3 + (3 +g 3 )k* + (2 +g 2 )X +g,=0 (11-110)

and with the prescribed eigenvalues it is

A 3
+4A 2 +6A +4 = (11-111)

Equating the like coefficients of the last two equations, we have

g\ = 4 g2 =4 gi = 1

or
G= [4 4 1] (11-112)

The state diagram of the overall system is shown in Fig. 11-5.

r O O c

Fig. 11-5. Linear control system with state feedback for Example 11-4.

Stabilizability*

We have shown that if a system is completely controllable, its eigenvalues

can be arbitrarily located by complete state feedback. This also implies that any
unstable system can be stabilized by complete state feedback if all the states are
588 / Introduction to Optimal Control Chap. 11

controllable. On the other hand, if the system is not completely controllable, as

long as the uncontrollable states are stable, the entire system is still stabilizable.

11.5 Design of State Observers'

Another area of design of control systems concerned with the concepts of

controllability and observability is the design of state observers. Suppose that in
the design of the state feedback controller discussed in Section 11.4, we deter-
mined the feedback matrix so that G
u(t) = -Gx(r) + r(t) (11-113)

However, in order to implement the state feedback, we still need to feed

back all the state variables. Unfortunately, in practice, not all the state variables
are accessible, and we can assume that only the outputs and the inputs are
measurable. The subsystem that performs the observation of the state variables
based on information received from the measurements of the input u{t) and the
output c(t) is called an observer.
Figure 11-6 shows the overall system structure including the observer. The
scalar control is ordinarily given by

u{t) = Er(t) - Gx(t) (11-114)

where E and G are matrices of appropriate dimensions. Figure 1 1-6shows that

the inputs to the observer are the output c{t) and the control u. The output of
the observer is the observed state vector x e (r). Therefore, the actual control is

u{t) = Er{t) - Gx,(r) (11-115)

We must establish the condition under which an observer exists. The

following theorem indicates that the design of the state observer is closely
related to the condition of observability.

^X U X
D

\e State-
G observer ,

Fig. 11-6. Linear system with state feedback and state observer.

Theorem 11-1. Given an nth-order linear time-invariant system that is

described by the dynamic equations,

±(t) = Ax(?) + Bu(t) (11-116)

c(0 = Dx(/) (11-117)

where x is an n-vector, u is the scalar control, and c is ap-vector. The state vector x
Sec. 11.5 Design of State Observers / 589

may be constructed from linear combinations of the output c, input u, and deriva-
tives of these variables if the system is completely observable.

Proof: Assuming that the system is completely observable, the matrix

T= [D' A'D' (A') 2 D' . . . (A')"

_1
D'] (11-118)

is of rank n. We shall show that the states can be expressed as linear combina-
tions of the output c(f), the input u{t), and their derivatives. Starting with Eq.
(1 1-1 17), we take the derivative on both sides with respect to /. We have
c(0 = Dx(0 = DAxO) + DBw(f) (11-119)
or
t(t) - DBm(0 = DAx(f) (11-120)

Repeating the time derivative process to Eq. (11-120), we get

c(0 - DBti(0 = DAx(0 = DA 2

x(0 + DAB M (?) (11-121)

Rearranging the last equation yields

c(0 - DBii(r) - DABm(0 = DA 2

x(/) (1 1-122)

Continuing the process, we have, after taking n derivatives,

c<">(/) = DBm'"- 1
^) - DABm<"- 2 >(0- ... - DA-^BmO) = DA"-'x(0 (11-123)

In matrix form, Eqs. (11-117) and (11-120) through (11-123) become

c D
c — DBh DA
c - DBri DABw DA 2

x (11-124)

DB^" 1
' - DABm<"- 2 »
DAnl B« DA"
Therefore, in order to express the state vector x in terms of c, u, and their
derivatives, the matrix

T= [D DA DA 2
. . . DA" 1 ]'

= [D' A'D' (A') 2

D' . . . (A')"-'D']

must be of rank n.

There are many ways of designing a state observer, and there is more than
one way of judging the closeness of x e (0 to x(f). Intuitively, the observer should
have the same state equations as the original system. However, the observer
should have u(t) and c(r) as inputs and should have the capability of mini-
mizing the error between x(t) and x c (f). Since we cannot measure x(/) directly,
an alternative is to compare c(t) and c e (t), where
c c (0 = Dx„(r) (H-125)

Based on the above arguments, a logical arrangement for the state observer
is shown in Fig. 11-7. The state observer design is formulated as a feedback
590 / Introduction to Optimal Control Chap. 11

"^ *e Xe
f
\+

Ge

Fig. 11-7. Block diagram of an observer.

control problem with G e as the feedback matrix. The design objective is to select
the feedback matrix G„ such that c,(0 will approach c(f) as fast as possible.

When c e (0 equals c(/), the dynamics of the state observer are described by

xjj) = Ax (?) +e Bu(t) (11-1 26)

which is identical to the state equation of the system to be observed. In general,

with u and c(f) as inputs to the oberver, the dynamics of the observer are rep-
resented by
±.(0 = (A - G.D)x.(0 + *«(0 + G.c(0 (1 1-127)

The block diagram of the overall system with the observer is shown in Fig. 1 1-8.

Since c(t) and x(f) are related through Eq. (11-117), the last equation is also
written as
±.(0 = Ax.(0 + Bu(t) + G D[x(0 - e x,(01 (1 1-128)

The significance of this expression is that if the initial values of x(t) and xe (t) are
identical, the equation reverts to that of Eq. (11-126), and the response of the
observer will be identical to that of the original system. Therefore, the design of
the feedback matrix G e for the observer is significant only if the initial condi-
tions to x(t) and x„(t) are different.
There are many possible ways of carrying out the design of the feedback
matrix G e. One way is to utilize the eigenvalue assignment method discussed in
Section 11.4. If we subtract Eq. (11-128) from Eq. (11-116), we have

[x(f) - ±,(0] = (A - G D)[x(/) -e x.(0] (1 1-129)

which may be regarded as the homogeneous state equation of a linear system

with coefficient matrix A — G C D. The characteristic equation of A — G e D and
of the state observer is then

|AI-(A-G D)| = e (11-130)

Since we are interested in driving x e (t) as close to x(t) as possible, the objective
of the observer design be stated as to select the elements of G e so that the
may
natural response of Eq. (11-129) decays to zero as quickly as possible. In other
 Sec. 11.5 Design of State Observers / 591

» X)
Observer

State feedback

System

-^OH B
HO
Fig. 11-8. Block diagram of a linear feedback system with observer.

words, the eigenvalues of A—GD C should be selected so that x e (t) approaches

x(t) rapidly. However, it must be kept in mind that the approach of assigning
the eigenvalues of A — G^D may not always be satisfactory for the purpose
of matching all the observed states to the real states, since the eigenvalues
control only the denominator polynomial of the transfer relation, while the
numerator polynomial is not controlled. In Section 11.4 the condition for the
arbitrary assignment of the eigenvalues of (A — BG) given A and B has been
established. We must first condition Eq. (11-129) so that the similarity between
the two cases may be established. It is easy to see that the eigenvalues of
(A — G D) are identical to those of (A — G D)'. The latter matrix is written
e e

(A - G,D) = (A - D'C;) (11-131)

Thus, invoking the results of Section 1 1.4, the condition of arbitrary assignment
of the eigenvalues of (A -
G„D) or of (A' -
D'G^) is that the pair [A', D'] be
completely controllable. This is equivalent to requiring that the pair [A, D] be
completely observable. Therefore, the observability of [A, D] ensures not only that
a state observer can be constructedfrom linear combinations of the output c, input
 :

592 / Introduction to Optimal Control Chap. 11

u, and the derivatives of these, but also that the eigenvalues of the observer can
be arbitrarily assigned by choosing the feedback matrix G„.
At this point an illustrative example may be deemed necessary to show
the procedure of the observer design.

Example 11-5 Consider that a linear system is described by the following transfer
function and dynamic equations

= (11-132)
W) (s + IX* + 2)

i(0 = Ax(0 + Bu(t) (11-133)

c(0 = Dx(0 (11-134)

where
o r
B
.-2 -3_
D= [2 0]

It is desired to design a state observer so that the eigenvalues of Al |

— (A — G D) =
e |

are at A = -10, -10.

Let the feedback matrix be designated as

gel
G = e (11-135)
LSel.

O xi O*io

Fig. 11-9. State diagram of the observer and the original system for Exam-
ple 11-5.
 Sec. 11.5 Design of State Observers / 593

Then, the characteristic equation of the state observer is

|AI-(A-GJ»!= + 2g " ~ X l

+ 2g e2 A + 3
2 (11-136)

= A 2 + (2g + 3)A + (6*„ + 2 + 2g. ) =

el 2

For the eigenvalues to be at A = —10, —10, the characteristic equation should be

A2 +20A + 100 = (11-137)

Equating like terms of Eqs. (11-136) and (11-137) gives

gel = 8.5

gel = 23.5

The state diagram for the observer together with the original system is shown in Fig.
11-9.
As was mentioned earlier, if the initial values of x(0 and x«(?) are identical, the

responses of the observer will be identical to those of the original system, and the
feedback matrix G„ will have no effect on the responses of the observer whatsoever.
Figure 11-10 illustrates the unit step responses of Xi(t) and xe i(t) for the following

1.2. 1 1 1 1 \ i
1

i n
gel = 8.5 xl
gel = 23.5

0.8 -

-
0.6 *el(t) («el=ge2=D

0.4 -

0.2
-

i
' 1 1 1 1 1
i

i 2 3 4 5 6 7

Time (seconds)

Fig. 11-10. State x\(f) and the observed states of the observer in Example
11-5.
594 / Introduction to Optimal Control
Chap. 11

initial states:
"0.5
x(0) = *.(0) =

Shown in the same figure is the response of x el (t) when the state observer is
designed for eigenvalues at A = -2.5, -2.5, in which caseg-,1 = 1 and^ 2 1. Since =
the state observer dynamics are now slower, it is seen from Fig. 11-10 that the x (t)
ei
response deviates more from the state x t
(t).

Figure 11-11 responses x 2 (t) and x e2 (t) for the two cases of observer
illustrates the
design. The reason that both x e2 (t) responses are not very close to the actual state
x 2 (t) due to the fact that the elements of the feedback matrix, g el and g e2 appear in
is ,

the numerator of the transfer function relation between X2 (s) — X {s) and x (0)
e2 2
- x e2 (0), but not in the relation between (s) — eX (s) and x (0) - jc€I (0). Taking
X t X t

the Laplace transform on both sides of Eq. (11-129) and rearranging, we have

X(s) - X«(J) = [si - (A -G e D)]-'[x(0) - Xc (0)] (11-138)

where
s - 3
-2-
[si - (A - G D)]"
e
2ge2 2gelJ (11-139)
|a-(A-G.D)|

2 3 4 5 6 7 8

Time (seconds)

Fig. 11-11. State x 2 (t) and the observed states of the observer in Example
11-5.

Thus, we notice from Eq. (11-139) that selecting larger values for g el and ge2 to
give faster transient response for the observer will not affect the numerator terms of the
X e i(s) response. However, a larger value of g e2 will increase the gain factor between

X 2 (s) - X
e2 (s) and x 2 (0) - x e2 (0). This explains the fact that in Fig. 11-11 the x e2 (t)
response for ge2 = 1 is a better observation of x 2 {t) than the one for g,, 2 =23.5.
Sec. 11.5 Design of State Observers / 595

A more versatile design method for the observer may be effected by the use
of a performance index on the error between x and x e A popular performance .

index is the quadratic form,

J =± \" {x - x.)'Q(x - x.) dt (11-140)

J

The matrix Q is symmetric and positive semidefinite. The elements of Q

may be selected to give various weightings on the observation of the states.

Closed-Loop Control with Observer

We are now ready to investigate the use of an observer in a system with

state-variable feedback. The block diagram of such a system is shown in Fig.
11-8.

From Fig. 1 1-8 the state equations of the entire system are written

= Ax - BGx + BEr
x e (11-141)

i. = (A - DG - BG)x„ +e DG,x + BEr (11-142)

A statediagram for the system is drawn as shown in Fig. 11-12 using Eqs.
(11-141) and (11-142). The initial states are included on the state diagram, and
the branch gains are matrix quantities. It can be easily shown that the system
with the observer is uncontrollable (state) but observable. However, the lack of
controllability is in all the states x and x e The reason for this is simple; the main
.

objective is to control x while keeping x e close to x. It is not necessary to control

x and x e independently.

OXe (fJ) = X e0

Fig. 11-12. State diagram of the system shown in Fig. 11-8.

596 / Introduction to Optimal Control Chap. 11

From Fig. 11-12 the transfer relations of the state vector and the observed
state vector are derived. These are
X(s) = [si - (A - BG)]-'BE/?(s) - BGA-'x, + [si - (A - G D - BG)]A-'x e

(11-143)

X e (s)
= [si - (A - BG)]- BE/?(i)
]
+ (si - A)A-'x e0 + G.DA-'x,, (11-144)

where
A= - (2A - G D - BG)j + [G DBG + A(A - G D - BG)]
5 2I e e e (1 1-145)

When x„ = xe0 it can be shown that ,

X(j) = X,(j) = [si - (A - BG)]-«BEi!(j) + [si - (A - BG)]-% (1 1-146)

Therefore, when the observer and the system have the same initial conditions, the
two systems will have identical state and output responses. More important is the
fact that the transfer relation in Eq. (11-146) is identical to that when the observer
is absent; that is, the true state variables are fed back for control purposes. This
means that when the of the original system and the observer are
initial states

identical, the responses of the system will not be effected by the feedback
matrix G e.
Since, in general, x, ^ x„, the design of the observer, and subse-
quently of G e, will affect the system response through the initial-condition term
due to x , as indicated by Eq. (1 1-143). However, the steady-state response is due
to the first term on the right side of Eq. (1 1-143), and the eigenvalues of A — BG
are not functions of G,.

Example 11-6 As an example of the design of a closed-loop system with observed

state feedback, consider the linear process that has the transfer
function
C(s) _ 100
(11-147)
U(s) s(s + 5)

The dynamic equations of the process are written

±(/) = Ax(/) + Bu(t) (11-148)

c(t) = Dx(r) (11-149)

where
"0 r " "
A= B=
Lo -5_ LlOOJ
D= [l 0]

It is desired to design a state feedback control that is given by

u = r-Gx (11-150)
where
G=fr, g 2 ] (11-151)

The eigenvalues of the closed-loop system should be located at; A = —7.07 ±y'7.07.
The corresponding damping ratio is 0.707, and the natural undamped frequency is
10 rad/sec.
Assuming that the states x and x 2 are
t inaccessible so that a state observer is to
be designed, the overall system block diagram is identical to that of Fig. 11-8, with
£=1.
 Sec. 11.5 Design of State Observers / 597

First, the pair [A, B] is completely controllable, so the eigenvalues of A BG —

can be arbitrarily assigned. Also, it is simple to show that the pair [A, D] is completely

observable, so that an observer may be constructed from c and u.

The characteristic equation of the closed-loop system with state feedback is

|Al-(A-BG)| =0 (11-152)
or
A2 + (5 + lOOg^A + KXte, = (11-153)

To realize the desired eigenvalues stated earlier, the coefficients of Eq. (11-153) must
satisfy the following conditions

5 + =
100^ 2 14.14

K%! = 100

Thus gx 1 and g 2 = 0.0914.

=
The state diagram of the overall system, which includes the observer, is shown in

Fig. 11-13.

rO

Fig. 11-13. State diagram of the system with observer in Example 11-6.

The design of the observer may be carried out in the usual manner. The charac-
teristic equation of the observer is

|AI-(A-G.D)] =0 (11-154)
where
~ge\
G =
e (11-155)
i-gel-
598 / Introduction to Optimal Control Chap. 11

Let us assume that the eigenvalues of the observer must be at A = —50, —50.
These eigenvalues are of much greater magnitude than the real parts of the eigenvalues
of the system. Therefore, the transient of the observer due to the difference between
the initial states x and x c0 should decay very rapidly to zero. The corresponding
values of gei and g e2 are found to be

g.i = 95
(11-156)
gel = 2025
The system of Fig. 11-13 is simulated on a digital computer with G and G„ as
indicated above.The responses of x (t) when r{t) is a unit step function are computed
t

and plotted as shown in Fig. 11-14 for x c0 = x = 0, and x = 0, x e = [0.5 0]'.

0.4 0.5 0.6 0.7 0.8 0.9

Time (second)

Fig. 11-14. Unit step responses of xi(t) and x el (t) of the system in Example
11-6.

When x e o = x , that with the real state-variable feedback. When x

the response is

^x e0 , and the elements of G

are as given in Eq. (11-156), the step response of *i(f)
e

is faster than the ideal Xi(t), and the overshoot is greater. Also shown in Fig. 11-14
is
the step response of x^t) when the observer is designed to have its eigenvalues at
X = — 10 and —10, which correspond tog ei =15 andg e2 = 25. The initial states of
the observer are again given by x e0 = [0.5 0]'. Note that since the observer dynamics
are much slower in this case, the corresponding step response of x^t) is also slower
 Sec. 11.6 Optimal Linear Regulator Design / 599

than the true response, but the overshoot is still greater. The x„i(t) responses are shown
to follow their corresponding x\(t) responses very closely, starting from the initial

state of *e i(0) =0.5.

It is interesting to investigate the effect of having various differences in the initial
values between Xi(t) practice, the initial value of Xi(t) (as well as that of
and x ei (t). In
the other state variables) not fixed and could be of any value. The initial values of the
is

observed states must be set so that they are as close to that of the real states as possible.
For illustrative purposes, we set x(0) = x = and x e2 (0) = but vary x el (0). Figure
11-15 shows the unit step responses of xi(t) when x e i(0) = 0, 0.25, 0.5, and0.8, with
G =
e [15 25]'.The case with ^(O) = *i(0) = is the ideal response. These curves
illustrate the importance of making x<, as close to x as possible.

1.2

x el =0.25

*10 ~*el0 ~

0.4 0.5 0.6 0.7 0.8 0.9

Time (second)

Fig. 11-15. Unit step responses of x\{t) of the system in Example 11-6 with
various values of x e \<y.

11.6 Optimal Linear Regulator Design 1

One of the modern optimal control design methods that has found practical
applications is linear regulator design theory. The design is based on the mini-
mization of a quadratic performance index, and the result is the optimal control
through state feedback. The basic advantages of optimal linear regulator design
 : 1

600 / Introduction to Optimal Control Chap. 1

are as follows

1. The solution is carried out without cut-and-try; a mathematical

algorithm leads to a unique solution.
2. Can be used to design linear time-varying or time-invariant systems
with multivariables.
3. Guarantees stability.

Consider that an nth-order linear system is described by the state equation

±(t ) = AxO) + Bu(» (1 1-1 57)

where x(r) is the n X 1 state vector and u(r) is the r X 1 control vector; A and B
are coefficient matrices of appropriate dimensions. In general, A and B may
contain time-varying elements. The design objective is that given the initial
state x(r ), determine the optimal control u°(r) so that the performance index

= " [x'(7)Qx(0
/ £ x'C^Px^) + £ f
J H
+ u'(?)Ru(0] dt (11-1 58)

is minimized. The performance index J is known as a quadratic performance

index, since P, Q, and R are required to be symmetric matrices. In addition, it
is required that P and Q be positive semidefinite and R be positive definite.

The first term on the right side of Eq. (1 1-158) is the terminal cost, which is
a constraint of the final state, \(t f ). The inclusion of the terminal cost in the
performance index is desirable since in the finite-time problem, t f is finite and
x(t f ) may not be specified.
The first term under the integral sign of / represents a constraint on the
state variables. The simplest form of Q that one can use is a diagonal matrix,
unless no constraint is placed on the states, in which case Q = 0. For an nth-
order system, a diagonal Q may be defined as
"?, ... 0"

q1 ...

Q= ?j ... (11-159)

.0 ... qn _
The rth entry of Q represents the amount of weight
the designer places on
the constraint on the state variable x (t). The larger the value of q„ relative to
t

the other qs, the more limiting is placed on the state x,(0- The second term under
the integral of Eq. (1 1-158) places a constraint on the control vector u(f). There-
fore, the weighting matrix R has the same significance for u(f ) as Q has for x(f).
Physically, the time integral of u'(r )R U (0 has the dimension of energy. Thus the
minimization of the performance index of Eq. (11-158) has the physical inter-
pretation of keeping the states near the equilibrium state x = 0, while holding
the energy consumption to a reasonable level. For this design, the reference
input is not considered and is set to zero, although the designed system may

still be subject to nonzero inputs. The term "regulator" simply refers to the
Sec. 11.6 Optimal Linear Regulator Design / 601

condition that the primary purpose of the design is that the system is capable of
damping out any initial state x(f ) quickly without excessive overshoot and
oscillations in the state responses.
The derivation of the necessary condition of the optimal control may be
carried out in a number of ways. In this section we shall use the principle of
optimality 16 and the Hamilton- Jacobi equation as development tools.

Principle of Optimality

Many algorithms and solutions in optimal control theory follow basic

laws of nature that do not need complex mathematical development to explain
their validity. One of these algorithms is the principle of optimality An optimal :

control strategy has the property that whatever the initial state and the control
law of the initial stages, the remaining control must form an optimal control with
respect to the state resulting from the control of the initial stages.
Figure 11-16 gives a simple explanation of the principle in the form of a
state trajectory of a second-order system. Let us consider that the control
objective is to drive the system from x(t ) to x(t f ), with x(f ) given, in such a way
that the performance index

= fV(x,u) dt (11-160)

is minimized. The state equations of the system may be that given in Eq. (1 1-157).

Let t x
be some intermediate time between t and t
f (t < f, < t ),
f and x^) is

the corresponding state, as illustrated in Fig. 11-16. We can break up the

performance index in Eq. (11-160) so that for the time interval tx < < t t
f , it is
expressed as

/, = f"F(x,u)A (11-161)

Xx(? )

-»- *1

Fig. 11-16. Principle of optimality.

 :

602 / Introduction to Optimal Control Chap. 11

The principle of optimality states that regardless of how the state reaches
x0,) at t = t u once x^) is known, then using x(/,) as the initial state for the
last portion of the state transition, in order for the entire control process to be
optimal in the sense of minimum /, the control for the time interval t x
<t <tf
must be so chosen that J a minimum. In other words, the
x
is last section of an
optimal trajectory is also an optimal trajectory.
Using another more readily understandable example, we may consider that
a track star desires to run a 440-yard distance in the shortest possible time. If he
plans his strategy based on dividing the total distance into four parts, the
optimal strategy for this minimum-time control problem, according to the
principle of optimality, is
Nomatter what the control laws of the first n parts, n 1, 2, 3, are, the =
remaining 4 —
n parts must be run in the shortest possible time, based on the state
of the physical condition at the time of the decision.

Hamilton-Jacobi Equation

In this section we shall use the principle of optimality to derive the

Hamilton-Jacobi equation, which is then used to develop the linear regulator
design.
To show that the theory is generally applicable to time-varying and non-
linear systems, we shall use the following state equation representation

x(/) = f(x,u,?) (11-162)

where x(t) is an n-vector and u(() is an r-vector.

The problem is to determine the control u(?) so that the performance index

J = f"F(x,u,T)A (11-163)
** to

is a minimum.
Let us denote the performance index over the time interval t <x< t
f by

/=JV(x,u,T)rfr (11-164)

and its minimum value by

S[x(t), t] = Min / = Min \" F(x, u, x) dx (1 1-165)

u u J t

Then
S[x(t ), t ] = Min J = Min
J
\" F(x, u, x) dx (1 1-166)
u n to

When ;

S[x(t f ), t f ] = Min f " F(x, u, t) dx = (11-1 67)

u J tr

Now let us break up the time interval from t to t f into two intervals: t <x<t
+A and t +A<x< t
f . Then Eq. (11-165) is written

A
S[x(t), t] = Min Q" F(x, u, t) dx + j" F(x, u, x) dx) (1 1-168)
 : : .

Sec. 11.6 Optimal Linear Regulator Design / 603

Applying the principle of optimality we have

j" +A
S[x(t), t] = Min F(x, u, t) dx + 5[x(f + A), + t A]) (1 1-169)
(

where, according to Eq. (11-165),

S[x(t + A), t + A] = Min

D J
\"
f +A
F(x, u, t) dx (1 1-170)

Let A be a very small time interval; then Eq. (11-168) can be written

S[x(t), t] = Min (F(x, u, 0A + S[x(t + A),t + A]) +e (A) (1 1-171)

o

where e (A) is the error function.

The next step involves the expansion of S[x(t + A), f + A] into a Taylor
series. This would require the expansion of x(t + A) into a Taylor series about

A = first. Thus
x(t + A) = x(t) + ^A + . . (11-172)

Thus Eq. (11-171) is written

S[x(t), t] = Min(F(x, u, t)\ + S[x(t) + ±(t)A + . . . , t + A]) + e„(A) (1 1-173)

u

The second term inside the minimization operation in the last equation is of the
form S[x(t) + h(j), t + A], which can be expanded into a Taylor series as
follows

S[x(t) + h(r), + t A] = S[x{t), t] + ^!k(() + *W>.f]

ox(.t) at
A
(11-174)
+ higher-order terms
Therefore,

S[x(t + A), t + A] = S[x(t) + f (x, u, f)A + . . . , f + A]

= S ^> *> + ti
d
-T§rf x < - ( > u
' ** + HM A
dJ}

+ higher-order terms (11-175)

where we have replaced x by f(x, u, t)

Substituting Eq. (11-175) into Eq. (11-173), we have

S[x(t), t] = Min(F(x,
V
u, r)A + S[x(t), + £ dSdm
}:*®' f (x, u, t)\
t]
'- 1
t]
t

'
(11-176)
+ ^^A) + fl (A)

where e^A) is the error function, which includes the higher-order terms in Eq.
(11-175) and e (A).
Since S[x(0, t] is not a function of the control u{t), Eq. (11-176) is sim-
plified to the following form after cancellation and dropping the common
factor A
- dS[ t]
= Min (F(x, ii, /) + £ ^gg^TXx, «, /)) + e 2 (A) (1 1-177)
f^'
 1

604 / Introduction to Optimal Control Chap. 1

As A approaches zero, Eq. (11-177) becomes

which is
-
known
^P = Min F(x, u,
as Bellman's equation.
(
+ £ S^/X*, -, 0) (1 1-178)

The necessary condition for the last equation to be a minimum is determined

by taking the partial derivative on both sides of the equation with respect to u,
and setting the result to zero. We have
<^(x,u,0 dSMt),t]dftx,v,t) =
f. .

(?u(f) ^h dx,(t) du(t)

ft (11-179)

or in matrix equation form,

dF(x,u,t)
"•"
dt(x, u, Q <7S[x(t), t] __ n (11-180)
du(t) du(t) dx(t)

where <?f(x, u, t)/du(t) denotes the matrix transpose of the Jacobian of f(x, n, t);

that is.

'Mi Mi Ml'
du t dui du t

df (x, u, Q
M
du 2
Mi
du 2
Mn
du 2 = (Jacobian of f)' (11-181)
du(t)

Mi Mi Ml
du r du T du r _
Once the optimal control u°(t) is determined or known, Eq. (11-178) becomes

dS[x °®' t]
+ Fix", u°, + ± dS t)
t]
m°, u°, = (1 1-182)
}£$jJj
where all the variables with superscript zeros indicate optimal quantities. The
sum of the two terms in the last equation
last is also known as the Hamiltonian
H(x°, u°, t). Thus Eq. (11-182) is written

dS[x\t), t]
+ H(x o^
u
o
f) = (11-183)

which is known as the Hamilton-Jacobi equation and is the necessary condition

for the control to be optimal in the sense of minimum /.

Derivation of the Differential Riccati Equation (Finite-Time Problem)

Now we are ready to use the Hamilton-Jacobi equation to derive the solu-
tion of the linear regulator problem. Consider the system

x(t) = Ax(r) + Bu(?) (11-184)

with the initial condition, x(f„) = x . The design problem is to find the optimal
control u°(t) such that the performance index

J = $x'(tf )Px(tf + £ )
\" [x'(t)Qx(t) + u'(f)Ru(0] dt (1 1-185)
 :

Sec. 11.6 Optimal Linear Regulator Design / 605

is a minimum. The matrices P and Q are symmetric and positive semidefinite,

and R is symmetric and positive definite.

Let the minimum of/ be represented by 5[x(?), t]. Then, from the develop-
ment of the last section we know that the necessary condition of optimal

control is that the Hamilton- Jacobi equation must be satisfied. For the present
case, Eq. (11-178) is written

_ dS[x(t),t] == M in ^_ x (/)Qx(0 + ^_u'«Ru(/) + [Ax(r) + Bu(0r^|^)

<

(11-186)

To carry out the minimization process, we differentiate both sides of Eq.

(11-186) with respect to u(f), and we have the optimal control after setting the
result to zero

u°(0 = -R-'B' dy//^ (11-187)

ox (t)

It remains to solve for dS[x°{i), t]/dx°(t) in order to complete the expression of

the optimal control. Substituting Eq. (11-187) into Eq. (11-186), we have
dS[x°(t),t] 1 _o
Yrto
_o,
rt
, 1
(
dS[x°(t), ?] V BR -, B >
(
dS[x°(t), t] \

, o
YfYA
, dS[x°(t), t] ( dS[x°(t), t] Y BR _ 1R ,
(
dS[x°(t), t] \

(11-188)
The last equation is simplified to

dS[x°(t),t] 1 _o,
rrt0x
o,
rt _ 1
(
dS[x°(t),tT\
'

m -i W dS[x°(t),t]
v
(11-189)
+ X WA dx°(t)

Since Q is a symmetric matrix, it can be shown that S[x°(t ), t] is a quadratic

function of x°(/). Thus we let

S[x°(0, t] = ±x°'(t)K(t)x°(t) (1 1-190)

where K(/) an n x n symmetric matrix.

is

The optimal control is determined in terms of K(t) by substituting Eq.

(11-190) in Eq. (11-187). We have

u°(0 = -R-'B'K(0x°(/) (11-191)

Substitution of Eq. (11-190) in Eq. (11-189) yields

—j x°'(t)^£x°(t) = i-x°'(0Qx°(0 - ^-x°'(r)K(0BR- B'K(0x (0 |

o

+ x°'(f)A'K(f)x o
(0 (11-192)

The last term in Eq. (11-192) can be written

x°'(0A'K(0x°(0 = £x°'(0[A'K(0 + K(0A]x°(0 (1 1-193)

606 / Introduction to Optimal Control Chap. 11

since both sides of the equation represent scalar quantities. Thus, using Eq.
(11-193) and comparing both sides of Eq. (11-192), we have
dKjt) - K(0A - A'K(f) - Q
= K(0BR-«B'K(0 (11-194)
dt

which is known The matrix K(t) is sometimes called the

as the Riccati equation.
Riccati gain matrix. The boundary condition needed to solve the Riccati
equation is obtained from Eq. (11-190) by setting / = t f Then .

K(t f ) =P (11-195)

Another important property of K(t) is that it is positive definite, since otherwise,

S[x°(t), t] would be negative.
In summarizing the solution of the optimal linear regulator problem, the
optimal control is given by Eq. (1 1-191) in the form of state feedback; that is,

u°(r) = -G(r)x°(0 (11-196)

where
G(0 = R-'B'K(0 (11-197)

is the time-varying feedback gain matrix. The Riccati gain matrix K(t) is solved
from the Riccati equation of Eq. (11-194), with the boundary condition of
K(? r ) = P. The state equation of the optimal closed-loop system is given by

x(0 = [A - BR-»B'K(r)]x(0 (1 1-198)

The block diagram of the feedback system is shown in Fig. 11-17.

^
J
u
x = Ax + Bu
X

R-'B'K(f)

Fig. 11-17. Implementation of the optimal linear regulator design.

A few remarks may be made concerning the solution of the linear regulator
problem. It should be noted that the present consideration has a finite terminal
time t and nothing is required on the stability, the controllability, or the
f ,

observability of the system of Eq. (11-184). Indeed, the finite-time linear

regulator problem does not require that the process be observable, controllable,
or stable.
The Riccati equation of Eq. (11-194) is a set of n 2 nonlinear differential
equations. However, there are only n(n — 1) unknowns in K(t), since the
matrix is symmetric.
Sec. 11.6 Optimal Linear Regulator Design / 607

We shall postpone the discussion on the solution of the differential Riccati

equation, Eq. (11-194), until we have investigated the infinite-time problem;
that is, t
f
= oo.

Infinite-Time Linear Regulator Problem

Now let us consider the infinite-time regulator problem for which the
performance index becomes

j = $ P [x'(0Qx(0 + u'(/)Ru(01 dt (1 1-199)

Notice that the terminal cost is eliminated from /, since as t approaches

infinity, the final state x(^) should approach the equilibrium state 0, so that the
terminal constraint is no longer necessary. The same conditions of positive
semidefiniteness and positive definiteness are placed on Q and R, respectively,
and all the derivations of the last section are still valid. However, the infinite-
time linear regulator problem imposes additional conditions on the system of
Eq. (11-184):

1. The pair [A, B] must be completely controllable.

2. The pair [A, D] must be completely observable, where D is any
n x n matrix such that DD' = Q.

The solution of the linear regulator problem does not require that the
process is stable; in the finite-time case, controllability is not a problem, since
for finite t
f, J can still be finite even if an uncontrollable state is unstable. On
the other hand, an unstable state that is uncontrollable will yield an unbounded
performance index if t f is infinite. The observability condition is required
because the feedback system of Eq. (11-198) must be asymptotically stable. The
requirement is again tied in with the controllability of the system in ensuring
that the state vector x(t) approaches the equilibrium state as t approaches
infinity, since observability of [A, D] is equivalent to the controllability of
[A', D']. A complete proof of these requirements may be found in the litera-
"
ture. 13 16
For the infinite-time case, the Riccati equation in Eq. (11-194) becomes a
set of nonlinear algebraic equations,

KBR^B'K - KA - A'K - Q = (1 1-200)

since as t f approaches infinity, dK(t)/dt =

0, the Riccati gain matrix becomes a
constant matrix. The solution of the Riccati equation, K, is symmetric and
positive definite.

Solution of the Differential Riccati Equation

A great deal of work has been done on the solution of the Riccati equation,
both for the differential form and for the algebraic form. In general, the alge-
braic Riccati equation of Eq. (11-200) is more difficult to solve than the differ-
ential equation of Eq. (11-194).
608 / Introduction to Optimal Control Chap. 11

The differential Riccati equation is generally solved by one of the following

methods:

1. Direct numerical integration.

2. Matrix iterative solution.
3. Negative exponential solution

The direct numerical integration method simply involves the solution of the
nonlinear differential equations backward in time on a digital computer,
knowing K(^). The matrix iterative method is effected by writing dK(t)/dt as

lim
K(f + A) - K(0
(11-201)
dt

Equation (11-194) can be approximated by a set of difference equations which

are then solved iteratively backward in time.

Negative Exponential Solution

Let us define the vector dS(x, t)/dx(t) as the costate vector y(t); that is,

dS(x,
dx(t)
t)
y(0 Mi) (11-202)

From Eqs. (11-187), (11-196), and (11-197), y(t) is related to the Riccati gain
K(t) and the state vector x(t) through the following relation:

y(t) = K(t)x(t) (11-203)

Taking the derivative on both sides of Eq. (1 1-203), we have

y(0 = K(t)x(t) + K(t)x(t) (1 1-204)

Substituting Eqs. (11-194) and (11-184) in Eq. (11-204) and simplifying, we have

y(0 = -Qx(?) - A'y(f) (11-205)

which is referred to as the costate equation.

Equations (11-198) and (11-205) are combined to form the canonical state
equations,
x(0" A BR'B' x(0
(11-206)
Q -A' Ly(0J
which represent 2« linear homogeneous differential equations. The 2« boundary
conditions needed to solve Eq. (1 1-206) come from the n initial states, x(? ), and

y(t f ) = K(t f )x(h) = P*^) (1 1 -207)

Let the coefficient matrix of the canonical state equations in Eq. (1 1-206) be
represented by
"
A -BR-'B"
M (11-208)
 : : :

Sec. 11.6 Optimal Linear Regulator Design / 609

The matrix M has the following properties

1. M has no purely imaginary eigenvalues.
2. If A, is an eigenvalue of M, then the negative complex conjugate of
A,- is also an eigenvalue ; that is,

k t
=-X? (11-209)

Let W be the 2« X In similarity transformation matrix which transforms M

into a diagonal matrix T. Then
A
T = W-'MW (11-210)
6
where A is an n X n diagonal matrix that contains the eigenvalues of M with
positive real parts (in the right-half plane).
W be partitioned as follows
Let

W w w w 12
"

(11-211)
v ; 2 2

where W n W W and W are n x n matrices. The transformed

, 12 , 21 , 22 state and
costate variables are denoted by 1(0 and y(0, respectively, and are related to the
original variables through the following relation

x(0" w 12 i(0*
(11-212)
i(0. lW«V wV,
Equation (11-207) leads to the following boundary condition for the
transformed state equations

Htf) = -(W 22 - PW I2 )">(W 21 - PW n )i(f,) (H-213)

The transformed canonical state equations are written
0" "1(0"
= -A
"1(0" !

(11-214)
i(0_ _ 6 1 "a. i(0.
quation are

±(t) = e- A *-'»±.(t f ) (11-215)

y( == e^'-'^itf ) (11-216)

Since it is awkward to work in backward time, let us introduce a time

variable, t, which represents the time from t until the final time tf :

r = tr
Let
1(0 = X(t) = X(tf - (11-217)

y(0 = t(r) = ?(*, - (11-218)

Equations (11-215) and (11-216) become

= e At^(0)
5fc(T) (11-219)

$(t) = e 't(0)
-A
(11-220)
610 / Introduction to Optimal Control Chap. 11

Since A represents a diagonal matrix with eigenvalues that have positive

real parts, the solutions of Eq. (1 1-219) will increase with r. Thus we rearrange
Eq. (1 1-219), and when it is combined with Eq. (1 1-220), we get
At
£(or "e " "X(T)
(11-221)
f(T)_ e" At M)
Now Eq. (11-221) represents solutions that have only negative exponentials.
Using the boundary condition of Eq. (11-213), the relation between "£(0) and
X(0) can be written

t(0) = -(W 22 - PW 12 )~'(W 2I - PW u )X(0) (11-222)

Substitution of Eq. (11-222) into Eq. (11-221), the relation between ¥(t) and
X(t) is established:

?(t) - -e~ A '(W 22 - PW 12 )-'(Wai - PW M )r A i(t) (11-223)

Let
H(t) = -e~ *(W 22 - PW 12 )-'(W 21 - PW n >-^ (11-224)

Equation (11-223) becomes

t(t) = H(t)X(t) (11-225)

In order to return to the variables x(t) and y(0, we substitute Eq. (1 1-225) in Eq.
(11-212) and using Eqs. (11-217) and (11-218), we have the relationship between
y(0 and x(0,
y(0 = [W 21 + W^H^fW, + W I2 H(r)]-'x(r) , (1 1-226)

Comparing the last equation with Eq. (11-203), the Riccati gain matrix is
written

K(0 = [W 21 + W 22 H( ?/ - 0][W„ + W 12 H(r,

- t)Y (11-227)
or
K(r, - t) = [W 21 + W 22 H(t)][W u + W.jHCt)]- 1
(11-228)

Although we have a noniterative solution for K(f), because it is time dependent,

in general, for high-order systems, even a computer solution will be quite com-
plex.

Solution of the Algebraic Riccati Equation

In the infinite-time case, the Riccati equation is a matrix algebraic equation

given by Eq. (11-200). Based on the negative exponential solution for the
finite-time problem obtained in Section 11.5, we can obtain the solution of the
algebraic Riccati equation by simply taking the limit as t approaches infinity,
since
T = t
f
— = t oo ~ t=oo (1 1-229)

Thus, from Eq. (11-224),

lim H(t) = lim H(t f -0 = (1 1-230)

andEq. (11-228) gives

K= lim K(f f - t) = WjjWr, 1
(11-231)
Sec. 11.6 Optimal Linear Regulator Design / 611

Although the above solution is of a very simple form, in general, to deter-

mine the matrices 2 W
and W, requires the solution of the eigenvalues and the
, ,

eigenvectors of the 2« x In matrix M. For high-order systems, a computer

solution becomes necessary.
In order to illustrate the solution of the Riccati equation, we shall first
consider the optimal linear regulator design of a first-order system and carry
out the design by pencil and paper. Then a higher-order system will be designed
with the aid of a digital computer.

Example 11-7 Consider that a linear time-invariant system is described by the

state equation

x(t) = -2x(0 + «(/) (11-232)

with the initial condition given as x(t ). It is desired to determine the optimal control
u°(t) such that

/ = \xKtf) + £ f" t* (0
2
+ « 2 (0] * = min. (1 1-233)

We shall first consider that the final time // is finite. From Eqs. (11-232) and (11-233),
we identify that A = —2, B = 1, P= I, 2= 1, and R= 1. Using the negative
exponential method, we form the matrix M in Eq. (11-208),
A ~BR- l
B'~ ~-2 -r
M= -A' _-l
(11-234)
Q _ 2_

The eigenvalues of M are X i

= — V^T and A2 = +/T. The eigenvectors of M are
found to be
r i
- r i ~|

Pi ~ P2 = (11-235)
_0.235_ _-4.235_
Then
-
W = r^n w xz
_
1 1 "
(11-236)
j\ w y%x 2 2- jo.:>35 -4.:>35j

It is important to note that the eigenvectors should be so arranged in that they W

correspond to the requirement of the matrix T in Eq. (11-210). Substituting the ele-
ments of W, P = 1, and A
= X 2 = */3~ in Eq. (11-224), we get
H(X) = -0.146<?" 2 ^ (11-237)

The time-varying Riccati gain is obtained by substituting H(x) in Eq. (1 1-228). Thus

Kit, T) =
0.235 + 0.6189e"
- 0.146e- ^3' 2
2 ^
r
(11-238)
1

We note that when t = t,,x =0, Kit,) = P = 1, which agrees with the boundary
condition on K(t,).
Figure 11-18 gives a plot of Kit) as a function of /. The steady-state solution of
Kit) is obtained from Eq. (11-238) by setting t — co. —
For the infinite-time problem, t, = co. Since the system in Eq. (11-232) is com-
pletely controllable, the optimal solution exists, and the constant Riccati gain K is

solved from Eq. (11-238) by setting t = oo. Thus

# = 0.235 (11-239)

which is also the steady-state value of Kit).

 '

612 / Introduction to Optimal Control Chap. 11

1.0

Steady-state K{t)
0.52
value of K(t) / - 0.5
0.245
0.235
|

0.306

I I ,

'/-I t
f
- 0.5 t,

t - 0.3
f

Fig. 11-18. Time-varying Riccati gain.

The optimal control for the finite-time problem is

u°(t) = -R- B'K(t)x(t)

l

(11-240)
= -K(t)x(t) < t tf

and that of the infinite-time problem is

u°(t) = -0.23540 (11-241)

The state diagram for the feedback system designed for the infinite final time is

shown in Fig. 11-19.

-0.235

Fig. 11-19. Optimal linear regulator system in Example 11-7.

Example 11-8 In this example we consider the linear regulator design of the control
of a simplified one-axis model of the Large Space Telescope (LST).
The LST, which will be launched into orbit in 1980, is an optical
space telescope aimed at observing stars that cannot be observed by ground-stationed
telescopes. One of the major control problems in the design of the LST control system
is the pointing stability of the vehicle. In addition to the accuracy requirement, the
LST vehicle should be maintained at its equilibrium attitude position during operation.
This a typical regulator design problem.
is

The block diagram of the simplified one-axis model of the system is shown in
Fig. 11-20. An equivalent state diagram of the system is shown in Fig. 11-21, from which
are written the state equations

±(/) = Ax(/) + B«(0 (11-242)

 Sec. 1 1 .6 Optimal Linear Regulator Design / 613

Control
moment Vehicle
Gimbal controller gyro dynamics

u(t) K, »*
1 1

H
K.s + K, s V J»* 2

Fig. 11-20. Block diagram of a simplified one-axis model of the LST.

Fig. 11-21. State diagram of the simplified LST system.

where
1 "

H
J.
A= 1

J.

-K„
-Kj j

0"

IKj.

The system parameters are defined as:

H = 600 ft-lb-sec control moment gyro angular momentum

Jg = 2.1 ft-lb-sec 2 gimbal inertia

K, = 9700 ft-lb/rad gimbal-rate loop integral gain

Kp = 216 ft-lb/rad/sec gimbal-rate loop proportional gain

/„ = 10 ft-lb/rad/sec
s
vehicle inertia

The design objective is to find the optimal control that minimizes the following
performance index:
J=\" [x'(0Qx(0 + Ru 2
(<)] dt (1 1-243)
614 / Introduction to Optimal Control Chap. 11

where
Qi
0"

1z
(11-244)
93
1i-
Since the states xi and x 2 represent the vehicle displacement and velocity, respec-
tively, and are of primary interest, more weights are placed on these variables in the
specification of the elements of Q. We let q = t
5 x 10 7 q 2
,
= 5000, q 3 = </ 4 = 1, and
R= l.

The optimal control is given by

u\t) = -.R->B'Kx(0 (11-245)

where K is the positive definite solution of the algebraic Riccati equation,

-KA - A'K + KBR'B'K - Q = (11-246)

The solutions of the Riccati equation and the optimal control are obtained on a digital
computer. The eigenvalues of the feedback system are

-2.75 -9700 -1.375 + /2.33 -1.375 -J2.33

The feedback matrix is

G= [7071 5220 10.6 0.99] (11-247)

The state diagram of the feedback system with state feedback is shown in Fig.
11-22. An input is added to the system through a gain of 7071. The latter is included so
that the output of the system will follow a step function input without steady-state
error. Figure 1 1-23 shows the responses of x x (0„) and x 2 (0 V ) when the input is a step
function with a magnitude of 5 x 10" 8 rad. The initial conditions of the system are
set to zero.

10.6

Fig. 11-22. State diagram of the LST system with state feedback from
linear regulator design.
 Sec. 11.7 Design with Partial State Feedback / 615

0.05 r

0.00 1.50 3.00 4.50 6.00 7.50 9.00

Time (sec)

0.08 r

o
X
? 0.041-

J_ _L
0.00 1.50 3.00 4.50 6.00 7.50 9.00
Time (sec)

Fig. 11-23. Time responses of xi(t) and xi(t) of the LST system when the
input is a step function.

11.7 Design with Partial State Feedback

It is demonstrated in Section 11.5 that an observer can be designed for an

observable system to observe the state variables from the output variable.
However, in general, the complexity of the observer is comparable to that of the
system. Tt is natural to seek a design procedure that relies on the use of feedback
from only the accessible state variables through constant feedback gains. It

should be pointed out that control is achieved by feedback through constant

gains from the accessible states, and no dynamics are involved in the controller.
Of course, the design problem would only have a meaningful solution if the
system is stabilizable with feedback from the accessible state variables.
Given the linear time-invariant system

H.0 = Ax(f) + B«(0 (1 1-248)

where
x(?) = n x state vector
1

u{t) = scalar control

it is assumed that the system is completely controllable.
Let the accessible state variables be denoted by the vector y(?) (m x i,

m< n), and

y(0 = Cx(?) (11-249)
616 / Introduction to Optimal Control Chap. 1

where C is an m X n matrix with constant coefficients. It is assumed that C has

full rank; that is, the rank of C should be at least equal to m.
The design problem is to find the partial state feedback control

u(t)=-Fy(t) (11-250)

where F is an 1 x m feedback matrix such that the designed system satisfies a

certain performance index. The performance index chosen for this design is

/ = P [x'(0Qx(0 + Ru*(t)] dt (11-251)

•> to

where Q is a symmetric positive semidefinite matrix and R is a positive constant.

The design objective is to find the optimal control u(t) such that / is minimized.
The reason for the choice of this performance index is that the design can be
associated with the linear regulator design.
Substitution of Eq. (11-249) in Eq. (11-250) yields

w(0=-FCx(f) (11-252)
or
i/(0=-Gx(0 (11-253)
where
G = FC = fe I g2 ... g„] (11-254)

It has been shown 13 that the control in Eq. (1 1-253) is optimal in the sense
of minimizing the J in Eq. (1 1 -25 1 ) for some Q and R if and only if

1 1 + G(joil - A)" B | > 1 (1 1-255)

for all real co.

It is interesting to note that the transfer relation between x(?) and u(t) is

X(s) = (si - A)-'Bf7(^) (1 1-256)

The loop transfer function of the system is G(sl - A) _1 B. Thus 1 +

G(sl —
A) -1 B is equivalent to the function 1 + G(s) in the classical control
notation, and the condition of Eq. (11-255) is equivalent to the sensitivity
relation of Eq. (9-74). In other words, the optimal system designed in the sense
of minimizing / in Eq. (11-251) is that the Nyquist locus of G(j'coI — A)~'B
must not intersect or enclose the unit circle centered at ( — 1, y0) in the complex
function plane. Therefore, the system designed based on the optimal linear
regulator theory must all satisfy Eq. (11-255).
The design strategy here is to utilize the condition of Eq. (1 1-255) to select
a weighting matrix Q that is G is then
positive semidefinite. The resulting
considered as the optimal feedback gain corresponding to that Q. The approach
follows that of the inverse problem 11 of the linear regulator design, which
involves the determination of the weighting matrices once the feedback matrix
is known.
For convenience, we R=
1, since it is relative to the magnitudes of the
let

elements of Q. Let p(s) represent the characteristic equation of the closed-loop

system with partial state feedback; that is,
Sec. 1 1 .7 Design with Partial State Feedback / 617

p(s) = I
si - (A - BG) |
(1 1-257)

Since the system is controllable, it can be represented in phase-variable canonical

form. The matrices A and B are written

1 ...
"0
1 ...

...
1
B = (11-258)

0.. 1

-a, —a 2
—a 3 — a„ ... — a„ 1

The characteristic equation of A is denoted by

q(s) = \sl-A\ = s* + a„s-~' + a„^s"- + 2
. . .+ a, (11-259)

The characteristic equation of the closed-loop system is written

p(s) = s° + (a. + ^>"- + ]

...+ (a l +g 1 ) (1 1-260)
Then

(si - A)'B (11-261)

'?(')

and
G(sl - A) >B
.g*s" g„_ 1
5°- 2
+ ... + g, (11-262)
q(s)

Thus
[l+G(*I-A)-'B]=2g (11-263)

In view of Eq. (11-255), we can write

1
1 + GC/col - A)-'B| 2
= [1 + G(;coI - A)-'B][1 + G(-jcol - A)~'B]
= p(Jco)p(-jco) > j (1 1-264)
q(Jco)q(-jco)

Subtracting 1 from both sides of the last equation, we have

pUa>)p(—JG>) qUco)q(-jco) -^ q- (11-265)
qUoj)q(-Jo>)
Let the numerator of the last equation be factored as

p(Joi)p(.-Jco) - q(j(o)q(-j<o) = d(jco)d(-jco) (11-266)

where
d(s) = a>"-' + d„^s"- 2 + . . . + a", (11-267)

Equation (1 1-267) is of the (n — l)st order since the s" terms in p(s) and q(s) are
618 / Introduction to Optimal Control Chap. 11

canceled in Eq. (11-266). Let

d2
D= (11-268)

Then
D(sl - A)~'B (11-269)

Equation (11-264) is written as

[1 + G(jcol - A)"«B][1 + G(-yo)I - A)~'B] = + 1

^-J^)
q(joj)q(—jco)
^l
(H-270)
= + 1 B'(-ycoI - AT'O'DOoI - A)"'B
has been shown by Anderson and
It Moore 12
that for G to be the optimal
feedback matrix, it is necessary that

[1 + G(jcoI - A)-»B][1 + G(-jcoI - A)"'B]

(11-271)
= + B'(-ycoI - AT 'QCAol -
l A)"'B
Comparing Eqs. (11-270) and (11-271), we have
Q = DD (11-272)

Equations (11-266), (11-272), and the condition that Q is positive semi-

definite form a design procedure for the partial state feedback. It should be kept
in mind that since the system may not be stabilizable, or the inverse problem
does not exist, an optimal solution may not always be possible.

Example 11-9 Consider the dynamic equations of a linear time-invariant system,

i(0= Ax(0 + Ba(0 (11-273)

X')=* 2 (0 (11-274)
where
"or B=
"0"

L-i o_ L i J
It is desired to design a closed-loop system by feeding back only x 2 (t) through a
constant gain, so that the performance index

/ = !""
[x'(0Qx(0 + uHt)] dt (11-275)
J t

is minimized. The control is u(t) = — Gx(t).

The characteristic equation of A is

q(s) = |il- A| =s + 2
1 (11-276)
and the characteristic equation of the closed-loop system with feedback from x 2 is

p(s) = \sl - (A - BG)| = s + g 2 s +2

1 (11-277)
where
G=[0 g2 ] (11-278)
Sec. 1 1 .7 Design with Partial State Feedback / 619

For the second-order system,

d(s) = dz s + di (11-279)

D= [rf, d2 ] (11-280)

Then the weighting matrix Q is of the form

-d\ did
Q= D'D = ,

(11-281)
-d\di
From Eq. (1 1 -266) we have
p(s)p(-s) = s* + (2 - gi)s 2 + 1
(11-282)
= q(s)q(-s) + d(s)d(-s) = s* + (2 dl)s* + d\ + 1

Therefore,
d t
=0 d\ = gi
and
"0 0" "0 0"
(11-283)
_0 d\\ Lo gi\
This result shows that Q must
be of the form defined in Eq. (11-283), or the optimal
solution does not exist for the problem given here. However, it does not mean that

given any other positive semidefinite Q an optimal linear regulator solution does not
exist.

The nonzero element of Q, d\, represents a constraint on the state x 2 which is the
derivative of x x . It is reasoned that if we place more constraint on x 2 , the overshoot of
the system will be reduced. This is verified by assigning different values to d2 and
,

observing the effect on the roots of the characteristic equation of the feedback system.

d\ = 1 p(s) =s +s+1
2
damping ratio C = 0.5
d\ =2: p(s) = s 2 + *fls -
damping ratio £ = 0.707
d\ =4: p(s) = s + 2s + 1
2
damping ratio £ = 1.0
In more complex systems, Eq. (1 1-266) leads to a set of nonlinear equations
that must be solved to obtain the elements of D and G. The problem is simplified
by observing that the elements in the last row of the matrix A in Eq. (11-258)
already represent feedbacks from the state variables. Therefore, it is logical to
absorb these coefficients in the feedback matrix G. Then the modified matrix
A would be
"0 1 ... 0"

A* = (11-284)

0... 1

...

The state equations of the system become

x(/) = A*x(r) + Bu(t) (11-285)

and the state feedback control is

u(t) = -G*x(r) (11-286)

620 / Introduction to Optimal Control Chap. 11

where
G* = G-[0 ... 1]A (11-287)

= igi — «i gi - a2 . . . g„ - a„]

It is simple to see that the characteristic equation of A* will always be

q(s) = s" (11-288)

The following example will illustrate this simplified procedure.

Example 11-10 Consider the linear time-invariant system

±(0 = Ax(0 + B«(0

where
1
0" "0"
A= 1 B =
-2 -3. _1_
We wish to find the feedback gain matrix G= [g t
0] such that the performance
index

/ = [x'(0Qx(0 + « 2 (01 dt
Jf~
to

is minimized. The form of G given indicates that only the state .vi is fed back for
control. Using Eq. (11-287), G* = fc, 2 3].

From Eq. (11-284), the characteristic equation of the closed-loop system is

p(s) = \sl - (A - BG)| = \sl - (A* - BG*)|

(11-289)
= s 3
+ 3.s
2
+ 2s + g x

Let
d(s) = d S 2 + d2 S + d,
3 (11-290)

From Eq. (11-266), we have

p(s)p(-s) - q(s)q(-s) = 5s 4 + (6g - 4)s 2 + g\ t

(11-291)
= dls* + (Idtds - di)s 4 2
d\

Thus equating the coefficients of corresponding terms in the last equation, we have
d\=g\ (11-292)

d\ = 5 (11-293)

2did3 ~dl=6g -4 1 (11-294)

Substitution of Eqs. (11-292) and (11-293) into Eq. (11-294) yields

d\ = 4 - 1.5278^ (11-295)

Since d\ must be positive, Eq. (11-295) leads to

di < 1.5278
= 2.618 (11-296)

From Eq. (11-292),

g, < 2.618 (11-297)

It is interesting to note that applying the Routh-Hurwitz criterion to the charac-

teristic equation of Eq. (11-289) shows that the closed-loop system with feedback from
Xi is stable for #i < 6. However, for the optimal partial state feedback control, g x
 . 1

Chap. 1 References / 621

must be less than or equal to 2.618. The difference between the two values of ^i is the
margin of relative stability.
The weighting matrix is given by
d\ didz d^df
D'D = did2 d\ d2 d3 (11-298)

-did 3 dzdi d\ _

whose elements have to satisfy Eqs. (11-292) through (11-294).

REFERENCES

Analytical Design

1 G. C. Newton, Jr., L. A. Gould, and J. F. Kaiser, Analytical Design of Linear

Feedback Controls, John Wiley & Sons, Inc., New York, 1957.

2. Chang, Synthesis of Optimum Control Systems, McGraw-Hill Book

S. S. L.
Company, New York, 1961.

Design by Pole Assignment

3. W. M. Wonham, "On Pole Assignment in Multi-input Controllable Linear

Systems," IEEE Trans. Automatic Control, Vol. AC-12, pp. 660-665, Dec. 1967.

4. F. M. Brasch and J. B. Pearson, "Pole Placement Using Dynamic Compensa-

tors," IEEE Trans. Automatic Control, Vol. AC-15, pp. 34-43, Feb. 1970.

5. E. J. Davison, "On Pole Assignment in Linear Systems with Incomplete State

Feedback," IEEE Trans. Automatic Control, Vol. AC-15, pp. 348-351, June 1970.
6. E. J. Davison and R. Chatterjee, "A Note on Pole Assignment in Linear Sys-
tems with Incomplete State Feedback," IEEE Trans. Automatic Control, Vol.
AC-16, pp. 98-99, Feb. 1971.
7. J. C. Willems and S. K. Mitter, "Controllability, Observability, Pole Alloca-
tion, and State Reconstruction," IEEE Trans. Automatic Control, Vol. AC-16,
pp. 582-593, Dec. 1971.

8. B. Sridher and D. P. Lindorff, "A Note on Pole Assignment," IEEE Trans.

Automatic Control, Vol. AC-17, pp. 822-823, Dec. 1972.

Design of Observers

9. D. G. Luenberger, "Observing the State of a Linear System," IEEE Trans.

Military Electronics, Vol. MIL-8, pp. 74-80, Apr. 1964.

10. D. G. Luenberger, "Observers for Multivariate Systems," IEEE Trans. Auto-

matic Control, Vol. AC-11, pp. 190-197, Apr. 1966.

11. D. G. Luenberger, "An Introduction to Observers," IEEE Trans. Automatic

Control, Vol. AC-16, pp. 596-602, Dec. 1971.

12. B. D. O. Anderson and J. B. Moore, Linear Optimal Control, Prentice-Hall,

Inc., Englewood Cliffs, N.J., 1971.
622 / Introduction to Optimal Control Chap. 1

Linear Regulator Design

13. R. E. Kalman, "When Is a Linear Control System Optimal?" Trans. ASME,

J. Basic Eng., Ser. D, Vol. 86, pp. 51-60, Mar. 1964.

14. R. E. Kalman, "Contributions to the Theory of Optimal Control," Bol. Soc.

Mat. Mex., Vol. 5, pp. 102-119, 1960.

15. M. Athans and P. L. Falb, Optimal Control, McGraw-Hill Book Company,

New York, 1966.

16. A. P. Sage, Optimum System Control, Prentice-Hall, Inc., Englewood Cliffs,

N.J., 1968.

PROBLEMS

Prove that the "integral of 2

11.1. 1 e\t) criterion,"

j = f" t
2
e\t) dt

can be expressed as

J = - F(s)F(-s) ds
*' J -j.

where F(s) = dE(s)/ds.

11.2. A unity-feedback control system is shown in Fig. PI 1-2.

r(t) /\ /\ e(t) K
c(t)
+ a)
+ \/ s(s

Figure Pll-2.

(a) Determine the integral square error (ISE),

J= {°°
e 2 (t)\dt
i

when K = 10, r(t) = u s (t) (unit step input), and a = 2.

(b) Find the value of a as a function of K
so that the ISE is minimized, when
r(t) = u s (t). Determine the damping ratio of the optimal system.
(c) If K is variable but a is held constant, find the optimal value of K so that
the ISE is a minimum.
(d) Repeat part (a) when the input is a unit ramp function, r(t) = tu s (t).
11.3. The transfer function of a linear process is

U(s) s2
 ; .

Chap. 11 Problems / 623

Determine the optimal closed-loop transfer function C(s)/R(s) such that

Je = I KO ~ c(t)]
z
dt = minimum

-i dt<l
J

The reference input r(t) is a unit step function,

11.4. A linear process is described by

x x
= x2
x2 =u
where x and x 2 are state variables and u
t
is the control. It is desired to mini-
mize the quadratic performance index

j-i J> - av) + « 'xt

2 2
] rfc

where jcd = constant. Determine the optimal control as a function of x it x 2 ,

and xd by means of the integral-square-error design. Draw a block diagram

,

for the completely designed system.

11.5. For the control system shown in Fig. PI 1-5, the input r{t) is a unit step function.

Figure Pll-5.

(a) Determine Ki and K2 so that the integral square error Je is minimized,

subject to Ju < 1

= f~
e\t) dt Ju [" d 2 (t) dt

The noise signal n(t) is assumed to be zero.

(b) If /•(/) = 0, and the values of K and K2 are as determined

t
in part (a),
find the maximum strength of the impulse disturbance n(t) = Nd(f) that
can be applied to the system such that /„ < 1

(c) Repeat part (a) with the following design criteria:

Je<\ Ju minimum
(d) Repeat part (b), and find the maximum strength of the impulse disturbance
N such that /„ is less than or equal to the minimum value found in (c).
624 / Introduction to Optimal Control Chap. 11

11.6. The control system shown in Fig. PI 1-6 has two free parameters in K\ and
K 2 . Determine the values of K 1
and K2 by parameter optimization such that
Je = e 1 {t) dt = minimum

J" = " 2(/) dt -25 '

J" I

The reference input is a step function with an amplitude of 0.5.

*~c(t)

Figure Pll-6.

11.7. Consider the linear process

x(0 = Ax(0 + B«(0

c(t) = Dx(r)
where
"1 0" "1"
B= D= [2 -U
Lo 0_ L i J

Design a state observer so that x(0 — x e(t) will decay as fast as <r 10t Find the
.

characteristic equation of the observer and the feedback matrix G„. Write the
state equation of the observer in matrix form.

11.8. A linear time-invariant process described by the state equation

is

x(t) = -0.5x(t) + u(t)

with the initial state x(t ) = x . Find the optimal control u°(t) that minimizes
the performance index

J = i J\"[2x\t) + uHt)]dt
to

where t = and tf = I sec.

11.9. Solve Problem 11.8 with t f = oo.

11.10. Consider the linear process

x{t) = 0.5x(t) + u(t)

It is desired to minimize the performance index

T
rr
J = (\e~<x 2 + e~'u 2 ) dt
J
Chap. 11 Problems / 625

Show that the optimal control is given by

1 1 _ ptp-T

11.11. A second-order process is described by the state equations

*i(t) = x 2 (t)

x 2 (t) = u(t)
Find the optimal state-feedback control that minimizes

/ = i.
r ( X\ + 2x lXz + 4x\ + u 1 ) dt

Draw a block diagram for the closed-loop system. Find the damping ratio of
the system.
11.12. Consider the linear time-invariant process

±(0 = Ax(/) + B«(0

where
"0 1

A= 1 B =
l_4 -4 1. 1J
It is desired to find the feedback gain matrix G = [#i g 3 ] such that the
performance index

J = [°°
[x'(0Qx(0 + u 2 (t)] dt = minimum
J to

Formulate the problem using the method described by Eqs. (11-284) through
(11-288). Find the bounds on g^ and g 3 such that solutions to the optimal
linear regulator problem with the specific partial state feedback exist.
 APPENDIX A
Frequency-Domain Plots

Consider that G(s)H{s) is the loop transfer function of a feedback control

system. The sinusoidal steady-state transfer function
is obtained by setting
s = jco
in G(s)H(s). In control systems studies, frequency-domain plots of the
open-loop transfer function G(jco)H(jco) are made for the purpose of analysis of
the performance of the closed-loop control system.
The function G(jco) is generally a complex function of the frequency co and
can be written

G(jco) = \G(jco)\ /GUa>) (A-l)

where G(jco) denotes the magnitude of G(jco) and /G(jco) is the phase of G(jco).
| |

The following forms of frequency-domain plots of G(jco) [or of G(jco)H(jco)]

versus co are most useful in the analysis and design of feedback control systems
in the frequency domain.

1. Polar plot: a plot of the magnitude versus phase in the polar coor-
dinates as co is varied from zero to infinity.
2. Bode plot (corner plot): a plot of the magnitude in decibels versus
co (or log 10 co) in the semilog (or rectangular) coordinates.
3. Magnitude- versus-phase plot: a plot of the magnitude in decibels
versus the phase on rectangular coordinates with co as a variable
parameter on the curve.

These various plots are described in the following sections.

626
 A.1 Polar Plots of Transfer Functions / 627

A.I Polar Plots of Transfer Functions

The polar plot of a transfer function G(s) is a plot of the magnitude of G(jco)
versus the phase of G(jco) on the polar coordinates, as co is varied from zero to
infinity. From a mathematical viewpoint, the process may be regarded as a map-
ping of the positive half of the imaginary axis of the .s-plane onto the plane of the
function G(jco). A simple example of this mapping is shown in Fig. A-l. For any
frequency co
co =
the magnitude and phase of G(jcoi) are represented by a
x ,

phasor that has the corresponding magnitude and phase angle in the G(jco)-

i/CO

s-plane

./co 2

-jOJ l

,,/ImC
G(/co)-plane

*-ReG

Fig. A-l. Polar plot shown as a mapping of the positive half of the/co axis
in the .s-plane onto the G(yeB)-plane.
628 / Frequency-Domain Plots App. A

plane. In measuring the phase, counterclockwise is referred to as positive, and

clockwise as negative.
To illustrate the construction of the polar plot of a transfer function, con-
sider the function

1
G(s) (A-2)
1 +Ts
where T is a positive constant.
Putting s = jco, we have

G(jco) = 1
(A-3)
1 +jcoT
In terms of magnitude and phase, the last expression is written

1
G(jco) /-tan-' coT (A-4)
*J\ + <o r 2 2

When a> is zero, the magnitude of G(jco) is unity, and the phase of G(jco) is
at 0°. Thus, at co =
0, G(jco) is represented by a phasor of unit length directed in
the 0° direction. As co increases, the magnitude of G(jco) decreases, and the phase
becomes more negative. As co increases, the length of the phasor in the polar
coordinates decreases, and the phasor rotates in the clockwise (negative) direc-
tion. When co approaches infinity, the magnitude of G(jco) becomes zero, and
the phase reaches —90°. This is often represented by a phasor with an infinitesi-
mally small length directed along the —90° axis in the G{ j'co)-plane. By sub-
stituting other finite values of co into Eq. (A-4), the exact plot of G(jco) turns out
to be a semicircle, as shown in Fig. A-2.
As a second illustrative example, consider the transfer function

G(jco) = + jcoT
1 2
(A-5)
1 +jcoT 1

where 7^ and T2 are positive constants. Equation (A-5) is written

/ImC G(/'co)-plane

ReC

V 1 + w2 T 2

Fig. A-2. Polar plot of G(jco) = 1/(1 +jcoT).

A.1 Polar Plots of Transfer Functions / 629

or -
G(jco) = yy /tan" 1
coT 2 tan~' coT (A-6)
+ co
2
T]
x

The polar plot of G(jco), in this case, depends upon the relative magnitudes of
T2 and r, . If T2 is greater than T, the magnitude of G(jco)
, is always greater than
unity as co is varied from zero to infinity, and the phase of G(ja>) is always posi-
tive. If T2 than Tj the magnitude of G(jco) is always
is less , less than unity, and
the phase always negative. The polar plots of G(j(o) of Eq. (A-6) that cor-
is

respond to the two above-mentioned conditions are shown in Fig. A-3.

/ImG G(/co)-plane

*-ReG
co = T2 m
(T2 > T,

,,/ImG
G(/co)-plane

zyr,
+-ReG

(T2 <T x )

Fig. A-3. Polar plots of C(Jco) = (1 +j<oT2 )/(l +ja>Ti).

It is apparent that the accurate plotting of the polar plot of a transfer func-
tion is generally a tedious process, especially if the transfer function is of high

order. In practice, a digital computer can be used to generate the data, or even
the final figure of the polar plot, for a wide class of transfer functions. However,
from the analytical standpoint, it is essential that the engineer be completely
computer data may be
familiar with the properties of the polar plot, so that the
properly interpreted. In some cases, such as in the Nyquist stability study, only
the general shape of the polar plot of G(jo})H(ja>) is needed, and often a rough
sketch of the polar plot is quite adequate for the specific objective. In general,
630 / Frequency-Domain Plots App. A

the sketching of the polar plot is facilitated by the following information:

The behavior of the magnitude and the phase at co == and at co = oo.
1.

2. The points of intersections of the polar plot with the real and imaginary

axes, and the values of co at these intersections.

The general shape of the polar plot may be determined once we have infor-
mation on the two items listed above.

Example A-l Consider that it is desired to make a rough sketch of the polar plot
of the transfer function

Substituting s =
jco in Eq. (A-7), the magnitude and phase of G(jco) at co = and
co =oo are computed as follows:

lim G(jco)
| |
= lim ^ = oo (A-8)

lim IG(jco) = lim /-)£ = -90° (A-9)

lim G(jco)
| |
= lim -^
CO
= (A-10)
<0-»oo

lim /G(jco) = lim A^L = -180° (A-ll)

Thus the properties of the polar plot of G(jco) at CO = and co = oo are ascertained.
Next, we determine the intersections, if any, of the polar plot with the two axes of the
G(y'C0)-plane.
If the polar plot of G(yCO) intersects the real axis, at the point of intersection, the
imaginary part of G(/co) is zero; that is,

Im[G(/co)] = (A-l 2)

In order to express G(jco) as

G(yco) = Re[G(yco)] + j Im[G(/co)] (A-13)

we must rationalize G(jco) by multiplying its numerator and denominator by the com-
plex conjugate of its denominator. Therefore, G(y'co) is written

= 10(-yco)(-/CO + 1) _ -10C0 2
_ 10C0 ,.
C(im\
KJ ~ co* + co
.

lA" 14
' jcoO'co + l)(-jco)(-jco + 1)
2 J
co* + co 2
'

which gives
- l
mGUco)]= (o( l l)
(A-15)

and

Re[G(7co)] = ^^L (A-16)

When we set Im[G(yco)] to zero, we get CO = oo, meaning that the only intersect
that the G(jco) plot has with the real axis of the plane is at the origin.
Similarly, the intersection of the polar plot of G(jco) with the imaginary axis is
found by setting Re[G(yco)] of Eq. (A-16) to zero. The only real solution for co is also
co = co, which corresponds to the origin of the G(/co)-plane. The conclusion is that the
polar plot of G(yco) does not intersect any one of the two axes at any finite nonzero
A.1 Polar Plots of Transfer Functions / 631

l/lmG

G(/co)-plane

-»-Re G

Fig. A-4. Sketch of the polar plot of G(s) = I0l[s(s + 1)].

frequency. Based upon this information, as well as knowledge on the angles of GO'©)
at co = and co = co, the polar plot of G(jo>) is easily sketched, as shown in Fig. A-4.

Example A-2 Given the transfer function

10
G(s) (A-17)
'
s(s + l)(.s + 2)
it is desired to make a rough sketch of the polar plot ofG(jco). The following calcula-
tions are made for the properties of the magnitude and phase of G(joo) at co = and
co = oo:

5
lim GC/eo)
| |
= lim (A-18)
HJ-.0 cu->0 CO

lim /Gt/c») = lim 14-

JCO
= -90° (A-19)
OJ-.0 t»-»0 /

Km |
G(yco) |
= lim -g = (A-20)

lim /GC/OJ) = lim /^j = -270 (A-21)

To find the intersections of the G(jco) curve on the real and the imaginary axes of the
G(ya>)-plane, we rationalize G(jco) to give

G(uiC0 ) = m-jco)(-jco + 1)(-/cq + 2)

' jcoUco + l)(jco + 2)(-jco)(-jco + IK-jco + (A " 22)
2)

After simplification, Eq. (A-22) is written

= -30a> 2 JlOaKZ - a)>)

w
G(ico)
' 9co* + co 2 (2 - co 2 2
9co* + co 2 (2 - co 2 2 (A-23)
) )
632 / Frequency-Domain Plots App. A

We set
= -30
Re[G(/co)] (A-24)
9co 2 + (2 - co 2 2
)

and
= -10(2 - co 2 )
Im[G(/w)] - CO (A-25)
9© 3 + co(2 1 2
)

Equation (A-24) is satisfied when

CO = oo

which means that the G(jco) plot intersects the imaginary axis only at the origin. Equa-
tion (A-25) is satisfied when
CO 2 =2
which gives the intersection on the real axis of the G(/C0)-plane when co = ±*J~2
rad/sec. Substituting co = «/2~ into Eq. (A-23) gives the point of intersection at

G(yV2)=-j (A-26)

The result of co = —^/2 rad/sec has no physical meaning, but mathematically

it simply represents a mapping point on the negative jco axis of the j-plane. In general,
if G(s) is a rational function of 5 (a quotient of two polynomials of .0, the polar plot

of G(jco) for negative values of CO is the mirror image of that for positive co, with the
mirror placed on the real axis of the G(y'co)-plane.
With the information collected above, it is now possible to make a sketch of the
polar plot for the transfer function in Eq. (A-17), and the sketch is shown in Fig. A-5.
Although the method of obtaining the rough sketch of the polar plot of a transfer
function as described above is quite straightforward, in general, for complex transfer

i/IraC

G(/oo)-plane

Re G

Fig. A-5. Sketch of the polar plot of G(s) = lOMs + l)(s + 2)].
 Bode Plot (Corner Plot) of a Transfer Function / 633
A 2

functions that may have multiple crossings on the real and imaginary axes in the trans-
fer function plane, the algebraic manipulation may again be quite involved. Further-
more, the polar plot is basically a tool for analysis; it is somewhat awkward for design

purposes. We shall show in the next section that approximate information on the polar
plot can always be obtained from the Bode plot, which is usually sketched without any
computations. Thus, for more complex transfer functions, other than using the digital
computer, sketches of the polar plots are preferably obtained with the help of the
Bode plots.

A.2 Bode Plot (Corner Plot) of a Transfer Function

The discussions in the last section show that the polar plot portrays a function
G(jco) in the polar coordinates in magnitude and phase as functions
terms of its

of co. The Bode plot, on the other hand, contains two graphs, one with the
magnitude of G(jco) plotted in decibels versus log co or co, and the other with the
phase of G(jco) in degrees as a function of log co or co. The Bode plot is also
known sometimes as the corner plot or the logarithmic plot of G(jco)- The name,
corner plot, is used since the Bode plot is basically an approximation method in
that the magnitude of G(jco) in decibles as a function of co is approximated by

straight-line segments.
In simple terms, the Bode plot has the following unique characteristics:

1. Since the magnitude of G(jco) in the Bode plot is expressed in deci-

bels, the product and division factors in G{jco) become additions and
subtractions, respectively. The phase relations are also added and
subtracted from each other in a natural way.
2. The magnitude plot of the Bode plots of most G(jco) functions en-
countered in control systems may be approximated by straight-line
segments. This makes the construction of the Bode plot very simple.

Since the corner plot is relatively easy to construct, and usually without
point-by-point plotting, it may be
used to generate data necessary for the other
frequency-domain plots, such as the polar plot, or the magnitude-versus-phase
plot, which is discussed later in this chapter.
In general, we may represent the open-loop transfer function of a feedback
control system without pure time delay by

0(5) = ^+^
+
s-(s
+
+ pj(s
*»>/••
p2 )
{'
(s
+ *->
+ p„)
(A-27)

where K is a real constant and the zs and the/?s may be real or complex numbers.
As an alternative, the open-loop transfer function is written

G{S) - (A 28)
s'(l + Ta sKl+Tb s)---(l+T„s)
where A" is a real constant, and the Ts may be real or complex numbers.
In Chapter 8, Eq. (A-27) is the preferred form for root loci construction.
However, for Bode plots, the transfer function should first be written in the form
634 / Frequency-Domain Plots App A

of Eq. (A-28). Since practically all the terms in Eq. (A-28) are of the same form,
without loss of generality, we can use the following transfer function to illus-
trate the construction of the Bode diagram

s(l +T a s)(l + j2{/u - ju

2
)

where K, T T 2 Ta
t , , , f, and ju are real coefficients. It is assumed that the second-

order polynomial, 1 + 2^/j. — /i

1
, // = co/co n , has two complex-conjugate zeros.
The magnitude of G(jco) in decibels is obtained by multiplying the logarithm
to the base 10 of G(joi) | |
by 20; we have

I
G(jco) |„ = 20 log.o G(jco) |
= 20 log I0 K + 20 log,
| + jcoT I I
„ 1 1 l |

+ 20 log, o 1 1 + j(oT - 20 log, ja> 1 1

(A-30) 1 |

- 20 log,„ 1 1 -
+ joaT.\ 20 log 10 + J2{ M - y}\ 1
1

The phase of G(j(o) is written

/G(jco) = /K + /l+ jcoT, + /l + Jq}T - 2 jjco

- /I + JmT - /l+j2j M -^ m

In general, the function G(jco) may be of higher order than that of Eq.
(A-29) and have many more factored terms. However, Eqs. (A-30) and (A-31)
would simply produce more' similar
indicate that additional terms in G(jco)
terms in the magnitude and phase expressions, so that the basic method of
construction of the Bode plot would be the same. We have also indicated that,
in general, G(jco) may contain just four simple types of factors:

1 Constant factor K
2. Poles or zeros at the origin ±p
(jco)
3. Poles or zeros not at co = (1 + jcoT) ±q
4. Complex poles or zeros (1 + j2£/z — 2 ±r
fi )

where p, q, and r are positive integers.

Equations (A-30) and (A-31) verify one of the unique characteristics of the
Bode plot in that each of the four types of factors listed may be considered as a
separate plot; the individual plots are then added or subtracted accordingly to
yield the total magnitude in decibels and phase plot ofG(jco). The curves may be
done on semilog graph paper or linear rectangular coordinate graph paper,
depending on whether co or log I0 co is used as the abscissa.
We shall now investigate the sketching of the Bode plot of the different
types of factors.

Constant Term, K
Since
KiB = 20 log I0 K= constant (A-32)
and
K>
£ = W
(0°

*<o (A " 33)

A.2 Bode Plol (Corner Plot) of a Transfer Function / 635

20 log 10 K dB
Magnitude
0dBf_ -
indB

90°
ArgK
Phase in 0°
degrees
-90° - -
- 180°
0.1 1 10 100
co rad/sec

Fig. A-6. Bode plot of constant K.

the Bode plot of the constant factor K is shown in Fig. A-6 in semilog coordi-
nates.

Poles and Zeros at the Origin, (jm) ±p

The magnitude of (jco) ±p in decibels is given by

201og, |Oco)
± '|= ±20/7 log, „ co dB (A-34)

for ctJ > 0. The last expression for a given p represents the equation of a straight
line in either semilog or rectangular coordinates. The slopes of these straight
lines may be determined by taking the derivative of Eq. (A-34) with respect to
log 10 co; that is,

dlog ioQ)
(± 2°P lo Sio ca) = ±20/7 dB (A-35)

These lines all pass through the 0-dB point at co = 1. Thus a unit change in

log 10 co will correspond to a change of ±20/? dB in magnitude. Furthermore, a

unit change in log, co in the rectangular coordinates is equivalent to 1 decade of
variation in co, that is, from 1 to 10, 10 to 100, and so on, in the semilog coordi-
nates. Thus the slopes of the straight lines described by Eq. (A-34) are said to
be ±20/7 dB/decade of frequency.
Instead of using decades, sometimes the unit octave is used to represent the
separation of two frequencies. The frequencies co, and co 2 are separated by an
octave if CO2/CO1 = 2. The number of decades between any two frequencies co,
and co 2 is given by
io Bio (a>i/<»i)
number of decades = = i og
°
("h) (A-36)
log,o 10 \coJ
Similarly, the number of octaves between co 2 and co, is

number of octaves = logi° W<o,) = 1

log
(coA (A . 3?)
log, 2 0.301 ° \coJ
Thus, the relation between octaves and decades is given by

number of octaves = 1
decades (A-38)
0.301
636 / Frequency-Domain Plots App. A

Substituting Eq. (A-38) into Eq. (A-35), we have

±20/7 dB/decade = ±2Qp x 0.301 ~ ±6p dB/octave (A-39)

For a transfer function G(jco) that has a simple pole at s =--

0, the magnitude
of G(jco) is a straight line with a slope of —20 dB/decade, and passes through the
0-dB axis at cu = 1.

The phase of (jco) ±p is written

l(jco)
±p = ±p x 90° (A-40)

The magnitude and phase curves of the function ±p

(jco) are sketched as
shown in Fig. A-7 for several values of p.

0Q
T3
s:

<4-
o

100
cj rad/sec

Fig. A-7. Bode plots of Uo>) ±p -

Simple Zero (1 -\- j<oT)

Let
G(jco) = + jaT
1 (A-41)

where T is a real constant. The magnitude of G(j(o) in decibels is written

I
G(ja>) |„ = 20 log , 1 GO) = 20 log V 1~T ~^T*
| ,
( A-42)
A.2 Bode Plot (Corner Plot) of a Transfer Function / 637

To obtain asymptotic approximations to the magnitude of G(jco), we consider

both very large and very small values of o>. At very low frequencies,
coT -€. 1,
Eq. (A-42) is approximated by

I
G(jco) dB |
= 20 log , 1 G(jco) |
~ 20 log , o 1 = dB (A-43)
since co 2
T2
is neglected when compared with 1.
At very high frequencies, cor> 1, we may approximate 1 + co
2
T2 by
co
z
T2 ; then Eq. (A-42) becomes

I
G(jco) U = 20 log I0 1 G(joi) |
~ 20 log I0 Jco 2 T 2
(A-44)
= 20 log, coT
Equation (A-44) respresents a straight line with a slope of
+20 dB/decade of
frequency. The intersect of this line with the 0-dB axis is found by equating Eq.
(A-44) to zero, which gives

co = 1
-=-
(A-45)

This frequency is also the intersect of the high-frequency

approximate plot and
the low-frequency approximate plot which is the 0-dB
line as given by Eq. (A-43)
The frequency given by Eq. (A-45) is also known as the corner
frequency of the
Bode plot of the transfer function in Eq. (A-41), since the approximate magni-
tude plot forms the shape of a corner at that frequency,
as shown in Fig. A-8. The
actual magnitude curve for G(jco) of Eq. (A-41) is
a smooth curve, and deviates
only slightly from the straight-line approximation. The
actual values for the
magnitude of the function 1 +
jcoT as functions of coT are tabulated in Table A-l

40 II
I II III I I
I
I I Ml

30

20

£0
10

10

-20

30-

40 _LL
0.01 0.1

Fig. A-8. Magnitude versus frequency of the Bode plots of G(s) = 1 + Ts

and G(s) = 1/(1 + Ts).
638 / Frequency-Domain Plots App A

Table A-l

°>T \osiqC0T |1 + jwT\ \l + ja>T\ /l + jwT

(dB)

0.01 -2 1 0.5°
0.1 -1 1.04 0.043 5.7°
0.5 -0.3 1.12 1 26.6°
0.76 -0.12 1.26 2 37.4°
1.0 1.41 3 45.0°
1.31 0.117 1.65 4.3 52.7°
1.73 0.238 2.0 6.0 60.0°
2.0 0.3 2.23 7.0 63.4°
5.0 0.7 5.1 14.2 78.7°
0.0 1.0 10.4 20.3 84.3°

Table A-2

Straight-Line
Approximation of
coT + j<oT\ |i+;a»n Error
(dB) (dBY {dB)

0.1 (1 decade 0.043 0.043

below corner
frequency)
0.5 (1 octave 1.0 1

below corner
frequency)
0.76 2 2
1 .0 (at the corner 3 3
frequency)
1.31 4.3 2.3 2
2.0 (1 octave 7 6 1

above corner
frequency)
10 (1 decade 20.043 20 0.043
above corner
frequency)

Table A-2 gives a comparison of the actual values with the straight-line approxi-
mations at some significant frequencies.
The error between the actual magnitude curve and the straight-line asymp-
totes is symmetrical with respect to the corner frequency \/T. Furthermore,
it is useful to remember that the error is 3 dB at the corner frequency, and 1 dB
at 1 octave above (2jT) and 1 octave below (0.5/T) the corner frequency. At 1

decade above and below the corner frequency, the error is dropped to approxi-
mately 0.3 dB. From these facts, the procedure in obtaining the magnitude curve
of the plot of the first-order factor (1 + jcoT) is outlined as follows:
A.2 Bode Plot (Corner Plot) of a Transfer Function / 639

1. Locate the corner frequency co = \\T.

2. Draw the 20 dB/decade (or 6 dB/octave) line and the horizontal
line at dB, with the two lines interesecting at co l/T. =
3. If necessary, the actual magnitude curve is obtained by locating the
points given in Table A-l.

Usually, a smooth curve can be sketched simply by locating the 3-dB

point at the corner frequency and the 1-dB points at 1 octave above and below

the corner frequency.

The phase of G(jco) = 1 + jcoT is written as

/G(jco) = tan"' mT (A-46)

Similar to the magnitude curve, a straight-line approximation can be made for

the phase curve. Since the phase of G(jco) varies from 0° to 90° we may draw a
line from 0° at 1 decade below the corner frequency to +90° at 1 decade above
the corner frequency. As shown in Fig. A-9, the maximum deviation of the
straight-line approximation from the actual curve is less than 6°. Table A-l
gives the values of /l +
ja>T versus coT.

Fig. A-9. Phase versus frequency of the Bode plots of G(s) = 1 + Ts and
G(s) = 1/(1 + Ts).

Simple Pole. 11(1 + jcoT)

When
1
G(Ja>) (A-47)
1 +jcoT
the magnitude, |
G(jco) |
in decibels is given by the negative of the right side of
Eq. (A-42), and the phase /G(jco) is the negative of the angle in Eq. (A-46).
Therefore, it is simple to extend all the analysis for the case of the simple zero
to the Bode plot of Eq. (A-47). We can write

coT < 1 | G(jco) dB |

- dB (A-48)

coT > 1 | G(jco) U= -20 log 10 coT (A-49)

Thus the corner frequency of the Bode plot of Eq. (A-47) is still at co = l/T.
At high frequencies, the slope of the straight-line approximation is +20 dB/
decade. The phase of G(jco) is 0° when co = and —90° when co approaches
640 / Frequency-Domain Plots App. A

infinity. The magnitude and the phase of the Bode of Eq. (A-47) are shown in
Figs. A-8 and A-9, respectively. The data in Tables A-l and A-2 are still useful
for the simple pole case, if appropriate sign changes are made to the numbers.
For instance, at the corner frequency, the error between the straight-line
approximation and the actual magnitude curve is —3 dB.

Quadratic Poles and Zeros

Now consider the second-order transfer function

G(s) =
,* + 2C«,., + a*
(A50)
=
1 + (2£/co„)s + (l/cofis*
We are interested only in the cases when £ <, 1, since otherwise, G(s) would
have two unequal real poles, and the Bode plot can be determined by consider-
ing G(s) as the product of two transfer functions each having a simple pole.
Letting s = jco, Eq. (A- 50) becomes

= (A" 51)
G(j<o)
[1-W1+W.J
The magnitude of G(j(o) in decibels is

2
20 log 10 G(jco) = -20 log 10 ^[l - + 4C
z
(A-52)
1 |

(gJJ (g)
At very low frequencies, co/co„ <C 1 ; Eq. (A-52) may be written as

I
G(jco) dB|
= 20 log 10 1
G(jco) |
=5 -20 log 10 1 = dB (A-53)

Thus the low-frequency asymptote of the magnitude plot of Eq. (A- 50) is a
on the 0-dB axis of the Bode plot coordinates.
straight line that lies
At very high frequencies, eo/con ^> 1 the magnitude in decibels of G(jco) in
;

Eq. (A-50) becomes

\G(jco)U = 20 log 10 \GU<o)\ = -20 log 10 ^/g-)

(A-54)
401og,„f-^\ dB

This equation represents the equation of a straight line with a slope of

—40 dB/decade in the Bode plot coordinates. The intersection of the two asymp-
totes is found by equating Eq. (A-53) with Eq. (A-54), yielding

-401og 10 (^-) = 0dB (A-55)

\co n /
and from which we get
co = co„ (A-56)

Thus the frequency, co =

©„, is considered to be the corner frequency of the
second-order transfer function of Eq. (A-50), with the condition that £ 1. <
The actual magnitude plot of G(jco) in this case may differ strikingly
from the asymptotic lines. The reason for this is that the amplitude and phase
 Bode Plot (Corner Plot) of a Transfer Function / 641
A. 2

curves of the G(ja>) of Eq. (A-50) depend not only on the corner frequency co„,
but also on the damping ratio £. The actual and the asymptotic magnitude plots
of G(jco) are shown in Fig. A-10 for several values of £. The errors between
the

shown A-l for the same set of £s. The standard

two sets of curves are in Fig. 1

40 Mill mi II Mil
1 1 1 1 Mil 1 1 1 1 1 1 I 1 1

30
f = 0.05

20 —
0.5 /
\ / al
« 10
-

3 -

- 10 r = 0.707

-20 -
1.0 -

-30 - -

in II III iiN Mil

-40 I i i l 1 1 1 i i
1
1 1 1 1

0.01 0.1 1.0 10 100

u = cj/w„

Fig. A-10. Magnitude versus frequency of Bode plot of G(s) = 1/[1

+ 2«5/o)„) + (sl<o„Y).

25

20

15

10
a
^ 5
3
5 o

-5

- 10

llll
- 15
I I I I

0.01 0.1

Fig. A-ll. Error in magnitude versus frequency of Bode plot of G(s) =

1/[1 + 2«.rM,) + (s/co„)*].
642 / Frequency-Domain Plots
App. A

procedure of constructing the magnitude portion of the Bode plot of a second-

order transfer function of the form of Eq. (A-50) is to first locate the
corner
frequency then sketch the asymptotic lines; the actual curve is obtained by
co„,

making corrections to the asymptotes by using either the error curves of Fig.
A-ll or the curves in Fig. A-10 for the corresponding
f.
The phase of G(jco) is given by

<aa --"' (^/['-(£)1 (A-57)

and is plotted as shown in Fig. A- 12 for various values of £.

0°
W4; Mil! II III!
—
1 1 1 1
1 1 1 ! 1

f = 0.05
-45° —
t^/0.1
f=1 '°//V*H
-90° _ 0.6// ^ 0.2

0.4 /
135° 0.3 —
180°

1 1 1 1 MM 1 1 1 1
1 1 1 INN 1 1 1 1 MM
0.01 0.1 1.0 100
u = w/w„

Fig. A-12. Phase versus frequency of Bode plot of G(s) = 1/[1 ---
2£(s/«„)
+ (*/«*)*].

The analysis of the Bode plot of the second-order transfer function of

Eq. (A-50) may be applied to a second-order transfer function with two complex
zeros. If

G(s)= 1 +%s co„ (On

1 ,
(A-58)

the Bode plot of G(jco) may be obtained by inverting the curves of Figs. A-10,
A-ll, and A-12.

Example A-3 As an illustrative example of the Bode plot of a transfer function, let
us consider

G(s)
10(5 + 10)
(A-59)
s(s + 2)(j + 5)

The first step is to express the transfer function in the form of Eq. (A-28) and set j
jco. Thus Eq. (A-59) becomes
10(1 +yo.ico)
GUco) = (A-60)
jco{\ +y0.5(U)(l +/0.26t>)

This equation shows that G(jco) has corner frequencies at co = 10, 2, and 5
rad/sec. The pole at the origin gives a magnitude curve that is a straight line with a
slope of —20 dB/decade and passing through the co = 1 rad/sec point on the co axis at
dB. The total Bode plots of the magnitude and phase of G(jco) are obtained by
 A.3 Magnitude-Versus-Phase Plot / 643

«
•o

Fig. A-13. Bode plot of G(s) = [10(j + 10)]/[4f + 2)(j + 5)].

adding the component curves together, point by point, as shown in Fig. A-13. The
actual magnitude curve may be obtained by considering the errors of the asymptotic
curves at the significant frequencies. However, in practice, the accuracy of the asym-
ptotic lines is deemed adequate for transfer functions with only real poles and zeros.

A.3 Magnitude-Versus-Phase Plot

The magnitude-versus-phase diagram is a plot of the magnitude of the transfer

function in decibels versus phase

its in degrees, with co as a parameter on the
curve. One of the most important applications of this plot is that it can be super-
posed on the Nichols chart (see Chapter 9) to give information on the relative
stability and the frequency response of the closed-loop system. When the gain
factor K of the transfer function is varied, the plot is simply raised or lowered
vertically according to the value of K in decibels. However, the unique property
of adding the individual plots for cascaded terms in the Bode plot does not
carry over to this case. Thus the amount of work involved in obtaining the
magnitude-versus-phase plot is equivalent to that of the polar plot, unless a
digital computer is used to generate the data. Usually, the magnitude-versus-
phase plots are obtained by first making the Bode plot, and then transferring the
data to the decibel-versus-phase coordinates.
As an illustrative example, the Bode plot, the polar plot, and the magnitude-
versus-phase plot of the function

G(s) = IOC? + 10)

(A-61)
s(s + 2)(s + 5)
 :

644 / Frequency- Domain Plots

App. A

'

/ Im
Phase
—_^^ G(s)-plane
crossover/^' 3
<5

co -> oo \ Re
~ 1 Jl *
V

v?/ \V^.
^y
'
Gain
crossover 1000
(a)

=
1 1 1
'
co o. \y
30

20
G(/u)

\
/
pa
J*
a CO = . o ^r co = 1 rad/sec
10
3~ \

W Gain crossover
10
co= 10 ^ Phase
crossover
20
to = 5.5
10 ,
*/ ,
i i i i

-270° -220° -180° -140° -90°

/g(/co)

(c)

Fig. A-14. G(s) = ri0(.s + 10)]/[4r + 2)(j + 5)]. (a) Polar plot, (b) Bode
diagram, (c) Magnitude-versus-phase plot.

are sketched as shown in Fig. A-14. The Bode plot shown in Fig. A-14(a) is appar-
ently the easiest one to sketch. The others are obtained by transferring the data
from the Bode plot to the proper coordinates.
The relationships among these three plots are easily established by compar-
ing the curves in Fig. A-14 without the need of detailed explanation. However,
for the purpose of analysis and design, it is convenient to define the following
terms

Gain-crossover frequency. This is the frequency at which the magnitude of

the transfer function G(jco) is unity. In logarithmic scale, this corresponds to
dB. The following interpretations are made with respect to the three types of
plots:
Polar Plot. The gain-crossover point (or points) is where G(jco)
[ |
= 1 [Fig.
A-14(a)].
 3

A. Magnitude-Versus- Phase Plot / 645

Bode Plot. The gain-crossover point (or points) is where the magnitude
curve of G(ja>) crosses the 0-dB axis [Fig. A-14(b)].
Magniwde-Versus-Phase Plot. The gain-crossover point (or points) is where
the G(jco) plot crosses the 0-dB axis [Fig. A- 14(c)].

Phase-crossover frequency. This is the frequency at which the phase of

G(jco) is 180°.

Polar Plot. The phase-crossover point (or points) is where the phase of G(jco)
is 1 80°, or where the G(jco) plot crosses the negative real axis [Fig. A-14(a)].
Bode Plot. The phase-crossover point (or points) is where the phase curve
crosses the 180° axis [Fig. A-14(b)].
Magnitude-Versus- Phase Curve. The phase-crossover point (or points) is where
the G(jco) plot interesects the 180° axis [Fig. A-14(c)].
 APPENDIX B
Laplace Transform Table

Laplace Transform Time Function

F(s) fit)

_1_
u(t) (unit step function)
s

1
t
s*

«!
t" (n = positive integer)

1
e -at
s + a

1 e~<" — e -6 '
Ct + a)(s + b) 6-a
<&n
j2 + 2Zcon s + co„ 2

1 1 <n-1 r -rr
(i + sry r»(«-i)!

7tB„ 2 e-'/r <o„ e-<°>*' sin (eWl - ^t —

TW rW 0)
.

0i n 2 1 - 2£7to„ + '
V(l- C
2
)d - 2£7ca„ -
+ Ts)(s* + 2Zco„s + co„2) *
(1
where * - tan-» ^i." C

646
App. B Laplace Transform Table / 647

Laplace Transform Time Function

F(s) fit)

(On
sin co„t
{s 2 + (On 2 )

(1 +
(On
Ts)(s 2 + C0n 2 )
\
^m"
+ T 2(On
2C r~' IT 1

'

vi+rw
1 lin w
(m t
w
<K\

where = tan con T

<f>
-1

(On 2 vl-( z
s(s 2 + 2Zco„s + a>„ 2 )
where $ = tan -1 ~ ^

C0„ 2
1 — cos (0„t
s(,s 2 + (0„ 2 )

1
1 - e-'/r
s(l + Ts)

1
1 -t-^-e-'i?
i(l + Ts) 2

rw
(On 2
!
1 - 2rCco„ + 7" W e -c<»,( sin (gW — 1 £
2t — 4>)

S(\ + Ts)(.S 2 + 2liC0„S + 2

(On ) Vi - C
2
d - 2crco„ + rw)
where * - tan"'
"/1 " C"
+ tan-»
Tc
?—
Wl ~ ^
—t l rc<»„

s 2 (s 2 +
(On 2
2Cco„s + co„ 2
'
V
i + ^^/r^"'™"'
1

5 "1 ^ 1 f2 '
«
)
where 0-2 tan" 1 v' 1 ~ ^
2

_r _ 2f T*COn 2
r
- 2C0)„r + T 2(On 2
(On 2
00„ 1

1
e- w sin (q,^ i _ ( 2 t _ 0)
5 2 (1 + Ts)(s 2 +
2r.(0„S + (On 2 ) eW(l - C
2
)d - 2i(0nT + T 2co„ 2 )
where - 2 tan"» V1 ~ ^ + r Wl ~^
<f>
-4
tan"' f
1 - TcO n C

1
t - 2T+ (t + 2T)e-<lT
s 2 (l + Ts) 2

^(On + a(On
co„ 2 (l + as)
(On^jl e -e*r gin (aWT^P/ + <f>)

S2 + 2{C0n S + C0„ 2 - acoWl ~^

where * tan"'
1 - af.(o„

co„ 2 (l 4- 05)
(s 2 +
(On 2 )
co„\/l + a 2 co„ 2 sin (co„t + 0) where <6 = tan -1 aa>„
648 / Laplace Transform Table App. B

Laplace Transform Time Function

F(s) fit)

(Qn 1 1 - 2gfra„ + a 2co„ 2 ,„,.,

'l -C 2 \l -
2rcco„ + t 2 cb„ 2
co„Hl + as) -r/T
(1 + Ts)is 2 + 2Z<o„s + co„ 2 )

where # - tan-
"J^g? - tan-. WJgl

(1 +
Q>„ 2 (1
Ts)(s 2
+ a?)
+ o„ 2 )
where
i +
<j>
rw
— tan -1
+
vi + r
a<»„ — tan
-1 a>„T
w ^
w

1 + Vi^T VI -2aCco„ + a 2 co„ 2 e^"-'

(Q„H1 + as)
x sin (m„Vl - 2t
+
t
s(s 2 + 2Cw„s + a>„ 2 )
<t>)

where <j> = tan-i ?^lz_C! _

- tar.-' ^L=£
1 flC(B„ -C

co„ 2 (l + as) 1 + TcQnHa - T) IT _ / 1+aW , , ~

s(l + Ts)(s 2 + co„ 2 )
—
where # = tan -1 acu» tan -1 a>„r

+ 2Coa>B + a 2e»„ 2 e~
1
Vd-{C 2 )(l - 2TCco„ + r 2 <o„ ^

j(1 +
ca„ 2 (l
J*)(i 2 +
+ as)
2r.co„s + co„ 2 )
=
x sin

tan"'
(o^rqp/ + « + r
[aawT^TW - «&»„)] __
^r£ + w e " /r
tf

-tan-.Z^TI^
— l rcc->B
tan- ^2
1 +as - -
s 2 (l + 7i)
/ + (a T)(l e-'/r)

_ e -w sin (oa n vT"- £

2f
+ $)
"»// 1-C
s2 + 2CwB j + <o 2
=
where # tan

COS co„t
S2 + 0)„ 2

=—-t sin <o„t

(s 2 + eo„ 2 ) 2 2c»„

;(C0S 0>„i/ — COS CO„2t)

(s 2 + 2
C0„l ){s
2
+ <o n2 2 ) ou„2" — <»»i

1
+ _ =_J___ _
(1 + 7S)(i 2 + (On 2 ) (1 + 7" 2 a>„ 2

=
)
e-'/r
7
VI + rw ; cos (co n , 0)

where ^ tan -1 <b„T

App. B Laplace Transform Table / 649

Laplace Transform 77me Funtion

F(s) /CO

+ (« r, T2 ) + b ri + Tl2 e-"r
1+ as + bs 2
/
-f
s 2 (l + ri5)(l +T2 5) b-aT2 + T2 2 ,,r,
r, - r2

, . /(l-aCco„-6co„2+26C 2 co» 2 ) 2 + con 2 (l-C 2 X«-26Cco„) 2

V o -
1
2
c )
co„2(l +
+ bs 2 as ) x e -c«»nf sin (av/l — C
2/
+ #)
s(s 2 .+ 2Cco„s + m„ 2 ) where
* - tan- »#,-- ^1",
Ao>„(2C 2+ - aCco„
1) 1
^ tan- ^-C" <*

s2
{s 2 + C0„2)2
=^-(sin ta„t + eo„/ cos <»„f
 APPENDIX O
Lagrange's Multiplier Method

In control system design it is often desirable to maximize or minimize a

function /(xi, x 2 , . .
.
, x„) with the constraint g(x u x 2 , . . . , x„) = g (constant).
Lagrange's method suggests that the desired value of x u x 2 , . . . , x„ can be
found by maximizing or minimizing the function

f=f(xi, x 2 , . . . , x„) + Xg(x u x 2 ,..., x„) (C-l)

where
= Lagrange's multiplier
X

In general, there may be m constraint equations of the form

&(x) = gt° i=\,2,...,m (C-2)

for m< n. The symbol x denotes the vector

x = [x u xz , . . . , x„\ (C-3)

With m constraint equations in Eq. (C-2) to maximize or minimize f(x), F

becomes
F =/(x) + A^,(x) + X 2 g 2 (x) + ... + X mgm (x) (C-4)

where X u X 2 , . . , X m are the Lagrange's multipliers.

The procedure of the Lagrange's method is outlined as follows for the
single-constraint case

1 Form the function

F = f(x) + Xg(x) (C-5)

2. Maximize or minimize F with respect to x u x2 , . . . , x„; that is,

650
App. C Lagrange's Multiplier Method / 651

simultaneously set

|^ = i=l,2,...,« (C-6)
ox,

These n equations give the desired values of x„ i = 1 , 2, . .

. , n, as

functions of A. Let these functions be denoted as

x, = /*,(A) i=\,2,...,n (C-7)

3. Substitute Eq. (C-7) into the constraint equation

g(xi,x 2 x„) = g (C-8)

Equation (C-8) now gives the value of A. Substituting this A into Eq.
(C-7) gives the desired optimal values of x„ i = 1 , 2, . . . , n.

4. Substitution of the optimal values of *,, x2 , . .

.
, x„ obtained from
the last step into/(x) gives the optimal value of/.

The justification of the Lagrange's method of adjoining the constraint to

the function to be minimized is as follows. Let a maximum, minimum, or saddle
point be described as a stationary point. Then the necessary condition for the
existence of a stationary point of a differentiable function f(x u x2 , , x„) is
that

$f- = (C-9)
ox,

Consider a case with just two parameters, Xi and x 2 that \%,f{x x 2 ) is to be ; x ,

maximized or minimized by choosing the proper values for x and x 2 with the t ,

constraint that
g(x u x 2 ) = g„ (constant) (C-10)

The necessary conditions for a stationary point are

#=0
axi
and J^
ax
= (C-ll)
2

Let us assume that x^ and x 2 are arbitrary functions of a dependent vari-

is, /and g can be written as/(x,, x 2 y) and g(x u x 2
able y; that , , y), respectively.
Then, we can form the following equations

df = _^idxJ ,^Ldx 2 (C-12)

dy dxi dy dx 2 dy
dg = dj^dx^ .dg^dxj
(C-13)
dy dxi dy dx 2 dy
The necessary conditions for a stationary point, stated in Eq. (C-ll), imply
that the total derivatives in Eqs. (C-12) and (C-13) must vanish at the stationary
point. Therefore,
652 / Lagrange's Multiplier Method App. C

Let us multiply the right side of Eq. (C-15) by an undetermined parameter

X and then add the result to Eq. (C-14). We have

If now X is so chosen that

& + *&-° (C- 17)

£+*£-° < c- 18 >

and
g(x ,x 2 )
i
= g (C-19)

then the necessary condition of a stationary point of/(x,, x 2 ) is satisfied. Thus

Eqs. (C-17) and (C-18) imply the maximization or minimization of F = f+ Xg
with respect to x, and x 2 .
Index

Block diagram (contd.)

feedback control system, 61
Absolute stability, 316, 365 multivariable systems, 62
Acceleration error constant, 266 Bode diagram, 322, 482, 538
AC control system, 12, 218 Bode plot, 519, 626, 633
Actuating signal, 1, 3 Bridged-T network, 558
Adaptive control systems, 57 Bridged-T network compensation, 557
Analog computer elements, 126
Analytical design, 574
Analytic function, 17
Asymptotic stability, 317, 365 Calculus of variation, 575
Auxiliary equation, 327 Caley-Hamilton's theorem, 146
Canonical form, 119
Canonical state equation, 608
B Cascade decomposition, 139
Cause-and-effect, 7
Back emf constant, 220, 223, 286 Characteristic equation, 117,275,317,
Backlash, 200 321, 322, 360, 376, 587, 617
Bandwidth, 463 Closed-loop control system, 3
second-order sy terns, 466, 470 Complex plane, 15
Bandwidth versus damping ratio curve, Complex variable, 15
467 functions of, 16
Bellman's equation, 604 Conditional frequency, 275
Block diagram, 58 Constant conditional frequency loci,
control systems, 59 277

653
654 / Index

Constant damping factor loci, 277 Decomposition of transfer function,

Constant-damping-ratio loci, 277, 451 136
Constant M loci, 485 cascade, 139
Constant natural undamped frequency direct, 137
loci, 277 discrete transfer function, 169
Constant N loci, 489 parallel, 139
Control system: Definite integral, table, 578
ac, 12 Delay time, 272, 282
classification, 11 Derivative control, 295
closed-loop, 3 Design of control systems, 504
continuous-data, 12 Diagonalization of a matrix, 118, 148
dc, 12 Digital controller, 14
digital, 13 Digital control systems, 13
open-loop, 2 Direct decomposition, 137
sampled-data, 13 Discrete-data control system, 13, 161,
Controllability, 109, 144, 585, 587 447
definition, 145, 148 Discrete state equation, 163
general concept, 144 z- transform solution, 165
output, 151 Discrete state transition equation, 164
state, 145 Discrete transfer function:
Controlled process, 1 decomposition, 169
Controlled variable, 1 Disturbance, 9
Convolution integral, 57, 104, 267 Dynamic equations, 97, 588
Corner frequency, 637 Dynamic systems, 95
Corner plot (see Bode plot)
Coulomb friction, 192
coefficient, 193, 199
Criticaldamping, 275
CSMP, 132 Edge-guide control system, 237
Cutoff rate, 463 Eigenvalues, 117, 118, 142, 585
Eigenvectors, 117, 118, 143
Electrical networks, 188
D Encircled, 333
Enclosed, 334
Damping constant (see Damping Energy, 196
factor) Equilibrium state, 317
Damping factor, 276 Error coefficients, 268
Damping ratio, 275 Error-sensing device, 60, 208, 285
DC control system, 12, 210 Error series, 267
DC motor, 220, 236 Error transfer function, 267
armature-controlled, 222, 248, 286
field-controlled, 220
permanent-magnet, 210
torque-speed curves, 225
Dead zone, 200 Feedback:
Decibel, 635 effect on external disturbance, 9
 Index / 65S

Feedback (contd.) Initial- value theorem, 21

effect on overall gain, 7 Input node, 67

effect on sensitivity, 8 Input vector, 54, 99

effect on stability, 8 Integral control, 300
effects of, 6 Integral square control, 574
Feedback control system, 3 Integral square error, 574
basic elements, 5 Inverse problem, 616
block diagram, 3, 5 Inverse z-transform, 43

types, 11 inversion formula, 44

Force-mass system, 191

Force-spring system, 191
FORTRAN, 132
Free-body diagram, 203
Frequency-domain analysis, 459 Jordan blocks, 123, 149
Frequency-domain plots, 626 Jordan canonical form, 123, 142, 150
Friction, 191
Coulomb, 192
rotational motion, 195
static, 192
viscous, 192 Lag-lead compensation, 552
Lag-lead network, 552
Lagrange multiplier, 575, 583, 649
Lagrange's multiplier method, 649
Laplace transform:
application to solving differential
Gain-crossover frequency, 478, 644
equations, 25
Gain margin, 473, 476
definition, 18
Gear train, 198
inverse, 19
Generalized error coefficients (see
table, 646
Error series)
theorems, 19
Laplace transform table, 646
Levers, 198, 200
H Linearization of nonlinear state
equations, 158
Hamilton-Jacobi equation, 601, 602, Linear regulator design, 573, 599
604 infinite-time, 607
Hessian matrix, 584 Linear spring, 191
Hurwitz determinants, 323 Linear systems, 11
Logarithmic plot, (see Bode plot)
Lyapunov's stability criterion, 322
I

Impulse response, 52, 55 M

Impulse response matrix, 55
Indefinite matrix, 38 Magnification curves, 465, 468, 469,
Inertia, 193 472
6S6 / Index

Magnitude-versus-phase plot, 626, N

643
Mass, 190 Natural undamped frequency, 275
Mass-spring-friction system, 204 Negative definite matrix, 38
Mathematical modeling of physical Newton's law of motion, 203
systems, 187 Nichols chart, 490, 493
Matrix: Noise signal, 9
addition, 32 Nonlinear control system, 11, 160, 363
adjoint of, 3 Nonlinear state equation, 158
algebra, 32 Nyquist criterion, 322, 330
associate law, 33 Nyquist path, 340
column, 28
commutative law, 33
conjugate, 31 O
definiteness, 38
definition, 27 Observability, 152, 589
determinant of, 29 definition, 153, 155
diagonal, 28 Observer (see State observer)
diagonalization of, 118 Octave, 635
distributive law, 35 Open-loop control system, 2
inverse of, 35 block diagram, 3
multiplication, 33 Operational amplifier, 246
nonsingular, 29 Optimal control, 572
null, 29 Output controllability, 151
order of, 28 Output equation, 97
quadrative form, 37 Output node, 67
rank of, 36 Output vector, 54, 99
row, 28
singular, 29
skew- symmetric, 30
square, 29
subtraction, 33 Parabolic input, 261
symmetric, 29 Parallel decomposition, 139
transpose of, 30 Parameter optimization, 583
unity, 28 Parseval's theorem, 575
Maximum overshoot, 272, 281 Partial-fraction expansion, 21-25, 43
Mechanical energy (see Energy) Partial state feedback, 615
Mechanical power (see Power) Peak resonance, 463
Mechanical system elements, 190 Performance index, 273, 509, 572
Mechanical systems, 203 Phase-crossover frequency, 476, 645
Minimum-time problem, 509, 573 Phase-lag compensation, 512, 535
Modal form transformation, 141 Phase-lead compensation, 512, 515
Multiloop feedback control system, Phase margin, 473, 477, 478
356 Phase-variable canonical form, 109,
Multivariable system, 2, 53 586, 617
 Index / 657

Polar plot, 444, 482, 488, 626, 627 Root loci (contd.)
Poles, definition, 17 construction, 380
Pole- zero cancellation design, 563 discrete-data control system, 447
Popov's stability criterion, 363 rules of construction, 411
Positional control system, 284, 298, time delay systems, 434
303 Root locus diagram, 290, 322, 375,
Positional error constant, 264 528
Positive definite matrix, 38 Rotary-to-linear motion, 195
Positive semidefinite matrix, 38 Rotational motion, 193
Potentiometer, 209 Routh array, 324
Power, 196 Routh-Hurwitz criterion, 322
Principle of argument, 335 Routh tabulation (see Routh array)
Principle of optimality, 601 Rudder control system, 4

Quadratic forms, 37
Sample-and-hold, 162
Sampled-data control system, 13, 171
R (see also Discrete-data control
systems)
Ramp input, 261 Sampler:
Rate feedback (see Tachometer 40
finite-pulse width,
feedback) ideal,40
RC Phase-lag network, 536 Sampling duration, 39
RC Phase-lead network, 516 Sampling period, 39
Relative stability, 316, 473, 483 Saturation nonlinearity, 160, 364, 366
Relay nonlinearity, 364 Sensitivity, 7, 497
Resonant frequency, 463 Sensitivity function, 9, 499
Resonant frequency versus damping Servoamplifier, 246
ratio curve, 466 Settling time, 272, 284
Riccati equation, 604 Signal flow graphs, 51, 64, 128
algebraic, 610 algebra, 69
differential, 604 basic properties, 66
solution of, 607, 610 construction of, 71
Riccati gain matrix, 606 definitions, 67
Rise time, 272, 282, 283 gain formula, 75
Root contour, 424, 529, 547 loop, 69
Root loci, 290, 375 path, 68
asymptotes, 384, 386 Signal-to-noise ratio, 9
basic conditions, 376 Similarity transformation, 118, 148
breakaway points, 392, 436 Single- valued function, 16
complementary, 376 Singularity, 17
complete, 376 Spectral factorization, 577
 658 / Index

Stability, 8, 316 Synchro control transformer, 213,215

absolute, 316 Synchro transmitter, 213
asymptotic, 317
definition, 317
multiloop systems, 356
nonlinear systems, 363
relative, 316 Tachometer, 219, 246
time delay systems, 360 Tachometer feedback, 302
zero-input, 317 Tension control system, 5, 235
Stabilizability, 587 Time delay systems, 360, 434
State concept, 95 Time-domain analysis, 259
State controllability, 112, 145-146 Time-invariant system, 12
State diagram, 126, 274 Time-optimal problem, 509
discrete-data systems, 167 Time-varying state equation, 173
sampled-data systems, 171 solution, 175
State equations, 96, 97, 173, 460 Time- varying systems, 12, 173
canonical form, 1 19 discrete approximation, 175
discrete-data systems, 161 Timing belt, 198, 200
electrical networks, 188 Torque constant, 221
matrix representation, 99 Torque-inertia system, 194
State feedback (see State-variable Torsional spring, 194
feedback) Tracking problem, 573
State observer, 588 Transfer function, 51
closed-loop control, 595 decomposition of, 136
State transition equation, 103, 133,317 discrete-data systems, 81
State transition matrix, 101, 164, 174, linear systems, 51
317 multi variable systems, 62
properties, 102 Transfer function matrix, 54, 460
significance of, 102 Transient response, 259, 271
State-variable feedback, 305 second-order system, 273, 280
State variables, 95 Translational motion, 190
definition, 96 Transportation lags, 242
State vector, 99 Two-phase induction motor, 225
Static friction, 192
coefficient, 192
Steady-state error, 260, 262, 510 U
Steady-state response, 259
Step input, 260 Unit step function, 19, 26
Step motors, 228 Unit step response, 272, 294, 307,
torque curve, 232 470, 471
variable-reluctance type, 228
Sun-Seeker system, 243, 520, 539
Suppressed-carrier modulated signal,
211
Suppressed-carrier modulation, 212 Vandermonde matrix, 120, 157
Synchro, 213 Velocity error constant, 265
 Index / 659

Viscous friction, 192 zero-input stability, 317

coefficient, 192, 199 zero-order hold, 171 (see also
Sample-and-hold)
zeros, definition, 17
W z-transfer function, 83
z-transform, 15, 39
Weighting function, 55 definition, 41
Weighting sequence, 82 final-value theorem, 46
initial-value theorem, 46
inverse, 43
table, 43
theorems, 45
zero-input response, 317 z-transform method, 165
 KUO

PRLN iud
I

0-13-054973-8 HALL