Sie sind auf Seite 1von 379

Library of Congress Cataloging-in-PublicationData

Ogata, Katsuhiko .
Discrete-time control systems / Katsuhiko Ogata. - 2nd ed.
p. cm.
Hncludes bibliographical references and inciex.
HSBN 0-13-034281-5
1. Discrete-time systems. 2. Control theory. H. IEltle.
A402.04 1994 94-19896
629,8'Sdc20 CHP

Editoriallproduction supervision: Lynda GriEitithsiT


Cover design: Karen Salzbach
roduction coordinator: David Dickeyl

O 1995, 1987 by Prentics-Hall, Inc.


Upper Saddle River, New Jersey 07458

All rights reserved. No part of this book may be


reproduced, in any form or by any means, 1-1 INTRODUCYION, 1
without permission in writing from the pubiisher. 9-2 DIGITAL CONTROL SY
1-3 QUANTIZING AND QUANTIZATIOI\I ERROR, 8
1-4 DATA ACQUISITION, CONVERSION, AND DISTRIBUTION SYSTEMS, 1 1
1-5 CONCLUDING COMMENTS, 20

Frinted in the Wnited States of America


1 0 9 8 7

I S B N D-23-034281-5 2-1 INTRODUCTION, 23


2-2 THE z TRANSFORM, 24
2-3 z TRANSFORMS OF ELEMENTARY FUNCTIONS, 25
2-4 IMPORTANT PROPERTIES AND THEOREMS OF THE z TRANSFORM, 31
Prentice-Hall InternationaI (UK) Limited, London 2-5 THE INVERSE z TRANSFORM, 37
Prentice-Hall of Australia Pty. Limited, Sydney
2-6 z TRANSFORM METHOD FOR SOLVING DIFFERENCE EQUATIONS, 52
Prentice-Hall Canada Inc., Toronto
Prentice-Hall Hispanoamericana, S.A., Mexico 2-7 CONCLUDING COMMENTS, 54
Prentice-Hall of India Private Limited, New Delhi EXAMPLE PROBLEMS AND SOLUTIONS, 55
Prentice-Hall of Japan, linc., Tokyo PROBLEMS, 70
Prentbce-Hall Asia Re. Ltd., Siizgapore
Editora Prentice-Eall do Brasil, Etda., Rio de Sanerio
ontents

6-3 OBSERVABILITY, 388


6-4 USEFUL TRANSFORMATIONS IN STATE-SPACE ANALYSIS AND DESIGN, 396
6-5 DESIGN VIA POLE PLACEMENT, 402
6-6 STATE OBSERVERS, 421
3-1 INTRODUCTION, 74 6-7 SERVO SYSTEMS, 460
3-2 IMPULSE SAMPLING AND DATA HOLD, 75 EXAMPLE PROBLEMS AND SOLUTIONS, 474
3-3 OBTAININGTHEzTRANSFORM BY THE CONVOLUTION INTEGRAL METHOD, 83 PROBLEMS, 510
3-4 RECONSTRUCTING ORIGINAL SIGNALS FROM SAMPLED SIGNALS, 90
3-5 THE PULSE TRANSFER FUNCTION, 98
3-6 REALlZATlON OF DlGlTAL CONTROLLERS AND DlGlTAL FILTERS, 122
EXAMPLE PROBLEMS AND SOLUTIONS, 138
PROBLEMS, 166

7-1 INTRODUCTION, 517


7-2 DIOPHANTINE EQUATION, 518
7-3 ILLUSTRATIVE EXAMPLE, 522
7-4 POLYNOMIAL EQUATIONS APPROACH TO CONTROL SYSTEMS DESIGN, 525
4-1 INTRODUCTION, 173 7-5 DESIGN OF MODEL MATCHING CONTROL SYSTEMS, 532
4-2 MAPPING BETWEEN THE s PLANE AND THE z PLANE, 174 EXAMPLE PROBLEMS AND SOLUTIONS, 540
4-3 STABlLlTY ANALYSIS OF CLOSED-LOOP SYSTEMS IN THE z PLANE, 182 PROBLEMS, 562
4-4 TRANSIENT AND STEADY-STATE RESPONSE ANALYSIS, 193
4-5 DESIGN BASED ON THE ROOT-LOCUS METHOD, 204
4-6 DESIGN BASED ON THE FREQUENCY-RESPONSEMETHOD, 225
4-7 ANALYTICAL DESIGN METHOD, 242
EXAMPLE PROBLEMS AND SOLUTIONS, 257
PROBLEMS, 288
8-1 INTRODUCTION, 566
8-2 QUADRATIC OPTIMAL CONTROL, 569
8-3 STEADY-STATE QUADRATIC OPTIMAL CONTROL, 587
8-4 QUADRATIC OPTIMAL CONTROL OF A SERVO SYSTEM, 596
EXAMPLE PROBLEMS AND SOLUTIONS, 609
PROBLEMS, 629
INTRODUCTION, 293
STATE-SPACE REPRESENIATIONS OF DISCRETE-TIME SYSTEMS, 297
SOLVING DISCRETE-TIME STATE-SPACE EQUATIONS, 302
PULSE-TRANSFER-FUNCTIONMATRIX, 310
DlSCRETlZATlON OF CONTINUOUS-TIMESTATE-SPACE EQUATIONS, 312
LIAPUNOV STABlLlTY ANALYSIS, 321
EXAMPLE PROBLEMS AND SOLUTIONS, 336 A-1 DEFINITIONS, 633
PROBLEMS, 370 A-2 DETERMINANTS, 633
A-3 INVERSION OF MATRICES, 635
A-4 RULES OF MATRIX OPERATIONS, 637
A-5 VECTORS AND VECTOR ANALYSIS, 643
A-6 EIGENVALUES, EIGENVECTORS, AND SlMlLARlTY TRANSFORMATION, 649
A-7 QUADRATIC FORMS, 659
6-1 INTRODUCTION, 377 A-U PSEU DO1NVERSES, 663
6-2 CONTROLLABILITY, 378 EXAMPLE PROBLEMS AND SOLUTIONS, 666
8-1 INTRODUCTION, 681
8-2 USEFUL THEOREMS OF THE z TRANSFORM THEORY, 681
8-3 INVERSE z TRANSFORMATION AND INVERSION INTEGRAL METHOD, 686
8-4 MODIFIED zTRANSFORM METHOD, 691
EXAMPLE PROBLEMS AND SOLUTIONS, 697

C-9 INTRODUCTION, 304


C-2 PRELIMINARY DISCUSSIONS, 704
C-3 POLE PLACEMENT DESIGN, 707
EXAMPLE PROBLEMS AND SOLUTIONS, 718
X Preface Preface

publisher. This book can also serve as a self-study book for practicing engineers w
discrete-time control theory by themselves.

tive cornrnents absut the material ín this book.


recent years there has been a rapid increase in the gital controllers in
ontrol systems. Digital controls are used for achieving performance-f~r
example, in the form of maximum productivity, maximu t 9minimum ~ 0 % or
minimum energy use.
recently, the application of cornputer control has made possible "intelli-
on in industrial robots, the optimization of fue1 economy '
and refinements in the operation of household appliances and m
rnicrowave ovens and sewing machines, among others. Decision-m
and flexibility in the control program are major advantages of digital control systerns.
The current trend toward digital rather than analog n t r d of dynamic systems
is mainly due to the availability of low-cost digital com ters and the advantages
found in working with digital signals rather than continuous-time signals.
continuous-time signal is a signal defined over a contin-
uous plitude may assume a continuous range of values or may
assume only a finite number of distinct values. The process o£representing a variable
by a set of distinct values is called quantization, and the resulting distinct values are
called quantized values. The quantized variable changes only by a set of distinct
steps.
n analog signal is a signal defined over a continuous range of time whose am-
plitude can assume a continuous range of values. Figure 1-l(a) shows a continuous-
time analog signal, and Figure 1-l(b) shows a continuous-time quantized signal
(quantized in amplitude only).
lntroduction to Bíscrete-Time Control Systems Chap. 1 ec. 1-1 introdlaction

numbers. (In practice, many digital signals are obtained by sampling analog signals
and then quantizing them; it is the quantization that allows these analog signals to
be read as finite binary words.) Figure 1-l(d) depicts a digital signal. Clearly, it is
a signal quantized both in amplitude and in time. The use of the
requires quantization of signals both in amplitude and in time.
The term "discrete-time signal" is broader than the term "digital signal" or the
term "sampled-data signal." 1can refer either to a digital
signal or to a sampled-data signal. "discrete time7' and
""dgital" are often interchan
used in theoretical study, wh
ware or software realizations.
In control engineering,
ysical plant or process or a nonphysical process such as an economic process.
t plants and processes involve continuous-time signals; therefore, if digital
controllers are involved in the control systems, signal conversions (analog to dibital
and digital to analog) become necess y. Standard techniques are available for such
signal conversions; we shall discuss em in Section 1-4.
Loosely speaking, terminologies such as discrete-time con
pled-data control systems, and digital control systems imply the
similar types of control systems. Precisely speaking, there are, of
in these systems. For example, in a sa -data control system both continuous-
time and discrete-time signals exist in the system; the discrete-time signals are
amplitude-modulated pulse signals. Digital control systems may include both contin-
us-time and discrete-time signals; here, the latter are in a numerically coded form.
th sampled-data control systems and digital control systems are discrete-time

any industrial control systems include continuous-time signals, sampled-data


signals, and digital signals. Therefore, in this book we use the term "discrete-time
control systems" to describe the control systems that include sorne forms of sarnpled-
Figwe 1-1 (a) Continuous-time analog data signals (amplitude-modulated pulse signals) and/or digital signals (signals in
signal; (b) continuous-time quantized numerically coded form).
signal; (c) sampled-data signal;
(d) digital signal.
. The discrete-time control systems consid-
ered in this book are mostly linear and time invariant, although nonlinear and/or
Notice that the anaiog signal is a special case of the continuous-time signal. In time-varying systems are occasionally included in discussions. A linear system is one
practice, however, we frequently use the terminology "continuous-time" in lieu of in which the principie of superposition applies. Thus, if y, is the response of the
"analog." Thus in the literature, including this book, the terms "continuous-time system to input xl and y2 the response to input x2, then the system is linear if and
signal" and "analog signal" are frequently interchanged, although strictly speaking only if, for every scalar a and P , the response to input al + Px2 is cuyl + ,By2.
they are not quite synonymous. A linear systern may be described by linear differential or linear difference
A discrete-time signal is a signal defined only at discrete instants of time (that equations. A time-invariant linear system is one in which the coefficients in the
is, one in which the independent variable t is quantized). n a diXrete-h-le signal9 differential equation or difference equation do not vary with time, that is, one in
if the amplitude can assume a continuous range of values, then the signal is called which the properties of the system do not change with time.
a sampled-data sigizal. A sampled-data signal can be generated by sampling an
analog signal at discrete instants of time. It is an amplitude-modulated pulse signal.
Figure 1-1(c) shows a sarnpled-data signal. Discrete-time control systems are control systems in which one or more variables can
A digital signal i:, a dlscrete-time signal with quantized arnplitude. Such a signal change only at discrete instants of time. These instants, which we shall denote by
can be represented by a sequence of nurnbers, for example, in t kTor tk (k = 091,2,. . . ), may specify the times at which some physical rneasurement
lntroduction to Discrete-Time Control Ystems Chap. 1 Sec. 1-2 Digital Control Cystems

is performed or the times at which the memory of a digital computer is read out. The controllers, a thorough knowledge of them is highly valuable in designing discrete-
time interval between two discrete instants is taken to be sufficiently short that the time control systems.
data for the time between them can be approximated by simple interpolation.
Discrete-time control systems differ from continuous-time control systems in
that signals for a discrete-time control system are in sampled-data form or in digital
form. If a digital computer is involved in a control system as a digital controller, any
sampled data must be converted into digital data. Egure 1-2 depicts a block diagram of a
Continuous-time systems, whose signals are continuous in time, may be de- tion of the basic control scheme. The sy
scribed by differential equations. Discrete-time systems, which involve sample feedforward control. In designing such
data signals or digital signals and possibly continuous-time signals as well, may be "goadness" of the control system depe
described by difference equations after the appropriate discretization of continuous- choose an appropriate performance index for a given case and design a controller
time signals. so as to optimize the chosen performance index.
esses. The sarnpling of a continuous-time signal replaces the .
Figure 1-3 shows a block diagram
origi -time signal by a sequence of values at discrete time points. A of the system are shown by the blocks.
sampling process is used whenever a control system involves a digital controller, ñhe controller operation is controlled by the clock. In such a digital control system,
since a sampling operation and quantization are necessary to enter data into such some points of the system pass signals of varying amplitude in either continuous time
a controller. Also, a sampling process occurs whenever measurements necessary for or discrete time, while other points pass signals in numerical code, as de
control are obtained in an intermittent fashion. For example, in a radar tracking
system, as the radar antenna rotates, information about azimuth and elevation is of the plant is a continuous-time signal. The error signal is con-
obtained once for each revolutio of the antenna. Thus, the scanning operation of 1 form by the sample -hold circuit and the analog-to-digital
the radar produces sampled data n another example, a sampling process is needed converter. The conversion is done at sampling time. The digital computer
whenever a large-scale controller or computer is time-shared by several plants in
order to save cost. Then a control signal is sent out to each plant only periodically
and thus the signal becomes a sampled-data signal.
The sampling process is usually followed by a quantization process. In the
quantiration process the sampled analog amplitude is replaced by a digital ampli-
tude (represented by a binary number). Then the digital signal is processed by the
computer. The output of the cornputer is sampled and fed to a hold circuit. The
output of the hold circuit is a continuous-time signal and is fed to the actuator.
shall present details of such signal-processing methods in the digital controlle
Section 1-4.
The term "discretization," rather than "sampling," is frequently used in the
analysis of multiple-input-multiple-output systems, although both mean basically

tant to note that occasionally the sampling operation or discretiza-


l i
1
fictitious and has been introduced only to simplify the analysis of Clock

control systems that actually contain only continuous-time signals. In fact, we often L _I
use a suitable discrete-time model for a continuous-time system. An example is a Digital controlier

digital-computer simulation of a continuous-time system. Such a digital-computer-


simulated system can be analyzed to yield parameters that will optimize a given
performance index.
ost of the material presented in this book deals with control systems that can
be modeled as linear time-invariant discrete-time systems. It is im
Noise
that many digital control systems are based on continuous-time
Since a wealth of experience has been accumulated in the design of continuous-time Figure 1-2 Block diagram of a digital control system.
lntroduction to Discrete-Time Control Cysterns Chap. ? igital Control Systerns 7

Analog-to-Digital Gonverter (AlD). An analog-to-digital converter, also


called an encoder, is a device that converts an analog signal into a digital signal,
usually a numerically coded signal. Such a converter is needed as an interface
between an analog component and a digital com onent. A sample-an
is often an integral part of a commercialliy available A/D converter. The conversion
of an analog signal into the corresponding digital signal (binary number) is an
ation, because the analog signal ke on an infinite number of values,
he variety of different nurnbers a finite set of digits
This approximation process is c ore on quantization
is presented in Section 1-3.)
Digital-to-Analog Converter (DIA). A digital-to-analog converter, also
called a decoder, is a device that converts a digital signal (numerically co
ipre 1-3 Block diagram of a digital control system showing signals in binary or graphic form. into an analog signal. Such a converter is needed as an interface between a digital
component and an analog component.
Plant or Process. A plant is any ical object to be controlled. Examples
are a furnace, a chemical reactor, and of machine parts functioning together
rocesses the sequences of numbers by means of an algorithm and produces new
to perform a particular operation, such as a servo system or a spacecraft.
quences of numbers. At every sampling instant a coded number (usually a binary
) must be converted to a physical A process is generally defined as a progressive operation or develop
number consisting of eight or more binary d
e Or anal% signal- The digital-to- marked by a series of gradual changes that succeed one another in a relatively
control signal, which is usually a continuous
sevence ~f numbers in numerical fixed way and lead toward a particular result or end. In this book we cal1 any
analog converter and the hold circuit convert
operation to be controlled a process. Examples are chemical, economic, and biolog-
code into a piecewise continuous-time signal. The real-time clock in the cornputer
ical processes.
synchronizes the events. The output of the hold circuit, a continuous-time signal, is
The most difficult part in the design of control systems may lie in the accurate
fed to the plant, either directly or through the actuator, to control its
modeling of a physical plant or process. There are many approaches to the plant or
The operation that transforms continuous-time signals into discrete-time
process model, but, even so, a difficulty may exist, mainly because of the absence
data is called sampling or discretizaíion. The reverse operation, the operation that
of precise process dynamics and the pr of poorly defined random parameters
transforms discrete-time data into a continuous-time signal, is called data-hold; it
amounts to a reconstruction of a continuous-time signal from the sequence of
in many physical plants or processes. , in designing a digital controller, it is
necessary to recognize the fact that the mathematical model of a plant or process
discrete-time data. It is usually done using one of the many extrapolation techniques.
in many cases is onIy an approximation of the physical one. Exceptions are found
Pn many cases it is ne by keeping the signal constant between the successive
in the modeling of electromechanical systems and hydraulic-mechanical systems,
sampling instants. ( shall discuss such extrapolation techniques in Section 1-4.)
since these may be modeled accurately. For example, the modeling of a robot arm
The sample-and-hold (SIH) circuit and analog-to-digital (AID) converter con-
system may be accompIished with great accuracy.
vert the continuous-time signal into a sequence of numerically coded binary words.
Such an AID conversion process is called coding or encoding. The combination of Transducer. A transducer is a device that converts an input signal into an
the SIH circuit and analog-to-digital converter may be visualized as a switch that output signal of another form, such as a device that converts a pressure signal into
closes instantaneously at every time interval T and generates a sequence of numbers a voltage output. The output signal, in general, depends on the past history of the
in numerical code. The digital cornputer operates on such numbers in numerical code input.
and generates a desired sequence of numbers in numerical code. The digital-to- Transducers may be classified as analog transducers, sampled-data transduc-
analog (DIA) conversion process is called decoding. ers, or digital transducers. An analog transducer is a transducer in which the input
and output signals are continuous functions of time. The magnitudes of these signals
efore we discuss digital control systems in detail, we may be any values within the physical limitations of the system. A sampled-data
need to define some of the terms that appear in the block diagram of Figure 1-3. transducer is one in which the input and output signals occur only at discrete instants
Sample-and-Hold (SIH). "Sarnple-and-hold" is a general term used for a of time (usually periodic), but the magnitudes of the signals, as in the case of the
sarnple-and-hold amplifier. It describes a circuit that receives an analog input signal analog transducer, are unquantized. A igital transducer is one in which the input
and holds this signal at a constant value for a specified period of time. Usually the and output signals occur only at discrete instants of time and the signal magnitudes
signal is elebrical, but other forrns are possible, such as optical and mechanical. are quantized (that is, they can assume only certain discrete levels).
lntroductisn to Discrete-Time Control ec. 1-3 Quantizing and Quantization

ns, As stated earlier, a si


discrete-time signal. A sa
in transforming a continuous-time g*al into a dkxete- where the FSR is the full-scale range. Note that the st bit of the natural binary
There are several different t es of sampling operations of practica1 impar- as the most weight (one-half of the full scale) he most significant
tance : SB). The rightmost bit has the least weight (112" t e full scale) and is
called the least significant bit (LS
n this case, the campling instants are equally spaced, or
tk = kT(k = O , 1 , 2 , . . . ). eriodic sampling is the II-lost conventional tYPe of
sampling operation.
Multiple-order sampling. The pattern of the tk's is repeated The least significant bit is t e quantization leve1 Q .
is, tk+r- tk is constant for al1 k .
Error. Since the nu igital word is finite, AID
ltiple-rate sampling. In a control system havi?
stant involved in one loop may be quite in a finite resolution. output can assume only
a finite number of levels, and therefore an analog number must be rounded off to
ence, it may be advisable to sample slow
the nearest digital level. Menee, any AID conversion involves quantization error.
time constant, while in a loop involving only small time constants the sampling
rate must be fast. Thus, a digital control system may have different sampling Such quantization error varies between O and t $ Q S error depends on the
fineness of the quantization level and can be made as s
riods in different feedback paths or may have multiple sam
quantization level smaller (that is, by increasing the
ndorn samplirzg. In this case, the sampling instaiats are ra there is a maximum for the number of bits n , and so there is always some error due
random variable. to quantization. The uncertainty present in the quantization process is called quan-
tization noise .
n this book we shall treat only the case where the sampling is periodic. To determine the desired size of the quantization level (or the number of output
states) in a given digital control system, the engineer must have a good understanding
of the relationship between the size of the quantization level and the resulting error.
The variance of the quantization noise is an important measure of quantization error,
since the variance is proportional to the average power associated with the noise.
The main functions involved in analog-to-digital conversion are sampling, amplitude Figure 1-4(a) shows a block diagram of a quantizer together with its input-
quantizing, and coding. hen the v a h e of any sample falls between two adjacent output characteristics. For an analog input x(t), the output y (t) takes on only a finite
"'perrnitted" output stat it must be read as the permitted output state nearest the number of levels, which are integral multiples of the quantization level Q .
actual value of the signal. The rOWss of representing a ~ontinuousor anal% sigrial In numerical analysis the error resulting from neglecting the remaining digits
by a finite number of discrete states is called amplitude quantization. That is, is called the round-off error. Since the quantizing process is an approximating
"quantizing" means transforming a continuous or analog signal into a set of process in that the analog quantity is approximated by a finite digital number, the
states. (Note that quantizing occurs whenever a physical quantity is represented quantization error is a round-off error. Clearly, the finer the quantization level is,
numerically. ) the smaller the round-off error.
The output state of each quantized sample is then described by a numerical Figure 1-4(b) shows an analog input x(t) and the discrete output y(t), which
code. The process of representing a sample value by a numerical code (such as a is in the form of a staircase function. The quantization error e(t) is the difference
binary code) is called encoding or coding. Thus, encoding is a process of assigning between the input signal and the quantized output, or
a digital word or code to each discrete state. The sampling period and quantizing
levels affect the performance of digital control systems. So they must be determined
carefully . Note that the magnitude of the quantized error is
. The standard number system used for processing digital signals
is the binary number system. In this system the code group consists of n pulses each
indicating either "on" (1) or "off" (O). In the case of quantizing, n "on-off" pulses For a small quantization leve1 ,the nature of the quantization error is similar
can represent 2" amplitude levels or output states. to that of random noise. nd, in effect, the quantization process acts as a source of
The quantization level Q is defined as the range between two adjacent decision random noise. In what follows we shall obtain the variance of the qu
points and is given by Scach variance can be obtained in terms of the quantization level
lntroduction to Diccrete-Time Ontrol Chala. 1 Sec. 1-4 Dala Acquicition, Conversion, and

signar e(t) may be plotted as shown in Figure 1-4(c). The average value of e(t) is zem,
or e(t) = O. Then the variance u h f the quantization noise is

n level Q is small co pared with the average amplitude o£


variance of the quantization noise is seen to be one-twelfth
of the square of the quantization level.

ith the rapid growth in the use of digital computers to perform digital control
actions, both the data-ac uisition system and the distribution system have become
an important part of the entire control system.
The signal conversion that takes place in the digital control system involves t
following operations:

ultiplexing and demultiplexing


. Sample and hold
Analog-to-digital conversion (quantizing and encoding)
. Digital-to-analog conversion (decodlng)
Figure 1-5(a) shows a block diagram o£a acquisition system, and Figure 1-5(b)
shows a block diagram of a data-distrib
n the data-acquisition system the input to the system is a physical variable such
as position, velocity, acceleration, temperature, or pressure. uch a physical variable
is first converted into an electrical signal (a voltage or current signal) by a suitable

Physical T o digital
variable controller

(a) Block diagram of a quantizer and its input-output characteristics;


(b) analog input x ( t ) and discrete output y ( t ) ; (c) probability distribution P(e) of
quantization error e(f). From digital
To actuator
controller

Suppose that the quantization level Q is small and we assume that the quan-
tization error e(t) is distributed uniformly between -4 Se
that this error
acts as a white noise. [This is obviously a rather rough assumption. However, since
the quantization error signal e(t) is o£ a small amplitude, such an assumption may Figure 1-5 (a) Block diagram of a data-acquisition system; (b) block diagram of a data-
distribution system.
be aiceptable as a first-order approximation.] The probability distribution P ( e ) of
Introduction to Discrete-Time Control Systems Chap. 1 Data Acquisition, Conversion, and

transducer. Once the physical variable is converted into a voltage or current signal, multiplexer is a multiple switch (usually an electronic switc ) that sequentially
the rest of the data-acquisition process is done by electronic means. switches among many analog input channels in bed fashion. The number
In Figure 1-5(a) the amplifier (frequentl instant of time, only one
Iows the transducer performs one or more of given input channel, the
the voltage output of the transducer; it converts a current signal into a vo for a specified period of
or it buffers the signal. The low-pass filter that follows the amplifier attenuates the time. During the connection time the sampl ircuit samples the signal
high-frequency signal wmponents, such as noise signals. (Note that voltage (analog signal) and holds its alue, while the analog-to-
are random in nature and may be reduced by low-pass filters. converts the analog value into digital ta (binary numbers). Each channel is read
common electrical noises as power-line interference are generally p in a sequential order, and the corresponding values are converted into digital data
be reduced by means of notch filters.) The output of the low-pass filter is an analog in tbe same sequence.
signal. m i s signal is fed to the analog multiplexer. The output of the multiplexer is
fed to the sample-and-hold circuit, whose output is, in turn, lexer. The demultiplexer, which is synchronized with the in
pling signal, separates the composite output digital data flrom the digital controller
digital converter. The output of the converter is the signal in
to the digital controller. into the original channels. Each channel is connected to a DIA converter to produce
The reverse of the data-acquisition process is the ata-distribution Process “%+, the output analog signal for that channel.
shown in Figure 1-5(b), a data-distribution system consists of registers, a demulti- er in a digital system convelrts an analog
plexer, digital-to-analog converters, and hold circuits. It conver the signal in digital
pulses. The hold circuit holds the value
form (binary numbers) into analog form. The output of the DI converter is fed tQ
of the sampled pulse signal over a specified period of time. The sample-and-hold is
the hold circuit. The output of the hold circuit is fed to the analog actuator, which, necessary in the AID converter to pro
in turn, directly control; the plant under consideration. input signal at the sampling instant.
In the following, we shall discuss each in b.+hal ComPonent h ~ ~ l v in e dthe available in a single unit, known as
signal-processing system. however, the sampling operation and
lexer. An analog-to-digital converter is the most expensive (see Section 3-2). It is common practice to use a single analog-to-digital converter
component in a data-acquisition system. The analog multiplexer is a device that and multiplex many sampled analog inputs into it.
performs the function of time-sharing an AID converter among many analog chan- practice, sarnpling duration is very short compared with the sampling period
nels. The processing of a number of channels with a digital controller is posible the sampling duration is negligible, the sampler may be considered an
because the width of each pulse representing the input signal is very narrow, so the mpler." An ideal sampler enables us to obtain a relatively simple mathemat-
empty space during each sampling period rnay be used for other signals. If many ical model for a sample-and-hold. (Such a mathematical model will be discussed in
signals are to be processed by a single digital controller, then these input signals must
be fed to the controller through a multiplexer. e 1-7 shows a simplified diagram for the sample-and-hold. Th
Figure 1-6 shows a schematic diagram of an analog multiplexer. The analog g circuit (simply a voltage memory device) in which an inpu
acquired and then stored on a high-quality capacitor with low leakage and low
dielectric absorption characteristics.
In Figure 1-7 the electronic switch is connected to the hold capacitor. Opera-
tional amplifier 1 is an input r amplifier with a high input impedance. Op-
erational amplifier 2 is the o amplifier; it buffers the voltage on the hold
capacitor .
To sampler
There are two modes of operation for a sample-and-hold circuit: the tracking
mode and the hold mode. n the switch is closed (that is, when the input signal
is connected), the operatin ing mode. The charge on the capacitor
in the circuit tracks the input voltage. the switch is open (the input signal is
disconnected), the operating mode is the hold mode and the capacitor voltage holds
constant for a specified time period. Figure 1-8 shows the tracking mode and the
hold mode.
Figure 1-6 Schematic diagram of an Note that, practically speaking, switching from the tracking mode to the hold
analog multiplexer. mode is not instantaneous. If the hsld command is given whde the circuit is in the
Introduction to Discrete-Time Control Cystems Chap. 1 Sec. 1-4 Data Acquisition, Conversion, and istribution Systemc

signal (usually in the form of a voltage or current) into a


coded word. In practice, the logic is based on binary dig
and the representation has only a finite
Arnp. 1 I
I
1 Arnp. 2 performs the operations of sample-and-ho
Analog
in the digital system a pulse is supplied every sampling period T by a clock. %he AID
Analog
input output converter sends a digital signal (binary number) to the digital contiroller each time
a pulse arrives.
Among the many ID circuits available, the following types are used most
frequently :

ve-approximation type

Sarnple-and-hoid
cornrnand
Each of these four types has its own advantages an
Figure 1-7 Sample-and-hold circuit. application, the conversion speed, accuracy, size
be considered in choosing the type of A / D convert
tracking mode, then the circuit will stay in the tracking mode for a short while before for example, the number of bits in the output signal must be increased.)
reacting to the hold command. The time interval during which the switching takes As will be seen, analog-to-digital converters art of thek feedback l o o p
place (that is, the time interval when the measured amplitude is uncertain) is called digital-to-analog converters. The simplest ty verter is the counter type.
the aperture time. The basic principle on which it works is that clock pulses are applied to the digital
The output voltage during the hold mode may decrease slightly. The hold mode counter in such a way that the o
droop may be reduced by using a high-input-impedance output buffer amplifier. the feedback loop in the AID con
Such an output buffer amplifier must have very low bias current. the output voltage
The sample-and-hold operation is controlled by a periodic clock. hen the output volt
the clock pulses are stopped. The counter output voltage is then the digital output.
es oofAnalog-to-Digital earlier, the PrQcess by The successive-approximation type of AID converter is much faster than the
which a sampled analog signal is quantized and converted to a binary number is counter type and is the one most frequently use Figure 1-9 shows a schematic
called analog-to-digital conversion. Thus, an AID converter transforms an analog diagrarn of the successive-approximarion type of

I npuí Sarnple to Hold rnode


I signal hold offset droop D/A Analog
converter reference

Digital
output

Tracking
mode 4 -ey; :pk
t
Analog
input
t
tioíd command Figure 1-43 Traclcing mode and hold
is given here mode. Figure 1-9 Schematic diagram of a successive-approximation-typeof BID converter.
lntroduction to Discrete-Time Control Sec. 'l-4 Data Acquisition, Conversion, and

The principle of operation of this type of AID converter is as follows. T


successive-approximation register (SAR) first turns on the most significant bit (ha
the maximum) and compares it with the analog input. The com
whether to leave the bit on or turn it off. If the analog input voltage is larger, tfae
most significant bit is set on. The next step is to turn on bit 2 and t
analog input voltage with three-fourths of the maximum. After n
c o q l e t e d , the digital output of the successive-approximation reg
those bits that remain on and produces the desired digital code.
ID converter sets 1 bit each clock cycle, and so it requires o d y n clock cycles t s
generate n bits, where n is the resolution of the converter in bits. (The number n
of bits employed determines the accuracy of conversion.) The time required for the
conversion is approxirnately 2 psec or less for a 12-bit conversion.
ctual analog-to-digital signal converters differ
from the ideal signal converter in that the former always have some errors, such as
offset error, linearity error, and gain error, the characteristics o£ which are shown
in Figure 1-10. Also, it is important to note that the input-output characteristics
change with time and temperature.
Finally, it is noted that commerciaI converters are specified for three basic
temperature ranges: cornmercial (0°C to 70°C), industrial (-25°C to 85"C), and
military (-55°C to 125°C).
onverters. At the output of the digital controller the
digital signal must be converted to an analog signal by the process called digital-to-
analog conversion. A DIA converter is a device that transforms a digital input (binary
numbers) to an analog output. The output, in most cases, is the voltage signal.
For the full range of the digital input, there are 2" corresponding different
analog values, induding O. Por the digital-to-analog conversion there is a one-to-one
~orrespondencebetween the digital input and the analog output.
Two methods are commonly used for digital-to-analog conversion: the method
usirig weighted resistors, and the one using the R-2R ladder network. The former
is simple in circuit configuration, but its accuracy may not be very good. The latter
Errors in AID converters:
is a little more complicated in configuration, but is more accurate, (a) offset error; (b) linearity error;
Figure 1-13 shows a schematic diagram o£ a DIA converter using weighted (c) gain error.
resistors. The input resistors of e operational amplifier have their resistance values
weighted in a binary fashion. hen the Iogic circuit recebes binary 1, itch
Notice that as the nurnber of bits is increased the range of resistor values becomes
(actually an electronic gate) co ects the resistor to the referente voltage. the
logic circuit receives binary 0, the switch connects the resistor to ground. The large and consequently the accuracy becomes poor.
digital-to-analog converters used in common practice are of the parallel type: al1 bits e 1-12 shows a schematic diagram of an n-bit DlA converter using an
-2 er circuit. Note that with the xception of the feedback resistor (which
act sirnultaneously upon application of a digital input (binary numbers).
is 3 R ) al1 resistors involved are either or 2R. This means that a high Ievel of
The D/A converter thus generates the analog output voltage corresponding
accuracy can be achieved. The output voltage in this case can be given by
to the given digital voltage. For the D/A converter shown in Figure 1-11, if the
binary number is b3b., bl bo, where each of the b's can be either a O or a 1, then the
output is
Ciircuii. She sarnpling operation pro-
duces an amplitude-modulated pulse signal. The function of the hold operation is
lntroduction to Biscrete-Time Control

Output

Figure 1-13 Outpur from a zero-order


hold.

ore sophisticated hold circuits are available than the zero-orde


r-order hold circuits and include the first-order hold a
gher-order hold circuits wiIl generally reconstnict a signal more
Figure 1-11 Schematic diagram of a D/A converter using weighted resistors. a zero-order hold, but with some disadvantages, as explained next.
The first-order hold retains the val e of the previous s a m p k as we
present one, and predicts, by extrapolation, the next sam
to reconstruct the analog signal that has been transmitted as a train of pulse sa generating an output slope equal to the slo
That is, the purpose o£ the hold operation is to fill in the spaces between sampling and present samples and projecting it from
periods and thus roughly reconstruct the original analog input signal. in J?igure 1-14.
The hold circuit is designed to extrapolate the output signal between successive As can easilv be seen from the figure, if the slope of the original signal
points according to some prescribed manner. The staircase waveform of the output not change much, the prediction is go
shown in Figure 1-13 is the simplest way to reconstruct the original input signal. The slope, then the prediction is wrong a
Id circuit that produces such a staircase waveform is called a zero-order hold. causing a large error for the sampli
cause of its simplicity, the zero-order hold is commonly used in digital control An interpolative first-order ho k d d , reconstructs the
systems. nal signal much more accuratel generates a straight-line
ut whose slope is equal to that joining the previous sample value and the present
sample value, but this time the projection is made from the cuirent sample point with

Output

Figure 1-12 n-Bit DIA converter using an R-2R ladder circuit


Sec. 7-5 Concluding Cornrtents

have been used in many srnall-scale control systems. Digital controllers ase often
superior in performance and lower in price than their analog counterparts.
Output Analog controllers represent the variables in an equation by continuous
ical quantities. The easily be designed to serve satisfactorily as non-decision-
e cost of analog computers or analog controllers increases
of the computations increases, if constant accuracy is to

There are additional advantages of digital controllers over analog controllers.


Egital components, such as sample-and-hold circuits, AID and DlA converters, and
highly reliable, and often compact
Figure 1-15 Output from an inter- ts have high sensitivity, are often
polative first-order hold (polygonal cheaper than their analog counterparts, and are less sensitive to noise signals. And,
hold).
as mentioned earlier, digital controllers are flexible in allowing programming
changes.

the amplltude of the prevlous sample. ence7 the accuracy in reconstructing the Digihl Ccpnkr~b industrial process control systems, it is gen-
original signal is better than for other hold circuits, but there is a one-sarnpling- erally not practica1 t ry long time at steady state, because certain
period delay, as shown in Figure 1-15. effect, the better aCcuracY is achieved at changes rnay occur in production requirements, raw materials, economic factoss,
the expense of a delay of one sampling period. From the viewpoint of the stability and processing equipments and techniques. Thus, the transient behavlor of indus-
of closed-loop systems, such a delay is not desirable, and so the interpolative trial processes must always be taken into consideration. Since there are interactions
first-order hold (polygonal hold) is not used in control system applications. amsng process variables, using only one p ariable for each control agent is
not suitable for really complete control. use of a digital controller, it ks
possible to take into account al1 process v together with econornic factors,
ents, equipment performance, and al1 other needs, and thereby
al control of industrial processes.
In concluding this chapter we shall compare digital controllers and analog controllers Note that a system capable of controlling a process as c o q l e t e % yas
used in industrial control systems and review digital control of processes. Then we will have to solve complex equations. The more complete the control, t
shall present an outline of the book. important it is that the correct relations between operating variables be known and
used. The system must be capable of accepting instructions from such varied sources
rs. Digital controllers operate only as computers and human operators and must also be capable of changing its control
on numbers. Decision making is one of their important Tunctions. They are often subsystem completely in a short time. Digital controllers are most suitable in such
used to solve problems involved in the optimal overall operation o£industrial plants. situations. %nfact, an advantage of the digital controller is flexibility, that is, ease
Digital controllers are extremely versatile, They can handle nonlinear control of changing control schemes by reprogramming.
equations involving complicated computations or logic operations. A very much In the digital control of a complex process, the designer must have a good
wider class of control laws can be used in digital controllers than in analog con- knowledge of the process to be controlled and must be able to obtain its mathemat-
trollers. Also, in the digital controller, by merely issuing a new program the oper- ical model. (The mathematical model rnay be obtained in terms of differential
ations being performed can be changed completely. Shis feature is particularly equations or difference equations, or in some other form.) The designer must be
important if the control system is to receive operating information or instructions familiar with the measurement technology associated with the output of the process
from some computing center where econornic analysis and optirnization studies are and other variables involved in the process. e or she must have a good working
made. knowledge of digital computers as well as modern control theory. Iíf the process is
Digital controllers are capable of performing complex computations with complicated, the designer must investigate severa1different approaches to the design
constant accuracy at high speed and can have almost any desired degree of compu- of the control system. In this respect, a good knowledge of simulation techniques
tational accuracy at relatively little increase in cost. is helpful.
Originally, digital controllers were used as components only in large-scale
control systems. At present, however, thanks to the availability of inexpensive The objective of this book is to present a detailed
microcornputers, digital controllers are being used in many large- and small-scale acco~intof the control theory that is relevant to the analysis an design of discrete-
control systemc. hn fact, digital controllers are replacing the analog controllers that time control systems. Our enphasis is on understanding the basic conceptc involved.
lntroduction to Discrete-Time Control Ystems ChaP '1

In this book, digital controllers are often designed in the form of pulse transfer
functions or equivalent difference equations, which can be easily implemented in the
form of computer programs.
The outline of the book is as follows. Chapter 1has presented introductory ma-
terial. Chapter 2 presents the z transform theory. ter includes z transforms
of elementary functions, important properties an
inverse z transform, and the solution of differe
method. Chapter 3 treats background materials
systems. This chapter includes discussions of impulse sampling and reconstruction
of original signals from sampled signals, pulse transfer functions, and realization of
digital controllers and digital filters.
Chapter 4 first presents mapping between the S plane and the z plane and then
discusses stability analysis of closed-loop systems in the z plane, followed by tran-
sient and steady-state response analyses, design by the root-locus and frequency-
response methods, and an analytical design method. Chapter 5 gives state-
space representation of discrete-time systems, the solution of discrete-time state-
space equations, and the pulse transfer function matrix. Then, discretization of
continuous-time state-space equations and Liapunov stability analysis are treated.
Chapter 6 presents control systems design in the state s
chapter with a detailed presentation of controllability and obs
present design techniques based on pole placernent, follow
full-order state observen and minimum-order state observe atical tool commonly used for the analysis an esis of discrete-time
chapter with the design of servo systems. Chapter 7 treats the
approach to the design of control systems. e b%in the c h a ~ t e withr diX~ssionsof systems is similar to that of the Lapla form in continuous-time systems.
Diophantine equations. Then we present the design of regulator systems and control
systems using the solution of Diophantine equations. The approach here is an
alternative to the pole-placement approach ombined with minimum-order observ-
ers. The design of model-matching control systems is included in this chapter. tions to linear difference equations become algebraic in nature. (Just as the Laplace
Finally, Chapter 8 treats quadratic optimal control problems in detail. transformation transforrns linear time-invariant differential equations into algebraic
The state-space analysis and design of discrete-time control systems, presented equations in S , the z transformation transforms linear time-invariant difference
in Chapters 5,6, and 8, make extensive use of vectors and rices- In studying h % e equations into algebraic equations in z .)
chapters the reader may, as need arises, refer to App which ~ummarizecthe The main objective of this chapter is to resent definitions of the z transform,
basic materials of vector-matrix analysis. Appendix ts materiak in z tEUls- basic theorems associated with the z transform, and methods for finding the inverse
form theory not included in Chapter 2. Appendix C treats pole-placement design z transform. Solving difference equations by the z transform method is also
problems when the control is a vector quantity. cussed.
In each chapter, except Chapter 1, the main text is followed by solved problems
and unsolved problems. The reader should study al1 solved probl Signak, Diserete-t signals arise if the system involves a
Solved problems are an integral part of the text. Appendixes A, sampling of continuous-time als. The samplled signal is x(O), x(T),
followed by solved problems. The reader who studies these solved problems will have x(2T), . . . , where T is the sampling period. Such a sequence of values arising from
an increased understanding of the material presented. the sampling operation is usually written as x(kT). If the system involves an iterative
process carried out by a digital computer, the signal involved is a nu
x(O),x(l), x(2). . . . The sequence of numbers is usualhy written as x(k), where tbe
argument k indicates the order in which the number occurs in the sequence, for
example, x(O), x(l), 4 2 ) . . . . Although x(k) is a number sequence, it can be con-
sidered as a sampled signal of x(t) when t e sampling period T is 1 sec.
The z Transform Sec. 2-3 z Transformc of Elementary Functions

The 2 transform applies to the continuous-time signal x(t), sampled signal The z transform defined by Equation (2-3) or (2-4) is referred to as the two-sided
x(kT), and the number sequence x(k). dealing with the z transform~if no z transform. In the two-sided z transfor e function x(t) is assumed to be
confusion occurs in the discussion, we occasionally use x (kT) and x (k) interchange- nonzero for t < O and the sequence x(k) is considered
ably. [That is, to simplify the presentation, we occasionally drop the explicit appear- h the one-sided and two-sided z transforms
ance of T and write x(kT) as x(k).] involves both positive and negative powers of z
one-sided z transform is considered in detail.
ection 2-1. has presented introductory remarks. For most engineering applications the one-sided z transform
ection 2-2 presents the definition of the z transform and associated subjects. venient closed-form solution in its r
Section 2-3 gives z transforms of elementary functions. mportant properties an infinite series in z-', converges
theorems of the z transform are presented in Section 2-4. radius of absolute convergente, in u ansform method for
computational methods for finding the inverse z transform are discussed in time problems it is not necessary each cify the values of z over which X(z)
2-5. Section 2-6 presents the solution of difference equations by the z transform ks convergent.
method. Finally, Section 2-7 gives concluding commeats. Notice that expansion of the right-hand side of Equation (2-1) gives

any continuous-time function x(t) may


zmkin this series indicates the position
The z transform method is an operational method that is very
working with discrete-time systems. In what follows we shall define
of a time function or a nurnber sequence.
In considering the z transform of a time function x(t), we consider
sampled values o£x(t), that is, x(O), x(T), x(2T), . . . ,where Tis the samplin
The z transform of a time function x(t), where t is nonnegative, os o£a
of values x(kT), where k takes zero or positive integers and Tis the sampli
is defined by the following equation: inversisn integral method (see Section 2-5 for details.)
m

For a sequence of numbers x(k), the z transform is defined by


m

n the following we shall present z transfor S of several elementary functions. It is


noted that in one-sided z transform theory, in sampling a discontinuous function x(t),
The z transforrn defined by Equation (2-1) or (2-2) is referred to as the one-sided we assume that the function is continuous from the right; that is, if discontinuity
z transform. occurs at t = O, then we assume that x(O) is equal to x(O+) rather than to the average
The symbol7 denotes "the z transform of." In the one-sided z transform, at the discontinuity, [x(O-) + x(O+)]/2.
we assume r (t) = O for t < O or x(k) = O for k < O. Note that z is a complex variable.
Note that, when dealing with a time sequence x(kT) obtained by sampling a Unit-Step Func~on. ket us find the z transforrn of the unit-step function
time signal x(t), the z transform X(z) involves T explicitly. Owevery £m a Imnber
sequence x(k), the z transform X(z) does not involve T explicitly.
The z transform of x(t), where -.a < t < .a, or of x(k), where k takes integer
values (k = 0, +-1, t 2 , ), is defined by S just noted, in sampling a unit-step function we assume that this function is
m continuous from the right; that is, l(0) = l . Then, referring to Equation (2-l), we
X(z) = 27 [x (t)] = Z [x (kT)] = have
k=-m
The z Transform Chap. 2 Sec. 2-3 z Transforms of Elementary Fcdnctions

Note that it is a function of the sampling period T


on ak. Let us obtain the z transform of x ( k ) as

Notice that the series converges if lzl > 1. fina"ir%the transform, the Variable where a is a constant. Referring to the definition of t e z mnsform given by
ator. It is not necessary to specify the region of z over which Equation (2-2), we obtain
suffices to know that such a region exists. The z transform m w

n ~ ( tobtained
) in this way is valid throughout the z plane
except at poles of X ( z ) .
t is noted that 1 ( k ) as defined by
k = 0,1,2 ,.-.

is commonly called a unit-step seqraence. -- z


-
on. Consider the unit-ramp function z-a
on. Let us find the z transform of
e-"', OSt
x(t) =
Notice that
x(kT) = k T , k = 0,1,2,. Since
Figure 2-1 depicts the signal. The magnitudes of the sam x ( k T ) = e-"kT, k = 0,1,2,. . .
values are proportional iod T . The z transforrn of the unit-r
function can be written as
m m m

on. Consider the sinusoidal function


sin o t , O5t
x(t) =
10, t<O
Noting that
e;"' = cos ot +j sin ot
e-;"' = cos ot - j sin ot

O T 2T 3T 4T t Figure 2-1 Sampled unit-ramp signai.


The z Transform Chap. 2 Cec. 2-3 z Transforms of Elementary Functions

The inverse Laplace transform of X e ) is


Since the z transform of the exponential function is
1 ( t ) = 1- e O 5t
7[e-"'] =
&
1- e-aTZ-l
Hence ,
we bave 1 1
- 1 X(Z) = Z[l - e-'] = -
1-
- 1- e-'Z-l

l'-j
Z-l

X(z) = ;S [sin wt] = ;S (elw' - eiw')]

en&. Just as in working with t lace transformatio*, a table


transforrns of commonly encountered functions is very useful for solving problerns
- z-l sin wT in the field of discrete-time systems. Table 2-1 is such a table.
1 - 2z-l cos oT + z - ~
- z sinoT
z 2 - 2 z coswT+ 1

Obtain the z transform of the cosine function


coswt, OClt
t<O

If we proceed in a manner similar to the way we treated the z transform o£ the


sine function, we have
x ( ~ )= Y wt] = $Y[e1." + e-]" 1

- 1 - z-' COS wT
1 - 2.2-' cos wT + z-'

Obtain the z transform of

Whenever a function in S is given, one approach for finding the corresponding z


transform is to convert X ( s ) into x(t) and then find the z transform of x(t). Another
approach is to expand X ( s ) into partial fractions and use a z transform table to find the
z transforms of the expanded terms. Still other approaches will be discussed in
The z Transform Chap. 2 Cec. 2-4 lmportant Properties and Theorerns of the z Transform

The use of the z transform method in the analysis of discrete-time control systems
may be facilitated if theorems of the z transform are referred to. In this section we
present important properties and useful theor
the time function x is z -transformable al

where a is a constant.
To prove this, note that by definition
m

. The z transform possesses an Pmportant prop-


erty: linearity. This means that, if f(k) and g(k) are z-transformable and a and P
are scalars, then x(k) formed by a linear combination
x(k) = af(W + 13gW

where F(z) and G(z) are the z tr


The linearity property can be
X(z) = Z[x(k)] = Z[af (k) + pg(k)]

This can be proved as follows:


m
m

x ( t ) = 0, for t < 0.
. The shifting theorern presented here is also ieferred to as
x(1cT) = x(k) = 0, for lc < 0.
on theorem. If x(t) = O for t < O and x(t) as Wz),
Unless otherwise noted, 1: = 0,1,2,3,. . . .
The z Transform Chap. 2 lrnportant Properties and Theorems of the z Trancform

Similarly,

and Z[x(k + n)]


= znX(z) - znx(0) - z n - ' ~ ( 1 )- zn-24 2 ) - .- - - zx(n-1) (2-13)
where n is a positive integer.
where n is zero or a positive integer. emember that multiplication of the z transform
To prove Equation (2-7), note that advancing the signal x(kT) by one step (1 sarnpiing per
m of the z transform X(z) by z-' has the effect of delaying the signal x(kT) by one step
(1 sampling period).

Find the z transforms of unit-step functions that are delayed by 1 sampling period
and 4 sampling periods, respectively, as shown in Figure 2-2(a) and (b).
By defining m = k - n, Equation (2-9) can be written as follows: Using the shifting theorem given by Equation (2-7), we have
m
z-'
r [i(t - TI] = Z - ~ Z [ I ( ~ )=] Z-l- --
1 -1*-1 - 1 - z-l

Since x (mT) = O for m < O, we may change the lower limit of the summation from
m = -n to m = 0.

(Note that z-' represents a delay of 1sampling period T, regardless of the value of T.)
Thus, multiplication o£ a z transform by z-" has t e 0f delaying the time
function x(t) by time nT. (That is, move the funct lo the right by time nT.) Obtain the z transform of
To p;ove-~quation(2-8)? we note that

For the number sequence x ( k ) , Equation (2-8) can be written as follows:

From this last equation, we obtain o 2T 3T 4T 5T 6T 7T 8T ' Faginre2-2 (a) UnitSstep function
7[ ~ ( k+ l)]= zX(z) - zx(Q) (2-1 1) delayed by 1 sampling period; (b) unit-
step function delayed by 4 sampling
Z[r (k + 211 = z l[x(k + 111 - zx(1) = z2X(Z) - z2x(0) - zx(1) (2-12) (b) periods.
The z Transform Chap. 2 Cec. 2-4 lmportant Properties and Theorems of the z Transform

Weferring to Equation (2-7), we have Noting that


Z [ x ( k - l ) ]= z-%(z) Z-' sin o T
Z [sin wt] =
1 - 22-' coswT +
The z transform of ak is
we substitute zeaTfor z to obtain the z transform of e-"' sin wt, as follows:
e-aT
z -1 sinwT
Z [e-"' sin wt] =
1 - 2 e - a T ~ - 'cos o T + e - 2 n T ~ - 2
and so
Similarly, for the cosine function, we have
1 - z-'
z [ f ( a ) ]= Z[ak-']
= z-'- - - 1 - Z-' cos wT
1-az-' 1-az-'
2' [cos wt] =
where k = 1,2,3, . . . .
1 - 22-' coswT zW2 +
y substituting zeaT for z in the z transform of cos wt, we obtain
1 - e-"Tz-' cos wT
Consider the function y ( k ) , which is a sum of functions x ( h ) , where h = 0 , 1 , 2 , Z [e-"' cos ot] =
such that
1 - 2 e - a T ~ - 'C O S ~+Te-2aTz-2
k

Obtain the z transform of te-"'.


Notice that
where y ( k ) = O for k < O. Obtain the z transform of y ( k ) .
First note that

Thus,

. If x ( t ) has the z transfor X ( z ) and if lim X ( z ) exists,


z-t m
Therefore, then the initial value x(O) of x ( t ) or x ( k ) is given by
Z [ y ( k ) - y ( k - l ) ]= Z [ x ( k ) ] x ( O ) = lim X ( z )
2 4m

is theorem, note that

which yields

ketting z --+m in this last equation, we obtain Equation (2-15). The be


signal in the neighborhood of t = O or k = O can thus be determined by the behavior
where X ( z ) = 2' [ x ( k ) ] . of X ( z ) at z = m.
The initial value theorem is convenient for checking z transform calculations
If x ( t ) has the z transform X ( z ) , then the z
for possible errors. ince x(0) is usually known, a check of the initiaI value by
transform of e -"'x(t) can be given by X(zeuT).This is known as the complex trans-
l i m X ( z ) can easily spot errors in X ( z ) , if any exist.
lation theorern . z-tm

To prove this theorern, note that


x
Determine the initial value x ( 0 ) if the z transform of x ( t ) is given by
"[e-"'x(t)] = 2 ~ ( k T ) e - " ~ ' z=- ~
k=O k=O
x(kT)(~e"')=
- ~X(zeaT) (2-14)

Thus, we cee that replacing z in X ( z ) by zeaTgives the z transform of e-"'x(t).


Examplie 2-6 By using the initial value theorem, we find
Given the z transforms of sin wt and cos wt, obtain the z transforms of e-"' sin wt and ( 1 - e-')z-'
x ( 0 ) = iim
e - "' cos wt , respectively, by using the complex translation theorem. z-m(l - z - l ) ( l - C T z - ' )= O
The z Transforrn Chap. 2 ec. 2-5 The Inverse z Transforrn

Referring to Example 2-2, notice that this X ( z ) was the z transform of


x ( t ) = 1 - e-'
and thus x ( 0 ) = 0 , which agrees with the result obtained earlier.

Pt is noted that the given X ( z ) is actually the z transform of

y substituting t = m in this equation, we have


X(W) = lim ( 1 - e-"') = 1
f-m

As a matter of course, the two resdts agree.


. lin this section we have resented important roperties and theo-
rems of the z transforrn that will e to be useful in solving many z transforrn
problerns. For the purpose of conv nt referente, these important properties and
theorems are sumrnarized in Table 2-2. y of the theoserns presented in this
table are discussed in this section. cussed here but included in the table
ence, are derived or proved in Ap

Taking the limit as z approaches unity, we have


m 1 The z transforrnation serves the same role for discrete-time control systems that the
Laplace transformation serves for continuous-ti control systems. For the z trans-
form to be useful, we must be familiar with thods for finding the inverse z
transform.
ecause of the assumed stability condition and t e condition that x(k) = O for
The notation for the inverse z transform is Z-l. The inverse z transform of
k < 0 , the left-hand side of this iast equation becomes
X(z) yields the corresponding time sequence x(k).
m
Irt should be noted that only the time sequence at the sarnpling instants is
k=O
[x (k) - x(k - l)] = [x(O) - x(- l)] + [ X (1) - x (O)] obtained frorn the inverse z transform. Thus, the inverse z transform of X(z) yields
a unique x(k), but does not yield a unique x(t). This rneans that the inverse z
t-[x(2) - x(l)l + - = X ( W ) = limx(k)
e *
transform yields a time sequence that specifies the values of x(t) only at discrete
k-+ m

instants of time, t = O, T, 2 T , . . . , and says not ing about the values of x(t) at al1
ence, other times. That is, many different ti e functions x(t) can have the sarne x(kT).
limx(k) = lim [(l- z-1)X(z)]
k-*m 1-2 1 X(z), the z transform of x(kT) or x(k), is given, the operation that
which is Equation (2-16). The final value theorem is very useful in determining the onding x(kT) or x(k) is called the inverse z transformation.
behavior of x(k) as k - + frorn its z transform X(z). n obvious method for finding the inverse z transform is to refer to a z transform

Determine the final value x(m) of extensive table of z tra


1 1 a sum of simpler z tr
X(z) = ----- - 1 - e-aTZ-1 a>O
1 - Z-1 presented in this section.)
Other than referring to z transform fables, four methods for obfaining the
by using the final value theorem.
inverse z transform are commonly available:
By applying the final value theorem to the given X ( z ) , we obtain
The z Transform ec. 2-5 The Inverse z Transform

Figure 2-3 Two different continuous-time functions, x l ( t ) and x Z ( t ) ,that have the
same values at t = O, T, 2T,. . . .

Direct division method


Gomputational method
. Partial-fraction-expansion method
nversion integral rnethod

In obtaining the inverse z transform, we assume, as usual, that the time


sequence x(kT) or x ( k ) is zero for k < 0.
efore we present the four methods, however, a few comments on poles and
zeros of the pulse transfer function are in order.
he. Hn engineering appiications of the z transform
method, X(z) may have the form

where the p,'s (i = 1 , 2 , . . . ,n ) are the poles of X(z) and the z,'s ( j = 1,2, . . . ,m )
the zeros of X(z).
The locations of the poles and zeros of X(z) determine the characteristics of
x(k), the sequence of values or numbers. S in the case of the plane analysis of
linear continuous-time control systems, we often use a graphical display in the z
plane of the locations of the poles and zeros of X(z).
Note that in control engineering and signal processing X(z) is frequently
expressed as a ratio of polynomials in z-', as follows:
The z Transform Chap. 2 Sec. 2-5 The lnverse z Transform

where z-' is interpreted as the unit del coefficients of the zWkterm in the series are the values x(kT) of the time sequence
properties and theorems of the z tra or the values of x(k) of the number sequence.
expressed in terms of powers of z , as given ,or in terms of powers h the present rnethod
of z-', as given by Equation (2-18), ,x(l),x(2), . . . in a se
e poles and zeros of an expression for the general term from a set of values of x(kT) or x(k).
als in z . For example,
Find x(k) for k = 0,1,2,3,4 when X(z) is given by

as poles at z = -1 and z = -2 and zeros at z = O and z = -0.5.


If X(z) is written as a ratio of polynomials in z-', however, the prece First, rewrite X(z) as a ratio of polynomials in z-', as follows:
be written as

Dividing the numerator by the denominator, we have


Although poles at z = - 1and z = -2 and a zero at z = -0.5 are clearly seen from
e expression, a zero at z = O is not explicitly shown, and so the beginner may fail
see the existence of a zero at z = O. Sherefore, in dealing with the pol
of X ( z ) , it is preferable to express X(z) as a ratio of polynomials in z ,
polynomials in z-l. In addition, in obtaining the inverse z transform by use o
inversion integral method, it is desirable t Tres§ as a ratio of ~ o l ~ n o m i a l ~
in z , rather than polynomials in z-', to av anY possible erKJrs determininb the
number of poles at the origin of function X(z)zk-l.
1 8 . 6 8 ~- +3.736~~~
- ~22.416~-~
. In the direet division rnethod we obtain the inverse Thus,
z transform by expanding X(z) into an infinite power series in z-l. This method is
useful when it is difficult to obtain the closed-form expression for the inverse z
By comparing this infinite series expansion of X(z) with X(z) =
transform or it is desired to find only the first several terms of x(k). we obtain
The direct division rnethod stems from the fact that if X(z) is expanded into
a power series in z-', that is, if x(0) = o

As seen from this example, the direct division method may be carried out by hand
calculations if only the first several terms of the sequence are desired. In general, the
method does not yield a closed-forrn expression for x(k), except in special cases.

Find x(k) when X(z) is given by


then x(kT) or x(k) is the coefficient of the zwkterm. ence7 the vahes of 4 k T ) 01
1 - z-l
x(k) for k = 0,1,2, . . . can be determined by inspectlon. X(z) = -- -
If X(z) is given in the form of a rational function, the expansion into an infinite z+l 1+z-l
power series in increasing powers of z-' can be accomplished by simply dividing the By dividing the numerator by the denominator, we obtain
numerator by the demminator, where both the numerator and denominato
are wrihten in increasing pdswers of z-l. lif the resulting series is conver
The r Transform Chap. 2 Sec. 2-5 The Inverse z Transform

By comparíng this infinite series expansion of X ( r ) with X ( z ) =


obtain = O, for k f O
x(O) = O Assume that x(k), the input to t e system G(z), is the Kronecker delta
x(1) = 1 input, or
x(2) = -1 x(k) = 1, for k = O
43)=3 0, for k f O
=
n(4) = -1 The z transform of the Kronecker delta input is
X(z) = 1
This is an alternating signal of 1 and -1, which starts from k = 1. Figure 2-4 shows Using the Kronecker delta input, Equation (2-19) can be rewritten as
a plot of this signal.

Obtain the inverse z transform of


X ( z ) = 1 4- 22-1 + 3ze2 + 4t-3
The transform X ( z ) is aiready in the form of a power series in z
Approach. MATLA can be used for finding t
has a finite number of terrns, it corresponds to a signal of finite length.
transforrn. Weferring to -20), the input X(z) is the z transform of the
we find
Kronecker delta input. the Kronecker delta input is given by
x(0) = 1
x(1) - 2
42)=3 where N corresponds to the end of the discrete-time duration of the process consid-
x(3) = 4 ered.
Since the z transforrn of the Kronecker elta input X(z) is equali to unity, the
Al1 other x(k) vaiues are zero. response of the system to this input is
En what follows, we present two corn
proaches to obtain the inverse z transform.
ce the inverse z transform of G(z) is given by y(O), y (l),y(2), . . . . Let us obtain
up to k = 40.
Difference equation approach To obtain the inverse z transform of G(z) with
follows: Enter the numerator and denominator as follows:
Consider a system G(z) defined by
num = [O
den = [l

n finding the inverse z transform, we utilize the Kronecker delta function So(kT), Enter the Kronecker delta input.
wbere
X =

Then enter the command

Y=
-1 0 4 Figure 2-4 Alternating signal of 1 and
-1 starting from k = 1. to obtaán the response y(k) from k = O tcs k = 40.
The z Transform Chap. 2 -5 The lnverse z Transform

e inverse z transforrn or the


response to the Kroneck rograrn 2-1.

33 plot the values of t e inverse z transfor


given in the folliowing.

onding plot is skown in Figure 2-5.

If this program is executed, the screen will show the output y(k) from k = O to 40
as follows:

Y=
Columns 1 through 7
O 0.4673 0.3769 0.2690 0.1 632 0.0725 0.0032
Response to Kronecker Delta Input
Columns 8 through 14
-0.0429 -0.0679 -0.0758 -0.0712 -0.0591 -0.0436 -0.0277
Columns 15 through 21
-0.01 37 -0.0027 0.0050 0.0094 0.01 1 1 0.01 O8 0.0092
Columns 22 through 28
0.0070 0.0046 0.0025 0.0007 -0.0005 -0.001 3 -0.001 6
Columns 29 through 35
-0.001 6 -0.001 4 -0.001 1 -0.0008 -0.0004 -0.0002 0.0000
Columns 36 through 41
0.0002 0.0002 0.0002 0.0002 0.0002 0.0001

computations begin from column 1 and end at column 41,


rather than from column O to column 40.) These values give the inverse z transforrn
of G(z). That is,
Y (0) = 0
y(1) = 0.4673
y ( 2 ) = 0.3769
re 2-5 Response of the system defined by Equation (2-20) to the Kronecker
delta input.
Sec. 2-5 The Inverse r Transforrn

1f oints (open circles, o)


need to plot(k,y,'ol) to plot(k,y,'o',k,y,'-'). efore we discuss the partial-fraction-expansion method, we shall review the shifting
theorem. Consider fhe following X ( z ) :

By writing z X ( z ) as Y ( z ) , we obfain
we can convert this equation into t
1
zX(z) = Y ( z )= -
1 - az-'
Referring to Table 2-1, the inverse z transforrn of Y ( z ) can be obtained as follows:
k) = O for k f O, T 1 [ Y ( z ) ]= y ( k ) = ak
Hence, the inverse z transform of X ( z ) = z-' Y ( z ) is given by
(O) and y ( l ) can be
k = -2 into Eqaiation (2-21), we fin$ 2-'[X(z)] = x(k) = y(k - 1)

Since y ( k ) is assumed to be zero for al1 k < O, we have

Consider X(z) as given by

To expand X(z) into partial fractions, we first factor t


of X(z) and find the poles of X(z):

Finding the inverse z transfor of Y(z) IXxV b3=2o


following difference equation for y ( k ) : then expand X(z)/z into partial fractions so t at each term is easily recognizable
in a table o£ z transforms. If the shifting theorem is utilized in
transforms, however, X(z), instead of X(z)/z, may be expanded
tions. The inverse z transform of X(z) is obtained as the sum
) = O for k # O. transforms of the partial fractions.
the initial data y (O) =
, FBRTRAN, or A comrnonly used procedure for the cas here all the poles are o£simple order
ation (2-22) can be solv
and there is at least one zero at the origin ( t is, bm = O) is to divide both sides
other .
o£X(z) by z and then expand X(z)/z into partial fractions. Once X(z)/z is expanded,
it will be of the form
presented here, which is paralle
Laplace transformation, is widely used in
The method requires that all terms in
recognizable in the table sf z tr The coefficient aican be determined by multiplying both sides of this last equation
To find the invesse z eran by z - p, and setting z = pi.This will result in zero for all the terms on the right-hand
side except ehe ai term, in which the multiplicative factor z - pihas been canceled
by the denominator.
The r Transform Chap. 2 Sec. 2-5 The lnverse z Transform

Noting that the two poles involved in the quadratic term of this last equation are
Note that such determination of a, is valid only for sim
complex conjugates, we rewrite X ( z ) as follows:
olves a multiple pole
n X ( z ) / z will have t

The coefficients cl and c2 are determined from Since


1 - e-aTz-' cos wT
2' [e-"" TOS wkT] = +
1 - 2eWaTz-'cos o T e-2aTz-2
e-aTz-l
o TwT
Z [e-"" sin wkT] = 1 - 2 e - a T ~ - 'cos sin + e-2aTz-2
by identifying e-"T = 1 and cos o T =$ in this case, we have wT = ~ 1 3and
Tt is noted that if X ( z ) / z involves a triple
sin wT = 1/3/2. Hence, we obtain
must include a term ( z + p J ( z - ~ 7 ~ () ~ . - -

Given the z transform and

where a is a constant and T is the sampling period, determine the inverse z transform Thus, we have
x ( k T ) by use of the partial-fraction-expansion method.
( k - 1 ) ~ 1 (k - 1 ) ~
The partial fraction expansion of X ( z ) / z is found to be x ( k ) = 4 ( l k P 1-
) 3(lk-') c o s+ (1.-') sin--
3
Xo=L-- 1 Wewriting, we have
z z - 1 z - e-aT
Thus,

The first severa1 values of x ( k ) are given by


From Sable 2-1 we find
r 1

Hence, the inverse z transform of X ( z ) is


x(kT) = 1 - k = 0,1,2,. ..

Let us obtain the inverse z transform of


Note that the inverse z transform of X ( z ) can also be obtained as follows:

by use of the partial-fraction-expansion method.


Since
We may expand X ( z ) into partial fractions as follows:
Sec. 2-5 The Inverse z Transform

and partial-fraction-expansion method rnay prove lo be simpler to apply. On the other

I
hand, in certain problems the partial-fraction-expansion approach may become
m
y-1
1 - z - 1z-l
+z-2 ] ;sí
=
krr
(1.) sin -3 laborious. Then, the inversion integral method proves to be very convenient.

krr 4 (k-l)rr
x(k) =
- 2 ~ sin-
?
3
+-
v'3
sin-
3 '
k = l,2,3,. . Qbtain x ( k T ) by using the inversion integral method when X ( z ) is given by
k 6 0
Although this solution may look difierent from the one obtained earlier, both solutions
are correct and yield the sarne values for x ( k ) . Note that
Pn is a useful techniq
verse z al for the z t r a ~ s f o
For k = O, 1 , 2 , . . . , X ( z ) z k - ' has two simple poles, z = zl = 1 and z = z2 = e-OT.
Hence, from Equation (2-24), we have

its center at the srigin of t x(k) =


-
-
1=1
[residue of
(Z
( 1 - e-OT)zk at pole z
- 1)(z - ePaT)
= z,
1
of complex variables. It c where
Kl = [residue at simple pole z = 11
m
[residue o£x(z)z"-' at pole z = zi af (2-24) -
i=l

K2 = [residue at simple pole z = e-OT]


-

Hence,
(2-25)
Ef X(z)zk-' contains a mult

Qbtain the inverse z transform of

by using the inversion integral rnethod.


Notice that

For k = 0 , 1 , 2 , . . . ,X ( z ) z k - ' has a simple pole at z = zl = e-"T and a double pole at


z = z2 = 1. Hence, from Equation (2-24), we obtain

1
Zktl
x(k) = [residue of at pole z = z,
I=I (z - z e-OT)
I ) ~ (-
= Ki + KZ
The z Transform -6 P Transform Method for

where P TRANSFORWLS OF x ( k + m) AND x ( k - m)


M, = [residue at simple pole z = e-."? Discrete function z Pansform
.k+l ,-a(k+l)T 1

( Z - ~ ) ~ -( e-"')
z ] = ( 1 - e-aT)2
K2 = [residue at double pole z = 11

= lim
(k + l ) z k ( z - e - a T )- zk+l
z-. I (z - ePaT)'

Soive the following difference equation by use of the z transform method:


x(k+2)+3x(k+I)+&(k)=O, x(O)=O, x(l)=l
First note that the z transforms of x ( k + 2), x ( k + l j , and x ( k ) are given,
respectively, by
Z [ x ( k + 2)] = z 2 X ( z ) - z 2 x ( 0 ) - zx(1)

Difference equations can be solved easily by use o£a Z [ x ( k + 1)] = z X ( z ) - zx(0)


nurnerical values of al1 coefficients and parameters are Z [ x ( k ) ]= X ( z )
expressions for x(k) cannot be obtained fr c o m ~ u t es0lLhX-l
r e*cePt for vepY 7
Taking the z transforms of both sides of the given difference equation, we obtain
special cases. The usefulness of the z "can ethod 6 t h t it enables tQ
z 2 X ( z j - z 2 x ( 0 ) - zx(1) + 3 z X ( z ) - 3zx(O) + 2%(z) = O
the clased-form expressisn for x(k).
Consider the linear time-invariant discrete-time system characterized by the Substituting the initial data and simplifying gives
following linear difference equation:

=--- 1 1
1 + z-' 1 + 22-'
where u ( k ) and x(k) are the systern's input and output, respedively, at the kth Noting that
iteration. In describing such a difference equation in the z plane, we take the z
transform of each term in the equation.
Eet us define
q x ( k ) ]= X(z)
Thenx(k + l),x(k + 2),x(k + 3), . . . andx(k - l ) , x ( k - 2),x(k - 3), . . . can be
expressed in terms of X(z) and the initial conditions. Their exact z transforms were Obtain the solution of the following difference equation in terms of x ( 0 ) and x ( 1 ) :
derived in Section 2-4 and are surnrnarized in Table 2-3 for convenient referente.
Next we present two exarnple problems for solving difference equations by t x(k + 2) + (a + b ) x ( k + 1) + a b x ( k ) = O
z transfcwm method. where a and b are constants and k = 0 , 1 , 2 , . . . .
The z Transform Chap. 2 Chap. 2 Example Problemc and Solutions

?'he z transform of this difference equation can be given by conventional analysis and design techniques available to analog (continuous-time)
control systems, such as the root-locus technique. Frequenc
[ z 2 X ( z )- z 2 r ( 0 ) - r x ( l ) ] + (a + b ) [ z X ( z )- zx(O)] + a b X ( z ) = O design can be carrie rting the z plane into
z-transforrned char
as the Sury stability
3 and 4.
Solving this last equation for X(z) gives
[z2 + (a + b)z]x(O)+ zx(1)
X(z) =
z2 + (a + b ) z + ab
Notice that constants a and b are the negatives of the two roots of the characteristic
shall now consider separately two cases: (a) a # b and ( Obtain the z transform of G k , where G is an n x n constant matrix.
(a) For the case where a # b , expanding.X(z)/z P t o partial fractions, we obtain y definition, the z transform of G ~ S
m

h [ G ~=] E~
k=O
~ z - ~

from which we get = 1 + GZ-' + G ~ ~ - " + 3 z - 3 + ...

The inverse z transform of X ( z ) @ves Note that G~ can be obtained by taking the inverse z transform of (
bx(0) + x(1) + x(1) (-b)k, or (21 - G)-' Z . That is,
x(k) =
b-a
(-a)" + ax(0)
a-b
a # b

where k = 0 , 1 , 2 , . . . .
) For the case where a = b , the z transform X(z) becomes Obtain the z transform of k 2
y definition, the t transform of k 2 is
m

Z [ k 2 ]=2 k 2 z - k = Z-' + 4z-2 + 9z-3 + l 6 z p 4+ e

k=O

= z-'(1 + z - l ) ( l + 32-' + 6z-2 + I O Z - ~+ 1 5 ~ +


- ~ )

The inverse z transforrn of X ( z ) gives ere we have used the closed-form ex


x(k) = ~ ( 0 ) ( - a+
) ~jax(0) +~(l)]k(-a)~-l, a = b involved in the problem. (See Appendix

where k = 0 , 1 , 2 , . . . .
Obtain the z transforrn of kak-' by two methods.

Method l . By definition, the z transform of kak-' is given by


m

In this chapter the basic theory of the z transform rnet een presenteda T'he
z transform serves the same purpose for linear time-i Saek-tirne sY5&Xls
as the Laplace transform provides for linear time-invariant continuous-time systems.
method of analyzing data in discrete time results in difference
z transforrn rnethod, linear time-invariant difference equations
into algebraic equations. This facilitates the transient response
analysis of the digital control system. Also, the z transform e t k d all0VJs L E t* We
Chap. 2 Exarnple Problerns and Solutions

ethod 2. The summation expression for the z transform of kak-' can also be written
as follows:
)7[kak-ll = 5kak-lz-k
k=O
= a-1
k=O
kakzk=
a
2
k=O
(a) -Ir

By using the final value theorem, we find

o~blemA-2-
k- m
Show that

2 ~ ( h=) E x ( k ) = lim X ( z )
2-1
h=O k=O

Next, to prove Equation (2-29), first define


k

and E
j ( k ) = h=i x ( h ) = x ( i ) + x(i i- 1) + + x(k)
where 1 r i r k - 1. Define also

Also show that


X ( z ) = x(i)zWi+ x ( i + 1)z-'"" + . - + x ( k ) z 7 * + - -
Then, noting that

where 1 5 i 5 k - 1.
we obtain
Solintaon Define 1-1
k

y(k) = 2 x(h),
h =O
k = O, 1,2,. . .
Since

the z transform of this last equation becomes


Y(Z) - z-' P ( z ) = Z ( z )
[Note that the z transform of x ( k ) , which begins with k = i, is X ( z ) , not X ( z ) . ] Thus,

Then, clearly -2-5


B(z) =-
1
1 - 2-1 [,(Z) - i(h)zh
h = ~ 1
Y ( k ) - Y ( k - 1) = x ( k ) Obtain the z transform of the curve x ( t ) shawn in Figure 2-6. Assume that the sampling
period T is 1 sec.
y writing the z transforms of x ( k ) and y ( k ) as X ( z ) and Y ( z ) , respectively, and by
taking the z transform of this last equation, we have ution From Figure 2-6 we obtain
x(0) = O
Y ( z ) - z-1 Y ( z ) = X ( z )
x ( 1 ) = 0.25
The z Transform Chap. 2 ap. 2 Example

Then

-
- of the individual terms give
The inverse z transforms

Then the z transform of x(k) can be given by


and therefore
~ ( k ) = 9 k ( 2 ~ - ' ) - 2 ~ + 3 , k = 0 , 1 , 2 ,...

Obtain the inverse z transform of

olution Expanding X(z) into partiaf fractions, we obtain

(Note that in this example X(z) involves a double pole at z = O. Hence the partial
Notice that the curve x(t) can be written as
fraction expansion must include the terms ll(z2) and llz.] By referring to Table 2-1,
) $ t - g t - 4 j 1 ~ t- 4)
~ ( t= we find the inverse z transform of each term of this last equation. That is,
where I(t - 4) is the unit-step function occurring at f = 4. Cince the sampling period
T = 1 sec, the z transform of x(t) can also be obtained as follows:
X(z) = Z [x(t)] = Z [$f ] - 7[a ( t - 4)l(t - 4)]

ence, the inverse z transform of X(z) can be given by


o-o-o=o, k=O
1-o-1=0, k =1
x(k) =
Consider X(z), where 2-1-0=1, k = 2
2k-'-O-O=2k-19 k=3,4,5, ...

Obtain the inverse z transform of X(z).


Sgalntigan We shall expand X(z)/z into partial fractions as follows:
Rewriting, we have

x(k)= /O9 1,
2k-1,
k =0,1
k=2
k = 3 , 4 , 5 ,...
To verify this result, the direct division method rnay be applied to this problem. Noting
that
we find
The z Transforrn Chap. 2 Chap. 2 Exarnple Probierns and

=-
(3
l lirn?
- q ! z - ~ dz
d2 '
(z - 1)3-
-
( Z zk113 I

Obtain the inverse z transform of


>
z-2
X(z) = (1 - z -1 >
3

Using the inversion integral method, obtain the inverse z transform of


utionr The inverse z transform of ~ - ~ / -( 1z - ' ) ~ is not available from most z
transform tables. It is possible, however, to write the given X(z) as a sum of z transforms
that are commonly available in z transform tables. Since the denominator of X(z) is
(1 - z - ' ) ~and the z transform of k2 is z-l(l + z-')!(1 - z-')~,let us rewrite X(z) as Sola~tion Note that

For k = 0, notice that X(z)zk-' becomes

Hence, for k = O, X(z)zk-' has t h ~ e esimple poles, z = zl = 1, z = z2 = 2, and z =


z3 = O. For k = 1,2,3, . . . , however, X(z)zbl has only two simple poles, z = zl = 1
from which we obtain the following partial fraction expansion: and z = z2 = 2. Therefore, we must consider x(0) and x(k) (where k = 1,2,3, . . . )
separately.
For k = O. For this case, referring to Equation (2-24), we have

The z transforms of the two terms on the rigkt-hand side of this last equation can be
found from Table 2-1. Thus,
lo
(Z - 1)(z - 2)z
at pole z = zi
1
where
It is noted that if the given X(z) is expanded into other partial fractions then the K1 = [residue at simple pole z = 11
inverse z transform rnay not be obtained.
As an aiternative approach, the inverse z transform of X(z) may be obtained by = -10
use of the inversion integral method. First, note that
(2 - 1)(z - 2)z
K2 = [residue at simple pole z = 21

z-2 (z - 1)(z - 2)z


Hence, for k = O, 1 , 2 , . . . ,X(z)zk-' has a triple pole at z = 1. Referring to Equation
(2-24), we have K3 = [residue at simple pole z = O]

x(k)
! zk
= residue of ---- at triple pole
(Z - 113
z = 1
The r Transform Chap. 2 Chap. 2 Example roblemc and Solutions

ence,
x(0) = K1 + K2 + K3 = -10 + 5 + 5 =O
For k = 1,2,3, . . . . Eor this case, Equation (2-24) becomes
10zk--'
at pole z = zi
( z - 1)(2 - 2 )
Method 2: Compuiational method (MATLAB approach). % ( z ) can be written as
= Ki + K2
where
K1 = [residue at simple pole z = 11 Hence, the inverse z transform of X ( z ) can be obtained with
Define

num = [l 2 O]
K2 = [residue at simple poIe z = 21
den=[l -2 11

If the values of x ( k ) for k = O, 1 , 2 , . . . ,30 are desired, then enter the Kronecker delta
input as follows:
Thus,
x(k)=K1+K2=-10+10(2k-1)=10(2k-1-1), k=1,2939
Mence, the inverse z transform of the given X ( z ) can be written
Then enter the command

An alternative way to write x ( k ) for k 2 O is


Program 2-3. [The screen will show the output x ( k ) from k = 0 té,
x ( k ) = 560(k) + 1 0 ( 2 ~ - '- l ) , k = 0 , 1 , 2 ,. . . k = 30.1 (MATLAB computations begin from column 1 and end at column 31, rather
where &(k) is the Kronecker delta function and is given by
for k = 0
"(w=jt: torkio
'4-2-1
Obtain the inverce z transform of

(2-30)

by use of the four methods presented in Section 2-5.


utiora
Method 1: Direcr division method. We first rewrite X ( z ) as a ratio of two polynomiak
in z-':

Columns 13 through 24

Dividing the numerator by the denominator, we get 37 40 43 46 49 52 55 58 61 64 67 70

X ( Z )= 1 +4 + +
~ 4 7 ~ - 2 ~ o z +-- ~ . e .
Columns 25 through 31
ence,
x(0) = 1
The z Transform Ckiap. 2 Chap. 2 Exarnple Problerns and

than from column O to column 30.) The values x ( k ) give the inverse z transform of X ( z ) . For k = O, 1,2, . . . ,X(z)zk-' has a double pole at z = P.
That is, (2-24), we have
x(0) = 1 ( 2 + 2)zk
residue of ----- at double pole z = 1
( 2 - 1)2
x(1) = 4
Thus,
x(2) = 7 - -

x ( k ) = -----i i m - [ ( z - I ) 2 W ]
(2 - l ) !Z-i dz
d
Method 3: Partial-fraction-expansion method. expand X ( z ) into the foliowing
= lim-[(z
Z-+I dz + 2)zk]
partial fractions: =3k+1, k = 0 , 1 , 2,...

Solve the following difference equation:


Then, noting that
2 x ( k ) - 2x(k - 1) + x ( k - 2) = u(k)
where x ( k ) = O for k < O and

y taking the z transform of the given difference equation,

2 X ( z ) - 22-l X ( z ) + Z - ~ X ( Z=) -
1
1 - z-'
we obtain
Solving this last equation for X ( z ) , we obtain
x(0) = 1
x(k)=3k+l, k = 1 , 2 , 3 ,...
which can be combined into one equation as follows: Expanding X ( z ) into partial fractions, we get
x(k)=3k+l, k = 0 , 1 , 2,...
Note that if we expand X ( z ) into the following partia1 fractions
Notice that the two poles involved in the quadratic term in this last equation are compiex
conjugates. Hence, we rewrite X ( z ) as follows:

then the inverse z transform of X ( z ) becomes


x(0) = 1
y referring to the formulas for the z transforms of damped cosine and damped sine
functions, we identify e-2aT= 0.5 and cos o T = 11V5 for this problem. Hence, we get
w T = ~ 1 4sin
, w T = l i d , and e-"= = l i d . Then the inverse z transform of X ( z ) can
be written as
x ( k ) = 1 - Se-akTcos wkT + $ewakTsin wkT
which is the same as the result obtained by expanding X ( z ) into the other partial
fractions. [Remember that X ( z ) can be expanded into different partial fractions, but
the final result for the inverse z transform is the same.]
from which we obtain
Method 4: Inversion integral rnethod. First, note that
The z TTansform Chap. 2 Chap. 2 Example roblems and Solutions

ence ,

Thus, the inverse z transform of X ( z ) gives


Consider the difference equation
x(k + 2) - 1.3679x(k + 1) + 0.3679x(k) = 0.3679u(k + 1 ) + 0.2642u(k)
where x ( k ) is the output and x ( k ) = O for k 5 O and where u ( k ) is the input and is
given by
u(k)=O, k<O
u(0) = 1
u ( l ) = 0.2142 Consider the difference equation
u ( 2 ) = -0.2142
u(k)=O, k=3,4,5 ,... where x ( 0 ) = O and x ( 1 ) = 1. Note that x ( 2 ) = 1, x ( 3 ) = 2 , x ( 4 ) = 3 , . . . . The series
U, 1,1,2,3,5,8,13, . . . is known as the Fibonacci series. Obtain the general solution
Determine the output x ( k ) . x ( k ) in a closed form. Show that the limiting value of x ( k + l ) / x ( k )as k approaches
infinity is ( 1 + 1/5)/2, or approximately 1.6180.
olintlon Taking the z transform of the given difference equation, we obtain olution By taking the z transform of this difference equation, we obtain
[ z 2 X ( z )- z 2 x ( 0 ) - zx(1)l - 1.3679[zX(z) - zx(O)] + 0.3679X(z)
= 0.3679[zU(z)- zu(0)l + 0.2642U(z) (2-31) olving for X ( z ) gives

By substituting k = -1 into the given difference equation, we find


x ( 1 ) - 1.3679x(O) + 0.3673x(-1) = 0.3679u(O) + 0.2642u(-1)
By substituting the initial data x ( 0 ) = O and x ( 1 ) = 1 into this last equation, we have
Since x ( 0 ) = x ( - 1) = O and since u ( - 1) = O and u ( 0 ) = 1, we obtain
x(1) = 0.3679u(O) = 0.3679
By substituting the initial data
x(0) = O , x ( 1 ) = 0.3679, u(0) = 1
into Equation (2-31), we get
~ " ( 2 ) - 0.36792 - 1.3679zX(z) + 0.3679X(z)
= 0.3679zU(z) - 0.36792 + 0.2642U(z)
The inverse 2 transform of X ( z ) is
Solving for X ( z ) , we find

Note that although this lact equation involves '~6the square roots in the right-hand side
The z transforrn of the input u ( k ) is of this last equation cancel out, and the values of x ( k ) for k = 0 , 1 , 2 , . . . turn out to
U ( Z )= Z [ u ( k ) ]= 1 + 0.21422-' - 0.2142~-~ be positive integers.
The z Trancform Chap. 2 Chap. 2 Example Problemc and Solutionc

The limiting value of x ( k + l ) l x ( k )as k approaches infinity is obtained as follows: The filtered output y shown next gives the Fibonacci series.

x ( k i1) X =
lim -
k-+m ~ ( k )
Columns 1 through 6
Since 1(1 -
o 1 1 2 3 5

Columns 7 through 12
Hence,
8 13 21 34 55 89
(1 + v5\""
x(k +
lim -= lim
') 1 -
- 1.6180 Columns 13 through 18
k-w x(k) 2
144 233 377 61 O 987 1597

Columns 19 through 24

Referring to Problem A-2-13, write a ATLAB program to generate the Fibonacci


2584 4181 6765 10946 1771 1 28657
series. Carry out the Fibonacci series to k = 30.
Solistiora The z transform of the difference equation 4301umns 25 through 30

46368 75025 121393 196418 31781 1 514229


is given by
, Column 31

Solving this equation for X ( z ) and substituting the initial data x(0) = O and x ( 1 )
we get
= 1, 1 832040

Note that column 1 corresponds to k = O and column 31 corresponds to k = 30. The


Fibonacci series is given by
The inverse z transform of X ( z ) will give the Fibonacci series. x(0) = 0
To get the inverse z transform of X ( z ) , obtain the response of this system to the
Kronecker delta input. MATLAB Program 2-4 wilf. yield the Fibonacci series.

% ***** The Fibona


% response of X(z) t
OO
/ x(Z) = 2/(2"2 - z - 1) ***** x(29) = 514,229
x(30) = 832,040
num = [O 1 01;
den = [l -1 -41;
u = [l zeros(1,3O)l; Consider the difference equation
x = fiIter(num,den,d
x(k + 2) + m ( k + 1) + ,4?x(k) = O (2-32)
The z Transform Chap. 2

-2
Obtain the z transform of k3.

Obtain the z transform of t2e-"'.

Obtain the z transform of the following x ( k ) :


x ( k ) = 9k(2k-1)- 2k + 3 , k = 0,1,2,. .
Figure 2-7 Region in the cup plane in
which the solution series of Equation Assume that x ( k ) = O for k < 0.
(2-32), subjected to initial conditions, is
finite, -5
Find the z transform of
Find the conditions on a and P for which the solution series x ( k ) for k = 0 , 1 , 2 , . . . ,
subjected to initial conditions, is finite.

where a is a constant.

Then, referring to Example 2-19, the solution x ( k ) for k = 0 , 1 , 2 , . . . can be given by -2-
Show that

The solution series x ( k ) for k = 0 , 1 , 2 , . . . , subjected to initial conditions x(O) and


x ( l ) , is finite if the absolute values o£a and b are less than unity. Thus, on the a p plane,
three critica1 points can be located: -2-7
Qbtain the z transform of the curve x ( t ) shown in Figure 2-8.

The interior of the region bounded by lines connecting these points satisfies the con-
dition / a / < 1, lb 1 < 1. The boundary lines can be given by 6, = 1, a - P = 1, and
a! -t- p = -1. See Figure 2-7. If point ( a , p) lies inside the shaded triangular region,
then the solution series x ( k ) for k = 0 , 1 , 2 , . . . , subjected to initial conditions x(0) and
x(l), is finite.

Obtain the z transform of -2-


Obtaln the inverse z transform of

where a is a constant.
The z Transform Chap. 2 Chap. roblerns

-1
Find the inverse z transfornl of Find the solution of the following difference equation:
x(k + 2) - 1.3x(k + 1) + 0.4x(k) = u ( k )
where x ( 0 ) = x ( l ) = O and x ( k ) = O for k < O. For the input function u ( k ) , consider
Use ( 1 ) the partial-fraction-expansion method and ( 2 ) the the following two cases:
a MATLAB program for finding x ( k ) , the inverse z transform of X ( z ) .

Given the z transform and

z-' u(0) = 1
X(z) =
( 1 - z-')(l + 1.32-' + O.4zw2) u(k) = O, k #O
determine the initial and final values of x ( k ) . Also find x ( k ) , the inverse z transform Solve this problem both analytically and computationally with MATLA
of X ( z ) , in a closed form.
-2-17
1 Solve the following difference equation:
Obtain the inverse z transform of x(k + 2) - x ( k + 1) + 0.25x(k) = u ( k + 2 )
where x ( 0 ) = 1 and x ( l ) = 2. The input function u ( k ) ic given by
u(k)=1, k = 0 , 1 , 2 ,...
Use ( 1 ) the invewion integral method and ( 2 ) the
Solve thic problem both anaiytically and computationally with
-12
Obtain the inverse z transform of
Consider the difference equation:
z - ~
X(z) =
( 1 - 2-l)(1 - 0.22-l)
x(k + 2) - 1.3679x(k + 1 ) + 0.3679x(k) = 0.3679u(k + 1) + 0.2642u(k)
where x ( k ) = O for k 5 O. The input u ( k ) is given by
in a closed form.
u(k)=O, k<0
u(0) = 1.5820
By using the inversion integral method, obtain the inverse z transform of
u ( 1 ) = -0.5820
u(k)=O, k = 2 , 3 , 4 ,...
Determine the output x ( k ) . Solve this problem both analytically and computationally
-2-1
with MATLAB.
Find the inverse z transform of

Use (1) the direct division method and (2) the MATLAB method.

Obtain the inverse z transform of

by use of the inversion integral method.


ec. 3-2 impulse ampling and Data

Discrete-time control systems may opera in discrete time and partly in


continuous time. Thus, in such control syst signals appear as discr
functions (often in the form of a sequence o or a nurnerical code) a
signals as continuous-time functions. In analyzing discrete-ti
plays an important role. To see why the z transform method is

shall consider a fictitious S er cornmonly called an


impu f this sarnpler is consider
that begins with t = O, with the sampling period equal to Tan
pulse equal to the sampled value of the continuous-ti
pictorial diagrarn of the i

sented by an arrow with an


Tke impulse-sampled output is a sequence of impulses, with the st~engthof
ulse equal to the magnitude of x(t) at the ding instant of time.
k T , the impulse is x(kT)6(t - e that 6(t - kT) = O
hall use the notation x"(t) to r
The z transform method is particularly useful for an signal x* (t), a train of impulses,
input-single-output linear time-invariant discrete-time infinite summation
presents background material necessary for the analysis and desig m

control systems in the zplane. The main advantage of the z transf x* (t) = x(kT)S(t - kT)
it enables the engineer to apply conventional continuous-time design methods to k=O

discrete-time systems that may be partly discrete time and partl or


Throughout this book we assume that the sampling operation is uniform; that
is, only one sampling rate exists in the system and the sampling period is constant.
If a discrete-time control system involves two or more samplers in the system, we shall define a train of unit impulses as ST(t),or
assume that al1 samplers are synchronized and have the same sarnpling rate or
sarnpling frequency.
k=O

er. The outline of this chapter is as follows. Section 3-1 The sampler output is equal to the product of t e continuous-time input x(t) and the
gives int ks. Section 3-2 presents a rnethod to treat the sampling train of unit impulses OT(t).Consequently, the sampler may be considered a rnodu-
operation as a mathernatical representation of the operation of taking samples x(kT) lator with the input x(t) as the modulating signal and the train of unit impulses &(t)
from a continuous-time signal x(t) by impulse modulation. This section includes as the carrier, as shown in Figure 3-2.
derivations of the transfer functions of the zero-order hold and first-order hold.
Section 3-3 deals with the convolution integral method for obtaining the z
transform. Reconstructing the original continu time signal from the sampled
signal is the main subject matter of Sectlon 3-4. ed on the fact that the Laplace
transform of the sampled slgnal is periodic, resent the sampling t h e ~ r e m .
ection 3-5 discusses the pulse transfer functio thematical modeling of digital
controllers in terms of pulse transfer functions is discussed. Section 3-6 treats the
realization of digital controllers and digital filters.
x*(t)

X( S ) 6r x"(s) Figure 3-1 Impulse sampler.


z-Plane Analycis of Discrete-Time Control Cysterns Chap. 3 Cec. 3-2 Impulse Sarnpling and Data Hold

Carrier . Let us summarize what we have just


signal x(t) is impulse sampled in a periodic manner,
signal may be represented by
m

x*(t) = 2 x(t)s(t - kT)


k=O

signal 1
1

Figure 3-2 Impulse sampler as a modulator.

Next, consider the Eaplace transform of Equation (3-1):


X"(s) = 2[x"'(t)] = x(O)%[6(i)] + x(T)Ce[G(t - T)] converts a continuous-ti
instants t = 0, T, 2T,. . . ,
S occurring at the sampling
od. (Note that between any
+ x(2T>ce[8(t - 2T)] + no information. Two signals
= x(O) + x ( T ) ~ - ~+' ~ ( 2 T ) e - ~
+~. '
m
sampled signal.)
= ~(kT)e-~" Data-hold is a process of generating a continuous-time signal h(t) from a
k=O discrete-time sequence x(kT). A hold circuit converts the sampled cignal into a
Notice that if we define continuous-time signal, which approximately reproduces the signaf applied to the
eTs -
sampler. %he signal h(t) during the time interval kT 5 t < (k + 1)Tmay be approx-
-z imated by a polynomial in r as follows:

where O 5 T < T. Note that signal h(kT) must equal x(kT), or

then Equation (3-2) becornes


m
ence, Equation (3-5) can be written as follows:
I

Hf the data-hold circuit is an nth-order polynomial extrapolator, it is called an


The right-hand side of Equation (3-3) ic exactly the same as the right-hand side of
nth-order hold. Thus, if n = 1, it is called a first-order hold. [The nth-order hold
t is the z transform o£ the seque e x(O), x(T), x(2T), . . . , gener-
uses the past n + 1 discrete data x((k - n)T),x((k - n + 1)T), . . . ,x(kT) to gen-
t = kT, where k = 0 , 1 , 2 , . . . . ence we may write
erate a signal h(kT t r).]
ecause a higher-order hold uses past sarnples to extrapolate a continuous-
nal between the present sampling instant and the next sampling instant, the
accuracy of approximating the ntinuous-time signal improves as the number of
and Equation (3-3) becomes
past samples used is increased. wever, this better accuracy is obtained at the cost
of a greater time delay. Hn closed-loop cont ystems, any added time delay in the
loop will decrease the stability of the system in some cases rnay even cause system
Note that the variable z is a cornplex variable and T is the sarnpling period. [ht should instability.
be stressed that the notation X(z) does not signify X ( s ) with s replaced by z , but The simplest data-hold is obtained when n = O in Equation (3-6), that is, when
ratber X' ( S = T.-' Ira z).]
z-Plane Analysis af Discrete-Tirne Control ystems Chap. 3 ec. 3-2 impulse Sampling and

Consider the sampler and zero-order hold shown in Figure 3-4(a). Assu
that the signal x(t) is zero for t < O. Then the out ut hl(t) is related to x(t) as follows:

Zero-order
hold

Figure 3-3 Sampler and zero-order hold Since

where O I( r < T and k = 0,1,2, . . . . Equation (3-7) implies that the circuit holds
the arnplitude of the sample from one sampling instantto the next. Such a data-hold
lace transform of Equation (3-9) becsmes
is cailled a zero-arder hold, or clamper, or staircase generator. The output of the
zero-order hold is a staircase function. In this book, unless otherwise stated, we
assume that the hold circuit is of zero order.
It will be seen later that the transfer function Gh of the zero-order hold may
be given by
1 - e-Ts
G,, = - Next, consider ehe mathematical mo shown in Figure 3-4(b). The output
S
of this model rnust be the same as tfiat of real zero-order hold, or
Figure 3-3 shows a sampler and a zero-order hold. The
input signal x(t) is sampled at discrete hnstants and the sampled signal is passed
through the zero-order hold. The zero-order hold circuit smoothes the sampled
signal to produce the signal h(t), which is constant from the last sampled value until
the next sample is available. That is,
+ t ) = x(kT),
h(k~ for O 5 t<T (3-8) From Figure 3-4(b), we have
e shall obtain a mathematical model of the combination of a real sampler and
zero-order circuit, as shown in Figure 3-4(a). Utilizing the fact that the integral of
an impulse function is a constant, we may assume that the zero-order hold is an Since
integrator, and the input to the zero-order hold circuit is a train of impulses. Then
a rnathematical model for the real sampler and zero-arder hold may be constructed
as shown in Figure 3-4(b), where GhO(s)is the transfer function of the zero-orcler
hold and x" (t) is the impulse sampled signal of x(t). Equation (3-11) may be written as

Zero-order
hold y comparing Equations (3-12) and (3-E?), we see that the transfer function of the
zero-order hold may be given by

Note that, mathematically, the system shown in Figure 3-4(a) is equivalent to the
(a) A real sampler and
hold; (b) mathematical
system shown in Figure 3-4(b) from the viewpoint of the input-output relationshi
61
model that consists of an impulse That is, a real sampler and zero-order hold can be aeplaced by a rnathernatically
(b) sampler and transfer function Gh0(s). equivalent continuous-time system that consists of an impulse sampler and a transfer
z-Plane Analysis of Discrete-Time Control ec. 3-2 impulse Sampling and

function (1 - e- Ts)ls The two sam fing processes will be distinguished (as "cey are
e
ño derive the transfer function of the first-order hold circuit, it is convenient
in Figure 3-4) by the rnanner in which the sampling switches are drawn. to assume a simple function for x(t). For
o not use first-order function, and a unit-ramp function
holds in control systems, it is worthwhile to see what the transfer function of uppose that we choose a unit-step fun
first-order holds may look like show that the transfer function of the -order hold shown in Figure 3-6(a)
first-order hold may be given by consists of stsaight lines that are
The output h(t) is shown in the
follows:

Mext we shafl derive Equation (3-14).


ave stated that Equation (3-6) describes the output of an nth-or
circuit, For the first-order hold, n = 1. Let us substitute n = 1into Equation (3-6). aplace transform of this last equation becomes
Then we have

whereO c: T < Tand k = 0 , 1 , 2, . . . . y applying the con


h((k - 1)2") = x((k - 1)T)
the constant al can be determined as Eollows:

ence, Equation (3-15) becomes

where 0 5 T < T. The extrapolation process of the first-order


Equation (3-16). The continuobis-time output signal h(t) obtained by use of the
first-order hold is a piecewise-linear signal, as shown in Figure 3-5.

First-order
hold

Figure 3-6 (a)


u , , Real sampler cascaded with first-order hold; (b) mathematical model
Figure 3-5 Input and output of a first-order hold. consisting of impulse sampler and GM(s).
r-Plane Analysis of iscrete-Time Control btaining the r Transform by the Convolution Integral Method

Figure 3-6(b) shows a mathematical model of the real sarnpler cascaded wit It is repeated that the impulse samplier is a fictitious sampler introduc
the first-order hold shown in Figure 3-6(a). The mathematical model consists of the the purpose of mathematical analiysi
impulse sampler and GR1(s),the transfer function of the first-order hol lement such a sampler that generates
pul signaX of this model is the same as the output of the real system.
is also given by Eq
place transform of ) to the first-order hold Ghl(s) is

Hn this section we shall obtain the z transform of x(t) by using the convolution integral
ence, the transfer function Chl(s) of the first-order hold is given by method.
Consider the impulse sarnpler shown in Figure 3-7. The output of the i
sampler is

x(t)¿i(t - kT) = x(t) (3-1 8)


k=O k=O

Note that a real sarnpler combined with a first-order hold is equivalent to an impulse Noting that
sampler combined with a transfer function (1 - e-TS)2(1FJ+ l)/(aS2).
Sirnilarly, transfer functions of higher-order hold circuits may be de-rived by
following the procedure presented. However, since higher-order
are not practica1 from the viewpoint of delay (which may cause system instability)
and noise effects, we shall not derive their transfer functions here. (The zero-order
hold is the simpiest and is used most frequently in practice.)
ag. Let us summarize what we have presented so far about impulse
sarnpling.

A real sampler samples input signal periodically and produces a sequence at X*( S ) is the Eaplace transform of product of two time functions, x ( t )
of pulses as the output. ile the sampling duration (pulse 6(t - kT). Note that it is not equal t e product of the two corresponding
sampier is very small ( will never become zero), the as Laplace transforms.
width, which implies that a sequence of pulses becomes a sequence of impulses The kaplace transform of the product of two Laplace-transformable functions
whose strengths are equal to the continuous-time signal at the sampling in- f (t) and g (t) can be given by
stants, simplifies the analysis of discrete-time systems. Such an assurnption is
f
valid if the sampling duration is very small compared with the significant time
constant of the system and if a hold circuit is connected to the output of the
sampler .
. When we transform eTsto z , the concept of impulse sampling (which is purely
a mathematical process) enables us to analyze by the z transform method
discrete-time control systems that involve samplers and hold circuits. This [For the derivation of Equatio l9), see Problern A-3-4.1
means that by use of the compiex variable z the techniques developed for the Let us substitute x(t - kT) for f(t) and g(t), respectively. Then
Laplace transform methods can be readily applied to analyze discrete-time the Laplace transform of
systems involving sampling operations.
3, As pointed out earlier, once the real sampler and zero-order hold are mathe-
rnatically replaced by the impulse sampler and transfer function (1 - e-Ts)ls,
the system becomes a continuous-time system. n i i s simplifies the analysis of
the discrete-time control system, since we may apply the techniques available
to continuous-time control systems. -- ..-
6, Figure 3-7 Impulse sampler
z-Plane Analysis of Discrete-Time Control the z Transform by the Convolution Integral Method

can be given by
2: p plane

where the integration is along the Iine from c - j m to c + j a and this line is paralle1
ne and separates the poles of X ( p ) from those of
is the convolution integral. Iat is a well-known fact
e evaluated in terms of residues by forrning a closed
contour cansisting of the line from c - j a to c + j m a ~ i da se icircle of infinite radius
in the left or right half-plane, provided that the integral along the added semicircle
is a constant (zero or a nonzero constant). That is, we rewrite E
follows:

Of
1 ure 3-8 Ciosed contour in the left
where T is a semicircle of infinite radius in the left or right half p plane. 1- e-Tts - p )
half of the p plane.
ñhere are two ways to evaluate t is integral ( m e using an infinite semicircle
in the M-half plane and the other an infinite semi roblem A-3-6 for the derivation of Equation (3-22).]
shall describe the results obtained by the two case for eTsin Equation (3-22), we have
analysis here, we assume that the poles
S) can be expressed as a ratio of polly

By changing the com lex variable notation from p to s , we obtain

where g (S) and p (S) are polynomials in s.


degree in s than q(sj, which means that
also assume that p(s) is of a higher residue of -X(s)z at pole of ~ () s
z - eTs
Assume that X(s) has poles SI,s2, . ,S,. f a pole at s
i
= sj is a simple pele,
then the corresponding residue Kj is
shall evalu-
ate the convolution integral given by Equation (3-21) using a closed contour in the
left half of the p plane as shown in Figure 3-8. Using this closed contour, Equation
(3-21) may be evaluated as follows: Noting that the denominator of X(s) is of higher If a pole at s = si is a multiple pole of order ni,then the residue Miis
degree in s than the numerator, the integral along fi vanishes.

Given
This integral is equal to the sum of the residues of ( p ) in the &sed contour.

obtain X ( z ) by use 0.6 the convolution integral in ihe left half-plane.


z-Plane Analysis of Discrete-Time Control ystems Chap. 3 Sec. 3-3 Obtaining the a Trancform by the Convolution Integral Method

Note that X ( s ) has a double pole at s = O and a simple pole at S = - 1.


Equation (3-23) becomes

residue of
1
X ( s > z at poole of X ( s )

--
-
- l o dL
lim - Z
s
s7
1
--- + Iim (S + 1)7
s s + z ~ e ~--i[ j
- z [ z - eTS+ (S + l ) ( - T ) e T s ] 1 z
1
---- zI
s (S + 1) z - eTs evaluating the integrals on the right
deir two cases separately: one case w
e of Equation (3-26), we need
denominator of X ( s ) is two or
= lim
*-+ O ( S + 1)2(z - eTSl2
+--(-1)2 z - e-T
more degrees higher in s than the numerat other case where the denom-
inator of X ( s ) is only one degree higher in s than the numerator.
Case 1: X ( s ) has a denominator two or more degrees hig
numerator. For this case, since X ( s ) possesses at least two more
we have

Then the integral along & is zero. Thus, in the present case
p plane. Let us choose the closed contour shown in Figure 3-9, whi
line from c - jm to c + jm and rR,
the portion of the sernicircle of infinite radius
in the right half of the p plane that lies to the right of t is h e . The closed contour
endoses al1 poles of 141 - e-T("p)],but it does not enclsse any poles of X ( p ) . Now X" ( S ) can be obtained as follows:
X " ( s ) can be written as

erivation of Equation (3-27), see Probiem A-3-7.1 Thus

Note that this expression of the z transforrn is useful in proving the sampling theorem
(see Section 3-4). owever, it is very tedious to obtain z transforrn expressions of
commonly encountered functions by this method.
Case 2: X ( s ) has a denominator one degree higher in s than the numerator. For
this case, lirirsX(s) = x(O+) =# O < and the integral along TRis not zero. [The
nonzero value is associated with the initial value x(O+) of x(t).]It can be shown that
the contribution of the integral along TRin Equation (3-26) is -fx(O+). That is,

Then the integral term on the right-hand si e of Equation (3-26) becomes

Show that X * ( s ) is periodic with period 2 7 1 . l ~ ~ .


Referring to Equation (3-29),

Figure 3-9 Closed contour in the right half of the p piane.


ec. 3-3 Obtaining the z Transform by the Convolution Integral Method

Hence,
1 1
X*(s + j-k) =-
m

Th=--
X(s + jw,k + j w h ) + -x(O+)
2
rz[gl(t)]= G ( z )
the z transform o£ x l ( t ) becomes
Let k + h = m. Then this last equation becomes
1
X 4 ( s + jmik) = -
- 1
X(s + jw,m) + x ( 0 - t ) = X'(s)
T,,-,
Therefore, we have X ( z ) = Z I G l ( s ) - e-Ts&;*(S)]
X*(s)=X*(s+jwsk), k = 0 , 1 , 2 ,... = /Z'[g1(t)l- Z[x1(t)]
Thus, X* ( S ) is periodic, with period 2.lr/wS.This means that, if a function X ( s ) has a = G 1 ( z )- Z - l G 1 ( z )
pole at s = si in the s plane, then % " ( S )has poles at s = si 5 jwsk ( k = 0,1,2, . . .).
= (1 - Z - ~ ) C ~ ( Z )

G ( s ) follows the zero-order hold. Then the product of the transfer function of the ( z =( S = (1 - ) Z [ ? ] (3-3 2)
zero-order hold and C ( s ) becomes
1 - e-Ts
X(s) = y a s )

In what follows we sha'il obtain the z transform of such an X ( s ) .


Note that X ( s ) can be written as follows:

then the z transform ciif the function


where
fxs)
G1(s)= -
S
can be obtained as follows. Since
Consider the function
X1(s) = e-TsG(s) x(s)= ( 1 - e - f i ) 2 T S +Ts2
'
G(s)

x1(t) = Iotgo(t- 4544 d~


where = Ts + 1 G ( s ) j
( 1 - z-1)2Z[T (3-33)
go(t) = %-'[e-fi] = 6(t - 7')
Equation (3-33) can be used for obtaining the z transform of the function involving
& ( t ) = %-'[G(s>] the first-order hold circuit.
Thus,

xl(t) = [6(t - T - r)gl(r)d r tain the z transform of


1 - e-Ts 1
X(s) = --
= g10 - TI S s+l
r-Plane Analysis of iscrete-Time Control riginal Signals from

Referring to Equation (3-32), we have spectrum of the sampled signal x"(t). The Laplace transform of x
obtained in Section 3-3 and is given by uation (3-27) or (3-29),
whether x(O+) = O or quency spectrum,
s in Equation (3-27). uency spectra, we nee
with the value of x(O+).] Thus,

Equation (3-34) gives the frequency

is periodic with period 2n-/O,,

sampling frequency is sufficiently high compar if a function X ( s ) has a pole at s = sl, then
onent involved in the continuous-time signal, t ( k = o, 1, 2, . . .).
Figures 3-11(a) and (b) show plots of t e frequency spectra X* (jw) versus

To reconstruct the original


minimum frequency that the sa
frequency is specified in the sam
time signal x(t) has a frequency spectrum as shown in
does not contain any frequency com
, IIf o,, defined as 2 d T , where T is the sampling period,
is greater than 2wl, or

where wl is the highest-frequency component present ontinauous-time sig-


nal x(t), then the signal x(t) can be reconstructed co
sigraal x* (t).
The theorem impIies that if ws > 2w1then, from t dge of the sampled
signal, it is theoretically possible to reconstruct exactl al continuous-time
signal. In what follows, we shall use an intuitive graphical approach to explain the
sampling theorem. For an analytical approach, see
To show the validity of this sampling theorem, we need to find the frequency

Figure 5-11 Plols of the frequency spectra / X":( j w ) l versus w for two values of sampling
-wl O 0, w Figure 3-10 A frequency spectrum. frequency w,: (a) ws > 2wl; (b)w, < 2wl.
a-Plane Analysis of Discrete-Time Control Sycterns Chap. 3 econstructing Original Signals f r o m Sampled Signals

for two values of the sampling frequency os. Figure 3-11(a) corresponds to o, > 2wl,
ile Figure 3-11(b) corresponds to ws < 2wl. Eac
consists of IX( jw)\lT repeated every ws = 2 ~ 1 2radlsec
"
of IX*(jw)] the component IX( jo)l/Tis called the prima
components, I x ( j ( o it o, k))llT, are called comple
If os > 2wl, no two components of IX*(jo)/
ill be repeated every w, sadlsec.
original shape of IX(jw)l o longer appears in the plot of
use of the superposition sf spectra. Therefore, we see bhat Figpaie 3-13 Frequency spectra of the signals before and after ideal filtering.
ignal x(t) can be reconst ted from the impulse-sampled
signal x"(t) by filtering if and only if os > 2wl. It is noted that if the sampling frequency is less than twice the highest-
It is noted that although the frequency component of the original continuous-time signal, then because of the
is specified by the sampling theor frequency spectrum overlap of the pri ary component and complementary com-
component present in the signal, ponents, everr the ideal filter cannot recons ct the original continuous-time signal.
closed-loop system a (In practice, the frequency spectrum of continuous-time signal in a control
at a frequency much system may extend beyond k i w,, even though the amplitudes at the higher frequen-
éo be lOo, to 20w,.) cies are small.)
r, The amplitude frequenc
pass filter GI(jw) is shown in Figure 3-12. The magnitu
S S
over the frequency range - os 5 o 5 o, and is zero ola
The sampling process introduces an infinite number realizable.
nents (sideband components) in addition to the primary c Since the frequency spectrum of the ideal filter is given by
will attenuate al1 such complementary components to
primary component, provided the s g frequency osis greater than twice the
highest-frequency component of t nuous-time ségnal.
reconstructs the continuous-time signal represented by the S the inverse Folarler transform of the frequency spectrum gives
shows the frequency spectra of the signals before and after ideal filtering.
The frequency spectrum at the output of the ideal filter is 112" times the
frequency spectrum of the original continuous-time signal x(t). Since t
S
has constant-magnitude characteristics for the frequency region - w, 5 w 5 w,, S
there is no distortion at any frequency within this frequency range. That is, there
is no phase shift in the frequency spectrum of the ideal filter. (The phase shift of the
ideal fifter is zero.)

- -1 sin -"-
o t
lrt 2

Equation (3-35) gives the unit-impulse response of the ideal filter. Figure 3-14
shows a plot of gI(t) versus t. Notice that the response extends from t = -a to t = '30.
This implies that there is a response for t < O to a unit impulse applied at t = O. (That
- - 0 Figure 3-12 Amplitude frequency is, the time response begins before an input is appllied.) This cannot be true in the
S
-
ws
2 2 spectrum of the ideal low-pass filter. physical world. Hense, such an ideal filter Is physlcally unrealizable. [lin many
z-Plane Analysis of Discrete-Time Control ec. 3-4 Reconstructing Original Signals f r o m Campled Signals

Figure 3-15(a) shows the frequency-response characteristics of the zero-order


. AS can be seen from Figure 3-15, ere are undesirable gain peaks at
of 3wS/2,5as/2, and so on. Notice

of the zero-order hold are not constant,


and zero-order hold, distortion of the
frequency spectra occurs in the system.
e phase-shift characteristics of the zero-order hold can be obtained as
follows. Note that sin (oU2) alte S positive and negative values as (U increases
from O to os, osto 2ws, 2wSto 30.4,
use 3-14 Impulse response g I ( t ) of ideal filter. bottom] is discontinuous at w = k &/a, where k = 1,2,3, . . . . Such a &con-
tinuity or a switch from a positive value to a negative value, or vic
considered to be a phase s ift of i180". lín Figure 3-15(a), phase
communications systems, however, it is possible to
to be -180". (It could be assumed to be 4-180" as well.) Thras,
adding a phase lag, which rneans adding a delay to t
systerns, increasing phase lag is not desirable from the viewpo
fore, we avoid adding a phase lag to approximate the ideal filter.]
Because the Ideal filter is unrealizable and because signals in practical. control
systems generally have higher-frequency components and are not id
limited, it is not possible, in practice, to exactly reconstruct a coné
from the sarnpled signal, no matter what sarnpling frequency is
words, practically speaking, it is not possible to recon ruct exactly a continuous-
time signali in a practical control system once it is sa

To compare the zero-order hold with the ideal filter, we shall obtain the frequency-
response characteristics of the transfer function of the zero-order hold.
ing jo for s in Equation (3-36), we obtain

The amplitude of the frequency spectrum of Gho(jw)is

The magnitude becomes zero at the frequency equal to the sarnpling frequency and Figure 3-15 (a) Frequency-response curves for the zero-order hold; (b) equivalent Bode
at integral rnultiples of the sarnpling frequency. dictgram when T = 1 sec.
z-Plane Analysis of iscrete-Time Control Cystems C econstructing Original

where

A modification of the presentation o£ the ncy-response diagram of


-15(a) is shown in Figure 3-15(b). The d shown in Figure 3-15(b)
ode diagram of the z o-order hold. The sampiing p I W s - -
WS 0 -
"S Ws W Figure 3-17 Diagram showing the
2 2 regions where folding error occurs.
to be 1 sec, or T = 2. Notice tude curve approa
ency points that are inte les of the sampling frequency o, =
phase curve [Figure 3-15(b), bottom]

TQ summarize what we have stated, the frequency spectrmm o£ t


Bn practice, signals in control systems have igh-frequency components, and some
the zero-order hold includes complementary components, since th
characteristics show that the magnitude of ChO( jo) is not zero for lo1 > os, except 4 folding effect will a h o s t always e e, in an electromechanical syste
at points where o = &os,o = &2os,o = +i3wS,. . . . Hn the phase curve there are some signal may be conta frequency spectrum of the signal,
phase discontinuities of +i180" at frequency points that are multiples of o,. Except therefore, may include lo as well as high-frequency noise
for these phase discontinuities, the phase characteristic is line comgonents (that is, nois e sampling at frequencies higher
Figure 3-16 shows the comparison of the ideal filter and than 400 Hz is not practical, the
For the sake of comparison, the magnit a Iow frequency. Rernember that al1 signals ith kequencies than U, ap
zero-order hold is a low-pass filter, alt quite good. e)£ten, as signals of frequencies between O and o,. fa&, in certaihicases, a signal of
additional low-pass filtering of the sign essary to effectively frequency may appear in the output.
remove frequency components highea than o,. 4 n the frequency spectra of an im sampled signal x*(t),where
The accuracy of the zero-arder epends on the sam- own in Figure 3-18, consider an arb frequency pointoz that falls
pling frequency o,, That is, the output of the hold may be made as close to the in the region of the overlap of the frequency spectra. The frequency S
original continuous-time signal as possibk by letting the sarnpling period T become o = o,comprises two components, IX*(jo2)l and Ix* (j(o, - @))l. The latter com-
as small as practically possible. ponent comes from the frequency spectrum centered at o = w,. Thus, the frequency
. The phenomenon of the overlap in the frequency spectra is known
as folding . Figure 3-17 shows the regions where folding error occurs. Tkie frequency
;o, is called the folding frequency or Nyq~tsrfrepency UN.That is,

-3ws -2w, w
- -w, O W,
- Ws 2w, 30, w
2 2

Comparison of the ideal filter and the zero-order hold. Figure 3-18 Frequency spectra of an impulse-sampled signal x* ( t ) .
r-Plane Analysis of iscrete-Time Control Sec. 3-5 The Pulse Transfer Function

led signal at o = 02 inch components not only al frequency


ncy o,- 02 (in general, no, where n is an integ
spectrum is filtered by a low-pass filter, such as a zero-o

component at o = n
it were a frequency
sin 3t
frequency component /
t frequency &~)2when
y o, - 02 (in general,

is not satisfied, then


x(t) sin t + sin 3 t
frequency high enough

osa. ñt is noted that, if the continuous-time signalx(t) involves


a frequency component equal to n times the sampling frequency o, (where n is an
integer), then that component may not appear in the sampled signal. For example,
if the slgnal
x(t) = xl(t) + x2(t) = sin t + s i n 3
where x,(t) = sin t and x4t) = sin 3t9 is sampled at t = 0 , 2 ~ 1 3 , 4 ~ /.3. ,. (the sa
pling frequency o,is 3 radlsec), then the sampled signal will not show the frequency
component with w = 3 radisec, the frequency equal to o,.(See Figure 3-19.)
Even though the signal x(t) involives an oscillation witb o = Plots of x,(t) = sin t, xZ(t)= sin 3t, and x ( l ) = sin t + sin 3t. Sampled
is, the component x2(t) = sin 311, the sampled signal does not show signal x(k), where sampling frequency w, = 3 radlsec, does not show oscillation with
frequency w = 3 radlsec.
Such an oscillation existing in x(t) between the sampling periods is called a hidden
oscillation .
is assumed to be a conitinuous-time signal y (t). f at the output there is another
sampler, which is synchronized in phase with the input r and operates at the
same sampling period, then the output is a train of impul assume that y (t) = O
for t < 0.
The transfer function for the continuous-time system relates the Laplace transform ñhe z transform of y(t) is
oE the continuous-time output to that of the continuous-time input, while the pulse
transfer function relates the z transform of the output at the sampling instants to that
of the sarnpXed input.
Before we discuss the pulse transfer function, it is appropriate to discuss
convolution summatian.
Convol n. Consider the response of a continuous-time system
driven by an ed signal (a train of impulses) as shown in Figure 3-20.
Suppose that x ( t ) = 0 for t < O. The impulse-sampled signal x*(t) is t
continuous-time system whose transfer function is G(s). The output of the system Figure 3-20 Continuous-time system G(s) driven by an impulse-sampled signal.
naiysis of Discrete-Time Control Systems Chap. 3 ec. 3-5 The Pulse Bransfer Function

In the absence o£ the output sampler, if we consi Equations (3-39) and (3-40) can be taken from
chronized in phase with the input sampler an changing the value of the summation. Therefore, Equations (3-39) and (3-40) can
iod) at the output and observe the sequence o£values taken by y ( t )only at instants be rewritten as follows:
t = k T , then the z transforrn of the output y" ( t )can also be given by Equation (3-38).
For the continuous-time system, it is a well-known fact that the output y(t) of
e system is related to the input x ( t ) by the convolution integral, or

g(t - ? - ) x ( ? - ) d ~ = l g >-x (~t ) g ( ? - ) d ~


Ht is noted that if G ( s ) is a ratio of polynomials in s and if the degree of the
where g(t) is the weighting function of the system or the impulse-response function denominator polynomial exceeds the degree of the numerator pofynomial onlly by
of the systern. For discrete-ti e sYstems we have a onvolution summation, whk 1the output y ( t ) Ps discontinuous, as shown in Figure 3-21(a).
is similar to the convofution integral. Since tinuous, Equations (3-41) and (3-42) yield the values immed
pling instants, that is, y (O+),y ( T + ) , . . . , y ( k T + ) . Such value
m

actual response curve.


If the degree of the denominator polynomial exce
m e y(t) of the system to tbe input x* ( t ) is the sum by 2 or more, however, the output y ( t )is con
hen y ( t ) is continuous, Equations (3-41) and (3
Ort<T
g(t)x(O) + g(t - T ) x ( T ) , Tit<2T
g(t - T ) x ( T ) + g(t - 2T)x(2T), 2T~t<3T

g(t)x(0) ig(t - T ) n ( T )+ + g(t - k T ) x ( k T ) , kT it < (k + 1)T

The values o£ the output y ( t ) at the sampling instants t = kT ( k = 0,1,2, . . .) are


given by
k

where g ( k T ) is the system's weighting sequence. [The inverse z transform of C ( z ) Figure 3-21 (a) Plot of output y(t)
is called the weighting sequence.] The summation in Equation (3-39) or (3-40) is (impulse response) versus t when the
called the convolution surnmation. Note that the sirnplified notation degree of the denominator polynomial
of G(s) is higher by 1 than that of the
numerator polynomial; (b) plot of
2T 3T 4T 5T output y(r) versus t when the degree of
is often used for the convolution summation. the denominator polynomial of G(s) is
Since we ascurned that xft) = O for t < O, we have x(kT - h T ) = O for h > k . higher by 2 or more than that of the
Also, since g ( k T - h T ) = O for h > k , we may assume that the values o£ h in (W numerator polynomial.
z-Piane Analysis of Discrete-Time Control ystems Chap. 3 ec. 3-"he Pulse iransfer Feanction

the sarnpling instants. The values y(k) in such a case portray the actual response Note that G(z) is also the z transfor
curve . necker delta input:

continuous-time part of the system has a


x(kT) = S,(kT) =
{u:
Since the z transform of the Kronecker delta input is
foa k = O
fork+

limsG(s) = O.
S"='

ulse on, From Equation (3-41) we then, referring to Equation (3-44), the response Y(z) to the Kronecker delta
input is

where g(kT - hT) = O for h > k . Henee, the z transfor Thus, tfme system's response to the Kronecker delta input is G(z), the z transform
of the weightirag sequence. This fact is parallel e fact that C(S) is the Laplace
transform of the systemysweighting function,

e signals in the system are starred (meaning that signals are impulse sampled)
others are not. To obtain pulse transfer functions and to analyze discrete-time
controf systems, therefore, we must be able to obtain the transforms of output signals
of systems that contain sampIing ope
uppose the impulse sampler is
whose transfer function is G(s), as sh 3. In the following analysis
where m = k - h and we assume that all initial conditions are zero m. Then the output Y($) is
a

g(rnT)z-" = z transform of g(t)


m=O Notice that Y ( s )is a product of X* (S), periodic with period 2?;r/o,,and G (S),
Equation (3-43) relates the pulsed output Y(z) of the syst which is riodic. The fact that th e-sampled sagnals are periodic can be
X(z). It provides a means for determining the z transform of seen fro fact that
for any input sequence. Divi ing both sides of Equation (3-43) by X(z) gives

In the following we skall show that in taking the starred Laplace transform of
G(z) given by Equation (3-441, the ratio of the output Y(z) and the input X(z), is Equation (3-45) we may factor out X * (S) so that
cafled the pulse trunsfer function of the discrete-time system. It is the z transform
of the weighting sequence. Figure 3-22 shows a block diagram for a pulse transfer
function C(z), together with the input X(z) and the out This fact is very important in deriving the pulse transfer function and also in
Equation (3-431, the z transforrn of the output signal can be obtained as the product simplifying the block diagram of the discrete-time control system.
of the system9spulse transfer function and the z transform of the input signal. So derive Equation (3-47), note that the inverse Eaplace transforrn of Y(s)
given by Equation (3-45) can be written as follows:

u pulse-transfer-function system. Figure 3-23 Impulse-sampled system.


r-Plane Analysis of Discrete-Time Controf Chap. ec. 3-5 The Pulse Transfer Function

v*W
_____)_
S, Y(z)
Fictitious
sarnpier

Then the z transform of y ( t ) becomes

where m = n - k . Thus, (a) Continuous-time system with an impulse sampler at the input; (b)
continuous-time system.

(3-48) Next, consider the system shown in Figure 3-24(b). The transfer function G(s) is
z transform can be understood as the starre Laplace transform with eTs given by
y z , the z transform may be considered to be a shorthand notation for the
starred Laplace transform. Thus, Equation (3-48) may be expressed as
ortant fact to rernernber is that the pulse transfer function for this system
which is Equation (3-47). have thus shown that by taking the starred Laplace [G(s)], because of the absence of the input sampler.
transform of both sides of uation (3-45) we obtain The presence or absence of the input sampler is crucial in determining the
To summarize what we have obtained, note that E pulse transfer function of a system, because, for exarnple, for the systern shown in
state that in taking the starred Laplace transfo igure 3-24(a), the Laplace transform of the output y (t) is
some are ordinary Laplace transforms and ot
the functions already in starred transforms can be factored out of the starred Laplace
transform operation. ence, by taking the starred Laplace transform of Y(s), we have
It is noted that systems becorne periodic under starred Laplace transform Y*( S ) = C * (s)X*( S )
operations. Such periodic systems are generally more complicated to analyze than
the original nonperiodic ones, but the former may be analyzed without difficulty if or, in terms of the z transform,
carried out in the z plane (that is, by use of the pulse-transfer-function approach). Y(z) = G(z)X(z)
while, for the system shown in Figure 3-24(b), the Laplace transform of the output
present general procedures for obtaining the pulse transfer function of a system thal
has an impulse sampler at the input to the system, as shown in Figure 3-24(a).
m is
The pulse transfer function G(z) of the system shown in Figure 3-24(a) is Y(s) = G(s)X(s)
which yields
Y" (S) = [G(s)X(s)f* = [GX(s)]*
? z-Plane Analysis of Discrete-Time Control Systerns Chap. 3 ec. 3-5 The Pulse Transfer Feinction

V(z) = Z[Y(s)] = z [ c ( s ) x ( s ) ] = Z[&;X(s)] =


Obtain the pulse transfer function of the system shown in Figure 3-24(a), where G(s)
is given by
The fact that the z transforrn of G(s)X(s) is not e ual to G(z)X(z) will be discussed
in detaPI Iater in this section.
In discussing the pulse transfer function, we a e that there is a sarnpler at Note that there is a sampler at the input of G(s).
ut o£ the elernent in consideration. The pres
e output o£the elernent (or the systern) does snot affect the pulse transfer function, Method d. G(s) involves the term (1 - e-Ts); therefore, referring to Equation (3-32),
because, if the sarnpler is not physicalfy presesnt a6 th we obtain the pulse transfer fttnction as foilows:
always possible to assurne that a fictitious sarnpler
eans that, although the output signal is con G(z) = Z f G ( s ) ] = Z
e output only at 1 = k?" (k = 0 , 1 , 2 , . . . ) a get sequence y (kT).
Note that only for the case where the input to khe*systemC(s) is an i
sannpled signal is the pulse transfer function given by

ExarnpIes 3-4 and 3-5 demonstrate t e methods h r obtaining the pulse From Table 2-1, the z transform of each o£the partial-fraction-expansion terms can be
transfen: function. found. Thus,

Obtain the pulse transfer function G(z) of the system shown in Figure 3-24(a), where
G(s) is given by

ethod 2. The given transfer function G(s) can be written as follows:

Note that there is a sampler at the input of G(s) and therefore the pulse transfer function
is C (z) = h [ G (S)].
Therefore, by taking the inverse Laplace transforrn, we have the following impulse
Method T. By referring to Table 2-1, we have response function:

Hence ,
e-kT + T - k = 1,2,3,. . .
= {o, k = o
Then the pulse transfer function G(z) can be obtained as follows:
Method 2. The impulse response function for the system is obtained as follows: w

Hence,

Therefore,
z-Plane Analysis of iscrete-Time Control Systems Chap. 3 ec. 3-5 The Pulse Transfer Function

er the system shown in Figure 3-25(b). From t

where

e starred Lapiace t

nce, by taking the starred La of ea& of these two equations, we In terrns of the z transform notation,

and the pulse transfer function y* ( t ) and input x* ( t ) is

Consequently,
Note that
( z ) $. G H ( z ) = Z [ C ; H ( s ) ]
ransfer functions of t
fn terms of the z transform notation, willl now verify this

e pulse transfer function between the output y*(t) and input x * ( t ) is t Consider the systerns shown in Figures 3-26(a) and (b). Obtain the pulse transfer
given by function Y(z)IX(z) for each of these two systerns.

Figure 3-26 (a) Sampled system with a sampler between elements G(s) = l/(s a )+
Figure 3-25 (a) Sampled system with a sarnpler between cascaded elements G(s) and H(s); and H(s) = l/(s + b ) ; (b) sampled system with no sarnpler between elements G(s)
(b) sampied system with no sampler between cascaded elements G(s) and H(s). and H(s).
Sec. 3-5 The Pulse Transfer Function

h r the system of Figure 3-26(a), the two transfer functions G(s) and M(s) are ence,
separated by a sampler. assume that the two sarnplers shown are synchronized and
kave the same sampling period. The pulse transfer function for this system is E(s) = R(s) - H(s)G(s)E*(s)
Then, by taking the starred kaplace transform, we obtain

Hence,
- - -
or

For the system shown in Figure 3-26(b), the pulse transfer function Y(z)lX(z)
is obtained as follows:
r- -..
Since

we obtain

Hence,
In terms of the z transform notation, the output can be given by

Clearly, we see that the pulse transfer functions of the two systems are different; that
is,
The inverse z transform of this last equation gives the values of t
G(z)H(z) $: GN(z) sampling instants. [Note that the actual output ~ ( tof)the system is a
Therefore, we must be careful to observe whether or not there is a sampler between . The inverse z transform of G ( z )will not give the continuous-
cascaded elements. ulse transfer function for the present closed-loop systern is
d ~ Tra~esfer
e Fune~a'csnof C s. In a closed-joop system the
existente or nonexistence of an output sampler within the loop makes a difference
in the behavior of the system. (If there is an output sampler outside the loop, it will
make no difference in the closed-loop operation.) Table 3-1 shows five typkal configurations for closed-loop discrete-time con-
Consider the closed-loop control system shown in Figure 3-27. In this system, ere, the sarnplers are synchronized and have the same sa
the actuating error is sampled. From the block diagrarn, period. For each configuration, the correspon
iscrete-time closed-loop control system
(that is, they do not have pulse transfer functions) because the input signal R ( s )
cannot be separated from the system clynamics. Although the pulse transfer function
may not exist for certain systern configurations, the sarne techniques discussed in this
chapter can still be applied for analyzing them.
ontmller. The pulse transfer function
of a digital controller may be obtained from the required input-output characteristics
of the digital controller.
Suppose the input to the digital controller is e(k) and the output is m(k). In
general, the output m(k) may be given by the following type of difference equation:

Figure 3-27 Closed-loop control system.


z-Plane Analysis of Discrete-Time Control Systems Chap. 3 Sec. 3-5 The Pulse Transfer Funetion

FIVE NPICAL CONFIGURATIONS FOR CLOSED-LOOP DISCRETE-TIME


CONTROL SYSTEMS
e z transform of Equation (3-51) gives
(z) + . - -+ a,z-"

ulse transfer function GD(z) of the digital controller may then be given by

The use of the pulse transfer function GD quation (3-52) enables


us to analyze digital control systems mi

3-28(a) shows a block diagram of a digital c


N D converter, digital controller , zero-order
continuous-time (piecewise-constaant) control
Figure 3-28(b) shows the transfer functions o
The transfer function of the dlgital contr
1 P-Plane Anaiysis of Discrete-Time Control Systerns Chap. 3 ec. 3-5 The Pulse Transfer Function

system the computer (digital controller) solves a difference equation whose input- signal, which is the difference between the in t and the feedback signal), integral
ut relationship is given by the pulse transfer function CD(z). control action (where the control action i
In the present system the output signal c(t) is fed
input signal r(t). The error signal e(t) = r(t
is converted to a digital signal thr
fed to the digital controller, which
desirable manner to produce the
This desirable relationship between t
by the pulse transfer function GD(z)
selecting the poles and zeros of GD(z),a number of input-output characteristics can
be generated.]
Weferring to Figure 3-28(b), let us define
ere e(t) is the input to the control1

Erom Figure 3-28(b), notice that

Bn eerms o£ the z transform notation,

Since

Define
and, therefore,

Figure 5-29 shows the function f(hT). Then


Equation (3-53) gives the closed-loop pulse transfer function of the digital control
system shown in Figure 3-28(b). The performance of such a closed-loop system can
be improved by the proper choice of GD(z),the pulse transfer function of the digital
shall later discuss a variety of forms for GD(z)to be used in obtaining Taking the z transform of this last eequation, we obtain
optimal performance for various given performance indexes.
In the following, we shall consider only a simple case, where the pulse transfer
function GD(z) is of the PID (proportional plus integral plus derivative) type.
ontmller. She analog
ustrial control systems for over half
a century. The basic princ (Por the derivation of this last equation, refer to Problern A-2-4.) Notice that
01 scheme is to act on the variable
to be rnanipulated through a proper combination of three control actions: propor-
tional control action (where the control action is proportional to the actuating error
z-Plane Analysis of Discrete-Time Control Systems Chap. Sec. 3-5 The Pulse Transfer Function

-29 Diagram depicting function f(hT).

Hence, trol scheme exhibits better response characteristics than


ontrol scheme. Another a vantage of the velocity-for
e((h - 1)T) + e(hT)
E(4 it is uceful in suppressing e cessive correctisns in proce
h=l 2
Then the z transforrn o£ Equation (3-55) gives Linear control laws in the form of 138 control actions, in both positional form

This last equation may be rewritten as follows:

Consider the control system with a digital PID coneroller shown in Figure 3-31(a). (The
PID controller here is in the positional form.) The trancfer function of the plant is
where assumed to be
KT
K p = K - - = K - K! = proportional gain
211 2
K~= --
KT =: integral gain
li
K~ = --
KTd = derivative gain
T
Pdotice that the proportional gain I;Lp for the phg> controller is smaller than
the proportional gain K for the analog PID controller by KJ2.
The pulse transfer function for the digital PID controller becornes

The pulse transfer function of the digital PID controller given by Equation (3-56)
is commonly referred to as rhe positional form of the PID control scheme. Figure 3-30 Block diagram realization of the velocity-form digital PID control scheme.
z-Plane Anaiysis of Discrete-Time Control Systerns Chap. 3

Digital PID Zero-order


controller hold Plant

Figure 3-31 (a) Block diagram of a control system; (b) equivalent block diagram.

and the sampling period T is assurned to be 1 sec. Then the transfer function of the
zero-order hold becomes
1 - e-"
C, ( S ) = -----
Since

we may redraw the block diagram of Figure 3-31(a) as shown in Figure 3-31(b).
Let us obtain the unit-step response of this system when the digital controller is
a PID controller with & = l9 & = 0.2, and KD = 0.2. The pulse transfer function of
the digital controller is given by

Then the cloced-loop pulse transfer function becomes

k
We shall use the MATLAB approach to obtain the unit-step response. re 3-32 Unit-step response.
iscrete-Time Control Sec. 3-5 The Pulse Trancfer Function 1

instants. In ordinar
e output will not vary very much between any two consecutive
sampling instants. Tn certain cases, however, we may need to find e
between consecutive sampling
Three rnethods for provi sponse between two consecutive sarnpiing

e transform method
ed z transform metkod
State-space method

Here we shall briefly discuss the ethod. The modified z trans-


form method is presented in App ers interested in the modified
z transforrn should read ection B-4.) The state-space method wili, be discussed in
Section 5-5.
for exarnple, the systern shown in

Equation (3-59) will give the continuous-time response c(t). ence, the response
al any time between two consecutive sampling instants can be calculated by the use
of Equation (3-59). [See Problem A-3-18 for sample calculationas of the right-hand
side of Equation (3-59).]
1 z-Plane Analysis of Biscrete-lime Control ec. 3-6 Realization of Digital Controllers and

In this section we discuss reali ulse transfer functions t


represent digital conttollers and
digital filters may involve either software or hardware or both. In general, "realiza- where
tion" of a pulse transfer function means eterrnining the physical layout for the al = -1
propriate combination of arithrnetic an storage operations. a:! = O
In a software realization we obtain co puter programs for the
involved. Pn a hardware realization we build a special- bo = Kp + KI + dg,
circuitry as digital adders, s (shift register§ with E& bl = -(Kp + 2KD)
sampling period T as a unit
Pn the field of digital b2 = KD
processing, a digital filter is a cornputational
algorithm that converts an input sequence of numbers into an output sequence in shall now discuss the irect propramming an
such a way that the characteristics of the signal are changed in so 1 filters. hn these programmings, coeffickn
fashion. That is, a digital filter processes a digital sigrmal by passing quantities) appear as rnultipliers in the block diagra
quency components of the digital input signal and rejecting undesirable ones. fhw diagram schemes where the coefficients a, and b, appe
general terms, a digital controller is a form of digital filter. called direct structures .
Note that there are important differences between the digital signal processing Consider the digital filter given by
used in communications and that used in control. In digital control the process sfer function has n poles and m zeros.
of signals must be done in real time. n communications, signal processing need n of the filter. The fact
be done in real time, and therefore delays can be tolerated in the processin Equation (3-60) can be seen easily, since frorn
improve accuracy. - a, a-, Y(z) + boX(z)
is sectioñ deals with the block n/(Z) = -al=-'Y(z) - a2z-2Y(z) -
ments, adders, and mulitipliers. + b, z-l X(z) + + b, z-"X(z)
diagram realizations will be discussed. Such block diagram realizations can b
Rearranging this Iast equation yields Equation (3-60).
as a basis fur a software or hardware design. Bn fact, once the block
realization is cornpleted, the physical realization in hardware or software is
forward. Note that in a block diagram realization a pulse tran
repiesents a delay of one time unit (see Figure 3-34.) (Note also that in the s plane
z-' corresponds to a pure delay e-".)
In what follows we shall deal with the digital filters that are used for filtering
and control purposes. The general form of the pulse transfer function between the
output Y(z) and input X(z) is given by
Y ( z ) = bo + blz-l + b2z-' + - + b , ~ - ~
G ( z ) = --
a m

X(z) 1 4- al z-' + a2z-2 + + a, z-" '


e e

where the al's and b,'s are real coefficients (some of them may be zero). The pulse
transfer function is in this form for many digital controllers. For example, the pulse
transfer function of the PID controller given by Equation (3-56) can be expressed
in the form of Equation (3-60), as follows:

Figure 3-34 Pulse transfer function


showing a delay of one time unit.
1 z-Plane Analycis of Diccrete-Time Control Systems Chap. 3

The type of realization here is called directprogramming . Direct


means that we realize the numerator and denominator of the pulse tra
using separate sets of delay elements. The numerator uses a set of m delay elements
and the denominator uses a different set of n delay elements. Thus, the total number
of delay elements used in direct prograrnming is m + n .
The number of delay elements used in direct programming can be reduced. In
fact, the nurnber of delay elements can be reduced from n + m to n (where n 2 m).
The programming method that uses a minimum possible number of delay elernents
standard programming .
practice, we try to use the minimum number of delay elernents in realizing
pulse transfer function. Therefore, the direct programming t
more rhan the minimum number of delay elements is more or less of ac
rather than of practica1 value.
ing. AS previously stated, the number of delay elements
required in direct programming can be reduced. In fact, the number of delay
elernents used in realizing the pulse transfer function given by Equation (3-60) can
be reduced frorn n + m to n (where n 2 m) by rearranging the block diagram, as
will be discussed here.
First, rewrite the pulse transfer function Y(z)/X(z) given by Equation (3-60)
as follows:

where

and

Then, draw block diagrams for the systems given by Equations (3-61) and (3-62),
respectively. To draw the block diagrams, we may rewrite Equation (3-61) as
Y(z) = b,N(z) + b , ~ - ~ N ( z+) - . + b,z-"H(z)
e (3-63)
and Equation (3-62) as
H ( z ) = X(z) - al z-1 H(z) - a, zW2
M(z) - - - a, z-" H(z) (3-64)
Then from Equation (3-63) we obtain Figure 3-36(a). Similarly, we get Figure
3-36(b) from Equation (3-64). The combination of these two block diagrams gives
the block diagram for the digital filter G(z), as shown in Figure 3-36(c). The block
diagram realization as presented here is based on the standard programming. Norice (a) Block diagram realization of Equation (3-63); (b) block diagram
that we use only n delay elements. The coefficients al, a2, . . a, appear as feedback f Equation (3-64); (c) block diagram realization of the digital filter
elements, and the coefficients bo,bl, . . . , b, appear as fee given by Equation (3-60) by standard propramminp.
z-Piane Analysis of Discrete-Time Control ystems Chap. 3 ec. 3-6 Realization of Di ¡tal Controllers and

The block diagrams in Figures 3-35 and 3-36(c) are equivalent, but the latter
uses n delay elements, while the former uses n + m delay elements. Obviously, the
latter, which uses a smaller number of delay elements, is preferred.
Fignre 3-37 Digital filter G(z) decomposed into a series connection of G,(z),
ents. Note first that the use of a minimal number of delay elements G ( z ) , . . . , Gp(z).
saves memory space in digital controllers. Also, the use of a minimal number of
conjugate complex zeros
poles and real zeros to pro
possible to group two real
The grouping is, in a sens irable to group severa
. The error due to the quantization of the input signal into a finite number of dis- to see which is best with respect to the num
crete levels. (Tn Chapter 1 we discussed this type of error, which may be con- the range of coefficients, and so forth.
sidered an additive source of noise, called quantization noise. The quantization To summarize, G(z) may be decornposed as follows:
noise may be considered white noise; the variance of the noise is (r2 = Q2/12.)
The error due to the accumulation of round-off errors in the arithmetic oper-
- 1 + eiz-l +f i ~ - ~
ations in the digital system.
,1 + ciz-l + diz-2
The block diagrarn for

small errors in the coefficients a, and b, cause large errors in the locations of
the poles and zeros of the filter.

These three errors arise because of the practica1 limitations of the number of
bits that represent various signal samples and coefficients. Note that the third type
are shown in Figures 3-3S(a) and (b), respectively. The block diagram for the digital
filter G(z) is a series connection of p component digital filters such as shown in
way, the system may be made less sensitive to coefficient inaccuracies. Figures 3-38(a) and (b).
For decomposing higher-order pulse transfer functions in order to avoid the
coefficient sensitivity problem, the following three approaches are commonly used. Eng. The second ap roacIS to avoiding the coefficient sen-
sitivity problem is to expand the pulse transfer function C(z) into partial fractions.
Series programming f C(z) is expanded as a sum of A , G1(z), G2(z), . . . ,Gq(z), or SO that
Parallel programming
Ladder programming where A is sirnply a constant, then the block diagram for the digital filter G(z) can
be obtained as a parallel connection of q + 1digital filters, as shown in Figure 3-39.
e shall discusc these three programmings next. ecause of the presence of the const term A , the first- and second-arder
The first approach used to avoid the sensitivity problem functions can be chosen in simpler forms. at is, C(z) may be expressed as
ransfer function G(z) as a series connection of first-ordes
transfer functions. If G(z) can be written as a product caf

G(z) = G1(z)G2(z) C;,(z)


then the digital filter for G(z) may be given as a series connection of the component
digital f h r s G1(z),G ~ ( z ).,. . , Gp(z), as shown in Figure 3-37,
In most cases the C,(z) (i = 1,2, . . . ,p) are chosen to be either first- or second- The block diagram for
order functions. If the poles and zeros of G(z) are known, Gl(z), G2(z),. . , C p ( ~ )
can be obtained by grouping a pair of conjugate complex poles and a pair of
r-Plane Analysis of iscrete-Time Control Systerns Chap. 3 Sec. 3-6 Realimation of Digiral Controllers and Digital Filters

decomposed as a parallel connection of


A, Wz), W ) , . . . , C&).

ion G(z) into the following continued-fraction forrn and to

The prograrnming method based on this scheme is calledi ladder programming . Let
igore 3-38 (a) Block diagram representation of Equation (3-65); (b) block us define
diagram representation of Equation (3-66).

and that for

are shown in Figures 3-40(a) and (b), respectively. The parallel connection of q + 1
component digital filters as shown in Figure 3-40 will produce the block diagram
for the digital filter G ( z ) .
iag, The third approach to avoiding ehe coefficient sensitiv- Then G(z) may be written as
ity probiem i s to implernent a ladder structure, that is, to expand the pulse transfer
z-Plane Analysis of Discrete-Time Control ystems Chap. 3 ec. 3-6 Realization of Digital Controllerc and

y the use of the feinctions 6 V 1 ( z ) 61B)(z),


, and 6 P 1 ( z ) ,t
may be written as follows:

Notice that GjB)(z)may be written as

The block diagram for G ! ~ ) ( given


z ) by Equation (3-70) is shown in Figure 3-41(a).
Similarly, the block diagram for G ( ~ ) ( zwhich
), may be given by

Q (a) Block diapram representation of Equation (3-67); (b) block


diagram representation of Equation (3-68).

shall explain this programming method by using a simple example where


2. That is,

(b)
Figure 3-41 (a) Block diagram for GIB'(z) given by Equation (3-70); (b) block
diagram for G(*'(z) given by Equation (3-71).
z-Plane Analysis of iscrete-Time Control Systems Chap. 3 ealimation of Digital Controllers and

structure. For example, a digital filter C(z) may be structured as a continue


(3-7 1) fraction expansion form around the origin in terrns sf z-l, as follows:

= AiX(Z)
Xi(z) - c(+R](z)~(z)
rnay be drawn as shown in Figure 3-41(b). Note that
1
6;$yz) = -
A,
y combining component digital filters as shown in Figure 3-42(a), it is
posslble to draw the block diagrarn of the digital filter G(z) as sho
3-42(b). [Note that Figures 3-42(a) and (b) correspond to the case
Also, instead of G(z), its inverse l/G(z) may be anded into continued-Eraction
en&. Digital filters based on la s in terrns of z or z-1 in order to carry out ladder programming.
with respect to coefficient sensitivity and accuracy. Realizatio
ture is achieved by expanding G(z) into continued fmctions ar
It is noted that the continued-fraction expansion given by Obtain the block diagrams for the following pulse-transfer-function system (a digital
not the only way possible. There are a few different ways to filter) by (1) direct programming, (2) standard programming, and (3) ladder program-
ming :

. Directprogramming. Since the given pulse transfer function can be written as

direct programming yields the block diagram showri in Figure 3-43. Notice that we need
two delay elements.
2. Standard programming. We shall firct rewrite the pulse transfer function as
follows:

where

and

Figure 3-42 (a) Component block diagrams for ladder programming of G(z) given by
Equation (3-69) when t~ = 2; (b) combination of component block diagrams showing ladder Figure 3-43 Block diagram realization of Y(z)lX(z) = (2 - 0.62-')1(1 + 0.5z-')
programming of G ( z ) . (direct programmins).
z-Plane Analysis of Discrete-Time Control Systems Ckap. 3 ec. 3-6 Realizarion of Digital Controllers and

Block diagram realizations of these last two equations are shown in Figure
3-44(a) and (b), respectively. If we combine these two diagrams, we obtain the block
diagram for the digital filter Y(z)/X(z), as shown in Figure 3-44(c). Notice that the
number of delay elements required has been reduced to E by the standard programming.
Ladder prograrnrnirzg . shall first rewrite the given Y(z)/X(z) in the ladder
form as follows:

5 Block diagram realization


of Y(z)/X(z) = (2 - O.62-')/(1 + 0.52-l)
Thus, A. = 2 and (ladder programming).

ence, we obtain

Referring to Figure 3-41(a) for the block diagram of G;~'(Z),we obtain the block
diagram of the digitai filter Y(z)/X(z) as shown in Figure 3-45. Notice that we need
only one delay element.

filters may be classified according to the duration of the irnpul


a digital filter defined by the following pulse transfer function:

where n 2 m. In terms o£ the difference equation,

The impulse response of the by Equation (3-721, where we


assume not al1 ai9sare zero, r of nonzero samples, although
their magnitudes rnay becorne negligibly small as k increases. This type of digital
filter is called an infinite-impulse response filter. Such a digital filter is also called a
recursive filter, because the previoies values of the output together with t
of the input are used in processing the signal to obtain the current
cause o£ the recursive nature, errors in previoies outputs rnay
sive filter may be recognized by the presence of both ai and b,
block diagram realization.
Next, consider a digital filter where the coefficients ai are al1 zero, or where

In terms of the difference equation,

Figure 3-44 (a) Block diagram realization of Y(z)lH(z) = 1 - 0.32-'; (b) block
y(k) = b,x(k) + b,x(k - 1) + e - . + b,x(k -m)
diagram realization of fI(z)/X(z) = 2/(1 + 0.52-'); (c) combination of block dia- The impulse response of the digital filter defined by Equadion (3-73) is limited to
grams in parts (a) and (b) (standard programming). a finite number of samples defined over a finite range of time intervals; dhat is, the
ec. 3-6 Realimation of Digital Controllerc and

impulse response sequence is finite. his type of digital filter is calle The characteristics of the finite-im ulse response filter can be summarized as
filter. lit is also called a nonrecursive filter or a rnoving-averagefilter. follows:
resent value of the o
. The finite-impulse The finite-impulse response filter Ps nonrecursive. Thus, because of the lack of
the block diagram re feedback, the accumulation of errors in past outputs can be avoided in the
prscessing of the signal.
ementation of the finite-impulse r filter does not require feed-
, so the direct programrn programming are identical.
Let us define the finite-impulse respon
Also, implementation may be speed convolution using the
digital filter as g ( k T ) . f the input x ( k T )is ap
fast Fourier transform.
can be given by
. The poles of the pulse transfer function of the finite-impulse response filter are
g(hT)x(kT - ha") at the origin, and therefore it is always stable.
If the input signal involves high-frequency components, then the number sf
= g (0)x(k?")+ g ( T ) x( ( k - 1)T ) + . - + g ( k T ) x( O ) (3-74) delay elements needed in the finite-impulse response filter increases and t
The output y ( k T ) is a convolutisn summation of the input signal and the impulse amount of time delay becomes large. (This is a disadvantage of the finite-
response sequence. She right-hand side of Equation (3-74) consists of k + 1terrns. impulse response filter compared with the infinite-impulse response filter.)
Thus, the output y ( k T ) is given in terms of the past k inputs x(O),
x ( ( k - 1 ) T ) and the current input x(kT). Notice that as k increases it is
not posible to process al1 past values of input to produce the current output. The digital filter discussed in Example 3-8 is a recursive filter
need to limit the number of the past values of the input to process. and realize it as a nonrecursive filter. Shen obtain the response
Suppose we decide to employ the N immediate past values of the inp to a cker delta input.
ividing the numerator of the recursive filter G ( z ) by the denominator, we
x ( ( k - 1 ) T ) ,x ( ( k - 2) T ) ,. . . , x ( ( k - N ) T ) and the current input x(kT). This
obtain
equivalent to approximating the right-hand side of Equation (3-74) by the
most recent input values including the current one, or
y ( k T ) = g ( O ) x ( k T ) i g ( T ) ~ ( ( k - 1 ) ? " ) + * ~ . + g ( N T ) x ( ( k - N ) T )(3-95
Since Equation (3-75) is a differen uation, the corresponding digital fálter ira tk
z plane can be obtained as foliows taking the z transform of Equation (3-75) By arbitrarily truncating this series at zW7,
we obtain the desired nonrecursive filter, as
follows:
we have
Y(.) = g(O)X(z) + g(T)z-'.X(z) i. - + ~ ( W T ) Z - ~ X ( Z ) (3-76)
Figure 3-46 shows the block diagram realization of this filter.

Figure 3-47 shows the block diagram for this nonrecursive digital filter. Notice that we
need a large number of delay elements to obtain a good leve1 of accuracy.
Noting that the digital filter is the z transform of the impulse response sequence,
the inverse z transform of the digital filter gives the impulse response sequence. By
taking the inverse z transform of the nonrecursive filter given by Equation (3-77), we
obtain
y ( k T ) = 2 x ( k T ) - 1.6x((k - 1 ) T ) + 0.8x((k - 2 ) T ) - 0.4x((k - 3 ) T )
+0.2x((k - 4 ) T ) - O.lx((k - 5 ) T ) + 0 . 0 5 ~ ( ( k- 6 ) T ) - 0.025x((k - 7 ) T )
For the Kronecker delta input, where x ( 0 ) = 1 and x ( k T ) = O for k # O, this last
equation gives
Y (0) = 2
Figure 3-46 Block diagram realization of Equation (3-76). y ( T ) = -1.6
aria! z-Piane Analvsis of Discrete-Time C o n t r ~Systems
l Chap. 3 Chap. 3 Exarnple Problerns and Solutions

Input and output curves


for a zero-order hoid.-

Obtain the expression for y(t). Then find Y ( s ) and obtain the transfer function
of the zero-order hold.
olution From Figure 3-49 we obtain
Figure 3-47 Block diagram for the digital filter given by Equation (3-77) (non- y ( t ) = x(O)[l(t)- l ( t - T ) ] + x ( T ) [ l ( t - T ) - l ( t - 2T)]
recursive form).
+ ~ ( 2 T ) [ l (-t 2 T ) - l ( t - 3T)I +
The Laplace transform of y ( t ) is

y(4T) = 0.2
y(5T) = -0.1
y(6T) = 0.05
y(7T) = -0.025
The impulse response sequence for this digital filter is shown in Figure 3-48.

where
m

The transfer function of the zero-order hold is thus

sequence for the digital filter given by


Equation (3-77). Consider a first-order hold preceded by a sampler. The input to the sampler is x ( t ) and
the output of the first-order hold is y(t). In the first-order hold the output y(t) for
+
kT 5 t < ( k l ) T is the straight line that is the extrapolation of the two preceding
sampled values, x ( ( k - l ) T ) and x ( k T ) , as shown in Figure 3-50. The equation for the
output y ( t ) is
t - kT
y ( t ) = -[ x ( k T ) - ~ ( ( -k 1)T)I
T
+ x(kT), kT 5 t < (k + l)T (3-78)
Consider a zero-order hold preceded by a sampler. Figure 3-49 shows the input x ( t )
to the sarnpler and the output y(t) of the zero-arder hold. In the zero-arder hold the Obtain the transfer function of the first-order hold, assuming a simple fianction
valrie ~f the last sample ic retained until the next sample Is talien. such as an impulse function at t = 0 as the input x(t).
e ,en 7-Plane Analvsis of Discrete-Time Control Systems Chap. 3 Chap. 3 Example Problems and Solutions

Since
X * ( s ) = T[x"((t)]= T[x(O)G(t)]= x(0)
the transfer function of the first-order hold is obtained as follows:

Consider the function

ure 3-50 Input and output curves


Show that s = O is not a pole of X(s). Show also that

&on. For an impulse input of magnitude x(0) such that x*(t) = x(0)6(t), the
,--+ . , / A (3-781 becomes as shown in Figure 3-51. The mathemat-
n ; x r ~ nhxr Fnila+inn has a simple pole at s = 0.
utisn Pf a transfer function involves a transcendental term eCTS,then it may be
replaced by a series valid in the vicinity of the pole in question.
For the function

Hence, let us obtain the Laurent series expansion about the pole at the origin. Since, in the
vicinity of the origin, e-Ts may be replaced by

substitution of Equation (3-80) into Equation (3-79) gives

which is the Laurent series expansion of X(s). From this last equation we see that S = O
is not a pole of X(s).
Next, consider Y ( s ) . Since
1-
Y ( s )=- s2
it may be expanded into the Laurent series as

We see that the pole at the origin (S = O ) is of order 1 , or is a simple pole.

Figure 3-51 Output curve of the first-


order hold when the input is a unit- Show that the Laplace transform of the product of two Laplace transformable functions
Slope = - -
T
#(O)
impulse function. f(t) and g ( t ) can be given by
z-Plane Analysis of Discrete-Time Control Cystems Chap. Chap. 3 Exampie Problems and Solutions

1 ciJm we have
.qf(t)g(t)l = , j-,-
i F(P)G(s - P) dP
SoiutB~sn The Lapiace transform of the product of f(t) and g(t) is given by

%[f (0s('11 = jomf(t)g(')e-s' ' f t

Note that the inversion integral is

f(r) = LjcYm
277-j
F(s)efl ds,
c-lm
t >O we have
1
where c is the abscissa of convergence for E(s). Thus, G(s - p ) =
1-e-~(~-~)

.qf(f)g(t)l =
1
T;;;j" m
I,,.C+Jm
F(p)eP'+ &)e--" dt
Notice that the poles of 141 - e-T(s-P'] may be obtained by solving the equation

Because of the uniform convergence of the integrals considered, we may invert the
order of integration:
-T(s - p ) = k j 2 7 ~ k , k = O,1,2 , . . .
so that the poles are
Noting that 2~r
p=skj-k=skjwsk, k = O , 1 , 2 ,...
T
j~g(i)e-'s-p"dt = G(s - p )
where ws = 2 d T . Thus, there are infinitely many simple poles along a line parallel to
the jw axis.
we obtain
The Laplace transform of x " ( t ) can now be written as
1
%If(t)g(t)l = ,ijc-jm
F(P)W
c+im
- P) dp

- -1 "+'" 1
-
2Trjj,-,m
X(p) 1 - e-ns - PI d~ (3-86)
Show that the Laplace transform of
m m
where the integration is along the line from c - j w to c + jw, and this line is parallel
~ " ( t=
) x(t)6(t - kT) = ~ ( t ) 6(t - kT) to the imaginary axis in the p plane and separates the poles of X ( p ) from those of
k=o k=o 1/[1 - e-T(s-P)].Equation (3-86) is the convolution integral. It is a well-known fact that
can be given by such an integral can be evaluated in terms of residues by forming a closed contour
consisting of the line from c - j w to c + j w and a semicircle of infinite radius in the
left or right half-plane, provided that the integral along the added semicircle is a
constant (zero or a nonzero constant). There are two ways to evaluate this integral (one
using an infinite semicircle in the left half-plane and the other an infinite semicircle in
the right half-plane); we shall consider these two cases separately in Problems A-3-6
and A-3-7.
ution Referring to Equation (3-83), rewritten as
1 c+lm
%[f(t)g(t)l = ,ijc-,-
F(P)G(s - P) dp
Referring to Equation (3-86), rewritten as
where
m
1
x w = ,ij,-,m
'+'" 1
X ( p I 1 - e-T(s-,>d~
f(t) = x(t) and g(t) = 6(t - kT)
k=o show that, by performing the integration in the left half-plane, X" (S) may be given by
and noting that
2 [ 8 ( t - k T ) ] = e-"'
z-Piane Analysís of Discrete-Time Control Systerns Chap. 3 Chap. 3 Example Problerns and Sol~atíons

By substituting z for eTs in Equation (3-87), we have Im


- -
X(z) = [residue of -
X(p)z at poie of ~ ( p )
z - eTp
By changing the compIex variable notation from p to S, we obtain
1 X p plane

X(z) = [residue of
1
where we assurned that X(z) has h different multiple poles and m - h simple poles
assume that the poles of X(s) lie in the left half-plane and that X(s) can
be expressed as a ratio of polynomials in S , or

where q (S) and p (S) are polynomials in s . We also assume that p (S) is of a higher degtee
in S than q(s), which means that
lirnX(s) = O
s-+m

shall evaluate the convolution integral given by Equation (3-86) using


a closed contour in the left half of thep plane as shown in Figure 3-52. Using this closed
Figure 3-52 Closed contour in the left half of the p plane.
contour, Equation (3-86) may be written as

By substituting z for eTs in Equation (3-89), we have

where the closed contour consists of the line from c - jm to c + jm and rL9 which án
X(z) =
[residue of -----
- eTp at p o i of i.(p j
zX(piz

By changing the complex variable notation from p to S, we obtain


turn consists of a semicircle of infinite radius and the horizontal lines at jm and -jm,
which connect the line from c - j m to c + j m with the semicircle in the left half of the
choose a value of c such that al1 the poles of X ( p ) lie to the left of the iine
from c - j a to c + jm and al1 the poles of 1/[1- e - T ( s - P ) ] lie to the right of this line. Let us assurne that X(s) has poles si, s2, . . . ,S,. Zf a pole at s = S, is a simple pole,
The closed contour endoses al1 poles of X(p), while the poles of 1/[1- e-T("p)] are then the corresponding residue Kiis
outside the closed contour.
Because we have assumed that the denominator of X(s) is of a higher degree in
s than the riurnerator, the integral along TL(the infinite semicircle in the Ieft half-plane
plus the horizontal lines at j m and -jm, which connect the line frorn c - jm to c + jm If a pole at S = si Is a multiple pole of order ni, then the residue Kiis
with the semicircle) vanishes. Hence,

Therefore, if X(s) has a multiple pole sl of order nl, a multiple pole s2of order n2, . . . ,
This integral is equal to the sum of the residues of X ( p ) in the closed contour. (Refer a multiple pole sh of order nh, and simple poles s,,+I,~h+2, . . . ,sm, then X(z) given by
to Appendix B for the residue theorem.) Therefore, Equation (3-90) can be written as
X(P>
1 - e-T('-p) at pele 0f X(~)
1 (3-89) h(z = 2 [residue of -X(s)z at pole of ~ ( s )
z - eTs 1
nalycis of Discrele-Time Control Chap. 3 Chap. 3 Example Problemc and Solutions

p plane
(3-93)

where n, is the order of the multiple pole at s = si.

Referring to Equation (3--86), rewritten as

show that by performing this integration in the right haif p plane, X* (S) may be given

C---y--J
Poles o f

X(P)
provided that the denominator of X(s) is two or more degrees higher in s than the
numerator. Show that if the denominator of X(s) is only one degree higher in s than
the numerator then

Solution Eet us evaluate the convolution integral given by Equation (3-86) in the right
half of the p plane. Let us choose the closed contour shown in Figure 3-53, which
+
consists of the line from c - j w to c jm and r R ,the portion of the semicircle of infinite ure 3-53 Closed contour in the right half of the p plane.
radius in the right half of thep plane that lies to the right of this h e . The closed contour
encloses al1 poles of 141 - e-T("p)], but it does not enclose any poles of X(p). Now
X"(s) can be written as
then the integral along TRis zero. Thus, in the present case

Therefore, Equation (3-96) simplifies to


Let us investigate the integral along r R , the portion of the infinite semicircle to
the right of the line from c - j m to c + jm. Since infinitely many poles of 1/[E - e-T(s-P']
lie on a line parallel to the jw axis, the evaluation of the integral along TRis not as simple
as in the previous case, where the closed contour enclosed a finite number of poles of The integral along the closed contour given by Equation (3-97) can be obtained by
X ( p ) in the left half of the p plane. evaluating the residues at the infinite number of poles at p = s k jws k. Thus,
In almost al1 physical control systems, as s becomes large, X(s) tends to zero at
kast as fast as 11s. Hence, in what follows, we consider two cases, one where the
denominator of X(s) is two or more degrees higher in s than the numerator and another
where the denominator of X(s) is only one degree higher in s than the numerator. The minus sign in front of the right-hand side of this last equation comes from the fact
that the contour integration along the path TRis taken in the clockwise direction. Using
Case 1: X(s) Possesses at Least Two More Poles Than Zeros. Referring to the theory L'Hopital's rule, we obtain
of complex variables, it can be shown that the integral along TRis zero if the degree
of the denominatorp(s) of X(s) is greater by at least 2 than the degree of the numeralor
q(s); that is, if X(s) possesses at least two more poles than zeros, which implies that
Iim sX(s) = x (O + ) =O
S- =
z-Plane Analysis of Discrete-Time Control Systems xample Problems and

Noting that

Weferring to a formula available in mathematical tables,

Thus,
and noting that

Note that this expression of the z transform is useful in proving the sampling theorem
(see Section 3-4). However, it is very tedious Lo obtain z transform expressions of we can rewrite Equation (3- 101) in the form
commonly encountered functions by this method.
Case 2: X ( s ) Has a Denominator One Degree Higher in s Than the Numerator. For x * (S
this case limw,sX(s) = x(O+) # O < m and the integral along TRis not zero. [The
nonzero value is associated with the initial value x(O+) of x(t).] It can be shown that
the contribution of the integral along PR in Equation (3-96) is -$x(O+). That is,

Then the integral term on the right-hand side of Equation (3-96) becomes

Thus, we have obtained X ( z ) by using the convolution integral in the right half-plane.
Consider the function [This process of obtaining the z transform is very tedious because an infinite series of
X(s + jws k ) is involved. The example here is presented for demonstration purposes
only. One should use other methods for obtaining the z transform.]
Qbtain A7(z) by using the convolution integral in the right half-plane.
Solution The Laplace transform of x ( t ) is
Obtain the z transform of
S
X(s) =
Clearly, iim-,sX(s) = x(O+) = 1, or the function has a jump discontinuity at t = 0. (S + q 2 ( s + 2)
Hence we must use Equation (3-95). Referring to this equation, we have by using (1) the partial-fraction-expansion method and (2) the residue method.
Solntlopa
1. Partial-fraction-expansion method. Since X ( s ) can be expanded into the form

qs) 2 1 2
= -- -- -
s+1 ( ~ + 1 ) S~+ 2
z-Plane Analysis of Discrete-Time Control Chap. 3 Example Problems and Solutions

By substituting jw for s in Ghl(s),we obtain

- Tjw t 1 -,
-- 4, sin2(Twl2)
T e o2

= tan-' Tw - To

where we have used the relationship T = 2 d w S .


At a few selected values of o , we have
IGh,(jo)/ = T /Ghl(jO) = O0

Figure 3-54 shows plots of the magnitude and phase characteristics of the first-order
hold and those of the zero-order hold. From Figure 3-54 it is seen that both the
zero-order hold and the iirst-order hold are not quite satisfactory low-pass filters. They
allow significant transmission above the Nyquist frequency, wh. = d T . It is important,
Magnitude and phase characteristics of the first-order hold and those
therefore, that the signal be low-pass-filtered before the sampling operation so that the
of the zero-order hold.
frequency components above the Nyquist frequency are negligible.

Iri terms of the z transform notation, we have

1 - e-T"
Y ( s ) = G ( s ) X *( S ) = -x*( 4 where

Show that
Y * ( s )= X * ( s )
Solutian By taking the starred Laplace transform of Equation (3-102), we have

Gís) Figure 3-55 Zero-order hold.


z-Plane Analysis of Discrete-Time Control hap. 3 Example Problems and Solutions

ence, utlon From the diagram we have


Y ( z )= X ( z )
Ln terms of the starred Laplace transform notation, this last equation can be written as M ( $ ) = Gl(s)E(s)
E ( s ) = R ( s ) - H(s)C(s)
ence,
Obtain the weighting sequence of the system defined by

Taking tbe starred Laplace transform of this last equation, we obtain


for n = 2 , 2, and 3, respectively.
M* ( S ) = [Gi R(s)]* - [Gi G,H(s)]*
oleation For n = 1, we have

Hence, the weighting sequence g,(k) is found to be


g ~ ( k=
) (-0)"

For n = 2, we obtain
1 - 1 - az-' + a 2 z - 2 - a 3 z T 3+ .. e

G$(Z)= - In terms of the z transform notation,


( 1 + az-l)' 1 4- az-'
= 1 - 2aZ-i + 3aZZ-' - 4a32-3 + - ..

Kence, the weighting sequence & ( k ) is


This last equation gives the discrete-time output C(z).
g4k) = (k -i- ~)(-a)~ The continuous-time output @ ( S ) can be obtained from the following equation:
For n = 3, we get

Notice that [CiR (s)]*/{I+ [Gl G2N ( s ) ] * )is a series of impulses. The continuous-time
output C ( s ) is the response of G2(s)to the sequence of such impulses. [See Problem
ence, the weighting sequence g 4 k ) is A-3-18 for details of áetermining the continuous-time output ~ ( t the ) , inverse Laplace
transform of C(s).]

Consider the svstem shown in Figure 3-57. Obtain the closed-loop pulse transfer
Obtain the discrete-time output C ( z )of the closed-loop control system shown in Figure function G(z)lR(z).Also, obtain the expression for @ ( S ) .
3-56. Also, obtain the continuous-time output C(s).

Figure 3-56 Discrete-time control system. Figure 3-57 Discrete-time control system.
1 r-Plane Analysis of Discrete-Time Control Systems Chap. 3 Chap. 3 Example Problems and Solutions

olution From the diagram we have Compare the polar plots (frequency-response characteristics) of the analog PID
C(s) = Gz(s)M*(s) controller with those of the digital PID controller.
ution For the analog PID controller, the frequency-response characteristics can be
M(s) = Gl(s)E*(s)
obtained by substituting j o for s in G(s). Thus,
E(s) = R(s) - N(s)C(s) = R(s) - N(s)G~(s)M*(s)
Taking the starred Laplace transforrns of both sides of the last three equations gives

rdys) = GT(S)E*(S)
E*(s) = R*(s) - HG2(s) For the digital PID controller, the frequency-response characteristics can be
obtained by substituting z = e'"= into Go(z):
Solving for C*(S) gives

= Kp + 1 - c o s o T&+ jsinwT
+ &(1 - cos wT +j sin wT)

sin wT
= K p + 5 2( l - j
1 - coswT
+ KD(l - cos wT + j sin wT) (3-104)

shall first compare separately the P action, the P action, and the D action of
controller with their counterparts in the digital controller. Notice that in the
proportional action (P action) the digital controller has a gain KI/2 less than the
corresponding gain in the analog controller, since Kp = K - SKI. See Figure 3-%(a).
In terms of the z transform notation, we have For the integral action (1 action) the real parts of the polar plots of the analog
controller and digital controller differ by KJ2, as shown in Figure 3-5$(b).
- -
C(z)
- Gl(~)G2(4 hen the proportional action and integral action are combined, then the real
R(z) 1 + Gl(z)HG,(z) parts of the polar plots for the analog P I action and the digital PI action become the
The continuous-time output C(s) can be obtained frorn the following equation: same, as shown in Figure 3-58(c).
The polar plots of the derivative action ( D action) for the analog controller and
the digital controller differ very much, as shown in Figure 3-58(d). Hence, there are
considerable differences in the analog D action and the digital D action.
The qualitative polar plot of the analog PHD controller can be obtained from
Equation (3-103) by varying o from O to m, as shown in Figure 3-59(a). Similarly, the
Consider the analog PID controller and the digital PXD controller. The equation far the qualitative polar plot of the digital PID controller can be obtained from Equation
analog PID controller is (3-104) by varying o from O to n-lT, as shown in Figure 3-59(b).
? .. Note that, although the polar plots of the analog PI controller and the digital PI
controller are similar, there are significant differences between the polar plots of the
analog PID controller and the digital PXD controller.
where e(t) is the input to the controller and m(t) is the output of the controller. The
transfer function of the analog PID controller is

In Section 3-5 we derived the pulse transfer function for the PID controller inpositional
The pulse transfer function of the digital PID controller in the positional form form. Referring to Figure 3-28, the pulse transfer function for the digital PID controller
is as given by Equation (3-56): was derived as

Using Pm(kT) = m(kT) - m((k - 1)T) derive the velocity-form P%Dcontrol equation.
z-Plane Analysis of Discrete-Time hap. 3 Example Problems and

Analog P action Digital P action

Analog I action Digital 1 action Analog PID controller Digital PID controller

(a) Polar plot of analog PXD controller; (b) polar pIot of digital PID
controller.

Analog P/ action Digital P I action

= K p [ e ( k T )- e ( ( k - l)?")] + K I e ( k T )
+ K D [ e ( k T )- 2e((k - 1)T ) + e ( ( k - 2)T ) ] (3-105)
where we have used the relationships K p = K - 4 K I , & = KTII;, and KD = KTdlT.
(For these reiationships, refer to the derivations of the positional form of the digital $ID
control equation.) Equation (3-105) takes into consideration the variation of the
positional form in one sampling period.
Analog D action Digital D action Suppose the actuating error e ( k T ) is the difference between the input r ( k T ) and
Polar plots of analog and digital controllers with (a) proportional the output c ( k T ) , or
action, (b) integral action, (c) proportional plus integral action, and (d) derivative e(kT) = r(kT) - c(kT)
action.
By substituting this last equation into Equation (3-105), we obtain
V m ( k T ) = K p [ r ( k T )- r ( ( k - 1 ) T ) - c ( k T ) + c ( ( k - 1)T)l
+ & [ r ( k T ) - c ( k T ) ] + K D [ r ( k T )- 2r((k - 1 ) T )
-t r ( ( k - 2 ) T ) - c ( k T ) + 2c((k - 1 ) T ) - ~ ( ( - k 2)IF)I (3-106)
z-Plane Analysis of Diccrete-Time Control Systems Chap. 3 ap. 3 Example Problems and

The velocity-form PID control scheme given by Equation (3-106) may be modified ints
a somewhat different form to cope with sudden large changes in the set point, Siince
the proportional and derivative control actions produce a large change in the controlker
output when the signal entering the controller makes a sudden large change, to suppress
such a large change in the controller output, the digital proportional and derivative
terms may be modified as discussed next.
If changes in the set point [input r(kT)] are a series of step changes, then - -

immediately after a step change takes place, the input r(kT) stays constant for a w (a)
until the next step change takes place. Hence, in Equation (3-106) we assume that

(Note that this is true if the input stays constant. But we assume that this holds true
even if a step change takes place.) Then Equation (3-106) rnay be modified to

The z transform of Equation (3-107) gives

Simplifying, we obtain

Equation (3-108) gives the velocity-form PID control scheme. The block diagram
realization of the velocity-form digital PIE) control scheme was shown in Figure 4-30.
Output points obtained
1.5 by sarnple calculations

he system shown in Figure 3-60(a). Obtain the continuous-time output c(t)


so that the output between any two consecutive sampling instants can be determined.
Find the expression fo-r the continuous-time output c(t). The sampling period Tis 1sec.
ution For the system shown in Figure 3-60(a), we have
C(s) = G(s)E*(s)
E(s) - R(s) - C(s)
Hence,

Figure 3-60 (a) Discrete-time control system; (b) plots of individual impulse
or responses; (c) plot of continuous-time output c ( t ) versus t.

Thus, c(t)

For the present system,


= (e--l[c(s)] = (e-'
[
G(s)
1:Z;$J

The continuous-tiae output c(t) can therefore be obtained as the inverse Laplace
transform of C(s):
z-Piane Analysis of Diserete-Time Control Chap. 3 le Problems and Solulions

ence, Figure 3-60(b) shows plots of individual impulse responses given by Equation (3-109).
F. 7
[Observe that c(t) consists of the sum of impulse responses that occur at t = O, t = 1,
t = 2, . . . with weighting factors 1, -0.3679, -0.6321, . . . .]
From Equation (3-109) we see that for time intervals O 5 t < 1 , l r t < 2,
Eet us define 2 r t < 3 , . . . the output c(t) is the sum of impulse responses as follows:
t - 1 + e-', Ost<l
( t - 1 + e-') - 0.3679[(t - 1) - 1 + e-('-')]l(t - l), 15 t < 2
e-') - 0.3679[(t - 1) - 1 t- e-"-"]l(t - 1)
Then the z transform expression for this last equation is
-0.6321[(t - 2) - 1 + e-''-2)]1(t - 2), 25t<3

Sample calculations for several values of t follow:


Referring to Equation (3-58) for the z transform of G(s), we obtain
1 '
1 - z-'
X(z) = (1 - z-')
' 0.3679~-' + 0 . 2 6 4 2 ~ - ~
(1 - 0.3679~-')(1 - z-')
-
- 1 - 1.3679~-1+ 0 . 3 6 7 9 ~ - ~
1 - z-' + 0 . 6 3 2 1 ~ - ~
Hence, noting that the sampling period T is 1 sec or T = 1, we have
1 - 1.3679e-" + 0.3679e-2"
X*(s) =
1 - e-" + 0.6321e-&
Therefore ,
1 1 - 1.3679e-" + 0.3679e-&
c(t, = +
T1[-s2(s 1) 1 - e-" + 0.6321e->

(1 - 0.3679e-' - 0.6321eT2 - 0.3996eP3"

+ 0e-4" + 0.2526e-~"+ 0.2526e-& +


The continuous-time output c(t) thus obtained is plotted in Figure 3-60(c).
Since
-- 1 - 1+_ 1
l --
s2(s + 1) s2 S S + 1 Consider the digital filter defined by
the inverse Laplace transform of this last equation is
r -I

Draw a series realization diagram and a parallel realization diagram. (Use one first-
Hence, we obtain order section and one second-order section.)
Solutiolra We shall firct consider the series realization scheme. To limit the coefficients
c(t) = ( t - 1 + e-') - 0.3679[(t - 1) - 1 + e-('-"]l(t - 1)
to real quantities, we group the second-order term in the numerator (which has complex
zeros) and the second-order term in the denominator (which has complex poles).
Wence, we group G(z) as follows:

(5-109 Figure 3-61(a) shows a series realization diagram.


P-Plane Analysis of Discrete-Time Control Systems Chap. 3 Example Problems and Solutions i

Next, we shall consider the parallel realization scheme. Expansion of G(z)/z into
partial fractions gives

Then G(z) can be written as follows:

Figure 3-61(b) shows a parallel realization diagrarn.

A slowly changing continuous-time signal x(t) is sampled every T sec. Assume that
changes in signal x(t) are very slow cornpared to the sampling frequency. Show that in
the z plane (1 - z-l)lT corresponds to "differentiation," just as s corresponds to
"differentiation" in the s plane.
antloa For a siowly changing signal x(t), the derivative of x(t) can be approxirnated
by

The z transform of this equation gives


1 1
V(z) = - [X(z) - z-l %(z)] = - (1 - z-l)X(z)
T T
from which we obtain the block diagram shown in Figure 3-62(a). This diagram
corresponds to the s plane differentiation shown in Figure 3-62(b). Note tRat the block
diagrarn shown in Figure 3-62(a) can be rnodified to that shown in Figure 3-63. (For
approxirnation of "integration" in the z plane, see Problems -3-25 through B-3-27.)

se 3-62 (a) Block diagram for "differentiation" in the z plane; (b) block
diagram for "differentiation" in the s plane.

P Block diagram reaiizations of the digital filter considered in Problem


A-3-19. (a) Series realization; (b) parallel realization.

Figure 3-63 Approximate


"differentiator" in the z plane.
r-Plane Analysis of Discrete-Time Control ystems Chap. 3 Chap. 3 Probterns

-3-
Consider a transfer function system
, ~ 2

Show that the circuit shown in Figure 3-64 acts as a zero-order hold.
Obtain the pulse transfer function by two different methods.

Obtain the z transform of

Use the residue method and the method based on the impulse response function.

Obtain the z transform of


R, << R, 1 - e-*" 1
X ( s ) = -----
S ( S + a)"
4 Circuit approximating a zero-order hold.

Consider the difference equation system


-2
Consider the circuit shown in Figure 3-65. Derive a difference equation describing the
system dynamics when the input voltage applied is piecewise constant, or where y (O) = O . Obtain the response y ( k )when the input x ( k )is a unit-step sequence.
Also, obtain the MATLA
e(t) = e(kT), kT 5 t < (k l)T
i-

Consider the difference equation system

where y ( k ) = O for k < O. Obtain the response y ( k ) when the input x ( k ) is a unit-step
sequence. Also, obtain the MATEAB solution.

O I C9 Obtain the weighting sequence g ( k ) of the system described by the difference equation
Figure 3-65 RC circuit . y(k)-ay(k-1)=x(k), -1<a<1
If two systems described by this last equation are connected in series, what is the
(Derive first a differential equation and then discretize it to obtain a difference equa- weighting sequence of the resulting system?
tion.)

Consider the system described by


Consider the impulse sampler and first-order hold shown in Figure 3-66. Derive the y ( k ) - y(k - 1) + 0.24y(k - 2) = x ( k ) + x ( k - 1)
transfer function of the first-order hold, assuming a unit-ramp function as the input x ( t )
to the sampler. where x ( k ) is the input and y ( k ) is the outpur: of the system.
Determine the weighting sequence of the system. Assuming that y ( k ) = O for
k < O, determine the response y ( k ) when the input x ( k ) is a unit-step sequence. Also,
obtain the MATLAB solution.
First-order
Figure 3-66 Impulse sampler and first- fioblem B-3-31
L order hold. Consider the cystern
r-Plane Analysis of Biccrete-Time Control kap. Problems

Obtain the closed-loop pulse transfer function of the system shown in Figure 3-68.
Obtain the response of this system to a unit-step sequence input. Also, obtain the
MATH,AB solution.
Zero-order K
-
hold s+a

Obtain the response y(kIT) of the following system:

where x ( t ) is the unit-step function and x" ( t ) is its impulse-sampled version. Assume
that the sampling period T is 0.1 sec. Discrete-time control system.

Consider the system defined by


Consider the discrete-time control system shown in Figure 3-69. Obtain the discrete-
time output C ( z ) and the continuous-time output C(s) in terms of the input and the
Using the convolution equation transfer functions of the blocks.

obtain the response y ( k ) to a unit-step sequence input u(k).


-3-1
Assume that a sampled signal X*(s) is applied to a system G(s). Assume also that the
output of C ( s ) is Y ( s ) and y(O+) = 0.

Using the relationship

Discrete-time control system


show that

Consider the discrete-time control system shown in Figure 3-70. Obtain the output
Obtain the closed-loop pulse transfer function of the system shown in Figure 3-67 sequence c ( k T )of the system when it is subjected to a unit-step input. Assume that the
sampling period T is 1 sec. Also, obtain the continuous-time output c(t).

Figure 3-67 Discrete-time control system. Figure 3-70 Discrete-time control system
z-Plane Analysis of Discrete-Time Control Systems Chap. 3

-3-1 Wealize this digital filter in the series scheme, the parallel scheme, and the ladder
scheme.
Obtain in a closed form the response sequence c(kT) of the system shown in Figure 3-71
when it is subjected to a Kronecker delta input r(k). Assume that the sampling period
T is 1 sec.
Referring to the approximate differentiator shown in Figure 3-63, draw a graph of the
output v(k) versus k when the input x(k) is a unit-step sequence.

Consider the system shown in Figure 3-73. Show that the pulse transfer function
Y(z)lX(z) is given by

Figure 3-71 Discrete-time control system.

Consider the system shown in Figure 3-72. Assuming that the sampling period Tis 0.2
sec and the gain constant K is unity, determine the response c(kT) for k = 0, 1, 2, 3,
and 4 when the input r(t) is a unit-step function. Also, determine the final value c(m).

Figure 3-73 Digital integrator without delay.

Assuming that y(kT) = O for k < O, show that

Thus, the output y(kT) approximates the area made by the input. Hence, the system
acts as an integrator. Because y(0) = Tx(O), the output appears as soon as x(0) enters
the system. This integrator is commonly called a digital integrator without delay.
use 3-72 Discrete-time control system. Draw a graph of the output y(kT) when the input x(kT) is a unit-ctep sequence.

Obtain the closed-loop pulse transfer function C(z)IR(z) of the digital control systern Consider the system shown in Figure 3-74. Show that the pulse transfer function
shown in Figure 3--30. Assume that the pulse transfer function of the plant is G(z). Y(z)/X(z) is @ven by
(Note that the system shown in Figure 3-30 is a velocity-form PPD control of the plant.)

Assume that a digital filter is given by the following difference equation:

Draw block diagrams for the filter using (1) direct programming, (2) standard prograrn-
rning, and (3) ladder programming.

Figure 3-74 Digital integrator with deiay.


Consider the digital filter defined by
Assurning that y(kT) = O for k < O, show that
y(kT) = T[x(O) + x(T) + .. + x((k - 1)T)I
The output y(kT) approximates the area made by the input. Since y(0) = O and
y(T) = 7jc(0), the output starts to appear at t = T. This integrator is called a digital
integrator with delay.
Draw a graph of the output y ( k T ) when the input x ( k T ) is a unit-step sequen

Consider the system shown in Figure 3-75. Show that the pulse transfer function
Y ( z ) / X ( z )is given by

Figure 3-75 Dipital bilinear integrator.

transfer function o£ the digital. controller.


Design techniques for continuous- trol systems based on conventional

summarizes transient and steady-state response characteristics of discrete-time con-


trol systems. The design technique based on the root-locus rnethod is presented in
Section 4-5. Section 4-6 first reviews the frequency-response method and then
presents frequency-response techniques using the w transforrnation for designing
discrete-time control systems. Section 4-19 treats an analytical design method.
173
esign of Discrete-Time Control Systems by Canventional Methods Cha ec. 4-2 Mapping Between the s

s. Note that since/z =


of z m to m. Consider a represen
The absolute stability and relative stability of the linear time-invariant continuous- on the jw axis in the s plane. As this point moves from -j2 asto j i o, on
time closed-loop control syste ocations of the closed-loop ere o, is the sampling fre uency, we have / z/ = 1, a n d b varies from - 71 to 71
poles in the s plane. For example, peles in the left half of the in the counterc1ocl;wise direction in the z plane. As the re
s pXane near the j u axis will exhib , and closed-loop poles s n from j i w, to j$ w, on the jw axis, the corresponding
ative real axis will exhibit the unit circIe once in the counterclockwise direction.
mce the complex variables z = eTs,the poje and zero moves from -m to m on the j o axis, we trace the unit
locations in the z plane are related to the pole arid zero locations in the s plane. number of times. Frsm this añialysis, it is clear that each
Therefore, the stability of the linear time-invariant discrete-time closed-loop system ane maps into the inside of the unit circle i
can be deterrnined in terms of the locations of the poles of the closed-loop pulse If of the s plane may be divid nt-5 an infinite n ~ b e of
r periodic
transfer function. It is noted that the dynamic behavior of the discrete-time control ian Figure 4-1. The primary s extends from jw.5 =: -j ose 0 j+
system depends on the sampling period T. In terms of poles and zeros in the z plane, ntary strips extend from ji j $ % , 2%to I ~ w ,. . , and fr
eir locations depend on the sampling period T. In other w , -j2ws to - j $ % ? . . . .
sampling period Tmodifies the pole and zero Ioeations in the In the primary strip, if we trace the sequence of points 1-2-3-4-5-1 in the s
response behavior to change. plane as shown by the circled numbers in Fig
]te centered at the origin of th
ne. ]en the design of a ng points 1, 2, 3, 4, and 5 in
are very important in predictirig the dynamic behavior of t nurnbers in Figure 4-2(b).
designing discrete-time control systems, the locations of th The area enclosed by any of the compiementary st
z plane are very important. In the following paragraphs we shall investigate how tbe unit circle in the z plane. This means that the correspo
locations of the poles and zeros in the s plane compare with the locations of the poies

rporated into the process, the co

Complementary
strip
s plane

This means that a pole in the s plane can be located in the z plane through ihe
1m
trancformation z = e". Cince the complex variable s has real part o and imaginary z plane
part w , we have Complementary
strip

and

From this last equation we see that poles and zeros in the s plane, where frequencies
differ in integral multiples of the sampling frequency 2 d T , are mapped into the same
locations in the z plane. This means that there are infinitely many values of s for each
Complernentary
strip $j
value of z .
Since u-is negative in the left half of the s plane, the left half of the s plane
corresponds to
Izl = e T u < 1
23-xjw axis in the s plane corresponds to zl = 1. That is, the imaginary axis in the
Sp h e (ihe line o- = O) corresponds to the unit circle in the z plane, and the interior Figure 4-1 Periodic strips in the s plane and the corresponding region (unit circle
of the unit circle corresponds to the left half of the s plane. centereti at the origin) in the z piane.
esign of Diccrete-Time Control ystemc by Conventional Methods Chap. 4

/z = 271. X 0.5= n== 1800


Thus, the spiral can be graduate
Figure 4-7(b)]. Once the sam
wd a1 any point on the spi
4-7(b), wd can be determi
specified as w, = 1071. radtsec, then at point 69

Hence, wd at point P is

Note that if a constant-damping-ratio line is in the second or t


in the s plane then the spiral decays within the unit circle in the z pl
if a constant-damping-ratio line is in the first or fourth quadrant in th
corresponds to negative damping), then the spiral grows outside
Figure 4-8 chows constant-damping-ratio loci for i= 0, c = 0.2,5 = 0.4,i= 0.6,
( = 0.8, and { = 1. The [ = 1 locus is a horizontal line between points r = O and
z = 1. (Note that Figure 4-8 shows only the loci in the upper half of the z
4 --S
which correspond to O 5 w 5 ms. 'F e loci corresponding to o, r 5 0 are the
rnirror irnages of the loci in the upper half of the z plane about the horizontal axic.)
Wotice that the constant {locj are normal to the constant wnloci in the S plane,
as shown in Figure 4-9(a). In the z plane rnapping, constant w, loca intersect constant
i spirais at right ancles, as shown in Figure 4-9(b). A mapping such as this, wbich
preserves both the size and the cense of angles, is called a conforma1 rnupping.

Im

z plane

-1.0 -0.5 O l= 1 0.5 1 .O Re

Figure 4-8 Constant-damping-ratio loci in the z plane.


ysternc by Conventional Methods C

z plane
W = W,

Im z plane

Figure 4-11 (a) A desirable region in the s plane for closed-loop pole locations;
(b) corresponding region in the z plane.

On the basis o£ the preceding discussions on mapping fsom the s plane to the z
plane, the desirable region can be mapped to the z plane as in Figure 4-11(b).
Note that if the dominant closed-loop poles of the continuous-time control cys-
tem are required to be in the desirabie region specified in the s plane, then the domi-
nant closed-loop poles of the equivalent discrete-time control systern must lie inside the
region in the z plane that corresponds to the desirable region in the s plane. Once the -12 (a) Diagram showing s plane poles at -u, k jw, and folded poies
discrete-time control system is designed, the system response characteristics must be appearing at -fi + j(w, + w,), - al IT j(wl F 2w,), . . . ; (b) z plane mapping of s
checked by experiments or sirnulation. If the response characteristics are not satisfac- plane poles at -a1 + jw,, - 02 I+ j(wl + u,), - cr, + j(wl 2 2w,), . . .
tory, then closed-loop poIe and zero locations must be modified until satisfactory results
are obtained.
entsLI, For discrete-time control systems, it is necessary to pay particular
attention to the sampling period T. This is because, if the sampling period is tos 1ong The stability of the system defined by Equation (4-3), as well as of other types of
and the sampling theorem is not satisfied, then frequency folding occurs and fhe discrete-time control systems, rnay be determined from the Iocations of the closed-
effective pole and zero locations will be changed. loop poles in the z plane, or the roots of the characteristic equation
Suppose a continuous-time control system has closed-loop po
s = - i jwl in the s plane. If the sampting operation is involved in this syst
4
if o1 > o,, where ws is the sampling frequency, then frequency folding occ
the system behaves as if it had poIes at s = -ul i j(wl t nw,), where n = 1,2,
3, . . . . This means that the sampling operation folds the poles outside the primar 1. For the system to be stable, the closed-ioop poles or the roots of the charac-
strip back into the primary strip, and the poles will appear at s = --ul i j(wl - @S teristic equation must lie within the unit circie in the z plane. Any closed-loop
see Figure 4-12(a). On the z plane those poles are mapped into one pair of conjuga pole outside the unit circle rnakes the system unstable.
complex poles, as shown in Figure 4-12(b). hen frequency folding occurs, oscdla- If a simple pole lies at z = 1, then the system becomes critically stable. Also,
eions with frequency o, - o l , rather than frequency a l , are observed. the system becomes critically stable if a single pair of conjugate complex poles
lies on the unit circle in the z plane. Any multiple closed-loop pole on the unit
circle makes the system unstable.
3. Closed-loop zeros do not affect the absolute stability and therefore may be
located anywhere in the z plane.

Thus, a linear time-invariant single-input-single-output discrete-time closed-


tbe stabiiity of linear time-invariant single-input-single-output discrete-time control loop control systern becomes unstable if any of the ciosed-loop poles lies outside the
systems. Cvnsider the following dosed-loop pulse-transfer function sg~stern: unit circle andlor any inultiple closed-loop pole lies on the unit circle in ihe z piane.
nalysís of Closed-loop ysterns in the z Plane

reveal the existence of any unstable roots (the roots that lie outside the unit circle
Consider the closed-loop control systern shown in Figure 4-13. Determine the stability in the z plane). However, these tests neither give the locations of unstable roots nor
of the system when K = 1. The open-loop trancfer function G ( s ) of the system is e the effects of parameter changes on the sy ability, except for the

Referring to Equation (3-58), the z transforrn of G ( s ) is

Since the closed-loop pulse transfer function for the cystem 1s

always real, the Jury test is preferred to the Schur-Cohn test.


the characteristic equation is ty Test. Hn applying the Jury stability test to a given charac-
1 + G ( z ) =O n P(z) = O, we construct a table whose ele
which becomes coefficients of P(z). Assume that the characteristic equatio
( z - 0.3679)(z - 1) + 0.36792 + 0.2642 = O in z as follows:

where a. > O. Then the Jury table becomes as given in 'Hable 4-1.
The roots of the characteristic equation are found to be Notice that the elements in the first row consist of the coefficients in P(z)
arranged in the ascending order of powers of z . The elements in the second row
ZI = 0.5 + j0.6181, z2 = 0.5 - j0.6181
consist of the coefficients of (z) arranged in the descending order of powers of z.
The elements for rows 3 through 2n - 3 are given by the following determinants:

the system ic stable.


4-1 GENERAL FORM OF THE JURY STABILITY TABLE

It is important to note that in the absence of the sampler a second-order syskem Row z0 z1 z z ... ~ n - 2 Zn-i Zn
is always stabie. In the presence of the sampler, however, a second-order system such
as this can becorne unstable for large values of gain. n fact, it can be dmwn that
the second-order system shown in Figure 4-13 will become unstable if K > 2.3925.
(See Example 4-7.)
ii@. ñhree stability tests can be applied
direc (z) = O without solving for the roots. Two
of them are the Schur-Cohn stability test and the Jury stability test. These two tests

Fignre 4-13 Closed-loop control system of Example 4-2.


esign of Discrete-Time Control Systerns by Conventianal Methods Ckap. 4 Cec. 4-3 Stability Analysis of Closed-Loop Systerns in the r

JURY STABILIN TABLE FOR THE FOURTH-ORDER SYSTEM

Wow zO zl 2 z3 2

Note that the last row in the table consists of t ee elements. @om:
systerns, 2n - 4 = 1 and the Jury table concists
otice that the elernents in any even-ra.umbered row are si
irnmediately prece ing odd-numbered row.

Test. A system with the c


P ( z ) = O given by Equation (4-51, ~ewrittenas

where a. > O, is stable if the following conditions are al1 satisfied:

It is noted that the value of cl (or, in the case of the nth-order system, the value
of q,) is not used in the stability test, and therefore the computation of c, (or q l ) may
be omitted.
Construct the Jury stability table for the EolIowing characteristic equation:
Examine the stability of the following characteristic equation:
where a0 > O. Write the stability conditions. P ( z ) = z4 - 1 . 2 +~ 0~ . 0 7 ~+ ~0.32 - 0.08 = O
Referring to the general case of the Jury stability table given by Table 4-1, a SurY Notice that for this characteristic equation
stability table for the fourth-order system may be constructed as shown in Table 4-2.
This table is slightly modified from the standard form and is convenient for the compu- ao = 1
tations of the b's and c's. The determinant given in middle of each row gives the vahe
of b or c written on the right-hand side of the same row.
The stability conditions are as follows:

Clearly, the first condition, la4/< a,,, is satisfied. Let us examine the second condition
for st ability:
ec. 4-3 Stability Analysis of Closed-Loop ystems in t h e r Plane

The second condition is satisfied. The third condition for stability becomes
P(-1) = 1 + 1.2 + 0.07 - 0.3 - 0.08 = 1.89 > O, n = 4 = even Examine the stability of the characteristic equation given by
ence the third condition is satisfied. P(z) = z3 - 1 . 1 -~0~. 1 ~+ 0.2 = o
We now cor~structthe Jury stability table. Referring to Example 4-3, we c
First we identify the coefficients:
the values of b3, b2, bi, and bo and c2 and co.The result is shown in Table 4-3. (Al
the value of ci is shown in the table, cl is not needed in the stability test and th
need not be computed.) Frorn this table, we get
1631 =: 0.994 > 0.204 = lbo/
lc2l = 0.946 > 0.315 =
Thus both parts of the fourth condition given in Example 4-3 are satisfied. Sin
The conditions for stability in the Jury test for the third-order system are as follows:
conditions for stability are satisfied, the given characteristic equation is stable,
roots lie inside the unit circle in the z plane.
As a rnatter of fact, the given characteristic equation P(z) can be factored
follows:
P(z) = (z - O.8)(z + 0.5)(z - 0.5)(z - 0.4)
As a rnatter of course, the result obtained above agrees with the fact that al1 roo
within the unit circle in the z plane. The first condition, la31 < ao, is clearly satisfied. Now we examine the second condition
of the Jury stability test:
JURY SIABILITY TABLE FOR THE SYSTEM OF EXAMPLE 4-4

Row zO z1 z2 z3 z4 This indicates that at least une root is at z = 1. Therefore, the system is at best critically
stable. The remaining tests determine whether the system is criticaliy stable or unstable.
(Tf the given characteristic equation represents a control system, critica1 stability will
not be desired. The stability test may be stopped at this point.)
The third condition of the Jury test gives

The third condition is satisfied. Wow we examine the fourth condition of the Jury test.
Simple computations give b2 = -0.96 and bo = -0.12. Hence,

The fourth condition of the Jury test is satisfied.


From the above analysis we conciude that the given characteristic equation has
one root on the unit circle (z = 1) and its other two roots within the unit circle in the
z plane. Hence, the system is critically stable.
Exa
A control system has the following characteristic equation:
P(z) = z3 - 1.32' - 0 . 0 8 ~+ 0.24 = O
Determine the stability of the system.
We first identify the coefficients:
esign of Discrete-Time Control ystems by Conwentional tabilily Analysis of Closed-Loop ystems in the z Plane

Clearly, the first condition for stability, /as/< ao, is satisfied. Next, we examine $he f gain K is set equal to 2.3925, then the system becomes critically stable (meaning
second condition for stability: that sustained oscillations exist at the output). The frequency of the sustained oscilla-
tions can be determined if 2.3925 is substituted for K in the characteristic equation and
P(1) = 1 - 1.3 - 0.08 + 0.24 = -0.14 <O the resulting equation is investigated. With K = 2.3925, the characteristic equation
The test indicates that the second condition for stability is violated. The system is becomes
may stop the test here.

The characteristic roots are at 2 = 0.2439 rt j0.9698. Noting that the sampling period
Consider the discrete-time unity-feedback control system (with sampling period T = 1 T is equal to 1 sec, from Equation (4-2) we have
sec) whose open-loop pulse transfer function is given by 2 7 ~/¿ = tan-'
ud =
@S
- =- --- -
27~ 271. 0.2439 - 1.3244 radisec
The frequency of the sustained oscillations is 1.3244 radisec.
Determine the range of gain K for stability by use of the Jury stability test.
The closed-loop pulse transfer function becomes

--w K(0.36792 + 0.2642) time control systems is to use the bi ion couplied with the Routh
R(z) - z2 + (0.3679K - 1.367912 + 0.3679 + 0.2642K stability criterion. The method requires transformation from the z
Thus, the characteristic equation for the system is complex plane, the w plane. who are familiar with the
stability criterion will find the
amount of computation require
Since this is a second-order systern, the Jury stability conditions may be written as criterion.
follows: The bilinear transformation defined by
1. 14 < ao
2. P(1) > O
3. P(-1) > O, n = 2 = even which, when solved for w , gives
e shall now apply the first condition for stability. Since a, = 0.3679 + 0.2642K
and a. = 1, the first condition for stability becomes
10.3679 4- 0.2642KI < 1
maps the inside of the unit circle in the z plane into t e left ha& of the w plane. ~ h i s
can be seen as foliows. Let the real pan of w be called o and the imaginary part w ,
so that
The second candition for stability becomes
P(1) = 1 + (0.3679K - 1.3679) + 0.3679 + 0.2642K = 0.6321K > 0 Since the inside of the unit circle in the z pIane is
which gives
K>O
The third condition for stability gives

which yields
we get

For stability, gain constant K must satisfy inequalities (4-61, (4-71, and (4-8). Hence,
2.3925 > K > O which yields
'The range of gain constant K Eor ctability is between O and 2.3925.
iscrete-Time Control ystems by Conventional Met ec. 4-4 Transient and

, tfie inside of the unit circle in t


Absolute stability is a basic r nt of al1 control systems. In
reliative stability and steady- acy are also required of any
whether continuous time or

P ( z ) = aozn+ al zn-' + . + a,-l z + a, = O

as follows:

+ + a,-l-----
Then, clearing the fractions by multiplying both sides of this last equation by
(w - l.)", we obtain
systems with energy storage cannot respond instantaneously and will always exhibit
Q ( w ) = bown + bl wn-l + - + b,-, w + b, = O transient response whenever they are subjected to inputs or disturbances.
Frequently, the performance characteristics of control system are specified
in terms of the transient response to a unit-step input, since the unit-step input is
easy to generate and is sufficiently e useful inforrnation on both the
transient response and the steady-state response characteristics of the systern.
e transient response of a system to a unié-step input depends on the initial
conditions. For conveniente in comparing transient responses of various systems, it
is a cornmon practice to use the standard initial condition: the system is at rest
systems design, because unstable or critically stable control systems are not de initially and the output and al1 its time-derivatives are zero. The response character-
As mentioned earlier, the arnsunt of computation required in this approach is istics can then be easily compared.
more than that required in the Jury stabifity test. Therefore, we shall not go a The transient response of a practica1 control system, w ere the output signaj
further on this subject here. We refer the reader to Problem A-4-3, whe is continuous time, often exhibits damped oscillations before reaching the steady
present method is used for stability analysis. state. (This is true for the majority of discrete-time or digital control systems because

If we are interested in the effect of a system parameter on the stability of


loop control system, a root-loc S diagram may prove to be useful
AB may be used to compute an plot a root-locus diagram.
It is noted that in testing the stability of a characteristic equation it
S,to find the roots of the characteristic equation directl

nt out that stability has nothing to do with the syslem9


ability to follow a particulax: input. The error signal in a closed-loop cmtr
system rnay increase without bound, even if the system is stable. (Wefer t
Section 4-4 for a diccussion of error constanes.)
Transient and Steady-State

Unit-step response cufve showing transient response specifications t,:


t r , t p , M,,, ands.

-15 (a) Unit-step response of ise time t,. The rise time is the time required for the respons
o kT the system shown in Figure 4-14; to 9O%, or 5% to 95%, or O% lo 100% of its
(b) discrete-time output in the unit-step situation. For underdamped second-order syste
(b) response. nly used. For overdamped systems a
O% to 90% rise time is cornmoniy
output c ( t ) of such a system to a unit-step input Peak time t,. The peak time is the time re for the response to reach the
shown in Figure 4-15(a). Figure 4-15(b) shows first peak of the overshoot.
Just as in the case of continuous-time control systems, Maximum overshoot . The maximum overshoot is the maximum peak value
of a digital contr01 system may be characterized not of the response curve measured from unity. Hf t e final steady-stat, value of
damped natural frequency, but also by the rise time the response differs from unity, then it is common to use the maximum percent
time, and so forth, in response to a step input. In fact, in specifying such transient overshoot. It is defined by the relation
response characteristics, it is common to specify the following quantities:
aximum percent overshoot =
TRANSIENT RESPONSE SPECIFICATIONS

The amount of the maximum (percent) overs


l. Delay time t,
stability of the system.
. Wise time tr . Settling time t,. The settling time is the time required for the response curve
Peak time t,
to reach and stay within a range about the final value of a size specified as an
. Maximum overshoot M, absolute percentage of the final value, usually 2%. The settling time is related
Settling time ts to the largest time constant of the control system.

The aforementioned transient response specifications in the unit-step response The time-domain specifications just given are quite important since most
are defined in what follows and are shown graphicaily in Figure 4-16. control systems are time-domain systems; that is, they must exhibit acceptable time
responses. (This means that the control system being designed must be modified
1. Delay time t,. The delay time is the time required for the response to reacb half until the traíident response is satisfactory.)
tbe firial value the very first time. Not al1 the specifications we have just defined necessarily apply to any given
iserete-Time Control y Conventional Met ec. 4-4 Trancient and Steady- tate Response Analysis

Consider the discrete-time control system defined by


-17 Unit-step response of the system defined by Equation (4-10).

friction, backlash, or deterioration or aging of components. Hn this section, however,


Obtain the unit-step response o£ this system. we shall not discuss steady-state error due to such causes.)
program for obtaining the unit-step sesponse is shown in MATLA
ny physical control system inherently S rs steady-state error Bn response
Program 4-1. The resulting plot oE c ( k ) versus k is shown in Figure 4-17.
in types of inputs. That is, a system ma e no steady-state error with step
inputs, but the same systern may exhibit nonzero stea y-state error in response
hether or not a given system will exhibit steady-state error in respo
to a given type ofinput depends on the type of open-loop transfer function of
system.
num = [O 0.4673 -0.339311; Gonsider the continuous-time control system whose o
den = 11 - 9 .XG!i' 0.66071; tion G(s)H(s) is given by
r = ones(l,$I);
k = 8: 40;
c = filter(num,den,r 1;
pior(k,c,W The term sN in the denominator represents a pole of multiplicity N at the origin. It
v = 10 40 O 1.6); is customary to classify the system according to the number of integrators in the
axic(vf;
grid open-loop transfer function.
title('UnitStep Response') A system is said to be of type O, type 1, type 2 , . . . , if W = O, N = 1,
xlabel ('k') N = 2 , . . . , respectively. Type O systems will exhibit finite steady-state errors in
response to step inputs and infinite errors in response to ramp and higher-order
inputs. Type 1 systems will exhibit no steady-state error in response to step inputs,
Steady-State Emm A nalyrd$is, An i ortant feature associated with transient finite steady-state errors in response to ramp inputs, and infinite steady-state errors
response is steady-state error. The steady-state performance of a stable control in response to acceleration and higher-order inputs. As the type number is increased,
systern is generally judged by tke steady-state error accuracy is improved. However, increasing the type number aggravates the stability
ation inputs. In what foltows we shall investigate a t problem. A comprornise between steady-state accuracy and relative stability (tran-
caused by the inability of a system to follow particular types of inputs. (It shou!d be skat response characteristics) is always necessary.
m e d thit, besides this type o£ steady-state error, there are errors that can be The concepts of static error constants can be extended to the discrete-time
mribnted tcs ather causes, such as imperfections iil cystem climpcsnemts, cí&lc control system, as discussed ln what follows.
Sec. 4-4 Transient and

y substituting Equation (4-12) into Equatim ('-PI), we obtain

(4-13)

As in the case of the continuous-time control system, we consider three types oE


inputs: un&-step, unit-ramp, an ~nit-accelerationinputs.
Static Posktion Error Constant. For a un&-ste input r ( t ) = 1(t), we have
1
(2) = ----
a - Z-l
y substituting this last equation into Equation (4-13) the steady-state actuating
steady-state values. FTorn the diagram we have the actuating error error in response to a unit-step input can be obtained as álollows:
e(t) = r(t) - b(t)
e shall consider the steady-state actuating error at the sampling instants. Note that
from the final valiue theorern we have osition error constant
lim e ( k T ) = lim [(1 - z - ' ) E ( z ) ]
--k 2-1

E;<sr the system shown in Figure 4-18, define Then the steady-state actuating error in response to a unit-step in
from the equation

and The steady-state actuating error in response to a unit-step input becomes zero if
& = which requires that G H ( z ) have at least one pole at z = 1.
Static Velocity Error Constant. For a unit-ramp input r(t) = t I ( t ) , we have
Then we have
Tz-'
( z ) = ( ~ -- 1 2
2 1
y substituting this last equation into Equation (4-131, we have
and

Now we define the static velocity error constant Kvas follows:

K, = lim (4-16)
z-+1 T
Then the steady-state actuating error in response to a unit-ramp input can be
given by

if K , = S , then the steady-state actuating error in response to a unit-ramp input is


Figure 4-18 Discrete-time control system. zem. This requires G ( z ) to possess a dorsble psle at z = 1.
Static Acceleration Error Constant. For a unit acceleration i n ~ u td r ) = Table 4-4 lists system ty nding steady-state errors in
response to step, ramp, and accel diserete-time control system
of the configuration shown in Figure 4-18.
The steady-state error analysis just presented applies to t
iscrete-time control system shown in Figure 4-18. For a differ
By substituting this last equation into Equation (4-13), we obtain configuration, it is noted that if the closed-loop discrete-time control system has a
,. r,, -,, 1 r 2 ( 1 + z-l)z-li .. T2 closed-loop pulse transfer function, t en the static error constants can be determined
by an analysis similar to the e just presented. Table 4-5 lists the static error
configurations of discrete-time control sy
e dehne the static acceleration error constant Knas follows: ntrol system does not have a closed-lo
static error constants cannot be define
Ka -- lim d from the system
241 T2
at the terms "position error," "velocity er
Shen the steadv-state actuatinp. error dy-state deviations in the output po
transients have died out the input a
at the same velocity, but have a finite position difference.

The steady-state actuating error in response to a unlt-acceleration input beco


zero if Ka = 33. This requires GN(z) to possess a triple -5 STATIC ERROR CONSTANTS FOR TYPICAL CLOSED-LOOP
CONFIGURATIONS OF DISCRETE-TIME CONTROL SYSTEMS
Equations (4-15), (4-17), and (4-19) give the ex
actuating errors of the discrete-time control system shown in Figure 4-18 at Closed-loop configuration VaIues of K p , K,, and K,
samplirig instants for a unit-step, unit-ramp, and unit-acceleration in
tively.
. It is irnportant to emphasize that the actuating error is the differ-
ence between the reference input and the feedback signal, noi t
tween the referente input and the out t. From the foregoing analysis we see that
a type O systern will exhibit a constant ady-skate actuating
step input and an infinite actuating error in response to r
&, = $5G ( z ) H ( z )
higher-order inputs. A type 1system will exhibit a zero stead
in response to a step input, a constant steady-state error in response to a ramp input, (1 - z-"G(z)H(z)
and an steady-state error response E_(, = iim
2-1 T
order inputs.
K, = ,lim ( 1 - ~ - l T) ~2 G ( z ) H ( z )
SYSTEWI TYPES AND THE CORRESPONDING 241

STEADY-STATE ERRORS IN RESPONSE TO STEP, RAMP, AND


ACCELERATION INPUTS FOR THE DISCRETE-TIME CONTROL P;P = G~(z)HGz(z)
SYSTEM SHOWN IN FIGURE 4-18
(1 -z I ) Gl(z)HGz(~)
1 Steady-state errors in response to K, = iim
2-1 T

System Step input Ramp input


Acceleration
input K~ = lim
Z-+l
'1 -Z - ' ) ~ G ~ ( Z ) H G Z ( Z )
T2
r(t) = 1 r(t) = t r ( t ) = St2
r\, = $m G l ( z ) G & ) H ( z )
Type O system 1 co co
1 + KP
K, = lim ( 1 - z-l)G1(z)G2(z)H(z)
Type 1 system O -1 m 241 T
K,'
1 ( 1 -Z-')~G~(Z)GZ(Z)EI(Z)
Type 2 system o O - I(, = lim
Z-+l TZ
Design of Discrete-Time Control Syslerns by Conventional Methods Chap. 4 Sec. 4-4 Transient and Steady-Sbte Responce Analysis

n examining transient response characteristics Since the system error is


E(z) = W(z) - C(z) = -C(z)
we find the error E(z) due to the
1
E(z) = - -
GD(4
osed-loop pulse transfer Thus, the larger the gain of GD(z)is, the smaller the error E(z).
function an integrator [which means that GD(z) has a pole at z = 11, then the steady-state
error due to a constant disturbance Bs zero. Thks may be seen as follows. Since for
a constant dicturbance of magnitude N we have

If 1 GD(z)C(z)I 1, then we find


if GD(z) invoIves a pole at z = 1, then it rnay be written as

ere e D ( z ) does not involve any zeros at z = 1. T


be given by

Hf a linear system is subjected to both the reference input and a disturbance


input, then the resulting error is the sum of the errors due to the reference input and
e disturbance input. The total error must be kept within acceptable limits.
Note that the point where the disturbance enters the system is very i nt
in adjusting the gain of GD(z)G(z). For exarnple, concider the system in
Figure 4-20(a). The closed-loop pulse transfer function for the disturbance is

To minimize the effects of disturbance N(z) on the system error E(z), the gain of
GD(z)G(z) must be made as large as possible. owever, for the system shown in
Figure 4-20(b), the closed-loop pulse transfer function for the disturbance is

to minimize the effects of disturbance N(z) on the system error E(z), the gain
of GD(z)G(z) must be made as small as possible.
Therefore, it is advantageous to obtain the expression E(z)/jV(z) before
concluding whether the g n of GD(z)G(z)should be large or to minimize the
error due to disturbances t is important to remember, howev hat the rnagnitude
of the gain cannot be determined solely from the disturbance considerations. It must
be determined by considering the responses to both reference and disturbance
Figure 4-19 (a) Digital closed-loop control system subjected to reference input and
disturbance input; (b) modified block diagram where the disturbance input is con- e frequency regions for the reference input and disturbance input are
sidered the input to the system. Wfficiently apart, a suitable fiiter rnay be inserted in the system. If the frequency
esign sf Discrete-Time ysterns by Conventional Met ec. 4-5 Design

To combine these two forms into one, let us define t


1 + F(z) = O Obtain the characteristic equation
where
d then rearrange this equation so arameter of interest, such as gain
pears as the multipiying factor in
at F ( z ) is the open-100 pulse transfer function. The characteristic equation
given by Equation (4-20) can then be written as

r of interest is gain K , wh
Since F(z) is a complex quantity, this last equation can be split into two equations
by equating first the angles and then the agnitudes of the two sides &oobtain
the closed-loop zeros consist of the
ASuGLE CONDITION: zeros of G(z) and the poles of H(z).]
ind the starting points and terminating points o
er of separate branches of the root loci. The
corresponding to K = O are open-loop poles and those cor
MAGNITUDE CONDITION : ence, as K is increased from zero to infinity, a root locaas starts
from an open-loop pole and terminates at a finite open-loop zero
IF(z)l = 1
zero at infinity. This rneans that a root-locus plot will have just as
The values o£z that fu as there are roots of the characteristic equation. [If the zeros at infi
of the characteristic e in the count, F(z) has the same number of zeros as poies.]
A plot of the point e angle condition calorre f the number n of closed-1 e same as the number of o
is the root locus. The r e characteristic equation (the closed-loop poles, then the nrrmber of ind 1 root Iocus branches termirrating
corresponding to a given value of the gain can be located on the root loci by open-loop zeros is equal to the number m of the loop zeros. The remaining
agnitude condition. %hedetails of applying the angle and magnitude conditions n - m branches terminate at infinity (at n - m cit zeros at infinity) along
to obtain the closed-loop poles are presented in the following. asymptotes.
. Determine the root loci on the real axis. Root ljoci on the real axis are
determined by open-loop poles and zeros lying on it. The conjugate compliex poles
cated, but actually it is not difficult if the and zeros of the open-loop pulse transfer function have no effect on the Iocation of
are applied. the root loci on the real axis because the angle eontribution of a pair of conjugate
y locating particuhr points and asymptotes and by computing angles 0 complex poles or zeros is 360" on the real axis. Each portion of the root Iocus on
departure from complex poles and angles of arrival at complex zeros, it is possib the real axis extends over a range from a pole or zero to another pole or zero.
to construct root Loci without difficulty. Note that while root loci rnay be conveniewt Bn constructing the root loci on the real axis, choose a test point on it. If the
tal number of real poles and real zeros to the ri of this test point is odd, then
drawn with a digital computer, if manual construction of the root locus plot
attempted, we essentially proceed on a trial-and-error basis. ut the number of tri is point lies on a root locus. The root locus an s complement form alternate
required can be greatly reduced if the establlshed rules are used. gments along the real axis.
Determine the asymptotes of the root lo the test point z is located far
Because the open-loop conjugate complex poles and conjugate complex zems
if any, are dways located symmetrically ut the real axis, the root loci are agway from the origin, then the anglies of al1 the complex quantities may be considered the
symmetric with respect to the real axis. same. One open-loop zero and one open-loop pole then each cancel the effects of
rice, we need only construct the uppe
half of the root loci and draw the mirro age of the upper half in the lower ha1 the other.
of the z plane. Remember that the angles of the complex quantities originating from Therefore, the root loci for very large values of z must be asymptotic to straight
t k open-loop poles and open-loop s and drawn to a test point z are meacured h e s whose angles are given as follows:
in the counterclockwise direction. shall now present the general rulec, and r 180°(2Pd + 1)
pii-oceduresfor constrclcting root loci.
L.,
Angle of asymptote = , N = 0 , 1 , 2,...
n-m
where (z)/A(z) = F ( z ) , then
n = number of finite poles of F(z) K = --A ( 4 (4-22)
m = number of b i t e zeros of F(z) Nz)
and the breakaway and break-in points (which correspond to multiple roots) can be
determined from the roots of

obtained as fo'ollows. Since

-
~ + p2
z ~ +-[(pl + + pn) - (z1 + Z2 + . + z , ) ] z ~ - ~ -+~ .

ne the angle of departure (or angle of arrival) of the root loci from
n-m

--ua = - (Pi + P2 + + pn) - (zi + zi + + zm)


n-m

Angle of departure =
480" - ( 8 , 3- 8, - 4)

If the characteristic equation


1 + F(z) = O
is written as

Figure 4-22 Diagram showing angle of departure.


ased on the Woot-Loctds Method

tbus found give the location at which the root loci cross the imaginary axis an Note, however, that if pole-zers canmcellatáon occurs in the feed-forward pu
e value of the corresponding gain transfer function, then the same e-zero cancellation occurs in the cl
. Any point on the root loci is a possible clos pulse transfer function. Consi hown in Figure 4-21,
-loop pole when the value of gain K sa assume
e magnitude condition enables us to G(z) = G ~ ( z ) G ~ ( z ) 9 (z) = 1
root location on the locus. The ma
pose pole-zero cancellation occurs in GD(z)Gl(z). Por exa

Then the closed-loop pulse transfer function beco

f gain K of the open-loop pulse transfer function is given in the problem, t


by applying the magnitude condition, Equation (4-24), it is possible to locate
closed-loop poles for a given K on each branch of the root loci by a trial-and-e
method.
It is important to note th ecause of the pole-zero cancellation, the third-order system becomes one of second
z) and the numerator of H(z
involve common factors then the corresponding open-loop poles and zeros t is important to summarize that the effect of pole-ze
cancel each other ,reducing the degree of the characteristic equation by one or ( 2 ) is different from that of pok-zem cancellation

The root locus plot of G(z)N(z) wi1Í not show al1 the roots of tfie charact transfer function (such as pole-zera cancellation in the
equation, but only the roots of the reduced equation. ant). In the forrner, the canceled of the closed-loop system
To obtain the complete set of closed-loop poles, we must add tfie cance ereas in the latter the canceled po
or polies of G(z)H(z) to those closed-loop poies obtaine system (in the Iatter the order of the S
of G'(z)Ei(z). The important thing to reme ber i~that a cancel& pele of gg(z poles).
is a closed-loop pole of the system. In what Eollows we shall
As an example, consider the case where G(z) and H(z) of the
\ 2 \ ,
svstern show
- - - -J - - - --- --- T on the relative stability of
in Figure 4-21 are given by the closed-loop control system. C own in Figure 4-23. Assurne
that the digital controller is of the integral type, or that

and
Let us draw root lscus diagrams for tke system for three values of the sam
T : 0.5 sec, 1 sec, and 2 sec. Let us also determine the critica1 value of K for eac
case. And finally let us locate the closed-loop poles corresponding to K = 2 for eac
Then, clearly, the pole z = -a of G(z) and the zero z = -a of of the three cases.
other, resulting in
z + c z+a z + c
G(Z)H(z) = (z + a)(z + b ) -z=+ d
(z + b)(z + d )
owever, the closed-loop pulse transfer function of the svstem is Integral
controller

c Figure 4-23 Digital control system.


e shall first obtain the z transforrn of Gh(s)GP(s): Equation (4-27) yiel
nce both K values are positive, z = 0.7788

Figure 4-24(a) shows the root locus diagrarn


value of gain K for this case is obtained by use of t
can be obtained from Equation (4-26) as follows:

lrn
z plane
--
( = 0.5 sec)

- 1 - e-T
--
z - e-=
The feedforward pulse transfer function becomes
Kz 1 - e-*
G(z) = G,(z)2?[Gh(s)G,(s)I = --- -
z- l ~ - e - ~
The characteristic equation is
1 + G(z) = O
or

1, Surnpling period T = 0.5 sec: For this case, Equation (4-25) become
0.3935Kz
G(z) =
(z - 1)(z - 0.6065)
Notice that G ( z ) has poles at z = 1 and z = 0.6065 and a zero at z = 0.
T(3 draw a root locus diagram, we first locate the poles
and then find the breakaway point and break-in point. Not
pulse transfer function with two poles and one zero results in a circ
centered at the zero. The breakaway point and break-in point are Im
writing the characteristic equation in the form of Equation (4-22), z plane

, = -
(z - 1)(z - 0.6065)
0.3935~
í T = 2 sec)

and differentiating K with respect to z and equating the result to zero:


dK =
- - z2 - 0.6065 = o
dz 0.3935z2
Hence,
z2 = 0.6065
Or Figure 4-24 (a) Root locus diagrarn for the system shown in Figure 4-23 when
T = 0.5 sec; (b) root locus diagram when T = 1 sec; (c) root locus diagram when
z=0.7788 and z=-0.7788 T = 2 sec.
Decign of Discrete-Time Control Syctems by Conventional Methods @ha

For the present case, T = 0.5 and this Iast equation becomes

Since the critica1 gain Mc corresponds to point z = - 1 , we substitute -1 for z in


Equation (4-28) :

or
K = 8.165
The critical gain M, is thus 8.3 65.
The dosed-loop poles correspon
zl = 0.4098 + j0.6623 and z2 = 0.4098 - j0.6623
These closed-loop poles are indicated by dots in the root locus diagram.
Sampling period 1' = I sec: r this case, Equatiorl (4-2 5 ) beca
follows: closed-loop pole is i , then in the s plane the closed-loop pule location (in the upper
half-plane) can be given by
s = -bn+ j u n V i - C2
Kence, G ( z ) has poles at z = 1 and z = 0.3679 and a zero at z = 0 .
e corresponding point in the z plane is
The breakaway point and break-in point are found to be z = 0.6065 and
z = -0.6065, respectively. The corresponding gain values are K = 0.2449 and z = exp [ T ( - i w n +j w n m ) ]
K = 4.083, respectively.
Figure 4-24(b) shóws the root locus diagram when T = 1 sec. T
z plane
of gain K is 4.328. The closed-loop poles corresponding to K = 2 are found to be
z, = 0.05185 + j0.6043 and z2 = 0.05185 - j0.6043
and are shown in the root locus diagram by dots.
3. Sampling period T = 2 sec: For this case, Equation (4-25) becomes

e see that G ( z ) has poles at z = 1 and z = 0.1353 and a zero at z = 0.


The breakaway point and break-in point are found to be z = 0.3678 and
z : -0.3678, with corresponding gain values K = 0.4622 and K = 2.164, respece
tively. The critical value of gain K for this case is 2.626.
Figure 4-24(c) shows the root locus diagram when T = 2 sec. The closed-loop
poles corresponding to K = 2 are found to be
z, = -0.2971 + j0.2169 and z2 = -0.2971 - j0.2169
These closed-loop poles are shown by dots in rhe root locus diagram. Figure 4-25 Closed-loop pole locations in the z plane shown with constant 5 loci.
aced on the Root-Lscus

from which we get closed-loop pulse transfer function for t


Izl e-Th rd pulse transfer function C(z) is given

and
& = TU-. = To, = 8 (rad)
For T = 0.5 sec and response can be given by
From Equations (4-29) and (4-30), the value of J can be calculate
in the case where the sampling period T is 0.5 sec, we have the elos
K = 2 at z = 0.4898 + j0.6623
lz/ = 'd0.4098' + 0.6623' = 0.7788
y solving
from whicb we obtain t response sequence c(kT) versus k T s
IzI = = 0.7788 ' Figure 4-26(a).
for the exponent, we find From Figure 4-25 we see that the angle of the Bine connecting the origin and
pole at z = 0.4098 + j0
Tb, = 0.25 ately 58.25". The angle 8
Also, termines the number of samples per cycIe a£ s atian. Note that

&= = 58.25" = 1.0167 rad


0.4098
Hence, for 8 = 58.25", we have 360°/8 = 36Q0/58.25"= 6.18 samples per cycle of
ence,
ed oscilliation, as seen fro
fi = T w , W = 1.0167 rad Similarly, for T = 1 sec and K = 2, the unit-step response is given by
From Equations (4-31) and (4-32), we obtain
TSmn 0.25
e unit-step response sequence c(kT) versus in Figure 4-26(b). Since
~ ~ ~ ' d i=-1.0167
52 e angle of the line connecting the origin and oop pole for the present
or case is 85.10°, as shown in Figure 4-25, we ha ately 360°/85.10" = 4.23
samples per cycle, which is ve
recommend eight or more sa
Finally, for T = 2 sec and R = 2, the unit-step response is given by
which ylelds
J = 0.2388
(From Figure 4-25 we graphically obtained 0.24 for S, which is The unit-step response sequence c(kT) versus k T is shown in Figure 4-26(c). Frorn
actual S value of 0.2388.) Figure 4-25, the angle of the line connecting the origin and the closed-loop pole for
It is irnportant to point out that in the second-order system t the present case is 143.87", and consequently we have 360"/143.87" = 2.50 samples
l' is indicative of the relative stability (for erarnple, in respect to the m per cycle, as seen from Figure 4-26(c). (Note that a slow sampling frequency such
overshoot in the unit-step response) only if the sa as 2.50 samples per cycle is unacceptable.)
high (so that there are eight to ten samplin Figure 4-26 has shown three different plots of the unit-step response sequence
frequency is not high enough, the maximu c(kT) versus kT. As can be seen from these plots, if the sampling period is srnall,
be much higher than would be predicted a plot of c(kT) versus kT will glve a fairly accurate portaait o£ the response c(t).
To compare the effects of different sampling However, if the sa eriod is not sufficiently small, then the plot of c(kT)
sponse, we shall compare the unit-step response se versus k T will not an accurate resullt. It is very important to choose m
T considered in the preceding analysis. adequate sampling d based on the satisfaction of the sampling theorem, systern
n of Discrete-Time Control ystems by Conventional Methodc Chap. ec. 4-5 Design

e static velocity error constant Kv is given by

=
Y
lim -----
Z+I
0.78702
0.52 (z - l)(z - 0.6065) 1
Thus, the steady-state error in res

es, = -
KJ = -4
'= 0.25

irnilarky, for the case where T = 1sec and = 2, the open-loop pulse transfes
function is

and the static velocity error constant Kv is given by


O 1 2 3 4 5 6 7 8 k T ( s e c )
(1 - z-')G(z)
Kv = lim m

and the steady-state error in response to a unit-ram


1
ess = K, = -2 = 0.5
Finaly, for the case where T = 2. sec and K = 2 , t
1 2 3 4 5 6 7 8 kT(sec) function is
Figure 4-26 (a) Unit-step response sequence of the system shown in Figure 4-23
when T = 0.5 sec and K = 2; (b) unit-step response sequence when T = 1 sec and
K = 2; (e) unit-step response sequence when T = 2 sec and K = 2.
anta the static velocity error constant
unit-ramp input are obtained , respectively , as
dynarnics, and actual hardware considerations. Note that barely satisfying the S
pling theorern is not sufficient. An acceptable rule of thumb is to have eight to
sarnples per cycle (six samples per cycle is marginal) if the system is underdamp
and exhibits oscillation in the response.
Next let us investigate the effect e sarnpling period T on the steady-
sball consider the unit-I-
case where the samplirr h r t s (a), (b); a*$ (c) of Figure 4-27 show, respectively, the plots of tke unit-ramp
open-loop pulse trancfer furrction is response sequence c(kT) versus kT for the three cases considered.
Design of Discrete-Time Control Systems by Conventional ec. 4-5 Design Based on the Root-Locus Method

ed sinusoidal oscillation). ncy is low (that is, less


les per cycIe of damped ), then predicting the
relative stability from the damping ratio value is erroneous.

Consider the digital control system shown in Figure 4-28. In the z plane, design a digital
controller such that the dominant closed-loop poles have a damping ratio 5 of 0.5 and
a settling time of 2 sec. The sampling period is assumed to be 0.2 sec, or T = 0.2. Obtain
the response of the designed digital control system to a unit-step input. Also, obtain
the static velocity error constant 6(, of the system.
For the standard second-order system having a pair of dominant closed-loop
poles, the settling time of 2 sec means that
4 4
settling time = - = -= 2
lw, 0 . 5 ~ ~
which gives the undamped natural frequency w, of the dorninant closed-loop poles as
wn = 4
The damped natural frequency wd is determined to be
wd = con- = -4 = 3.464

Since the sampling period T is 0.2 sec, we have

[Notice that there are approximately nine samples per cycle of damped oscillation
(31.4213.464 = 9.07). Thus, the sampling period of 0.2 sec is satisfactory.]
shall first locate the desired dominant closed-loop poles in the z plane.
Weferring to Equations (4-29) and (4-30), for a constant-damping-ratio locus we have

and
12 = Twd = 27T- @d

@S

From the given specifications (5 = 0.5 and wd = 3.464), the magnitude and angle of the
dominant closed-loop pole in the upper half of the z plane are determined as follows:

Figure 4-27 (a) Unit-ramp response sequence of the system shown in Figure 4-23
when T = 0.5 sec and K = 2; (b) unit-ramp response sequence when T = 1 sec and
K = 2; (c) unit-rarnp response sequence when T = 2 sec and K = 2.

The three cases we have considered demonstrate that increasing the samp
period T adversely affects the system's relative stability. (It may even cause insta
ity in some cases.) It is important to remember that the damping ratio i of t
ched-loop poles of the digital control system is indicative of the relative stddit
only if the sampling frequency is sufficiently high (that is, eight or more s a r n p k P Figure 4-28 Digital control systern for Example 4-9.
ased on the Woot-Locus Methad

and
3 464
& = 27.r-31.42 = 0.6927 rad = 39.69"
is the gain constant of the controller.
lf we decide to cancel the pole at z = 0.6703 by the zero of the controller at
can now locate the desired dominant closed-loop pole in the upper half of the z
z = - a , then the pole of the controller can be detesmined (from the condition that the
e, shown in Figure 4-29 as point P. Note that at point P
controller must provide +51.26") as a point at z = 0.2543 (@ = -0.2543). Thus, the
z = 0.6703/39.69" = 0.5158 + j0.4281 pulse transfer function for the controller may be determined as
Noting that the sampling period T is 0.2 sec, the pulse transfer function G(z)
the plant preceded by the zero-order hold can be obtained as follows:
The open-loop pulse transfer function now becomes

eyuarion can oe written as


LIUS I ~ S L

Next, we locate the poles (z = 1and z = 0.6703) and zero (z = -0.8760) The gain constant K can be determined from the following rnagnitude condition:
on the z plane, as shown in Figure 4-29. If point P is to be the location for the
dominant closed-loop pole in !he upper half of the z plane, then the sum of the angl IGD(z)G(z)/,= 0.5158 + j0.4281 = 1
at point P must be equal to t180". However, the surn of the angle contributions at go Hence ,
P is
O.O1758(z + 0.8760)
= 1
17.10" - 138.52" - 109.84" = -231.26" (z - 0.2543)(z - 1) z=0.5158+~0.4281

Hence, the angle deficiency is which gives


-231.26" + 180" = -51.26" K = 12.67
The controller pulse transfer function must provide $51.26". The pulse transfer fun The designed digital controller is
tion for the controller may be assumed to be

r plane The open-loop pulse transfer function for the present system is

Hence, the closed-loop pulse transfer function is

The response to the unit-step input R(z) = 1/(1 - z-l) can be obtained from

Figure 4-30 shows the unit-step response sequence c ( k T ) versus k T . The plot
shows that the maximum overshoot is approximately 16% (which means that the
damping ratio is approximately 0.5) and the settling time is approximately 2 sec. The
@re4-29 Root locus diagram of the system considered in Example 4-9. digital controller just designed catisfies the given specifications and is satisfactory.
ecign Gased on the Frequency-

Familiarity with the relati ane pole and zero Iocations


and the time response charact ning discrete-time control

between O and 1 i

Similarly, in the s plane a closed-loop poPe on the negative real axis neai: the
origin increases the settling time. In the z plane, su
to a closed-loop pole on the positive real axis betwee
pole in the z plane bekween O and 1 (Pn particular, n
0.4 settling time. The psesence of a closed-loop pole or
between O and -1 in the z plane, however, affect
0.2 slightly .

O 0.4 0.8 1.2 1.6 2.0 2.4 2.8 3.2 3.6 4.0
k T ísec)

-30 Unit-step response sequence of the system designed in Exampie 4-9. N

The static velocity error constant M, of the system is given by The freqeaency-response concept plays the same powerful role in digital control
systems as it does in continuous-time control systems. s stated earlier, it is
assumed in this book that the reader is familiar with conventional frequency-
se design techniques for continuouc-time control systems. In fact, familiarity

= 2.801
1 de diagrams (logarithmic plots) is necessary in the extension of the comen-
tional frequency-response techni ues to the analysis and design of
control systems.
Frequency-response methods have very frequently been used in the
If it is required to have a large value of K v , then we may include a lag cornpen-
sator. For example, adding a zero at z = 0.94 and a pole at z = 0.98 would raise t-
sator design. She basic reason is the simplicity of the rnethods. In pe
frequency-response tests on a dáscrete-time system, it i nt that the system
K.. value three times, since (1 - 0.94)1(1 - 0.98) = 3. (It is important that the pole and
zero of the lag compensator lie on a finite number of allocable discrete points.) A lag have a low-pass filter before the sampler so that sideban
compensator, which has a pole and a zero very close to each other, does not significan ti^ response o£ the linear time-invariant system to a sinuso
change the root locus near the dominant closed-loop poles. The effect of a lag compen- quency and modifies only the amplitude and phase of
sator on the transient response is to introduce a srnall but slowly decreasing transienl amplitude and phase are the only two quantities that must be dealt with.
component. Such a srnall but slow transient, however, is not desirable frorn the vkw- In the following, we shall analyze the response of the linear time-invariant
point of disturbance or noise attenuation, since the response to disturbances would not discrete-time system to a sinusoidal input; this analysis will be followe
attenuate promptly. definition of the sinusoidal pulse transfer function. Then e discuss the design of a
Finally, it is noted that although the designed system is of the third order, it acts discrete-time control systern in the w plane by use of a
as a second-order system, since one pole of the plant has been canceled by the zero of
the controller. Because of this, the present system has only two closed-loop poles. The
dominant closed-loop poles are the only closed-loop poles in this case. If a pole and a Earlier in this book we stated that the frequency response of G ( z )can be obtained
zero do not cancel each other, then the system will be of the third order. by substituting z = elwTinto G ( z ) .In what follows we shall show that this is indeed
en& lt is important to note that the poles of t h p ~ ~ ~ C P A - n1llsc
~ A ~ T P
true.
transfer function determine the natural modes of the svstem. Gonsider the stable linear time-invariant discrete-time system shown in Fig-
frequency response behaviors, however, are stro 4-31. The input to the system G ( z ) before sampling is
the c~o~ed-loop pulse irancfer function. u (í) = sin wt
Design of Discrete-Time Control Systems by Conventional Methods Chap. ased on the Frequency-Responce Method

Stable linear
time-invariant
discrete-time

The inverse z transform of this Iast equation is

-31 Stable linear time-invariant discrete-time system

The Iast term on the right-hand side of E uation (4-35) repsesents t


The sampled signal u ( k T ) is response. Since the system G ( z ) has been assurned to be stable, al1 transient
u ( k T ) = sin kwT response terms will disappear at steady state and we wPlI get the following steady-
state response x,,(kT):
The z transform of the sampled input is
z sin wT
U ( z ) = Z' [sin k o T ] = ( z - e J w T ) ( ~- e-lwT)
bjected to a sinusoidal in
The response of the system is given by
z sin wT
X ( z ) = C ( z ) U ( z )= G ( z )( z - e lwT)(z- e-~wT)
ase angle, is given by
-
-- az a~
+ ----- S = @(o)= / C ( e J w T )
2 - elwT z - e-~wTi [tesrns due to poles of G ( z ) ]
Multiplying both cides of Equation (4-34) by ( z - e p T ) / z ywe obtain In terms of G(eiwT),Equation (4-36) can be written as follows:
x,(kT) = 1G (elwT)lsin (kwT +/ G(ejwT))
[terms due to poles of G
have shown that G(ejwT) indeed gives the magnitude and phase of the frequency
The second term on the right-hand side of this Iast equation a response of C ( z ) .Thus, to obtain the frequency response of G ( z ) ,we need only to
approaches elwT.Since the system considered here is stable, the third term on th substitute elwTfor z in G ( z ) .The function G(eiwT)is coznrnonly called the sinusoidal
right-hand .side also approaches zero as z approaches e'"'. Hence, by letting pulse transfer function. Noting that
,-
approach elw',we have
we find that the sinusoidal pulse transfer function G(eiwT)
is periodic, with the perio
equal to T.
The coefficient E, the complex conjugate of a , is then obtained as follows:
C (e-jwT>
Consider the system defined by
x ( k T ) = u ( k T ) + ax((k - 1)T), O <a <1
Let us define
~f (el"T) Mele
P;
where u ( k T ) is the input and x ( k T ) the output. Obtain the steady-state output x ( k T )
when the input u ( k T ) is the sampled sinusoid, or u ( k T ) = A sin k o T .
Then The z transform of the system equation is
X ( z ) = U(z)+ az-'X(z)
Equation (4-34) can naw be written as By defining G ( z ) = X ( z ) / U ( z ) ,we have
iscrete-Time Control Sec. 4-6 Design Based on the Frequency-Responce Method

Let us substitule eJwTforz in G(z). Then the sinusoidal pulse transfer hnction G(e1"3
can be obtained as
1
G'(eJ"*) = 1 - ae-jmT-
- 1
1 - a coswT + ja sinwT
The amplitude of G(eJuT)is

IG(eJuT)l= M = 1
VI + a2 - 2a c o s w ~
and the phase angle of G(eJWT)
is
a sin wT
/G(eJ"T) = 6, = -tan-'
1 - a coswT
Then the steady-state output x,,(kT) can be written as,follows:
x,(kT) = AM sin (kwT + 8)
- A
i
sin kwT - tan-'
2\/1 + a2 - 2a c o s w ~
-32 Diagrarns showing mappings from the s plane to the z plane and from the ,z plane
to the w plane. (a) Primary strip in the left half of thes plane; (b) z plane mapping of the primary
strip in the s plane; (c) w plane mapping of the unit circle in the z plane.
ne. Before we ca
apply our well-developed frequency-response methods to the ana lane, z varies Pom 1 to - 1 along
discrete-time control sgstems, certain rnodifications in the z pl to 03 along the imaginary axis in
necessary. Since in the z plane the frequency appea Although the left half of the w plane corre o the left half of the s plagie
frequency response in the z plane, the sim and the imaginary axis of the w plane correspond
completely lost. Thus, the direct application of freque there are differences between the two plane
worthy of consideration. In fact, since the z transformat behavior in the s plane over the frequency rang maps to the range
complementary strips of the left half of the s plane into th where v is the fictltious frequency e. This means that,
conventional frequency-response methods, which d frequency response characteristi filter will be repro-
plane, do not apply to the z plane. discrete or digital filter, the fiequency scale on which the response
The difficulty, however, can be overcome by transforming the pulse tran occurs will be compressed from an infinite interval in t e analog filter to a finite
function in the z plane into that in the w plane. The transformation, cornrnonly ca interval in the digital filter.
the w transformation, a bilinear transforrnation, is defined by Once the pulse transfer function G ( z ) is transformed into G ( w ) by means of
1 + (TI2)w the w transformation, it may be treated as a conventional transfer function in w.
Z =
1- (2'12)~ Conventional frequency-response techniques can then be used in the w plane, and
SO the well-established frequency-response design tec niqUes can be applied to the
where T is the sampling period involved in the discrete-time control system und design of discrete-time control systerns.
consideration. By converting a given pulse transfer function in the z plane into noted earlier, v represents the fictitious frequency
rational function of w , e frequency-response methods can be extended to discret onal frequency-sesponse techniques may be used to
time control systerns. y solving Equation (4-37) for w 9 we obtain %heinver for the transfer function in w. (In the brief review of the
relationship
section, we shall use the fictitious frequency v as the variable.)
w = - -2- 2 - 1 Although the w plane resembles the s plane geometrically, the frequency axis
T z +1 in the w plane is distorted. The fictitious frequency v and the actual frequency o are
Through the z transforrnation and the w transforrnation, the prim related as follows:
the k i t half of the s plane is first mapped into the inside of the unit circl
plane and then mapped into the entire left half of the w plane. The two
processes are depicted in Figure 4-32. (Note that in the s plane we conside
primary stiip.) lvotice that the origin of the z plane is mapped into
w = -2lT in the w plane. 1Votice also that, as s varies from O to joJ2 along
ecign Based on the Frequency-

< Y
J

Gís)
Equation (4-39) gives the relationship between ihe actual frequency o a
fictitious frequency v. Note that as the actual fre uency o moves from -4 @S t -34 Transfer-function system of Example 4-11.
uency v moves from -co to O, and as w moves from O to w,, v S
Referring to Equation 44-39), the actual freauencv w can he traaacii~t~d
k f r th
~

correspondlng bandwidth in the w pllane is (212') tan. ( - \ , r \J

corresponds to G( jw,), where wl = (%T) tan-"(v, n2). figure 4-33 s


tionship between the fictitious frequency v times i T and the actual frequency w f
the frequency range between O and w,. S
Notice that in Equation (4-39) if wT is small then By use of the bilinear transforrnation given by Equation (4-371, or
v=o
This means that for small wT the transfer functions C(s) and G(w) resemble ea
other. Wote that this is the direct result of the inclusion of the scale factor 21T G(z) can be transformed into G(w) as follows:
Equation (4-38). The presence of this scale factor in the transformation enables
to maintain the same error constants before and after the w transformation. (T
means that the transfer function in the w plane will approach that in the s
hes zero. See Example 4-11, which follows.)

Consider the transfer-function system shown in Figure 4-34. The sampling period Notice that the Iocation of the pole of the plant is s = -10 and that of the pole
assumed to be 0.1 sec. Obtain G(w). in the w plane is w = -9.241. The gain value in the s plane is 10 and that in the w plane
The z transform nf Gfr) i<: is 9.241. (Thus, both the poie locations and the gain values are similar in the s piane and
the w plane.) However, G(w) has a zero at w = 2lT = 20, although the plant does not
have any zero. As the sampling period T becomes smaller, the w plane zero at w = 2l-F
approaches infinity in the right half of the w plane. Wote that we have
1o
lim G(w) = iim -
w-+o s-OS + 10

This fact is very useful in checking the numerical calculations in transforming G ( S ) into
G(w)-

To sumrnarize, the w transforrnation, a bilinear transformation, maps the


inside of the unit circle of the z plane into the left half of the w plane. The overall
result due to the transformations from the s plane into the z plane and from the z
plane into the w plane is that the w plane and the S plane are similar over the region
of interest in the s plane. This is because some of the distortions caused by the
transformation from the s plane into the z plane are partly cornpensated for by the
transformation from the z plane into the w plane.
Figure 4-33 Relationship between Note that if
fictitious frequency v times 4 T and
actual frequency w for the frequenc
range between O and 4 w,.
esign Based on the Frequency-Wesponse Method

where the si's and bi7sare constants, is transformed into the w plane by the trans- which corresponds to O 5 v 5 m. Then, noting that v = in t
formation to o = 1 o, in thesplane, it can be said that lim,,, /C(jv)lcorr
1 + (TI2)w ( j o + lo)/, which is a constant. (Ht is irnportant to note tha
z = enerally not equal to each otker.) From the pole-zero polnt of view, it can be sai
1 - (TI2)w
at when IG(jv)l is a non constant cnt v = m it is implied that G(w) contains
then G ( w ) takes the form
e same number of pole
In general, one or m G(w) lie in the right f the w plane. The
resence of a zero in the right half of the w ) is a nonminimum
where the aiysand the Piysare constants (some of phase transfer function. Therefore, we must be careful in drawing the phase angle
m may be zero). Thus, G(w)
is a ratio of polynomials in w , where the degrees of numerator and denorninator curve in the Bode diagram.
may or may not be equali. Since G( jv) is a rational ction of v , the Nyquist stability
criterion can be applied to G(jv). En terms of th iagram, the conventional approach to the analysis and design of control systems is particularly useful for the
straight-line approximatiori to the magnitude curve as'well as the concept of the following reasons:
phase margin and gain margin apply to C(jv).
Design by means of ode diagram the low-frequency asymptote of the magnitude curve is
put-single-sutput he static error constants K P , K v , or
lar, if the transfer function is in a facto transient response can be translated into those of the
ode diagram can be drawn and reshaped are well known. n terms of the phase margin, gain margin, bandwidth, and
As stated earlier, the conventional freq so forth. These specifications can easily be handled in the
transfer funclions in the w plane. Recalh that the particular, the phase and gain margins can be rea directly from the Bode
separate plots, the logarithmic magnitude IG diagrarn.
/ G ( jv) versus log v. Sketching of the logarith The design of a digital compensator (or digital controller) to satisfy the given
of G(jv), so that it works on the principle in terrns of the phase margin and gain margin) can be carried
instead of multiplying individual terms. Fam e diagram in a simple and straightforward manner.
be applied, and therefore the magnitude
stsaight-line asymptotes. Usirng the Bode di
controller rnay be designed with conventio
It is irnportant to note that there rnay be a difference in the phase lag-lead compensation techniques.
magnitudes for C (jo) and G( jv). The high-frequency asymptote of Phase lead compensation is cornmonly used for irnproving stability margins.
magnitude curve for G(ju) rnay be a constant-decibel line (that is, a horizon The phase lead compensation increases the system bandwidth. Thus, tbe system has
On the other hand, lim,,,G(s) = O, then the magnitude of G ( j o ) alw a faster speed to respond. However, such a systern usin phase Iead compensation
proaches zero ( - m d as o approaches infinity. For exarnple, referring to Exarnp may be subjected to high-frequency noise problems ue to its increased high-
4-11, we obtained G ( w ) for G(s) as follows: frequency gains.
hase lag compensation reduces the systern gain at higher frequencies without
reducing the system gain at lower frequencies. The S em bandwidth is reduced and
thus the system has a slower speed of response. cause of the reduced high-
The high-frequeracy magnitude of G(jv) is frequency gain, the total system gain can be increased, and thereby low-frequency
gain can be increased and the steady-state accuracy can be improved. Also, any
bigh-frequency noises involved in the system can be attenuated.
In some applications, a phase lag compensator is cascaded with a phase lead
whlle the high-frequency magnitude of the plant is compensator. The cascaded compensator is known as aphase lag-lead compensator.

1-
lirn j o
10
+ PO =O
By use of the lag-lead compensator, the low-frequency gain can be increased (which
means an improvement fn steady-state accuracy), while al the same time the system
bandwith and stability margins can be increased.
The difference in the Bode diagrams at the high-frequency end can be explained a Note that the PID controller is a special case of a phase lag-lead controller.
fo!iows. First. recail that we are interested only in the frequency range O 5 o 5 The PD control action, which a£fectsthe high-frecpency region, increases the phase
techniques for PID controllers basically follow those of phase lag-lead compe
tors. (lín industrial control systerns, however, each PID control action in the
controller may be adjusted experimentally.) Figure 4-35 Digital control system.

That is,

Ht is important that the sampling period T be chose


is to sample at the frequency 10 times that of the b
system. (Although digital controls and signal prscessing use similar approaches
ing continuous-time signals, t
numbers of bits, if the number of bits employed is insufficient, the pole and the field of signal
locations of the filter rnay not be realized exactly as desired and the resulting , while in the field of
re generally low. Sra
quencies is mainly due to tbe different
trade-offs in these twu fields.)
Substitiste w = jv into G(w) and plot the Bode diagra
ode diagram the static error constants,

t the low-frequency gain of the discrete-time controller (or


transfer function GD(w)is unity, determine the system gain
by satisfying the requirement for a given static error constant. Then, by using
conventional design techniques for continuous-time control systerns, deter-
mine the pole(s) and zero(s) of the digital controller transfer function. [CD(w)
is a ratio of two polynomials in w .f Then the open-loop transfer function of the
designed system is given by GD(w)G(w).
. Transform the controller transfer function CD(w) into GD(z) through the
inaccuracies in the locations of poles and zeros, (The important thing to remernber bilinear transformation given by Equation (4-38):
is that the poles and zeros of the filter in the z plane rnust lie on a finite number of 22-1
allowable discrete points,)

Then
GD(z) = GD(w)~w=(~,r)(z-~)i(z+i,
8, First, obtain G(z), the z transforrn of the plant preceded by a hold. Tbe* is the pulse transfer function of the digital controller.
transforrn G(z) into a transfer function G(w) through the bllinear transfwma-
6,Realiize the pulse transfer function GD(z) by a computational algorithm.
tion given by Equation (4-37):
1 + (TJ2)w esign procedure just given, it Is important to note the
z =
following:
1 - (7-/2)w
Design of Discrete-Time Control ystems by Conventional Metho ased on the Frequency-

The transfer function G(w) is fer function. Hence Thus,

K[0.01873(ii) 1 - O.lw + 0.01752]


consideration the nonmi ( ) = (1 + "1.T
The frequency axis in the w plane is distorted. 1 - 0.lw
- 1.8in7('iu'")
1- 0 . 1 ~
+ o81g

Eictitious frequency v and the actual frequency w is - K(-0.000333w2 - 0.09633~+ 0.9966)


2 wT w2 + 0 . 9 9 6 9 ~
v = - tan-
T 2
f , for exarnple, a bandwidth wb is specified, we need to desi -
K(l+ &)(1-5)
w(w + 1)
A simple phase-lead compensator will probably satisfy al1 requirements. There-
2 @bT ~

fore, we shall try lead compensation. (If Iead compensation does not satisfy al1 require-
vb = T tanl ments, we need to use a different type of compensation.)
Now let us assume that the transfer function of the digital controller GD(w) has
unity gain for the low-frequency range and has the following form:
Consider the digital control system shown in Figure 4-36. Design a digital contr
in the w plane such that the phase margin is 50°, the gain margin is at least 10 dB
the static velocity error constant Kv is 2 sec-l. Assume that the sampling peri
0.2 sec, or T = 0.2.
(This is a phase-lead compensator.) It is one of the simplest forms of the digital
First, we obtain the pulse transfer function G(z) of the plant that is preceded by
controller transfer function. (Other forms may be assumed as well for this problem.)
the zero-order hold:
The open-loop transfer function is

The static velocity error constant K, is specified as 2 sec-l. Hence,


Kv = lim wGD(w)G(w)= 16' = 2
w-io
The gain K is thus determined to be 2.
By setting K = 2, we plot the Bode diagram of G(w):
- K(0.01873~+ 0.01752)
z2 - 1.8187~+ 0.8187
Next, we transform the pulse transfer function G ( z )into a transfer function G(w
by means of the biliriear transformation given by Equation (4-37):

z =:
1 + (T12)w -
--1 + 0 . 1 ~
1 - (2'12)~ 1 - 0 . 1 ~
Figure 4-37 shows the Bode diagram for the system. For the magnitude curves
we have used straight-line asymptotes. The magnitude and phase angle of G(jv) are
shown by dashed curves. (Note that the zero at v = 10, which lies in the right half of
the w plane, gives phase lag.) The phase margin can be read from the Bode diagram
(dashed curves) as 30" and the gain margin as 14.5 dB.
The given specifications require, in addition to K, = 2, the phase margin of 50"
and a gain margin of at least 10 dB. Let us design a digital controller to satisfy these
specifications.

Design of lead compensator. Since the specification calls for a phase rnargin of
50", the additional phase-lead angle necessary to satisfy this requirement is 20". To
achieve a phase margin of 50" without decreasing the value of K , the lead compensator
Eipwe 4-36 Digital control system of Example 4-12. must corrtribute the required phace-lead angle.
ased on the Frequency-Response Method

To find the frequency point where the magnitude is -4.425 dB, we substitute w = jv
in G ( w ) and find the magnitude of C ( j v ) :

y trial and error, we find that at v = 1.7 the magnitude becomes approximately
-4.4 dB. We select this frequency to be the new gain crossover frequency v,. Noting
that this frequency corresponds to 11(V%\), or

and
ar = 0.3534

The lead compensator thus deterrnined is

The magnitude and phase angle curves for G D ( j v )and the magnitude and phase angle
curves of the compensated open-loop transfer function G D ( j v ) G ( j v )are shown by solid
curves in Figure 4-37. From the Bode diagram we see that the phase rnargin is 50" and
the gain margin is 14 dB.
The controller transfer function given by Equation (4-40) will now be trans-
formed back to the z plane by the bilinear transformation given by Equation (4-38):

Thus,

The open-loop pulse transfer function of the compensated system is


ased on the Frequency-

Unit-Step Response of Desígned System


1.6- a 1

0.0891z2+ 0.0108~- 0.0679


z3 - 2.28S5z2 + 1.84602 - 0.5255 1.4 -
0.0891(z + 0.9357)cz - 0.8145)
-
-
1.2 -
(Z - O.S126)(z - 0.7379 - j0.3196)(z - 0.7379 + j0.3196) O
o O o
o
o
1- O O O O O O O O o O O O O O O O O O o O O O O O O O O o o ~ ~

-O2 0.8 - O

0.6 - o
poles acts as dominant closed-loop poles. (Tke system behaves as if it is a secsn
0.4 -
o
unit-step response of this sy
the unit-step response curve 0.2 -
exhibits a mLurimum oversh o
06
O 5 10 15 20 25 30 35 40

k (Samplingperiod T = 0.2 sed


under normal operation of this systern.
Figure 4-38 Plot of unit-step response of the designed system.

the o value varies from O to o,. 4


titious frequency v varies from O to m,
since v = (21T) tan (oT12). Thus
hit-step response of designed systern ----------
system for O S o 5 o, is similar 4
ing analog control system for O r v 5 a.
Since C(jv) is a rational function of Y , it is basically the same as G(jo). In
determining possible unstable zeros of the cha
stability criterion can be applied. Therefore,
line approximation to the magnitude curve
concept of phase margin and gain margin ap
Compare transfer functions G(w) and G(s). As w
of the presence of the scale factor 21Tin the w
ing static error constants for G(w) and C(s)
scale factor 2/T, this will not be true.)
The w transformation may generate one or more right half-plane zeros in
G(w). If one or more t half-plane zeros exist, then G (w) is a nonminimum
phase transfer functio ecause the zeros in the right half-plane are generated
by the sample-and-hol peration, the locations of these zeros depend on the
sampling períod T. The effects of these zeros in the right half-plane on the
response become smaller as the sampling period 1" becomes smaller.
In what follows, let us consider the effects on the resgonse of the zero
in the right half-plane at w = 2lT. The zero at w = SlTcauses distortion in the
frequency response as v approaches 217'. Since
2 oT
v = - tan-
' $ 2
esign of Discrete-Time Control ysteems by Conventional Methods Chap. ec. 4-7 Analytical Design Method

then, as u approaches 'EIT, tan (wT12) approac


wT 57-

and thus

The main reason why the control actions of analog controllers are limited is that ther
are physical limitations in pneumatic, hydraulic, and eliectronic components. Suc
limitations may be completely ignored in designing digital controllers. Thus, man
control schernes that have been irnpossible with analog controls are possible
digital controls. In fact, optimal control schemes that are not possiblie with an -39 (a) A digital control system; (b) diagram showing equivalent digital
controllers are made possible by digital control schemes. control system.
In this section we specifically present an analytical design method for digi
coneroflers "eat will force the error sequence, when subjected to a speci
time-domain input, to become zero after a finite number of sampling p sarnpled every time interval T. Lhe input to the digital controller is the error signal
in fact, to become zero and stay zero after the minimum possible number of sarn e(kT). ñhe output of the digital controller is the control signal u(k2"). The control
periods. signal u(kT) is fed to the zero-order hold, and output of the hold, u(t), which
Lf the response of a closed-loop control system to a step input e is a piecewise continuous-time signal, is fed to t plant. [Although the sampler at
minimum possible settling time (that is, the output reaches the final value in the mi the input o£ the zero-order hold is not shown, the signal u(k2")is first sampled and
mum time and stays there), no steady-state error, and no ripples bet fed to the zero-order hold. As mentioned earlier, the zero-order hold shown in the
sampling instants, then this type of response is cornmonly iagram is a sarnple-and-hold esired to design a
The deadbeat response will be discussed in this section. ( GD(z) such that the closed-loop control system will exhibit the minimum possible
response again in Chapter 6, where we discuss the pole placement technique ana t settling time with zero steady-state error in response to a step, a ramp, or an
design of state observers.) acceleration input. It is required that the output not exhibit intersampling ripples
The discussions that follow are lirnited to the determination of the con after the steady state is reached. The system must satisfy any other specifications,
algorithms or pulse transfer functions o£ digital controllers for single-input-sin if required, such as a specification for the static velocity error constant.
output systems, givea desired optirnal response characteristics. For optimal con Let us define the z transform of the plant that is preceded by the zero-order
of multiple-input-multiple-outputsystems, see Chapter 8, where the state-spa frold as G ( z ) , or
approach is used.
Design of Digital CesntroUersfeso- Minimu
Error. Consider the digitai control system shown in Figure 4-39(a). The erro Then the open-loop pulse transfer function becomes G D ( z ) G ( z )as, shown in Figure
signal e([). which is rhe difference between the input r ( r ) and the output c ( f ) .i 4-39(b). Next, define the desired closed-loop pulse transfer function as F ( z ) :
iscrete-Time Control Systerns by Conventional

G+w(z) dition to the physical bility conditions, we must pay attention to


-
-
=qz) the stability aspects of the system ifically, we must avoid canceling an unstable
1 + Go(z)G(z)
pole of the plant by a zero o£the d
Since it is required that the system exh with zero steady-state any error in the pole-zero cancellation
error, the system must exhibit a finite ce, the desired ciosed- will become unstable. Similarly, the di
loop pulse transfer function must be cel plant zeros that lie outside the unit ciscle.

F(z) =
a o z N + a l z N - '+ + aN at will happen to the closed-loop puIse transfer
zN function F(z) if G(z an unstable (or critic
z = a outside (or o it circie. [Note that t
or equally , if G (z) inv r more unstable-or
E(z) = a. + alz-l + +aN~-N define
hat F(z) must not conta G(z) = -
GiW
series expansion of F(z) z-a
e for a physically real where Gl(z) does not include a term that cancels with z - a . Then the closed-loo
p pulse transfer function pulse transfer function becomes
the digital controller pulse transfer function GD(z).That is, we find t
function G,(Z) that willl satisfy Equation (4-41). Solving Equation (4-41) foa:GD(z) Gl(zb
G D ( ~z_,
)
we obtain G D ( ~ ) G ( ~=
) = F(z) (4-44)
+ G,(z)G(z) Gdz)
1 + GD(z)__
G D ( ~=)
G(z)[I - J;(z)l
Since we require that no zero of GD(z) cancel the unstable pole of G(z) at z = a,
The designed system must be physicaily realizable. The conditions for we must have
realizabifity place certain constraints on tbe closed-loop pulse transfer funct
and the digital controHer pulse transfer function GD(z).The conditions for 1 - z-a
1 - F(z) =
realizability may be stated as follows: Gdz) z -a + GD(z)Gl(z)
1+ G D ( ~ ) z _ .
The order of the numerator of GD(z>must be equal to or lower t at is, 1 - F(z) must have z = a as a zero. Also, notice t
f the denominator. (Otherwise, the controller requires future if zeros of C(z) do not cancel poles of GD(z), the zeros of G(z) become zeros of
roduce the current output .) F(z). [F(z) may involve additional zeros.]
f the plant &;,(S) involves a transportation lag e-L" then the designed ciosed Eet us surnmarize what we have stated concerning stability.
loop system must involve at Ieast the same magnitude of the transportation la
(Otherwise, the closed-loop system would have to respond before an input w 31. Since the digital controller GD(z) should not cancel unstable (or critically
glven, which is impossible for a physically realizable system.) stable) poles of G(z), al1 unstable (or critically stable) poles of G(z) must be
If G(z) is expanded into a series in z-l, the lowest-power term of the serie included in 1 - F(z) as zeros.
expansion of E(z) in z-' merst be at least as large as that of G(z). For (z) that lie inside the unit circle may be canceled with poles of
if an expansion of G(z) into a series in z-1 begins with %hez-l term, at lie on or outside the unit circle must not
first term of F(z) given by Equation (4-42) must be zero, or a. must ence, al1 zeros o£ G(z) that lie on or outside
that is, fhe expansion has to be of the form the unit circle must be included in F(z) as zeros.
F(z) = alz-l + a2z-2+ + aNzlN Now we shall proceed with the design. Since e(kT) = r(kT) - c(kT), referring
where N 2 n and n is the order of the system. This means that the plant carm to Equation (4-41) we have
respond instantaneously when a control signal of finite magnitude is a p p k
the response comes at a delay of at least one sampling period if the crri
expansion of C(z) begins with a term in z-l. Note that for a unit-step input r(t) = l(t)
esign of Discrete-Time Control Cysternc by Convenlional

c(t L nT) = constant, for ste


C(t nT) = constant, for ra
For a unit-ramp input v ( t ) = tl(t), ?(t nT) = constant, for acceleration inputs
ion must be satisfied
tion on c(t), C(t), or
And for a unit-acceleration input r(t) = it21(t), ontinuous time and
time function; therefore, to have no ripples in the out
y state must be either constant or mono
ing) for step, rarnp, or acceleration i
S, in general, z transforms of such time-domain
ten as
Since the closed-loop pulse transfer function F(z) is a polynomial in z-', all
the closed-loop poles are at the origin or at z = O. The mult
R(z) = (1 -P(z)
2-1)4+1 pole at the origin is very sensitive to syst
Although a digital control system design
where P(z) is a polynomial in z-l. Notice that for a unit-step input
with zero steady-state error in response t
q = O ; for a unit-ra p input, P(z) = Tz-' and q = 1; and for a uni
transiemt response characteristics for the
+
input, P(z) = T2z-l(l z-l) and q = 2.
inferior or sometimes unacceptable tran
By substituting Equation (4-46) into Equation (4-451, we obtain
types of input. (This is always true in
control system will exhibit the best resp
E(z) = it is designed for, but will not exhibit o

20 ensure that the system reaches steady state in a finite number o£sam hich an analog controller is discretized, an increase in the
and maintains zero steady-state error, E(z) must be a polynomiali in z- sampling period changes the system dynamics and may lead to system instabil-
number of terms. Then, by referring to Equation (4-47), we choose the ity. On the other hand, the behavior of the digital control systern we are
1 .- F(z) to be of the form designing in this section does not depend on the ch
Since the inputs r(t) considered here are time-d
2 - F(z) = (1 - ~ - l ) q + ~ A J ( z )
inputs, ramp inputs, and acceleration inputs), th
where N(z) is a polynomlal in z-1 with a finite number of terms. Then chosen arbitrarily. H;or a smaller sampling period,
an integral multiple of the sarnpling period T ) be
E(z) = P(z)N(z) a very small campling period T , the magnitude of Knltr~lsignal will become
excessively large, with the result that saturation nomena will t&e place in
which is a polynomial in z.-' with a finite number of terms, This means that is section will no longer
m, and the design method presented
signal becomes zero in a finite number of sampling periods,
ence, the sampling period T should not be too small. On the other
From the preceding analysis, the pulse transfer functlon of the digital co
hand, if the sampling period T is chosen too iarge, the system may behave
can be deterrnined as follows. y first letting F(z) satisfy the physical realiza unsatisfactorily or may even become unstable when it is subjected to suffi-
and ctability conditions and then substituting Equation (4-48) into Equation (4-4
ciently time varying inputs (such as frequency-domain inputs). Thus, a compro-
we obtain
mise is necessary. A rule of thumb is to choose the smallest sampling period
T such that no saturation phenomena occur in the control signal.
G"(') = G(z)(l - Z - ' ) ~ + ' N ( ~ )

Equation (4-50) gives the pulse transfer function of the digital controller tkat Consider the digital control systern shown in Figure 4-39(a), where the plant transfer
produce zero steady-state error after a finite number of sarnpling function %(S) is given by
For a stable plant G,(s), the csndition that the output not exhibit inters
r&@es after the settlirag time is reached may be written as fcsllows:
esign of Discrete-Time Control yctemc by Conventional Methodc Chap. 4 -7 Analytical Design Method

Design a digital controller G D ( z )such that the closed-loop- system


. will exhibit a
deadbeat response to a unit-step input. (in a deadbeat response the system should n
exhibit intersampling ripples in the output after the settling time is reached.) T
sarnpling period T is assurned to be 1 sec. Then, using the diktal controller G D [ z )
designed, investigate the response of this system to a unit-ramp input.
The first step in the design is to determine the z transform of the plant that For U ( z )to be a series in z-' with only two terrns, F ( z ) must be of the following form:
preceded by the zero-order hold:
F ( z ) = (1 + 0 . 7 1 8 1 ~ - ~ ) 4
z-~ (4-54)
where Fl is a constant. Then U ( z ) can be written as follows:

Equation (4-55) gives U ( z ) in terrns of El. Once constant fi is determined, U ( z ) can


be @ven as a series in z-1 with only tws terms.
Now we shall determine N ( z ) , F ( z ) , and K . y substituting Equation (4-52) into
Equation (4-531, we obtain
1 - alz-' - a2z-" (1 - z-l)N(z)
The left-hand side of this last equation must be divisible by 1 - z-'. If we divide the
Now redraw the block diagram of the system as shown in Figure 4-39(b). Define t left-hand side by 1 - z-', the quotient is 1 + ( 1 - al)z-' and the remainder is
closed-loop pulse transfer function as F(z). or
(1 - al - a2)z-2. Hence, N ( z ) is determined as

and the remainder must be zero. This requires that


Notice that if G ( z )is expanded into a series in z-' then the first term will be 0.3679~-
Hence, F(7) must begin with a term in z-l. 1-al-a2=0
Referring to ~ q u a t i o n(4-42) and noting that the system is of the second ord
Also, from Equations (4-52) and (4-54) we have
(n = 2), we assurne F ( z ) to be of the following form:
F ( z ) = al z-' + a2zW2= (1 + 0 . 7 1 8 1 ~ - ' ) z -F,~
Hence,
al + a2z-l = ( 1 + 0.7181z-')fi
Division of the left-hand side of this last equation by 1 + 0.7181z-' yields the quotient
al and the remainder (az - 0.7181ai)~-1.By equating the quotient with f i and the
remainder with zero, we obtain

and

Solving Equations (4-57) and (4-58) for al and az gives


al = 0.5820, a2 = 0.4180
Thus, F ( z ) is determined as

and
f i = 0.5820

Equation (4-56) @ves


Design of Discrete-Time Control yctems by Conventional Methods Chap. ec. 4-7 Analflical Design Method 1

The digital controller pulse transfer function G D ( z )is tben determined from Equatio
(4-50), as follows. By referring to Equations (4-51), (4-54), and (4-60),

With the dioital


" controller thus desioned.
" the closed-loo~~ u l s etransfer functio
I L ....
becomes as follows:

Hence,

Figure 4-40 Responses of the system designed in Example 4-13. (a) Plots of c ( k )
versus k , u ( k ) versus k , and u ( t ) versus t in the unit-step response; (b) plots of c ( k )
versus k , u ( k ) versus k , and u ( t ) versus t in the unit-ramp response.

The signal u ( k ) becomes constant ( b = 1) for k 1 2. Hence, the cystem output will not
exhibit intersampling ripples. Figure 4-40(b) shows plots of c ( k ) versus k , u ( k ) versus
k , and u ( t ) versus t in the unit-ramp response.
Note that the static velocity error constant Kv for the present system is

0.58202-' + 0.4180z-'
= lim =
Z-1 1 + 0.41802-'
esign of Discrete-lime Control ystemc by Conventional Methods Ckap. ec. 4-7 Analytical Design Meahod

Thus, the steady-state error in the unit-ramp response is (If a catisfactory result is not obtained, we must assume N > 3.) Since the input is a
step function, from Equation (4-48) we require that
1 - n(z) = (1 - Z-')N(z) (4-62)
which is indicated in Figure 4-40(b).
Note that the presence of a critically stable pole at z = 1 in the plant pulse transfer
function G(z) requires 1 - F(z) to have a zero at z = 1. owever, the fmction
1 - F(z) already has a term 1 - z-l and therefore satisfies the stability requirement.
The requirement that the static velocity error constant be 4 sec-l can be written
as follows:

where we used Equation (4-50) with q = O. Notice that from Gquation (4-62) we have
F(1) = 1. Hence, Kv can be written as follows:

Since the system output shouId not exhibit intersampling ripples after the settiing time
is reached, we require U(z) to be of the following form:

Because the plant transfer function G,(s) involves an integrator, b must be zero.
Consequently, we have
U(Z) = bo + blz-' + b2z-'
Also, frorn Figure 4-39(b), U(z) can be given by

For U(z) to be a series in z-' with three terrns, F(z) must be of the foilowing form:

where fi(z) is a first-degree polynomial in z-l. Then U(z) can be written as follows:

From Equations (4-61) and (4-62), we have

~ 1 - z-', the quotient is 1 + (1 - al).¿-' +


Hf we divide 1 - a, z -'- azz V 2- a3z - by
(1 - a, - a2)zeZand the remainder is (1 - al - az - a 3 ) ~ - ~Hence,
. N(z) is deter-
mined as
W(z) = 1 + (1 - al)z-' + (1 - al - az)zw2 (4-66)
iccrete-Time Control Systems by Conventional Met

and the rernainder must be zero, so that


1-al-a,-a3=0
Note that from Equation (4-63) we require W ( 1 ) = i.Therefore, by su
z-' = 1 into Equation (4-66), we obtain
2a1 + az = 2.75
Also, Equation (4-64) can be rewritten as
The unit-step response sequence has a malramum overshoot of approximately 50%. The
F ( z ) = al 2-' + a2z-' + a32-3 = ( 1 + O . 7 1 8 1 ~ - ~ ) ~&-(' z ) settling time is 3 sec.
Hence, Notice that from Equation (4-65) we have
al + a 2 z y 1+ a 3 z W z= ( 1 + 0 . 7 1 8 1 ~ - ' ) R ( z ) U ( z ) = 2.7181(1 - 0.36792-')(1.26184 - 0.679792-')
Division of the left-hand side of this last equation by 1 $ 0.7181~-' yields the qu
[a, + (az - 0.7181al)z-'1 and the remainder [a3- 0.7181(a2 - 0.7181a1)]z-2 Thus, the control signal u ( k ) becornes zero for k 2 3. Consequently, there are no
equating the quotient with F,(z) and the remainder with zero, we obtain intersampling ripples in the response. Figure 4-41 shows plots of c ( k ) versus k , u ( k )
Fl(z) = al + (a2 - 0.7181al)z--' versus k , and u ( t ) versus f in the unit-step response. Notice that the assumption of
N = 3 , that is, the assumption o£ F ( z ) as given by Equation (4-61), is satisfactory.
and
a3 - 0.7181(az - 0.7181al) = O
Soiving Equations (4-67), (4-68), and (4-69) for al, a2, and a3 gives
al = 1.26184, a2 = 0.22633, a3 = -0.48816
Thus, F ( z ) is determined as
F(z ) = 1.261842-' + 0.22633z-' - 0.48816~-~
and
fi(z) = 1.26184 - 0.679792-'
Equation (4-66) gives
N ( z ) = 1 - 0 . 2 6 1 8 4 ~ --
~ 0.48817~-~
The digital controller pulse transfer function GD(z) is then determined from Equ
tion (4-50) as follows:
F(z
G"(z) = G ( z ) ( l - z - ' ) N ( z )

- ( 1 + 0 . 7 1 8 1 ~ - ~ ) ~ - ' ( 1 . 2 6 1-
8 40.679802-')
+0-7181z-')2-'
( 1 - 2-1)(1 - 0 . 2 6 1 ~ 4 ~ --1 0.48817z--2)
( 1 - z V ' ) ( 1 - 0.36792-')
( 1 - 0.53872-')(1 - 0.36792-')
= 3.4298 ---
( 1 - 0.84182-')(1 + 0 . 5 7 9 9 ~ ~ ' )
With the digital controller thus designed, the system output in response
unit-step input r ( t ) = 1 is obtained as follows:
C(z)= F(z)R(z)
Figure 4-41 Plots of c ( k ) versus k ,
= (1.2,61842-' + 0.22633.T2 - 0 . 4 8 8 1 6 ~1' --~1z)-- - ~ u ( k ) versus k , and u ( t ) versus t in the
unit-step response of the system de-
= 1.2618~-' + 1 . 4 8 8 2 ~+~Y~ 3+ z - ~+ . signed in Example 4-14.
esign of Discrete-Time Control ystems by Conventional [Vethods
[Vethod: Chap.

Next, let us investigate the response of this system to a unit-ram


unit-ramp input:
C
C (( zz )) =
=FF (( zz )) R
R (( zz )) we note thát the latter irnproves the ramp response characteristics at the e
-. 1 the settling time. (The latter system requires one extra sampling period to
= (1.26184z-' + O.22633z-' - 0.4$$16zz3)( 1 - , 1, steady state.) Note also that the former as beteer step response characteri
z is, a shorter settling time and no overshoot onn the objectives of the
= 1.2618z-' + 2 . 7 5 0 0 ~ -+~ 3 . 7 5 0 0 ~ -+~ . system, we rnay choose one over the other. response
response characteristics
characteristics
foollows:
In the unit-ramp response, the control signal U ( z ) is obtained as foollovc- are required, then the system should be d the ramp input as the

F ( z ) qz) =-F(z) - 1 -
U(Z) = -= -
2-'

G(z) G ( z ) G ( z ) 1 - z-' 1 - z-l


z -l
= (3.4298 - 3.
3.10962-'
10962-' 4- 0 . 6 7 9 8 ~ ~ ~ ) -
1 - 2-'
= 3.42982-1 + 0 . 3 2 0 2 ~ -+~ z w 3+ k W-t-4 z - ~+
The signal u ( k ) becomes constant ( b = 1) for k 2 <
2 3. Hence, the system output Show that geometrically the patterns of the poles near z = 1 in the z plane are similar
exhibit intersampling ripples. Figure 4-42 shows plots of c ( k ) versus k , u ( k ) v to the patterns of poles in the s plane near the origin.
and u ( t ) versus t in the unit-ramp response. Wotice that the steady-state
steady-stai err
unit-ramp response is e, = 1/K, = $, as indicated in Egure 4-42.
z = p
Near the origin of the s plane,
z = eTs = 1+ +. + -.-
L T ~ ~ ~

z-l=Ts
Thus, geometrical patterns of the poles near z = 1 in the z plane are similar to the
patterns of poles in the s plane near the origin.

the system described by


y ( k ) - 0.6y(k - 1) - 0.81y(k - 2) + 0.67y(k - 3) - 0.12y(k - 4) = x ( k )
where x ( k ) is the input and y ( k ) is the output of the system. Determine the stability
of the system.
2 Soliaation The pulse transfer function for the system is

o
X(Z) 1 - 0.62-' - 0.81z-' + 0.672-3 - O . l 2 ~ - ~
-2 z4
-
+ 0.672 - 0.12
e

z4 - 0 . 6 ~- ~0.812'
Define
P ( z ) = z4 - 0 . 6 -
~ ~0.81z2 -i- 0.672 - 0.12
2 =aoz4+a1z3+a2z2+a3z-t-a4, ao>O

oO Figure 4-42 Plots of c(k c ( k ) versus k > Then we have


) u ( k ) versus k , and u(t) vversus
c r in the
unit-ramp response of the
response of the system ao = 1
-2 unit-ramp
designed in Example 4-lr 4-14. = -0.6
Design of Discrete-Time Control ystems by Conventional Methods Chap. 4 Exarnple roblerns and Soieitions

a, = -0.81 By dividing both sides of this last equation by -0.14, we obtain


a3 = 0.67 w 3 - 7.571w2 - 3 6 . 4 3 ~- 14.14 = 8
a4 = -0.12 The Wouth array for Equation (4-71) becomes as foliows:
The Jury stability conditions are: one sgin, w3 1 -36.43
l. la4/< ao. This condition is clearly satisfied. c h a n g e L w2 -7.571 -14.14
. P(1) > O. Since
P(1) = 1 - 0.6 - 0.81 + 0.67 - 0.12 = 0.14 > O
the condition is satisfied. Routh stability criterion states that the number of roots with positive real parts
79- P(-1) > O. Since is equal to the number of changes in sign of the coefficients of the first column of the
array. Since there is one sign change for the coefficients in the first column, there is one
P(-1) = 1 i- 0.6 - 0.81 - 0.67 - 0.12 = O
root in the right half of the w plane. This means that the original characteristic equation
the condition is not satisfied. P ( - 1) = O implies that there is one root at z = - given by Equation (4-70) has one root outside the unit circle in the z plane. The system
lbsi > lbo/.Since is unstable. (Compare the amount of computation needed in the present method and
that needed in the Jury stability test. See in particular Example 4-6.)

Lonsider the system defined by

the condition is satisfied.


5. /c2j > /COI. Since

The sampling period T is 0.5 sec. Using , plot the unit-ramp sesponse up to
k = 20.
slaritioarn The unit-ramp input u may be written as
u=kT, k = 0 , 1 , 2 ,...
In the MATLAB program, this input can be given as

where N is the end of the process considered.


A M-ATLAB program for plotting the unit-ramp response of the system consid-
ered is given in MATLAB Program 4-3. The resulting plot is shown in Figure 4-43.

Show that if the characteristic equation for a closed-loop system is written as

where A ( z ) and B ( z ) do not contain K, then the breakaway and break-in points can
be determined from the roots of
Clearing the fractions by rnultiplying both sides of this last equation by ( w - 1j39
get
-0.14w3 + 1.06w2 + 5 . 1 0 ~+ 1.98 = O where the primes indicate differentiation with respect to z .
iscrete-rime Control Systems by Convenlional 4 Exarnple Problems and Colutions "2

Sslution Let us write the characteristic equation as


f(z) = A ( z ) + KB(z) = O (4-72)
Suppose that f ( z ) = O has a multiple root o£ order r . Then f ( z ) may be wriRen as
f(z) = (2 - z , ) ~ ( z- 22) ' ' ' (2 - z p )
If we differentiate this equation with respect to z and set z = z l , we get

This means that multiple roots of f ( z ) will satisfy the following equation:

where

Solving Equation (4-73) for K , we obtain

Unit-Ramp Response
1O

This particular value of K will yield rnultiple roots of the characteristic equation. li we
substitute this value of K into Equation (4-72), we obtain

If this last equation is solved for z , the points where rnultiple roots occur can be
obtained. 8 n the other hand, from Equation (4-72) we have

and

If dK1dz is set equal to zero, we get the same equation as Equation (4-74). Therefore,
k
the breakaway or break-in points can be simply determined from the roots of
Figure 4-43 Unit-ramp response of the system considered in Problem -4-4-4.

It should be noted that not al1 the solutions of Equation (4-74) or of dKidz = O cor-
respond to actual breakaway or break-in points. Cuch a point for which dKldz = O is
hap. 4 Exarnple Problerns and Solutions

an actual breakaway or break-in point if and only if the value of K at this point is
positive value.

Discuss the procedure for designing lead compensators for digital control syste
the root-locus method.

Figure 4-45 Digital control system.

first obtain the z transform of G(s). Referring to Example 3-5, we get


F- 7

Next we construct root locus diagrams for the three cases considered.
dominant closed-loop poles. l. Sampling period T = 1: For T = 1, Equation (4-75) becomes
K[(1 - 1 + e-')z-' + (1 - e-' - e - ' ) ~ - ~ ]
G(z) =
(1 - z-l)(l - e-lz-')
4. This angle must be contributed by the lead compensator if the new
is to pass through the desired locations for the dominant closed-loop po
3. Assume the lead compensator GD(z) to be
1+ az Wotice that G(z) possesses a zero at z = -0.7181 and poles at z = 1 and
GD(z) = KDa ----- O<a<l
1 + a7-z ' z = 0.3679. The breakaway point is at z = 0.6479, and the break-in point is at
If static error constants are not specified, determine the location of t z = -2.0841. The root-locus diagram for this case is shown in Figure 4-46(a).
zero of the lead compensator so that the lead compensator will con The value of gain K of any point on the root Ioci can be determined from the
necessary angle 4. If no other requirements are imposed on the syst magnitude condition
make the value of a as large as possible. A larger value of a generally res
a larger value of Kv, which is desirable. (If a particular static error const
specified, it is generally simpler to use the frequency-response appr
If we choose a point z on the root loci, the value of K at that point can be
5. Determine the open-loop gain of the compensated system from the
condition. calculated by substituting the value of z into this last equation. (This means that
with this value of K that particular point becomes a closed-loop pole.) The critical
gain is found to be K = 2.3925.
Once a compensator has been designed, check to see whether or not al1
formance specifications have been met. If the compensated system does not me Sampling period T = 2: For the sampling period T = 2, we have from Equa-
performance specifications, then repeat the design procedure by adjusting the tion (4-75)
pensator pole and zero until al1 such specifications are met. If a large statkc
constant is required, cascade a lag network or alter the lead compensator to a lag
compensator.
The pulse transfer function G(z) in this case possesses a zero at z = -0.5232 and
poles at z = 1 and z = 0.1353. The breakaway point is at z = 0.4783, and the
Draw root locus diagrams in the z plane for the system shown in Figure 4-45 f break-in point is at z = -1.5247. The root-iocus diagram for this case is shown
following three sarnpling periods: T = 1 sec, T = 2 sec, and T = 4 sec. in Figure 4-46(b). The critical gain IC for stability is Ik = 1.4557.
iscrete-Time Control ha p.

z plane The root-locus diagram is shown in Figure 4-46(c). The critica1 gain for stability
K = 2.3925 is K = 0.9653.
Frorn the three cases considered, notice t
period is, the larger the critica1 gain K for stabiiity.
= 1 sec)

Consider the digital control system shown in Figure 4-47, where the plant is of the first
order and has a dead time of 2 sec. T e sampling period is assumed to be 1sec, or T = 1.
Design a digital PI controller such that the dominant closed-loop poles Rave a
damping ratio 5 of 0.5 and the number of samples er cycle of damped sinusoidal
oscillation is 10. Obtain the response of the systern to a unit-step input. Also, obtain
the static velocity error constant Kv and find the steady-state error in the response to
a unit-ramp input.

\ I z plane

I
-47 Digital control system.

Soltiiition The pulse transfer fianction of the plant that Is preceded by a zero-order
hoId is

z plane
Unit circle
\ /'-

The digital PI controller has the following pulse transfer function:

6 Root-locus diagrams for the system shown in Figure 4-45 when (a)
T = 1 sec, (b) T = 2 sec, and (c) T = 4 sec.

3. Sampling period T = 4: For the case of T = 4, Equation (4-75) gives The open-loop pulse transfer function becomes

The breakaway point ic at z and the break-in point is at z = -0.945


kap. 4 Example

z plane

(Note that this point is the intersection of the 5 = 0.5 lacus and the line from the origin
having an angle of 36".)
If point P is to be the closed-loop pole locatisn in the upper half of the z plane,
then the angle deficiency at point B is

The controller zero must contribute f93.52". This means that the zero of the
controller must be located at z = 0.5881. Therefore,

Hence, the PI controller is deterrnined as follows:

where K = M, -t K,. Gain constant K is determined from tke magnitude condition:


Figure 4-48 Pole and zero locations in the z ~ l a n eof the svstem considered in

We locate the open-loop poles in the z plane as shown in Figure 4-48. There is on Thus,
open-loop zero involved in this case, but its location is unknown at this point. + Ki = 0.5070
Since it is required to have 10 sarnpies per cycle of damped sinusoidal oscillatio
the dominant closed-loop pole in the upper half of the z plane must lie on a 1 From Equations (4-76) and (4-77), we find that
from the origin having an angle of 360°/10 = 36". Frorn Equations (4-1) and (4 K, = 0.2982 and K, = 0.2088
rewritten as
Hence, the PI controller just designed can be given by

Finally, the open-loop pulse transfer function becomes

the desired closed-loop pole location can be determined as follows. Noting tha

The closed-loop pulse transfer function becomes


or w ~ / w ,= 0.1. Since 5 is specified as 0.5, we have

The response c ( k T ) to the unit-step input can be obtained easily by use of


MATLAB. A MATLAB program for plotting the unit-step response is shown in
The closed-loop pole is located at point P in Figure 4-48, where (at point P ) MATLAB Program 4-4. The resulting plot is shown in Figure 4-49.
n of Discrete-Time Control Chap. 4 Exarnple Problerns and Solutions

Digital
controller S(S + 1)
% U n it-step response ----------

num = [O O O 0.3205 -0.18851;


den = [l -1 3 7 9 0.3679 0.3205 -0.1 8851;
r = oneo;(1,571;

Figure 4-33 Digital control system.

title('Unit-Step Response') We shall first locate the desired closed-loop poles in the z plane. Referring
xlabel('k'f to Equations (4-1) and (4-2), for a constant-damping-ratio locus we have
yIabel('c(k)')

- - Unit-Step Response and

Since we require eight samples per cycle of damped sinusoidal oscillation, the dominant
closed-loop pole in the upper half of the z plane must be located on a line having an
angle of 45" and passing through the origin as shown in Figure 4-51. (Note that the
number of samples per cycle is 360'18. Hence, eight samples per cycle requires 8 =
360"/8 = 45".) Thus,

which gives
d - 1
-U =
o-), 8
Since the sampling period T is specified as 0.2 sec, we have

Therefore,
9 Plot of c ( k T ) versus kT for the system designed in Problem A-4-8.
(Sampling period T = 1 sec.)

y letting 5 = 0.5 and substituting Equation (4-79) into Equation (4-78), we obtain
IZI = e-0.4535 = 0.6354
Corisider the system shown in Figure 4-50.
that the dominant closed-Ioop poles of the system will have a damping ratho 5 Hence, we can locate the desired closed-loop pole in the upper half of the z plane, as
0.5. We also want the number of samples per cycle of damped sinusoidal oscillati~n shown by point P in Figure 4-51. Note that at point P
be 8. Assurne that the sampling period T is 0.2 sec.
Wsing the root-locus method in the z plane, determine the pulse Izl /Z = 0.6354/450 = 0.4493 + 10.4493
of the digital contrdler. Obtain the response of the designed system to Next, we obtain the pulse transfer function G ( z ) of the plant that is preceded by
klso obtain ehe static velocity error csnstant K v . a zero-order hold:
.4 Exarnpie Problems and 71
Design of Discrete-Time Control Cystems by Onventional Methods
and choose the zero of the controller to cancel the pole at z = 0.8187. Then the pole
of the controller can be determined easily from the angle condition as z = 0.1595. Thus,
we have

The open-loop pulse transfer function of the system is therefore obtained as follows:

The gain constant K can be determined from the magnitude condition:

Hence, we have determined the pulse transfer function of the digital controller to be

The open-loop pulse transfer function is

The closed-loop pulse transfer fiinction is

Because of the cancellation of a pole of the plant and the zero of the controller, the
order of the system is reduced from third to second. The system has only a pair of
conjugate complex closed-loop poles.
Figure 4-52 shows the unit-step response sequence c ( k T ) versus k T . The plot
shows the maximum overshoot to be approximately 16.5%.
Finally, the static velocity error constant Kv is determined as follows:

1 - z-'
z-1

= lim
,i [-------
1 - 2-' 0.2610(1 + 0.93562-')z-'
0.2 ( 1 - 0.15952-')(1 - z - ' )
Chap. 4 Exarnple Problerns and

Figure 4-53 Digital control system in the w piane.

Using the gain KD thus determined, draw a ode diagram of Gl(w),the gain-
adjusted but uncompensated system. Evaluate the phase margin.
Determine the necessary phase lead angle 4 to be added to the system.
. -
Add 5" 12" to & to compensate for the shift of the gain crossover frequency.
Define this added & as 4,. Determine the attenuation factor cx from the foollowing
equation:
1-a
sin cb, = -
l + a
ermine the frequency point where the magnitude of the uncompensared
em Gi@) is equal to -20 log (U&). Select this frequency as the new gain
crossover frequency. This frequency corresponds to v, = ll(V"&),and the max-
imum phase shift +,occurs at this frequency.
Determine the cornvr frequencies of the lead compensator as follows:

Zero of lead compensator: Y -1


= T

1
Bole of lead compensator: Y = -
a7
7. Check the gain margin to be sure it is satisfactory. kf not, repeat the design process
by modifying the pole-zero Iocation of the compensator until a satisfactory result
is obtained.

The primary function of a lag cornpensator is to provide attenuation in the


high-frequency range to give a system sufficient phase margin. The phase lag character-
istic is of no consequence in lag compensation.
Assume the following form for the lead compensator:
1 + TW
GD(w) = KD------ O<a<l
1 + arw9 . Assume the following form for the lag compensator:
The open-loop transfer function of the compensated system may be writt

The open-loop transfer function of the compensated system may be written as


--

where G l ( w )= K D G(w).Determine gain KD to satisfy the requirement o


given static velocity error constant.
Exarnple Problerns and

z plane

Root-locus diagram for the system designed in Problem A-4-11.

the root-locus diagram, constant 5 loci for 6 = 0.5 and 0.6 are superimposed. From
the diagram it can be seen that the damping ratio 5of the closed-loop poles is approx-
imately 0.55.
The line connecting the closed-loop pole in the upper half of the z plane and the
origin has an angle of 37". ence, the number o£samples per cycle of damped sinusoidal
oscillation is 360°/37" = 9.73.

Consider the digital control system shown in Figure 4-57> where the plant transfer
function is 11s'. Design a digital controller in the w plane such that the phase margin
is 50" and the gain margin is at least 10 d . The sampling period is 0.1 sec, or T = 0.1.
After designing the controller, obtain the static velocity error constant K v . Also, obtain
the response of the designed system to a unit-step input.
shall first obtain the z transform of the plant that is preceded by the
zero-order hold:

Next, using the bilinear transformation given by


iscrete-Time Control ystems by Conventional hap. 4 Example Problems an

and gain margin. By applying a conventional design technique, it can be seen that the
following lead network will satisfy the requirements:

The addition of this lead network modifies the ode diagram. The gain crossover
frequency is shifted to v = 4. Note that the maximum phase lead 4, that this lead
network can produce is 61.93", since
re 4-57 Digital control system. 1-L
4, = sin-l---6 =. sin-' 0.8824 =: 61.93"
l + &
we transform G ( z ) into G(w):
At the gain crossover frequency v = 4, the phase angle of GD(jv)G(jv)becomes
-191.31" i61.93" = -129.38". Thus, the phase margin is 50.62". The gain margin is
found to be approximately 13 dB. ence, the given design specifications are satisfied.
now transform the controlier transfer function CD(W>into GD(z).
the biíinear transfsrmation
Thus,

we obtain
Figure 4-58 shows the e diagram of G ( j v ) thus obtained. Wotice t
margin is -2". ft is nec

Hence, the open-loop pulse transfer function becomes

The static velocity error constant Kv is obtained as follows:

Thus, the static velocity error constant K,, is infinity. There is no steady-state error in
the ramp response.
The closed-loop pulse transfer function of the system is

v (radlsec)
Figure 4-59 shows the unit-step response. Notice that the zero of the digital controller
at z = 0.9048 is close to the double pole at z = 1. A pole-zero pair near point z = 1
gure 4-58 Bode diagrarn for the system considered rn Problem A-4-12. creates a long tail with srnall amplitude in the response.
Chap. 4 Example Problerns and

1.6 1.8 2.0 2.2


k T ísec)

Figure 4-59 Plot of c ( k T ) versus kT for the system designed in Problem A-4-12.

1 (a) Desired output c ( t ) in response to a unit-step input; (b) plot of u ( f )


versus t .
Consider the digital control system shown in Figure 4-60. The plant transfer fu
involves a transportation lag e-'". The delay time is 5 sec, or L = 5. Th
c ( t ) in response to a unit-step input is as shown in Figure 4-61(a). The The z transform of the plant that is preceded by the zezo-order hold is
zero to the final value in 10 sec (measured frorn t = 5 e0 t = 15) and

course, choose the sampling period to be 2.5 sec, 1 sec, or another value. In
example, however, to simplify our presentation, we set the sampling period at 5 S
Notice that there is no unstable or critically stabie pole involved in G(z). Therefore,
there is no stability problem involved in this case.
Let us define the closed-loop pulse transfer function as F(z):

In the present case the output c ( t ) in the unit-step response is specified as shown in
Figure 4-61(a). Since h[l - e-0.1(15-5)]= h ( l - e-') = 1, we have h = 1.5820. from
the deadbeat response curve shown in Figure 4-61(a), we obtain
c(0) o
=

Figure 4-60 Digital control system. c(1) = o


Design of Diccrete-Time Control Systernc by Conventianal hap. 4 Exarnple Problerns and Solutions

Taking the inverse z transform of U ( z ) ,we find that u ( k ) is constant for k r 2. Thus,
there are no intersampling ripples in the output after the settling time is reached. The
signal u ( t ) versus t is plotted in Figure 4-61(b).
from which we get

Consider the digital control system shown in Figure 4-62. Design a digital controller
G D ( z )such that the closed-loop system will exhibit the minimum settling time with zero
steady-state error in a unit-ramp response. The systern should not exhibit intersampling
ripples at steady state. The sampling period Tis assumed to be 1 sec. After the controller
is designed, investigate the response of the system to a Kronecker delta input and a
Noting that unit-step input.
Soliution The first step in the design is to determine the z transform of the plant that
is preceded by the zero-order hold:

we obtain
F ( z ) = 0.6225z-' + 0 . 3 7 7 5 ~ -=~ 0.6225(1 + 0 . 6 0 6 5 z - l ) ~ - ~
Once F ( z ) is determined, the pulse transfer function of the digital controller can b
obtained frorn Equation (4-80):
Now define the closed-loop pulse transfer function as E ( z ) :
GD(z) =
G ( z ) P - F(z)l
Notice that from Equation (4-48) we have
Notice that if G ( z ) is expanded into a series in z-' then the first term will be 0 . 5 ~ ~ ' .
1 - F(z) = (1 - zV1)N(z) Hence, F ( z ) must begin with a term in z-':
or
1 - 0.6225zp2 - 0 . 3 7 ' 9 5 ~ -=~ ( 1 - z-')IV(z) where N r n and n is the order of the system. Since the system here is of the second
By dividing (1 - 0 . 6 2 2 5 ~ --~ 0 . 3 7 7 5 ~ - ~by) ( 1 - z-'), N ( z ) can be deterrniined order, n = 2.
follows: Since the input is a unit ramp, from Equation (4-48) we require that

N(z) = 1 + z-' + 0 . - 3 7 7 5 ~ - ~ 1 - F ( z ) = ( 1 - z-')'IV(z) (4-81)


Consequently, Notice that G ( z ) has a critically stable double pole at z = 1. Therefore, from the
1 - F ( z ) = ( 1 - z-')(1 + z-' + 0.37752-') stability requirement, 1 - F ( z ) must have a double zero at z = 1. However, the
function 1 - F ( z ) already involves a term ( 1 - z-')', and therefore it satisfies the
and
stability requirement.
o ,, 0.6225(1 + 0.6065~-')2-~ Since the system should not exhibit intersampling ripples at steady state, we
require U ( z ) to be of the following type of series in z-':
U ( Z )= bo + blz-' + b 2 z V 2+ - - + b N - l ~ - N ++l b +
( ~ - f N~- ' + z-~-' + - .)

This last equation gives the pulse transfer function of the digital controller. Since
must be unity at steady state, u ( t ) , a continuous-time signal, must be constant after
steady state is reached.
Eet us determine U ( Z ) :
,,, . C(z) 0 . 6 2 2 5 ~ ~+' 0 . 3 7 7 5 ~ - ~ /I - n367~~~21

Figure 4-62 Digital control system.


lesign oC Discrete-Time Control Systems by Conventional Methods Chap. roblems and Ssllations

Because the plant transfer function C;,(s) involves a double integrator, b must be ze and the remainder is set equal to zero:
(Otherwise, the output increases parabolically, instead of linearly.) Consequently,
have al - az + a3 = O (4-86)

U ( Z )= bo + b l 2-' + . . . + b ~ - l y solving Equations (4-84), (4-85), and (4-86) for al, a,, and a', we obtain

From Figure 4-62 U ( Z )can be given by al = 1.25, a2 = 0.5, a3 = -0.75


Wence ,

and
$ ( z ) = 1.252-' - 0.75z-' = 1.252-'(1 - 0.62-')
and F ( z ) is determined as follows:
F ( z ) = 1.252-' + O . ~ Z - -~ 0 . 7 5 ~ - ~
For U ( z ) to be a series in 2-' with a finite nurnber of terms, F ( z ) rnust be
by 1 + z-':
F ( 2 ) = ( 1 + z-')Fi(z) The digital controller G u ( z ) is then determined from Equation (4-50):

Then U ( z ) can be written as follows:


U ( z )= 2F(2)
where & ( Z )is a polynomial in z-' with a finite number of terms.
By cornparing Equations (4-81) and (4-82) and by making a simple analysis,
see that F ( z ) must involve a term with at least z - ~ Hence,
. we assurne
~ ( z =) al z-' + a 2 z - 2 + a32-'
This assumed form of F ( z ) involves the minimurn number of terms; the tra ith the digital controller thus designed, the system output in response to a
response will settle in three sampling periods. unit-ramp input is obtained as follows:
We shall now determine constants a l , a2, and a3. Frorn Equation (4-81), we
1 -. al 2-1 - a22-2 - a32 - 3 = ( 1 - 2 - 1 12 N ( z )
If we divide the left-hand side of this last equation by ( 1 - z-')', the quotient is
(2 - a1)z'-'. The remainder is [2(2 - a l ) - ( 1 + Q ~ ) ] Z --~ [(2 - a l ) + a 3 ] ~ - 3He
.
N ( z ) is deterinined as
N(z) = 1 + ( 2 - al)z-' Hence,

and the rernainder is set equal to zero: c(0) = o

[Y(:! - al) - ( 1 + a 2 ) ] ~-- ~( 2 - al + a 3 ) ~ - =3 O


To satisfy this last equation, we require that
2(2 - a l ) - (1 + a2) = 0
Notice that from Equation (4-83) we have
2-al+a3=0
From Equation (4-K!), we have
a 1 . Y 1 + a22-2 + a3z-3 = ( 1 + z-')F1(z)
Jf we divide the left-hand side of this last equation by 1 + 2-l, the quotient
al z-' + ( u z - ul)z-'. The remainder is (al - a2 + a3)z-3. Thus, the control signal u ( k ) becomes zero for k s 3. Consequently, there are no
intersalñrpling ripples in the response at steady state. Figure 4-63 shows plots of c ( k )
F1(z) = a l z - ' 4 (a2 - a l ) z W 2 versus k , u ( k ) versus k , and u ( t ) versus t in the unit-ramp response.
Design of Biscrete-Time ontrol Systems by v. Chap. 4 Example Problems and

For the unit-step input,


C ( Z )= F ( z ) R ( z ) = (1.252-' + 0.52-' - 0 . 7 5 ~ - ~1) --12-'
- ~z - ~+ z - ~+ z-' +
= 1 . 2 5 ~ ~+' 1 . 7 5 ~ +

The maximurn overshoot is 75% in the unit-step response. Notice that

= 1.25(1 - 0.62-')(2)(1 - z-')


= 2.5 - kv'+ 1 . 5 z - ~
The control signal u ( k ) becomes zero for k 2 3. Gornsequently, there are no inter-
sarnpling ripples after the settling time is reached. Figure 4-64(a) shows plots of c ( k )
versus k , u ( k ) versus k , and u ( t ) versus t in the response to the Kronecker delta input.
Figure 4-64(b) shows similar plots in the unit-step response. Notice that when the
system is designed for the ramp input the response to a step input is no longer deadbeat.

(sed
-63 Plots of c ( k ) versus k ,
u ( k ) versus k , and u ( t ) versus t in the
unit ramp response of the system
designed in Problem A-4-14,

Next, let us investigate the response of this system to a Kronecker d e i ~ aI I ~ J U ~


a unit-step input. For a Kronecker delta input.
C ( z ) = F ( z ) R ( z ) = F ( z ) = 1.252-' + 0.5z-' - 0 . 7 5 ~ - ~
Notice that U ( z ) in this case becomes as follows:

Figure 4-64 (a) Plots of c ( k ) versus k , u ( k ) versus k , and u ( t ) versus t in the response to
The control signal u ( k ) becomes zero for k 2 4. Heme, there are no iritef-samp the ICronecírer delta input of the system designed in Problem A-4-14; (b) plots of c ( k ) versus
ripples after r 3 4T = 4, k , u ( k ) versus k, and u ( t ) versus t in the unit-step response of the same system.
Design of Discrete-Time Contr

Solve Problem B-4-4 by using the bilinear transforrnation coupled with the Wouth
stability criterion.

the regions in the s plane shown in Figures 4-65(a) and (b). Draw t
corresponding regions in the z plane. The sampling period T is assumed to be 0.3 s Consider the system
(The sampling frequency is w, = 2v/T = 2 ~ 1 0 . 3= 20.9 radhec.)

S plane Suppose that the input sequence {x(k)) is bounded; that is,
= constant, k = 0,1,2,. ..
Show that, if al1 poles of G(z) lie inside the unit circle in the z plane, then the output
y(k) is also bounded; that is,
ly(k)l 5 M2 = constant, k = 0,1,2,. ..

State the conditions for stability, instability, and critical stability in terms of the weight-
ing sequence g(kT) of a linear time-invariant discrete-time control system,

Consider the digital control system shown in Figure 4-66. Plot the root loci as the gain
Kis varied from O to m. Determine the critical value of gain Kfor stability. The sampling
period is 0.1 sec, or T = 0.1. hat value of gain K will yield a damping ratio of the
closed-loop poles equal to OS? With gain M set to yield 2; = 0.5, determine the damped
natural frequency wd and the number of sarnples per cycle of damped sinusoidal
oscillation.

Figure 4-65 (a) Region in the s pla


bounded by constant w lines and
constant CT lines; (b) region in the s
plane bounded by constant 5 lines,
constant o lines, and a constant c

Figure 4-66 Digital control system for Problem B-4-8.


Gonsider the following characteristic equation:
z3 + 2 . 1 i-.
~ 1.442
~ + 0.32 = O
Referring to the digital control system shown in Figure 4-67, design a digital controller
Determine whether or not any of the roots of the characteristic equation lie ou
G D ( z )such that the damping ratio 2; of the dominant closed-loop poles is 0.5 and the
unit circle centered at the origin of the z plane.
number of sarnples per cycle of damped sinusoidal oscillation is 8. Assume that the
sampling period is 0.1 sec, or T = 0.1. Determine the static velocity error constant.
Determine the stability of the following discrete-time system: Also, determine the response of the designed system to a unit-step input.
Y(z)
- - - z-~
X(z) 1 + 0.52-' - 1 . 3 4 ~ --k~0 . 2 4 ~ ~ ~ Consider the control system shown in Figure 4-68. Design a suitable digital controller
hob8em B-4-4 that includes an integral control action. Lhe design specifications are that the damping
Consider the discrete-time closed-loop control system shown in Figure 4-13. Determln ratio of the dominant closed-loop poles be 0.5 and that there be at least eight samples
the range of gain K for stability by use of the Jury stability criterion. per cycle of damped sinusoidal oscillation. The sampling period is assumed to be 0.2
hap. 4 Prablems

Digital control system for Problem B-4-9. -70 Digital control system for Problem B-4-12.
Figure 4-67

Weferring to the system considered in Problem 8-4-9, redesign the digital controller
so that the static velocity error constant K , is 12 sec-', without appreciably changing
the locations of the dominant closed-loop poles in the z plane. The sampling period is
assumed to be 0.2 sec, or T = 0.2. After the controller is redesigned, obtain the
unit-step response and unit-ramp response of the digital control system.

Consider the digital control system shown in Figure 4-71. Draw a Bode diagram in the
w plane. Set the gain K so that the phase margin becomes equal to 50'.
Figure 4-68 Digital control system for Problem B-4-10. K so set, determine the gain margin and the static velocity error constant K V .The
sampling period is assumed to be 0.1 sec, or T = 0.1.
sec, or T = 0.2. After the digital controller is designed, determine the static ve1
error constant K v .

order and has a dead .time of 5 sec. By choosing a reasonable sampling period T, d
a digital PI controller such that the dominant closed-loop poles have a dampin
jof 0.5 and the number of samples per cycle of damped sinusoidal oscillation is 1
L L
Digital control system for Problem B-4-14.

Using t diagram approach in the w plane, design a digital controller for the
ign specifications are that the phase margin be
and the static velocity error constant KV be 20
sec-l. The sampiing period is assumed to be 0.1 sec, or T = 0.1. After the controller
is designed, calculate the number of samples per cycle of damped sinusoidal oscillation.

Design a digital proportional-plus-derivative controller for the plant whose tra


function is l/sZ,as shown in Figure 4-70. It is desired that the damping ratio 5 0
dominant dosed-loop poles be 0.5 and the undamped natural frequency be 4 rad
The sampling period is 0.1 sec, or T = 0.1. After the controller is designed,
the mmber of samples per cycle of damped sinusoidaf oscillation. Fignre 4-72 Digital control system for Problem B-4-15.
Consider. the digital control system shown in Figure 4-73. Using the Bode diag
approach in the w plane, design a digital controller such that the phase margin is
the gain margin is 12 dB or more, and the static velocity error constant is 5 sec-l.
sampling period is assumed to be 0.1 sec, or T = 0.1.

In Ghapters 3 and 4 we were concerned wieh conventional methods for the analysis

Figure 4-74 Digital control system for Problem B-4-17.

which are mostly time varying andlor nonlinear.


Consider the digital control system shown in Figure 4-75. Design a digital contr A modern control system may have many inputs and many outputs, and these
G D ( z such
) that the system output will exhibit a deadbeat response to a unit step i
(that is, the settling time will be the minimum possible and the steady-state error
be zero; also, the system output will not exhibit intersampling ripples after the set
time is reached). The sampling period T is assumed to be 1 sec, or T = 1.

the state space can be carried out for a


ec. 5-1 lntroduction

function such as the impulse function, ste function, or sinusoi


state-space methods enable the engineer to include initial condi For linear time-varying discrete-time systems, t e state equation an
his is a very convenient and useful feature that is not

ble, state vector, and state

Seate, The state of a dynamic syst is the smallest set of variables (called where
state variables) such that the knowledge hese variables at t = to, together with
letely determines the behavior of the x(k) = n-wector (state vectsr)
e knowledge of the input for t $: to, c
system for any time t 2 to. Note that the concept of state is by no y(k) = m-vector
%tis applicable to biological systems, eeonomic syste
u(k) = r-vector
G(k) = n x n matrix (state matrix)
maki H(k) = n x r matrix (input matrix)
system. Ef at least n variables xl,xs, . . . ,x, ar
atrix (output matrix)
behavior of a dynamic system (so that once the
state at t = to is spe D ( k ) = m x r matrix (direct transmission matrix)
then such n variabl
arance of the variable k in the arguments of m
Note that state variables need not be
implies that these matrices are time vas-ying. If
quantities. Variables that do not represenh
not appear explicitly in the matrices, they are
neither measurable
constant. That is, if the system is time invariant,
in choosing state variables is an advantage o
speaking, however, it is convenient to choos
state variables, if this is possible at all, becau w(k + 1) = Gx(k) + Hu(k) (5-1)
feedback of al1 state variables with suitable
Vee$or, H f n state variables are needed to com
As in the discrete-time case, continuous-time (linear or nonlinear) systems may
f a given system, then these n state variables can
be represented by the following state equation and output equation:
components of a vector x. Such a vector is called a state vector. A state vector Bs thus
a vector that determines uniquely the system state x(t) for any time t z to, once the
state at t = to is given and the input u(t) for t 2 to is specified.
aee. The n-dimensional space whose coordinate axes consist of the
For linear time-varying continuous-time syste
XI axis, x2 axis, . . . , x,, axis is called a staie space. Any state can be represented b
equation are given by
a point in the state space.
ons, %nstate-space analysis we are concerned wit
tYpes are involved in the m ling of dynamic systems: 1
variables, output variables, and state variables. If tbe system is time invariant, then the last ~ W Oe u a t i ~ n sare simplified to
state-space representation for a given systern is not unique, except that the numb
of state variables is the same for any of the different state-space representations
the same system.
For time-varying (linear or nonlinear) discrete-time systems, the state e q d
may be written as Figure 5-l(a) shows the block diagram representation of the discrete-ti
system defined by Equations (5-1) and (5-2), and Figure 5-l(b) shows the contin-
x(k + 1) = f [x(k), un(k), k] UOUS-time control system defined by Equations (5-3) (5-4). Notice that the basic
and the output equation as coarfigurations of the diserete-time and continusus- systems are the same.
iscrefre-Time Systems

(a) Consider the discrete-time system described by


y(k) + aly(k - 1) + aly(k - 2) + + any(k - n)
=bou(k)+blu(k-1)+--.+bnu(k-n) (5-5)

Y(z) bo + bl z-l
-=:
+ . - + b, z-"
~ ( z ) 1 + alz-' + . - - + anz-" (5-6)

Y(z)
-=
bozn + bl zn-l + a . + bn
U(Z) zn + al zn-1 + - . + a, (5-7)

(b)

following representations:

Controllable canonical form


. Observable canonical form
. Diagonal canonical form
. Jordan canonical form

hod. (See Problem A-5-1.) The observable canonical form can be obtained by
tions of linear time- -time systems. Section 5-3 first tr nested programming rnethod. (See Problem A-5-2.) The diagonal canonical
orm and Jordan canonical form may be obtained by use of the partial-fraction-
procedure and by the z transform approach. Then it presents a method for com
ing (zI - G)-'. Section 5-3 concludes with discussions of the solcltdon of the li
time-varying discrete-time state equation. Section 5-4 deals with the pulse tran
fmction matrix. Section 5-5 first treats the discretization of linear continasous-e
state-space equations. Then it discusses time response between two concecu by the fdowing eqimations:
State-Space Analysic ec. 5-2 State-Space epresentationc of Díscrete-Time Systems

Tbe state-space representation given by Equations (5-12) and (5-13) is called an


observable canonical form. [For the derivation of Equations (5-12) and (5-13), see
Problem A-5-2.1 Notice that the n X n state rnatrix of the state e
Equation (5-12) is the transpose of that of the state equation defi
uations (5-8) and (5-9) are the staf.e equation and output equation, (5-8).
The state-space representation given by Equations (5-8) and (5-9) Note that if we reverse the order of the state variables, tha
-- -
called a controllabl ical form. [For the derivatidn of Equations (5-8) a
(5-9), see Problem 1
Note that if w e the order of the state variables, that is, we define ne
state variables according to the fashion
then the state equation and the output equation becorne as follows:

then the state equation can be written as follows: i,-,(k + 1 )


-al -a2 -
-an Xl(k) 1 Xn(k 4- 1 )
1 O - . e O O i2(k) O
= O 1 . a . O O X3(k) + O

O O - - S O --~,ik)~ b~

The output equation can be given by


Xdk)
X2(k)
y(k)=[1 O S . - O O] f + bou(k)
i n - l(k)
in(k) .
kquations (5-10) and (5-11) are also in the controllable canonical form.
Equations (5-14) and (5-15) are also in the observable canonical form.

The state-space representation of the discrete


time system given by Equation (5-5), (5-6), or (5-7) may be put in the followin . If the poles of the pulse transfer function given by
form: Equation (5-5), (5-6), or (5-7) are al1 distinct, then the state-space representation
may be put in the diagonal canonical form as follaws:
tate-Space Analysis C epresentationc of Discrete-Time

r X,W i
ence,

[For the derivation of Equatlons (5-16) and (5-17), see


transfer function

are distinct, then the state equation and out

transformation.
Gonsider the system defined by

Let us define a new state vector X(k) by

is a nonsingular matrix. [Note that, since both x(k) an k(k) are n-dimen-
sional vectors, they are related to each other by
Then, by substituting Equation (5-22) into ation (5-2O), we obtain

Premultiplying both sides of Equation (5-23) by

et us define
The state-space representations in the controllable canonical form, observable can
ical form, and diagonal canonical form become as follows:
CONTROLLABLE CANONICAL FORM:
uation (5-24) can be written as follows:

imilarly, by substituting Equation (5-22) into E uation (5-21), we obtain

y defining

we can write Equation (5-26) as


r ,J

DIAGONAL CANONICAL FORM: have thus shown that the state-space representation given by Equations (5-20)
The given pulse trancfer hnction Y ( z ) I U ( z )can be expanded as follows: "nd (5-21),
n n n
V V C
W W W
ec. 5-3 Solving

From Equations (5-35) and (5-43) the state transition matrix

Therefoire, we flrst obtain ( z

resent the solution of a discrete-time state equation by t


Consider the discrete-time syste described by Equation (5-28):
x(k + 1) = G x ( k ) + Hu(k) (5-
Taking the z transforrn of both sides of Equation ( 5 - N ) , we get
z X ( z ) - zx(0) = G X ( z ) + HtJ(z)
where X ( Z ) = Z [ x ( k ) ]and U(z) = Z [ u ( k ) J Then
.
- G ) X ( z ) = zx(0) + HU(z)
Premultiplying both sides of this last equation by ( z - G)-', we obtain The state transition matrix ( k ) is now obtained as follows:
(5-
Eking; the inverse z transforrn of both sides of Esuation (5-41) gives

Comparing Equation (5-30) with Equation (5-42), we obtain


Gk = Z-'[(zI - 6)-lz ]
and
k- 1
b_=k-j-1
Equation (5-45) gives the state transition matrix.
Next, compute x(k). The z transform of x ( k ) is given by
where k = 1 , 2 , 3 , . . . .
Notice that the solution ( z ) = (zlj - G)-lzw(0) + ( z
inverting the matrix ( z - G), which rnay be accomplisfied by analiytical meam
by use of a computes routine. The solution also reauires the inverse z transfor
Since

Obtain the state transition matrix of the following discrete-time system:


we obtain z2
x(k + 1) = G x ( k ) + H u ( k )
y ( k ) = Cx(k) -z2 + 22
where z-1 z -1
Hence

Then obtain the state x ( k ) and the output y ( k ) when the input u ( k ) = 1 for k O
2 , . . . . Assume that the initial state is given by
olvirig Discrete-Time S

Thus, the state vector x(k) is given by

(The trace of an n x n matrix is t


Finally, the output y(k) is obtained as follows: For a higher-order determina
into the form given by Equation (5
Equation (5-50) lo compute the si's

ent method is convenient for computer

Determine the inverse of the matrix (zI - G), where


0.1 0.1 o

Also, obtain Gk.


[
G = 0.3 -0.1
O O
-0.2
-0.3

From Equation (5-46), we have

Although the determinant / z - 6 1 can be expanded easily, here for demonstration


purposes let us use Equation (5-50) to compute a l , a2, and a3. First, notice that

a, = -trG = -tr

Then, from Equation (5-49), we obtain


[Y a -1.J
0.3 -0.1 -0.2 = 0.3

where

Hence

W,
=

=
GWn-2
GH,-,
+ a,-i
-i- a, I=
a- = -!tr GHl = -+ ([H:: O. 1
--o1 -:.2][0.3
-0.3
0.4 0.1
O ii.2 -~2]

Note that a,, a2, . . . a, are the coefr'icients appearing in the determinant &'e "03 -1.021
by Equation (5-47). The si's can also be given (see Problern A-5-13) by use of th = -4 tr 0.09 0.01 0.02 = -0.04
trace, as follows: [:.o7
ee. 5-3 Solving Diserete-Time State-Space Equations

m$. Consider the

uation (5-52) may be found easily by recursion, as follows:


x(h + 1) = G(h)x(h)+ H(h)aa(h)
x(h + 2 ) = G(h + I)x(h + 1) + A(h + l ) u ( h + 1)
= G(h + f ) G ( h ) x ( h )+ G(h + l ) H ( h ) u ( h )+

Eet us define the state


defined by Equation (5-52) a

e k = h , h + 1,h + 2 , . . . . t can be seen that t e state transition


,h ) is given by the eqisation

ution of Equation (5-52) becomes

k j +l ) ( j ) ( ) k >h (5-55)
j= h

Notice that the first term on the right-hand side of Equation (5-55) is the contribu-
tion o%the initial state x(h) to the current state x ( k ) and that the second term is the
) , + l ) , . . . ,u(k - 1).
contribution of the input ~ ( h u(h
Equation (5-55) can be verified easily. Referring to Equation (5-54), we have

If we substitute Equation (5-56) into


k

j= h
me-Space Analycic Ch ec. 5-4 Pulse-Transfer-Function Matrix

where x(k) is an n-vector, u(k) is an r-vector, (k) is an m-vector, G is an n x n


matrix, H is an n x r matrix, is an m x n matrix, and D is an m x r
j= h
the z transforrns of Equations (5-58) and (5-59), we obtain
= G(k)x(k) + H(k)u(k)
Thus, we have shown that Equation (5-55) is the solution of Equation (5-52).
Once we get the soiution x(k), the output equation, Equation (5-53), becom
as follows: Pdoting that the definition of the pulse transfer function calls for the assumption of
zero initial conditions, here & i d state x(O) is zero. Then
(k, J + I)H(j)n( j) + D(k)u(k), k > we obtain
If @(k)is nonsingular ecf, so that the inverr;e of q ( k ,
exists, then the inverse of (h, k ) , is given as follows:

where
= [G(k - I)G(k - 2) e
G(h)]-l
= ~ - l ( h ) ~ - l (ih 1) e -~ - y -
k 1)
(z) is called the pulse- nsfer-function matrix .
. A sumrnary on the state transition matrix transfer function rnatrix z) characterizes the in

Since the inverse of matrix (z - G) can be written as

G(k - 1 ) q k - 2) . G(h), k >h

ulse-transfer-function matrix (z) can be given by the equation

f G(k) 1s nonsingular for al1 k values considered, then


i ) foranyi,j,k Clearly, the poles of (z) are the zeros of lz
f G ( k ) is singular for any value of k , then cbaracteristic equation of the dáscrete-time syst
(j,), fork>j>i
or
zn + alzn-l + a2zn-2+ . - + an-lz + a, = O
where the coefficients ai depend on the elements of G.
A single-input-single-output discrete-time system may be modeled by a pulse tran have shown that for the system defined by
fer function. Extension of the pulse-transfer-function concept to a multiple x(k + 1) = Gx(k) + Hu(k)
multiple-output discrete-time system gives us the pulse-transfer-function ma
this section we shall investigate the relationship between state-space represen
and representation by the pulse-transfer-function matrix. the pulse-transfer-function matrix is
Pulse-Transfgr-Functa'o~ The state-space representation of an
order linear time-invariant discrete-time system with r inputs and m outputs can
given by ]in Section 5-2 we sh representations for a
&en system are rela defining a new state
x(k + 1) = Gw(k) + Hu(k) vector %(k)by using a si
y ( k ) = Cw(k) + Dn(k) x(k) = P%(k)
iscretimatiom of Continuocas-Time tate-Space Equations

ecause of the convergence of t Ak tklk!, the series can


be differentiated term by ter
d
- - t A3 t 2 + ... + ----
- A + ~ 2 +- ~k t k - l
+ ...
dt 2! ( k - l)!
, fi are related, res

~=iej,
= t
The pulse-transfer-function matrix F(z) for the system
and (5-62) is is can be proved as follows:

In particular, if s = - t , then
eAte-At = e-At eAt = eA(t-t) =

inverse of eA'ise-"'. Since the inverse af e"' always exists, e*' is nonsingular.
That is, itdoes not depend on the particular state vector x ( k ) c s important to point out that
representation.
The characteristic equation lz
transformation, since
shall next obtain the solutiorr of the continuous-time state e
Thus, ihe eigenvalues of G are invariant under similarity transformation.
where x is the state ctor (n-vector), ia the input vector (r-vector),

In digital control of continuous-time plants, we need to convert and premultiplying both sides of this last equation by e-"', we obtain
state-space equations into discrete-time state-space equations,
can be done by introducing fictitious samplers and fictitious hol
continuous-time systems. ?'he error i
negligible by using a sufficiently small sa ntegrating the preceding equation between O and t gives
time constant of the system.
e-*tx(t) = X(O)+ /'e-*'
iscretization of Continuous-Time State-

Equation (5-64) is the solution of Equation (5-63). Note that t


state equation starting with t e initial state x(tO)is

where h = T - t .
shall present a procedure for discr
assume that the input vector n(t)
Note that the sampling operation

Consider the continuous-ti then Equation (5-72) becomes


x((k + 1 ) T ) = G ( T ) x ( k T )i- (5-75)

Hn the following analysis, esentation, we use the notation kT a


( k + 1)T instead of k an
(5-66) will take the forrn
x((k + 1 ) T ) = G ( T ) x ( k T )+ W(T)u(kT) trices C and D are constant atrices and do not depend on the sampling

Note that the matrices G and H depend on the sampling perio atrix A is nonsingular, t T ) given by Equation (5-74) can be simpli-
are constant matrices. fied to
To determine G ( T ) and

so that all the components of


consecutive sampling instants, or
u(t) = n ( k T ) , for k T 5 t < kT +T
In the state-space approach, notice that by assuming the input vector a ( t ) to
be constant between any two consecutive sampling instants, the discrete-ti
model can be obtained simply by integrating the continuous-time state equa-
tion over one sampling period. The discrete-time state equation given by
and Equation (5-68) is called the zero-order hold equivalent of the continuous-time
state equation given by Equation (5-66).
n general, in converting the continuous-time system equationr into a discrete-
"O
ime system equation, some sort of approximation is necessary. It is important
multiplying Equation (5-71) by eATand subtracting it from Equation (5-70) gives to point out that Equation (5-75) involves no approximation, provided the
input vector u(t) is constant between any two consecutive sampling instants,
[ ( k + 1 ) 7 e-*T
x((k + 1 ) T ) = eATx(kT)+ eA(k+l)T
, , , , \ / -
as assumed in the derivation.
JkT
Notice that for T << 1, G ( T ) = G(O) = eAO= . Thus, as the sampling period
%nce from Equation (5-69) ~ ( t=) u ( k T ) for kT i t < kT 3- T , we may substitut T becomes very small, G ( T ) approaches the identity matrix.
U(T) = u ( k T ) = constant in this last equation. [Note that u(t) mai jump
t = kT + T and thus u(kT + T ) rnay be different from u(kT). Such a jump in u(
at r = kl' + T , the upper Iirnit of integration, does not affect the value of the inWP Consider the continuous-time system given by
in
"*
this last equation, because the integrand does not involve impulse function
nence, we rnay w i t e
iscretization of Continuaus-Time State-Space E

Obtain the continuous-time state-space representation of the system. Then discre


the state equation and output equation and obtain the discrete-time state-space reDr
sentation of the system. Also, obtain the pulse transfer function for the syst&n by u
Equation (5-60).
The continuous-time state-space representation of the system is simply
L
Thus, r 1

Now we discretize the state equation and the output equation. Referring to Eqjuatio
(5-73) and (5-74), we have
G(T) = e--aT I h e output equation becomes - ..
rT I ..-aT
H(T)= jo =
a'
Hence, the discretized version of the system equations is hen the sampiing period is 1sec, or T = 1, the discrete-time state equation and the
output equation become, respectively,
1 - e-aT
x(k + 1) = e-OTx(k) + -
a u@)

Y (k) = x(k) and


Referring to Equation (5-60), the pulse transfer function for this system is
- ...

The pulse-transfer-function reprecentation of this system can be obtained from Equa-


\- - /
a a ( l - e-aTz-") tion (5-60), as follows:
This result, of course, agrees with the z transform of C(s) where it is preced
by a sampler and zero-order hold [that is, where the signal u(t) is sampled and fed
a zero-order hold before being applied to G(s)]:

Eea
Obtaim the dascrete-time state and output equations and the pulse transfer hnct
(when the sarnpling period T = 1) of the following continuous-time svstern:

Note that the same pulse transfer function can be obtained by taking the z
which may be represented in state space by the equations transform of C(s) when it is preceded by a sampler and zero-order hold. Assuming
T = 1, we obtain
- - F -.

The desired discrete-time state equation will have the form


x((k + 1)T) = G(T)x(kT) + H(T)u(kT)
where matrices G(T) and B(T) are obtained from Equations (5-73) and (5-74)
6- 1 2 .
iscretization of Continuous-Time

equation

into

As another example, consi es the following system:

then, assurning the sampling period to be 0.05 sec, we obtain G and H as E


Assuming that the sampling period T is 0.05 sec and wit out specifying the forrnat,
we get the following discrete-time state equation:
x(k + 1) = G x ( k ) +
where matrices G and H can be found in the following computer out

A = [O 1 0 0
20.601 O O O
o 0 0 1
-0.4905 O O O];
B = [O;-1;0;0.5];
[G,H] = c2d(AfB,0.05)

space equation
x(k + 1) = G x ( k ) + Wu(k)
ec. 5-6 Liapunov Stability Analycis li

Consider the system discussed in Example 5-5. Obtain the discrete-time state equation
4, the z transform solution of the discrete-time sy and the output equation at t = kT + AT. Also, obtain the specific expressions for the
state equation and output equation when 2" = 1 sec and
lin Example 5-5, the matrices G ( T ) and H(T) were obtained as follows:

shall show that th

Consider the time-invariant continuous-time system defined by

To obtain the state equation and the output equation at t = kT + da", where
O < AT < T , we first convert G ( T ) to G(Air) and H(T) to H(AT) and then substitute
Eet us assume that the input u is sampled and fed to a zero-order hold G(ibT) and M(dT) into Equations (5-80) and (5-811, as follows:
u ( ~ )= u ( k T )for kT T < k T + T. Referring to Equation (5-651, the solu
the state equation starting with the irnitial state x(to) is

1 '(1 -
)
e-''' 1
EOr T = 1 and d T = 0.5 we obtain the state equation and output equation as follows:

[ E ] [ :::(7:][:rl:l]
= + [u:~P~Y]u(~)

y(k + O S ) = 11 0.31hlj[:~:i] + (00920)u(l)


G(AT) = eAAT

Then we obtain
n important role in &hestability analysis of control
equations. There are two methods o£ stability
e first method and the second method; both apply
mination of the stability of dynamic systems described by ordinary
. The first method consists entirely of procedures
tions of the differential equations or difference
. The second rnethod, on the other h a n 4 does
ential or difference equations. This is the reason
second method is so useful in practice.
Although there are many powerful stability criteria available for control sys-
s, such as the Jury stability criterion and the Routh-Hurwitz stability criteria,
puter calcu'tation.) Y are limited to linear time-invariant systerns. The second method of Liaprtnzov,
ec. 5-6 Liapunov Stability Analysis

on the other hand, is not limited to linear time-invariant systems:


both linear and nonlinear systems, time invariant or time varying. In this example, we give severa1 scalar functions and their classifications according to
find that the second method of Liapunov is indispensable for the the foregoing definitions. Here we assume x to be a two-dimensional vector.
of nonlinear systems for which exact solutions may be undtainable. (It is cautlone
however, that although the second method of Liapunov is applicable to any n + x:
1. V ( s ) = x: positive definite
x;
linear system, obtaining successful results may not be an easy task.
imagination may be necessary to carry out the stabifity analysis o£
. V(x) = x: + ----
1 + xi positive definite
systems.) 3. V(x) = (xi + x2)2 positive sernidefinite
The second method of Liapunov is also called t . V(x) = -x: - (xl+ x2)' negative definite
the classical theory of V(x) = x1x2 + xi indefinite
that 1 energy is continuall ons. The Liapunov function, a scalar function, is a positive
equilibrium state is reached.
The second method of Liapunov is based on a generalization of t a time derivative
system has an asymptotically stable equilibrium state, tkien the stored
system displaced within a domain of attraction decays with increasin
finally assumes its minimum value at the equilibrium state. For purely mat nctions do not include t
systems, however, there is no simple way of defining an "' explicitly, then we denote them by V(xl, ~ 2.,. . ,x ~ ) or
, V(x).
circumvent this difficulty, Liapunov introduced the Liapunov fu Notice that ~ ( xt ), as ac of V ( x ,t) with respect to t
energy function. Shis idea is, however, more general than that of en along a solution of the system. s that V(x, t) is a decreasing
widely applicable. In fact, any scaiar function satisfying the hypothe function of t. A Liapunov fuai en systern. ( k r this reason,
orems (see Theorems 5-1 through 5-6) can serve as a Lia the second method of Liapunov is a more powerful tool than conventional energy
efore we discuss the Liapunov function further, it is necessary to defin ons. Note that a system whose energy E decreases on the average but not
positive definiteness of scalar functions. at each instant is stable, but that E is not a Liapunov function.)
er in this section we shall show that in the second method of Eiapunov the
ons. A scalar function V(A)is said
positive definite in a region f2(which includes the or vior of V(x, t) and that of its time derivative ~ ( xt), = d V ( x ,t)/dt give
e state space) if v ( ~
for al1 nonzero states x in the region 92 and if V ( information about the stability of an equilibrium state without having t
A time-varyirig function V(X,t) is said to b Note that the simplest positive definite function is of a
ve definite in a regi
(which includes the origin of the state space) if it is bounded from below
time-invariant positive definite function, that is, if there exists a positive de
function V(x) such that
In general, Liapunov functions may not be of a simple quadratic form. For any
V(x, t) > V(x), for a l t 2 t, Liapunov function, however, the lowest-degree terms in V must be even. This can
for al1 t 2 to be seen as follows. If we define

negat
then in the neighborhood of the origin the Iowest-degree terms done will become
to be dominant and we can write V(x) as
origin and at certain other states, where it is zero.
sns. A scalar function V(w) is
If we keep the i,'S fixed, V(X1, X2, . . . ,in-l,
1) is a fixed quantity. For p odd, x[ can
nite if - V(x) is positive semidefinite.
assume both positive and negative values near the origin, which means that V(x) is
A scalar function V(x) is said to be i not positive definite. Hence, p must be even.
ositive and negative values, no matler In what lollows, we give definitions of a syste , an equilibrium state, stability,
s~nallthe region $2 is. asymptotic §tal-iility, and instability.
pace Analysis Chap. ec. 5-6 Liapunov Stability Analysis

An equilibrium state xeof the syste


is said to be asyrnptotically stable if it is stable in t
solution starting within S ( 6 ) converges, without leaving S ( € ) , to w, as t increases
where x is a state vector (an n-vecto indefinitely.
functions of xl, x2, . . ,x,, and t . (Note that w In practice, asymptotic stability is more im ortant than mere stability. Also,
since asymptotic stability is a local conce simply to establish as
a model to present basic materials on stabilit
oes not necessarily mean that the syste
largest region o£ asy
e dornain of attractio
ble trajectories originate
nating in the domain of attraction is asym
totic stability holds for all states (al1
from which trajectories originate, t e equilibriurn state is
said to be asymptotically stable in the large. That e equilibrium state x, of the
system given by Equation (5-82) is said to be as tically stable In the large if
it is stable and if every solution converges to x, as ses indefinitely. Obviously,
a necessary condition Eor asymptotic stability in e is that there be only one
uilibrium state in the whole state space.
In control engineering problems, asymptotic stability in t
feature. -hf the equilibrium state is not asymptotically stable in the large, then the
problem becomes one of determináng the largest region of asymptotic stabiiity. This
is usually very difficult. For al1 practica1 purposes, however, it is sufficient to deter-
mine a region of asyrnptotic stability large enoragh that no disturbance will exceed it.
n equilibrium state x, is said to be unstable if for sorne real
number E > O and any real number 6 > O, no atter how small, there is always a
state xo in S(6) such that the trajectory starting at this state leaves S(€).

Ilx - xell 5 r graphical representation of the foregoing definitions will clarify their meanings.
where llx - x,ll is called the Euclidean norrn and is defined as follows: Let us consider the two-dimensional case. Figures 5-2(a), (b), and (c) show
equilibrium states and typical trajectories corresponding to stability, asymptotic
I/x - = [ ( x ~- le)' + ( ~ 2- ~ 2 , +) ~ . + (xn - x,,)~]~"
e

Let S(6) consist of al1 points such that


Ilx, - xell 5 6
and let S ( € ) consist of al1 points such that

An equilibrium state x, of the system of Equation (5-82) is said to be s


sense of Eiapunov if, corresponding to each S ( € ) ,there is an S(6) such t
tories starting in S(6) do not leave S ( € ) as t increases indefinitely. The re
6 depends on E and, in general, also depends on h. If 6 does not depen
equilibriurn state is said to be uniformly stable.
Wlat we have stated here is that Figure 5-2 (a) Stable equilibrium state and a representative trajectory; (b) asymp-
$ ( E ) , there must be a region S ( 8 ) such totically stable equilibrium state and a representative trajectory; (c) unstable equi-
S ( E ) as t iilcreases indefinitely. librium ctate and a representative trajectory.
ec. 5-6 Liapunov Stability Analysis

. V(x, t) is positive definite.


Q(x, t) is negative definite.

The conditions of this theorem may be modified as follows:

entically in t 2 to for any to and any wo =#


lution starting from xo at t = to.

ly stable in the large.

move toward the origin.


To prove stability (but not asymptotic sta-
V(x) = C biliay) of the origin of the systern defined by Equation (5-&2), the following theorem
may be applied.

. Suppose a system is described by

for al1 t. If there exists a scalar function VCx,t) having continuoras


atives and satisfying the conditions

V(x, t) is positive definite.


. v(x, t) is negative semidefinite.
then the equilibrium state at the origin is uniformly stable.
It should be noted that the negative semidefiniteness of V(x, t) [ ~ ( xt), 5 O
along the trajectories] means that the origin is uniformly stable but not necessarily
~niformlyasymptotically stable. Hence, in this case t e system may exhibit a limit
where cycle operation.

. If an equilibriurn state x = of a system is unstable, then there


Tf there exists a sealar fimction V(x, t ) having continuous first partial derivatives function W(x, t) tkat determin s the instability of the eqluilibrium
satisfying the conditions state. We shall present a theorem on instability in t
tate-Space Analycic ec. 5-6 Liapunov

w = f(x, t )
where

systems, is algebraic and does not require factoring of the characteristic polynomial,
e-invariant systems the
W(x, t ) is positive definite in the same region. gives not just sufficient conditions, but t
stability or asymptotic stabiiity.
ty analysis o£linear time-invariant systems, it is aspmed
arks. A few comments are in order when the Liapunov st that if an eigenvalue A, o£ matrix A is a complex quantity then A must have A,, the
complex conjugate of A,, as its eigenvalue. Thus, any complex eigenvalues of A wili
appear as conjugate complex pairs. Also, in the following discussions on stability,
we shall use the conjugate transpose expression, rather than the transpose expres-
sion, of matrix A, since the elernents of rnatrix A may include complex conjugates.
tions but are not necessary conditions. The conjugate transpsse of t is a conjugate of the transpose:

Consider the following linear time-invariant system:

includes this equilibrium state, it does not necessarily mean that the where x is a state vector (an n-vector) and A is an n x n constant matrix.
are unstable outside the region 0. is nonsingular. Then the only equilibrium state is the origin,
For a stable or asyrnptoticauy stable equilibrium state, a Liapunov fun stability of the equilibrium state of the linear time-invariant system can be investi-
with the required properties always exists. gated easily with tke second method of Liapunov.
For the system defined by Equation (5-84), let us choose as a possible Liapunov
function
proaches to the investigation of the asyrnptotic stability of linear time-i
systems. For exarnple, for a continuous-time systern described by the equ
K = Ax is a positive definite f x is a real vector, then
chosen to be a positive definite real syrnmetric matrix.) The time derivative of V(x)
it can be stated that a necessary and sufficient condition for the asymptotic along any trajectory is
of the origin of the system is that al1 eigenvalues of A have negative real p
that the zeros of the characteristic polynomial

have negative real parts.


Sirnilarly, for a discrete-time system represented by the equation
x(k + 1) = Gx(k) Since V(x) was chosen to be positive definite, we require, for asyrnptotic stability,
that p(x) be negative definite. Therefore, we require that
.%te-%pace Analysis Chap. 5-6 Liapunov Stability Analysis

ong any trajectory, then


be chosen to be positive semidefinite.
) = posltfve definite
If an arbitrary positive definite matrix is chosen for
e asymptotic stability of the system of E idefinite matrix if ~ ( xdoes
) not vanish identical
the matrix equation
For a test of positive definiteness of an n x n rnatrlx, we apply Sylvester
criterion, which states th essary and sirfficient con
rminants of all tke successive pri
r example, the following n x n
are al1real, then the Hermitianm matrix beco

r- -7
'
p11 Pl2

Fin p2n a Pnn

where pi, denotes the complex conjugate of pij. as a root of the chasacteristic equarion, and if for every sum of two roots
a13 the successive principal rninors are positive,
Pil Pi2 - then the elements of ined. (Note that for a stable matrix
F12 p22
... ... the sum Ai + hk is always nonzero.)
e . .
" '
9

In determining whether or not t ere exists a positive definite Hermitian or


positive definite real sym , it is csnvenient to choose
tead of first specifying a positive definite matri is the identity matrix. Then tke ele are determined from
is positive definite, it is c
then examine whether
is tested for positive definiteness.
e definite. Note that po
shall strrnrnarize what Determine the stabiiity of the equilibriurn state of the following system:
Consider the system described by
x = AA
where x is a state vector (an n-vector) and A is an n x re co The system has only one equilibrium state at the orágin.
necessary and sufficient condition for the equilibrium state into Equation (5-851, we have
totically stable in the Iarge is that, giv any positive definite
positive definite real symmetric) matrix
(or a positive definite real symmetric) rnatrix Noting that A is a real matrix, must be a real symmetric matrix. This last equation
may then be written as follows:

The scalar function x*Px is a Liapunov function for this system. [Note that in
r systern considered, if the equilibrium state (the origin) is asymptoticaliy stabl
it is asyrnptotically stable in the large.]
where we have noted thatpzl = p12and made the appropriate substitution. If the matrix
wks. In applying Theorem 5-4 to the siability analysis of linear tirn B turns out to be positive definite, then x*Bw is a Eiapunov function and the origin is
~tinerous-timesysterns, several important rernarks rnay be rnade. asymptotically stable.
tate-Space Analysis ec. 5-6 Liapunov Stability Analysis

Equation (5-86) yields the following three equations:


-2p11 + 2p12 = -1
- 2 ~ 1 1- $12 + pzz = 0 e equilibrium state x = totically stable in the large and V(x) is a
-4~12- 8p22 = -I kiapunov function.
Solving for the p y s ,we obtain Note that in this theorem c 2 may be replaced by
p1i = g, p12 = -&, pzz = &$ V(x) 5 O for all x, and V(x) does not vanish i entically for any soltution
Hence, sequence {x(kT)} satisfying Equatlon (4-88).

V(x) need not be negative definite if it oes not vanish identicalliy


on any solution sequence o£ the differesrce equation.
By Sylvester's criterion, this matrix is positive definite: Hence, we condude that t
origin of the system is asymptotically stable in the large.
It is noted that a Liapunov function for thic system is Gonslder the discrete-time system described by

where w is a state vector (an n-vector) and G is an n X n constant nonsingular rnatrix.


= & (23~:- 14x1x2 + 11x2) is the equilibrium state.
and p(x) is given by by use of the second method of Liapunov.
Ti(,) = -x; - x,2 ket us choose as a possible Liapunov ftunction

positive definite Herrnitian (or a positive definite real sy


time systems. As in the case of continuous-time S
V(x(k)) = W(x(k + 1)) - V(x(k))
te-time systems based on the second met

V(x(h-7')) = V(x(k + 1)T) - V(x(kT))


Consider the discrete-time system
Since V(x(k)) is chosen to be positive clefinite, we require, for asymptotic stability,
at AV(x(k)) be negative definite. Therefore,
where
x = n-vector
where
) = positive definite
L = sampling period
totic stability of the discrete-ti e system of Equation (5-89),
Suppose there exists a scalar function V(x) contlnuous in x such that
systems, it is convenient to specify first
1. V(X) > O for x J. O. definite real symmetric) matrix Q and
2, AV(x) < O for x f O, where then to see avhether or not the P rnatrix deterrnined from
A V ( x ( k T ) ) = ~ ( x ( k+ 1)T) - V(x(kT)) = V ( f ( x ( k ~ ) ) )- ~ ( r ( k T ) ) G*PG - E" = -Q
ec. 5-6 Liapunov

l l ~ l l= 0, for x =

w(k + 1) = Gx(k)
A function f (w) is said to be a contraction if
A necessary and sufficient condition for t
ically stable in the large is that, given any
for some set of values of

Consfder the folowing di

ere x is an n-vector and (x) is also an n-vector. Assurne that (x) is a contraction
Then the origin of the system of Equation (5-91) is
asymptotically stabie in the large, and one of its LPa
V(x) = 11d1
stability of an equilibrium state of a discrete-time system obtained by
This can be seen as follows. Since V(x) = llxll is ositdve definite and
continuous-time systern is equivalent to that of the original continuous
Consider a continuous-time system
k = Ax is negative definite beca is a contraction for al1 w, we fin
and the corresponding discrete-time system Liapunov function, and rem 5-5 the origin of the system
stable in the large. (See
x((k + 1)T) = Gx(kT)
where
G =
Consider the following system:
Ef the continuous-time system is asymptotically stable, that is, if all the eigenv
of the matrix A have negative real parts, then
llGnll-+O, as n m
-+
Determine the stability of the origin of the system.
and the discretized system is also asymptotically stable. This is becau . Then, referring to Equation (5-90), the Liapunov
are the eigenvalues of A then the e" 's are the eigenvalues of 6. (Note t stability equation becomes
if hiT is negative.)
Ht should be noted here that, if a continuous
is discretized, then in certain exceptional cases hi
is found to be positive definite, then the origin x = is asymptotically stable
ing on the choice of the sampling period T. That i
in the large.
time system is not asymptotically stable, the equivalent discretized system From Equation (5-92) we obtain the following three equations:
to be asyrnptotically stable if we look at the v a h
instants. This phenomenon occurs only at cert
If the value of Tis varied, then such hidden instability shows up as
See Problem A-5-15.
p,, - 2p12 = -1
ConéPac~on. A norrn of ra denoted by I/x/lmay be though from which we get
the Iength of the vector. There are severa1 different definitions of
however, has the foilowing properties:
pace Analysis Chap. hap. 5 Example Problems and

Consequectly, ntáasn The given system can be modified to

= [; $1
By applying Sylvester's criterion for the positive definiteness of matrix
positive definite. Hence, the equilibrium state, the origin x =
. . -
in the large.
This last equation can be written as follows:
Note that instead of choosing to be I we could choose
semidefinite rnatrix, such as
(61 - al b ~ ) z - 1+ (b2 - a2b o ) ~ +
-~ + (b, - a, bo)z-"
.+
1 + alz-l + a2z-2 + + a,z-" U(z>
Let us define
(bl - al bo)zll + (b2 - a2b o ) ~ -+
2 + (b, - a, bo)z-"
AV(x) = -x:(k) Y(Z) =
1 + alz-' + a2zW2+ e + u,,z-~ U(z>

Then Equation (5-95) becomes

ket us rewrite Equation (5-96) in the following form:


O -0.5
[l -1 1;: : ] [ -:] - [;: Y;] -[::]
= (bi - al bo)z-'
~ ( 4
+ (b2 - a2bo)z-' + + (b, - a, bo)z-,
By solving this last equation, we obtaln

From this last equation the following two equations may be obtained:
Q ( ~=) -al;-' Q ( z ) - a2z-2Q ( z ) - - - unz-" Q ( z ) + U ( z )
and
F ( z ) = (bl - al b0)z-' Q ( z ) + (b2 - a2 bo)z-' Q ( z ) + ..
+ (b, - a, bo)z-" Q ( z )
Now we define the state variables as follows:
(Direct programming method) Gonsider the discrete-time system defined by
Y-
- ( z ) - bozn + b, zn-' + + bn
U ( Z ) Z" + alzn-1 + . - + a,
Show that a state-space representation of this system may be given by
Xn-'(2) = z -Q
~ (z)
X n ( z ) = 2-l Q ( z )
Then clearly we have
z X * ( z )= X2(z)
zXz(z) = &(z)
State-Space Analysis C Chap. 5 Exarnpie Problerns and Sslutions

In terms of difference equations, the preceding n - 1 equations become aakiora Rewrite the pulse transfer Eianction as follows:
x l ( k t 1) = x,(k) Y ( z ) - bo U ( z ) + zw1[alY ( z ) - bl U ( z ) ]
xz(k + 1) = x 4 k ) + z-'[a, Y ( z ) - bZ U ( z ) ] + - + ~ - ~ [Ya( z, ) - b, U ( z ) ]= O
or
~ , - ~ (+k 1) = x,(k) Y ( z ) = boU(z) + z-'(b1 U ( z ) - al Y ( z ) + z-'{b2 U ( z ) - a, Y ( z )
By substituting Equation (5-100) into Equation (5-98), we obtain + Z-'[b3 ~ ( z-)a 3 ~ ( z+) . -11) (5-106)
zX,(z) = - a l X n ( z ) - a 2 X n - - l ( z-
) . - a n X 1 ( z )+ U ( z ) Now define the state variables as follows:
which may be transformed into a difference equation: X,(Z) = z-'Lbl U ( z ) - al Y ( z ) + X n - l ( z ) ]
x n ( k + 1 ) = - a n x l ( k ) - a n - l x 2 ( k ) - ~ ~ ~ - a l x n ( k ) + ~ ( k(5) Xn-,(z) = z-l[bZU ( Z )- a, Y ( z ) + X n V 2 ( z ) ]
'
Also, Equation (5-99) can be rewritten as follows: (5-107)
Y(Z) = (b1 - al bo)X,(z) + (bi - a,bo)Xn--l(z)+ . . + ( b , - c. bo)X1(r) &(z) = ~ - l [ b , -U~( z ) - Y ( z )+ Xi(z)]
By use of this last equation, Equation (5-97) can be written in the form X l ( z ) = z-'[b, U ( z ) - a, Y ( z ) ]
y ( k ) = (b, - a, bo)xi(k) + (bn-i - a,-, bo)xz(k) Then Equation (5-106) can be written in the form
+ ... +(bi-aibo)x,(k)+bou(k) (5 Y ( z ) = boU(z) + & ( z ) (5-108)
Combining Equations (5-101) and (5-102) results in the state equation giv By substituting Equation (5-108) into Equation (5-107) and multiplying both sides of
Equation (5-93). The output equation, Equation (5-103), can be rewritten in the the equations by z , we obtain
given by Equation (5-94).
~ X n f z =) X n - ~ ( z )- a l x n ( z )+ (bi - al bo)U(Z)
z X n - i ( z ) = X , - ~ ( Z )- a 2 X n ( z )+ (b2 - a2bo)U(z)

(Nested programming method) Consider the pulse transfer function system defin
z X z ( z ) = X ~ ( Z-) a,-, X n ( z ) + (b,-1 - anTlb o ) U ( z )
qZ) Y ( z ) bo + blz-' + - . + b,z-"
= -- =
U(z) l + a l z - ' + . - ~ + a n z - " z X 1 ( z ) = -a, X,(z) + ( b , - a, bo)U ( z )
Show that a state-cpace representation of this system may be given as ~ L ~ o w s : Taking the inverse z transforms of the preceding n equations and writing the resulting
- 9 -z equations in the reverse order, we obtain
x l ( k t 1) O O O O -a, xl(k)
xz(k+l) 1 O
a - .

O O xz(k)
xi(k + 1) = -a, x,(k) + (b, - a, bo)u( k )
- ... ... ..
V . -

..
,-
xz(k + 1) = x l ( k ) - a n - l x n ( k )-t- (b,-l - a,-I b o ) u ( k )
~ , - ~ (+k 1) O O + - - 1 O -a2 ~,-~(k)
xn(k+l) O O O 1 -al x,(k)
- m
- -, ~ , - ~ (+k 1) = x,-,(k) - a 2 x n ( k )+ (6, - a2b o ) u ( k )
b, - a, bo x,(k + 1) = ~ , - ~ ( k-)a l x n ( k )+ (bl - al bo)u(k)
bn-i - 6,-1 bo
+ Also, the inverse z transform of Equation (5-108) yields
b2 - a, bo
y ( k ) = x, ( k ) + bo ~ ( k )
bi - al bo
P
Rewriting the state equation and the output equation in the standard vector-matrix form
x1 ( k ) gives Equations (5-104) and (5-105), irespectively.
x2W
y ( k ) = [O O O 11 !
x,- 1 ( k ) (Partial-fraction-expansion programming method) Consider the pulse transfer func-
- xn(k) tion system given by
State-Spaw Analycis Ch Chap. 5 Exarnple Problerns a n d

b o z n + bi zn-' + - - - + bn
-Y=( z )
Also, Equation (5-113) can be written as
U ( Z ) zn + al zn-' + . . . + a, Y ( z ) =: bo U ( Z )+ C ~ X , ( Z+) Q X ~ ( Z+) + cnXn(z) (5-116)
Show that the state equation and output equation can be given in the following The inverse z transforms of Equations (5-115) and (5-116) become
canonical form if al1 poles are distinct.
xl(k + 1) = p,xl(k) + ~ ( k )
xi(k + 1) 0 .".
x2(k + 1) = p2 ~ 2 ( k +) ~ ( k )
2 ( k+
kx,(k : 11)) ] = F O 0 ~ ~ : ~ ' l ] + [ ~ ] u ( k ) (5-117)

Pn xn ( k )
xn ( k + 1) = pn xn( k ) + ~ ( k )
and
and
xlw
y(k) = [cl c2 cnl/(k)] + bou(k) y ( k ) = c , x i ( k ) + c Z X ~ (+~ ) + c n x n ( k )+ b o u ( k ) (5-113)
Rewriting the state equation and the output equation in the form of vector-matrix
xn (k)
equations, we obtain Equations (5-109) and (5-110).
Solation The system puise transfer function can be modified as follows:
Y-
- (z) buzn + bl 2"' +
+ bn e e
(Partial-fraction-expansion prograrnming method) Consider the pulse transfer func-
U(2) ~"+a~z"-l+..-+a, tion system defined by
(bl - al bo)zn-l + (b2 - a2bo)zn-' + . - + (bn - anbo)
,
( z ) - bozn + bl zn-l +
= bo r -Y- + b,
( 2 - pí)(z - pz) (z - pn) U ( Z ) zn + al zn-1 + - + a, m

Since al1 poles of the puise transfer function Y ( z ) / U ( z )are distinct, Y ( z ) / U ( z ) Assume that the system involves a multiple pole of order m at z = p, and that all other
- -
expanded into the following form: poles are distinct.
Show that this system may be represented by the followii g state equation and
output equation:
where

Equation (5-112) can be written in the form

Y ( z ) = bo U ( z ) + A U ( z ) + L?L
-pi z -p2
U(z)+ m - + --LL.
z -pn
w1
Let us define the state variables as follows:
1
X , ( z ) = -u(4
z -p1

Solutisn Since the system pulse transfer function can be written in the form

1
X,(z) = -U ( z >
z -pn
Then Equation (5-114) can be rewritten as = bo + (bl - al bo)zn-l + (b2 - a2b o ) ~ n -+2 . - + (bn - a, bo)
(2 - P I ) ~ ( Z- pm+l) * (2 - pn)
zX,(z) = plxl(z) + U ( z )
zX2(z) = p2&(z) + U(z)
ap. 5 Exarnple Problems and Sslutions

we obtain xrn+i(k + 1) = pm+ixm+i(k)+ ~ ( k )

The output equation given by Equation (5-122) can be rewritten as follows:


Y ( z ) = C I X ( Z )+ ~ ~ X Z +
( Z ). . + cmXm(z) + cm+lXrn+l(z)
Let us define the first m state variables X l ( z ) ,&(z), . . . ,& ( z ) by the equations + cm+zXm+,(z) + + cnXn(Z) + bo U ( Z )
By taking the inverse z transform of this last equation, we get
y(k) = c l x i ( k ) + czxz(k) + . + crnxrn(k) + crn+ixrn+,(k)
+ cm+2xm+2(k)+ ' ' + ~n x n ( k ) + bo ~ ( k ) (5-127)
Rewriting Equations (5-126) and (5-127) in the standard vector matrix form, we obtain
Equations (5-119) and (5-120), respectively.

Using the nested programming method (refer to Problem A-5-2), obtain the state
and the remaining n - m state variables ;%i,+l(z),Xm+2(z), equation and output equation for the system defined by
1
Xm+ 1 ( z) = -
-

Z
M-
-

- pm+i
W )
1 Then draw a block diagram for the system showing al1 ctate variables.
Xm +2 ( Z ) = ------ U(Z)
2 - Pm+2 sHu&ion The given pulse transfer function can be written as
Y ( z ) = z-'{U(z) - 4 Y ( z ) + z-'[SU(z) - 3 Y ( z ) ] }
Define

Notice that the m state variables defined by Equation (5-123) are related each to th
next by the following equations:

The state equation can therefore be given by


Xm-i(z>- 1
Xm(z) -pi
and the output equation becomes
By tahing the inverse z transforms of al1 of Equation (5-125), the last equation - -.
Equation (5-123), and al1 of Equation (5-124), we obtain

Figure 5-3 shows the block diagram for the systern defined by the state-space equations.
The output of each delay elernent constitutes a state variable.

Obtain a state-space representation of the system shown in Figure 5-4. The sampling
period T i s 1 sec.
ap. 5 Example Problems and Solutions

Figure 5-5 Modified block diagram for the system shown in Figure 5-4.

Figure 5-3 Block diagram for the system considered in Problem A-5-5.

from which we get


Y

+ 1)
olution Vde shall first obtain the z transform of the feedfonvard transfer h n
x (k
[xi(k + l)] [ O 0 6 3 2 x (k)
-1 1' ][xi(k)] +

- 0.3679(z + 0.7181)
-
(Z - 1)(z - 0.3679) Obtain a state-space representation of the following pulse-transfer-function system:

diagram modification. Use the partial-fraction-expansion programming method. Also, obtain the initial values
Let us expand G(z) into partial fractions: of the state variables in terms of y (O), y(l), and y(2). Then draw a block diagram for
the system.
1 - z -'
0.6321 --- 0.6321~-'
G(z) = -- rihtion Because we need the initial values of the state variables in terms of y (O), y (l),
z - 1 z - 0.3679 1 - Z-' 1 - 0.3679~-' and y (2), we slightly modify the partial-fraction-expansion programming method pre-
Figure 5-5 shows the block diagram for the system. Let us choose the output ~f e sented in Section 5-2. Let us expand Y(z)/U(z), zY(z)/U(z), and z2 Y(z)/U(z) into
unit delay element as a state variable, as shown in Figure 5-5. Then we obtain partial fractions as follows:
Consequently, we have the state-space equations as follows:

Then we have

The initial data are obtained by use of Equation (5-129), as follows:


Yow iet us define the state variables by the following equation:
P -f r =?

?*he block diagram for this system is shown in Figure 5-6.

Then the state variables X3(z),&(z), and &(z) are related to Y(z), zY(z
as follows: Obtain a state-space representation of the followingpulse-transfer-function system such
that the state rnatrix is diagonal:

z2 Y(z)
Then obtain the initial state x(0) in terrns of y(O),y(l),y(2) and u(O), u(l), u(2).
Erom Equatiom (5--12$), we obtain
Solutisn Let us first divide the nurnerators of the right-hand sides of Y(z)IU(z),
(2 + 1)2Xi(z) = U(z) zY(z)IU(z), and z2 Y(z)/U(z) by the respective denominators and expand the remain-
(z + 1)Xz(z) = U(z) ing terms into partial fractions, as follows:

(Z 2)X3(Z) = U(Z)
Noting that

we get

.(z) is given by the equation


Y(z) = 5Xi(Z) - j&(Z) .t5X3(Z) Figure 5-6 Block diagram for the system considered in Problem A-5-7.
State-Space Analysis .5 Example Broblerns an

The output Y ( z ) Is given by

In vector-matrix notation, the state space equations become

[ =
Rewriting, we have O O x1(k)

~3(+
k 1) -2O -30 ] [x3(k)
x2(k)]+[~]u(k)

The initial data are obtained from Equation (5-131) as follows:


Y@) -
Y (1) - ~ ( 1-) 2u(O)
y(2) - u(2) - 2u(l) + 6u(O)

Figure 5-7 shows the block diagram for the present system.

Let uc define the state variables X1(z),&(z), and % ( z ) as follows:

Then we have

Notice that Equation (5-130) can be written as


zX&) = -X,(z) + U(z)
zX2(z) = -2X4z) + lJ(z)
z x 3 ( z ) = -3X3(Z) + u(Z)
from which we obtain
xi(k + 1) = -xi(k) + u ( k )
xz(k + 1 ) = -2x2(k) + ~ ( k )
x3(k + 1 ) = -3x3(k) u(k)
-i Figure 5-9 Block diagram for the system considered in Problem A-5-8
xarnple Problemc and Solutionc

y assumption, the greatest common divisor of the matrix adj (A1 - A) is


Let A be an n x n matrix, and let its characteristic equation be d (A). Therefore,
lhli - A / = A" + al A"-' + - + a,-~ A + a,
e =O

Show that matrix A. satisfies its characteristic equation, or that where the greatest common divisor of the n2efements (which are functions of A) of
is unity. Since
A" + alAn'-l + . + a n W i A+ a, I =
e

- A) adj (A1 - A) = \Al - Al


(This is the Cayley-Mamilton theorem.)
we obtain
(5-132)
from which we find that / A - Al is divisible by d ( A ) . Let us put

where
(5-133)
Then the coefficient of the highest-degree term in A of $(A) is unity. From Equations
(5-132) and (5-133), we have
Note also that

Wence, we obtain

Note that $(A) can be written as follows:


$(A) = g ( A ) W ) +
where a ( A ) is of Iower degree than $(A). Since $(A) =
. Since 4 ( A ) is the minimal polynomial, a ( h ) must be identically zero, or
4 0 ) = g(h)rp(h)
Note that because $ ( A ) = 0 we can write

This proves the Cayley-Hamilton theorem.

Referring to Problem A-5-3, it has been shown that every n x n matri and we obtain
own characteristic equation. The characteristic equation is not, however
scalar equation of least degree that A satisfies. The least-degree polyn ( 4=g w w
as a root is called the minimal poíynomial. That is, the minimal polynomia Note that the greatest common divisor of the n2 elements of
matrix A is defined as the polynomial $(A) of least degree:
g(A) = 1
+(A) = A" + al A"-' + +a,-~h+a,, m r n
Therefore,
$(A) = +(A)
Then, from this last equation and Equation (5-133), we obtain
The minimal polynomial plays an important role in the computation of polyno
an n x n matrix.
Let us suppose that d ( A ) , a polynomial in A, is the greatest common
al1 the elements of adj ( A 1 - A). Show that if the coefficient of the highest-de It is noted that the minimal polynomial $ ( A ) of an n X n matrix A can be
in h of d ( A ) is chosen as 1 then the minimal polynomial $ ( A ) is given by determined by the following procedure:

B. brnn adj (A1 - A) and write the elements of adj ( A l - A) as factored polynomi-
als in A.
State-Space Analysis hap. 5 Exarnple robfems and Solurions

Determine d(A) as the greatest common divisor of al1 the elemen Next, consider the matrix . The characteristic polynomial is given by
). Choose the coefficient of the highest-degree term in A of d
be 1. If there is no common divisor, d(A) = 1.
3. The minimal polynomial +(A) is then given as \ A

simple computation reveals that matrix has three eigenvectors, and the Jordan
has n distinct eigenvalues, then the minimal polynomia
racteristic polynomial. Also, if the multiple eigenvalues
linked in a Jordan chain, the minimal polynomial and the characteristic polyno
identical. If, however, the multiple eigenvalues of A are not linked in a Jorda
the minimal polynomial is of lower
'CJsing the following matrices A
than the characteristic p
S examples, verify the foregoing state Thus, the multiple eige
[H % i]
ues are not linked. To obtain the minimaí polynomial, we
about the minimal polynomial when multipie eigenvalues are involved. first compute adj (A1 -
(A - 2)(X - 1) o
(A - 2)(A - 1)
3(A - 2) (A - 212
iatioan First, consider the matrix A. The characteristic polynomial is give from which it is evident that
A-2 -1
-A\- O
O
A-2
-3
y
A-1
¡=(A-?)'(A- ence,

Thus, the eigenvalues of A are 2, 2, and 1. Ht can be shown that the Jordan c
form of A is

[H d 81
and the multiple eigenvalues are linked in the Jordan chain as shown. (For the
As a check, let us compute +(

for deriving the Jordan canonical form of A, refer to Appendix A.) For the given matrix , the degree of the minimal polynomiai is lower by 1 than that
To determine the minirnal polynomial, let us first obtain adj ( of the characteristic polynomial. As shown here, if the multiple eigenvalues of an n x n
given by matrix are not linked in a Jordan chain, the minimal polynomial is of lower degree than
(A-2)fA-1) (A+11) 4(A-2) the characteristic polynomial.
- 2)(A
[ :
(A - 1)
3(A - 2) (A - 2)"
Notice that there is no common divisor of al1 the elements of adj (A Show that by use of the minimal polynomial the inverse of a nonsingular matrix A can
d(h) =: 1. Thuc, the minimal polynomial +(A) is identical with the chara be expressed as a polynomial in with scalar coefficients as follows:
nornial, s r

- Al (A - 2)2(h - 1)
= h3 - 5h2 + 8A - 4 where al, a2, . . . , a m are coefficients of the minimal polynomial

A simple calculation proves that

[: : :]
A3 - 5a2+ 8A - 41 = Then, obtain the inverse of the following matrix

but
A = 3 -1 -2
Fi2 - 3A + 21 f
Thus, we have shown that the minimal polynornial and the characte Solution Fsr a nonsingular matrix A, its minirnal polynomial $(A) can be written as
of this matrix A are the same. follows:
hap. 5 Example Problems and Sdlatisns

) =Am + alAm-' + . e - +a,-lA+am where


where a, iO. Hence,
1
= --(Am+ alikm-' + + a,-,A2 + am-IA)
am
Premultiplying by A-', we obtain

which is Equation (5-134). and al, a2, . . . ,a, are the coefficients appearing in the characteristic polynomial given
Eor the given matrix A, adj (h - A) can be given as follows: by
-6 1 = zn + al zn-' + a2znW2 + + a,
Show also that
al = -trG
Clearly, there is no common divisor d(h) of al1 elements of adj (A1 -
d ( h ) = 1. Consequently, the minimal polynomial +(A) is given by the eq
a2 = -2 trGHl

Thus, the minimal polynomial +(A) is the same as the characteristic polyno
Since the characteristic equation is 7% simplify the derivation, assume that n = 3. (The derivation can be easily extended

-
- Al = h3 + 3h2 - 7A - 17 = O to the case of any positive integer n.)
we obtain
$(A) = h3 + 3h2 - 7A - 17 - z2G+Z ~ -Wz G~H ~ + z H 2 - GHz
y identifying the coefficients ai of the minimal polynomial (which is the sam
characteristic polynomial in this case), we have
al = 3, a, = -7, a3 = -17
The inverse of W can then be obtained from Equation (5-134) as follows:

:I3[n
1
+ a l A + a,I) = -(A2
17
+3 The Cayley-Hamilton theorem (see Problem A-5-9) states that an n x n matrix G
7 O -4 satisfies its own characteristic equation. Since n = 3 in the present case, the character-
+ -.
istic equation is
-6 1 = z3 + a 1 z 2 + a2z + a3 = O
and G satisfies the following equation:

=A[! -;; ]
6 -4
G3 + al gy2 + a 2 G 4- as

[$ -8 ];
-

-
17
6. -4.

-
17
-L.
17
Hence Equation (5-136) simplifies to

Consequently,
z2 + H1z + H2) = (z3 + a l z 2 + a 2 z C a3)I = jz

Show that the inverse of z - G can be given by the equation


adj (z1 - G )
(z1 - G)-' =
IzP - G /
-
- 1zn-' + H1 + I ~ ? J ~ z+~ - ~ - + Hn-1
lzli - G / which is Equation (5-135) when n = 3.
Chap. 5 Exarnple

Next, we shall show that


al - trG
=

a, = - S tr GH1
a3 = - S tr GHs
We shall transform into a diagonal matrix if G involves n Iinearly inde
eigenvectors (where = 3 in the present case) or into a matrix in a Jordan c
forrn if G involves fewer than rz linearly independent eigenvectors. That is,
-1
=D = matrix in diagonal form
Let us write
or
= matrix in a Jordan canonical form
are nonsingular transformatioh matrices. L O O ~ 3 1
Since the following derivation applies regardless of whether matrix where an asterisk denotes "either O or 1." Then
transformed into a diagonal matrix or into a matrix in a Jordan canonical form,
use the notation
- ~ G T= fi
where represents either a diagonal matrix or a matrix in a Jordan canonical form
the case may be.
In the following we shall first show that
trG = trD
a3 = -p1p2p3
tr GHI = tr fihl Notice that
tr GH2 = tr fiG2
where
ii1=n+a1~
ii2= nE9, + a,I
Then we shall show that
al = -trD
n *
a, = - S ir DHI
a3 = - 4 ir DE2
A A

Wotice that since


= 3plp2p3 = -3a3
Thus, we have shown that
we have

Notice also that

Now we have
Consider the following oscillator system:
tr G = tr TBT-' = tr B
trGH1 = trG(G + alI) = trG2 + tralG
State-Space Analysis Chap. 5 Example Problems and

Qbtain the continuous-time state-space representation of the system. Then Noting that D is zero, we have
the system and obtain the discrete-time state-space representation. Also o
pulse transfer function of the discretized system.
z
s i n w~ 1-l
- cos wT
[ 1
ution From the given transfer hnction, we have
y + w Z y = w2u
- ( 1 - cos wT)(z + 1)
Define
-
Xl =y z2 - 22 cos wT +1
1 Hence,
X2 =-j
w
Then we obtain the following continuous-time state-space representation of the syst
m
U ( z1
= F ( r ) = ( 1 - cos O J T ) (+~ 2-')z-'
1 - 22-1 cos W T + z-'
Note that the pulse transfer function obtained in this way is the same as that obtained
by taking the z transform of the system that is preceded by a zero-order hold. That is,

The discrete-time state-space representation of the systern is obtained as fo 1 1 - z P 1cos wT


Noting that
- ( 1 - cos w"(l + z-')z-l
-
1 - 22-' c o s o T + z-'
Thus, we get the same expression for the pulse transfer function. The reason for this
we have is that discretization in the state space yields the zero-arder hold equivalent of the
-1
continuous-time system.

Consider the system shown in Figure 5-8(a). This system involves complex poles. Tt is
stable but not asymptotically stable in the sense of Liapunov. Figure 5-8(b) shows a
--
s2 + w2 s2 + w2
cos wT
-sin wT
sin wT
cos wT 1
and

-sin o h cos wh
-.
1 - cos wT

Hence, the discrete-time state-space representation of the oscillator systern be


as follows:

-sin oT cos wT x2(k7')


+ [
cos COT sin W T ] [ ~ ~ ( ~ T 1) -]
sin wT
COT]~(

xl(kT)
Y = 1 o~[x2(ki)]
The pulse transfer function of the discretized system can be obta
Equation (5-60):
Figure 5-8 (a) Continuous-time system of Problem A-5-15; (b) discretized version
F ( z ) = C(z1 - G)-'H +D of the system.
Chap. 5 Exarnple Problerns and Soluaions

The pulse transfer function of the discretized system shown in Figure 5-8(b) is
discretized version of the continuous-time system. The discretized systern is also sta
but not asymptotically stable.
Assuming a unit-step input, show that the discretized system may exhibit hid
oscillations when the sampling period T assumes a certain value.
&y =#- 7

e-" " -
= (1 - z-')+&]
- -

aa The unit-step response of the continuous-time system shown in Figure 5-8


1s
Hence, the unit-step response is obtained as follows:

Hence ,
y (t) = cos rt
[Notice that the average value of the output y(t) is zero, not unity.] The respsnse y
The response y (kT) becomes oscillatory if T f nn-sec ( n = 1,2,3, . . . ). For example,
versus r is shown in Figure 5-9(a).
the response of the discretized systern when T = $71. sec becomes as folIows:

Hence,

A plot of y(kT) versus k T when T = n- sec is shown in Figure 5-9(b). Clearly, the
response is oscillatory. If, however, the sampling period T is rr sec, or T = n-, then

The response y(kT) for k = 0,1,2, . . . is constant at unity. A plot of y(kT) versus kT
when T = n- is shown in Figure 5-9(c).
Notice that if T = n- sec (in fact, if T = nrr sec, where n = 1,2,3, . . . ) the
unit-step response sequence stays at unity. Such a response may give us an irnpression