Ogata, Katsuhiko .
Discretetime control systems / Katsuhiko Ogata.  2nd ed.
p. cm.
Hncludes bibliographical references and inciex.
HSBN 0130342815
1. Discretetime systems. 2. Control theory. H. IEltle.
A402.04 1994 9419896
629,8'Sdc20 CHP
publisher. This book can also serve as a selfstudy book for practicing engineers w
discretetime control theory by themselves.
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 1l(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 "discretetime signal" is broader than the term "digital signal" or the
term "sampleddata signal." 1can refer either to a digital
signal or to a sampleddata 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 continuoustime 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 14.
Loosely speaking, terminologies such as discretetime con
pleddata 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 discretetime signals exist in the system; the discretetime signals are
amplitudemodulated pulse signals. Digital control systems may include both contin
ustime and discretetime signals; here, the latter are in a numerically coded form.
th sampleddata control systems and digital control systems are discretetime
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.
Discretetime control systems differ from continuoustime control systems in
that signals for a discretetime control system are in sampleddata 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 12 depicts a block diagram of a
Continuoustime systems, whose signals are continuous in time, may be de tion of the basic control scheme. The sy
scribed by differential equations. Discretetime systems, which involve sample feedforward control. In designing such
data signals or digital signals and possibly continuoustime 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 continuoustime signal replaces the .
Figure 13 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 continuoustime 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 analogtodigital
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 largescale controller or computer is timeshared 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 sampleddata 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 continuoustime signal and is fed to the actuator.
shall present details of such signalprocessing methods in the digital controlle
Section 14.
The term "discretization," rather than "sampling," is frequently used in the
analysis of multipleinputmultipleoutput systems, although both mean basically
control systems that actually contain only continuoustime signals. In fact, we often L _I
use a suitable discretetime model for a continuoustime system. An example is a Digital controlier
signar e(t) may be plotted as shown in Figure 14(c). The average value of e(t) is zem,
or e(t) = O. Then the variance u h f the quantization noise is
ith the rapid growth in the use of digital computers to perform digital control
actions, both the dataac 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:
Physical T o digital
variable 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 15 (a) Block diagram of a dataacquisition system; (b) block diagram of a data
distribution system.
be aiceptable as a firstorder approximation.] The probability distribution P ( e ) of
Introduction to DiscreteTime 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 dataacquisition process is done by electronic means. switches among many analog input channels in bed fashion. The number
In Figure 15(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 lowpass filter that follows the amplifier attenuates the time. During the connection time the sampl ircuit samples the signal
highfrequency signal wmponents, such as noise signals. (Note that voltage (analog signal) and holds its alue, while the analogto
are random in nature and may be reduced by lowpass filters. converts the analog value into digital ta (binary numbers). Each channel is read
common electrical noises as powerline 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 lowpass 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 sampleandhold 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 dataacquisition process is the atadistribution Process “%+, the output analog signal for that channel.
shown in Figure 15(b), a datadistribution system consists of registers, a demulti er in a digital system convelrts an analog
plexer, digitaltoanalog 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 sampleandhold 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
signalprocessing system. however, the sampling operation and
lexer. An analogtodigital converter is the most expensive (see Section 32). It is common practice to use a single analogtodigital converter
component in a dataacquisition system. The analog multiplexer is a device that and multiplex many sampled analog inputs into it.
performs the function of timesharing 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 sampleandhold. (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 17 shows a simplified diagram for the sampleandhold. Th
Figure 16 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 highquality capacitor with low leakage and low
dielectric absorption characteristics.
In Figure 17 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 sampleandhold 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 18 shows the tracking mode and the
hold mode.
Figure 16 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 DiscreteTime Control Cystems Chap. 1 Sec. 14 Data Acquisition, Conversion, and istribution Systemc
veapproximation type
Sarnpleandhoid
cornrnand
Each of these four types has its own advantages an
Figure 17 Sampleandhold 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, analogtodigital converters art of thek feedback l o o p
place (that is, the time interval when the measured amplitude is uncertain) is called digitaltoanalog 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 highinputimpedance 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 sampleandhold 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 oofAnalogtoDigital earlier, the PrQcess by The successiveapproximation 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 19 shows a schematic
called analogtodigital conversion. Thus, an AID converter transforms an analog diagrarn of the successiveapproximarion type of
Digital
output
Tracking
mode 4 ey; :pk
t
Analog
input
t
tioíd command Figure 143 Traclcing mode and hold
is given here mode. Figure 19 Schematic diagram of a successiveapproximationtypeof BID converter.
lntroduction to DiscreteTime Control Sec. 'l4 Data Acquisition, Conversion, and
Output
Output
have been used in many srnallscale 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 nondecision
e cost of analog computers or analog controllers increases
of the computations increases, if constant accuracy is to
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 onesarnpling erally not practica1 t ry long time at steady state, because certain
period delay, as shown in Figure 115. 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 closedloop systems, such a delay is not desirable, and so the interpolative trial processes must always be taken into consideration. Since there are interactions
firstorder 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 largescale
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 smallscale 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 DiscreteTime 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 closedloop systems in the z plane, followed by tran
sient and steadystate response analyses, design by the rootlocus and frequency
response methods, and an analytical design method. Chapter 5 gives state
space representation of discretetime systems, the solution of discretetime state
space equations, and the pulse transfer function matrix. Then, discretization of
continuoustime statespace 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
fullorder state observen and minimumorder state observe atical tool commonly used for the analysis an esis of discretetime
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 continuoustime 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 poleplacement approach ombined with minimumorder observ
ers. The design of modelmatching 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 timeinvariant differential equations into algebraic
The statespace analysis and design of discretetime control systems, presented equations in S , the z transformation transforms linear timeinvariant 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 vectormatrix 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 poleplacement 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, Diseretet signals arise if the system involves a
Solved problems are an integral part of the text. Appendixes A, sampling of continuoustime 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. 23 z Transformc of Elementary Functions
The 2 transform applies to the continuoustime signal x(t), sampled signal The z transform defined by Equation (23) or (24) is referred to as the twosided
x(kT), and the number sequence x(k). dealing with the z transform~if no z transform. In the twosided 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 onesided and twosided z transforms
ance of T and write x(kT) as x(k).] involves both positive and negative powers of z
onesided z transform is considered in detail.
ection 21. has presented introductory remarks. For most engineering applications the onesided z transform
ection 22 presents the definition of the z transform and associated subjects. venient closedform solution in its r
Section 23 gives z transforms of elementary functions. mportant properties an infinite series in z', converges
theorems of the z transform are presented in Section 24. 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)
25. Section 26 presents the solution of difference equations by the z transform ks convergent.
method. Finally, Section 27 gives concluding commeats. Notice that expansion of the righthand side of Equation (21) gives
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 (22), 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 ,..
l'j
Zl
 1  z' COS wT
1  2.2' cos wT + z'
The use of the z transform method in the analysis of discretetime 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
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,
Find the z transforms of unitstep functions that are delayed by 1 sampling period
and 4 sampling periods, respectively, as shown in Figure 22(a) and (b).
By defining m = k  n, Equation (29) can be written as follows: Using the shifting theorem given by Equation (27), we have
m
z'
r [i(t  TI] = Z  ~ Z [ I ( ~ )=] Zl 
1 1*1  1  zl
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(28)? we note that
From this last equation, we obtain o 2T 3T 4T 5T 6T 7T 8T ' Faginre22 (a) UnitSstep function
7[ ~ ( k+ l)]= zX(z)  zx(Q) (21 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) (212) (b) periods.
The z Transform Chap. 2 Cec. 24 lmportant Properties and Theorems of the z Transform
Thus,
which yields
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 z1)X(z)]
k*m 12 1 X(z), the z transform of x(kT) or x(k), is given, the operation that
which is Equation (216). 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
Figure 23 Two different continuoustime functions, x l ( t ) and x Z ( t ) ,that have the
same values at t = O, T, 2T,. . . .
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 continuoustime 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. 25 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 (218), ,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 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 closedforrn expression for x(k), except in special cases.
n finding the inverse z transform, we utilize the Kronecker delta function So(kT), Enter the Kronecker delta input.
wbere
X =
Y=
1 0 4 Figure 24 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
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
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 21, 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 (221), we fin$ 2'[X(z)] = x(k) = y(k  1)
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
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 partialfractionexpansion 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  eaT
Thus,
I
hand, in certain problems the partialfractionexpansion approach may become
m
y1
1  z  1zl
+z2 ] ;sí
=
krr
(1.) sin 3 laborious. Then, the inversion integral method proves to be very convenient.
krr 4 (kl)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 = eOT.
Hence, from Equation (224), we have
Hence,
(225)
Ef X(z)zk' contains a mult
1
Zktl
x(k) = [residue of at pole z = z,
I=I (z  z eOT)
I ) ~ (
= Ki + KZ
The z Transform 6 P Transform Method for
( Z  ~ ) ~ ( e"')
z ] = ( 1  eaT)2
K2 = [residue at double pole z = 11
= lim
(k + l ) z k ( z  e  a T ) zk+l
z. I (z  ePaT)'
= 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 24 and are surnrnarized in Table 23 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 (continuoustime)
control systems, such as the rootlocus 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
ztransforrned 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  ~
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) =
ba
(a)" + ax(0)
ab
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
k=O
where k = 0 , 1 , 2 , . . . .
Obtain the z transforrn of kak' by two methods.
In this chapter the basic theory of the z transform rnet een presenteda T'he
z transform serves the same purpose for linear timei Saektirne sY5&Xls
as the Laplace transform provides for linear timeinvariant continuoustime systems.
method of analyzing data in discrete time results in difference
z transforrn rnethod, linear timeinvariant 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[kakll = 5kaklzk
k=O
= a1
k=O
kakzk=
a
2
k=O
(a) Ir
o~blemA2
k m
Show that
2 ~ ( h=) E x ( k ) = lim X ( z )
21
h=O k=O
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
where 1 5 i 5 k  1.
we obtain
Solintaon Define 11
k
y(k) = 2 x(h),
h =O
k = O, 1,2,. . .
Since
Then

 of the individual terms give
The inverse z transforms
(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 21,
) $ 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 unitstep 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)]
x(k)= /O9 1,
2k1,
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
The z transforms of the two terms on the rigkthand side of this last equation can be
found from Table 21. 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
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 (224) 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(2k1)=10(2k11), k=1,2939
Mence, the inverse z transform of the given X ( z ) can be written
Then enter the command
(230)
Columns 13 through 24
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, (224), 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 ) !Zi dz
d
Method 3: Partialfractionexpansion method. expand X ( z ) into the foliowing
= lim[(z
Z+I dz + 2)zk]
partial fractions: =3k+1, k = 0 , 1 , 2,...
2 X ( z )  22l 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:
ence ,
Note that although this lact equation involves '~6the square roots in the righthand 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
kw x(k) 2
144 233 377 61 O 987 1597
Columns 19 through 24
Solving this equation for X ( z ) and substituting the initial data x(0) = O and x ( 1 )
we get
= 1, 1 832040
2
Obtain the z transform of k3.
where a is a constant.
Then, referring to Example 219, the solution x ( k ) for k = 0 , 1 , 2 , . . . can be given by 2
Show that
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 27. 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.
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 partialfractionexpansion method and ( 2 ) the the following two cases:
a MATLAB program for finding x ( k ) , the inverse z transform of X ( z ) .
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.
217
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  2l)(1  0.22l)
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
21
with MATLAB.
Find the inverse z transform of
Use (1) the direct division method and (2) the MATLAB method.
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 continuoustime design methods to k=O
er. The outline of this chapter is as follows. Section 31 The sampler output is equal to the product of t e continuoustime input x(t) and the
gives int ks. Section 32 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 continuoustime signal x(t) by impulse modulation. This section includes as the carrier, as shown in Figure 32.
derivations of the transfer functions of the zeroorder hold and firstorder hold.
Section 33 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 34. ed on the fact that the Laplace
transform of the sampled slgnal is periodic, resent the sampling t h e ~ r e m .
ection 35 discusses the pulse transfer functio thematical modeling of digital
controllers in terms of pulse transfer functions is discussed. Section 36 treats the
realization of digital controllers and digital filters.
x*(t)
signal 1
1
Consider the sampler and zeroorder hold shown in Figure 34(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:
Zeroorder
hold
where O I( r < T and k = 0,1,2, . . . . Equation (37) implies that the circuit holds
the arnplitude of the sample from one sampling instantto the next. Such a datahold
lace transform of Equation (39) becsmes
is cailled a zeroarder hold, or clamper, or staircase generator. The output of the
zeroorder 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 zeroorder hold may
be given by
1  eTs
G,, =  Next, consider ehe mathematical mo shown in Figure 34(b). The output
S
of this model rnust be the same as tfiat of real zeroorder hold, or
Figure 33 shows a sampler and a zeroorder hold. The
input signal x(t) is sampled at discrete hnstants and the sampled signal is passed
through the zeroorder hold. The zeroorder 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 (38) From Figure 34(b), we have
e shall obtain a mathematical model of the combination of a real sampler and
zeroorder circuit, as shown in Figure 34(a). Utilizing the fact that the integral of
an impulse function is a constant, we may assume that the zeroorder hold is an Since
integrator, and the input to the zeroorder hold circuit is a train of impulses. Then
a rnathematical model for the real sampler and zeroarder hold may be constructed
as shown in Figure 34(b), where GhO(s)is the transfer function of the zeroorcler
hold and x" (t) is the impulse sampled signal of x(t). Equation (311) may be written as
Zeroorder
hold y comparing Equations (312) and (3E?), we see that the transfer function of the
zeroorder hold may be given by
Note that, mathematically, the system shown in Figure 34(a) is equivalent to the
(a) A real sampler and
hold; (b) mathematical
system shown in Figure 34(b) from the viewpoint of the inputoutput relationshi
61
model that consists of an impulse That is, a real sampler and zeroorder hold can be aeplaced by a rnathernatically
(b) sampler and transfer function Gh0(s). equivalent continuoustime system that consists of an impulse sampler and a transfer
zPlane Analysis of DiscreteTime Control ec. 32 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 firstorder hold circuit, it is convenient
in Figure 34) by the rnanner in which the sampling switches are drawn. to assume a simple function for x(t). For
o not use firstorder function, and a unitramp function
holds in control systems, it is worthwhile to see what the transfer function of uppose that we choose a unitstep fun
firstorder holds may look like show that the transfer function of the order hold shown in Figure 36(a)
firstorder hold may be given by consists of stsaight lines that are
The output h(t) is shown in the
follows:
Firstorder
hold
Figure 36(b) shows a mathematical model of the real sarnpler cascaded wit It is repeated that the impulse samplier is a fictitious sampler introduc
the firstorder hold shown in Figure 36(a). The mathematical model consists of the the purpose of mathematical analiysi
impulse sampler and GR1(s),the transfer function of the firstorder 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 firstorder 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 firstorder hold is given by method.
Consider the impulse sarnpler shown in Figure 37. The output of the i
sampler is
Note that a real sarnpler combined with a firstorder hold is equivalent to an impulse Noting that
sampler combined with a transfer function (1  eTS)2(1FJ+ l)/(aS2).
Sirnilarly, transfer functions of higherorder hold circuits may be derived by
following the procedure presented. However, since higherorder
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 zeroorder
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 Laplacetransformable functions
whose strengths are equal to the continuoustime signal at the sampling in f (t) and g (t) can be given by
stants, simplifies the analysis of discretetime 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
discretetime control systems that involve samplers and hold circuits. This [For the derivation of Equatio l9), see Problern A34.1
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 discretetime the Laplace transform of
systems involving sampling operations.
3, As pointed out earlier, once the real sampler and zeroorder hold are mathe
rnatically replaced by the impulse sampler and transfer function (1  eTs)ls,
the system becomes a continuoustime system. n i i s simplifies the analysis of
the discretetime control system, since we may apply the techniques available
to continuoustime control systems.  ..
6, Figure 37 Impulse sampler
zPlane Analysis of DiscreteTime 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 wellknown 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 halfplane, 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 38 Ciosed contour in the left
where T is a semicircle of infinite radius in the left or right half p plane. 1 eTts  p )
half of the p plane.
ñhere are two ways to evaluate t is integral ( m e using an infinite semicircle
in the Mhalf plane and the other an infinite semi roblem A36 for the derivation of Equation (322).]
shall describe the results obtained by the two case for eTsin Equation (322), we have
analysis here, we assume that the poles
S) can be expressed as a ratio of polly
Given
This integral is equal to the sum of the residues of ( p ) in the &sed contour.
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 (326), we need
denominator of X ( s ) is two or
= lim
*+ O ( S + 1)2(z  eTSl2
+(1)2 z  eT
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 39, 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  eT("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
Note that this expression of the z transforrn is useful in proving the sampling theorem
(see Section 34). 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 (326) is fx(O+). That is,
Hence,
1 1
X*(s + jk) =
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 )  eTs&;*(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 zeroorder hold. Then the product of the transfer function of the ( z =( S = (1  ) Z [ ? ] (33 2)
zeroorder hold and C ( s ) becomes
1  eTs
X(s) = y a s )
Referring to Equation (332), we have spectrum of the sampled signal x"(t). The Laplace transform of x
obtained in Section 33 and is given by uation (327) or (329),
whether x(O+) = O or quency spectrum,
s in Equation (327). uency spectra, we nee
with the value of x(O+).] Thus,
sampling frequency is sufficiently high compar if a function X ( s ) has a pole at s = sl, then
onent involved in the continuoustime signal, t ( k = o, 1, 2, . . .).
Figures 311(a) and (b) show plots of t e frequency spectra X* (jw) versus
Figure 511 Plols of the frequency spectra / X":( j w ) l versus w for two values of sampling
wl O 0, w Figure 310 A frequency spectrum. frequency w,: (a) ws > 2wl; (b)w, < 2wl.
aPlane Analysis of DiscreteTime Control Sycterns Chap. 3 econstructing Original Signals f r o m Sampled Signals
for two values of the sampling frequency os. Figure 311(a) corresponds to o, > 2wl,
ile Figure 311(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 313 Frequency spectra of the signals before and after ideal filtering.
ignal x(t) can be reconst ted from the impulsesampled
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 continuoustime 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 continuoustime signal.
closedloop system a (In practice, the frequency spectrum of continuoustime 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 312. 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
highestfrequency component of t nuoustime ségnal.
reconstructs the continuoustime 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 continuoustime signal x(t). Since t
S
has constantmagnitude 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 (335) gives the unitimpulse response of the ideal filter. Figure 314
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 312 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 lowpass filter. physical world. Hense, such an ideal filter Is physlcally unrealizable. [lin many
zPlane Analysis of DiscreteTime Control ec. 34 Reconstructing Original Signals f r o m Campled Signals
To compare the zeroorder hold with the ideal filter, we shall obtain the frequency
response characteristics of the transfer function of the zeroorder hold.
ing jo for s in Equation (336), we obtain
The magnitude becomes zero at the frequency equal to the sarnpling frequency and Figure 315 (a) Frequencyresponse curves for the zeroorder hold; (b) equivalent Bode
at integral rnultiples of the sarnpling frequency. dictgram when T = 1 sec.
zPlane Analysis of iscreteTime Control Cystems C econstructing Original
where
3ws 2w, w
 w, O W,
 Ws 2w, 30, w
2 2
Comparison of the ideal filter and the zeroorder hold. Figure 318 Frequency spectra of an impulsesampled signal x* ( t ) .
rPlane Analysis of iscreteTime Control Sec. 35 The Pulse Transfer Function
component at o = n
it were a frequency
sin 3t
frequency component /
t frequency &~)2when
y o,  02 (in general,
In the absence o£ the output sampler, if we consi Equations (339) and (340) can be taken from
chronized in phase with the input sampler an changing the value of the summation. Therefore, Equations (339) and (340) 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 (338).
For the continuoustime system, it is a wellknown fact that the output y(t) of
e system is related to the input x ( t ) by the convolution integral, or
where g ( k T ) is the system's weighting sequence. [The inverse z transform of C ( z ) Figure 321 (a) Plot of output y(t)
is called the weighting sequence.] The summation in Equation (339) or (340) 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.
zPiane Analysis of DiscreteTime 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:
limsG(s) = O.
S"='
ulse on, From Equation (341) we then, referring to Equation (344), 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 discretetime
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
In the following we skall show that in taking the starred Laplace transform of
G(z) given by Equation (3441, the ratio of the output Y(z) and the input X(z), is Equation (345) we may factor out X * (S) so that
cafled the pulse trunsfer function of the discretetime system. It is the z transform
of the weighting sequence. Figure 322 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 (3431, the z transforrn of the output signal can be obtained as the product simplifying the block diagram of the discretetime control system.
of the system9spulse transfer function and the z transform of the input signal. So derive Equation (347), note that the inverse Eaplace transforrn of Y(s)
given by Equation (345) can be written as follows:
v*W
_____)_
S, Y(z)
Fictitious
sarnpier
where m = n  k . Thus, (a) Continuoustime system with an impulse sampler at the input; (b)
continuoustime system.
(348) Next, consider the system shown in Figure 324(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 (348) may be expressed as
ortant fact to rernernber is that the pulse transfer function for this system
which is Equation (347). 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 (345) 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 324(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 pulsetransferfunction approach). Y(z) = G(z)X(z)
while, for the system shown in Figure 324(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 324(a).
m is
The pulse transfer function G(z) of the system shown in Figure 324(a) is Y(s) = G(s)X(s)
which yields
Y" (S) = [G(s)X(s)f* = [GX(s)]*
? zPlane Analysis of DiscreteTime Control Systerns Chap. 3 ec. 35 The Pulse Transfer Feinction
ExarnpIes 34 and 35 demonstrate t e methods h r obtaining the pulse From Table 21, the z transform of each o£the partialfractionexpansion terms can be
transfen: function. found. Thus,
Obtain the pulse transfer function G(z) of the system shown in Figure 324(a), where
G(s) is given by
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 21, we have response function:
Hence ,
ekT + 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,
zPlane Analysis of iscreteTime Control Systems Chap. 3 ec. 35 The Pulse Transfer Function
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,
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 326(a) and (b). Obtain the pulse transfer
given by function Y(z)IX(z) for each of these two systerns.
Figure 326 (a) Sampled system with a sampler between elements G(s) = l/(s a )+
Figure 325 (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. 35 The Pulse Transfer Function
h r the system of Figure 326(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 326(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 closedloop systern is
d ~ Tra~esfer
e Fune~a'csnof C s. In a closedjoop 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 closedloop operation.) Table 31 shows five typkal configurations for closedloop discretetime con
Consider the closedloop control system shown in Figure 327. 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
iscretetime closedloop 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 inputoutput 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:
ulse transfer function GD(z) of the digital controller may then be given by
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 inputoutput characteristics can
be generated.]
Weferring to Figure 328(b), let us define
ere e(t) is the input to the control1
Since
Define
and, therefore,
Consider the control system with a digital PID coneroller shown in Figure 331(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 (356)
is commonly referred to as rhe positional form of the PID control scheme. Figure 330 Block diagram realization of the velocityform digital PID control scheme.
zPlane Anaiysis of DiscreteTime Control Systerns Chap. 3
Figure 331 (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
zeroorder hold becomes
1  e"
C, ( S ) = 
Since
we may redraw the block diagram of Figure 331(a) as shown in Figure 331(b).
Let us obtain the unitstep 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
k
We shall use the MATLAB approach to obtain the unitstep response. re 332 Unitstep response.
iscreteTime Control Sec. 35 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
Statespace method
Equation (359) will give the continuoustime response c(t). ence, the response
al any time between two consecutive sampling instants can be calculated by the use
of Equation (359). [See Problem A318 for sample calculationas of the righthand
side of Equation (359).]
1 zPlane Analysis of Biscretelime Control ec. 36 Realization of Digital Controllers and
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 (356) can be expressed
in the form of Equation (360), as follows:
where
and
Then, draw block diagrams for the systems given by Equations (361) and (362),
respectively. To draw the block diagrams, we may rewrite Equation (361) as
Y(z) = b,N(z) + b , ~  ~ N ( z+)  . + b,z"H(z)
e (363)
and Equation (362) as
H ( z ) = X(z)  al z1 H(z)  a, zW2
M(z)    a, z" H(z) (364)
Then from Equation (363) we obtain Figure 336(a). Similarly, we get Figure
336(b) from Equation (364). The combination of these two block diagrams gives
the block diagram for the digital filter G(z), as shown in Figure 336(c). The block
diagram realization as presented here is based on the standard programming. Norice (a) Block diagram realization of Equation (363); (b) block diagram
that we use only n delay elements. The coefficients al, a2, . . a, appear as feedback f Equation (364); (c) block diagram realization of the digital filter
elements, and the coefficients bo,bl, . . . , b, appear as fee given by Equation (360) by standard propramminp.
zPiane Analysis of DiscreteTime Control ystems Chap. 3 ec. 36 Realization of Di ¡tal Controllers and
The block diagrams in Figures 335 and 336(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 337 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 roundoff errors in the arithmetic oper
 1 + eizl +f i ~  ~
ations in the digital system.
,1 + cizl + diz2
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 33S(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 338(a) and (b).
For decomposing higherorder 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 339.
e shall discusc these three programmings next. ecause of the presence of the const term A , the first and secondarder
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 firstordes
transfer functions. If G(z) can be written as a product caf
The prograrnming method based on this scheme is calledi ladder programming . Let
igore 338 (a) Block diagram representation of Equation (365); (b) block us define
diagram representation of Equation (366).
are shown in Figures 340(a) and (b), respectively. The parallel connection of q + 1
component digital filters as shown in Figure 340 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
zPlane Analysis of DiscreteTime Control ystems Chap. 3 ec. 36 Realization of Digital Controllerc and
(b)
Figure 341 (a) Block diagram for GIB'(z) given by Equation (370); (b) block
diagram for G(*'(z) given by Equation (371).
zPlane Analysis of iscreteTime Control Systems Chap. 3 ealimation of Digital Controllers and
= AiX(Z)
Xi(z)  c(+R](z)~(z)
rnay be drawn as shown in Figure 341(b). Note that
1
6;$yz) = 
A,
y combining component digital filters as shown in Figure 342(a), it is
posslble to draw the block diagrarn of the digital filter G(z) as sho
342(b). [Note that Figures 342(a) and (b) correspond to the case
Also, instead of G(z), its inverse l/G(z) may be anded into continuedEraction
en&. Digital filters based on la s in terrns of z or z1 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 continuedfraction expansion given by Obtain the block diagrams for the following pulsetransferfunction 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 :
direct programming yields the block diagram showri in Figure 343. Notice that we need
two delay elements.
2. Standard programming. We shall firct rewrite the pulse transfer function as
follows:
where
and
Figure 342 (a) Component block diagrams for ladder programming of G(z) given by
Equation (369) when t~ = 2; (b) combination of component block diagrams showing ladder Figure 343 Block diagram realization of Y(z)lX(z) = (2  0.62')1(1 + 0.5z')
programming of G ( z ) . (direct programmins).
zPlane Analysis of DiscreteTime Control Systems Ckap. 3 ec. 36 Realizarion of Digital Controllers and
Block diagram realizations of these last two equations are shown in Figure
344(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 344(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:
ence, we obtain
Referring to Figure 341(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 345. Notice that we need
only one delay element.
Figure 344 (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 (373) 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. 36 Realimation of Digital Controllerc and
impulse response sequence is finite. his type of digital filter is calle The characteristics of the finiteim ulse response filter can be summarized as
filter. lit is also called a nonrecursive filter or a rnovingaveragefilter. follows:
resent value of the o
. The finiteimpulse The finiteimpulse 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 finiteimpulse r filter does not require feed
, so the direct programrn programming are identical.
Let us define the finiteimpulse 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 finiteimpulse response filter are
g(hT)x(kT  ha") at the origin, and therefore it is always stable.
If the input signal involves highfrequency components, then the number sf
= g (0)x(k?")+ g ( T ) x( ( k  1)T ) + .  + g ( k T ) x( O ) (374) delay elements needed in the finiteimpulse 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 righthand side of Equation (374) consists of k + 1terrns. impulse response filter compared with the infiniteimpulse 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 38 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 righthand side of Equation (374) 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 )(395
Since Equation (375) is a differen uation, the corresponding digital fálter ira tk
z plane can be obtained as foliows taking the z transform of Equation (375) 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 ) (376)
Figure 346 shows the block diagram realization of this filter.
Figure 347 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 (377), 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 346 Block diagram realization of Equation (376). y ( T ) = 1.6
aria! zPiane Analvsis of DiscreteTime C o n t r ~Systems
l Chap. 3 Chap. 3 Exarnple Problerns and Solutions
Obtain the expression for y(t). Then find Y ( s ) and obtain the transfer function
of the zeroorder hold.
olution From Figure 349 we obtain
Figure 347 Block diagram for the digital filter given by Equation (377) (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 348.
where
m
Since
X * ( s ) = T[x"((t)]= T[x(O)G(t)]= x(0)
the transfer function of the firstorder hold is obtained as follows:
&on. For an impulse input of magnitude x(0) such that x*(t) = x(0)6(t), the
,+ . , / A (3781 becomes as shown in Figure 351. 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, eTs may be replaced by
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
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(r) = LjcYm
277j
F(s)efl ds,
clm
t >O we have
1
where c is the abscissa of convergence for E(s). Thus, G(s  p ) =
1e~(~~)
.qf(f)g(t)l =
1
T;;;j" m
I,,.C+Jm
F(p)eP'+ &)e" dt
Notice that the poles of 141  eT(sP'] 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=skjk=skjwsk, k = O , 1 , 2 ,...
T
j~g(i)e'sp"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 = ,ijcjm
F(P)W
c+im
 P) dp
 1 "+'" 1

2Trjj,,m
X(p) 1  ens  PI d~ (386)
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  eT(sP)].Equation (386) is the convolution integral. It is a wellknown 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 halfplane, 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 halfplane and the other an infinite semicircle in
the right halfplane); we shall consider these two cases separately in Problems A36
and A37.
ution Referring to Equation (383), rewritten as
1 c+lm
%[f(t)g(t)l = ,ijc,
F(P)G(s  P) dp
Referring to Equation (386), rewritten as
where
m
1
x w = ,ij,,m
'+'" 1
X ( p I 1  eT(s,>d~
f(t) = x(t) and g(t) = 6(t  kT)
k=o show that, by performing the integration in the left halfplane, X" (S) may be given by
and noting that
2 [ 8 ( t  k T ) ] = e"'
zPiane Analysís of DiscreteTime Control Systerns Chap. 3 Chap. 3 Example Problerns and Sol~atíons
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 halfplane 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
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
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 (390) can be written as
X(P>
1  eT('p) at pele 0f X(~)
1 (389) h(z = 2 [residue of X(s)z at pole of ~ ( s )
z  eTs 1
nalycis of DiscreleTime Control Chap. 3 Chap. 3 Example Problemc and Solutions
p plane
(393)
show that by performing this integration in the right haif p plane, X* (S) may be given
CyJ
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 (386) in the right
half of the p plane. Let us choose the closed contour shown in Figure 353, which
+
consists of the line from c  j w to c jm and r R ,the portion of the semicircle of infinite ure 353 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  eT("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
Noting that
Thus,
and noting that
Note that this expression of the z transform is useful in proving the sampling theorem
(see Section 34). 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 (396) is $x(O+). That is,
Then the integral term on the righthand side of Equation (396) becomes
Thus, we have obtained X ( z ) by using the convolution integral in the right halfplane.
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 halfplane.
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 (395). Referring to this equation, we have by using (1) the partialfractionexpansion method and (2) the residue method.
Solntlopa
1. Partialfractionexpansion method. Since X ( s ) can be expanded into the form
qs) 2 1 2
=   
s+1 ( ~ + 1 ) S~+ 2
zPlane Analysis of DiscreteTime Control Chap. 3 Example Problems and Solutions
 Tjw t 1 ,
 4, sin2(Twl2)
T e o2
= tan' Tw  To
Figure 354 shows plots of the magnitude and phase characteristics of the firstorder
hold and those of the zeroorder hold. From Figure 354 it is seen that both the
zeroorder hold and the iirstorder hold are not quite satisfactory lowpass filters. They
allow significant transmission above the Nyquist frequency, wh. = d T . It is important,
Magnitude and phase characteristics of the firstorder hold and those
therefore, that the signal be lowpassfiltered before the sampling operation so that the
of the zeroorder hold.
frequency components above the Nyquist frequency are negligible.
1  eT"
Y ( s ) = G ( s ) X *( S ) = x*( 4 where
Show that
Y * ( s )= X * ( s )
Solutian By taking the starred Laplace transform of Equation (3102), we have
For n = 2, we obtain
1  1  az' + a 2 z  2  a 3 z T 3+ .. e
Notice that [CiR (s)]*/{I+ [Gl G2N ( s ) ] * )is a series of impulses. The continuoustime
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 A318 for details of áetermining the continuoustime output ~ ( t the ) , inverse Laplace
transform of C(s).]
Consider the svstem shown in Figure 357. Obtain the closedloop pulse transfer
Obtain the discretetime output C ( z )of the closedloop control system shown in Figure function G(z)lR(z).Also, obtain the expression for @ ( S ) .
356. Also, obtain the continuoustime output C(s).
Figure 356 Discretetime control system. Figure 357 Discretetime control system.
1 rPlane Analysis of DiscreteTime Control Systems Chap. 3 Chap. 3 Example Problems and Solutions
olution From the diagram we have Compare the polar plots (frequencyresponse 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 frequencyresponse 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 frequencyresponse 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) (3104)
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 35$(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 continuoustime output C(s) can be obtained frorn the following equation: same, as shown in Figure 358(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 358(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 (3103) by varying o from O to m, as shown in Figure 359(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 (3104) by varying o from O to nlT, as shown in Figure 359(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 35 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 328, the pulse transfer function for the digital PID controller
is as given by Equation (356): was derived as
Using Pm(kT) = m(kT)  m((k  1)T) derive the velocityform P%Dcontrol equation.
zPlane Analysis of DiscreteTime hap. 3 Example Problems and
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.
= 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 ) ] (3105)
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 (3105) 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 (3105), 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 (3106)
zPlane Analysis of DiccreteTime Control Systems Chap. 3 ap. 3 Example Problems and
The velocityform PID control scheme given by Equation (3106) 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 (3106) 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 (3106) rnay be modified to
Simplifying, we obtain
Equation (3108) gives the velocityform PID control scheme. The block diagram
realization of the velocityform digital PIE) control scheme was shown in Figure 430.
Output points obtained
1.5 by sarnple calculations
Figure 360 (a) Discretetime control system; (b) plots of individual impulse
or responses; (c) plot of continuoustime output c ( t ) versus t.
Thus, c(t)
The continuoustiae output c(t) can therefore be obtained as the inverse Laplace
transform of C(s):
zPiane Analysis of DisereteTime Control Chap. 3 le Problems and Solulions
ence, Figure 360(b) shows plots of individual impulse responses given by Equation (3109).
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 (3109) 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
Draw a series realization diagram and a parallel realization diagram. (Use one first
Hence, we obtain order section and one secondorder 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 secondorder term in the numerator (which has complex
zeros) and the secondorder term in the denominator (which has complex poles).
Wence, we group G(z) as follows:
Next, we shall consider the parallel realization scheme. Expansion of G(z)/z into
partial fractions gives
A slowly changing continuoustime 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  zl)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
se 362 (a) Block diagram for "differentiation" in the z plane; (b) block
diagram for "differentiation" in the s plane.
3
Consider a transfer function system
, ~ 2
Show that the circuit shown in Figure 364 acts as a zeroorder hold.
Obtain the pulse transfer function by two different methods.
Use the residue method and the method based on the impulse response function.
where y ( k ) = O for k < O. Obtain the response y ( k ) when the input x ( k ) is a unitstep
sequence. Also, obtain the MATEAB solution.
O I C9 Obtain the weighting sequence g ( k ) of the system described by the difference equation
Figure 365 RC circuit . y(k)ay(k1)=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.)
Obtain the closedloop pulse transfer function of the system shown in Figure 368.
Obtain the response of this system to a unitstep sequence input. Also, obtain the
MATH,AB solution.
Zeroorder K

hold s+a
where x ( t ) is the unitstep function and x" ( t ) is its impulsesampled version. Assume
that the sampling period T is 0.1 sec. Discretetime control system.
Consider the discretetime control system shown in Figure 370. Obtain the output
Obtain the closedloop pulse transfer function of the system shown in Figure 367 sequence c ( k T )of the system when it is subjected to a unitstep input. Assume that the
sampling period T is 1 sec. Also, obtain the continuoustime output c(t).
Figure 367 Discretetime control system. Figure 370 Discretetime control system
zPlane Analysis of DiscreteTime Control Systems Chap. 3
31 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 371
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 363, draw a graph of the
output v(k) versus k when the input x(k) is a unitstep sequence.
Consider the system shown in Figure 373. Show that the pulse transfer function
Y(z)lX(z) is given by
Consider the system shown in Figure 372. 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 unitstep function. Also, determine the final value c(m).
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 372 Discretetime control system. Draw a graph of the output y(kT) when the input x(kT) is a unitctep sequence.
Obtain the closedloop pulse transfer function C(z)IR(z) of the digital control systern Consider the system shown in Figure 374. Show that the pulse transfer function
shown in Figure 330. 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 330 is a velocityform PPD control of the plant.)
Draw block diagrams for the filter using (1) direct programming, (2) standard prograrn
rning, and (3) ladder programming.
Consider the system shown in Figure 375. Show that the pulse transfer function
Y ( z ) / X ( z )is given by
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 uis negative in the left half of the s plane, the left half of the s plane
corresponds to
Izl = e T u < 1
23xjw 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 41 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 DiccreteTime Control ystemc by Conventional Methods Chap. 4
Hence, wd at point P is
Im
z plane
z plane
W = W,
Im z plane
Figure 411 (a) A desirable region in the s plane for closedloop 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 411(b).
Note that if the dominant closedloop poles of the continuoustime control cys
tem are required to be in the desirabie region specified in the s plane, then the domi
nant closedloop poles of the equivalent discretetime 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
discretetime 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 closedloop poIe and zero locations must be modified until satisfactory results
are obtained.
entsLI, For discretetime 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 (43), as well as of other types of
and the sampling theorem is not satisfied, then frequency folding occurs and fhe discretetime 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 continuoustime control system has closedloop 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 closedioop 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 closedloop
see Figure 412(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 412(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 closedloop pole on the unit
circle makes the system unstable.
3. Closedloop zeros do not affect the absolute stability and therefore may be
located anywhere in the z plane.
reveal the existence of any unstable roots (the roots that lie outside the unit circle
Consider the closedloop control systern shown in Figure 413. 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 openloop trancfer function G ( s ) of the system is e the effects of parameter changes on the sy ability, except for the
where a. > O. Then the Jury table becomes as given in 'Hable 41.
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:
It is important to note that in the absence of the sampler a secondorder syskem Row z0 z1 z z ... ~ n  2 Zni Zn
is always stabie. In the presence of the sampler, however, a secondorder system such
as this can becorne unstable for large values of gain. n fact, it can be dmwn that
the secondorder system shown in Figure 413 will become unstable if K > 2.3925.
(See Example 47.)
ii@. ñhree stability tests can be applied
direc (z) = O without solving for the roots. Two
of them are the SchurCohn stability test and the Jury stability test. These two tests
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 evenra.umbered row are si
irnmediately prece ing oddnumbered row.
It is noted that the value of cl (or, in the case of the nthorder 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 41, a SurY Notice that for this characteristic equation
stability table for the fourthorder system may be constructed as shown in Table 42.
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 righthand 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. 43 Stability Analysis of ClosedLoop 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 43, 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 43. (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 43 are satisfied. Sin
The conditions for stability in the Jury test for the thirdorder 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 44
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,
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 discretetime unityfeedback control system (with sampling period T = 1 T is equal to 1 sec, from Equation (42) we have
sec) whose openloop 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 closedloop 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 secondorder 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 (461, (471, and (48). Hence,
2.3925 > K > O which yields
'The range of gain constant K Eor ctability is between O and 2.3925.
iscreteTime Control ystems by Conventional Met ec. 44 Transient and
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 wnl +  + 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 unitstep input, since the unitstep input is
easy to generate and is sufficiently e useful inforrnation on both the
transient response and the steadystate response characteristics of the systern.
e transient response of a system to a uniéstep 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 timederivatives 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 A43, 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 discretetime or digital control systems because
15 (a) Unitstep response of ise time t,. The rise time is the time required for the respons
o kT the system shown in Figure 414; to 9O%, or 5% to 95%, or O% lo 100% of its
(b) discretetime output in the unitstep situation. For underdamped secondorder 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 unitstep input Peak time t,. The peak time is the time re for the response to reach the
shown in Figure 415(a). Figure 415(b) shows first peak of the overshoot.
Just as in the case of continuoustime 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 steadystat, 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 aforementioned transient response specifications in the unitstep response The timedomain specifications just given are quite important since most
are defined in what follows and are shown graphicaily in Figure 416. control systems are timedomain 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
isereteTime Control y Conventional Met ec. 44 Trancient and Steady tate Response Analysis
(413)
E;<sr the system shown in Figure 418, define Then the steadystate actuating error in response to a unitstep in
from the equation
and The steadystate actuating error in response to a unitstep input becomes zero if
& = which requires that G H ( z ) have at least one pole at z = 1.
Static Velocity Error Constant. For a unitramp input r(t) = t I ( t ) , we have
Then we have
Tz'
( z ) = ( ~  1 2
2 1
y substituting this last equation into Equation (4131, we have
and
K, = lim (416)
z+1 T
Then the steadystate actuating error in response to a unitramp input can be
given by
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 420(b), the closedloop 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 419 (a) Digital closedloop 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 DiscreteTime ysterns by Conventional Met ec. 45 Design
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 closedloop 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 openloop poles and those cor
MAGNITUDE CONDITION : ence, as K is increased from zero to infinity, a root locaas starts
from an openloop pole and terminates at a finite openloop zero
IF(z)l = 1
zero at infinity. This rneans that a rootlocus 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 closed1 e same as the number of o
is the root locus. The r e characteristic equation (the closedloop 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 openloop 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 closedloop poles are presented in the following. asymptotes.
. Determine the root loci on the real axis. Root ljoci on the real axis are
determined by openloop poles and zeros lying on it. The conjugate compliex poles
cated, but actually it is not difficult if the and zeros of the openloop 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 trialanderror 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 openloop 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 openloop zero and one openloop 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 openloop poles and openloop 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)
piioceduresfor constrclcting root loci.
L.,
Angle of asymptote = , N = 0 , 1 , 2,...
nm
where (z)/A(z) = F ( z ) , then
n = number of finite poles of F(z) K = A ( 4 (422)
m = number of b i t e zeros of F(z) Nz)
and the breakaway and breakin points (which correspond to multiple roots) can be
determined from the roots of

~ + p2
z ~ +[(pl + + pn)  (z1 + Z2 + . + z , ) ] z ~  ~ +~ .
ne the angle of departure (or angle of arrival) of the root loci from
nm
Angle of departure =
480"  ( 8 , 3 8,  4)
tbus found give the location at which the root loci cross the imaginary axis an Note, however, that if polezers canmcellatáon occurs in the feedforward pu
e value of the corresponding gain transfer function, then the same ezero cancellation occurs in the cl
. Any point on the root loci is a possible clos pulse transfer function. Consi hown in Figure 421,
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 polezero cancellation occurs in GD(z)Gl(z). Por exa
The root locus plot of G(z)N(z) wi1Í not show al1 the roots of tfie charact transfer function (such as polezera cancellation in the
equation, but only the roots of the reduced equation. ant). In the forrner, the canceled of the closedloop system
To obtain the complete set of closedloop poles, we must add tfie cance ereas in the latter the canceled po
or polies of G(z)H(z) to those closedloop 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 closedloop 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 421 are given by the closedloop control system. C own in Figure 423. 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 closedloop 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 closedloop pulse transfer function of the svstem is Integral
controller
lrn
z plane

( = 0.5 sec)
 1  eT

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 (425) 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 breakin point. Not
pulse transfer function with two poles and one zero results in a circ
centered at the zero. The breakaway point and breakin point are Im
writing the characteristic equation in the form of Equation (422), z plane
, = 
(z  1)(z  0.6065)
0.3935~
í T = 2 sec)
For the present case, T = 0.5 and this Iast equation becomes
or
K = 8.165
The critical gain M, is thus 8.3 65.
The dosedloop poles correspon
zl = 0.4098 + j0.6623 and z2 = 0.4098  j0.6623
These closedloop poles are indicated by dots in the root locus diagram.
Sampling period 1' = I sec: r this case, Equatiorl (42 5 ) beca
follows: closedloop pole is i , then in the s plane the closedloop pule location (in the upper
halfplane) 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 breakin 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 424(b) shóws the root locus diagram when T = 1 sec. T
z plane
of gain K is 4.328. The closedloop 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 (425) becomes
and
& = TU. = To, = 8 (rad)
For T = 0.5 sec and response can be given by
From Equations (429) and (430), 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 426(a).
for the exponent, we find From Figure 425 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
=
Y
lim 
Z+I
0.78702
0.52 (z  l)(z  0.6065) 1
Thus, the steadystate error in res
es, = 
KJ = 4
'= 0.25
irnilarky, for the case where T = 1sec and = 2, the openloop pulse transfes
function is
Consider the digital control system shown in Figure 428. In the z plane, design a digital
controller such that the dominant closedloop 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 unitstep input. Also, obtain
the static velocity error constant 6(, of the system.
For the standard secondorder system having a pair of dominant closedloop
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 closedloop poles as
wn = 4
The damped natural frequency wd is determined to be
wd = con = 4 = 3.464
[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 closedloop poles in the z plane.
Weferring to Equations (429) and (430), for a constantdampingratio 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 closedloop pole in the upper half of the z plane are determined as follows:
Figure 427 (a) Unitramp response sequence of the system shown in Figure 423
when T = 0.5 sec and K = 2; (b) unitramp response sequence when T = 1 sec and
K = 2; (c) unitrarnp 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
chedloop 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 428 Digital control systern for Example 49.
ased on the WootLocus Methad
and
3 464
& = 27.r31.42 = 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 closedloop 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 429 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 zeroorder hold can be obtained as follows:
The openloop pulse transfer function now becomes
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 429. If point P is to be the location for the
dominant closedloop 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
r plane The openloop pulse transfer function for the present system is
The response to the unitstep input R(z) = 1/(1  zl) can be obtained from
Figure 430 shows the unitstep 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
@re429 Root locus diagram of the system considered in Example 49. digital controller just designed catisfies the given specifications and is satisfactory.
ecign Gased on the Frequency
between O and 1 i
Similarly, in the s plane a closedloop poPe on the negative real axis neai: the
origin increases the settling time. In the z plane, su
to a closedloop 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 closedloop 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)
The static velocity error constant M, of the system is given by The freqeaencyresponse concept plays the same powerful role in digital control
systems as it does in continuoustime control systems. s stated earlier, it is
assumed in this book that the reader is familiar with conventional frequency
se design techniques for continuouctime control systems. In fact, familiarity
= 2.801
1 de diagrams (logarithmic plots) is necessary in the extension of the comen
tional frequencyresponse techni ues to the analysis and design of
control systems.
Frequencyresponse 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
frequencyresponse tests on a dáscretetime 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 lowpass 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 timeinvariant system to a sinuso
change the root locus near the dominant closedloop 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 timeinvariant
point of disturbance or noise attenuation, since the response to disturbances would not discretetime 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 discretetime control systern in the w plane by use of a
as a secondorder 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 closedloop poles. The
dominant closedloop poles are the only closedloop 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 timeinvariant discretetime system shown in Fig
frequency response behaviors, however, are stro 431. The input to the system G ( z ) before sampling is
the c~o~edloop pulse irancfer function. u (í) = sin wt
Design of DiscreteTime Control Systems by Conventional Methods Chap. ased on the FrequencyResponce Method
Stable linear
timeinvariant
discretetime
Let us substitule eJwTforz in G(z). Then the sinusoidal pulse transfer hnction G(e1"3
can be obtained as
1
G'(eJ"*) = 1  aejmT
 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 steadystate 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 welldeveloped frequencyresponse methods to the ana lane, z varies Pom 1 to  1 along
discretetime 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 frequencyresponse 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 frequencyresponse techniques can then be used in the w plane, and
SO the wellestablished frequencyresponse design tec niqUes can be applied to the
where T is the sampling period involved in the discretetime control system und design of discretetime 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 frequencyresponse methods can be extended to discret onal frequencysesponse techniques may be used to
time control systerns. y solving Equation (437) 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 432. (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 (439) 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 Transferfunction system of Example 411.
uency v moves from co to O, and as w moves from O to w,, v S
Referring to Equation 4439), the actual freauencv w can he traaacii~t~d
k f r th
~
Consider the transferfunction system shown in Figure 434. 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 = 2lF
approaches infinity in the right half of the w plane. Wote that we have
1o
lim G(w) = iim 
w+o sOS + 10
This fact is very useful in checking the numerical calculations in transforming G ( S ) into
G(w)
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 polezero 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
straightline 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 lowfrequency asymptote of the magnitude curve is
putsinglesutput 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
stsaightline 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 laglead compensation techniques.
magnitudes for C (jo) and G( jv). The highfrequency asymptote of Phase lead compensation is cornmonly used for irnproving stability margins.
magnitude curve for G(ju) rnay be a constantdecibel 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 highfrequency noise problems ue to its increased high
411, 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 highfrequeracy magnitude of G(jv) is frequency gain, the total system gain can be increased, and thereby lowfrequency
gain can be increased and the steadystate accuracy can be improved. Also, any
bighfrequency noises involved in the system can be attenuated.
In some applications, a phase lag compensator is cascaded with a phase lead
whlle the highfrequency magnitude of the plant is compensator. The cascaded compensator is known as aphase laglead compensator.
1
lirn j o
10
+ PO =O
By use of the laglead compensator, the lowfrequency gain can be increased (which
means an improvement fn steadystate accuracy), while al the same time the system
bandwith and stability margins can be increased.
The difference in the Bode diagrams at the highfrequency end can be explained a Note that the PID controller is a special case of a phase laglead 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 highfrecpency region, increases the phase
techniques for PID controllers basically follow those of phase laglead compe
tors. (lín industrial control systerns, however, each PID control action in the
controller may be adjusted experimentally.) Figure 435 Digital control system.
That is,
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 (437):
1 + (TJ2)w esign procedure just given, it Is important to note the
z =
following:
1  (7/2)w
Design of DiscreteTime Control ystems by Conventional Metho ased on the Frequency
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 lowfrequency range and has the following form:
Consider the digital control system shown in Figure 436. 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 secl. Assume that the sampling peri
0.2 sec, or T = 0.2.
(This is a phaselead 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 zeroorder hold:
The openloop transfer function is
z =:
1 + (T12)w 
1 + 0 . 1 ~
1  (2'12)~ 1  0 . 1 ~
Figure 437 shows the Bode diagram for the system. For the magnitude curves
we have used straightline 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 phaselead 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 436 Digital control system of Example 412. must corrtribute the required phacelead angle.
ased on the FrequencyResponse 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 magnitude and phase angle curves for G D ( j v )and the magnitude and phase angle
curves of the compensated openloop transfer function G D ( j v ) G ( j v )are shown by solid
curves in Figure 437. 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 (440) will now be trans
formed back to the z plane by the bilinear transformation given by Equation (438):
Thus,
O2 0.8  O
0.6  o
poles acts as dominant closedloop poles. (Tke system behaves as if it is a secsn
0.4 
o
unitstep response of this sy
the unitstep response curve 0.2 
exhibits a mLurimum oversh o
06
O 5 10 15 20 25 30 35 40
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
timedomain 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 zeroorder hold, and output of the hold, u(t), which
Lf the response of a closedloop control system to a step input e is a piecewise continuoustime 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 zeroorder hold is not shown, the signal u(k2")is first sampled and
mum time and stays there), no steadystate error, and no ripples bet fed to the zeroorder hold. As mentioned earlier, the zeroorder hold shown in the
sampling instants, then this type of response is cornmonly iagram is a sarnpleandhold esired to design a
The deadbeat response will be discussed in this section. ( GD(z) such that the closedloop 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 steadystate 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 singleinputsin 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 zeroorder
of multipleinputmultipleoutputsystems, see Chapter 8, where the statespa frold as G ( z ) , or
approach is used.
Design of Digital CesntroUersfeso Minimu
Error. Consider the digitai control system shown in Figure 439(a). The erro Then the openloop 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 439(b). Next, define the desired closedloop pulse transfer function as F ( z ) :
iscreteTime Control Systerns by Conventional
F(z) =
a o z N + a l z N  '+ + aN at will happen to the closedloop 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 unstableor
E(z) = a. + alzl + +aN~N define
hat F(z) must not conta G(z) = 
GiW
series expansion of F(z) za
e for a physically real where Gl(z) does not include a term that cancels with z  a . Then the closedloo
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 (441). Solving Equation (441) foa:GD(z) Gl(zb
G D ( ~z_,
)
we obtain G D ( ~ ) G ( ~=
) = F(z) (444)
+ 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 closedloop pulse transfer funct
and the digital controHer pulse transfer function GD(z).The conditions for 1  za
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 eL" 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 closedloop 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 zl, the lowestpower 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 z1 begins with %hezl term, at lie on or outside the unit circle must not
first term of F(z) given by Equation (442) 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) = alzl + a2z2+ + 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 (441) 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 zl. Note that for a unitstep input r(t) = l(t)
esign of DiscreteTime Control Cysternc by Convenlional
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 steadystate 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 (447), 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 timed
2  F(z) = (1  ~  l ) q + ~ A J ( z )
inputs, ramp inputs, and acceleration inputs), th
where N(z) is a polynomlal in z1 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 (448) into Equation (44
ciently time varying inputs (such as frequencydomain 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 (450) gives the pulse transfer function of the digital controller tkat Consider the digital control systern shown in Figure 439(a), where the plant transfer
produce zero steadystate 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 DiscreteTime Control yctemc by Conventional Methodc Chap. 4 7 Analytical Design Method
and
and
f i = 0.5820
The digital controller pulse transfer function G D ( z )is tben determined from Equatio
(450), as follows. By referring to Equations (451), (454), and (460),
Hence,
Figure 440 Responses of the system designed in Example 413. (a) Plots of c ( k )
versus k , u ( k ) versus k , and u ( t ) versus t in the unitstep response; (b) plots of c ( k )
versus k , u ( k ) versus k , and u ( t ) versus t in the unitramp response.
The signal u ( k ) becomes constant ( b = 1) for k 1 2. Hence, the cystem output will not
exhibit intersampling ripples. Figure 440(b) shows plots of c ( k ) versus k , u ( k ) versus
k , and u ( t ) versus t in the unitramp response.
Note that the static velocity error constant Kv for the present system is
0.58202' + 0.4180z'
= lim =
Z1 1 + 0.41802'
esign of Discretelime Control ystemc by Conventional Methods Ckap. ec. 47 Analytical Design Meahod
Thus, the steadystate error in the unitramp response is (If a catisfactory result is not obtained, we must assume N > 3.) Since the input is a
step function, from Equation (448) we require that
1  n(z) = (1  Z')N(z) (462)
which is indicated in Figure 440(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  zl and therefore satisfies the stability requirement.
The requirement that the static velocity error constant be 4 secl can be written
as follows:
where we used Equation (450) with q = O. Notice that from Gquation (462) 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 439(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 firstdegree polynomial in zl. Then U(z) can be written as follows:
 ( 1 + 0 . 7 1 8 1 ~  ~ ) ~  ' ( 1 . 2 6 1
8 40.679802')
+07181z')2'
( 1  21)(1  0 . 2 6 1 ~ 4 ~ 1 0.48817z2)
( 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
unitstep input r ( t ) = 1 is obtained as follows:
C(z)= F(z)R(z)
Figure 441 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
unitstep response of the system de
= 1.2618~' + 1 . 4 8 8 2 ~+~Y~ 3+ z  ~+ . signed in Example 414.
esign of DiscreteTime Control ystems by Conventional [Vethods
[Vethod: Chap.
F ( z ) qz) =F(z)  1 
U(Z) = = 
2'
zl=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.
o
X(Z) 1  0.62'  0.81z' + 0.6723  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+a3zta4, ao>O
The sampling period T is 0.5 sec. Using , plot the unitramp sesponse up to
k = 20.
slaritioarn The unitramp input u may be written as
u=kT, k = 0 , 1 , 2 ,...
In the MATLAB program, this input can be given as
where A ( z ) and B ( z ) do not contain K, then the breakaway and breakin 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 .
iscreterime Control Systems by Convenlional 4 Exarnple Problems and Colutions "2
This means that multiple roots of f ( z ) will satisfy the following equation:
where
UnitRamp Response
1O
This particular value of K will yield rnultiple roots of the characteristic equation. li we
substitute this value of K into Equation (472), 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 (472) we have
and
If dK1dz is set equal to zero, we get the same equation as Equation (474). Therefore,
k
the breakaway or breakin points can be simply determined from the roots of
Figure 443 Unitramp response of the system considered in Problem 444.
It should be noted that not al1 the solutions of Equation (474) or of dKidz = O cor
respond to actual breakaway or breakin points. Cuch a point for which dKldz = O is
hap. 4 Exarnple Problerns and Solutions
an actual breakaway or breakin 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 rootlocus method.
Next we construct root locus diagrams for the three cases considered.
dominant closedloop poles. l. Sampling period T = 1: For T = 1, Equation (475) becomes
K[(1  1 + e')z' + (1  e'  e  ' ) ~  ~ ]
G(z) =
(1  zl)(l  elz')
4. This angle must be contributed by the lead compensator if the new
is to pass through the desired locations for the dominant closedloop 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 + a7z ' z = 0.3679. The breakaway point is at z = 0.6479, and the breakin point is at
If static error constants are not specified, determine the location of t z = 2.0841. The rootlocus diagram for this case is shown in Figure 446(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 frequencyresponse appr
If we choose a point z on the root loci, the value of K at that point can be
5. Determine the openloop 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 closedloop 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 (475)
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 445 f breakin point is at z = 1.5247. The rootiocus diagram for this case is shown
following three sarnpling periods: T = 1 sec, T = 2 sec, and T = 4 sec. in Figure 446(b). The critical gain IC for stability is Ik = 1.4557.
iscreteTime Control ha p.
z plane The rootlocus diagram is shown in Figure 446(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 447, 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 closedloop 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 unitstep input. Also, obtain
the static velocity error constant Kv and find the steadystate error in the response to
a unitramp input.
\ I z plane
I
47 Digital control system.
Soltiiition The pulse transfer fianction of the plant that Is preceded by a zeroorder
hoId is
z plane
Unit circle
\ /'
6 Rootlocus diagrams for the system shown in Figure 445 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 (475) gives The openloop pulse transfer function becomes
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 closedloop 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,
We locate the openloop poles in the z plane as shown in Figure 448. There is on Thus,
openloop 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 closedloop pole in the upper half of the z plane must lie on a 1 From Equations (476) and (477), we find that
from the origin having an angle of 360°/10 = 36". Frorn Equations (41) and (4 K, = 0.2982 and K, = 0.2088
rewritten as
Hence, the PI controller just designed can be given by
the desired closedloop pole location can be determined as follows. Noting tha
Digital
controller S(S + 1)
% U n itstep response 
title('UnitStep Response') We shall first locate the desired closedloop poles in the z plane. Referring
xlabel('k'f to Equations (41) and (42), for a constantdampingratio locus we have
yIabel('c(k)')
Since we require eight samples per cycle of damped sinusoidal oscillation, the dominant
closedloop 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 451. (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 A48.
(Sampling period T = 1 sec.)
y letting 5 = 0.5 and substituting Equation (479) into Equation (478), we obtain
IZI = e0.4535 = 0.6354
Corisider the system shown in Figure 450.
that the dominant closedIoop poles of the system will have a damping ratho 5 Hence, we can locate the desired closedloop 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 451. Note that at point P
be 8. Assurne that the sampling period T is 0.2 sec.
Wsing the rootlocus 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 zeroorder hold:
.4 Exarnpie Problems and 71
Design of DiscreteTime 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 openloop pulse transfer function of the system is therefore obtained as follows:
Hence, we have determined the pulse transfer function of the digital controller to be
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 closedloop poles.
Figure 452 shows the unitstep 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'
z1
= lim
,i [
1  2' 0.2610(1 + 0.93562')z'
0.2 ( 1  0.15952')(1  z  ' )
Chap. 4 Exarnple Problerns and
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:
1a
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:
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 polezero Iocation of the compensator until a satisfactory result
is obtained.
z plane
the rootlocus 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 closedloop poles is approx
imately 0.55.
The line connecting the closedloop 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 457> 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 unitstep input.
shall first obtain the z transform of the plant that is preceded by the
zeroorder hold:
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 457 Digital control system. 1L
4, = sinl6 =. 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 458 shows the e diagram of G ( j v ) thus obtained. Wotice t
margin is 2". ft is nec
Thus, the static velocity error constant K,, is infinity. There is no steadystate error in
the ramp response.
The closedloop pulse transfer function of the system is
v (radlsec)
Figure 459 shows the unitstep response. Notice that the zero of the digital controller
at z = 0.9048 is close to the double pole at z = 1. A polezero pair near point z = 1
gure 458 Bode diagrarn for the system considered rn Problem A412. creates a long tail with srnall amplitude in the response.
Chap. 4 Example Problerns and
Figure 459 Plot of c ( k T ) versus kT for the system designed in Problem A412.
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 closedloop pulse transfer function as F(z):
In the present case the output c ( t ) in the unitstep response is specified as shown in
Figure 461(a). Since h[l  e0.1(155)]= h ( l  e') = 1, we have h = 1.5820. from
the deadbeat response curve shown in Figure 461(a), we obtain
c(0) o
=
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 461(b).
from which we get
Consider the digital control system shown in Figure 462. Design a digital controller
G D ( z )such that the closedloop system will exhibit the minimum settling time with zero
steadystate error in a unitramp 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 unitstep input.
Soliution The first step in the design is to determine the z transform of the plant that
is preceded by the zeroorder 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 (480):
Now define the closedloop pulse transfer function as E ( z ) :
GD(z) =
G ( z ) P  F(z)l
Notice that from Equation (448) 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 (448) we require that
This last equation gives the pulse transfer function of the digital controller. Since
must be unity at steady state, u ( t ) , a continuoustime 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
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 (486)
U ( Z )= bo + b l 2' + . . . + b ~  l y solving Equations (484), (485), and (486) for al, a,, and a', we obtain
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 (450):
(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 A414,
Figure 464 (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 iritefsamp the ICronecírer delta input of the system designed in Problem A414; (b) plots of c ( k ) versus
ripples after r 3 4T = 4, k , u ( k ) versus k, and u ( t ) versus t in the unitstep response of the same system.
Design of DiscreteTime Contr
Solve Problem B44 by using the bilinear transforrnation coupled with the Wouth
stability criterion.
the regions in the s plane shown in Figures 465(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 timeinvariant discretetime control system,
Consider the digital control system shown in Figure 466. 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
closedloop 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.
Digital control system for Problem B49. 70 Digital control system for Problem B412.
Figure 467
Weferring to the system considered in Problem 849, redesign the digital controller
so that the static velocity error constant K , is 12 sec', without appreciably changing
the locations of the dominant closedloop 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
unitstep response and unitramp response of the digital control system.
Consider the digital control system shown in Figure 471. Draw a Bode diagram in the
w plane. Set the gain K so that the phase margin becomes equal to 50'.
Figure 468 Digital control system for Problem B410. 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 closedloop 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 B414.
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
secl. 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.
In Ghapters 3 and 4 we were concerned wieh conventional methods for the analysis
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) = nwector (state vectsr)
e knowledge of the input for t $: to, c
system for any time t 2 to. Note that the concept of state is by no y(k) = mvector
%tis applicable to biological systems, eeonomic syste
u(k) = rvector
G(k) = n x n matrix (state matrix)
maki H(k) = n x r matrix (input matrix)
system. Ef at least n variables xl,xs, . . . ,x, ar
atrix (output matrix)
behavior of a dynamic system (so that once the
state at t = to is spe D ( k ) = m x r matrix (direct transmission matrix)
then such n variabl
arance of the variable k in the arguments of m
Note that state variables need not be
implies that these matrices are time vasying. If
quantities. Variables that do not represenh
not appear explicitly in the matrices, they are
neither measurable
constant. That is, if the system is time invariant,
in choosing state variables is an advantage o
speaking, however, it is convenient to choos
state variables, if this is possible at all, becau w(k + 1) = Gx(k) + Hu(k) (51)
feedback of al1 state variables with suitable
Vee$or, H f n state variables are needed to com
As in the discretetime case, continuoustime (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 ndimensional space whose coordinate axes consist of the
For linear timevarying continuoustime 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, %nstatespace 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
statespace representation for a given systern is not unique, except that the numb
of state variables is the same for any of the different statespace representations
the same system.
For timevarying (linear or nonlinear) discretetime systems, the state e q d
may be written as Figure 5l(a) shows the block diagram representation of the discreteti
system defined by Equations (51) and (52), and Figure 5l(b) shows the contin
x(k + 1) = f [x(k), un(k), k] UOUStime control system defined by Equations (53) (54). Notice that the basic
and the output equation as coarfigurations of the diseretetime and continusus systems are the same.
iscrefreTime Systems
Y(z) bo + bl zl
=:
+ .  + b, z"
~ ( z ) 1 + alz' + .   + anz" (56)
Y(z)
=
bozn + bl znl + a . + bn
U(Z) zn + al zn1 +  . + a, (57)
(b)
following representations:
hod. (See Problem A51.) The observable canonical form can be obtained by
tions of linear time time systems. Section 53 first tr nested programming rnethod. (See Problem A52.) The diagonal canonical
orm and Jordan canonical form may be obtained by use of the partialfraction
procedure and by the z transform approach. Then it presents a method for com
ing (zI  G)'. Section 53 concludes with discussions of the solcltdon of the li
timevarying discretetime state equation. Section 54 deals with the pulse tran
fmction matrix. Section 55 first treats the discretization of linear continasouse
statespace equations. Then it discusses time response between two concecu by the fdowing eqimations:
StateSpace Analysic ec. 52 StateSpace epresentationc of DíscreteTime Systems
O O   S O ~,ik)~ b~
r X,W i
ence,
transformation.
Gonsider the system defined by
is a nonsingular matrix. [Note that, since both x(k) an k(k) are ndimen
sional vectors, they are related to each other by
Then, by substituting Equation (522) into ation (52O), we obtain
et us define
The statespace representations in the controllable canonical form, observable can
ical form, and diagonal canonical form become as follows:
CONTROLLABLE CANONICAL FORM:
uation (524) can be written as follows:
y defining
DIAGONAL CANONICAL FORM: have thus shown that the statespace representation given by Equations (520)
The given pulse trancfer hnction Y ( z ) I U ( z )can be expanded as follows: "nd (521),
n n n
V V C
W W W
ec. 53 Solving
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 DiscreteTime S
a, = trG = tr
where
Hence
W,
=
=
GWn2
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 (547). The si's can also be given (see Problern A513) by use of th = 4 tr 0.09 0.01 0.02 = 0.04
trace, as follows: [:.o7
ee. 53 Solving DisereteTime StateSpace Equations
k j +l ) ( j ) ( ) k >h (555)
j= h
Notice that the first term on the righthand side of Equation (555) 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 (555) can be verified easily. Referring to Equation (554), we have
j= h
meSpace Analycic Ch ec. 54 PulseTransferFunction Matrix
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 nsferfunction matrix .
. A sumrnary on the state transition matrix transfer function rnatrix z) characterizes the in
~=iej,
= t
The pulsetransferfunction matrix F(z) for the system
and (562) is is can be proved as follows:
In particular, if s =  t , then
eAteAt = eAt eAt = eA(tt) =
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 continuoustime state e
Thus, ihe eigenvalues of G are invariant under similarity transformation.
where x is the state ctor (nvector), ia the input vector (rvector),
In digital control of continuoustime plants, we need to convert and premultiplying both sides of this last equation by e"', we obtain
statespace equations into discretetime statespace equations,
can be done by introducing fictitious samplers and fictitious hol
continuoustime 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 ContinuousTime State
where h = T  t .
shall present a procedure for discr
assume that the input vector n(t)
Note that the sampling operation
Note that the matrices G and H depend on the sampling perio atrix A is nonsingular, t T ) given by Equation (574) can be simpli
are constant matrices. fied to
To determine G ( T ) and
Now we discretize the state equation and the output equation. Referring to Eqjuatio
(573) and (574), we have
G(T) = eaT 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 discretetime state equation and the
output equation become, respectively,
1  eaT
x(k + 1) = eOTx(k) + 
a u@)
Eea
Obtaim the dascretetime state and output equations and the pulse transfer hnct
(when the sarnpling period T = 1) of the following continuoustime 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 zeroorder hold. Assuming
T = 1, we obtain
  F .
equation
into
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. 56 Liapunov Stability Analycis li
Consider the system discussed in Example 55. Obtain the discretetime state equation
4, the z transform solution of the discretetime 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 55, the matrices G ( T ) and H(T) were obtained as follows:
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 zeroorder hold G(ibT) and M(dT) into Equations (580) and (5811, as follows:
u ( ~ )= u ( k T )for kT T < k T + T. Referring to Equation (5651, 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(~)
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 RouthHurwitz stability criteria,
puter calcu'tation.) Y are limited to linear timeinvariant systerns. The second method of Liaprtnzov,
ec. 56 Liapunov Stability Analysis
negat
then in the neighborhood of the origin the Iowestdegree 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, . . . ,inl,
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 §taliility, and instability.
pace Analysis Chap. ec. 56 Liapunov Stability Analysis
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 twodimensional case. Figures 52(a), (b), and (c) show
equilibrium states and typical trajectories corresponding to stability, asymptotic
I/x  = [ ( x ~ le)' + ( ~ 2 ~ 2 , +) ~ . + (xn  x,,)~]~"
e
w = f(x, t )
where
systems, is algebraic and does not require factoring of the characteristic polynomial,
einvariant 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 timeinvariant 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:
includes this equilibrium state, it does not necessarily mean that the where x is a state vector (an nvector) 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 timeinvariant system can be investi
with the required properties always exists. gated easily with tke second method of Liapunov.
For the system defined by Equation (584), let us choose as a possible Liapunov
function
proaches to the investigation of the asyrnptotic stability of linear timei
systems. For exarnple, for a continuoustime 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
r 7
'
p11 Pl2
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
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 54 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
~tineroustimesysterns, several important rernarks rnay be rnade. asymptotically stable.
tateSpace Analysis ec. 56 Liapunov Stability Analysis
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
ere x is an nvector and (x) is also an nvector. Assurne that (x) is a contraction
Then the origin of the system of Equation (591) is
asymptotically stabie in the large, and one of its LPa
V(x) = 11d1
stability of an equilibrium state of a discretetime system obtained by
This can be seen as follows. Since V(x) = llxll is ositdve definite and
continuoustime systern is equivalent to that of the original continuous
Consider a continuoustime system
k = Ax is negative definite beca is a contraction for al1 w, we fin
and the corresponding discretetime system Liapunov function, and rem 55 the origin of the system
stable in the large. (See
x((k + 1)T) = Gx(kT)
where
G =
Consider the following system:
Ef the continuoustime 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 (590), 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 (592) 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 A515.
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
= [; $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 + alzl + a2z2 + + 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>
From this last equation the following two equations may be obtained:
Q ( ~=) al;' Q ( z )  a2z2Q ( 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 discretetime system defined by
Y
 ( z )  bozn + b, zn' + + bn
U ( Z ) Z" + alzn1 + .  + a,
Show that a statespace representation of this system may be given by
Xn'(2) = z Q
~ (z)
X n ( z ) = 2l Q ( z )
Then clearly we have
z X * ( z )= X2(z)
zXz(z) = &(z)
StateSpace 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 (5100) into Equation (598), we obtain + Z'[b3 ~ ( z)a 3 ~ ( z+) . 11) (5106)
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) = zl[bZU ( Z ) a, Y ( z ) + X n V 2 ( z ) ]
'
Also, Equation (599) can be rewritten as follows: (5107)
Y(Z) = (b1  al bo)X,(z) + (bi  a,bo)Xnl(z)+ . . + ( b ,  c. bo)X1(r) &(z) = ~  l [ b , U~( z )  Y ( z )+ Xi(z)]
By use of this last equation, Equation (597) can be written in the form X l ( z ) = z'[b, U ( z )  a, Y ( z ) ]
y ( k ) = (b,  a, bo)xi(k) + (bni  a,, bo)xz(k) Then Equation (5106) can be written in the form
+ ... +(biaibo)x,(k)+bou(k) (5 Y ( z ) = boU(z) + & ( z ) (5108)
Combining Equations (5101) and (5102) results in the state equation giv By substituting Equation (5108) into Equation (5107) and multiplying both sides of
Equation (593). The output equation, Equation (5103), can be rewritten in the the equations by z , we obtain
given by Equation (594).
~ 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 statecpace 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)
bni  6,1 bo
+ Also, the inverse z transform of Equation (5108) 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 vectormatrix form
x1 ( k ) gives Equations (5104) and (5105), irespectively.
x2W
y ( k ) = [O O O 11 !
x, 1 ( k ) (Partialfractionexpansion programming method) Consider the pulse transfer func
 xn(k) tion system given by
StateSpaw Analycis Ch Chap. 5 Exarnple Problerns a n d
b o z n + bi zn' +    + bn
Y=( z )
Also, Equation (5113) can be written as
U ( Z ) zn + al zn' + . . . + a, Y ( z ) =: bo U ( Z )+ C ~ X , ( Z+) Q X ~ ( Z+) + cnXn(z) (5116)
Show that the state equation and output equation can be given in the following The inverse z transforms of Equations (5115) and (5116) 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 ) (5117)
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 ) (5113)
Rewriting the state equation and the output equation in the form of vectormatrix
xn (k)
equations, we obtain Equations (5109) and (5110).
Solation The system puise transfer function can be modified as follows:
Y
 (z) buzn + bl 2"' +
+ bn e e
(Partialfractionexpansion prograrnming method) Consider the pulse transfer func
U(2) ~"+a~z"l+..+a, tion system defined by
(bl  al bo)znl + (b2  a2bo)zn' + .  + (bn  anbo)
,
( z )  bozn + bl znl +
= bo r Y + b,
( 2  pí)(z  pz) (z  pn) U ( Z ) zn + al zn1 +  + 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
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 (5114) can be rewritten as = bo + (bl  al bo)znl + (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
Using the nested programming method (refer to Problem A52), 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 (5123) are related each to th
next by the following equations:
Figure 53 shows the block diagram for the systern defined by the statespace equations.
The output of each delay elernent constitutes a state variable.
Obtain a statespace representation of the system shown in Figure 54. The sampling
period T i s 1 sec.
ap. 5 Example Problems and Solutions
Figure 55 Modified block diagram for the system shown in Figure 54.
Figure 53 Block diagram for the system considered in Problem A55.
+ 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 statespace representation of the following pulsetransferfunction system:
diagram modification. Use the partialfractionexpansion 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 partialfractionexpansion programming method pre
Figure 55 shows the block diagram for the system. Let us choose the output ~f e sented in Section 52. 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 55. Then we obtain partial fractions as follows:
Consequently, we have the statespace equations as follows:
Then we have
Then the state variables X3(z),&(z), and &(z) are related to Y(z), zY(z
as follows: Obtain a statespace representation of the followingpulsetransferfunction 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 (512$), we obtain
Solutisn Let us first divide the nurnerators of the righthand 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
[ =
Rewriting, we have O O x1(k)
~3(+
k 1) 2O 30 ] [x3(k)
x2(k)]+[~]u(k)
Figure 57 shows the block diagram for the present system.
Then we have
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
where
(5133)
Then the coefficient of the highestdegree term in A of $(A) is unity. From Equations
(5132) and (5133), we have
Note also that
Wence, we obtain
Referring to Problem A53, 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 leastdegree 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 (5133), 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 highestde 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.
StateSpace 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 highestdegree 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
A2 1
A\ O
O
A2
3
y
A1
¡=(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
(A2)fA1) (A+11) 4(A2) 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
[: : :]
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
which is Equation (5134). 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
$(A) = h3 + 3h2  7A  17  z2G+Z ~ Wz G~H ~ + z H 2  GHz
y identifying the coefficients ai of the minimal polynomial (which is the sam
characteristic polynomial in this case), we have
al = 3, a, = 7, a3 = 17
The inverse of W can then be obtained from Equation (5134) as follows:
:I3[n
1
+ a l A + a,I) = (A2
17
+3 The CayleyHamilton theorem (see Problem A59) states that an n x n matrix G
7 O 4 satisfies its own characteristic equation. Since n = 3 in the present case, the character
+ .
istic equation is
6 1 = z3 + a 1 z 2 + a2z + a3 = O
and G satisfies the following equation:
=A[! ;; ]
6 4
G3 + al gy2 + a 2 G 4 as
[$ 8 ];


17
6. 4.

17
L.
17
Hence Equation (5136) simplifies to
Consequently,
z2 + H1z + H2) = (z3 + a l z 2 + a 2 z C a3)I = jz
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
Now we have
Consider the following oscillator system:
tr G = tr TBT' = tr B
trGH1 = trG(G + alI) = trG2 + tralG
StateSpace Analysis Chap. 5 Example Problems and
Qbtain the continuoustime statespace representation of the system. Then Noting that D is zero, we have
the system and obtain the discretetime statespace representation. Also o
pulse transfer function of the discretized system.
z
s i n w~ 1l
 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 continuoustime statespace representation of the syst
m
U ( z1
= F ( r ) = ( 1  cos O J T ) (+~ 2')z'
1  221 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 zeroorder hold. That is,
Consider the system shown in Figure 58(a). This system involves complex poles. Tt is
stable but not asymptotically stable in the sense of Liapunov. Figure 58(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
xl(kT)
Y = 1 o~[x2(ki)]
The pulse transfer function of the discretized system can be obta
Equation (560):
Figure 58 (a) Continuoustime system of Problem A515; (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 58(b) is
discretized version of the continuoustime system. The discretized systern is also sta
but not asymptotically stable.
Assuming a unitstep input, show that the discretized system may exhibit hid
oscillations when the sampling period T assumes a certain value.
&y =# 7
e" " 
= (1  z')+&]
 
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 nnsec ( n = 1,2,3, . . . ). For example,
versus r is shown in Figure 59(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 59(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 59(c).
Notice that if T = n sec (in fact, if T = nrr sec, where n = 1,2,3, . . . ) the
unitstep response sequence stays at unity. Such a response may give us an irnpression
Viel mehr als nur Dokumente.
Entdecken, was Scribd alles zu bieten hat, inklusive Bücher und Hörbücher von großen Verlagen.
Jederzeit kündbar.