Eliahu Ibrahim Jury - Theory and Application of The Z-Transform Method (1973)

THEORY AND APPLICATION
OF THE z-TRANSFORM METHOD

I~. I. J URY, DEPARTMENT OF ELECTRICAL ENGINEERING,
IINIVERSITY OF CALIFORNIA, BERKELEY, CALIFORNIA
I{(>BERT E. KRIEGER PUBLISHING CO.

IIUNTINGTON, NEW YORK
PREFACE
I luring the curly Ilfties. considerable interest and activity arose umong
C'lIl!illeCrS und systems theorists in the relatively new area of discrete
,y,ICIlI theory, This activity and aroused interest were mainly motivated
h)' Ihe adv<lnccmcnt made in digital computer technology and its wide-
',pn'ad application in systcm analysis and design, Several methods of
,llIalysis of discrete systems were proposed and applied, Among them
Wl're the transform techniques and operators' methods, One transform
1lIt'lhml. which found wide application in the analysis. is the z-transform
1IlI'lhm!. It represents the counterpart of the laplace transform as applied
III nllllintiUlIS system theory, In this text the z-transform method is
1'\II'II!>ivcly ~Ieveloped and applied to many areas of discrete system
IIII'III'Y,
I hI' ,uhjec\ matter is mainly addressed to engineers and systems
Ilwllli,I,; however. some of the material developed would also be of
IJIII'I~",1 In applied mathematicians, Whenever it is found necessary. most
III IIU' IhC'llrellls alld lemmus are proved 011 a somewhat rigorous mathe-
111011 h.d ha,i!>, Jlowever, to limit the text and to place morc cmphasis on
IIII' .ll'l'lh'alioll!>, Ihe ri~lIr is sllllletimcs sacrificed, ('onsclluently, certain
1111'1111'11" an' tOlldled nil hricl1y without prnnl's, Cnnver~encc prnhlcms
,II" lI11t '"'I'nroll!>ly p"rs"ed, hut lIrc left til Ihe reader ror rurther study,
"III 11111',1 .. I' Ih~' III1PI'll\'~'1I ll1alC'rial till' pC'rtiJlent rdi:rell~'c, arc ~'ited SIl
111.11 II ... 11';11"'1 ~'aJll'a,ily lilld Ihe ri~ol'Utis dcrivlllilllls or Ihl' thl'ory,
C'h,ll'lI'l I .11"'11' ..1'" ill dl:lail thl' .:,lrallsl'lIrll1 Ihl'ury alld it-. 1IJ1l1lilka-
"till', ,11,,1 IIII' IIIII,," ...d ,'tl;\II .. l'ollll, I',xll'n .. ivl' Ii .. " III tlll'lIlI'lII', 01 rilles
,II" "llh"1 011'11\'1'" III 1"1"1'1111'" III ;11'l1"ailll 1IIl' n'a"I'1 With till' ,Il'taill'"
II'," .. I 1111',1111'1111111
\'11
viii PREFACE
In Chapter 2, the z-transform method is applied to the solution of linear

difference or difference-differential equations. Several forms of difference
equations are given with examples from network. circuit, and control
theory.
The problem of stability of linear discrete systems and the root dis-
tribution within the unit circle are discussed in detail in Chapter 3. This
stability discussion is not yet available in any other text; it also represents
some of my original work that is available only in papers. Thus the
contents of this chapter could be very useful to applied mathematicians
as well as to engineers and systems theorists.
Further development of the z-transform theory is considered in Chapter
4, with the derivation of the convolution of the z-transform and the
modified z-transform. The application of the convolution theorem.
and of the two-sided z-transform to statistical study and design of discrete
systems are demonstrated.
The application of the convolution z-transform to the solution of
certain types of nonlinear discrete systems or nonlinear difference equa-
tions is emphasized in Chapter 5. Examples of the application of this
method are also described.
(n Chapter 6 the periodic modes of oscillation of certain nonlinear
discrete systems are emphasized and applied. The stability study of limit
cycles and their identification is also included.
The use of the z-transform method and its modification as applied to
the approximate solution of differential equations or continuous systems
is examined in Chapter 7. The methods developed are well suited for
digital computer analysis.
Finally, in Chapter 8 are found various examples of areas in discrete
system theory. These areas include nonlinear sampled-data feedback
system, discrete antenna array theory, information and filtering theory,
economic systems, sequential circuits. and discrete Markov processes.
The main emphasis is the application of the z-transform theory. developed
in the preceding chapters, to various disciplines of discrete system theory.
Extensive tables of z-transform, modified z-transforms. total square
integrals, and summation of special series are in the Appendix. Further-
more, there are many problems that illustrate further the application of
the z-transform theory and indicate the method of the proofs of cert:lin
theorems omitted from the text.
Some of the chapters have heen 11I\I~ht :It :1 ~r:ulU:ltc seminar fur a
olle-lIllit. one-scmcstcr l'uursc at the Universily III' Califurnia. 'Thc pre-
n"luisitc is all ullder,~radllatl' "(1111's,' ill lim'al' sysll'lI1 analysis. If used as
a lI'xt.1 hl'lil'w thalnlllh"lI1all'rilll":1II h,'a,h"llIalc'IYl'llv,'n'" ill a tWII"Ullit,
lilli' .,,'IIII,.,Il', rnlll'il' lit IIII' ,~, III I UIIIl' Ie'wl,
PREFACE ix
The writing of this book was aided considerably by my students whose
contributions and efforts are very much emphasized. To mention but a
few, I would like to thank Professors M. A. Pai and S. C. Gupta, Doctors
C. A. Galtieri and T. Nishimura. and Messrs. A. Chang. T. Pavlidis. and
A. G. Dewey.
The suggestions and valuable comments of Professor W. Kaplan have led
to much improvement in the mathematics and they are greatly appreciated.
Since the major part of this material grew both from my research
activities and those of my students, I would like to thank the Air Force
Office of Scientific Research for their generous support and interest in
our group at the University of California. The patience and efforts of
Mrs. L. Gilmore in typing this manuscript are gratefully appreciated and
acknowledged.
I wish to acknowledge the very helpful private discussions by cor-
respondence, over many years, with my colleague Professor Va. Z. Tsypkin
of the Institute of Automatics and Telemechanics in the U.S.S.R.
t hope that the contents will serve the useful purpose of providing a
unified and integrated transform method which can be applied extensively
lind fruitfully by systems engineers and scientists for many years.
/oi·hr/lory 1964 E. I. JURY
CONTENTS
z-TRANSFORM DEfiNITION
AND THEOREMS
1.1 Discrete Time Function and z-Transform Definitions 2
1.2 Properties of z-i'ransforms 3
1.3 Inverse z-Transform and Branch Points 9
1.4 The Modified z-Transform 15
1.5 Relationship between Laplace and z-Transforms 20
1.6 Application to Sampled-Data Systems 28
1.7 Mean Square Value Theorem 29
I.K Equivalence between Inverse Laplace and Modified z-Trans-
forms 31
1.'1 Other Transform Methods 36
Appendix. A Method of Determining the Coefficients of the
.:-Tnlllsform Expansion 41
2 .:-TRANSrORM METHOD OF SOLUTION

OF LINEAR DIffERENCE EQUATIONS 45
.'.1 LillcClr Dilference Equations with Constant Coefficients 45
.'.) SlIlution of Difference Equations Whose Coefficients Are
Pcrillllic Functions 48
.'. , l.incClr l>illcrcncc-DiITerential Equations 51
.'..1 I )illi.'I'l'lIl·C ':'Iuatinns with Periodic Coefficients 57
.'.... Tillll'·Var'yin~ Dillcrcncc E(llIutions S9
.',(1 '1 inl!" Varying .:-Translill'ln lind thc Systcm FUllction 66
.', I 1)01111'" .:-"('rallsliH·lIIaliun lind SCllutilln IIf Parlinl Dil1'erencc
73
xii CONTENTS
3 STABILITY CONSIDERATION FOR

LINEAR DISCRETE SYSTEMS 79
3.1 Definition of Stability 80
3.2 "Stability Condition for Linear Time-Varying Discrete
Systems 81
3.3 Tests for Stability 82
3.4 Stability Test Directly Applied in the z-Plane 84
3.5 Determinant Method 8S
3.6 Critical Stability Constraints for System Design 94
3.7 Number of Roots of a Real Polynomial Inside the Unit
Circle 9S
3.8 Relationship between the Determinant Method and Hurwitz
Criterion 97
3.9 Table Form 97
3.10 Division Method 109
3.1 I Aperiodicity Criterion for Linear Discrete Systems 112
3.12 Theorems Related to Stability and Number of Roots 116
Appendices
l. Derivation of the Table Form of Stability 121
2. Singular Cases in Determinant and Table Forms 128
3. Summary of the Stability Criteria 136
4 CONVOLUTION z-TRANSFORM 142
4.1 Complex Convolution Theorem 142
4.2 Complex Convolution Theorem for the Modified z-Transform 145
4.3 Applications of the Convolution Modified or z-Transform
Method 147
Appendices
I. Proof of Complex Convol ution Formula 167
2. Derivation of Total Square Integrals Formula 168
5 CONVOLUTION z-TRANSFORM APPLIED
TO NONLINEAR DISCRETE SYSTEMS 174
5.1 Assumptions 17S
5.2 Convolution z-Transforms of Certain Functions 176
5.3 Method of Solution for Secon<f- nnd Higher-Order Equations 177
5.4 Illustrative Exumples 179
(, I'FRI()()I(' MC)()I:S OF OS(,(I.I.ATION
IN NONI.INI"AI( I>IS( 'IUTI: SYSTIMS IKI)
tt.! I illlil ('yd,' A'lIIly .. i, of NOllli'"'''1" J)i",n'l(' Sy,II'III' I KI)
td Appli,"aliflllllfIJII'I'IIIIIIIIIIII'lIlnl 1'11'111111111 loSp"'"11I1' I"~ .. ,ilpl,'" II)tl
CONTENTS xiii
6.3 Limitation on the Period of Limit Cycles of Relay Mode
Oscillations 203
6.4 Stability Study of Limit Cycles 207
6.S Forced Oscillations in Nonlinear Discrete Systems 208
6.6 Direct z-Transform for Determining True Oscillation 213
6.7 Periodic Solution of Certain Nonlinear Difference Equations 21S
7 z-TRANSFORM METHOD IN
APPROXIMATION TECHNIQUES 219
7.1 A pproximation Methods 219
7.2 Initial Conditions Nonzero 223
7.3 Integrating Operators 229
7.4 z-Forms and Modified z-Forms 231
7.5 The Choice of the Sampling Period 238
7.6 Analysis of the Error 239
7.7 Low-Pass Transformation for z-Transforms 240
7.8 Applications to Time-Varying Differential Equations 241
7.9 Application to Nonlinear Differential Eq"uations 243
7.10 Other Numerical Techniques 246
H APPLICATIONS TO VARIOUS
AREAS OF SYSTEM THEORY 248
H.I Nonlinear Sampled-Data Feedback Systems 248
1<'2 Analysis of Discrete Antenna Array by z-Transform Method 254
IU Application to Information and filtering Theory 258
X.4 z-Transform Method Applied to Problems of Economics 263
X.5 Linear Sequential Circuits 267
X.(I Application to Discrete Markov Processes 270
APPENDIX
Tahle I z-Transform Pairs 278
Tahle" Pairs of Modified z-Transforms 289
TlIhle III Total Square Integrals <c 297
Tahle IV Closed Forms of the Function I nr x ll , x<1 300
n 0
1'1« IIII.I:MS 301
INUI·X 321
1
z-TRANSFORM DEFINITION
AND THEOREMS
I he techniques of the z-transform method are not new, for they can be
arlllally traced back as early as 1730 when DeMoivrc l introduced the
"lIncept of the "generating function" (which is actually identical to the
Iransform) to probability theory. The concept of the generating function
\\;1" latcr extensively used in 1812 by Laplace 2 and others in probability
Ih,'ory. In <\ much later article by H. L. Se<\I,3 <\ historical survey of the
\1'." of the generating function in probability theory was presented.
1("n'llIly, the development and extensive applications of the :-transform3I ,35
,.1\' IIIl1ch enhanced as a result of the use of digital computers in systems.
I h,'''" ~ystems are referred to as discrete, because of the discrete nature of
Ilw \igllals (lr information flowing in them. Thus a new discipline of
',\",1\'111 Ihc(lry is being developed, to be known as discrete system theory.
Ilw matl'rial here is devoted for the most part to discussing the various
1"n·l .. III' this discrete theory.
I hi' -I ransfnrm method constitutes one of the transform methods that
• ,III J", applicd to thc solution of linear difference equations. It reduces
II ... ',"ll1lilln~ IIl'such c{luations into those of algcbraic equations. The
I "I'I,u'" Iran!ool'llrm method, which is well developed fur thc solution of
.hlll-lI'nllal "'Illations and ('xtcnsively lI~l'd in Ihe Ii ICl'lIt 111'1', can he
1II,,,I,h,'dlo ,'\ll'lId il~ applkahilily tn Ilisl'I'cll' S),Sll'IIlS. SlIch IIImlilil'alions
1t,lIo' 1I",lIlh'" ill inll'milldll/'. Ih,' vadou), a~)'IIl'iakd Il'an"l'lIrlll 1l'l'ill1i111lCS
\\ It II h ,IIC' h, idly di""II)'M',1 ill Ih,' lasl sl'l'lion (II' Ihi)' dlap"'1'.
I h." dmph" i.. mainly .II'vlIlI'cl III IIIl' ,kwlol"III'UI III' Ilw Ilwol')' of
11,111,,1111111 ancllh., 1I11I1It1ic'cl IIall .. 1'01'111 , Many "",I'ul 1111'0,,'111 .. 1I'lah'"
III Iho",,' 11;111',1',,"11', IIIC' "11111'1 cl"llv,',1 III' !oolah',1. III ;11""111'", IIlh,'1'
1111'1111'111',1111' 1I.1'""III'C'clll' Ih,' !,rllhle'lII ""l'lill" 11'11111'" III Ih, .. ,hap"'I,
2 THEORY AND APPLICATION OF THE z-TRANSFORM
1.1 Discrete time function and

z-transform definitions14-16.24-26
In many discrete systems, the signals ftowing are considered at discrete
values of I, usually at nT, n = 0, 1,2, ... , where T is a fixed positive
number usually referred to as the sampling period. (n Fig. 1.1, a con-
tinuous function of time /(1) is shown where its values at 1 = nT are
indicated. The study of such discrete systems may be carried through by
using the z-transform method. This method will be extensively developed
in this and other chapters with its modifications, extensions, and
applications.
Definition
Let T be a fixed pOSitive number (it could be taken as unity). Let/(/) be
defined for this discussion for t ~ O. This case will be extended in
Chapter 4 to cover values of I which are also negative. The z-transform
of/(I) is the function
00
,1[f] = ~(z) = I f(nT)z-n, for Izl > R =-1 (1.1)
n=O p
p = radius of
convergence of
the series
of the complex variable z. We use the symbol,1 to denote the z-transform
off
Since only the values/n = /(nT) of/at nTare used, the z-transform is
actually defined for the sequence {In}.
(1.2)
' ...... _-
i:
J i:
51::::: ...., i:~I i:
;...1
e.... ~I ~
.~.
\!!..
.~
I ~ I
n 'I' "'I' 1'1' 4'" ',',.
I HilllU 1.1 nl~'IC'IC' 111111 "ClIlIiIl"CI"~ hllwIlClII\.
Z-TRANSFORM DEFINITION AND THEOREMS 3
The series in equation (1.1) can always be considered as a formal series
to be'manipulated in certain ways and not necessarily to be summed.
If/(r) has a jump discontinuity at a value nT. we shall always interpret
lenT) as the limit of/(t) as 1-+ nT+, and we shall assume the existence of
this limit. for n == 0, 1,2, ... for all/(/) considered.
EXAMPLE
To obtain the z-transform Of/(l) == t, we use equation (1.1) as follows:
...,
§(z) == 1[f) == I f(nT)z-n == Tz-1 + 2Tz-1 + 3Tz- 3 + ...
11-0
== Tz-J [1 + 2z-1 + 3z- 2 + ... ] == Tz 2'

(z - 1)
for Izl >1 (1.3)
1.2 Properties of z-transforms5 ,14l,25,26

In the following we shall show a few properties and theorems related to
the z-transform. Some theorems witt be presented whose proofs could
be easily obtained as an exercise in the problem section. Their use will
enable us to develop the z-transform method and indicate its applications
in the following chapters.
Linearity of the z-transJorm

For all constants CI and C2. the following prop~rty holds:
...,
1(cdl + cd2) == ~ [cdl(nT) + cs.fa(nT)]z-1I
limO
00 CO
== C1 ~ /1(nT)z-fl + C2 I .fa(nT)z-n
flUO fl=O
(1.4)
Thus 1 is it linear operatur on the Iineur !lpace of ull ::-transformable

functions /(t), (I :?: 0).
If" III 0' :4' (:),
.)I{(/' '1')1·-:1:1'(:) /(UI)I

Proof' By definition
GO
1[f(t + T)] = I f[(n + l)T)z-n

n-O
GO 00
= zft-O
I I[(n + l)T]c'"+ll = z I l(kT)z-k
k-I
(1.6)
where k = n + 1. By adding and substracting /(0+) term under the

summation sign of equation (1.6), we can write the summation over the
range from k = 0 to k = 00. Thus
1[/(1 + T)]" = zl~/(kT)Z-k - I(O+)J = z[§"(z) - 1(0+)]

(1.7)
Extending these procedures, we can readily obtain for any positive
integer m the following results:
1[/(1 + mT)] = zm[§"(z) - ],1/(kT)Z-"] (1.8)
COROLLARY: If 1[/(1)] = .F(z), then
1[/(t - nT)u(t - nT)] = Z-II§"(Z) where U(/)

= unit step function for n = 0, 1,2, . .. (1.9)
00
.1[/(1 - nT)u(1 - nT)] =I [f(m - n)Tu(m - n)T]z-,n

",.-0
00
= z-n I {t[(m - n)T]u[(m - n)T]}z-''''-"' (1.10)

m-O
Letting m - n = k, we obtain
GO
.1[1(1 - nT)u(1 - nT)] = Z-II I f(kT)II(kT)z-"

It--II
= z-n I'" l(kT)z-k = z-"§"(z) (I. J I)

"=0
The use of the shifting theorem is important in the solution of difference
equations as indicated in Chapter 2. Following a similar procedure. we
can easily obtain the z-transform of the forward difference as well as the
backward difference as follows:
.1[£\~f("T») = (z - 1)":'(::) -:: '"l(:
I
- I)'" I 1£\1(0'1") (1.12)
I "
where
£\ I(" 'I') - f( PI + 1)"1' - le".,., ( I.ll)
Furthermore,
I[V'r(nT)] = (I - Z-l)k~(Z) (1.14)
where
V/(nT) == l(nT) - I(n - I)T (1.15)
Complex scale change

If the %-transform of1(/) is ~(z),
1[e- 0 'l'(/)] = .~(e,.r%). (1.16)
Proof' From the definition of the z-transform we have
co co
1[e-"Y(t)] = I e-tJnTf(nT)z-fI == 1: f(nT)(~Tz)-n (1.17)
'1-0 ftleO
Finite summalion1,.26,28
Tu obtain the %-transform of
(1.18)
linit we define
ft-l
1:" f(kT) ~ g[(nT)], or g[(n - 1)T] =I f(kT) (1.19)
k~O k-O
We can write a relation between successive values of the sum by noting
Ihis definition and equation (1.19),
K(nT) = g(n - I)Tu(n - I)T + l(nT) (1.20)
Al'plying the z-transform to this eq~ation, we have
~1(z) = zl(9(z) + F(z) (1.21)
SlIlving I'ur ~q(::). we finally oblain
~1(::) = J[i/(kT)] = _%_,:F(z), for Izl >1 ( 1.22)

k D :: - 1
It,ililll ,,,,tI./ilwll'tll,,,·.fl u l
1111111 Iht' tft'liniliun til' Ihc~-ll'Iln!ll'nrm,
.#(,.) _ i
11·11
f('''/'): " - 'I') +1(2~1') ... , . .
,nOn +/(1..1 (1.2,\'
We readily notice that the initial value is obtained as:

1(0) =. lim !F(z) (1.24)
Iff(O) = 0, we can obtainf(T) as the lim %?(z).

For the final value, let us write z-co
..
1[/(t + T) - I(t)] = lim I [/(k + l)T - l(kT)]z-t (1.25)
""CO k-O
The transform of the left-hand side is obtained from equation (1.4) and (1.5)
lli~ ..
%?(z) - zf(OT) - !F(z) = lim ! [/(k +
I)T - l(kT)]z-1:
""'00 t-O
(1.26)
We now let z - I for both sides of equation (1.26), assuming the order of
taking the limits may be interchanged.
..
lim (z - l)!F(z) - I(OT) = lim ! [/(k + l)T - l(kT)]
1~1 n-'cc 1::=0
= lim ([/(1 T) - I(OT))
"-CO
+ [/(2T) - 1(1 T)]
+ ... + [/(nD - I(n - l)T]
+ [/(n + l)T - l(nT)]}
= lim [-/(OT) + fen + l)T]
.. -co
= -f(OT) + f(oo) (1.27)
so that we finally obtain
lim l(nT) = lim (z - l)!F(z) (1.28)
""'00 1-1
if the limit exists.

,
Complex multiplication (real convolution)

Iff1 andft have the z-transform !FI(Z) and .F.Jz), then
!F1(z)!Fa(z) = 1~/I(kT)fl(n - k)T] (1.29)

Proof: By definition
co
!F1(z)!F.(z) =I Il(kT)z-t!F1(z) (1.30)
1:-0
but from equation (1.11)

Z .t,~.(z) - ICI.(1 - k'l')) (Ul)
11/, en Ie, "'- O. fnf n· Ie
Hence
eo
'1(Z):F2(Z) =I fl(kT)I[ia(t - kT)J
It~O
eo eo
= I
k~O
/l(kT) I Ia(n - k)T]z-1I
n~O
=! {i /l(kT)h[(n -
n-O k~O
k)T]}z-n (1,32)
but 12 (n - k) T = 0, for n < k and ~erefore
~l(z)j~'iz) = 1[~/1(kT)/2(n - k)TJ (1.33)
Complex differentiation or multiplication by I

If ~(z) is the z-transform off, then
l[tfJ = - Tz !!.. ~(z) (1,34)

dz
ao eo
J[if] =I (nT)f(nT)z-n = -TzIf(nT)[-nz-n-l] (1.35)
II~ n~
The term in the brackets is a derivative of z-n with respect to l.
eo d d ao
,[if] = - Tz I f(nT) - z-n = - Tz - I f(nT)z-fi
II~O dz dz n-O
d
= -Tz -:F(z) (1.36)
dz
Similarly. we cun write
(1.37)
Wllt'l'l'
k >0 and integer (1.38)
As II l'llrnllary In Ihis Ihcnrclll we can deduce
I "')'/(11) I
)'" ,=;: It (/~'Jf(:)
.. _....- ( 1,39)
'. ,/(: I)k
wlll"l' 11Il" rUIlction nUl iN lIivon by
"fA I _ 1/(1/ • 1)(11"- 2) •. ,(II ., k -j' I) ( 1.4U)

8 THEORY AND APPLICATION OF THE z..TRANSFORM
As special cases of the preceding.
1[( -1)"n 1lt1(n - k + 1)] = zdltF(z)

--
dzlt
(1.41)
and
dltF(z)
l[n(n + 1)(n + 2) ... (n + k - 1)/(n») = (-t)ltzlt-It-
dz
(1.42)
Special summation theorem
00
If F(z) = I /(n1)z-", then
71-0
F(zlt) = I[V!t(n1)] (1.43)

where
[nlltl
/1(nT) = I
m-O
/(mT),
and [n/k] denotes the largest integer in n/k.

Proof' From the definition of the z..transform we can write
[",it] ] & 00 [nllt) 00 [nllt]
1 m~o /(mT) = 1:.0'1.-" ",~o /(mT) = ~o ~o z-,(mT)
[ (1.45)
Since n varies in integer values from zero to infinity and m also varies
in integer values, we can write for the right-hand side of this equation,
IL!l/(mT)] == io ,,~,/(mT)z-" == l/(mT) 7I!ltz- n
«: 00
== ! /(mT)z-7Jlk 1: z-n (1.46)
", .. 0 11_0
However,
00 'I.
2'1.-" =--, for 1'1.1> 1
.. ~o 'I. - 1
Therefore equation (1.46) can be written
[nllt) ] z z
J[ ! f(mT) == - 2 f(mT)z-mk =
00
-F(zlt) (1.47)
".-0 Z - 1 ",-0 'I. - 1
From equation (1.14), we have for k = 1
'I. - 1
:J[Vf(nT)] = -.~(=) ( 1.4K)
.::
lIsing this cqulltinn in (1.47) we tinnily nhillin
:J'(:') - .JIVf.(,"nl ( 1.49)
%-TRANSFORM DEFINITION AND THEOREMS 9
This theorem is very important in obtaining the inverse z-transform of
special functions of ~ (z) containing essential singularities. These functions
are discussed in detail in the next section.
We introduce the following additional theorems whose proofs are left
as an exercise to the reader.
Differentiation with respect to

second independent ('ariable
1[.!
oa f(t, a)] = .!
oa 9'"(%, a) (1.50)
Second independent ('ariable limit t'alue
.llim f(t. a)] = lim .'F(z. a) (LS1)

L;. ... oo 4~tI'O
Integration wilh respect to

second independent t'ariable
J[1:1f (t, a) daJ = f~1 :F(z, a) da (1.52)
if the integral is finite.
1.3 Inverse ~-transform and branch points 14 •a7

The discrete functionf(t) at t = nT or f(nT) can be obtained from F(z)
hy " process called the inverse z-transform. This process is symbolically
delluted as
(1.53)
or
where :~(z) is the ::-trnnsform f(t)t or f(nT).
In the following, we discuss the several methods from which we can
IIhlain /(tr1') from :~(::) or the inverse z-trunsform,
'J111' pm"c'r ,vc'r;c',I' mc·tlmt!
WIIl'II .... ( ,) is I~iv"n :as a funl'lillll an:llylic I'm 1:1> R (lind lit ;: ,... 'I').
Iii,' v:III", 111'/(11 /') ,'an h,' j'l'mlily Ilhl:ainl'd as the l'IICnki,'nt III' .. " ill Ihe
I"'WI'I "'I i," "'Pilll,illll (T:aylm"s ",· .. il·s) nl' :~ (~) IlS " l'ullctiull "I' .: I.
10 THEORY AND APPLICATION OF THE 'I.-TRANSFORM
From equation (I.l) it is observed that
§l(z) = f(OT) + f(T)z-1 + .. .f(knz-k + .. .f(nnz-n + ...

(1.54)
Thus it is noticed that f(nT) can be read off as the coefficient ofz- n and
so can values off at other instants of time.
If §l(z) is given as a ratio of two polynomials in 'I.-I, the coefficients
f(On, . •. ,f(nT) are obtained as follows:
§l(z) = Po + P1Z-1 + p"z-I + ... + P..z- n

qo + qlZ- 1 + qaz-a + ... + qnz-n
= f(OT) + f(1 T)Z-1 + f(2T)z-2 + ... (1.55)
where
Po = f(OT)qo
\
PI = f(l T)qo + f(OT)ql

(1.56)
Pn = f(nT)qo + f(n - I)Tql + fen - 2)Tqa + ... + f(OT)q ..

It is also observed thatf(nT) can be obtained by a synthetic division of
the numerator by the denominator.
Partial fraction expansion

If SI"(z) is a rational function of 'I., analytic at 00, it can be expressed by
a partial fraction expansion,
(1.57)
The inverse of this equation f(nT) can be obtained as the sum of the
individual inverses obtained from the expansion, that is,
We can easily identify the inverse of a typical !F/t;(z). from tablest or

power series and thus obtainf(nT).t
t Sec Appendix, Tnblc I.

t A 11c:lerlllilllllll 1I11:lhllll fur IIhllllllllll: I("n fl'lllll('IIIIIIlillll (I. "1"1) i~ inlhc: Appendix !If
Ihllll'llllpter.
Complex integral formula

We can also represent the coefficient f(nT) as a complex integral.
Since $&"(z) can be regarded as a Laurent series, we can multiply ,~(z) in
equation (1.54) by Z"-I and integrate around a circle r on which 1zl =
Ro' Ro > R, or any simple closed path on or outside of which ~ (z) is
analytic. This integral yields 2'If i times the residue of the integrand, which
is, in this case, I( nn,
the coefficient of z- I. Hence
i$&"(Z)Z,.-1 dz = f(nT)' 2'1Tj (1.59)

or
f(nT) = ~.1 $&"(Z)Z"-I dz, (n = 0, 1, 2, ... ) (1.60)
21T} h'
The contour r encloses all singularities of .~(z) as shown in Fig. 1.2.

The contour integral in equation (1.60) can be evaluated when .~(z) has
only isolated singularities by using Cauchy's integral formula,
f(nT) = _1_. .1 $&"(z)zn-I dz = sum of the residues of §(Z)Z,.-I

2'1T) J..
(1.61)
The following cases are discussed for the inversion formula.
TIfE POLES OF .~(z) ARE SIMPLE. Assume that
.~(z) = .;f'(z) (1.62)

~(z)
When ~ (z) has simple zeros only, the residue at a simple singularity lJ is
given by
lim (z - a).:F(z)z,,-1
: 'n
= lim [(Z -
:-a
a) ,;fez)
~(Z)
zn-1J (1.63)
Imaginary
I'/)"~!; of
:!i ( .. )
I It ;IIIU' U ('1111111'" 11110',:"1111111 III IIII' . plll\lll

12 THEORY AND APPLICATION OF THE Z-TRANSFORM
POLES OF ~(Z) ARE NOT GIVEN IN A FACTORED FORM, BUT ARE SIMPLE.
The residue at the singularity am is
[~(z) 7I-IJ (1.64)

~I(Z) Z :QIJ..
where
~'(z) = d~(z) (1.65)
dz.
/F(Z) HAS MULTIPLE POLES. The residue at a kth-order pole of !F(z) is
given by the following expression: residue at kth-order pole at a,
b l = _1_ d"-l [.~(z)(z _ Q)"Z7l-IJI (1.66)

k - I! d?!,-1 '_11
.~(Z) has essential singularitier7

In some cases when ~(z) has an essential singularity at z other than
infinity, we can utilize the preceding theorems of the z-transform to
obtain the inverse without power series expansion. This is best illustrated
by the following examples.
If !F(z) is given as
~(z) = e-Iz/zI-1/2.' (1.67)
to obtain its inverse, that is,/(nT) = rl[~(z)J.
We let ~(z) = e-z /, and """(z) = e- 1/ 2• 2 ; then by using the real convo-
lution theorem on p. 6,
n
lenT) = ,-l[~(Z)~(z)J =rI h(kT)g[(n - k)TJ (1.68)
1<-0
The inverse of e-z /, can be readily obtained from the series expansion
(1.69)
Furthermore, from the theorem on p. 8, we have for k = 2,

,-I[e-l/t •
I
J = h(nT) = 'i/ I
(n/21 (
m"O
-=-- 1)m
2'"m!
(1.70)
Using equations (1.69) and (1.70) in equation (1.68). we obtuin (rcplucing

k by s)
(1.71)
The expression in the bracketed term of equation (1.71) is different from
zero only for s even, and then it is equal to
(-I )"/2
(1.72)
2,/2(S/2)!
Putting s = 2k in equation (1.71) and using (1.72), we finally obtain for
the inverse z-transform
["/21 (l)'a-J,: X,,-21t H ( )
=
l(nT) rl[e-Z/Ze-1/2Z'] ! - = = ~
k=O k! (II - 2k)! 21- n!
(1.73)
where H" is the Hermite polynomial of degree II.
ff(:;) is irrational/unction 0/"Z"37
The inverse z-transform can be obtained either by power series expansion

or the integral formula. If we use the latter form, we must be specially
careful in the integration because of branch points in the integrand. As
in the previous case, we shall illustrate the procedure by using the following
example.
Let ji"(z) be given as
$(z) = e~ b)" (1.74)
where Ot can be any real value and assumed to be noninteger.

To obtain/(nT) we use the two procedures of contour integration and
power series expansion. To simplify the integration process we introduce
~l change of variable to normalize the constant b to be unity.
Let
::=b.'l (1.75)
Then .:F(::) becomes
.'I/(b/I) = (11; I)'" (1.76)
The function .f:(h!l) hm; :1 hmllch cut in the !I-phme thut extends from
Icrn h) minus unily liS lIhUWIl in Fig. 1.3. By lIsing c'Iulltiun (J.lIO), the
inwl"sc is ~ivcll
.
1~1I'1') = .J II;F(lII/)]
,
.I,~.
= 2fT.i I. (It:f.-..!)'',,II 1,/11
I' /1 • •
( 1.77)
wh,'rc Ih,' dusCo',J ,'unlnm' I' iN Ill! shllwn in Fi~. 1.3.

W,' c.·un "I\\ily shllw Ihlll III Ihe Ii mil !I • n the inlc~1'II1 II 1"11 II III I Ihc
"nlllll'·II·dc.' lie' /) ih /l'III, Fmlhc.·I"II1IlI"c.', Ihc intcl-Il"IIllllnnl-lll'A III 111'111 1(· .....
14 THEORY AND APPLICATION Of THE Z-TRANSFORM
Imaginary
y-plane
Real
A c
-1.0 ) o
z;; e-jlr
FIGURE 1.3 Contour integration enclosing branch point or S:(by).
Therefore f(nT) can be obtained by integrating around the barrier.
f(nT) = b ll . [ fO(xe- S"_ + 1"\x ..- 1e- SIIII dx

2.,,) Jl xe i'tt J
+ fl(xe Sr : 1\- x.. -1eSwn dX] (1.78)
Jo xe 'lr J
This equation can also be written as
f(nT) = b".[ fl X"- 1- 2 (1 _ x)Glelr(n-GlI dx

2.,,) Jo
-il x,,-l-_{l - x)-e-I,,(n-Gl) dxJ
= b" sin (n - <x}." fl X"-l-«(t _ x)- dx (1.79)

." Jo
By utilizing the following identity,
B(m. k) = r(m)r(k) _ fl xm-1(1 _ X)k-l dx (1.80)
rem + k) Jo
we obtain for fCnT)
lenT) = b1l sin (n - <x)." r(n - <x)r(ot + 1) (1.81)
." r(n + I)
By using the identity
r(m)r(1 - m) = -?!-
sm ."m
(1.82)
we finally obtain
l(nT) = b"
rCn
not + l)
+ l)r(<x - n + 1)
= (<x) b"
n
(1.83)
and for at negative integer
(ex) _ ~'('.I.. - ~) ( .... 1)" (1.84)

" 1'(" + 1)1'( ·"11)
Z-TRANSFORM DEFINITION AND THEOREMS J5
We can also obtain J(nT) from the Taylor's series expansion of F(z)
as follows:
~(z) = (z ~ bJ = (1 + bz- 1 )-
«: 1 d"(1 + bz-I)'·I _
=I-
II~O n! (dz- 1)" .-1_
z II (1.85)
0
Equation (1.85) yields
ao J
~(z) = I - ex(ex - I)(ex - 2) ... (0: - n +I)b"z- II (1.86)
11=0 n!
We know that
r(O( + I) = at(~ - 1')(7. - 2) ••• (ex - n + J). r(cx - n+ 1)
(1.87)
and
r(n + I) = n! (1.88)
By using equations (1.87) and (1.88) in equation (1.86).
~(z) = (z + b, = i r(at + l)b" Z-II

z J r(n + l)r(at - n + J)
""0
= I (at) b"z-" (1.89)

n II~O
From the definition of the z-transform we readily ascertain
J(nT) = (:) b" (1.90)
The entries of Table I in the Appendix readily yield the inverse z-

transforms ofJ(nT). They have been obtained by using the theorems and
properties of the z-transforms discussed in the preceding sections.
1t should be noted from the inverse theorem thatJ(nT) can be obtained
from the power series expansion without having to evaluate the poles of
.'F(z). This feature offers a decisive advantage over the continuous case
using the Laplace transform. Thus the z-transform is used for approxi-
mating a continuous function, which will be discussed in Chapter 7.
1.4 The modified z-lmnsform lol ,2H,31,32
In nlllny applicalions of discretc systcms and pnrticularly ill Ihe IISl~ tlf
,Ii~itlll l·nrnJllIll~rs in cClntrul systems, Ihe uutput hl'lwccn I hl' slllllpJinl~
instants is vl~ry impnrtant. In SlUdyill~ hyhrid systems (miKl'd dil~itllJ :lIId
111111111", sy!ih'lIls) thl' oUlpll1 iN II \·CllltiIlIlClIlS flllll·tiun IIf lin,,", 11111111111" thl'
.:-trllnsfllnu IIll'lhud is IIClI 'Iuih' IIdl'llllll'" fur II nitklll "Iudy ClI IlIId,
I(t)
FIGURE 1.4 A fictitiously delayed output to scan values of {(I) other than at I = nT.
systems. However, the z-transform can be easily modified to cover the

system behavior at all instants of time; The extension of this method is
called the modified z-transform method.
The modified z-transform is also important in the study of linear
systems for periodic inputs, in sampled-data systems with pure delay, in
limit cycle analysis of discrete systems, in the solution of difference
equations whose coefficients are periodic functions (with period equal to
unity), in the solution of mixed difference-differential equations, in
approximation techniques for continuous systems, and in summing up of
infinite convergent series. The introduction of the modified z-transform
as well as its applications will be discussed both in the remainder of this
chapter and in the following chapters.
Modified z-trans/orm definition l 4.?JJ

To obtain the values of /(/) other than at 1 = nT, (n = 0, I, 2•... ) we
can delay the function /(1) by a fictitious negative delay !l T as shown in
Fig. 1.4. By letting !l vary between zero and unity we obtain all the
points of /(1) for 1 = (n - A)T, n = 0, I, 2, ... and 0 ~ !l ~ I. As will
be shown in Section 1.5 in order to avoid any convergence problem in the
integral evaluation of tile modified z-transform from the Laplace transform
of /(1), and also in order to utilize the existing extensive tables of the
modified z-transform, we make the following change of variable:
A = I - m. O~m~ I
With Ihis chllngc of varillhle. ( hl'l~lIml'S
I - en ." I ., m)'I', n - n, 1.2 •... , n <~ ,'t < I e1. 1)1 )

z-TRANSFORM DEFINITION AND THEOREMS 17
We can also scan the values of the continuous function between the
sampling instants by fictitiously advancing the functionf(t) by the amount
'YfT. such that
1 = (n + 77)T, n = 0, 1.2, ... ( 1.92)
The preceding description of the time is utilized in some other works,
thus avoiding any convergence difficulties in the integration process as
will be explained in Section 1.5.
The modified z-transform off is defined as follows:
.~(z, m) ~ Jm<!) = i [f(n - I + m)T]z-lI,

,,-0
(1.93) ° ~ m ~ I
This definition also relates to the modified z-transform of the function
+
.I [(1/ - I m)T).
By using equation (1.9), the preceding equation can be written as
':F(z, m) = Z-1 ~"" [fen + m)T)z-n. ( 1.94)

n=O
I-'rom this equation by letting m = 0, we readily deduce

ex:
z.~(z, m)lon'o = ~ f(t1T)z-II
n~O
= .~(z) ( 1.95)
Therefore the z-transform is obtained as a special case from the modified

transform. Furthermore, if f(t) has no discontinuity at the sampling
illstants (or jumps), the :-transformation can also be obtained as follows:
.?="(z) = .:F(z. m)l ... 1 (1.96)
Whenf(l) has a discontinuity at the sampling instants. the lime function
Il'Iatl'd to this equation yields the value at the left side of the discontinuity,
Ihal i~. at 1= nT , for /I = 1,2, ' .. and the value zero at 1/ = O. This
'0111 Ill' readily ascertained by noting equation (1.7).
SlInilar to the .::-transform we can derive several properties of the
IIIlIdilil'd ::-transform, These properties and important theorems only arc
',1;111'" in thl' ftlllowin~ disl'ussions and the proofs clIn be e~,sily derived
by IIH' n'adl''', Thl' steps for the derivation follows exactly as for the
Ilall~fllrlll,
rlII'O/',·II/.\' ,."/,/{",, 10 1/11' //Ior/i/i,'" ,>I""II.~j;l/'Il/ll.~';
I~H IIAI VAIlII IIIIIIHI M
11111./111 I 111),1 IIIII',J I~. III' II I) II

II
m 'u
'11 •
",
I. 'II
]8 THEORY AND APPLICATION OF THE Z-TRANSFORM
Special case, the response over the first interval, n = 0, 0 s m s 1, is

lim fen
,.. ... 0
+ m)T = lim z.F(z, m)
Z:~Q)
(1.98)
O~m:sa O:S:m:S:l
FINAL VALUE
limf(n, m)T = lim (z - l).F(z, m) (1.99)

ft-Q) I~l
if the limit exists.

REAL TRANSLATION
1 ... [f(1 - kT)] = z-Ic.F(z, m), k - 0, 1,2, ... (1.100)

1",[f(1 - aT)] = z-IJF[Z, m + I - a), 0 ~ m $ a
= .F(z, m - a], a$ m~ I (1.101)
where
o < a < I, and zero initial conditions off
LINEARITY
where a l are constants
independent of t
(1.102)
COMPLEX SCALE CHANGE
1 ... [e HTf(t)) = e:!:bT(m-l).F(ze±bT, m) (1.103)

DIFfERENTIATION WITH RESPECT TO m
1m [1.-
om fen, m)TlJ = 1.-
om .F(z, m) (1.J04)
MULTIPLICATION BY IT<
1 ... [11(1)] = r[(m - 1).F1(z, m) - z:z !Fl(Z, m)J (1.105)

where
.F1(z, m) = 1",[tk-Y(I)] (1.106)
and k integer larger than zero.
DIVISION BY 1
1 [-f(t)J
m t
= -T1 z"'-11zco z-m=(
or z m ) d z +)'1m -f(t)
' 1-0 t '
0$ m $ 1 (1.107)
INTEGRATION WITH RESPECT TO I
1m [I ' 1 o
f(l) dl ..,. . T ..
. z .- I
il C)
o'lt(z, til) ,I", +.,. i"'
41
:i>(z, m) (I",
( 1.1(8)
SUMMATION OF SERIES
....
I/(n + m)T == lim ~(z, m), (l.l09)
»-0 %-1
if the sum exists.
INVERSE MODIFIED Z-TRANSFORM. The continuous time function
1(1)1/=<I1-.'I,n)" can be obtained from the modified z-transform by a
process called inverse modified z-transformation, that is,
1(')II~(n-1+"')T == I(n, m)T == J';;,I[~(Z, m)1 (1.110)
Methods similar to those for the inverse z-transform exist for the inverse
modified z-transform, namely the integral formula and the power series.
The integral formula yields the time function in a closed form,
1(1)11-<n-1+ml1' = ~ ~(z, m)z ..

,( -l dz, 0 ::; m ~I
2'ITJk
t == (n - I + m)T (1.111)
where the closed contour r encloses in a counterclockwise direction the
pules of .~(z, m).
The power series method yields the continuous function in a piecewise
form.
z/F(z, m) == !oem) + II(m)z-l + h(m)z-Z + ...
+ In(m)z-" +, .. , Izl > R 0 ~ m ~ 1 (1.1 12)t
where lo(m) with 0 ~ m ~ 1 represent the continuous time function in
the first sampling interval, II(m) represents the same function in the
second interval. and so forth. Actually. the time function is related to
./,,(m) as follows:
f(t) == I(n + m)T == f~(m) (1.113)

II should be noted that since m can be considered as a constant in the
contour integration of equation (1.111), the same tables for the inverse
,:-transform are readily applicable to the inverse modified z-transform.
MAXIMUM OR MINIMUM POINTS OF I(t) == I ..(m). To obtain the maxima
III'minima poinls we clin differentiate the modified ::-transform with
H'sp~~cl In m tn ohtuin. using the power series form.
o;;,·J'"(::. m)
Om
=.('( m) +.,"() +I'(z In;:)
II 1m z
1 a
+ ... -+ l:,("'):: " -I- ' ., (1.114)
I '1""' ,:"III'III'lclIlII "'/.(111) I~ IIhlilllll'" IIHill,: Ihl' IIc'll'llIlIlI;l1I1 funll IIiN"II""" III Ihr
"1'IM'III'h, I" Ihi" dllll"c'"
If we let
J~(m) = 0, J;(m) = 0, ... , J~(m) = 0 (1.115)
the solution of this equation for 0 < m < 1 yields the maxima or minima
points. The sign of the second derivative determines which points are
maxima and which are minima. The preceding theorem is very important
in determining the quality of response in discrete systems.
MODIFIED z-TRANSFORM OF A kth DERIVATIVE
J",[flk)(t)] = Tt10"
amt§"(z, m), provided that (1.116)
IimJ'nl(t) = 0 for
1-0
and
X {z't-l[z§"(z, m) - ~>JI(m)z-"J} (1.117)
where h is.a positive integer and lim pnl(t) = 0 for 0 ~ n ~ k - 1.

1-0
1.5 Relationship between Laplace

and z-transforms 14 ,29,34
The one-sided Laplace transform of a function "J" is defined as follows:
F(s) ~ ~[f) =ii(t)e-" dt, Re [s] > an (1.1 18)
where aa is the abscissa of ahsolute convergence associated withJ(t).

The inverse Laplace transform is represented by the following Bromwitch
integral, that is,
1 i<+/<O
J(t) = ~-l[F(s») = -. F(s)el'ds, c > am
21rJ c-J<o
for t >0 (1.119)
We can obtain both .F(z, m) and .F(z) directly from F(s) by an integral
transformation symbolically denoted as follows:
:F(:::. m) /~: ,Z",I/'"(·v)I ( 1.12()

lind
(1.121)
z-TRANSFORM DEFINITION AND THEOREMS, 21
The purpose of this section is to show the preceding relationship and
develop the required transformation to enable us to obtain the z-transform
directly from the Laplace transform,
The function f<nT) shown in Fig. 1.1 can be considered as a sum of
rectangular pulses of area hf(nT), where h is the width of the rectangle
(or pulse). Such a sum approximates L hf(nT) o(t - nT), since o(t - nT)
should be considered as the limit of a pulse of unit area. Thus to convert
{(liT) into a train of pulses
f(O) !J(r),f(T) !J(t - T),f(2T) o(t - 2T), ... ,f(nT) 6(t - nT)
(1.122)
a scale factor I/h is needed for such a conversion. Therefore we can

II'place the sampled functionf(nT) by the impulse functionf*(t) provided
Ih., s('alc factor (or pulse width) is accounted for. The definition of such
all impulse function is given (assuming a full impulse occurring at t = 0+).
<t:- 00
f*(t) = f(t)! O(t - nT) =! l(nT) O(t - nT) ( 1.123)

n=O n~O
I aking the Laplace transform of this equation and noting that
(~(t) = 0, t ¢ 0, and fO(t) dt = 1,

for a <0 <b (1.124)
111111
..'/'I~(t - kT)] = 1 00
O(t - kT)e-" dt = e'- ks7 ' ( 1.125)
\\1' IIblain
ae
,...(.~) ~ ,Y'[!*(t)] = 2 !(nT)e- nT• ( 1.126)
n-O
II"" 1''111al ion is readily recognizcd as the ::-transform ofJ. if we replace

• I. bv IIt'IWI' we eSlablish
.'~(:) - /'.(
, .\' >1 • '/,' III : ( I. ) 27)
",
II \\T dl'I\1I1!- ~ "(1 II/')·-'~ '/'( I), WI' I'l'adily t'~lahli .. h Ih(~ ,',,"m'd ion
III~"',~ I I)', I "I.' I :I1lt1 1,1]7) Iwlw('rll IIII' .'·Irall .. rmm IIlId Iht, 1111'1111'1'
11,,"',1111111
-'I ) I, ",. - ""I,~).!l '1'1(,·(111 - .'I'I/It) It"lnl (I J.lH)

By using the convolution theorem for the Laplace transform and

assuming that f( t) contains no impulses and is initiaUy zero, we obtain
(1.129)
where
F(P) = F(s) 1._" = 9'[f(t)] I.~" (1.130)
1 T = 1 + e-T• + e- 2T• + ' , , e- nT• + ' , ,

1- e- •
= ~[bT(t)], for le-T'I <1 (1.131)
and
For the case F(s) has only one degree higher denominator than numer-
ator, Eq, 1.129 should be modified,40
1 f.c+J oo 1
F·(s) = ~(z) 1._e = -, T'
21T} c-loo
F(p)
1- e
-TC.- )
l>
dp + 1f(0+)
(1.132)
The addition of !/(O+) is required in view of our definition that a full

impulse occurs at t = 0+,
Both equations (1.129) and (1.132) yield a relationship between ~(z)
and F(s), hence we define the following transformation :40
~(z) ~ .('[F(s)]
= [~ r+C /oo
F(p) ~TC.-l>1 dp + V(O+)] eT'_~ (1.133)
21T} JC-lae 1- e
El'O/uotion of .('[F(s)]
To evaluate §(z), we assume first that F(s) has two degrees in s higher
denominator than numerator; thus we can use equation (1.129), The
path of integration in equation (1.129) should lie in an analytic strip which
does not enclose or pass through the poles of the integrand, This is
assured in view of the restriction imposed on c,
In effecting the line integral, we can readily enclose in a negative sense
(clockwise direction) the poles of 1/1 - (' .Tc. pI in the right half of the
p-plane, or alternatively we may enclose in a positive sense the left half of
t Where ,r - Rr 1.\'1. rI.,. - IIh ....·I\'1I or IIh."llIlr l'''"Y!',!:!'I!"!' til' /'In, If.. , - IIhwi~~u
CI""h~lllulr l'llllyr'W'lIl'"
,,1"\,111 II!'II' "., - II, " - It I' 1/'1 .
the plane. Because of the assumed form of F(s) the integrals on both the
infinite semicircles are zero.
If we integrate along the left half p-plane as shown in Fig. 1.S equation
(1.129) becomes
·~(%)I.~,7'O = -.
I
2fT)
t F(p) I- T ( ' dp
1 - e "'J)
(1.134)
If f~s) has only simple poles, the integral using Cauchy's formula yields
I he sum of the residue of the function in the closed path, that is,
~(z) = rontAor
~ residue of A(p)
B(p) 1 - e
~ IJ
T I
e- , .~e7'o
(1.135)
m",
where
F(p) = F(s)/.=» = ~~:~ (1.136)
When B(s)/.~P = 0 has simple roots only, this equation becomes
(1.137)
wh~'rc S •• Sz, S3, ••• , s" are the simple roots of B(s) = 0, and
B'(s,,) = dB
ds
I sa ••
(1.138)
WIII:re /'ls) has branch points in addition to regular singularities, the

llallsful'lncan also be obtained using equation (1.129); however. here
Imaginary
c+jfJO x
.f ....... ~
jt"'''- x p-plane
II / Poles 01
!\.·(II)
X
x
I
....... Poles 0 1 - - -
,, ,,'MY.
)( x;'- 1- ,.-T/o-p)
I 0,0
,.' x
\ q "
'~ .......
--- ...•.... ...
/
I Ie .IIIU I \ ",.Ih IIllIIh'/:..IIII ... III 1111' "'1'1111111111 Ih"" ""11\1'

, p-plane
Poles of
\
.,
\
\
F(pJ
I
x ---.l'2 .. I Real
X
J-
I
xt T 'f
x I
I
x JI
}$.k ....
c -joo
'"
FIGURE 1.6 Path of integration in the right half of the p-plane.
special care is required to evaluate the line integral. An example will be

presented later.
The integral in equation (1.129) can also be evaluated by the contour
integration which encloses only the poles of 1/1 - e-'l'II-PI in the right
half of the p-plane as shown in Fig. 1.6.
Thus equation (1.129) can be represented by
1 , I
F*(s) = ..2'[J*(I)] = -.
217)
F(p)
] -
- T ( ) dp
e I-P
(1.139)
The integral of equation (1.139) is equivalent to the negative sum of

the residue according to Cauchy's theorem,
F*(s) = -(sum of the residue of the

integrand at the poles enclosed) (1.140)
By evaluating the residues at the infinite poles of 1/1 - r 7'CS-PI, we

finally obtain
1 k~oo
F"'(s)=- I F(s+jkw r ) (1.141)
T k--co
where
(I. 142)
If we replace in this equation ,,'1'. = :, we readily ohtain the ;:-transli,rm

,~(.;:)
= Z[F(.\,)J, The'two fOI'IIlS III' ( I. I.Vi an" 1.141) arl' l'lluivalent clln
he relldily verified till' hpcciflc ('""mple" III' 1-'(.\,),
For F(s) has one degree in s higher denominator than numerator. the
integral along the infinite semicircle in the left-half plane is no longer
zero. whereas in the right-half plane it is still zero. There equation (1.132)
should be used. which yields on the left-half plane the same as equation
(1.135). whereas for integration in the right-half plane we obtain 1·1.·10
I k~'"
pes) = - ~ F(s + jkwr) + U(O+) (1.143)
T k:-..,
It should be noted that here the infinite summation is not absolutely
convergent. However. if the sum is evaluated by taking pairs of terms
~'vrresponding to equal positive and negative values of the index k. the
~llIll converges to a definite value,
EXAMPLE
(iiven F(s) = 1'( I - /3)SII 1",2R, ~ > o.
f/ > O. and noninteger, we obtain
..... (.:) ~ Z[F(s)] as follows:
\Ising the contour integration of equation (1.129). we can write for
,. . I
1 iC+iOC 'rJI 1I-lr(1 R)
.F(z) = Z[F(s)] = -. e p -.1:- I' dp (1.144)
27TJ c-;-,; I - Z e
The integrand has a branch cut (for ~ not integer) in the left half of
lIu' plane. Using Cauchy's formula the integral of equation (1.144) is
"lllIivalcnt 10 integration around the branch cut as shown in Fig. 1.7.
lit-lin'
(1.145)
c +joo
fl-plane
J' u
-om
Rf'1I1
\
\ f'" 1,'·-'·
\ ,,
, "-
..... ,
111,1 'itl I I I',"h III IIIh'I~IIII""1 ,111111111111 hllllll h 1111
Setting p = xe/'ff on r I' and p = xe -/.. on r 2' and noting that the integral
around r 0 vanishes, we have
g&"(z) =- r(1 - ~}[iO e-«zxfJ-1e;flrr

2'TTj "" 1 - z-le- z
dx +1"" e-«Zx/J-1e-1fJrr' ]
0 1 - z-le- z
dx
r(l - {J) sin {J'TTi"" e-«"xfJ - 1

= dx
'TT 0 1 - z-le- Z
= 1-
r(fJ)
i""
0
e-a"xfJ - 1 dx
1 - z-l e-.,
(1.146)
i"
We note from integral tables that
-If'" fJ-l
e Xl
o 1 - z- e-'"
dx = r(p)lI>(z-l, p, (X), {J > 0, (X > 0 (l.I47)
This equation is also defined for {J integer. Therefore

g&"(z) = lI>(z-\ p, (X), for (X >0 and P> 0
(and also integer) (1.148)
It is known that the definition of lI>(z-1, {J. (X) is
lI>(z-x, p, (X) = i
n=O (n
1
+ (X)fJ
z-n (1.149)
which is the definition of the z-transform oif(t) = 1/(1 + (X)II, for P> 0,
a > 0, and T = 1. This result is expected because
f(t) = g.>-l[F(s)] = g.>-l[r(l - P)S,s-le,,·] = 1/(t + (X),s
for all p's different from I, 2. 3. 0 0 0 0
Althougp (z - I, ~. Ct) can be represented only in a summation form,

however. if Z-1 = I, the function <1>(1, P. (X) = '(P, (X) is tabulated for
certain (X and po This function is referred to as generalized Riemann-zeta
function. Therefore the z-transform does not exist in a closed form;
00
however. the infinite summation I I/(n + (Xt can be calculated numer-

ically for certain -x's and p's. ,,=0
£L'aluation of ~(z. m) ~ Zm[F(s)]
The relationship between the Laplace transform and the modified z-

transform can be readily obtained similarly to the relationship of the
z-transform. From equations (I. 93) and (1.123), we can write
o~(z. m)11 fT. A P(s. m) = .:1'1/*(/. m>l
'-j :/'1/( t 'f" III /') ,~ -,.(1) I. 0 .-~ m ..::
(I.I 'til)
This equation can also be written as
S;(z, m)I"ae7" = !l'[f(t - T + mT) ~T(t - T)]

=e- Ts .2"[/(t+mn8 T (t)]. Osm<l (1.151)
From the convolution theorem of Laplace transform this equation is

clluivalent to
:!F(z, m)I ••
1
= -. Z-l
lei-loO F(p)e mpT
,.T. t
-TIs- I dp,
27T} 0-1oe 1- e P
o~ m < 1 (1.152)
It is noticed from the preceding that to evaluate this integral we require
the change of variable parameter from m through A = I - m, so as to
ICCI the term e mpT such that the integral vanishes on the left haIf infinite
\cmi-circle. Furthermore, equation (1.152) constitutes the relationship between
:~ (2:, m) and F( s)j thus we define
- A
~(z, m) = Zm[F(s)) = -.t
Z-l
lC+i F(p)e mll'lo
", , I
-Tls- ,) dp,
27T} c-Joe 1- e I
o~ m < t (1.153)
Inlcgrating in the left-half plane and assuming F(s) has regular singular-
cq uation (1.153) becomes
1111''',
_
, pnlraof
. F( )emllT
.'1" (z m) = Z-l ~ resIdue of ---,,,-P,,-:::---:
l_e llTz- 1 '
Jo'(p)
~ m<1 °
(1.154)
Wllt'n I'~s) = A(s)/B(s) has simple poles, equation (1.154) can be
1'\ Jlll'~sed as
N A(s) e",snT ]
:~(z, m) = Z-l [ ~. --" ---..,..- o~ m < 1
n ·1 B'(s,,) I - e-TIS-'nl z~.T"
(I.I 55)
\\ 111"11' ·\'s, .f:!, ••• ,s-" arc thc simple roots of R(s) = 0, and B'(s,,) = ~: I
,. "'n
I.dlh' II of Ihe Appcndix lists extensive forms of the modified z-
II,tll',l'ollll for variolls forms of (i(s) or F(.f). This lahle can also he lIsed
I" "blalll :(/(.') ,:-I(i(s)llIsin~ the relalionship in clilialion (1.157).
''1110111011\ (I.I ~4) :Inti (1.155) arc :llsCl valid for", . , I, pl'llVilkti /-'(.1')
h,I·. 1"01 dITI'I'I'~ In .\' highl'" tll'lIolllillalnr Ihall 11111111'1'01101'. Il0wl'wl'. if
II I' Jril·. olliv lillI' tll'/~I't'I' in ,I' high,'!, tll'lIolllillalor Ihall 1I1111H'1':t11l1'.
''III,SIl'"l'' (I 1'1·1) :11111 (1,1 "'I) yidd rill' Ilwi!' inwl~I's, if tli.,nlllll1l1lilil·S
I \1',1. IIII' "lilli,,,, III 1111' "'1'1 .. illl· "I' 1111' di~I·ClIllillllilil·'. Thr vllltll' al I 0
'lllllllid Ill' III~I'II a', 1111 -. \I .111111 .... /1·111
28 THEORY AND APPLICATION OF THE Z-TRANSfORM
Evaluating the integral of equation (1.152) in the right half of the

p-plane, we (with due care for convergence) obtain similar to equation
(1.141) the infinite series form
F*(s, m) ~ ,!R[/*(r, m)]
=! k~OO F(S + jk 21T)e-l'+I.lZw,nHI-mIT,

Tk.-oo T
O~ m < It (1.156)
If F(s) is of two degrees higher denominator than numerator, this
equation can be extended for m = I.
The z-transform can be obtained as a special case from the preceding
by noting
.~(z) = z!F(z, m)lm_o (I.157)
and for /(0+) = 0
.~(z) = ~(z, m)lm_l (1.1 58)
The conditions for obtaining F(s) knowing either :F(z, m) or !F(z),that
is, Z.;;1, Z-I,are discussed in Chapter 4.
].6 Application to sampled-data systems14 - 16 ,ZO

One of the basic engineering applications of the z-transform theory is in
the field of sampled-data or digital control systems. Such systems are
used more and more often in modern technology. We shall briefly obtain
the z-transforms of certain systems configurations.
I. Let the sampled data systems be presented as shown in Fig. 1.8. The
z-transform of the output is readily obtained by applying the z-transform
to C(s) as follows:
C(s) = E*(s)G(s) (1.159)
Tv
,... ........., ---0 '8(z)
oE~) X~ I
G(s} ~ I 0 Cis). '8 (z. m)
FIGURE 1.8 A sampled-data system.
t It should be nOled Ihlll ir we lei", == I inside the SIInlJ11illilln lind before slimming,
this equation should he IlIl1llilird us
1"·f.I, -
I A
, ~ I'
," (
.f
J.,,)
I II.,. I 6,(11".
I. ".
Z-TRANSFORM DEFINITION AND TflEOREMS 29
T .-
,..../~ - - - < ) 'If(l)
I
R(s) +
I (.'(s). ~ (z. m)
FIGURE 1.9 A sampled-data feedback control system,
or
1"(,(:;) ,;1 Z[C(s)] = 1,"(=Y1(z), where t(::) = £*(s)l. r- I hI!
( 1.160)
lhe modified z-transform of equation (I.J 59) is
'fi(.:, 11/) ~ Z",[C(s)] = t(z)~(z, m) (1.161)
2, l.et the sampled-data feedback system be presented as shown in

li~, 1,9, The z-transform of the output transform ('(s) is given as
1"6'(.:) ~ Z[CCs)] = Z[£*(s)G(s)) = l,""(zY-1(z) (1.162)
0(::) = Z[£(s») = Z[R(s) - C(s)H(s») = Z[R(s) - £*(s)

x G(s)H(s)] = ,'1I(z) - t (zY1l ..*' (.:)
(1.163)
SlIhstituting the preceding in equation (1.162), we obtain
f(,(Z) = ;7f(z) :-1(z) (1.164)

1 + ,;r.'~(z)
Similarly, for the modified z-transform of the output we obtain
ft.'(
, :,'" ) = "Y [r»)
,,"'n \. (s = ';/J()
,'71 Z
~(z" m) ( 1.165)
1+ ,*,'~(z)
W,' I'an alslI ohtain Ihe syslem transfer function of any configuration
III ·.al1lpl~d-data or di~ital control system, In later chapters some of these
W·.II·I .... will h,l' .. I udicd in nHlI'I.' detail.
1.7 Mcan stillarc valuc Ihcllrcm I4 .:17 .:lH

IIII' 1lIlIlIwilll~
thl'lIn'lll with it, l·\tl'n~ill'" is wry tlwflll in til\' ,tlldy uf
tll'.III'II' W',II'II'"III parti"lIlar. till' 1\I('ail "Ilian' valtll' III' tla,' "lIl1lillllllllS
"11111 III ',nlllpkd 1IIIIn \'111111111 ,y,h'IlI' 1'1111 hI' n':ulily IIhlailll'll.
· 30 THEORY AND APPLICATION OF THE z-TRANSFORM
THEOREM:
(1.166)
Proof: From the infinite series form of the z-transform, we can write the
following relations:
1 "'QOO
Z[G(s}] = - 2 G(s + jkw,} + !g(O+} (t.t67)
T",--oo
1
Z[G(s}G( -s)] =....:.
"'.I
-00 G(s + jkw,}G( -s - jkw,) (J.168)
T "'--00
Letting s - jw in this expression and noting that G(jw) is the conjugate
of (I( -jw), we find that
1
Z[G(s}G( -S»)._I", = - 2 IG(jw + jkw,W
'=00 (1.169)
T "'--00
From the infinite series form of the modified z-transform, the following
can be written: .
G*(jw, m) =.!. "ioo G(jw + jnw,)e-m-m)(",+"""IT,
Tn--oo
o~ m < 1 (1.170)
Let
(1.171)
Then
G*(jw, m} =! "'ioo G{jw + jkw,}eifl-m)(",tk""IT,
T",--oo
o~ m < 1 (1.172)
Multiply equation (1.170) by (1.172) and note that since G(jw) is the
transform of a real valued function,
~(jw) = G( - jw) (1.173)
Thus
1 ["_00
IG*{jw, m)1 2 ="2 2 IGUw + jnw,W1
T "--00
n=eQ) 1c-co
+ 2 k-2
tl· 'f! fl.
G{jw + jnflJ,)
IIlk
X (;( -jill - jklfl,k Ilk ,,1,.. ,11 "'1'1'1.

()<"'~I (1.174)
The mean value of the second summation term, evaluated from m = 0
to m = I, is zero since for n ~ k, e-ilrw,(l-m)T is orthogonal to e- jnw,(I-III)T
with respect to the interval (0, 1) in m. Therefore
1 R=CO
IG*(j(l), m)1 2 = I~(z, m)/.=e/wr I2 = I i L IG(j(l) + jnw r)1 2
T 11=-00
(1.175)
By noting equations (1.169) and (1.175), the following theorem is shown:
Z[G(s)G( -s)] = T f'~(z, m)I.=,I""1 2 dm = T I~(z, m)I.~./Q)TI2,

o~ m ~ 1 (1.176)
The following two corollaries can be shown similar to the preceding proof:
Z[F(s)F( -s)) = T f.5F(Z, m).5F(z-I, m) dm, 0 ~m~1

(1.177)
:llId
Z[F(s)H( -s») = T f.5F(Z, m)Jt'(z-l, m) dm, o~ m ~ 1

(1.178)
Thc use of the extension of this theorem will be demonstrated in
('haplcr 4.
1.8 Equivalence between inverse Laplace
and modified z-transforms 14 ,39
1111111 clllHition (1.111), it has been indicated that we can obtain the
nllllillUoUS time function from the modified z-transform as follows:
lell, m)T ~ 1;;.l~(z, m)] = ~

27rJ )"
r !F(z, m)z"-l dz,
t = (n - I + m)T. (1.179)
o~ m ~ t, II = integer
\\ 111'11'
:1- (;;, m) :1 J ",II(l) I :!.\: ~",[/'ls)] (1.180)
I III Ihl'II1111 I'l', Ihe same funclion can l)l' ohtained from e1lualitlll (1.118)
11\1 II ',III f'. IIIl' iIIV1'I'SC 1.11 plncl' I!'ailSI'llI'll!. lIenee we hll VI: the 1'1111 owing
hlt'l1lll y
1(1l1, I" '1 ",I',. .'" ..'/' 'II"(.~)I ....' J",'I:~(;:, IIIlI.
n':III<I, " - ill\t')I,I'" O.II! I)
In some practical situations the form of F(s) contains combinations of
both eT • and s (mixed form), and thus the inverse Laplace transform
yields the time function as an infinite series. However, by using the
inverse modified z-transform, we obtain a closed form for the time
function.
This procedure is advantageously utilized to obtain (I) a closed form
for a convergent infinite Fourier series, (2) a closed form solution for the
response of a linear time invariant circuit to general periodic inputs, and
(3) the steady-state response (in a closed form) of such circuits by applying
the final value theorem for the modified z-transform. To explain in
detail these applications, we shall discuss the following three cases.
Summation of a Fourier series
Assume F(5) as follows:

1 1
F(s) - -- (1.182)
- 1 -e-T. s + a
The inverse Laplace transform is expressed as
1 iC+iGO e"
c(t) = - ds for t > 0 (1.183)
21Tj c-Ico (1 - e-T·)(s + a)
The integrand in this equation has simple poles at $ = -a and s =
± j(27rklT), k = 0, I, 2, . . .. Evaluating the integral by the sum of the
residues, we obtain
e-at 1 2 GO
c(t) =
I - ~T
+- aT
+ Tk-I
-I
X a cos (21TkIT)t + (27rk/T) sin (27rk/T)t (1.184)
a2 + (27rk/T)2
The steady-state part of this equation is
c(t)•• = ..!... + ~ i a cos (27rk/T)t + (2'1rk/T) sin (27rkIT)r
, aT TIt=l a2 + (27rk/T)2
(1.185)
If the substitution t = (n - I + m)T is inserted in this equation. the
expression is
() 1 2 ~ a cos 27rkm + (27rk/T) sin 27rkm
ct.,=-+-k ( 1.186)
aT TIt=l a2 + (27rklT)2
On the other hand, the modified z-trunsform of F( ...) is readily obtained as
:: t -"",.
.'F(z, m) = ":",(F(.~)J '" I I ~",(' ,~ II ) '"
I
..-- ',,.'
'
I:: /I ,.,' l , .. t ..
lI<m<1 (1.IH7)
The steady-state response is obtained by applying the final value theorem
-amT
C(/)$"
.-1
= lim (z - I)SF(z, m) = e
1- e
-a1' , 0< m <1
( 1.188)
Multiplying the numerator and denominator respectively by e4 T , we get
e"TU-mJ
c( I)
."
= ---=--
e"T _ I
( 1.189)
By equating the steady-state part of this equation to the steady-state

part of the inverse Laplace transform, we obtain
i a cos 27Tkm + (27Tk/T) sin 27Tkm = I eClTU-mJ _ ..!.. ,
k-l 02 + (27Tk/T)2 2 eaT - I 2a
0< m < I (1.190)
I hw, a dosed form of a convergent infinite series is obtained by the final
""I Ill' theorem of the modified z-transform,
Response of a linear circuit for periodic inpul 39
111'1 iVl' Ihe general equation for i(I). the current in the RL circuit shown in
I 'I'. 1.1 () when e(1), the applied voltage, is the periodic saw tooth voltage
"I,"" 11 in Fig. 1.11. Assume zero initial conditions, The Laplace trans-
""111 or the voltage £(s) is given as
i (t)
II \I . _1_,_, [~( I - e·-I,,) _ khe-''']
- (' ~. 52 S
R
I - (I + 0.4S)e-O.4.
(1.191)
(' ',. .. eW
IIII' 1IIIIl'Il1 Iransrorm /(s) is I.
II \ I
I';h) n.s 1 - (I + 0.4.~)e-O.4.
.,..
I< I 1..\ I 5 (5 + 0.5)
-" 2
FIGURE 1.10
(1.192)
w,· ;apply IllI' Illodilkd .::-lransl'ol"lll to Ihis "'I"alioll 10 ~CI
() ..'i.~ _, I· - (I +- 0.45),' n ..\.

I " I" 5 2(S I 0.5)
." ( I I. Il .1\ ),' n 4.,.)
~ '" ~~( \ I 0 'I)
I
II 1')\)
34 THEORY AND APPLICATION OF THE %-TRANSFORM
R= 1 ohm
L- 2 henries
.(t) h .. 0.4 second
T= 1.0 second
k-l
/11
~----!_ _~--:~ __
/ InT+h
..£o_...1: __ _+_
T+h nT Time
FIOURE 1.11
In view of the convergence required for the modified %-transform of the

last term in this equation for m < 0.4, we have to divide the response into
the following two regions:
1. o :s; m :s; 0.4 (1.194)
In this region the modified z-transform .f(%, m) is readily obtained by
utilizing equation (I.IOI) as follows:
J(%, m) == _1_[(2% - 1.185) e- ml2 + m - 2J

z - 1 % - 0.607
1
== - - "0(%' m) (1.195)
z-1
The steady-state current for region 1 is
i.. == lim J o(z, m) == 2.07e- mI2 + m - 2 (1.J96)
The total response for region 1 is obtained by using the inverse modified
%-transform as follows:
.(n, m )T ==
I -. i
1 -Z"-l- (2z - t.185 e-mIl + m -
2.".) r z - 1 z - 0.607
d% 2)
== 2.071e- mll + m - 2 - 0.07(0.607)n-le-mll,
t == n - 1 + m, 0 :s; m :s; 0.4
n == integer (1.197)
2. 0.4 :s; m :s; t (1.198)
In this region we obtain
i .. == 0.114 e m,l (1.199)
;(n, m)7'- ().1I4,· 111,111 -- «(U07}"). I •• (II - 1 + /PI).
0.4 "'" ". "'" I (1.2(JU)
Therefore, by using the modified z-transform and dividing the inverse
into two regions, we obtain the solution in a closed form. However, if we
had applied the inverse Laplace transform to ./(s), the response would be
obtained as infinite series which is difficult to plot and calculate exactly.
Steady-slale response of a linear time-invariant circuil"

Consider the steady-state voltage of vo(t) across the condenser in Fig. 1. I2a
for an input of the form of rectified sine wave shown in Fig. 1.12h. Vo(s)
in this example equals
V.o(s) = k 1 + e-'I'· _1_ = 1 V. (s) (1.201)
S2 + (.".IT)2 1 - e-'I's s + a J _ e-'I'. 01
where
(1.202)
The steady-state voltage is given by
vo•• = lim
,-I
<... [V01(S») = lim
,-1
2kZm[
(s
2
+ (.".IT)J 2)(s + a)J .
(1.203)
By using partial fraction expansion of + (""IT)2J(s + a) or avail-
1/[52
ahle tables of modified z-transforms, Zm[l/(s2 + (.".IT)2)(s + a»), can be
readily obtained.
Thus the steady-state voltage is obtained as follows
Vo•• = [2 +Q
k
(""IT)
2J [ 2e-(J~'I''I' -
I - e (J
cos.".m + QT
'IT
sin .".mJ,
o~ m ~ J (1.204)
III slllllmurizing the application of the final value theorem of the
IIIl11lilictl :-transform to obtain the steady-state response, the following
IWII t':llil'S arc discussed,
, - II HI
.x uilt}
,/ 'j
( ) ",ft) 110(0
I !
(Ill (M
IlCilllU· 1.1'-
CASE 1
(1.205)
where F 1(s) is a rational function of s and has a final value /ls' that can be
obtained from the lim sF(s). Hence the steady-state response
.-0
(J .206a)
CASE 2
(1.206b)
The sum of all T" is less than T.

In general, the steady-state f.s of this equation is composed of several
regions depending on (T,,); thus in obtaining the modified %-transform of
F3(s) the several regions should be observed. Mathematically. this
indicates that care should be taken in observing the convergence of the
integral along the infinite semicircle in evaluating the modified z-transform.
The example on p. 33 illustrates such a case where equation (1.101) can
be readily applied in calculating the modified z-transform for the various
regions.
1.9 Other transform methods
In this section various transform methods or operators other than the
%-transform which have been used in the analysis of discrete systems are
briefly discussed and enumerated. Pertinent references are given for each
of these methods and the reader can pursue further study as needed. The
close relationship between some of these methods and the z.tnlllsform can
be readily ascertained after digesting the material of this chapter. Finally.
this section will serve as an historical background to the z-transform and
other associated transform methods.
Laplace-Slieltjes inlegral4.5
This transform integral is defined as
L,[f(t)] = F(e,T) £ Loo/(t)e-,T drx(t) ( 1.2(7)
where rx(/) is the staircase function shown in Fig. 1.13.

The integral in equation (1.207) cun be rewritten as
I·.r/WI = f.· ..f(/k

.. II
., ,1,,(1) • i
" II
I<I/TIc' ".,..••
" I jIll (I ;"IIIK)
Z-TRANSfORM DEfiNITION AND THEOREMS 37
OI(t)
,..--
I
3T
2T
T
-3T
I
-TI
I
I
I
I
I
I
__ ...JI
FIGURE 1.13 Staircase function cr.(,).
III Ihi!; method. absolute convergence of the series is assumed, that is,
coo
I Ij(nT)z- 7I lz_ t f" ~ M (1.209)
71-0
wh,'re M is finite. This condition is satisfied, for example, if j(t) is a
1'"lynomial, but not if j(t) = e,l. The magnitude of Iz-"I"js e-" 7'o, and
II (1.2(9) holds for some a, say ao, it would hold for any a > ao.
!lased on this definition. the use of Dirac (delta) function in the defini-
111111nr the =-lransform could be bypassed and thus retain a certain rigor
III Ihl' mathematics of this transform. It may be noted that the Laplace-
Sllc·lc.ics integral reduces to the one-sided Laplace transform when ~(I) = I.
71,(, xeneraliud Laplace transformS

IIII'. IIIdhnd, which has been introduced by W. M. Stone for the opera-
111111,11 ,,"lution of difference equutions. is defi!led as follows:
. \ or: 1-'(11)
/.I/·WIJo: =f(·~) = l . - (1.210)
" /I S"II
\\llh III,· l'IIIHlilinn limf{.\') C'" n. The symhol H dennle" a shiflin~

" .'1
"111"1,11111 1lt-lilll'lI hy IIIl' rl'ialiclll HI. I II FI·"(I.),

lIy 1I"IIII~ IIII'. 11:I1I .. 1'11I1tI. Stiliit' ha" llt-wIIlP"cl all c'Xll'lI,iw 1i .. 1 III'
11'\11',1111111 l'ail .. (I"(/,) 1(,\,)1. whidl ,1It'l'I""dy u'lah'clll' IIIC' . 110111'01'''1111
111111 ..... ",n'pl 1111 1\ lill'lIlI ' - c,"'··,
;y=Ix
4 r--
I
3
2
FIGURE 1.14 A jump function.
The jump function method?

This method was introduced by Gardner and Barnes and is similar to
the Laplace-Stieltjes integral discussed earlier. The transform in this
case is deli ned as
6 (00
yes) == 9"[Iy(x)] == Jo Iy(x)e-·:e dx (1.211)
where Iy(x) denotes the jump function of y(x). If y(x) == :c, the jump
function of y(:c) is as shown in Fig. 1.14 and gives
00
Iy(:c) == I:c == Irp(:c - r) (1.212)
r-l
where p(:c) denotes a unit pulse of duration one, that is,
p(:c) = u(:c) - u(:c - I) (1.213)
where u(:c) is the unit step function.

In general, the jump function Iy(:c) can be written as
00
Iy(:c) = !y(r)p(:I: - r).
r-O
where p(:I: - r) £, u(:c - r) - u(:c - r - J) (1.214)
Dirichlet series transform8

This transform has been proposed by T. Fort and has the following
definition:
'r,
f(.~) .£ /)[,,(1)) ... ~ m "tI(I),
,·-0
fur III > I,!I > I, I - intcoller (1.21~)
z-TRANSFORM DEFINITION AND THEOREMS 39
It is of interest to note that this transform reduces to the z-transform
when m = e, 1 = nT, that is,
..,
/(s) =I a(nT)z-II £ ~(z)I,_.'" = J[a(nT)] (1.216)
.. -0
This change of notation could also be performed without influencing

the application of the method.
Discrele Laplace Iransjorm 9 •41

The discrete Laplace transformation, which is denoted by the operator
symbol ~, is a transformation applied to the sequence (fll}' that is,
00
·~{fn} ~ ~ i"e-"' == F*(s) (1.217)

,,=0
This transformation can also be described as a Laplace transformation
applied to a certain pseudofunctionl*(t) described as
01)
1*(t) == /(1) 2 ~(t - nT) ( 1.218)

n=9
I"hcrcfore the!!) transformation may be considered as a special case of the

I apllicc transformation.
The discrete Laplace transformation has been applied extensively to
discrete system theory by Tsypkin and his colleagues in the U.S.S.R.
More recently. Doctsch U has presented an illuminating exposition of this
1II1'Ihud and its relationship with both Laplace transform and the z-
1IIIIlsfunn methods.
1.ollren/-Cauchy·s transjorm lO
II .... lransfurm is again related to the z-transform or the generating
'ullrlinn method and is defined as
(1.219)
S"!/~ "P,II',lIm· II •44
:\ 11,.","1 III",ralll" Oil sl"llIl'lIl'l'S is 111l~ shift opcralor H, dclillCll hy Ihe

,,'1,1111'11" II" I FII,. Thl'lI II". 1':11" I /'711,. 111 gl'lIl'l'I1l,
I,. "", . - II"" (I.n!))

If u(t) is a function of the continuous variable t and if T is a given
constant, we define the shift operator E here by the formula
Eu(t) = U(I + T) (1.221)
or, more generally,
ERu(t} = u(t + nT) (I.222)
In applications to difference equations, the shift operator plays the
same role as the differentiation operator D in differential equations.
B. M. Brown has applied this method extensively to the analysis and
design of discrete systems.
,- Transform method!3
This transform, which is applicable to sampled-data systems and to
certain difference equations, is defined as
co
ya, e) = ~~Y ..+a} = I y,,+&(l + ,)-R,
'--I.
, =e •- t,
n-O
0 ~ e~ 1 n = 0, I, 2, . . . (1.223)
The inverse integral transform is given by
y,,+& =~ .1
2'1TJ Jpole8 or Ule
Ya, e)(1 + ,)"-1 d"
Intesrand
o s: e s: I, n = 0, I, i, ... (1.224)
The advantage of this transform is that the Laplace transform is readily
obtainable as a limiting process, that is,
Yes) = .9'[y(t)] = r"" y(l)e- sf

dt = lim TY(', 0) (1.225)
Jo 7''''0
Mikusinski's operational calculus42.43

Mikusinski's operational calculus is the most general method that en-
compasses both the continuous and the discrete theory. The z-transform
method can be obtained as a special case of a series of the translation
operator defined by Mikusinski. In this operator the numerical cocm-
cients are the values of the function at the sampling instants and the
exponents are 0, T, 2T, ... , nT. . . .. We can define for the functiun
[j(t)) the operator
,"
f· = ll<'''I')/.,,'I' ( 1.22f,)
.. II
For IIltT we can write e-" 7\ and thus the identity between the operator
f* and the z-transform can be established. The modified z-transform can
be also obtained from Mikusinski's calculus by introducing a parametric
operator with respect to the parameter l:!. 1 - m. This operator is =
defined in the whole domain 0 ~ l:!. < I, 0 ~ t < co.
In concluding this section of the survey of the various transform
methods, we may mention that any of the methods could be effectively
used in the treatment of discrete systems. They represent one and the
same thing with different form or notation. Which method is preferred
depends on the ease and familiarity to the reader.
Appendix. A method of determining the

coefficients of the z-transform expansion 9 •47
From equations (1.55) and (1.56), (letting qo = I). we obtain the time
l'ocfficients f(11 T) as follows:
f(OT) = Po
f(l T) = Pl - f(OT)ql
f(2T) = P2 - '1./(1 T) - qd(OT) (I)
f(nT) = Pn - L" f[(n

;~1
- i)T]q;
hjuation (I) can be written in a determinant form as follows:

o 0
ql 0
f(nT) = q2 ql o
for 11 ~ I (2)
q" q" 1
,llId 110 n ~-= Pn.
Iqllalillll (2) is readily used ror f,,(m) by suhslitliling for I'll' Jl,,(m).
11I1I11I'11II1l1l',llll'l'lIl'l1kit'III'/;"A(k>- I)t'all hcoblaillcd as
"
III I A ~ I;, ,A ,If, (.1 )
, I
REFERENCES
1. DeMoivre, Miscellanes Analytica de Seriebus et Quatratoris, London, 1730.
2. Laplace, P. S., Theorie Analytique de Probabilities, Part I: Du Calcul des Functions
Generatrices, Paris, 1812.
3. Seal, H. L., "The Historical Development of the Use of Generating Functions in
Probability Theory," Mitteilungen der Jlereinigung Schweizerincher Verischerungs
Mathematiker, Bern, Switzerland, Vol. 49, 1949, pp. 209-228.
4. Widder, D. V., The Laplace Transform, Princeton University Press, Princeton,
New Jersey, 1946.
S. Helm, H. A., "The z-Transformation," B.S.T. Journal, Vol. 38, No.1, 19S6, pp.
177-196.
6. Stone, W. M., "A List of Generalized Laplace Transforms," Iowa State College
Journal of Science, Vol. 22,1948, p. 21S.
7. Gardner, M. F., and J. L. Barnes, Transients in Linear Systems, John Wiley and Sons,
~ew York, 1942, Chapter 9.
8. Fort, T., "Linear Difference Equations and the Dirichlet Series Transform," Am.
Math. Monthly, Vol. 62, No.9, 1955, p. 241.
9. Tsypkin, Y. Z., Theory of Pulse Systems, State Press for Physics and Mathematical
Literature, Moscow, 19S8 (in Russian).
10. Ku, Y. H., and A. A. Wolf, "Laurent-Cauchy Transform for Analysis of Linear
Systems Described by Differential-Difference and Sum Equations," Proc. IRE, Vol.
48, May 1960, pp. 923-931.
II. Brown, B. M., "Application of Operational Methods to Sampling and Inter-
polating Systems," IFAC Proc., Vol. 3, pp. 1272-1276.
12. Van Der Pol, Balth, and H. Bremmer, Operational Calculus Based on Two-Sided
Laplace Integral, Cambridge University Press, London, 1950.
13. TsChauner, J., Einfiihrung in die Theorie der Abtastsysteme, Verlag R. Oldenbourg,
Munich,1960.
14. Jury, E. r.. Sampled-Data Control Systems. John Wiley and Sons, New York, 19S8.
IS. Ragauini, J. R., and G. F. Franklin, Sampled-Data Control Systems, McGraw-
Hill Book Co., New York, 1959.
16. Tou, J. T., Digital and Sampled-Data Control Systems, McGraw-Hili Book Co.,
New York, 1959.
17. Jury, E. I., and F. J. Mullin, "A Note on the Operational Solution of Linear Differ-
ence Equations," J.F.I., Vol. 266, No.3, September 1958, pp. 189-205.
18. Neufeld, J., "On the Operational Solution of Linear Mixed Difference Differential
Equations," Cambridge Phil. Soc., Vol. 30, 1934, p. 289.
19. Bridgland, T. F., Jr., "A Linear Algebraic Formulation of the Theory of Sampled-
Data Control," J. Soc. Indust. Appl. Math., Vol. 7, No.4, December 1959, p. 431.
20. James H., N. Nichols, and R. Phillips, Theory of SerIH)meC'hani.fnrs, M.I.T. Rad.
Lab. Series No. 2S, McGraw-Hili Book Co., Ncw York, 1947, Ch. S.
21. Meschkowski, H., Di,ffc',,'"z,'ng',';c'I""!I{I'", (illttcngcn Vllndenhucsch 111111 Ruprecht,
Basclund Stuttgllrt, JlJ~').
22. M."l1tel, I'., 1.1'e"III.r .mr 11'.1' H"I'/IIUII,'I',' rt 1.l'lIf.r AI'I,III·lItf",I.f, (illUlhicl' Villllrll,
1'lIri~, 1·'~7.
23. Miller, K. S., An Introduction to the Calculus of Finite DiJference and Differential
Equations. Henry Holt and Co., New York, 1960.
24. Kaplan, W., Operational Methods for Linear Systems, Addison-Wesley Publishing
Co., Reading. Massachusetts, 1962.
25. Aseltine, John A., Transform Method in Linear Systems, McGraw-Hili Book Co.,
New York, 1958.
26. Cheng, D. K., Analysis of Linear Systems, Addison-Wesley Publishing Co., Reading,
Massachusetts. 1959.
27. Jury. E. r.. "Contribution to the Modified z-Transform Theory," J. Frankli"
Itlsl .• Vol. 270, No.2, 1960, pp. 114-124.
28. Tsypkin, Y. Z., DiJferenzenleichungen der Impuls-und Regeltchnik, Veb Verlag
Technik. Berlin, 1956.
29. Doetsch. G .• Handbuch dl.'r Laplace Transformation, Birkhauser Verlag, Basel and
Stuttgart, 1956.
JO. Friedland. B., "Theory of Time-Varying Sampled-Data Systems." Dept. of Electri-
c;11 Engineering Technical Report T-19/B. Columbia University, Ncw York. 1956.
.11. Barker, R. H., "The Pulse Transfer Function and its Application to Sampling
Scrvo Systems." PrOf. 1.£,£,. (London), Vol. 99. Part IV. 1952. pp. 202-317.
12. Lago. G. V.• "Additions to z-Transformation Theory for Sampled-Data Systems."
"'rtll/.~. A.I.E.E., Vol. 74, Part II. 1955. pp. 403-408.
1.1. Truxal, J. G., Automatic Feedback System Synthesi.s, McGraw-Hill Book Co.,
New York, 1955.
1.\. I.awden. D. F.• "A General Theory of Sampling Servo Systems." PrOf. I.E.E.
II "mlon). Vol. 98. Part IV, 1951, pp. 31-36.
1\. Ragazzini. J. R., and L. A. Zadeh. "Analysis of Sampled-Data Systems." Trans.
·I.I.!:'./:·.• Vol. 71. ParI II. 1952. pp. 225-232. i
lit. l'l'ccman, H.• and O. Lowenschuss. "Bibli6graphy of Sampled-Data Control
S~'~ICIllS ami :-Transform Applications," I.R.E. Trans. IAulomalic Control (PGAC).
Mardi 1958. pp. 28-30. .
t I JIII'Y, F. 1.. and C. A. Gaitieri, "A Note on thefnvenk z-Transform." I.R.E. Trans.
I 'II'/lil l1l1'rJI:r (Ctlrrl'sp(mc/{'I/('e), Vol. CT-9, Septe r 1961, PP', 371-374.
tK. Mllri. M.• "Statistical Treatment of Sampled- ala Control Systems for Actual
1(.111<111111 Inputs," Trun.~. ASME, Vol. 80. No.2, February 1958.
N . .Iury. E. 1., "A Notc on the Steady-State Respon~ of Linear Time-Invariant Sys-
"·111·.... I.R.I,·. Prllc. «·tll'rI'.rptlllt/I'l/fl.'), Vol. 48, No.5, May 1960.
·111 Wi1t~, (', 11.. 1'l'il/rip"·.I· tlf FC'l'db(lck etlllimi. Addison-Wesley Publishing Co.
1I.·.ulilig. M;I~sachllsc\ls. 11)(,0 (Appcmlix n. p. 261).
·11 I )OIcI,,:h. (;uslav, (;lIid.· 111111.· Al'l'li('(llitllu ..rl.(ll'lm·,· Tr(l/I.rji'rm, D. Viln Nostrand
t 01 01'1(.1. (·h. 4.
I,' l\'ltl.u,ill,J..i •.Iall. 01",,,,,'i,,,,,,1 (,,,1"11111,,. i>cl'J:alllnn I'rcs~, Ncw Yllrk. V.. 1. K, 1')5').
II 1'.1'."'111. Iialln·hm·. "lh'lallll/l,hip 1ll'IWC\'1I MiJ..u~ill\J..i·s 0p"rauIIlIal ('"I,'ullls alld
lilt' 1\1."hll.·.1 11.",,1 .."11." Ma''''1 IIfS.-i,·tIl·,· I'I"i",·I. ''''1'1. .. I' I .I .• tllllVl't',ily Clf
t ,lhlllllll;l. "'·It.,·I,·y ..hllll· 1'1,.. (111111"1 ',up"lvhlllll .. llh,· :lIl1hlll).
+I lit "" II. II 1\1. /I,.. 1111111"'/11,,1,...,1 n"·,,, I' ,./1 ",,',1/ ,\"·,It,·"". I.. hll Wlh-y .111.1 S....•••
I It \\ \ til". 1"" I. t h I I
4S. Volgin, L. N., Elements of the Theory of Control Machines, Soviet Series, Moscow,
1962.
46. Kuo, B. C., Analysis and Synthesis of Sampled-Dala Control Systems, Prentice-
Hall, Englewood Cliffs, New Jersey, 1963.
47. Tsypkin, Y. Z., Theory of Linear Sampled-DQtQ Systems, State Press for Physics and
Mathematical Literature, Moscow, 1963 (in Russian).
48. Tamburelli, 0., Alternate;:..Transform, Alta Frequenza, Vol. 30. No. lQ, October
1961, pp.74S-7S0.
2
z-TRANSFORM METHOD OF
SOLUTrON OF LINEAR
DIFFERENCE EQUATIONS
The field of linear difference equations has been studied extensively by

many mathematicians in the last century or more. Several excellent
textbooks! 8 have been written on this subject that can be studied by the
interested reader. Several methods exist for solution of difference
equations such as the classical method, the matrix method, the recurrence
method, and the transform methods. In this chapter we shall apply the
:-transform method (or widely known among mathematicians as the
!tenerating function method) for the solution of only certain types of
linear difference equations. The formulation of the difference equation
can be expressed in several forms such as the backward and forward
dilTerence method or the translational form. The theorems developed in
the preceding chapter will be readily applied to any of the several forms
used in the mathematical formulation of these equations. Several examples
l'nlln network and system theory will be presented and solved, using the
,lilTerence equations as the basis of description.
Although this chapter is necessarily a brief study of difference equations,
",'vera I types of linear difference or difference-differential equations,
usually encountered in physical. economic. and physiological systems,
will he studied. This study will be based primarily (although other
IIll'thmlli exist) Oil the ::-trunsl'orm or the modified :-lransform theory.
2.1 I.incur dilTcrcllcc c«.Juuti()ns

with ,,'OI1Slillll c()cllicicnts2H .:1I
Ilw rC1l1nwill~ ,IiITcrcncc Cllllillil1l1.
,.,
(·I1I1!.icll'"
,." ",.1'"
~ I' ......
v
'.11" t. (1.1 )
, II ,."
46 nn·:ORY AND APPLICATION OF THE Z-TRANSFORM
tI,lind h, are constants, Y .. is the forcing function and is known for all
villues of n ~ O. It is desired to find z". z" and Y,. are both functions of
thr pnmmeter 1= nT. Assuming T = I in all the discussions of this
dUlpter. I = n. where n is an integer equal to or greater than zero.
Tilking the z-transform of equation (2.1). we obtain
(2.2)
wIH~I'l' .'1'(:) ~ j[znJ. qv(z) ~ 1[Y,.], and zl' Ys are the initial conditions.
SCllvill~ equation (2.2) for ~(z), we obtain
:"(z) =
'W(z) i
1-0
biZ; + f QiZi(I1Z/Z-/) -
;-0 1-0
f b•.zi(!yr- /)
;=0 1-0
p
IQ;Zi
i-O (2.3)
The inverse z-transformation zn A 1-1[.~(z)J can be accomplished by
""illf~ Cauchy's integral formula, the power series expansion, the partial
II IIl·tiCln expnnsion, or the available table of inverse z-transforms. It is
"l'plln~III frum the foregoing that initial conditions are readily inserted
IIl1hlllllllicnlly ut the start, thus avoiding evaluating the integration con-
IIIIIlIls liS in the clnssical solution of difference equations. This fact with
Iht' Wit' III' lahles of z-transform pairs indicates that this method of solution
I... " ,'(,l'llIin ndv~\Otages over the other methods.
Thl" l"xtellsinn of the z-transform to solution of simultaneous linear
,hllrlrlll't' "'JlUltions with constant coefficients can be performed in a
11111111111 fllshilln ns for the preceding case.
II.IIISTR.ATIVE EXAMPLE·'
II Irll"il'ed tn find the current in the nth loop i" for the ladder network
III
IIhuWII ill H~. 2,1. Assume all the resistances except the load R£ have
Ihr slime villue It
lie II Ilet I I Iluhl....... IWill It.

Z-TRANSFORM SOLUTION OF LINEAR DIFFERENCE EQUATIONS 47
The equation for the loop with current in+ \
(2.4)
Equation 2.4 is true for any n except -I and k - I (the beginning and
end loops). Equation (2.4) with end conditions is sufficient to describe
the network.
Applying the z-transformation to equation (2.4) and dividing by R, we
obtain
,F(z) - 3:f(z) + 3z;0 + z2.1(Z} - Z2;0 - zi\ = 0 (2.5)
Solving for ..f(z), we have
...f(z) = z(zio - 3;0 + i 1) (2.6)

Z2 - 3z + I
We can eliminate one of the initial conditions by writing the equation
for the first loop of Fig. 2.1.
2Rio - Ri\ = V (2.7)

or
'I. = 2''0 --RV (2.8)
I'Iws .1'(.:::) becomes
fez) = z{z - [J + (V/Rio»}io (2.9)*

z2-3z+1
h'ol11 tables of inverse z-transform (Table I) we readily obtain ; .. as
1"lIows:
i" = ,:y-l[..I'(Z)] = io[COSh won + I - ~Rio) sinh won J (2.10)
I 3
l'US 1 /"11 = - (2.1 J)
2
!oillh /"n ::
J"5., (2.12)
2
• IIII'. 1'11",1111111 nlll hI' III", Wlilh'n inn 1'111'111 ~lIilllhll' fill IlIhll' rnllil'~ ;IS
II I ,'16, 11'/ Rill) I

II \. I I
The value of io can be found by substituting equation (2.1O) into the
equation for the end loop and solving for i o• The current in can also be
obtained from the power series expansion of (2.9) in powers of Z-I.
2.2 Solution of difference equations whose

coefficients are periodic functions ll8 ,31
In the solution oftQis type of difference equation the modified z-transform
iii readily applied. Consider the following type of difference equation:
p (/
x{t)
(n+p) s: t s: (n+P+ll
! h(t)xnH + i-O
= i-O ~ gi(t)Yn+i (2.13)
I'ere /;(1) and gi(l) are arbitrary periodic functions of the independent
vllriahle I with periods equal to unity.
UcclIuse of the periodicity of/;(I) and gi(t), we can replace these functions
hy 1;(111) and gi(m) , respectively, where m is a parameter which varies
frolll zero to unity. With this change of variable equation (2.13) becomes
p (/
x(n + p + m) = i-O
~ };(m)xn+i + ~ gi(m)Yn+i
i-O
(2.14)
I'wm Ihe definition of the modified z-transform, we have
","[x(" - 1 + m)) ~ ,q-(z, m), (2.15)
.1mfx(n + k + m)] £ Zk+l[,q-(Z, m) - Ix(j + m - 1)z-/J,

i-O
o~ m ~ 1 (2.16)
'I'hl' lIIotlificd z-tran!.form of the difference equation in (2.14) gives
Z"'I[.,y·(Z, m) - Ix(j +m - I)Z-IJ

. I' 0
= i~.r.(m)Zfr(Z) - XXSZ-1J
I i R.(m):tr·:q(Z) - }::lIIZ '1,

• " I-~
():5:m~1 (2.17)
nllih ,n., IIl1tl ;n:,

III) lilt' Ullknown in Ihis ('tllllilion. lIt1wc"vl'r, :1'(z)
ht' ,11'11'11111111',1 hy noling Ihlll :V'(:) - :n·,III)I", ,. Thlls,sllhslillllin~~
1'1111
", - I III rllllllllOIl p.1 II gavt''I Il'lIllily :n'). Tht' n"lIlliu/', ""pn" .. itlll
1.111 Ill' 1111111,1.11111'11 III () Ill, whit-II ellll Ihl'lI ht' '1IIlvI'cI 1'111 .n . , m). Thill
z-TRANSFORM SOLUTION OF LINEAR DIFFERENCE EQUATIONS 49
would finally give
,q"(z, m) =
C?!I(Z)[
11+1
f g1(1)zi f h(m)z' + I
.-0 .-0
Q
gi(m)zi
]
z z,,+1 _ f h(l)z' i-O

t~o
x [~oh(m)zi % ziz-i + i~glm)zi ~ y;Z-i] (2.18)
The time function is obtained by the inverse modified z-transform

t = (n - J + m),
0!5: m !5: J (2.19)
ILLUSTRATIVE EXAMPLE
hnd ;r(t) which is a solution of the following equation.
x(t) = :1:"+1(2.582 - 1.582e-1 - t) - x,,(J - e-I )
+ Yl/+l(e- I + t - 1) + 0.418Yn(1 - e-') (2.20)

Wc CCln construct the coefficients of X/t and ?Ik to be periodic functions
Wllh pcriod one by replacing t in these functions by m. This gives the
I .. llowing clJmuion:
x( n + J + m) = xn+ 1(2.582 - 1.582e- m - m) +- xn(l - e- m )

+ ?I" I(£' + III - 1)
I III
+ 0.418!1,,(1 - (' ./11), 0 !5: m !5: 1 (2.2L)
1\ 111'11' III" ~'lldlkil'lIl!1 ur ."" II' ."". /1" .1_ !/" Clrc pcriodic.
IIII' IIlllllilkd :-lransJ'1lI'1II ul'CtluHtion 12.21) givcs
l.}I:f'(1..m) Ix(rn 'l)+x(mh Il}
(2.~K2 1.5.'2c' III m)l,
1·"'( l) xIII (I c' "').'1"( l)
. Ie,. "' I '" 1) 1.1 .y( 1.) )'11 I
I 1I.oIIH II ,. "''''''11.1 c.'.J ,,)
1.6
1.4
1.2
t 1.0
0.8
-=-
wO. 6
0.4
0.2
°0~~--~--~~~~--~--~~~~9
ra_
FIGURE 2.2 Particular solution for 1/.. = a unit step.
To obtain ~(%). we replace m = 1 in equation (2.22) to obtain
~(%) = (0.368% + 0.264)tzV(z) + 222:0 + %(2:1 - 2:0 - 0.368yo)

Zl - 21 + 0.632 Zl - 21 + 0.632
(2.23)
If we assume y,. = unit step (Yo = I) and 2:0 = 0, we readily obtain
from equation (2.21), by letting n = -1, m =
I, the value 2:1 = 0.368.
Therefore ~(z) becomes
o'!'(%) = (0.36821 + 0.264)% (2.24)

(z - 1)(21· - Z + 0.632)
Substituting (2.24) in equation (2.22) and noting that 2:(m - I) = O. we

obtain
%2[~(Z, m) - x(m)z-I] = (2.582 - l.S82e- m - m)~(z)
- (1 - e-m)~'(z) + (1 - e- m)-%-
%- 1
%2
+ (e- m + m - 1 ) - -
21 - 1
- (e- m + m - 1)% (2.25)
The initial value 2:(m) can be obtained from equation (2.21) by letting
n = -I to give
x(m)losms 1 = yo(e- m + m - 1) = e-'" +m - I (2.26)
Substituting equation (2.24) and (2.26) in equation (2.25) an~l. expandinl
in powers of Z·". we ohtain the particulnr Rulution .t·pC z. m) "~ f()lIuw!l:
z:"p(:' III) .... (I' III ·1 III" I): II -I- (O..\fIH I O.t,]2m)..: I
-f (!.f"2 O.M1,. Ill).. • -I· . .. (7..27)

The coefficient of Z-k represents x,,(t) for k ~ t < k + I. The final
value of X,,(I) which is known to exist is obtained from applying the final
value theorem to !r(z, m) to give
lim x(n - t + m) = lim (z - l)~,,(z, m) = 1 (2.270)
n"'''X) z-I
The particular solution is shown in Fig. 2.2.
2.3 Linear difference-differential equations5 ,36

In many applications of control systems a pure delay (or transportation
lag) often appears in the controlled process. The mathematical description
of such systems is carried through the use of mixed difference-differential
equations. In the following discussion only linear constant coefficient
equations involving one-time delay are considered. Extension to other
cases can be performed using the same or other techniques.
The modified z-transform of the continuous functions C(/) can be
written in the following form:
z<if(z, m} = co(m) + c1(m)z-1 + ...
co
+ c ..(m)z-" +... = 11-0
I c,,(m}z-n
t = (n + m)T, 0 ~ m ~ 1 (2.28)
From this series expansion of H(z, m), we may note that
co(l) = Cl(O), Cl(1) = cz(O), e..(I) = c ..+1(O),
for n = 0, I, 2, ...
for continuous outputs as assumed in this case. Using the preceding form
of the modified z-transform of C(I), we can also write the transforms of
the derivatives of c(l), c(l + n) with T = I, from equations (1.116) and
(1.117) as
a
.1m[c'(t)] = am [<if(z, m)], (2.29)
,1m[c'(t
a
+ n)] = 'om Im[C(1 + n)]
= ~ {ZII l[ZW(Z. m) -'i\'I,(m)::: pl} (2.30)

om I' II
Similm rclatitlnships could he writtcn for highcl'-urdcr derivatives.

In Ihe soluliun III' dilTcrcnce-diiTcrential c'llIalillns. it is t~llnvenitml to
l'IloUSt' Iht' sampling p,~rill(l as t~'1'1lI1 ICI Ih,' lin,,! tlt'lay itsdf. Thi~ will
yicltillu' J"L'sJl"nM' ill II dClSt~tl UI" l'UI,,:isl~ I'urm wilh Ih,' minimulII III' lahul'.
Wt' !.hall 11111,11 all' lilt, pml'I',lmt' hy sulYinl~ lilt, l'ulluwlIIl~ Iwu t'lIlIlIIplt's
III ""11111.
EXAMPLE I: FIRST-oRDER EQUATION

Consider the difference-differential equation
c'(t) + ke(t - I) == 0 (2.31)
The boundary condition is specified for a sampling interval; here let
c(t) == Pfor 0 ~ I ~ I, where both k and Pare constants. We must also
assume that there is no discontinuity at t == O.
By taking the modified z-transform of equation (2.31),
~ c;f(z, m) + kz-l~(Z, m) == 0 (2.32)

am
that is,
a
z - tff(z, m) + kfif(z, m) == 0 (2.33)
am
This can be solved as a first-order differential equation in m, treating z as
constant.
The solution of equation (2.33) is
fif(z, m) = A(z)e-'k/~)m (2.34)
where A(z) is to be determined. Now from the definition of the modified
z-transform in equation (2.28)
fif(z, 1) + coCO) = ~(z, 0) (2.35)
that is,
A(z)e-k/. + (J = zA(z)
Therefore
zA(z) = p. (2.36)
1 - z-le-kl •
Substituting in (2.34)
t"A(
ZlI>'Z,m ) = __ ...:.P--:-...,....,.... e-U:/dm (2.37)
1 - ,,-Ie-Ill.
Expansion of.<2.37) in powers of Z-I ,-le/& gives
COl
zfif(z, m) = (Je-Ikldm 1: z-fle-Ikls)n
ft.O
= PI z-fte-Ikls)'n+ml
COl
(2.38)
n-O
The coefficient of z-n will give the respon5c (',,(m) and if we cVlllulltC
this for all n we get the cnmplclc rcspnnllc.
UI(..
z·.,M. t" ) - I'11-0
" ..
... M
","~ I-(Ie/:)(II f·",W (2.11)
v~n III
Z-TRANSFORM SOLUTION OF LINEAR DIFFERENCE EQUATIONS 53
This can be simplified as
f'L7(
Z'l)' z, m) ~ Z-/~
= (J £.. £..
[-k(/ +m- p)]p (2.40)
1-0 ".0 p!
The coefficient of Z-I is
ez(m) = {J ±
'P~O
(-k)"(/ +m-
p!
p)'P (2.41)
Equation (2.41) is useful if one is interested in the response for large

intervals of time. Generally, the process of obtaining a general time
function such as (2.41) in a closed forin is very difficult if not impossible.
In many engineering problems, we are interested only in the transient
response. This can be obtained by actually expanding equation (2.38)
and identifying the coefficients of Z-I, Z-2, •.• , etc. This approach is
illustrated in the following example. The coefficients of Z-I, Z-2, •.• will
give the responses for the 1st, 2nd, ... sampling intervals in terms of m
which ranges from 0 to 1.
EXAMPLE 2: FEEDBACK CONTINUOUS SYSTEM WITH TIME DELAy31
Consider the feedback system shown in Fig. 2.3. The forward transfer
function is G(s)e-~~ = [(s + 0.3)/S2] e-'. The overall transfer function can
he written as
c(.~)
-=
(s+ 0.3)e-· (2.42)
R(s) S2 + (s + 0.3)e-·
Alllwugh we can obtain the modified z-transform of this equation directly,
however. for sake of illustration we formulate first the difference equation.
The difference equation is. for r( t) = tu( t).
c"(t} + e'(1 - 1) + 0.3c(l - 1) = r'(t - 1) + 0.3r(1 - J)
= U(I - J) + 0.3r(1 - J)
(2.43)
I 'IlIlltion (2.43) is also equivalent to
e"H(t + I) + r'(I) + 0.3c(r) = U(I) + 0.3r(t) (2.44)
wllh Ih~· inlli"r conditions specified as
('(I) = 0, (·'(t) =0 for 0 ~ r ~ 1 (2.45)
e(t)
I I( IIIIU ], I ('lIl1liIlIlIlU~ It-I"'dhll~'k s'yslem wilh p"re ddllY.

54 Tlllln"V AND A.'LIC'ATION 01' 'H' ..'UNIPOI..
Taking the mmliflcd ,·trlUllIl'nrm ul' (1.44)
z
a' ~(z, m) + ama 'l(z, m) + 0.3'.-(a, ttl)
amI
= _1_ + 0.3[ mT + __
T_ ..] (2.46)
%- 1 %- 1 (% - 1)1
Treating z as constant, (2.46) is solved as a differential equation in m, land
noting that T the sampling period = I, we obtain
~(z , m) = A1e'lm + AI e..... + (z _1 1)1 + (%_m_

_ 1)
('2.47)
where Al and AI are functions of % and $1> Sa are roots of the characteristic
equation
ZSI + s + 0.3 = 0,
.
that IS, S1,I =
-1 ± I - 1.2%
2%
J
(2.48)
Again from the definition of the modified z-transform we have
tCf(z, m)I"'=1 = [ztCf(z, m) - co(m)]I",_o (2.49)

and
~
am <;f(z, m)1 m-l
== ..!. [ztCf(z, m) -
am co(m)]1
m-O
(2.50)
provided that C'(I) is continuous at t = 1,2, ....

Substituting (2.47) in (2.49) and solving for Al and As yields
1
ZAI = - - - - - , - - - - (2.51)
(SI - S.)(Z-l e'l - 1)
-1
zAa = ---~-- (2.52)
(SI - S.)(%-l e,· - 1)
The complete solution for H(z, m) is

1 1
z~(z m) == e'I'" - - - - - - , - - - -
, (SI - SI)(%-le'l - 1) (SI - SI)(%-l e,· - 1)
X e,.m + Z +~ (2.53)
(z - 1)1 (z - 1) ,/
I
Expanding equation (2.53) in negative powers of z and identifying the
coefficients of z-l, %-2•••• gives the system response for all time. Although
this looks formidable, considerable simplification will result if we are
..1'IAN.'UIM ICILllTION CJ. I.INIIAI 1'"'I'II.IINe'• ..,LlATIUNIl "
1111..',.1,&1 lit Ih...."nll'lO nilly I" •.unpltll", In.umt., lilt Ihlll
:f.'C:,O) - 'I' I --,

(: -, I)· (sa - ,Y,)(: ',"\ - I)
=
(z -
.:
1)1
- _1_[ I z-ne,·n -l: z-ne,tn]
SI - S8 n-O n-O
Z 1
= - -- ICICo
(z-nens, - z-nen'l )
(z - 1)1 SI - S8 n~1
z ~. e"" - en••
= -
(z - 1)1 ".1
"z -ft
SI - sa
(2.54)
1,'l'Il1n tables53
(2.55)
where
-1 ± ../1 - 1.2z
SI,I = 2z (2.56)
This reduces (2.54) to
Although (2.57) is in a sense "closed" form, it is not useful in evaluating

Ihe response since one has to expand the infinite product to identify
coefficients of Z-I, Z-I, etc. It is easier to proceed directly from (2.54) and
simplify it by actual expansion.
en•• - e'lIl nl n3 2 a
---- = n +- (S, + Sa) + - (SI + SIS8 + Sa)
SI - Sa 2! 3!
•
~ M + s~sa + SIS~ + S~) •.•
+ 4! .
-nil (SIII-I + s,11-2Sa + ... S211-1 ) + . . .

+ q! (2.58)
56 THEORY AND APPLICATION OF THE til-TRANSFORM
The polynomials in SIS. can all be calculated in terms of til and it is
noted that z appears only in powers of Z-I. %~(z, 0) can then be simplified
as
7
+ !!... (Z-8 - 1.5z-5 + 0.54z-4 - 0.027z-3)
7!
8
- ~ (z-' - 1.8z-8 + 0.9z-5 - 0.108z-4)
8!
+ !!...•(Z-R _ 2.1z-7 + 1.35z-8 - 0.27z-&
9!
+ 0.OO81z-4) + ...J (2.59)
By identifying the coefficients of z-k in this expression, the response up

to the fifth sampling in~tant can be obtained as
c(O) = 0 c(l) = 0 c(2) == 0.55
c(3) = 2.08 c(4) == 4.145 c(S) == 5.4135 (2.60)
This result, as is to be expected, agrees with those obtained by other
methods. 5O More terms in the expansion in (2.59) have to be taken to
evaluate responses for longer times.
In this brief discussion the usefulness of modified z-transform has been
demonstrated in solving linear constant coefficient difference-differential
equations. It is particularly useful in calculating transient responses of
continuous systems having time delays, and this method avoids having to
evaluate roots of transcendental equations, but requires manipulation of
series. Although in Example 2 only the response at sampling instants
has been calculated, it is, however, necessary to use the modified z-trami·
form theory to arrive at the constants AI and AI' This method can he
easily extended to higher-order systems huving more tlmn one but other-
wise commensurahle timl' deluYII.61
• 1\ ~llCl\lld he 111111'11111111 "10(',0) I. 11111 tll'1I1I1I11I1I .,I'lh" I 11'1111\1'111'111, 111111 i'l, '(.(~).
z-TRANSFORM SOLlJTlON OF LINEAR DIFFERENCE EQUATIONS 57
2.4 Difference equations with periodic coefficients25 ,28

Consider the following equation:
Zn+l = a,.z" + Yn (2.61)
II" is a periodic coefficient whose period is N; that is, 0t = a.\,+i for
i = 0, I, 2, 3, ... , N - 1.
To solve this equation we shall first form a constant coefficient difference
l'(luution that describes the solution for n = kN, k = 0, 1,2, ... , and
then use this result to find the solution .for all other values of n. This
pmccdure, in essence, removes the periodicity of the coefficients from
the analysis.
I.et
;[YkNJ = Yo + YNZ-N + Y2NZ- 2N + Ya.vz- aN + ...
+ YlcNz - kN + ...
l[YkN+lJ - Yl + YN+l Z- N + Y2N+l Z- 2N + YaN+1 Z- aN + ...
,)'[]/(klll.v-d = YN-l + Y2.v_1z-N + Ya,v_ 1z-2N + Y4.\,_lZ-3N + , , ,

(2.62)
With this pattern for z .. we solve this periodic coefficient difference
t·tlwltiun. Let n - kN so that an = 00 (that is, an = ak .,· and letting at
Ihe' <;1:lrl k = 0), We can then write for equation (2.61)
X",."II - + Y"'N
a"z",.\'
,1:"',\'12 = a1z"',Hl + Ykl\'+1 = al0 rk.'i + aIYk.\' + Y",.\'+l
;1'''',\'13 = a2x k.V·.,. + YkNI.2 = a2a ,aOxk.\, + a2a'Yk.V + a2Y"',V+!
+ Yk.'·,,2
"'U't II,\' = a,\, 'IXI'" 111,\'·-' + Yllc+lJN-l = (0.\,_l all'_2' , , aO)zk,"

+ (aN I" • a.)ll""v + (a,\·._1 .• , 0a)Y",,\, ...
'I- (ll,,' I ' " aa)lllt;v 12 + ' . , + 0,\, 1111A·.II," 2

+ ,'lIlt Ill.\' I (2.63)
lilt' 1;I';f ('cJllatinll is aC'flmlly :a cnnstClnt eClcnicient difference cqmltion
1I1\'1I1\,11I1~ ,"" :tilt! fhe v:arinus furcing terms, The ::-tr:ansfnrlll method
I ,III III' u';"IIlv applic·tl. '
'1.1', ,f·') ·"01 . ('" I'" (10):7'4",'(::) I ('" ,.,. "I)!IIA.\('~)
I (II, I'" ".)'.'11" 1'(") " •. ' I II\, 1';'1/" III ,. ~( .• ) I ·'l/IA'II" ,(.~)
a.M)
S8 THEORY AND APPLICATION OF THE Z-TRANSFORM
or
~"N(Z) == N 1 [zN:co + (aN-I' •• )q;'tNCz) + ...
z - (aN-I' •• ao)
+ QN-ll{!J(lc+lIN-a(Z) + I{!JIt+lIN-l(z»). (2.65)
We l"eadily obtain :C"N by inverse z-transformation of this equation.
Knowing :CItN we can find :CIt N+l from the first of the series of equation
(2.63). Similarly, :CItN+1 is obtained from the second, and so forth. We
finally obtain :CU:+lI,v.
Solution of simultaneous difference

equations with periodic coefficients
Consider the two following difference equations where a". b", e", and d"
are periodic coefficients of period N, and P" and Q" are known for all
n(n ~ 0).
:C"+1 == a":C,, + b"y" + P" (2.66)
(2.67)
A solution for these two equations can be obtained following the

procedure for a single equation; that is, the 2N equations which corre-
spond to a complete period of the coefficients are written and each
equation is substituted into the following equation, as in (2.63). This
results in two constant coefficient difference equations,iWhich involve :Ck .\·,
YItN and various forcing function terms. These twp equations can be
simultaneously solved for the two unknowns using the!. indicated procedure
of the z-transform method. \
EXAMPLE
Find :c" which is a solution of the following equation:
(2.68)
with at == 0.5 for i == 0, 1,4,5,8,9, ... ,and a, == -0.5 for i == 2,3,6,7,

10, II, ... , and y" == (..... l)".
Here N == 4
4
I[Y,.;v) == I + z-4 + z-8 + ... == ~ (2.69)
z - I
'1
(.
r'I,,~· )== -I .... :; . -
I
, I f · "
z
I
--- "'"
-~
~ ..
z"- 1. (2.70)
IlttV,_ - IIIt,v. !I. v I. - !I•.v, I (2.71)

Using this procedure. we readily obtain for ~Ii:.v(z)
!l'kN(Z) =, % -
1 ,
0.5
X [ Z4zo + 0.125z' + (0.25)( -z') - 0.5z' - ~J
z'-1 z4-1 z4-1 z4-1
(2.72)
with Zo = 0
fr () -1.625z4 (2.73)
Ii:." Z = (%4 _ 0.0625)(z' - 1)
The inverse of this equation yields
26.0 41i:
Z411: = - - [l - <t) ] (2.74)
IS
Since we know
%,111:+1 = 0.5z411: + Y1.k (2.75)
it readily follows that
13.0
Z4k+1 =--
IS
[1 - m ]+ I
4k
(2.76)
Similarly.
13.0 4k
%'k+2 =- - [1 - (!) ] - ! (2.77)
30
%411:+8 = -
13.0
[l - (i) ]
4k
+ 4~ (2,78)
60
2.5 Time-varying difference equations7,10,23,25

111 the following section, we shall first apply the ordinary z-transform to
linear difference equations with rational coefficients and then, in the next
...,·"tinn. the time-varying z-transform will be discussed and applied. The
ordinary ::-trunsform reduces the solution of the difference equation to
Ihl' solution of a differential equation in the variable z. By obtaining
:1 (:), we then apply the inverse z-transform (or use of tables) to obtain
IIII' I ime solution. A general solution of such equations exists only for
IIIl' lirst order eClse, whereas higher-order equations can be solved only
I". pa!'lieular cases. We shall illustrate the procedure by considering the
1.. IIC1will~ Nth-order difference equation with rational coefficients in the
\ .11 ia hi,' ".
11'1 Ih,' "int'n'lIce 1:(IIHllioll oe represented as
,
~ (II," I "~A )/'·~l( II) 'X(II) (2.79)
t •
where the shift operator E denotes Ef(ll) = f(n I), and g(n) is a known +
forcing function.
Applying the z-transform to equation (2.79), we get, using equation
(1.34)
(2.80)
where ~(z) is a polynomial in z containing all the initial values of equation

(2.79) and '~l(Z) = 1[[(n + k)] = Zk§(Z) - (initial values).
Equation (2.80) can also be conveniently written as
d.F(z) + B(z) §(z) = D(z} (2.81)
dz A(z) A(z)
where
.\'
A(z) = - Ia~+] (2.82)
k~l
.V
B(z) = I (ka rr + bk)Zk (2.83)
k=l
and
D(z) = fI(z) + <8(z) (2.84)
It may be noted that equation (2.81) is a first-order differential equation
in the variable z. The solution of this equation is generally known and is
given by the following:
.~(z) = exp[-J B(z) dzJf D(z)
A(z) A(z;) -
exp[f B(z) dzJdZ
A(z)
+ K'~XP[ - f !~:~ dZ] (2.85)

where K is the integration constant.
In general. A(z) can be represented as
A(z} = -a.v(z - Otlyl(z - Ot2)T, ••• (z - Otm)Tm
m m
= -a.\, II (z - IX./Yi, with I r i = N +1 (2.86)
t-l i-·l
Using equation (2.86), we can write
-f B(z) dz =
A(z)
-fi I td t, J (j -
1 [d J- 1 B(z)(z - Oti)'i]
1)! dz H A(::) : a,
(/::
X --.-..... .
(:: _ IX,)" "I
rI.:
~ ~('/, (~.H7)
I
III ",
I' I
"
, (: 1'1,)'1 ,I I
z-TRANSFORM SOLUTION OF LINEAR DifFERENCE EQUATIONS 61
where
C _ 1 [di-l B(z)(z - «XI] (2.88)
II - (j _ I)! dzi-l A(z) '~"I
Performing the integration in equation (2.87), we have
- f B(z)
A(z)
'" [rl-I C
-dz=~ ~_IJ
I
'_I'_I,,-i(z-Cl,),1-1
-Kln(z-Cl) ,
I I
]
where K, = Ciri (2.89)

:lnd
B(z) J m
[ f -.4(z) • m rl-I C·
exp - dz = II (z - Oti)-1I:1 II II exp Ii r. (2.90)
i~l ;~1 i-I (Z - Ot;) ,-I
Where all the roots of A(z) are distinct, that is,
'I = '2 = ... = rm = I, m=N+1 (2.91)
we ubtain from equation (2.90)
exp - [f -
B(z)
A(z) i~1
] "'+1
dz = II (z - Ot,)-r.i (2.92)
where
Ci = [B(Z)(Z - Oti)] (2.93)
A(z) Z~"I
IIII' Ihis case, equation (2.85) reduces to
•:F(z) = II (z -
,\'+1
i~1
Oti)-(;/
• J D(z)·\·+1
-
.4(z)
II(z - Otit, 1dz
1=1
.
.\'+1
+ K II (z - Oti)-C'i (2.94)
i=1
I he inverse z-transform of equation (2.94) yields fen). the solution of

IIII' '~~'lIcml Nth-order time-varying difference equation given in (2.79).
II lIIay hI.' noled thut if CI is noninteger, as is often the case, .?(z) will be
1111 1IIIIIillnni function of z. In Table I of the Appendix many entries for
, ( ) III' such n furm ure tabulated and thus can be conveniently used to
,,11111111 Ihl.' linal solution.
nCAMI'I.I;
III 11m "X:llllple: we shall nn:lly'-c u sumpled-dutu feedback system with
"II III hie' /~:lill. This syslem is shuwn in Fig. 2.4.
,0\·.·.11111111/'. I.~·I·II illilinl cnlUliliulll; :tIIt! lelting (0 .. = a. we h.lYC for the
,,\",11'111
I /I
1'(11 I I) ,,1'(11) - '1(" )
C"L *'
61(z) T-l T-l ~(z)
~ -j f--A:.-j .!:p H ..... t---ofr+-o. 1

c(ra)
0
'fI(z)
FIGURE 2.4 Time-varying sampled-data system.

and
v(n) = ~ [r(n) - c(n)] (2.96)

n+1
Eliminating v(n) from these two equations and denoting k - a) kl' = (I
ex
b = k - a, we have the following time-varying difference equation:
(n + I)e(n + I) + (bn - a)e(n) = knr(n) (2.97)
Applying the z-transform to equation (2.97) we get using equation (1.34)
z(z + b) d~(z) + a"'(z) = kz d9t(z) (2.98)

dz dz \
Assuming a step input. that is, 9t(z) =
b=~
*' - 1 arict for k = 2a, that is,
I
-z- ~(z) = -k
z+a (z
f zdz
+ a)2(z - 1)2
+K (2.99)
where K is the integration constant.

We now discuss the following three cases.
1. a ,t: -I, for this case (equation 2.99) yields
~(z) = k {Z(l - a} + 2a + 1- az + a In z - I}
(I + a)1I z(z - J) t +a z z+a
+K z+a (2.100)
z
Since c(n)ln =0 = lim 'if(z) = 0, K = O. Hence equation (2.100) becomes
z-co
~(z) = k {zC1 - a) + 2a +1- a z + a In z - t}

(I + a)2 z(z - 1) 1+ a z z+a
The inverse of this equation yields (2.101)
('(n) =
(I
k
+ tI)
I {< I - tl)lI(n - l) + 2(III(n - 2)
+1-
I ··f
tI [R(II)
(I
-t (111(11 .- I U} (2.102)
where g(n) is defined as
g(n) = 0, for n = 0
= ~ [(_a)" - 1], for" ~0 (2.103)

n
Equation (2.102) yields after using g(n)
c(O) = 0, c(1) = 0
c(n) =
k {
(1 + a) -
I-a
-- ---
1
(l + a)2 1 + a n(n - I)
X [n(a + I) - I + (-a)"J}, for n ~2 (2.104)
2. a = -1, k = -2. For this case equation (2.99) becomes
-z- <6(z)
z- 1
= -k f zdz
(z - I)
4 + ko (2.105)
Since c(O) = 0 = lim ~(z), we have ko = O.

Equation (2.105) yields
3l- I
fi'(l) = k 6l(l- 1)2 (2.106)
Thc inverse of this equation gives

e(O) = 0, e(l) = 0
k(2n-1)
e(n) = , for n ~ 2 (2.107)
6
.\. h = 0, a = 1, or (or. = 0, k = J). For this case, assuming a step
inJlIII. e<'luution (2.98) becomes
:1 d'6'(z) + ~(z) = _ z (2.108)
dz (z _ 1)2
TIll' ~nlllti()n of this equation yields
(2.109)
"11 nhtnin Ihe integmtinll constant Kt~. we use the following limiting
111'01' 1'111 III\' illili:11 vulllc:
lim '("(.) dU), I ·1· 1\;' = () (2.11())
I I."
111'111'"
1\'-
II (,'.111)
Substituting this in equation (2.109) and finding the inverse z-transform.

we finally obtain:
I
c(n) = u(n) - - (2.112)
n!
where u(n) is the unit step at discrete times.
Simplification of equations be/ore

z-transjorm application7.l8.46
In certain cases the rational coefficients difference equation could be
reduced to a constant coefficients equation by certain function substitu-
tions in the original equation. This reduction simplifies the application
of the z-transform to obtain the time solution.
A typical unmixed difference-differential equation with rational coeffi-
cients which may be reduced to one with constant coefficients is
(n + k + q)(n + k - I + q) ... (n + 1 + q)Y~".l.k
+ 0k_l(n + k - 1 + q) ... (n + 1 + q)y~t;':~1 + ...
+ ol(n + 1 + q)Y~~1 + 0oY" = /,,(1) (2.113)
where
Ik' d lr ( ) (2.114)
Y,,-u = dl k Yn+1 I
and n is a parameter.
The substitution function is
%
Yn = J"'(1 + ~ + q) (2.115)
Substituting this in equation (2.113), we obtain the following constant

coefficient difference equation:
110'
%n+k + 0k-I%n+ .-. + • • • + 01%,,+1
1lt-1I /11 + 00%"
= r(l + q + n)/"(I) = g,,(r) (2.116)

By applying both the Laplace and z-transform to this equation. we can
obtain the solution.
I. EXAMPLE
Obtain the solution 11.. (1) of the following unmixed difference-differential
equation:
!I:,':. +./I
a
I I
!I" ,0. with !I"IU)
/I! •
a l'U list illl I
1}..117)
z-TRANSFORM SOLUTION OF LINEAR DIFFERENCE EQUATIONS 6S
Let
X" X"
(2.lIS)
JI" = I'(n + I) = n!
Substitutillg the above in (equation 2.117), we have
Cll
X,,+1 + ax,. = 0 (2.119)
Applying the Laplace transform to equation (2.119), and using the
initial value %,,(0) = y,,(O)n!, we get
SX'Hl(S) + aX,,(s) = I (2.120)
Defining the z-transform of X,,(s),
oc
.-l'(z, s) = J'[X lIes)] =,,=0
! X ,,(s)z-" (2.120a)
and applying it to this equation with s constant, we have

z
sz.'l'(z. s) - szXo(s) + a.'l'{z, s) = -- (2.121)
z- I
This equation reduces to
.tr(z, s) = szXo(s) + z (2.122)
sz +a (z - 1)(sz + a)
The inverse z-transform of this equation yields
X,,(s) = (- -a)"[
s
Xo(s) - -I-]
s+a
+--
1
s+a
(2.123)
We can write for Xo{s) the following form:

1
Xo(s) = - - + K1(s) (2.124)
s+a
where K.(s) is determined from the given initial condition.
If we let K. (s) = K. / s, the inverse Laplace transform of equation (2.123)
lives
;r ,,(1) = ,r"' + K 1 (_~)n t" (2.125)
n.
I J!oill~ eq lIHlion (2.118), we obtain for the solution
c.
f/,,(I) = - -
PI!
,,1
+ K, (-")"1"
---.-
(,,,)2
(2.126)
1 0' Ih,' ,':lM' A·.(.~) () (or 1\. = 0). Ihis etlll:tlinll reduces tn
I' '"
f/"I II (2.127)
II!
2.6 Time-varying z-transform

and the system function 10 ,J3,25
In this section the solution of a general linear ~ime-varying difference
equation will be indicated by using the z-transform of the system function.
For the cases where closed form solution is not obtainable and the system
is slowly varying, an approximate method based on the system function
is used. The discussion follows closely Zadeh's work 1 J.12.26 as applied to
time-varying differential equations and the use of the time-varying
transform methods.
Assume a time-varying difference equation as follows:
[a/n)E-" + a,._I(n)£-11I-1I + ... + a1(n)E- 1 + ao(n)]cn
= [bA(n)E-.l + bA_ 1(n)E- CA- 1l + ... + b1(n)E- 1 + bo{n)]rn
(2.128)
with p ~ i. and Cn , r n representing the output and input sequences of the
system described by equation (2.128). The delay operator E-l is defined
as E-'1n = In-/('
Equation (2.128) could be expressed in a condensed form
A(E-I, n)c" = B(E--l, n)rn (2.129)
Since the system H is linear, we may define its impulse response
hen, n - r) = h(II, k) by
(2.130)
where the Kronecker delta function 6nk is defined as
if n = k} (2.131)
if n ~ k
Using the superposition theorem for linear systems, the output Cn is
given as
n
Cn = !h(n, k)r/( (2.132)
k=O
It should be noted that hen. k) is sometimes written as hen, n - r) as
indicated. The impulse response hen, k) is expressed from equation
(2.132) as the amplitude of the pulse occurring at the output at the nth
instant as a result of an input pulse of unit amplitude applied at the kth
instant.
In analogy to the one-sided z-transfnrm. we define II limc-vllrying
z-transform of h(fl. fl - r):
. ,
.*'(11. z) ,:'- ",111(", 1/ r)l ~ ',(11, " r).~ r (2.13.\)
, II
Setting k = /I - r in this equation. we obtain
» n
,~(n, z) = L hen, k)z-"+k = z-n 2. hen, k);;k (2.134)*
k=-» k=O
The summation on the right-hand side of equation (2.134) represents

the response of H to an input signal having the sequence of amplitude
:::(l, ;;1, .:;2..... Hence the system transfer function .~(n. z) may be
alternately defined by
£(n, z) = response ~f H to z" (2.135)

z
Input-output relationship
The output at the nth instant may be obtained from the consideration ot
'(,',,(z), where
~ ft(;;) = Yl(z).:tt'(n, z) (2.136)

where
to
:3I(z) ~ J[r k] = L rkz- k (2.137)
k~O
The output sequence is given by the following inversion integral
e" = _1
21Tj
rf~(z).;;fl(z,
J1'
n)z,,-I dz (2.138)
when r is a counterclockwise closed contour in the z-plane which encloses

I he singularities of the integrand.
II should be noted that the index n in equation (2.138) is treated as a
parameter; thus C n could be obtained from the ordinary table of inverse
.:-transforms, that is, Table I of the Appendix. Furthermore, if'tf,,(z) is
l'''panded in power series. the coefficient of r k is not equal to Ck except for
J.. .0 II. Thus equation (2,136) does not represent the z-transform of the
OUlput for all sampling instants,
The :::-tr:lOsl'orm of the output (sometimes referred to as the ",-trans-
forl11)1U can can he ohtained by using the double-difference system function
III analogy to the hifrequency system function for the systems described
by dill"l'n..mlial elluations.20 In addition. as noticed. the method of the
1lIl'l'l'ding Sl'l'tioll yields in some cases the :-transform of the output.
• 1111'. l'll"allly huhb ht,,'mlsl' ill Ilwsc IIiSt'liSsinns wc arc :lsslll1lin~ Ihal nn inpul exisls
1''''1'1'11111/: " n,
68 nlEORY AND APPLICATION OF THE z-TRANSFORM
Determination of ~(z, n)
The system transfer function .JtP(z, n) satisfies a difference equation which
sometimes is easier to solve than the original difference equation (2.128).
To obtain the equation for Jf7(z, n), we take the z-transform of equation
(2.130) with respect to the index k to get
GO co
I A(E- I, n)h(n, k)ZTc = IB(E-\ n)~ntZk (2.139)
k~O Tc-O
and noting that hen, k) = 0 for k > n, this yields

A(E-I, n).Jt"(z, n)z" = B(E-l, n)zn (2.140)
Before rewriting equation (2.140) in a more convenient form, we utilize

the following identities:
(2.141)
and
(2.142)
since A(E-l, n) and B(E-l, n) are polynomials in E-l, using equations

(2.141) and (2.142), the left- and right-hand sides of equation (2.140) become
(2.143)
and
(2.144)
Therefore ~(z, n) is given from equations (2.143), (2.144), and (2.140) as
(2.145)
Equation (2.145) is a ,uth order difference equation with nonconstant

coefficients. By adding and subtracting A(Z-l, n)~(z, n) to equation
(2.145). we may obtain
B( -1 ) [fa,(n)z-i - f ai(n)z-iE-i]~(Z' n)
~(z n) = z, n + .:=.:.;;;o_.:..O_ _ _-"i-;;;.:O~----==----
, A(Z-I, n) A(Z-l, n)
(2.14()
or
II
1 a;CII): i[-'po(:, II) - .tI'(::, II - 01
..0°(.,.,.," ) --- /1(: I, '1)
.R
,. I
A(: I, II) A(.: 1,'1)

0.147)
The second term on the right side of equation (2.147) is small compared
to the first if either .J?"(z, n) changes slowly with n or if the changes are
small. This suggests that equation (2.l47) can be solved by iterative
methods. Thus if .71'k(z, n) is the kth iterated solution we have
,.
I aj(n)z-ipr'k(Z, n) - :It'iz, n - i)]
'W'k+l(Z, n} = ,7I'0(Z, n} + .:;;;1~;.o.I_ _ _ _ _-:-_ _ _ _ __
A(Z-l, n)
(2.148)
where
-Vi> (., ) _ 8(%-1, n)
.J( 0 -, n - (2.149)
A(z-t, n)
is called the Jrozen transfer Junclion.
Equation (2.148) is very useful in obtaining an approximate solution
for slowly varying discrete systems. In some cases the frozen transfer
function will give a good indication of the response behavior. Such
!oituations may arise in sampled-data systems where the sampling rate
l'IHlIlges very slowly. Another application of this technique is in the analysis
II\' automatic gain control where the pulse-frequency modulation varies
according to a sinusoidal law. Such a system will be studied as an
illustnttive example.
Furthermore, if .11 o{z, n) has its poles inside the unit circle for al\ n,
all the iterative solutions HI> H 2 • ••• will 'have their poles inside the unit
I'Irde and vice versa.
The following example was studied by Tratakowskii l3 using a similar
approach. A more detailed analysis of this system will be discussed and
n:lt't conditions on stability as well as the response approximation
IlIt· ... t·nted. The system under study is referred to as "pulse feedback with
\',11 iahle pulse repetition frequency." This system is shown in Fig. 2.5
,11111 its etjuivalent feedback system in Fig. 2.6. The automatic amplifi-
I al ion control in u receiver of pulse signals of constant amplitude when an
......
(h'III,1) /I c y (Output)
L-_r-----'--,---o
r-
I II iI/IO ." I \OllIIl'h' fur II,,· v;II'iahlc pulse I'eetlback system.
T-Y
PFM H;i, I
FIGURE 2.6 Block diagram of the variable pulse feedback system.
RC circuit is used for filtering in the AVe leads to the study of such a
system as shown in Fig. 2.5.
The pulse-frequency modulator (PFM) yields the following equation
In < t < In+1

(2.150)
where In is the time of sampling which is determined by
In - 1._ 1 = T(l + Msin On) (2.151)
where T is the average sampling period.

When 10 = 0, h is a positive constant with the following constraint:
h ~ T(I -IMD (2.152)
The relation between y and :r for Fig. 2.6 is given by the following
differential equation:
y + fly = kx (2.153)
The solution of equation (2.153) yields
y(t) = y(to)e-P(Ho) + ke-p'l'X(T)ePr dT (2.154)

'0
In particular when t = Itc and '0 = 'n-I, since X(/) = 0, for ' 11 _ 1+ h <t<
In' equation (2.154) yields
y(t",} = y(tn_l)e-fJ(I.-tn-l) + ke- fJt • f'.-1

'·-I+h
q(tn_l)e fJr d'T (2.155)
Using equation (2.151) in (2.155), we obtain

y(t n) = y(t,,_l)e-PTU+.l l sin 0,,)
+ ~ q(t "_I)e- fJTU + ill sin 0,,)( eflh - I) (2.156)
Let
ex = /0: l' =,/"

" kll :--=
"
ii' !lU,,):,= !I(II), q(I,,) = q(II)
(2.157)
and assuming oc« 1 or fJh « I, we have from equation (2.156)
yen) = [I - oc(1 + M sin On)]u(n - I)
+ koOty[1 - oc(1 + M sin On)]q(n - I) (2.158)
From Fig. 2.6, we have for the feedback equation
q(n) = u(n) + y(n) (2.159)
Substituting equation (2.159) in equation (2.158), we get
yen) = (1 + k oocy)[1 - oc(1 + M sin !In)]y(n - I)
+ koocy[l - Ot(l + M sin Un)]u(n - I) (2.160)
The frozen transfer function is simply obtained by taking the z-transform
of equation (2.160) and considering n as a parameter [this is seen from
equation (2.149)]. Therefore
,Yfo(z, n)koocy[1 - oc(1 + M sin On)]z-l
= (2.161)
1 - (I + koocy)[l - oc(1 + M sin On)]z-l
If we assume a slow pulse-frequency modulation, that is,
)
ocM« 0 (2.162)
Ihe system function £'(z, n) can be approximated by the frozen transfer

fUllction or by .K1(z, n). The response of the system for a step input has
h~'ell studied for different 0 and oc/tJ and the exact and approximate (using
Ihe frozen transfer function) responses are shown in Figs. 2.7 and 2.8.
As expected. the smaller the quantity T with respect to 2fT/!). the better
" Ihe approximation.
hluation (2.160) describes a first-order time-varying difference equation,
which can be written as
yen) = a(n)y(n - 1) + b(n)u(n - 1) (2.163)
From this equation if a(n) and ben) are bounded (as is usually the case)
.11111 '\upposing a(n) ¢ 0.* then the system is stable if and only if
I ,\'
lim - ! In la(n)1 < 0 (2.164)
N~'" N n l
A ~ul1icient condition for stability requires that Ia(n) I < I for all n. In
11110,('xamplc, ItI(tI)1 < I. is the same as the condition that the pole of the
111I/l'1I Ir:lllsfl'r function [see equation (2.161)] should lie inside the unit
• lid,'. Ilence Ihe suflicicnt condition for stability (for M> 0) is
I Il( I - M~.- 2 < ko <! I - M (2.165)

)'111)(1 , .. M) - 1 y 1- et() - M)
• I h., \llIlIr l'lIlIIlitilili I'll. Mlihilil.Y ellll he pmyell IIll1n if u(,,) ". o.
For the plots of Figs. 2.7 and 2.8 the preceding condition is satisfied.
Thus the system is stable.
To obtain a better approximation, the first iterative solution from
equation (2.148) could be used, and for this example
~ (
.:n 1
) _
z, n - on
.yp (
0 z, n
) + Ql(n)z-lpof'o(z, n) - .Yt'o(z, n - 1)]
ao(n) + a1(n)z-1
(2.166)
where
Ql(n) = -(1 + k oocy)[1 - oc(1 + M sin On)] (2.167)
ao(n) = 1 (2. 167a)
0.25,..------..------,------.------,
"
I \
I ~lst order approx.
I \
0.20 t-------+--+~---t-----.f....-4i_-----I
Oth order approx.
0.15 f------+---!--'~_i__--+--~__f+l_+-----I
0.10 f-----+-H---lH~-+-__f_+__!___f__1___+_----l
0.05 t--.",."----+------+-------I-------I
a =0.2
ko"Y = 0.1
M=0.4
n= 7r/2
OL-_____ _____ _____ _ _ _ _. _ J
~
~
~
o 25 50 15 100
/I
FIGURE 2.7 i\XIICllllld npprnxllllilio relll,ulIlIC ur .Ylllcmlll Fig. 2.b rUf II 11101' Inpul.
0.15
0.10
Oth order approx.~ ~
~
!:"'Exact
~.-=.
0.05 I 1st order approx.
a=0.2
= 0.1
k{j'y
M=0.4
0=11'/100
o
o 50 100 150 200 250
n
FIGURE 2.8. Exact and approximate response of system in Fig. 2.6 for a step input.
and
.,g> ( ) _ B(Z-l, n) _ ko«y[(l - «(1 + M sin On»)
.n 0 z, n - -
A(z-t, n) ao(n) + a 1(n)z-1
= __b..!,;l(~n:.....)_ _ (2.168)
ao(n) + a1(n)z-1
It may be noticed that if the pole of '~o(z, n) is inside the unit circle for
all n, so is the pole of £'1(Z, n) and similarly for other higher-order
iterative solutions. The response using the first-order iterative solution is
also shown in Figs. 2.7 and 2.8.
2.7 Double z-transformation* and solution

of partial difference equations21 ,23,4.1,49
Partial difference equations arise in many applications such as probability
theory, boundary value problems, lattice network, and transmission lines.
The systematic solution of such equations can be performed using the
duuble z-transformation (or two-dimensional z-transform).
This transform is defined as follows:
00 co
U(z, w) ~ a'w[u(n, m») ~ ~ ! u(n, m)z-nw-m (2.169)
n-O m=O
where ,)'". denotes the zw-transform of u(n, m).
• AIN" rdcrrcd Itl ill Illllthclllllticnlliternture 05 bivariate generating function."

We apply the zw-transform to the following functions to obtain

co GO
II11[u(n + 1, m)] == ....
I I u(n + I, m)z-"w- m
0 ...=0
== z[U(Z, w) -I u(O, m)w- .... -0

m] (2.170)
'l11[u(n, m + 1)] == w[U(Z, w) - I u(n, O)Z-"J (2.171)

.. -0
II11[u(n + 1, m + 1)] == zw[U(Z, w) - I u(n, o)z-n

..-0
- m~oU(O, m)w- m + U(O,O)J (2.172)
Similarly. we can extend this procedure to obtain the general zw-

transform of u(n + a, m + b). lbis transform is given as follows and
can be easily derived from procedure given,
CIO CIO
+ a, m + b)] == I I
1",[u(n
?I~O .-0u(n + a, m + b)z-"w- m
a-I CIO
-I I u(n, m)z-"w- m
.. =0 .... -0
(2.173)
Equation (2.173) can be readily used to obtain zw-transformation of
any order linear partial difference equations. By successive inversion
with respect to wand z, the discrete time function u(n, m) can be obtained.
This procedure is illustrated by the following example.
We consider in this example a partial difference equation which arises in
probability theory. Obtain u(n, m) which denotes the probability of In
successes in n trials and which satisfy the following partial difference
equation:
u(n + I, m + I) - quell, In + I) - pll(n, m) == () (2.174)
The boundary condition on Ihe n-Itxill ill u( n, 0) ." q" for n .- O. The
number p denotes the prubllhility fur III1CCClIlI in independent scrics or
z-TRANSFORM SOLUTION OF LINEAR DIFFERENCE EQUATIONS 7S
trials, and q = 1 - P denotes the probability of failures. Applying zw-
transform to equation (2.174),
zw[U(Z, w) - I u(n, O)z-" - I u(O, m)w-

,,-0 m-O
m + U(O,O)J
- qw[U(z, w) - Jou(n, O)Z-"J - pU(z, w) = °
(2.175)
From the formulation of the problem it is evident that u(O, m) =
(0,l,m=O
m > O}. Therefore equation (2.175) becomes
zw[U(Z, w) - I (~)-"J
,,-0 q
- qw[U{Z, w) - ,,=0
I (~)-"J
q
- pU(z, w) = 0 (2.176)
III'
U(z, w)(zw - qw - p) = (zw - qw) I00

,,~O
(z)-"
-
q
(2.176a)
lienee,
U(z, w) = ~o (;f" _ ~o (;f" (2.177)

1-(zw -Pqw) I-!(-P)
%-q W
Tllk ing the inverse w-transform of (2.177) we obtain
'1/(%, m) = I (z)-"(
00
- - p)tn = (p/z)m I00 (q)"
- (2.178)
,,-0 q z- q 1 - 4/% ,,~O Z.
=~ (p)m( 1-;q)-m(1-;q)-1 =~(p)tn( 1-;q)-l-m

"I"
'f/(:, m) = (;r[1 + (-I ~ m) (~)(-J)

+ (-I ;- m)~(_1)2 + ...J (2.179)
IIII' \'(1"lIid,'nt III" Ihe:: " term is
,,"',," "'I (-"I.-- '""')1(-1)" ' ', (2.180)

Therefore the inverse z-transform of equation (2.179) becomes

u(n, m) = pmq"-m
X [(-1- m)(-2 - m) ... (-n + m - m>J(_I),,-m
(n - m)!
m lI_m[n(n - 1) ... (m
=pq + 1)J (2.181)
(n - m)!
Finally, u(n, m) yields
u(n, m) = pmqn-m (:1) (2.182)
REFERENCES
I. Doole, G., A Treatise on the Calculus of Finite Dijferenus, G.E. Stechert, New
York. 1926.
2. Jordan, C., Calculus of Finite Differences, Chelsea Publishing Co., New York. 1960.
3. Milne·Thom~on, L. M., The Calcuills of Finite Difference, Macmillan Co., London,
1951.
-"'''4. Goldberg. S., Introduction to Difference Equations with Illustrative Examples from
Economics. Psycllolo,f{Y and SoC'iolo.IfY. John Wiley and Sons. New York, 1958.
"l5. Pinney, E., Ordinary DiJjerence·Differential Equations. University of California
Press, 1959.
6. Gelfund. A. 0., Differenzenrechnung. Moscow·Leningrad. 1952 (in Russian).
7. Levy, H., and F. Lessman. Finite Difference Equations, Macmillan Co.. New York,
1961.
8. Nor1und, N. E.• J'orlesu".f{l'n uber Differenzenrechnung, Verlag Von Julius. Springer.
Berlin. 1924.
9. Samuelson, P. A., FOllndation of Economic Analysis, Harvard University Press: 1947.
10. Friedland, B., "Time·Varying Sampled·Data Systems," Proc. Second Midwest
Symp(l.fium on Circuit Theory, Michigan State University, East Lansing. December
3-4, 1956.
11. Zadeh, L. A., "Circuit Analysis of Linear Varying·Parameter Networks," J. Appl.
Phys., Vol. 21. November 1950, pp. 1171-1177.
12. Zadeh, L. A., "On Stability or Linear Varying Parameter Systems," J. Appl.
Phys., Vol. 22, Apri119SI, pp.402-405.
13. Tratakowskii, G. P., "Application of Simulation in Analyzing Linear Pulse Systems
with Variable Parameters," At'lomatica i Telemekh, Vol. 20, May 19S9, pp. 559· S(lf!.
~ 14. Aseltine, J. A., "A Transform Method ror Linear Time-Varying Systems," J. AI'I,I.
Phys. Vol. 25, June 1954, pp. 761-764.
15. Cooke, R. c., Infinite Matrices a/l(l Seqll('n('(' .\i'u(·(·.~, MlIcmillllll ('11., 1.1\lldllll, 1')~().
16. Ho, E. C .• and H. Davis. "Gcncrali/cli Op~"alin'ml ('alclllu~ fill' Time-Varyinl:
Networks," Dept. of F.ngineering Repnrt ~4 71. JIII.V 11)';4, IInivl''',ity nf ('lllifn,""',
Los Angeles.
17. Bubh, I-'. W .• "A New l.inl'lli 01"'111""11111 (·lIk"I"·.... II.S.A I. h·,·hlli,·,,1 It"I"'1I
Nil. MHI. May I ')~ I. J)uylnll. Ol"n_
18. Madwed, A., "Number Series Method of Solving Linear and Nonlinear Differential
Equations," Instrumentation Laboratory Repon No. 6445-T-26, April 1950, M.I.T.,
Cambridge, Massachusetts.
19. Raymond, F. H., "Analyse du Fonctionnement des Systemes Physiques Discontinus
(ct son Application aux Servomechanismes)," Ann. Teluomm., Vol. 4. July, August,
October 1949, pp. 250-256. 367-314, 347-357.
m. Poincare, H., "Sur les Equations Lineaires aux Differentielles Ordinaires et aux
llilTcrences Finies," Am. J. Math., Vol. 7, 1885, pp. 213-217, 237-258.
~~1. I.awden, D. F., "On the Solution of Linear Difference Equations," Mathematkal
(it/,-,·I1('. Vol. 36, 1952, pp. 193-1971.
n. nelhnan, R., and K. L. Cooke, Diffcrential-Di.Q;rence Equations, Academic Press.
N.·\\" York and London, 1962.
.. I. Mcschkowski. H., DifferenZf!llgleichungen, Vandenhoeck and Ruprecht, GoUingen.
1"\11 •
.'01 1luetsch, G., Handbucll der Laplace Transformation, Basel und Stuttgart. 1956.
"" lIurnegel, R. G., "Analysis of Aperiodically-Sampled-Data Feedback Control
';y·,It·Ill'." Ph.D. Thesis, Dept. of Electrical Engineering, Cornell University. Ithaca,
N.·\\ Yurk, 1954 .
.'" /;lIlch, L. A., "Frequency Analysis of Variable Networks," Proc. IRE, Vol. 38,
1"',11, PI'. 2t}1 ·298 .
.'1 1\;lpllln, W., Operational Methods for Linear Systems, Addison Wesley Publishing
I II, Ih·;nling, Massachusetts. 1962.
.'N .IIIIV. L I.. and F. J. Mullin, "A Note on the Operational Solution of Linear
I hll"II'lIn' I.quations." J. Franklin Inst., Vol. 266, No.3, September 1958, pp. 189-205 •
.., A·.•·lline. J .• Transform Method in Linear System~, McGraw-Hili Book Co., New
\ .. ,~. I·I~K.
Itl 1\111111,'1. I'.• I.('t;ons sllr les Remrrences et leurs Applications, Gauthier-Villars, Paris,
1'" I
II hn y. I . I.. SlIInpl('d-Data Control Systems, John Wiley and Sons, New York. 1958.
I' ,..·111"",,', J.• D(lfital Processes for Sampled-Dota Systems. John Wiley and Sons,
I 10·\\ \ "' J,., 11/(,2.
II I "'1. ·1 .. "I incar Difference Equations and the Dirichlet Series Transform,"
I", """" "'II/II"~I" Vol. 62. No.9, 1955, p. 641.
... N'·lIh·hl. I., "On the Operational Solution of Linear Mixed Difference-Differential
1'1".11'"11":' ("I/I"";/{~/' "lri/. SIIC., Vol. 30, 1934, p, 289.
" 111"\\11, II, IIII' MIIII,('Itltltical TI,e/lfY of Linear Systems, John Wiley and Sons,
'11.\ \ "' J,., 1'11,1.
". 1',11. 1\1 A," I hc Operatiunal Sulution of Difference-Differential Equations Using
110. 1\1",h' .... 1 1':11,,1''''''11,'' IR/:, 1'(jA(', October 1962, pp. 124-126.
I' I., ",111111". I., "Oh"1 ciuc Vcmllgcll1cincrung der Exponentialreihe," Math .
. ,1/" I" ,\'111 1", 1'11.2. PI'. 2.1'1 242.
II, I., h.III'II". I, "I "'·III'.,·"'I~' "'",ktiullen von Zugeordneten Kugelrunktionen,"
11.1/1. .', ,/I. I" , V"I/H. 1')(,2. 1'1'. 1.11 1.\4.
1'1 I ,.10 1\ I, .0\ , II~I,""'I I I,IIIU·,'"d,·",,,II·illl/'ti,,,r.v, Me(imw-JliII nook C<, .. New York.
7M '1'1 II "It V "NI) A1'1'1 ... AIIUN III 1111' • I" ANllt". M
41. Miller, K. S., All """"1",·"",, t" tl,,' •"",,,1,,, ,./ I,,,,,,,
IJlI/,'/I'"", .111,1 11'/"'/1'11'"
Equatio".f, Henry Holt lind ('11 •• Nt'w ~'II''', 1'''011.
42. Ku, Y. H., and A. A. Wulf, "l.lIlIrelll-( ·lIlId.y "'11'''/'111111\ , ... A,"I1V\I' "' IllIrlll
Systems Described by Differential-Difference lind S\l1lI I ''1"11lilln\,'' I'll"'. IHI , VIII . • 'H.
May 1960, pp. 923-931.
43. Jury. E. I., Discussion of the above reference, Pror. IRE, Vol. 4K, Nil. 12. Url'('I11'''''
1960.
44. Mikusinski, Jan, Operational Ca/rulus, Pergamon Press, Landltll, I')!I').
45. Jury, E. I., "Contribution to the Modified z-Transform," J. Frallklill Illst., VIII. 21t1,
August 1960, pp. 118-129.
46. Tschauner, J., Einfuhrung in die Theorie der Abtasysteme, Verlag R. OldenhulII1:,
Munich, 1960.
47. Cypkin. J., Theory of Pulse Systems, State Press for Physics and Mlllhel1lllticlll
Literature, Moscow, Russia, 1958 (in Russian).
48. Stone, W. M., "A List of Generalized Laplace Transforms," Iowa Statl' C"II':I{"
Journal of Srience, Vol. 22, April 1948, pp. 215-225.
49. Feller, W., An Introduction to Probability Theory and Its Appliratiol/s, John Wiley
and Sons, New York (second edilion), 1959, Chapter XI.
50. Wloka, J., "Ober die Anwendung der Operatorenreihnung auf Lineare DitTerentilll-
DitTerenzengleichungen mit Konstanten Koeffizienten," J. fur die Riene lind Augewandtl'
Mathematik, Vol. 202, OClober 1959, pp. 107-128.
51. Fi4!vet M., On the Analysis of Feedback Systems with Transportation Lo,f Using
the Modi/h'd z-Tran~rorm, Master of Science Project, Dept. of E. E., University of
Californill, Berkeley, June, 1963.
52. Alper P., TWI} Diml'"sional z-Transforms, Technological University Eleclronics
Labamtory, Delfl-Netherlands, August, 1963.
53. Gradstein, E. S., and E. M. Ryzik, Tables of Integrals, Sums and Products, State
Press for Physical and Mathematical Literature, Moscow, 1962, p. 37, Formula 1.223.
S I A BII.ITY (,ONSIDERATION
I'()I{ I.INEAR DISCRETE
SYSTI~MS
'ltll' "I IIII' main pl'Ohlcllls in the study of linear systems is the detel'-
111111.111"11 .. I' stahility. For linear continuous systems, the problem of
·.I,lhlllly 1\'lh'l'~S til the necessary and sufficient condition for the roots of
II" /'oY'oI\'1lI dHlnlt;tcristic equation) to lie in Ihe left half of the s-plane. 8
II ....... nllltlilinlls huve been studied in detail and as a result the criteria of
IIIIIWIII 1~llllthH and of Licnard-Chipart8.38 completely solve the problem
,III'llylll:ally. For linear discrete systems the condition of stability reduces
", .. hlallling the necessary and sufficient condition for the roots of F(z)
1I1I·.I'II'll' system characteristic equation) to lie inside the unit circle in
II ... ·pJ:lnc.!I,17 Although these conditions have been studied by Schur.28
C 'lIhll,~ and Marden,· the final stability criteria to be obtained from these
!"IIWI'S arc mther complicated and not readily applicable for system anal-
~",I" and design. Therefore in this chapter a complete study of this problem
I', IlIlIk,rtaken where the final solution of the stability criteria in the forms
"IIII,terminunt,l table,l2 and division 14 methods are obtained.
Siahility tests within the unit unit circle are very important in many
IIl'lds, For instance, the stability of difference equations with constant or
1"'1 iotiically varying coefficients,21 the stability of limit cycles in nonlinear
tI"l'l'cle systems, the stability test of linear systems with randomly varying
parameters, the convergence of iterative computational rule using digital
1·lImplIlers. 21 the stability of periodic regimes in nonlinear systems with
JlII:~ewise linear characteristics, and many 'others all reduce to stability
1l'~ls within the unit circle. In view of these applications. it is felt that
I'IIncerted efforts should be directed toward stability tests within the unit
!'irde as well as the conditions on the roots distribution of £(z) within the
unit circle.
79
NO IIlIORV ANI! AI'I'IIIAIlCtN III 1111 .IIlANlilltllM
3.1 ()eliniliull tlfllillhilllyll

A linear discrete system is considered It) he "'"hl(~ it III 1111 hllll III h',1
inputs there always correspond bounded outputs.·
The stability condition can be obtained if we sepunlh~ IIw 111111'"1
(system response) into steady-state and transient terms. Thi, t:UII h,'
easily established if we write the output z-transform (6'(.::) <IS
~(z) == Er(z)~(z) (.1.1 )
where 8(z) and .~(z) are the z-transforms of the input and the syslem
transfer function respectively. The output related to the steady-stule p:lfl
(or due to the input function) can be obtained from the inverse ::-trunsful'm
formula as the sum of the following typical term:
residue of lEr(z) . .Jf'(z)]l,.=.... z::,-l (3.2)
where Z'II is a typical singularity of Er(z). Since the input is bounded, that
is,
Iz".1 ~ 1 (3.3)
then the output part resulting from this input is also bounded.
Similarly, a typical transient term of the output can be written
residue of Pt"(z) . 8(z)]I._I", ZItn-l (3.4)
where ZIt is a typical singularity of ~(z). From this expression, it is
evident that in order for the ouptut to be bounded, the following con-
ditions should exist.
(3.5)
or all the poles (singularities) of .Te(z) should lie inside or on the unit
circle.
When one singularity is on the unit circle, that is, Iz,,1 = I, then from the
physical point of view such a system should be regarded as unstable, since
a very small change in the physical constants may throw the critical
singularity from the boundary into the exterior of the unit circle, causing
actual instability. Summarizing this remark, we may state the following
fundamental theorem: If all the singular points of the discrete system
transfer function ~(z) are located inside the unit circle, the system is
stable. If at least one singular point lies outside or on the unit circle, the
system is unstable.
• These outputs are considered only at discrete intervals or time (sampling instants).
For a continuous output or a discrete system, the modified z-transrorm or the output
should be used instead of equation (3.1).
It I 41111 II ~ ' ....... 1111 It" 1111'" 11111 II NI AII III!I' III II ""'111 M'I KI
I hi" I "IlIhll"" "I ·,1,1111"1" Il',ullly 111'1'''''' III "I""I"II'T 1"1"01111111\ Wllh
,,,",.1,1111 ,11111 1"'11,"111 1111'11" 11'111'., 11\1\\'\'\"'1, filII' '.hllllitl Ill' I'XI'II'I\Cd
III .\,1111111". IIII' I h.II,I' 11,,,',111' l'll'lallllll hll ,'OIl'll III Ih,'''' 1'111'111\.
I "I 111111' \ III \ III". ,II',I'II'I!' .. y.. h'JII!' (III II IIII'·VllI yi"~ di IIcl'cncc ell uations),
,, .. 1"11""1"" 1111 ,.llIhlhly ill'p"I"h, 'III the Irilnsfer function .Y/ (:::, II) for
,III \ ,11111", .,1 1/ I hi' 11,'1"1 IIlIlIallllll of stability for the time-varying case is
I.llh"1 III\IIhl',1 illI" 1\ 111"'11,,,,<1 separately.
\.,' Stahility l'Onditiull for lincar

111m' Vll ryi ,,/!. d i!o.lTetc systems
I h,' "1'1111111"" "I' .,Iahility for this case is the same as that for the time
1',1'>l' disl'll~sed in the preceding section. It can be presented as:*
111\111 101111
..\ "y',lI'lII I~ 'Iahk if and only if its impulse response 11(11, k) satisfies the
1"II"wlII)~ I'OIl"itioll; for some N the following inequality exists:
"
~11r(",k)l < N < 00, (3.6)
',,1 .111 II,
SlIh'I' IIIl' input 'k
is bounded, there is an M such that Irkl ~ M for all
" rhlls frolll the output relationship,
k,,1 " "

~ 2Ih(lI, k)lIl'A'1 ~ M 2111(11, k)l ~ NM (3.7)
kO k"O
Ilw. proves the sufficiency of the stability condition,
III prove the necessity of the condition, suppose it is not satisfied.
II,,'" for any fixed P < 00 there is a 110 such that
"0
2Ih(no, k)1
k ,0
> P. (3.7a)
('hoose the inputs 'Ic'
I'k = {+1
-I
if h(no. k) ~ 0
if h(lIo, k) <0
(3.8)
Wl' Ihen have

no no
C no = 2 r"h(n o, k) = I Ih(lIo, k)1 > ,P (3,8a)
,,=0 k=O
I'hcrcfore for all P < 00 we can find an input uniformly bounded by
and an 110 such that en
o
> P; hence the system is not stable.
• An alternate and more precise definition of stability is: A system is stable if for any
input sequence uniformly bounded by I, that is, Ir.1 ~ I. for all k. there is a P < :tJ
'llch that 1('.1 ~ P for all n.
If every hen, k) is bounded, it is necessary only to find the limit of

!" Ih(n, k)1
k~O
as n -+ 00.
From the definition of Jf"(z, n), it follows that the preceding stability
requirement is equivalent to the condition that ~(z, n) be analytic for
Izi ~ 1, which can be shown as follows:
From the definition of £(z, n) we have
n
.;F(z, n) == !h(n, k)zlr-n (3.8b)
A.-O
Letting k == n - j this is equivalent to
n
~(z, n) == I hen, n - j)Z-1 (3.8e)
I~O
For Izl ~ I we have

n n
.;>F(z, n) ~ ! Ih(n, n - i)llz-/1 ~! Ih(n, n - i)1 < N < 00
1-0 ;.0
(3.8d)
Therefore if the system is stable, .Jt"(z. n) is analytic on and outside the
unit circle. Furthermore, if Jt'(z, n) is analytic for Izi ~ 1, it requires
that the radius of convergence of the series is strictly less than unity.
From Cauchy-Hadamard 33 theorem we know that such a radius is given
by
lim sup Ih(n, n _ i)1 1/i (3.8e)
It is also known that if this limit is strictly less than unity, the series
{Ih(n, n - j)l} is absolutely convergent and hence for all n there exists an
N such that .
n
! Ih(n, n - i)1 ~ N < 00 (3.S/)
1=0
Hence the necessary and sufficient condition for the stability of linear
time-varying discrete system is that equation (3.8 f) be satisfied for all n,
By a similar reasoning the stability requirement for the time invariant case
discussed in the preceding section can be similarly established.
3,3 Tests for stability

We shall.conccnlralc in Ihis seclion Ilnly nn Ihe varillus lInalylil' It"".
Uowever, we IIllly nWl1linn Ihal Ih,:n' ,'xi,1 !.I'vl'lal/~r:lphical h'~h, ~lIl'h a ..
lilt! NY'I"i~1 tlia/\r:l1II alltl Ihe 1'11111-111"11'; 1I"'lhllll, which nrc Ih!.l'IIS~'~" in
"cillil d!ll'wlll'l'c ill III,' 1I1\'llIllIu'.
STABILITY CONSIDERATION FOR LINEAR DISCRETE SYSTEMS 83
The test involved is the determination of the necessary and sufficient
conditions for the roots of the following polynomial in z, which represents
the system characteristic equation to lie inside the unit circle in the z-plane:
F(z) = ao + a1z + azz'l. + ... + a,l' + ... + aflz"
with an >0 (3.9)
The various analytic tests can be summarized as follows.
Bilinear Iransformation:5.9.17
If the closed region 1%1 ~ I is mapped into the region Re [w] ~ 0 of an
auxiliary w-plane, the condition for stability becomes exactly the Hurwitz
condition. The bilinear transformation
z=--
w+l (3.10)
w-l
is the simplest possible mapping of the unit circle into the left half of the
,,·-plane. In this manner the nth-degree polynomial in z is converted to a
ratio of nth-degree polynomials in "' and the numerator is tested for its
fl'WS in the left half of the w-plane.
Let the system characteristic equation in the z-plane be defined by (3.9)
with il ll > O. When mapped into the w-plane by (3.10), the zeros of the
"'-polynomial are given by
b .. >0
(3.11)
The rclutionship between the a,. and b,. is
(3.12)
WIlI'II'
(3.13)
IIII', plOl"I'llme involve!; a complicated ~\Igebraic manipulation since

.1 ... hllllll'ar 'ransfCll'lIlatilln. and the limll constraints on the original
)I"IYII""l1al I"lI'lIidl'nl~ :1I"l~ ralhl~" involved. Thus this procedure has not
1""," .... 1"11 ·.III"I·I· ..·.l'lIl1y fur dl'sil~1I ur panllllctcr c1ll1n~e fur systems higher
.1It111 ,.1'1'1111""1 1111,,1 mdl·... lIC1wl~vt'r. il Ims the advanta~e thai we can
84 'l'lIIiOR\' ANI) AI'I'I.IC'A'I'rUN UP '1'1111 .·lMANIII'UMM
-2,0
FIGURE 3.1 A shifted unit circle in the C-plane.
apply the Routh-Hurwitz criterion or the simpler criterion of Lienllrd·

Chipart to the transformed polynomial in the w-plane.
In some cases the discrete system can be described in the '-planc'"
which requires a stability test for the shifted unit circle of Fig. 3.1. A
criterion similar to that of Hurwitz has been developed for the shined
unit circle which offers an alternate method of testing the stability of
linear discrete systems. 37.38
3.4 Stability test directly applied in the z-plane

A logical and useful approach to the problem of stability is to be ahlc t(l
obtain a stability criterion directly in the z-plane. thus avoiding ltny
transformation as in (3.10). Such a stability criterion was first established
by Schur-Cohn in 1922 and further discussed by Fujiwara3 in 1924.
However. the original form of this criterion is quite complicated; Cohn'.
has 2n - 1 constraints and requires the evaluation of two determinllnll
of order n, whereas in the Hurwitz and Lienard-Chipart criteria, only n
z-plane
FlcnIlU~:\.2 1/"" c:lrdD In IIID '-rlnne,

"
N'AIIIIIIY C"UN,um UIIClN IUK UNIAK IIllIC"KIII lIVlIlI M!I H~
1"1'""11"111"1" IIl1ll 1111'" IIIh·cmltl· deterlnillll'" IIrc retluired. ThwI this

"1 uri 11111 h'I_ IIClI rmvell ~IICCe!illl'lIl und is very rarely used by engineers
ICII hl~hc,"-urdcr lIy!ilel1l!l.
I he' IIri/tilllll 1'111'111 ClI' Ihe
Schur-Cohn criteritm for complex coefficients
C'IIII hl' !illlh'ci lis I'lIlInws. The rolynomial
/0"(.:) .... tI., + ".: + Clz:1 + ... + akzk + ... + anz"
h'I" 1111 il!! '"lIl1ls· inside the unit circle, as shown in Fig. 3.2, if and only if
II", r"II"willl-t cnl\!ltrnints ure fulfilled.
IA.I < 0, k-odd
(3.14)
1/\,1> 0, k-even
wh"I\' Ih(' 1I11111'ix tlk represents·
a" 0 0 0 0- 0_-1 On-hi
al 00 0 0 0 0- O_-hl
k = I,
2,,, .,
Dk._1 0 0 an
n
Ok I Ok. "I 00
fl., .... -°t_1 ak. =
"- 0 0 0 00 a I' •• complex
a. I an 0 0 0 00 Ot-I con-
jugate
ofDt
-
On 00
(l.IS)
Itrcclllly. this criterion was significantly simplified, and three forms of

,,11I"ilily crileria were derived; the analogs of the Hurwitz-Lienard.Chipart,
Mlllllh, lind Ihe continued fraction expansion methods for the continuous
,·.I"C'. These urc: formulated in the following discussions.
3.5 Determinant method 1,15.30
I hc' clrlrl'lninllnt method developed in this section is similar to the Hurwitz
III 1'~'"I1"(I.(,hipllrt mel hod for the continuous case. Here the stability
,"11111" minl!lllre I'urmulated in algebraic equations in terms of the coefficients
III Ilcr 1'"lynmniul /'l:). The derivation of this method is based on the
111111111 simplilicillinns of lhe Schut-Cohn determinants which arise by
' ..... 1111111\/1, Ihlll the cnef1icienls of F(z) are all real, an assumption which is
• III IIIf J'III1I1WhIK lll"I:"N"iUIiN. Ilia rCKll1I \If II plllynomilllllrc also referred to as zeros.
K(, 1'111:11"" ANn Allllllc'AlinN nl' lilt: '-IIANlltnIM
satisficd in rhylliclllllyNI~'nul, Thr flvC' "'"IU1' ,,1"'rllll"lIth"'''' wlll"',lInlllly

led to lhe concise fnrOlullllinl\ ttl' Ihe ~'nll!lll'lllnill lin' hi "!lIy ,11"",,",",,1
1111
follows.
1. The determinant I~tl givcn in equlltinn (3.15) elln he rreHelll~',II"
= IXt + Ykll Xt - Ytl
I~tl (.l.It,)
where Xt and Yk are the following matrices:
aD al aa a,\:_l
0 aD a1 aa a"'_2
Xt = (3.17)
0 0 0 a1
0 0 0 aD
and
a..-l:+1 a .._1 a ..
a..-k-ta a.. 0
Yk = (3.18)
a.._l 0 0
all 0 0 0
In the first simplification it is readily noticed that the Schur-Cohn
determinant I~tl is reduced to the product of two k-order determinants
which is considerably easier to evaluate than the 2k-order determinant
I~kl. If the ak's are complex. this simplification is no longer possible.
The determinants I Xt ± Ykl can be written in terms of Ale and Ble
which we designate as stability constants as follows.
IXt + Ylel = At + Bk (3.19)
(3.20)
In terms of the stability constants, the stability condition can be written
IAkl > IBtl. k-eVen)

k = 1.2, ...• n (3.21)
IAtl < IB",I. k-odd
• The proofs of thes'e simplifications are available in the literature.I,II,II,1G For brevity
of the discussions we only state the useful results.
",,11.·.1 1111 "rllllill I" "IWIII.·.. "'IIIIIIIp, Ih,' "llIhlhl,Y '·1111 .. 11111 .... WI' Ilhlllin
Ih,' IIlh"1 'I111I1,lIlh"IIIIIIIII,
) I h.· """1111,1 11111I1,hll"1I111111 I" Ihe' elluivnlr4'cc ClI' Ihe IUsl cunslruint
I f,,1 111,,1111 1\ !lilliI'll' III1Kih,"y ~·III1"'l'IIilll. This lIimplificaliun is based
1111 Ih,' I'lIl1uwillt'. ich'Jllily:
(3.22)
Sllin' (A". II" .)2 is II pusilive quantity, the last constraint can be
/1'1'1""1',1 hy
I''(I)N -I) > 0, n-even

(3.23)
/.'(1 )/-l - I) < 0, n-odd
II' w,' 1,'1 II" he pOllitive and multiply if necessary, F(z) by minus one,
lit,· lillilll'unsimint IA"I ?2 IB"I can be equivalently replaced by
1-'(1) > 0, (-I)"F(-I) >0 (3.24)
As :111 addition to this property we can also write
An + B/I = F(I)(A"_1 - B n _ I ), (3.25)

All - B" = F(-I)(A"_1 - B"-I) (3.250)
1 The third property which led to the formulation of the stability

1'I11lstmints for values of k of only half of equation (3.21) is given.
A! - B! = At-lAtH - Bt-IBH1 = ![(A k + l + Bk+l)(AA~1 - B4_ 1)

+ (AH1 - Bk +1)(AI<-I + BI<-I)],
k = 2, 3, 4, ... , n - 1 (3.26)
It is noticed from this property that by forcing certain restrictions on
the A's and B's before and after a certain k as required by the stability
condition, we can dispense with Ai ~ Bi.
4. The following property which can be obtained from the second and
Ihird properties is
A,,_1 + B,,_I = l[F(l)(A"_2 - 8,,_'1.) + F(-I)(A n _ 2 + B .. _2)],

n ~ 3 (3.27)
The importance of this equation lies in the fact that we can terminate
the last constraint by A .. _I - B"_1 ~ 0 only. This is evident since the
constraint on A .. -I + 8 .. _ 1 is necessarily satisfied when the constraints
88 THDORY AND APPI.ICATIDN 0' THI '·TAANIPOIM
=
on F(I), F(-I), and A"..• B" .• a" •• tl.nod. Th, U.,ruhllil or thl.
property will be shown in the dl.cuilion of tho critical onnllrlh" •• will
as in the final formulation of the Itability con.t'ltinta r", low.a,d,r
systems.
=
5. The final property relates to the determinants IXII::i: Y"I - A" B".
and states that by expanding the matrix I X 11_ I :t: Y" _" we can obtain III
other determinants for other values of k from this expansion. Thl. ,.
evident if the determinant is written in detail; we can notice thut It
includes inside itself all the other determinants involved in the stability
constraints. The matrices Xn _ 1 and Yn-l can be easily remembered Ir
their form is noticed as written here in detail.
00 01 Oz °n-Z
0 ao 01 a.._3
0 0 ao a.._4
X .._I = (3.28)
0 0 0 00
and
u. 0 ..-8 0 ..-1 a..
a8 0 ..-1 a .. 0
O. a.. 0 0
Yn- I I: (3.29)
a" 0 o o 0
It is noticed that the matrix X.._I can be easily generated by starting

with the first row from ao up to 0 ..- 2 and the second row the same as tho
first displaced by one column. This process is continued until we obtain
the n - 2 row and column. For the matrix Y..- 1• we start the first row
from the extreme right by a .. and continue up to aa. and similarly, for tho
second row displaced by one column. Continue this process up to the
first column and (n - 2)th row which has the entry a".
THE NEW STABILITY CRITERION. I •80 From the preceding simplificationll
and properties the simplified stability criterion can be written in tho
following form for n-odd and n-cven respectively:
F.?_-C
.'. . _: =)'-,_: = =Il._:
,.f,_:
Gt=~ c: =a» ~ ±a. ~=Ds D&::do Ds=Q. 011,-» =....-: &"-1 =..
k=II-3
=41,
:a.
. a. ± a,
~Q~
411 ± a• o. ± o. 0, ± a, a,:: 0, 11"-11 = II.
Q ... -s
-.-"
~
±o. ±a, k=6 0, ... 11..-5
±a. ±a, 410 ± a,,_s 411 ± 41 ..-, at ± 41.-3 a, ± 41,,_, a, ± 41,,-1 as ± 41"1 Q.-I II....
k =4
i I
=! ±a7 ±as ±a,,-4 Do ± Q.-3 Ql ± Q.-I Q. ± 41,,-1 a» ± a. a, I Q.... G .._,:, (lJlt
k=2
I
! ±an - 3 ±an_1 I 410 ± 41"-1 Ql :I:: an a. a,
:1::41.. _. :1::41,,-1 :1::41" 410 Ql at
:1::41'_1 " :1::41" 0 0 410 411

±a. 0 0 0 0 410
:1::41,,-2 :1::41.-1 ±a. Os

±a.-l ±a. 0 0 0 0 0 Do a,
±a" 0 0 0 0 0 0 0 Qo
00
>D
00 ± 02 01 ± a3 °2 ± O~ 03 ± Os O( ± 0. Os ± 01 0n-3 ± ° 71- 1 ° 71-2 ± °n
k=n-3
±03
±04
00 ± a4
±05
a1 ± Os Q2 ± o. a3 ± a, 04 ± 08 Qn-4.± On I On-a
° 71-4
±Os ±Q& k =5 ° 71 -5
±Os ±o, lao ± ° 71 -4 01 ± Qn-3 Q2 ± ° 71-2 °3 ± Qn-J 04 ±an I ... ° 71 -6
k =3
±Qn-3 I 00 ± On-2 01± 0n-l 02 ± On I °3 (3.31)
k =1
±Qn-2 ±On-l 100 ±On I °1
Q2
±an_ 1 ±an 0 00 01
±On 0 0 0 °0
±On-tl ±On_l °1 °1
=a,._1 : ±a.. 0 0 0 0 lie a1
::6, 0 0 0 0 0 0 lie
,nt\llll.l.'V C'ONIIUJ'I,."'.'IUN "Uk UN'!",. mlle'.u"" IIYS'r'!MS 91
Tilt' .'lIhlllly &!unilirulntH for II-odd IIrc
1-'(1) > 0, F(-I) < ° (3.32)
k - 2, 4, 6, ... , n - 1
(3.33)
The IIllIhility constraints for n-even are
1-'(1) > 0, F(-I»O (3.34)

(_I)t(k 11)12 (At - Bit) > 0, (_1)t(k+1I/ 2(A k + Bt ) < 0,
k - 1, 3, 5, ... , n - 1 (3.35)
.,'rum the form of these determinants it is noticed first that the

11II11,'i\'~1i IIrc symmetrical around the center. This phenomenon may be
Ill1rihlllc~ll() the stability conditions for the unit circle which is symmetric
1I1111111d Ihc origin in the z-plane. A similar symmetry also exists if the
"c'l'rliNIII'Y lind sufficient conditions are sought for the roots of any poly-
1111111 ill I In lie along the real axis in the complex plane. This will be
11t~"II!t:'l'd Illter in connection with the conditions that the roots of F(z)
III' h,'lween 0 and I in the z-plane. The second point to be "bserved is
Ihlll III "hlllin the stability constraints both determinants IX" ± Ytl are
hI hc' l'lIpllndcd, Lastly, in the final constraints both An-I ± B n- 1 ~
nrc'llllI hl~ !mtisfied. This requires the solution of two algebraic equations
°
.. ,. clrp.rce n - I. In the following discussion the last two points will be
.11tI1"iIIl'd.
"VAl IIA'I'ION 011 TilE STABILITY DETERMINANTS. 30 To avoid the need of
"I"II1!1illll of two determinants of order n - 1, we need expand only one
,16-11" lIIinllnl by using the following artificial procedure. Replace the a,:s
III ", hy ",.'N und then expand only IXn- 1 + Yn- 1 1. From this expansion,
WI' "1111 "blnin 1111 other determinants by a substitution process. For
111111 II 1I"l', I X" • + Y" .1 can be obtained by letting ao - 0, ai -+ ai-l for
, - I. 1. •.•• , n in IXn-l + Yn-II and then dividing by (-l)"-Ib n. Simi-
111I1y. wr l~lI11 ublnin I XII • 9 + Yn-al directly from IXn_1 + Yn-II by letting
III Ih,' Inlll'r (nner expansion) a o = 0, a l - 0, Q i -+ QH and dividing by
,,: All nlher delerminants can be obtained by a similar substitution
IIIIII'I'~", Tn idelllify the A,,'s and B,t's from IXI: + Ykl, perform the
1IIIIIIWillPoI'fClcesIi, Afler expansion, examine every term which is a product
III ":~ 111111 ".·s; if' it contains an even number (including zero) of b/s.
11_~1p.I1 II III A,; nlhcrwise, aNsign it to Bt . After collecting the terms of
.4, 111111 II, ,'epillce 1111 lhe ",'N by (he a/s,
92 THEOR.Y AND' APPLICATION OF THE Z-TR.ANSFORM
In this example we shall obtain the stability constraints for the fifth-
degree case. that is
F(z) = ao + alz + a.z2 + aar' + a,z' + aszS. as> 0
(3.36)
We form X, + Y. by replacing the a/s of Y, by b/s:
ao + b. al + b. aa + b. Q3 + bs
ba a o + bt Q1 + b" a2
IX. + Y,I = be bs al
(3.37)
au
b" 0 0 Qo
After expansion, let ao = 0, al = 0, a. - a i - Z and divide by -b~ to

obtain IX. +Y.I.
To identify At ± B, and A. ± 8 2 we can perform the following manipu·
lation. The constants At + B4 can be readily obtained from IX4 + Y.I
by replacing all the b/s by a/s. Replacing the sign of all the terms which
contain odd number of ba's by negative signs and replacing all the b,'s
by a.'s, we readily obtain A4 - B.. Similarly, Aa ± B2 can be obtained
from IX. + ral.
The stability constraints for the fifth-degree case are
F(I) > 0, F(-I) <0 (3.38)
(3.39)
ALTERNATE FORM OF THE STABILITY CRITERION. To avoid having to
satisfy for two inequality equations of degree n - I. we can write the
stability constraints in different forms so that only one equation of degree
n - I need be satisfied. This can be readily accomplished by noting the
second, third, and fourth properties among the stability constraints. The
final stability constraints for any n are given as
II-odd
F(I) > O. F(-I)<O (J.40)
(-l)"U'~ 1I/2(At - Bt ) > 0, (_1)tCt+ll,2(AI; + B,,) < 0,
k=I.3.S ..... n-2 (JAI)
PAl)
n-even
F(I) > 0, F(-l»O (JA "
k -. 2.4.6 •... . 1/' 7. (.1.44)
('.4~'
SUMMARY OF STABILITY CONSTRAINTS FOR LOW-ORDER SYSTEMS}·I& In
the following the stability constraints in terms of the coefficients of the
low-degree polynomials are presented. using the alternate form.
n=2
F(z) = a o + alz + aaZ2, a2 >0 (3.46) .
F( I)> 0, F( -1) > 0 (3.47)· . .;
ao - 02 < 0 (3.48) .'
n=3
F(z) = 00 + alz + aaZ2 + oaza, 03> 0 (3.49)
F(l) > 0, F(-I) <0 (3.50)
100 1< aa (3.51)
2 I
ao - a 3 < aoal - al 0 3 (3.52)
,,=4
F(z) = ao alz + aaZ2 + aaZ3
+ + a,z', (3.53)
F(I) > 0, F(-I) > 0 (3.54)
a! - a: - ooaa + ala. < 0 (3.55)
a: - a: + 0oa 3 - ala, <0 (3.56)
a~ + 20 0 a.a, + alaaa. - aoo: - a.a: - aoa~ - a:a.
- a:all - a~o, + a: + aOalaa > 0 (3.57)
I Ising algebraic manipulations the constraints for n = 4 could be more
'1lIlIplificd so that instead of the two second-degree equations (3.55 and
I. '1(,). we obtain the following two first-degree equations:
01 - 03 < 2(04 - ao) (3.58)

0, - a3 > 2(ao - a,) (3.59)
" 5
F(:;) = ao+ alz + aaZ2 + a3z3 + o,z' + arl', 05> 0 (3.60)
1:( I) > 0, F( - I) < 0 (3.61)
I(lul < or. (3.62)
(tI.II,(l6 - Qoa: - a 2 a: + a: + 0 00 805 - 0 0 0:) =F
( 2
",(16 - ooaaOa - + OoOa
3
05
2
+ aOa&
2 »0
- °oo,a, < 0 (3.63)
(II: lI~)1 - (CloU, - al05)2 + (0 00, - 010&)(0! + 0 103
- ,,~ - U2U, - a: + 0:> + (a003 - aaos)
x (11 11/,. - U,06 - UOU3 + 020,) - (0002 - 0a05)
X 1(11: - ":) - 2(1/ 11114 - U.(16)] > 0 (3.64)
• WI' III11Y IIl1tr tllllt 1"111111 chi\ rll"llllllll we IIlIv,," II. t· II. ::.-- O.
04 '1IIIIlIIt\' ANn AI'I'I It AIII1N en 1'111:'" KANIII'UKM
Theile cn".ln,lnlli rlUl 1,11111 he 1I1"'1,lIlIed hy IIIM"hllll,· lIu"ulmhuhlll_ 1111
Ihlll h'lItall" nf the twn third.daMr.:r "'IUllllm'lI (.UI.\) wr hllvo
(1M)
In concluding this scclinn it shnuld be pointed nul Ilml thc dctcl'lnilUu"

stability mclhou cnuld be useful for numcricllllcliting oJ' pnlyncuninlll flit
stability and with its alternatc form they yicld the numhcr tlJ' runlH IIlIlIIldc
the unit circle if the system is unstllhlc. This will hc i11l1strlltcd Intcr In tho
discussion of the number of ruots outside Ihe unit cil'cle. If Ihe !itllhility
constraints arc further simplificd as indicutcd for" ..., 4 lind" - S (1I11i1l"
equations (3.58). (3.59) and (3.65» this informutiun un the nUlnh,'" til'
roots is lost at the advantage of more simplified sttlbilily cunstrninlH. Fur
system design this simplification is only of academic vullic. for with hi~hrr
degree polynomials the final constraint. that is.
<00
A n_ 1 - Bn _ 1 >
is of major importance, as will be discussed in dettlil in the folluwlnM
section.
3.6 Critical stability constraints for system dcsign211
In certain cases where the discrete system [whose chaructcristic cllllnlinn
is F(z)] is initially stable, and if one or more parameters (for c'UIIUI,Il'.
gain) are changed so that it becomes unstable, then only certain conslrllinlll
need be satisfied. These constraints are called "critical construints" nnd
are discussed in this section.
Assume the characteristic equation of the linear discrete system III
described as follows:
with II" >U

C:\'M)
This equation can be also written in terms of its roots Zj as
1,2, .•• ,n
F(z) = Qn
,
II
t
(z - z,) (.U.1 )
We can readily ascertain Ihut F( I) ;> () und ( -- I)" F( - I) >. () &:CllIlIt illll"
two of the critical cnnstruinf..'i, hCCIlII!iC Wt' "1111 Wl'itl'
1,1•.... "
/o'(I) .... CI" II (I /,) (HIM)
III A1111 II" I'tINlIllJlllA III1N 11111 IINI All UIIII'IlI II IIVNII MN I)~
111111
'.. , ...• "
1),"'( I) - II" II (I +- :.) (3,69)
II " n'lIclily IIlIliw,1 I'rum these two equutions that F(l) = 0 and
N I) • n "I1I1!.tilllte the critic,,1 conditions when the real roots move
"1I1~lIlt· Ih,' unil drde lit ;: = I and z = - J,
l'ullht'l'lIIl1n\ the (Jther crilicnl constraint which yields the condition
\\ 111'11 tht' l'IIlIIpll.'x runts lIIove outside the unit circle is given as20
1,2, ...• 11.
·1
'" I
.. /l , , 1-- (-I)"'" IIIBa,,·1
"
IT
,<,t
(I - z,z )
,It
(3,70)
11", 'A' (cllnjugnte), Ihcn '::,z" = 1::,1 2 and An,_, - Bn ._ 1 changes its
~I/~II Wlll'lIl'ver /I pnir (or udd pairs) of complex conjugate roots move
1I1I ..... h· the unil circle.
WI ... II tilt' "Ilmplc" ruots arc of even multiplicity, the equation
A" • - 11" 1=0

h "h" nilkal although it docs not change its sign. In all these cases the
',~',Ic'1IIh'·l'IlIllI.'S ullst:lblc whenever An .1 - B n_ 1 = O. Therefore we can
"l1l1ll1llll'i/l' Ihe preceding critical constraints discussions as follows. 20
I III 1111 illilially !.tahle linear discrete system, a certain parameter is changed
'II' Ihlll I,tahilily ccuscs, the following are the critical conditions which
vlI'hl Ih,' IlliIximum variation of parameters for stability limit.
1-'(1) ~ 0, (-WF(-l) i!: 0 (3.71)
(,.- 1)"1" 11I2(A" I - B/I ,I) ~ 0 (3.72)
II .... t1i'l'ussinn shows that for analytic tests for stability when applied
III Ihi' .h',i,'.11 III' cerluin systems, it is not necessary to consider all the
',I,IIIIIIIV l"IlI1!.lraillls hut only the critical ones. There the solution of the
111",11''',1 "c·/',rt·c c'luutinn is at most of (II - l)th degree; furthermore, this
''11111111111 lIt'cd he solved I'or only real roots, which can be easily done
"I'll'hll'lIl1y. III a fullllwillg section an illustrative example indicates the
npl'llI'lllillll IIIllI impnrlance ()f this point .
.t,7 Numher (If mots or a real polynomial

i.. ,itll' the unil drde l ,4
IIII' ',lllhllily nih·rio .. cliSl'US!\l'd curlier in this chapter yields the neces-
~,IIv 111111 .. 1IIIidt·1l1 l'lIndilillll lill' al\ the mots of Fe::) In lie inside the unit
96 TIII:ORY ANI) APIII.IC'AIiUN 01· 1111; .·IIANltnll ..
circle. To obtain inl'ornultinn un Ihe nlllnhrr ul' mu'" Ihlll 11" Inilid. ,t"
unit circle if the system is unstllble, we ~an mudily Ihe I"c\tllltli lll,·
cussions to cover this casco Following the work uf Schur·Cuh" I'rr"c"lld
in equation (3.14) and using some of the: pruperliell "I' the IIt .. hlllly
constants At'S and B,:s we can state the foll()wing.
The number of roots of F(t.) that lie inside the unit circle iA ellUII
to the number of variations of the inequalities of the folluwing C:UI\-
straints:
n-even n-odd
1>0 1>0 >()
IAll ~ IBII ~ A. < ()
IAll ~ IBII IA.I ~ IB.I
1.0411~ IB21 <=> AIAa ~ BJBa IA.I ~ IBal ~ A.A. ;.; B.B.
IAal ~ IBal IA.I ~ IB.I
1.04.1 ~ IB.I <=> A8 A, ~ BaB,. IAIiI ~ IB,I <=> A.A. ~ B.B.
IA,I ~ IB..I IA.I ~ IB.I
IA .._II ~ IB_11 1.0411-11 ~ IB_II

IA"I ~ IB"I ¢> F(l)F( -1) ~ ~ IA"I ~ IB"I ~ F(I)F( -I) ~ ~
It is noticed from the preceding that for n-even we do not have
evaluate the even determinants but can use the relationship
'n
A: - Bf = A.t-IA~l - B.t-lBI>+1
== ll(At+l + BI>+t)(A.t-t - Bt_l )
+ (AI>+1 - B~I)(A~l + B._I»)
which are obtained from the calculated odd determinants. Similarly,
odd determinants. are avoided for n-odd. It is assumed thlll in lid.
,h.
counting of number of the inequality variations, none of the cunstllllill A,
are equal to Bt . Thus the number of ruuts outside the unit cirt:lc I.
equivalent to n minus the number inside the unit circle.
To determine the number of roots insidc Ihe unil circle when A.
Bt-l but A" == Bt • a speci .. 1 prucedllre is relillil'cd I"r thill Hingulllr CIINI.
I"
~I AI'" II V C'IINIUIIIIlA IIUN 'U" IINI'AM IIIIIC "tI II lIVlIlI-Mli 97
" ,II'IIl,h,.1 ,11""",,""11 III' Ih,' 1II1I,1iIil'lIli.H1" r\'tllli"'~,1 til ,I\'tcrmine Ihe
'I
1IllIlIh,'r III 111111'1 ,h"'\'II",,,\',1 in "PIl\'nlli" 2.
J,M Rclnt inn:o.hip hetween the determinant

1II"'lhlltl and lIurwitl. critcrion l ,20
I hi' 1'I'llIlllIn,lnp h'~lwc"1I sUllie uf the stability constraints and Hurwitz
1'1111,,11"111111 .. mill~~ Ih,' hilinear transfurmation can be established. If we
\\ IIh' IIII' rllllllwilll~ lIurwit.l. pulynolllial obtained from F(z), with Q n > 0
hv lI .. illl'. Ih,' hilinc:lf Irnnsfurmation
(3.73)
hu = (-I )nF(Z)/',,_1 (3.74)

(3,75)
\\ h."", 1.\:,' II is Ihe next-to-Ielst Hurwitz determinant. The relationship

II "I h:" I,..~'n wrilied by expansion for low-order systems and can be
1"'III',ally tI~'l1Iun~tmted through the use of Orlando's formula~,21) and the
1IIIIIII'al' Il'all~limnaliun .
.'.1) Table furm:!·ll.III,I:I.:l4.11:;

In 1111'. 'I'~OI ill II an alternate form of the stability test for the unit circle
,', •h·.1'1 I ....,·, I. rhb furm is presented in a table form similar to the Routh
III"'" 11'01'" IiiI' thc ~llIhilily in the left half of the s-plane. It requires
"\lIlnallllll IIf unly secund-order determinants and can be easily pro-
1~""lIlIlI'tI 1111 a digilal computer. The table form is particularly useful
III' 1111 II lI'ril'a I leslin~ uf polynomials as well as for design purposes. The
1IIIII1IIIalillll III' Ihis furm is bused on the early work of Rouche, Cohn,
1I1I.lllIh·1 1111 Ihl' work uf Marden. The derivation is presented in Appendix
I II', \\1'11 iI" • hI' varillus reccnt mOllifications and their extensions.
IIII' III·,·,', ..:!r}, alltl sullicient cllndition for the following polynomial
Q II >0
(3.76)
hI h,I\I' all II .. rll .. l~

im.itlc Ihc unit circle can be obtained by forming the
,II I lahl,'. (II !lhlluld be lIoted that the clements of row 2k + 2
111111',111\,1111',
I ""'.1·.1 III Ilu' 1'1"111"1110; IIf Ih .., ruw 2k + I written in reverse order.
A II. I. .', .... 1
Row to Zl Zl
J 00
2 an
3 bo
4 bn - 1
---1--- - - - - - - - - - - - - - - - - - - - -
5 Co
6 Cn _ 1 Cn-s Cn_, Co
---1----------------- (3.77)
---1--- --------------------
2n-S s, ! s, i
So $3
2n - 4 i $11
S3 St
' . _ ••• 1
$0
---1--- --- --- --- --- --- -----
where
'0 = I So
S3
'2 = I So
S3
The stability constraints are
F(1) > 0, (-I)nF(-I) >0 (3.78)
laol < an, or a.. > 100 1
Ibol > Ibn_II
leol > len-21
Idol> Idll-al (n - I) constraints
It is noticed from the table that the dotted entry r 1 is not needed for
calculation; hence the corresponding computation involving the dotted
entries in rows 2n - 4 and 2n - 5 becomes redundant.
The number of roots inside the unit circle, if none of the preceding
constraints becomes equal, is given by the number of the products Pk ,
k = I, 2, ...• n, which are negative, where
(3.79)
PII = [1001 - all] Ibol - [lbn-lil ...
X Ir21J(1r~ - ,:1 - Irorl - r,rall (3.80)
[lrol -
The difference between n and this number yields the number of roots
outside the unit circle.
EXAMPLES
(I) F(z) = 3 - 2z - =z2 + Z3, 11 = 3 (3.81)
By applying the constraint 1'(-1) we notice that F(z>I, __
1 > 0; thus
Ihe system is unstable. Therefore the table need not be computed. How-
ever, to obtain information on the number of roots outside the unit
circle we form the following table:
Row ZO %1
31
3 -2 i- 2i
'2 - 2
3
L~~.! 3 (3.82)
5
3 8
2
P. = 1001 - an = 3 - 1 > 0 (3.83)

1'2 = 2[1hol - Ihll .. .11 = 2(8 + :> > 0 (3.84)
p .• = pus. pos. {l82 - (~)21 - 18(-:) - :(~>I) >0 (3.85)
11"'I'l'furc nu ruul exilits inside the unit circle and all three roots are
1IIII',i,h' Ihe unit circle. It may be noted in passing that the condition on
II';; I'ia II'". - '.'lIlI nmy also be obtained from F(1)F(-I) and the
IlIlIllIplklllilln IIrlhe IIlher cIIllslmints. For inlitance in this case the sign
III Ihl\ h' .. ", i~ l'lluivulcnl III the liign or /0'( I)F( -I), Ihat is,
"Mn II": - ':1 - 1"lIr. - r.,'.11 "'" S~1l f( -I W( I) (3.116)

100 THEORV AND APPLICATION 01' TIIH a-nAN.POIM
In view of the preceding relationship the dolled entry In lho lalll IIno
need not be calculated.
(2) F(z) = 0.0025 + 0.08z + 0.4126.;11 - 1.368.;3 + .;'. n-4

(3.87)
Row ZO ZI Z2 Z3 .;4
1 0.0025 0.08 0.4126 -1.368 I

2 I -1.368 0.4126 0.08 0.0025
...................
3 ~ -1 1.368 -0.41161l -0.0834
4 -0.0834 0.4116 1 1.368:
: ................... :
-1
5 0.9936 !':i402'i
: ................:
0.5256
(3.88)
Stability test
F(I) = 0.1271 > 0, F( - 1) = 2.703 >0 (3.89)
laol < a .., 0.0025 <1 (3.90)

Ibol > Ib'HI, 1 > 0.0834 (3.91)
ICol > IC n-21, 0.9936> 0.5256 (3.92)
The system is stable.
Simplifications of the stability constraints l 2.24

An alternate simplified form of the stability constraints given in equation
(3.78) can be obtained by noting that •
bo = a~ - a! < 0 (3.93)
and
b~ > b~_t, or Ibol > Ibn-II (3.94)
These two constraints can be equivalently written as
bo - bn - 1 < 0, bo + bn-l <0 (3.95)
Similarly, for
do = c~ - C!_2 > 0 (3.96)

and
(3.97)
NI'AlIII.II'Y t!UN.II)ItItA I'IUN I'UK LINhAlt UlS('KI! 1'1; lIYS"rIlMII 101
we can write
tiD + d,. .3 > 0 (3.98)
Similar rclutioll!.hipli can ulso be obtained for the other constraints. Thus
the stability conslrnint fur f,z) can be summarized as follows:
/I-even
F( I) > 0, F( -1) >0
00 - an < 0, 00 + an > 0
('0 - (',,-2 > 0, Co + C,._2 > 0
eO -£>n-4>0,
n constraints
So - S2 > 0, So + S2 > 0
n-odd
F(I) > 0, F(-I) <0
bo - bn- 1 < 0, bo + bn _ 1 < 0
do - d n_ a > 0, do + dn- 3 > 0
/0 - /n-5 > 0, /0 +/n-5 > 0 (n - I)
(3.99)
constraints
'0 - r > 0,
2
This form can be further simplified by noting that each of the above
constraints is divisible by the preceding constraint with alternate signs;
that is. ('0 - ('II 2 is divisible by ao + ant do - d n - 3 by bo + bn - 1 • etc.
This property is utilized in the next section to show the equivalence
between the table form and the determinant method.
R('/tltiollship helll'e£>11 Ihe lab/e form

tile' c/('/('rmiIlQIII method 24
(11/(/
The relutinnship hetween the stability criterion introduced earlier and

'he: '" hie form will now he established. The importance of this relationship
102 '1'lIIn\tY ANI) AI'I'IIC'A'IIUN UI' 1111: "'RANlltuIlM
is thut the: eVllhllltinn (If hi~h~r urdcl' d,'lerrninllllh IIIlIy h" 1111111'1111",1 hy
obtaining their value directly 1'rIlIlI Ihe 1IIIlIe I'urlll.
To establish the identity between the A" ± II" uf Ihe prcccdin~ crilcrinn
and the coefficients from the tuble, we may nute the rulluwill~ hy ,lircci
examination of the terms Al ± Bl and All ± 8 2 lind the curresJlllllllinlt
terms of the table, that is,
(3, WO)
(J.IOI)
Furthermore, by regular division, we can further identity the I'ull(lwill~

relationships.
A. =F Bol == do =F tl" 3 _ _.
(a: - a:')(Aa ± II.)
(3.102)
and
A =F B = eo =F ell -5 (3,103)
5 5 (a: _ a~)2(b: - b~_I)(A3 ± B3)
Using the induction method, we can show the following general relation-
ship.
A =F B _ 10 =F III-Ir+l
" ao - alla)1c-3(b0ll
" - (I - bl11-1 )k-4(Co2 - CII_I
I )"-5 ' • ,
X (h: - h:-k+3)(A/r-Z ± B". •)
(3,104)
where the I's and h's are the appropriate entries in Table (3.77).
By imposing the stability constraints on the coefficients of the tuble for
n-odd and II-even respectively. we readily obtaih the constraintli on
A" ± B" which are noticed to coincide with the determinant method.
This will be illustrated for n = Sand n = 6, respectively.
n = 5:
The stability constraints for this case in terms of the construints in
equation (3.99) are
(J.IOS)
(J,IO()
1-'(1) > 0, ,.~ -I) < () (l.W7)

NIA"'IIIY C'UNIiIItl'MAIlIIN IUK IINI A" UilltlU II IIVIIINt!!'! In.l
111111 lilt 111111 .... I It, - "" t "., A. I II. - (till j tly )/I(f1ii _. "r.)(A w =F 112 )]
,,,,,I "" - II::
,,~. I h"~,, \'I1I111lmlnl!o III lerms of A's nnd 8's CIIIl be
wrllll'lI hy 111111111' rfllm "" f ". < () Ihlll hll < 0; we finally obtain for
II-~
Au -t 11K < () (3.108)
(3.109)
F( I) > o. F{ - I) <0 (3.110)

II .. (,:
III this casc the constraints arc
(In - (18 < O. 00 + 0& > 0 (3.111)
(3.112)
(3.113)
F(I) > 0, F( -I) >0 (3.114)
lJ!>ing the relationship introduced in equations (3.102) and (3.103), we

ohtain
AI - BI < 0, AI + Bl > 0 (3.115)
A3 - B3 > O. A3 + B3 < 0 (3.116)
Ar• - B.i < O. A5 + B. > 0 (3.117)
F(!) > 0, F(-I) > 0 (3.118)
Following the foregoing procedure and utilizing equation (3.104), the

clluivalcncc for any n can be readily established.
The stability constraints for n-even and n-odd from the table form
could be written in an alternate form that coincides· with similar forms
from the determinant form. This form can be obtained for n-odd by
starting with constraints 0 0 - 0" < 0, 00 + an > 0, and thus the final
constraint is 'II - '2
> O. Similarly, for II-even starting with bo - b n_ 1 <
o ~Ind ho + hIt I < 0, the final constraint is So - S2 > O.
To show ihat '0 '2
+ > 0, or (so + S2) > 0 is automatically satisfied,
wc rewrite equation (3. \04) for k = n - I to obtain
'0 + '2 = (An 1 + B n- 1)

X [(a~ - CI!)"-'(b: - b!._I)"-s ... (An_a - Bn - 3 )]
(3.119)
104 rur.ORY ANn AJlI'I.IC'AI'IUN III' lltll .·rltAN.I1,"M
Since all the constraint!! in the bracketed tern,,, "r" IIltth"I,," fur .,"hlilly
from the constraints before '0 + '., the con!ltruill' '" .. '. iN tl"p""""'"
only on A,,+l + B"_I' From the properly on p, K7, thuliN,
A,,_l + B,,_1 = i[F(I)(A""'2 - B,,2) + F(-I)(An" + I," .>1,
0,120) ,,~:\
we notice that A"_l + B.. _l is satisfied from F(l), fX-I), and All aT
'0 '2
B,,_2 constraints; thus + or (so + S2) constraints are also siltisficd
from F(l), F( -I), and the preceding constraints.
Therefore we can state that the critical constraints for stability from the
table form are
F(I) ~ 0, (3.1200)
and
'a - '2 ~ 0 (3.120b)
To facilitate the derivation of Ale ± B/c from Table (3.77), we can further
simplify this table as foIlows:
Row 1.0 1. 1 Zl ... %,,-k . .. %"-1 Zn
1 00 °1 °1 ... Dn-t . .. 0,,-1 an

2 an a n-l an-t .. . at ... QI Do
3 bo b, bl bn _ 1
4 b'l-! b.. _t b..- a bo
5 Co C1 C. ('n-I
6 C.. _I C.. _I C,,_, Co
7 ~ d'I d; ... d~_.

8 d~_, d~_. d~_5 ... d: (3.121)
9 r~ r; ... e~_,
10 e~_t e~_1 ... e'•
2n - 5 .I'~ .1'; .1'; .1';

2n-4 .I'~ .1'; .1"1 So
--
2n - 3 ,~
" I
" t
where h/c and Cle are the same as in Table (3.77) and the change occurs in
die' _-I Co C,,_t-1c I..!. ' _I d~

e/c- d~-~1c I.!. , ... ,
C,,_I CI; bo d~-3 dl; Co
(3.122)
'11 Aim" y • ClN_IIII:U IIUN IHIt IINI AM 11I!l1"M1 II ToI\'ToIII MS \()~
I hI' .11''''111111111111 .. III.' 11111111,.1 hv Iht' ,hvi!oiull uf the lirst terlll of two
111\\ ~ h,'IIIII' 11""',11111.1111', 1·"II ..tlllt·li,,1I the st:,hility clIllslraints from this
Itlh'" Ill" II 1a'l1 I 11"111 III SI·h .... -( '"hll determinants.
1~31 = d~ < 0,
< 00 n-even
'"
I4Jo n _l 1 ='0'2 -'2'2 > n-odd
(3.123)
hlllll Ih,' pn'~'t'dillg t:lble we can easily identify Ak ± Bk. as follows:
A, ± B. = do ± d~-3
bo T bll _ 1
As ± B" = (e~ ± e~_,)(ao ± a,,)
Co T Cn-Z
A ± B = (f~ ±.f~-, )(bo ± b"_t)

6 6 (d~ T d~_3)
(3.123a)
Similarly. all the others Ak ± Bit. can be readily obtained from the
~illlplilied
form of the Table (3.121).
Gt'Ilt'rotion oj the stability consl,ainlsz'I,Cl

In the following discussion the stability constraints which are obtained
frum the table form will be easily generated without the use of the table.
This generation is based on a simple rule which can be easily remembered
and applied,
GENERATION RULE
Given the following real polynomial (whose coefficients are given by
symbols) whose roots are to be tested for stability within the unit circle.
F(z) = 00 + a1z + az%2 + ' .. + 0,,_2Z"-2 + O"_IZ"-1 + 0nz",
an> 0 (3.124)
Perform the following manipulations.
I, Form the first row from the original polynomial coefficients by
replacing only a ll _l and an byb n_1 and b" respectively. The second row is
formed from the first by reversing the coefficients and relabeling certain
of the 01t.'S by bk·s. Form the third row by expanding the second-order
matrix formed from the first column and all the other columns (starting
106 TIIIIORY ANI) APPI.I(~A1·ION 01' 'rllil I-TIANlmIM
first with the column at extreme risht). ThoAO ara .umn1ltrI7.ed ft. fullow.:
Row
1
z'
a.
Zl
al
z·
a.
_
-- ....__I"a".,.•....... . ..
.
I" I
II" I
I"
II"
b.. _J
2 b" b".1
-_ . _b........... fli
".
3 a: -b! aaOl - b".lb" a.b, - b"a".1 a.b" • - "all" tlob" I - "I""
2. Starting with D~ - b! form a sequence of expressions, each fmm the

preceding one, by the substitution given in row 3 with relabeling sume
coefficients, that is by replacing
DO - 0: - b~ b" - Dob,,_1 - Dlb"
01 - DoDl - b.._1b..
b.._1 - Dohn-a - Dshn
De - ODDs - b.._zb,. (3.124t1)
Continue this process to generate the required n - I constraints. After

performing this process, replace all the bit's by DIt'S to obtain the tlnnl
constraints. These n - 1 constraints with F( -I) > 0, (- 1)" F( -I) > 0
constitute the necessary and sufficient conditions for the roots of F(.) tn
lie inside the unit circle.
By applying the preceding rule we shall generate the stability constraint"
for the following fifth-degree polynomial:
F(z) = Do + D1Z + DaZ'J. + 03'1.3 + D.Z· + 0&'1.1, Ofi >0
(3.125)
1. Form the following three r!>ws:
Row
I ao a, al as h. hi
2 b, b, b, bl a, flo
3 Qi - bl aoQI - b,b, aob. - asb, aoba - a.b. a.h, - n,".

2. Form the following four constraints by replacing the various coo",-
cients from the third row, that is,
aD-aD - II "5 . ~ Do". - tI."a
, b' }
01 - aoal - h.ha h. ~ l'U"3 - 11."11
(lIH,,)
al -.~ aoo_ - "a"" "11 •4'u". - 11.1,.
aa .. ~ tlotl, - "_"a I,. • "0". - "."11
IIt'AhlUl'Y C'UNlllUltU'UUN t,(llt UNIIAI mAC'111I'11 IIYII'I'IIMII 107
"~a .
Th. """.trAlnt" .,,"
- "',- IX
": - ,,: •(u: - ,':)' - (aob, - a,h.>·
- Ct' - (aob, - albrl = P
(' •(" - 1(,,: - ":)(1/0". - a,b.)
- (aOul - b.b,)(aob, - a 1b,»)2 = "
" .". - {[(a: - ,,:)1 - (aob. - a1 b,)I][(a: - b:)(aobll - a.b,)
- (uoa. - h.h,)(aob, - alb,» - [(a: - b:)(aoal - b,b$)
- (uob. - a,ba)(aob. - a 1 b,))[(a: - b:>(aob. - a 2 b,)
- (l,oU, - b.ba)(a.b. - alb,)W
- "I _ AI = ~ (3.126)
l4y re'rll,cin~ "lIlhe h/t'ft by a/t's we finally obtain the stability constraints
i"( I > > 0, F(-I) <0 (3.127)
IX ~ ,,~ - u: < 0 (3,128)
II = Ct l - (aoo, - alaS" >0 (3.129)
" := (I' - [(a: - a:)(aOa3 - ala.)
- (aoa l - a,a,)(aoa, - ala,»)1 >0 (3.130)
~ == "I _ AI > 0 (3. t 31)
Thclle' cun!ilnaint!i could be simplified to obtain
','( I) > 0, F( - J) <0 (3,132)
(tI: - a:> ± (aDa, - 0laa) <0 (3. t 33)
"±A>O (3,134)
The' I'rcc~'din~ !ihuw!i that for n-odd, we can obtain these expressions
d'rl'~'lly f'rnrn Ihe polynomiul coefficients by certain substitution of two
""'1'11 III "limc, Thi!i cun be accomplished by forming rows 4 and 5 to
"hlllin Ihc rClluired !iuhstitution,· Similarly, for n-even we can skip the
Inll'l'Inrtlilllc Cnn!ilrllinl!i by using two steps at a time.
Tlar rl'I"Iinn!ihip " ± Acan be also written as
" I A ..... (,': - ":)(,,: - I':> T (aoa. - ula,;)](A, ± B.)
(3.135)
• I "' 11I~11"'l·". In Ih,. l'll\t "" • (,,: - ":,. - (n.,h. - a,".)',
"1 •I,,: '.:11",01', ".".' - (flJ.. . ".".)(".1'. - II,'''), II. ~" ..
10K I'IIWlty ANI) AI'Plle'AnUN CJt Iliit '-IIAN.tUM,..
where A, andB, are given in I'"
ellflier dlll~u"!llnn. ""n~" Iho "IIIIIIIII,hll
in equation (3.134) can be replaccd [noting "'Iulllinn (.'.1 .I~)j by A I I II.:>
0, A, - B. > o.
Furthermore. the constraint A, % B, can be ailio ubillined Ihun lhe
following equivalent form of,.. % A. that is,
A, ± B. + IXclI
I
yCIII
..L
a IIt'"t
;;I; (ll3t,).
(a~ - a:)[(a~ - a:) T (aoa, - a,QaH
The terms Xru and y~2) are obtained from (3.17). (3.18) for It. .. 2 by
replacing the elements of X2 , that is, ao. Q10 by the two step suhstituliun
using row 3 and replacing the elements of Ya• that is. b,. h~, by II !timilllr
substitution as indicated earlier. After this substitution. the "It'S IIrc
replaced by a,;5 to expand the matrix.
Therefore we can replace (or obtain) any of the kth-generated cnll-
straints by a certain relationship from the expansion of the IIIl1tr;II
IXLj~j % Y~~)bt-lit' which indicates the relationship between the tnble furlll
and the determinant method. The preceding equivalence could be Itlsn nl'
advantage in numerical testing of polynomials where we are not retlui!'cd
to expand high-order determinants.
In conclusion, we discuss how to obtain the stability constraintli by It
certain generating rule which can be easily applied. This rule is silllilltr
to Routh rule number 2 for the continuous case. 10 The pmof nf thi"
generation process is not presented, but it can be deduced from the identity
of the table form and the determinant method.
lIIu.rlratiL'e design example

The critical stability constraints developed in an earlier section are nnw
applied to a design problem. For the sampled data feedback syslem
shown in Fig. 3.3, we shall obtain the maximum value of the gain It. fur
the stability limit. .
For the parameters given, the z-transform of the open-loop tmnsler
function is given as
~l(Z) = <[1 -s e-'l'· e-U&,'l'
s(s
k
+ 1)
]
= 0.2223k (z + 0.03)(% + 1.755) (3.Ll7)

%2(% _ 1)(% - 0,368)
• Another form of this equation can be written as

.'II . 6.--
· IX :i: r"ll
A. :!: B - - - -•.- ...... -
":-":
1114111111V C'IINIIIIII:UIIUN l'IIM IINI"" 1)1111 MI II IIVIIIIM'I lUI)
Hf.'
II •
I () ••
X .
-1";-';';'
_ •
('(II)
)
-I. ...._... _ _ __
7'.1
b' • 1.25
l'I(lIIRI U A SlIlIlplcll-daC;I system with pure delay.
I hi' "Y'Ih'1I1 l'Iull"lu'l~'rislic C4uution for stability investigations is

H.-) - I + ~ql(z) = Z, - 1.368::3 + (0.368 + 0.2223k)%2
+ 0.3974kz + O,0123k = 0 (3.138)
I hI' 1II"I",,'tlilllt i). II fourth-dcgree polynomial and we notice that if
II. A • , Ihl' ).y).ICIlI is stable, This can be readily ascertained from a
_~"I .. h III Ihl' "lilli-locus us u function of k, Hence we apply the critical
111111111111111'1 for" .. 4 10 obtain the maximum gain.
HI) ~ 0, k>O
n I) > n, k. < 16.75
I hi' IIlh"1 nilicnl <:ullslruint from the determinant method is
or k < -247 -5.5 < k < 0.72
1111111 1111' IlIhle, form Ihc critical constraint
('" ('2 ~ ()
I hi .. 1'Il1I'llrninl clln he written as

",," ('~ = (II" + a4)(A 3 - 8 3) ~ 0 (3.139)
!'\ltlll' 1111' l'l'ilklll (crm is A3 - 8 3, in both cases we obtain the maximum
t!IIIII 11'1 A11111_ :< n.72,
.\, J() Division mcthod'4,24

IIII' 1111111 rorlll or ohlaining a stahility test is using the simple method of
11111'\11111 Ill' polynomials, The derivation of this method can be obtained
1IIII'I'IIv I'l'lItn (Ill' IlIhlc form as follows.
Slabilily 11',\'1
A IIC'C't'\~III'Y 111111 )'1I11k-icIII condition for the real polynomial

n:)· II" ~ til': + Cl~;:~ + . , . + ak::'" + . , , + a..z".
with an >0 (3.140)
110 THIOIlV AND A"LlrATION O' TN. "TIANlI'OIM
to have all its roots Insldo the unit circle In tho I· plano t. obtained ••
follows.
Form the inverse polynomial F·(z) as
I'
F·{z) - znF ( ~) - an + aft_Iz
+ an-tz' + ... + an_",1e + ... + a.,zn (3.141)
Obtain the stability constraints ctle by simple division as shown.
F(z) Fl(z) FI(z)
F*(z) - CI.o + F:(z) - CI.o + ctl F:{z) - CXo + Otl + Ot,
Fa(z) ,(z)
+ Fa(z)
-.- ... - Oto + ctl + ct•••• + Ot,,_11 + 'F"..• ,,(,)
(3.142)
where FI(z), FII(z) • •• are the remainder polynomials obtained from tho
division and Fr(z), Ft(z) . .. are their inverses, as defined in equation
(3.141).*
We now satisfy the following constraints for the stability test.
F(I)
>00
> 0, F(-I) < n-even
n-odd
k = 0, I, 2, ..• , n - 2 (3.143)
The proof of the test lies in establishing the identity between the Otle's and
the stability constants aD, boo Co•••• of the table as follows:
aD
Oto - -
an
bn- I
Otl--
bo
(3.144)
• To obtain the inverse of F,(z). we factor out the con\mon fllclnr , or Its ruwerll hornr.
we apply (3.141); also. ror F.(,) and FI(a)•••••
I rAlllU rY r'ONIIUItRA I'IUN 'Oil 1.INteAl mlle'IUTI' IIYII'fIlMIi III
"r"m Ihl ""IhUlly cnnltrlll"llI hnl""lcd prcvlnully on Iho ,,'II, h'" , , .•
Ih.~"I"'lrltl"llI In hi hnpullcd "n tho «.11 arc readily clltablillhed u!ling
"I""I1"n (.tl44).
I h. n"mher nf' rnutll nf 1-'(:) l",deJe the unit circle is equal to the number
..I' prmhlL'l1I I't which nrc ncglltive, where Pit ill defined as
lit - (1«111- 11[_1

locI I
- IJ ... [_I
loc,l
- I] ... [_1
locII--II
- IJ
k = 1,2,3, ... , n (3.145)
\\lth
I', - lI'Xol - 1J. PI = [locol - 1][_1

locil
- IJ ' ... (3.146)
."d I, .. ,I ill uhulinccJ from

f" II:). - .« I
+ 1-'"
- (3.147)
,•. - (.) " 1,'*"
.. I "
J;)CAMI'I.li
/0'(:) - 0.0025 + 0.08: + 0.4126z 2 - 1.368z3 + Z4 (3.148)
,..-(:) _ I - 1.368: + 0.4126z2 + 0.08z3 + 0.0025z' (3.149)
In thlll L'IU~C we: need obtain only OCo, OCIt OC2 as follows.
1 - L36H: + 0.41 26z1 + 0.08z3
./- () 002S-.) 0.0025 + 0.08z + 0.4126z2 - 1.368z3 + Z" (00025
. .. 0.0025 - 0.0034z + 0.001z2 - 0.002Z3 ~ Oz" .
: - 1.36H.:1I + 0.41 16:3

0.0834% + 0.4116z1 - __ 1.368z3 + z" (
+ O.OH34:' ) 0.0834: _ 0.114%2 +\6~Oj42;~.+ 0.007z' 0.0834
.... _---,
. -- ......
U.993: 1 -~j .4()22.:~:'
. ---.+ O.S2S6%')0.S256Z1 -':f.:.~}~~+ 0.993Z'(0.53
II .. hllIlM he: ,,,,tcd tlml in obtaining /xo, OCI' and OC2. the terms encircled in
dult",1 h""11 need nut he: cilicuillted.
III Ihlll l'lI!!l',
"'u - ().(K)2~ QC, - 0.0834 ex. - 0.53 (3.150)
112 'fllllORV ANI) APPLJ(~A'rI()N or 1'1111 .·raANI..mlM
For stability constraints,

F(I) = 0.1271 >0 F( - 1) - 2.703 >0
0.0025 < 1
0.0834 < 1
0.53 < I (3.151)
Therefore the system described by equation (3.148) is stable.
3.11 Aperiodicity criterion for

linear discrete systems 18 •13
An aperiodicity condition in discrete systems is obtained when ull the
roots of the characteristic equation lie on the real axis in intervul (0. I)
in the z-plane. Such an aperiodicity condition yields only a finite number
of maxima and miliima in the system response (less than n. the order nf
the system). This condition is important when excessive oscillations nrc
not desirable as, for example, in instrumentation systems.
For aperiodicity criterion, we obtain the necessary and sufficient
conditions (I) for the roots to be real and (2) to require that these root!!
be between (0, I).
The necessary and sufficient conditions for the roots to be real can he
obtained by using either the Romanov conditions,ls which require Ihllt
the discriminants Do. D1 ••• • , D"_I (to be obtained from Table 3.1) be
positive, or alternately, by using Fuller's conditions34 which arc bused
on Trudi's determinants. In the latter method, for the roots to be relll it
is necessary and sufficient that AI< > 0, k = 2, 3, ... , n, where Ao - I.
AI == no" (always positive) and An is given by
a" Qn-l 0.0
0 AI
a" Q,,-1 Q,,-t °"_1 a._.
AI
0 0 a" Q,,-1 Q .. - I Q .. - I ",
Al
A. = 0 0 0
~ (n - 1)"'-1 (n - 2)a •• "1
0 na.. (n - 1)11.. -1 (n - 2)a"_1 (n - 3)/1"" 0
0 na .. (n - 1)11"-1 (n - 2)11"_1 (n - 3)a"" (n - 4)a" •
na. (II - 1)11.... 1 0
(3.1~2)
In the second form we obtain" - I conditiuns to sutisfy for the rout!!

to be real. The D,,'s of Romllnnv ure ubtuined frnm Tublc 3.1.
To obtain the necessllry und lIuf1\ciellt C()lltiilitlllll for the roul!! ul' (.l.140)
III AlIIIII V I 'UNIUlltMA IIUN I'IIM IINI:AM IIlIlnnll: IIVIIII'MII 11.1
hi III' 11,,,1 1111,1 In II" h"twrl''' Itrn ""el unily, thC' f'nlinwinJot IIelelitinlllll
I ",,"11 II In ... 1111 e'III"'1 1I1Il' "I' Ih,' IWtl I1Iclhtlllloo outlinccl cun he uscd,
1II1111f',,, ""II/,Y/flrll/(Illfm II
1111' 1"1'1"'IIlIl~ Irnn"l'nrllllllinn invulves mapping the real segment [0, 1)
II' III" ,."llInl' ""ltl Ihc IIcglllivc rcul axis of the II'-plane. It is given by
IIII' '"II"wlll~ I I'll lI!>fllrlllltt ion :
w :::
. - '. or w =-- (3.153)
II' ,I z- 1
II 1'111111111111 (.l.I S.l) is suhslitutcd into (3.140). we obtain
"'t(lI') _ /1,,11''' + R" IW" I + ... + Bn_kll'k + ' , . B t ", + Bo
(3.154)
II,,·"'hllillmhip IlI'tween the (l's of (3.140) and the B's of equation (3.154)
I, 1"\1'11 !ly
(3.155)
\\ h"I" r!/',! (r - k)! (3.156)

111'111 ItlllllllllllV'S or Fuller's conditions that the roots lie on the
""l'IIII\I' 1I'1I11I\i!oo, we rClluire thc additional constraints that
II" A > n. k = n, I •...• n (3.157)
11",,,'1,,", Ihc IIcccloosary und sufficient condition that all the roots of
I , '·11111,,· Iwlween l.ero and unity in the z-plane is given as
/),>0, k.=0.1.2 •...• {1I-1) (orAk>O.
k=2.3 •... ,n) (3.158)
111111
" (' 1')
I )A '~A II. = 0, J, ... , n
II" r > 0, k (3.159)
1'1"11111111' ('\'I~K) IInel (J.159) yield 2n + 1 constraints.

I 'AMI'II'
WI' IIhllllll lilt' "llL'rillllil'ily condition for the following polynomial:
'I) I II,.: I- tI:!~:!
lin + U3::3, Q3 = 1 (3.160)
I "11,111111,,, fill' I'l'al mills
Ie "'"I1I1I1V', \I('ll'rlllillants
.lll~lIl1 > () (3.161)
'111ull~ - 211: + IOClllCl,Cl2 - 9C1~ >0 (3.162)
""n": '4(1~ I IK"lIll,l1~ - 271/~ > 0 . (3.163)
114 'I'IWURY ANI' AI'PI.IC 'A I'IUN ()Jo I'III~ '·I'RANIII'URM
TABLE 3,1 ROMANOV'S <':ONDI'I'ION l'O OU'I'AIN I).,

k-O 2 4
•. _.4
i
Q~ 4aua~ ala, - SQoa, ,a. - 6"a'"
o+
- 2aoaz a,a, - 3aoa3 Qlaa - Q
, -. :
la: - 4aoa,- 2asa. - Saoa. - 2a.a, - 6aoU, - la.a, - 7a"", -
2a1Q. 3ala, 4al a, Sala,
3a: - 6aoa.- 3a,a, - 7aoal - 3aaa, - 8aoU. -

2
4ala, - laaa, Sala, - 3a2a, 6alal - 4a,a,
4a: - 8aoU.- 4a,a, - 9aoU. -

3 6ala7 - 4a.a, - 7ala8 - Sa.a? -
laaas 3a,a,
Sa: - 10aoa,o- -
4 8a,a. - 6a.a. -
4a.a7 - 2a,a,
9
1This matrix is symmetric with respect to ttte majn diagonal.
ID~ is formed by the first k rows and columns,
alf the system is of lower order than 10, put the higher order coefficients equal to zero
2, For aperiodicity conditions in addition to the preceding (through

transformation)
1 + a2 + al + aD > 0 (3,164)
a2 + 2al + 3ao < 0 (2.165)
al + 3ao > 0 (3.166)
aD <0 (3,167)
In this case Fuller's determinants yield
A2 = a: - 3a1 > 0 (3,168)
A3 = Q~a: - 4a~ - 4a oa: + 18aoalQ2 - 27a: >0 (3,169)
11"lIll.1tY "'NlII'•• ArIUN tnll UN.. ". UIIIC'IlJ:IJt I""UMII Il~
7
"
~
II, ttl I""., "'''' - H"..". ".u. - 9ao'l. "I". - IOaoGlo
.. __ .._---
'.'" H"ull, 2"111 , - ""ot1. - 2(a80. - 4a,as - 2aaae - 9a l a l0 loa
b",II, 7a,u. 500010)
---
••11. '1",,11_ ~- lusa, - 1000010 - 3aao. - 901010 - 3aaae - 80.0.0 3aa
-1"", •.- ~t"UI 8a,ae - 60aa a 7aaa,
-- -
,II',11. - IlIt,,,II, - 40.0, - 901010 - 4a~a8 - Sasa.o - 40,0. - 7asal0 4a,
H,
',"- - (,11.11. -
411.U,
7a.a, - Saaa, 60.0,
--
"~,'" ... 90,11,0 -
"I,',"- - SII,a. -
SOia, - Sa,alo-
60.0, - 4a,ail
Sa,a. - 7aaal0 -
Sa,a,
Sala, - 60,010 So,
30,0,
60,0, - 70,0,0 - 60,0, - 60,010 - 60,0, - Sasolo 60,

'",_".
': - 80,010 -
50,0, - 3aaa, 40,0,
'" - 40,0,20,0, -
7a~ - 60,010 - 70,0, - 50,010 - 70,0, - 40,0 10 70,

40$0, - 20.0, 30,0,
-.
So: - 40,0 ' 0 - 80,0, - 30,010 Saa
20,0,
-
9a: - 20.010 90,
10
Ilnd factor out from the nth column (and row) a. (finally a:).
Critical constraints
Similar toclitical stability constraints, we can obtain the critical aperiodi.
city constraints, which are given as
An ~ 0 (3.170)
or
(3.171)
These equations indicate that when a parameter is changed for an
initially aperiodic system, the determinant An is the first to become zero
when two roots become complex.
1t6 "III!()RY ANI) API'I.U'A'rrUN ni' '1'1111 .·'I'UNltI'URM
When the roots remain relll hut mOYe Cllltsidt" the re~itln (n, I), tht"
following critical constraint is satisfied:
F(1) ~ 0 0,172)
and
F(-I) ~ 0 (3,17.1)
Equations (3,t71) to (J.t73) yield the critical constraints for uperiodicity,
3,12 Theorems related to stability

and number of roots2 •26,28.32
In this section, certain theorems are introduced which give a quick
check on stability and on the number of roots within the unit circle,
Monotonic conditions
If the coefficients of F(z) are such that
(3.174)
then all the roots are within the unit circle; thus the system is stable,
This can be easily verified since the constraints given in equation (3,7M)
arc readily satisfied, that is, ho < 0, Co > 0, do > 0, ' , , .
As a corollary to the preceding theorem we can state the following, If
o < a" < 0 ..._1 < , , , < 01 < 0 0 (3,175)

then F(::.) has all its roots outside the unit circle. This can be easily
verified from the formula of number of roots inside the unit circle.
Furthermore, if the coefficients of F(z} are all positive, the roots lie in the
ring m ~ Izl ~ M, The values m and M are, respectively, the smallest
and the biggest of the n-numbers38
(3.176)
Relationships between coejJicients2 •2S

If the coefficients of F(z) are such that
lOki> 10,,1 + 10,,-11 + . , . + law I + IUk .1 + ... + Itl.l -I Ifl•• I

(3,177)
then the polynomial is unstable,
IUAIIII" Y • ClNIIIII,IItA "UN WM UNI'AM UtIl"M'IIt' NYII ... MII 117
III 1'111\1" Ihl" 1111'111"'111, II-!
,J.( l) _ ",,,It (3.178)
.lII.1
VI(O:) - ""~" I ... I· tlkl.:'·11 + tlk .:£0 • + ... + 0 0, (3. I78a)
II WI' I'III1~id~r Ihis Cllullliun fur 1::1 = I. we have
I"I(~)I .~ ItlA·1 > 1",,1 + ... + lawl + lak .. 11 + ...
1- Itlol ~ 1",.:" + ... + akHzkll + ak_1zk - 1
+ ... + aol = Itp(z) I (3.179)
SUU'I' "1(.:) I VI(:) = /0"(:), we can write
,.., .• ) vl{::)
.." 1 -I- . -- (3.180)
"1(.:) if,(:)
IIII\\"'VI'r.I~I(.:)1
,,1(." )
< I; therefore in the mapping of equation (3.180) when
,h",,'rih,·s Ih,' hllundury of the unit circle, the origin is not enclosed.
WI' k IIIIW 111111 ,"'::) = tI,.:'· has k roots inside the unit circle (at the origin);
Ihll ... 1I"lillg '~'Illaliull (3.180). F(z) should also have the same number of
1111,1', III .. ill .. Ihe IIl1il circle. Therefore the remaining roots of F(z), that is,
" A. ·,IlIlIIld lie uutside the unit circle, and thus the system is unstable.
AI ;II;m(l.\' pri"dple 23
IIII'. 1I1l'lh",I is II se I'u I in giving the stability conditions for certain

I'"IYlllll1linls F(.:) whose coefficients (with an = I) are less than unity.
II 1"- huscd IlII giving precisely the largest root of an equation which
.1'"11111111"" Ihe furlll of F(%). For stability considerations the form of the
.IIII ..lll1illl is givcn by
ItI.. kl::;; I'"~ 1_ p", k = 1,2, ... ,n - I,
an = I and 0 ~ p <1 (3.181)
111111
ItllIl ~ /1" , (3.182)
A "III hi lily dllirl fur different values of k as a function of p and also for
,II n""'11 I vullles tll'l/ up to 7 is given in Fig. 3.4. The calculation of the
, 111111 ,'.111 hc l'xtcllded to higher-degree polynomials of F(z).
Ut\MI'I.I!
I II .1I1I.,1.llh' Ihe applkatioll of the chart we test the following polynomial
III. ·.llIh,hly:
F( .. ) .,:4 I 11.2.:3 + 0.01:: + 0.032z -
2 0.024 (3.183)
11K '1'111'0"'1' ANU AIII'W'A'I'InN ClI' 1'1111 .·rUN.llfll,..
III
(I."
OK
0.7
0.6 i
-I
0:
Q,
0.5 !
:t:
0.4~
0.3
0.2
0.1
0.9
0.8
0.7
0.6
...~
...
Q, c:t
-I
...
I
Q,
on
0.5 C
...'u
:r:
0.48
0.3
0.2
0.1
FIGURE 3.4 Stability chart for linear discrete systems.

III A"" II \' nIN_II'flU linN 111M IINIrAM Itllll HIli NYIUI Mil IIV
1'111111 Ihr d,,"1 wr "",I
111111 lUI \'rlllll" mit""·,, "' vlllul'S "I' " <
H.M
Ih,' 1111"1111111111'" (I.IKI) 111111 1 UK.') 11ft' 11111 i!lIl",1 , IIlIcI helll'e the system
h"l1 "lIlh 11111'" 111,,111," Ih,' IIl1il l'lI'de,
NIIIII/"'" I~I """':i III.vitl,' tl drdl' tI/ radius < 124

(J
I II "hlllill illflll"lIlIIliulI UII Ihe number of roots inside a circle of radius

". I. WI' IIlIIy IIpply IIcerlain transformation on F(z), that is, z' = z/a so
Ihlll Ihl' d""'I' "I' mdius /1 becomes equivalent to the unit circle. Hence
Ihl' 1"lIl'l'l'Ilill~ ,liscIIssiul1 can be readily applied to this case.
I 111'. !,1'1Il'l'clure is useful in testing the relative stability of the character-
ft
1,,111' plllynomial. The constraints for F(z) = I
1'~C)
apZlI
.,
= 0 to have all its
111111 .. lII~jllc = (J < I will be obtained by replacing in the constraints
Izi
'"I1I1c1 'ill' " ~.: I the following:
(3.184)
/\\ an extension of this theorem, we can also obtain the stability

1"IIIIIIiliClIIS or the number of roots of a shifted unit circle. This is based
1111 Ih,' following theorem by Marden.
To uhtain the !lumber of roots in the circle 12 - sl < I, replace the
",,'s "I' 1-'(::) by
(1" .... ~
It~p
~ (k)P .~It-/>Ok (3.185)
where thc binomial (k) = k!/p! (k - p)!

.p
III particular. we obtain the number of roots in the unit shifted circle
a~ shown in Fig. 3.1 by letting s = 1 in the shifting transformation.
Theorem related to F(z) haL'ing roots on the

Ulliteireli' Izl = I and reciprocal roots'l.,4.31
In the preceding case when F(z) is such that in the table constraints
certain ho = h n- k or Fit 01 == 0, the following theorem is useful.
If for some k < n, Pi "#- 0 for all j ~ k, (that is, go"#- gn-ItH in the
table form). but Fit., 1(%) == 0, then F(z) has n - k zeros at the zeros of
I'~(::)' These zero!; are either on the unit circle Izl = 1 or occur in pairs
that are inverse relative to the unit circle or are a combination of both.
Furthermore F(z) has p additional zeros in the circle. where p is the
number of negative Pi forj = 1.2, .•. ,k, and it hasq = k - P additional
zeros outside the circle.
120 .... "'UKY ANn AIIIIIIC'AIIUN UI' 11111 '·llANII.... M
liXAMIII.I!
Let F(z) == 1 + lz + Zl + lzl
1.0 '1.1 Zl 1.1
--
F(z) 1 2 1 2
F*(z) 2 1 2 1
(3.111(1)
F1(z) -3 0 -3
Ff{z) -3 0 -3
F2{z) 0 0
It is noticed from the table that 100 1 < la"l, FI(z) == 0; thus one rool
is inside the unit circle and two roots are on the unit circle (since no
reciprocal roots in F1{z) exist). To obtain the location of these roots we
solve for
F1(z) == -3 - 31.· == -3(1 + '1.1) == 0 (3.187)
Therefore the roots are at '1.1 ,1 == ±j.
When F(z) has no zeros on the unit circle, but ho == h,,_t, the number p
of zeros of F(z) in the unit circle may be obtained by a limiting process
as follows·
We replace Ft{z) by the polynomial
h(z) == Ft{rz) (3.188)
which, because r == 1 ± e and e is a sufficiently small positive quantity,
has as many zeros in the circle as does Ft{z).
EXAMPLE
F(z) == -2 + 3'1. + lz2 (3.189)

Here it is noticed that a~ - a! == 0 [and F(z) has no zero on the unit
circle]; therefore we replace z in F(z) by rz == (1 - e)z, with e == 0.1;
thus
h(z) == -2 + 2.7'1. + 1.6lz1 (3.190)
In the preceding case a~ - a! == 4 - (1.62)' > 0, and
h{l) > 0, ftC -I) < 0, or Ib.1 < Ib,,_11(3.191)
Therefore PI> 0, PI < 0, one negative sign; hence one root exists
outside the unit circle and the other root inside the circle. It should be
noted that the same results are obtained if e is considered to be negative,
that is, r == 1 + e == 1.1.
III "'11111 ~ IIINII,,,,''''' IIIIN ... 11 IINI All 111"'"1 II "" .. 11 M'I 1],1
-"'\"'1111 1I .... htlll~ 1111' IIVllllllhll' Whl'l,'ltV "\1""'11"" III' 1IIIIII'"i"l'IWt' III'
"'III~ 1111 .111' 111111 ,lid" "'111 hi' ,11''''11111111,,1.1 h,''o'' 1I",lhll'l!. which ClIll be
"1'1'111'11 1111 h'''IIII'~ Ih,' IIlIlIIh"1 "I /t'I'II'o elll Ihe IInil circle lire discussed
III AI'I"'I"I" "
,4/'/"'11'//\ I. /)c'ril'tlli"" I~/'I/Il' l"hIt' flm" of sfobilily4

I" 1111'. I1l'l'e'mll'l WI' shull derivc Ihe table form of the stability test by first
IIhllllllllll~ I'll' 1'C1II,liliullS nil the number of roots inside the unit circle.
III 1111'. 1111111111· ... Ih,' slubility condition is obtained as a special case when
"II Ih.· 1111'" Ill' 1-'(.:) = () IIrc within the unit circle, Before we derive the
'OIII,hllllll' 1'111' the Ilumber of roots, certain properties of F(z) and its
111\ I· ..... ,. +(,'1 us well as uther associated polynomials are introduced,
I ,'I II", "I1I1 .. idercd real polynomial to be tested for the number of roots
III' ,Ie',",Ie'" a, F(.:) and written as follows:
f(.:) = lIo + tl l Z + a"z' + ... + 0/(1," + ... + anz n
II
= ao IT (z - z;)
I 1
III mltliliull, consider the sequence of polynomials Fo(z) = F{z), F1(z),
1 .•1 I.". I'~(::) whose properties will be studied in detail. If we associate
III ":lI'h III' Ihe polynomials Fj(z) the polynomial Fj*{z) whose roots are the
11''''prlll'al with respect to the unit circle of Fj{z), Fil,(z) can be obtained
hlllll I)::) as follows:
F'H{Z) = a~JlF,(z) - a!:~JF:(z), j = 0,1,2, ' , . ,n - 1

(A.I)
h-I
Fi(z) = 2 o~ilzk,
k~O
with ,,~OI written as a k (A,2)
and
j = 0, I, ... , n - 1
(A.3)
The following properties can be deduced from (A,I) and (A.3),
I. If Fi(z) is of degree n - j, then FH 1 is at most of degree 11 - j - I in z.
2. The coefficients of F j ,.1 can be obtained from the coefficients of FJ as
follows:
(A.4)
In each polynomial Fj(z), the constant real coefficient aW> is denoted
by hi so that
.I.
r1 i+l -_ [awJ'
0 - [a IJ
n-i> J' - l
- a 0J+1) (A.S)
122 l'lIIi()RY ANI) APltI.lC·A'I'IUN UI' 11111 "fUNII'URM
3. The zeros of "j-(I) arc the inverMe!l of' the 1.erUIi of' ,.~(.). r,hulv. In
the circle Izl = l. In particular. it' ,.~(:) hll!! nl' lern nn tha 111111 ~lr,~I.
and Pi zeros inside the unit circle, then I'i-(:) hall (II - J - p,) lernl
the unit circle and no zero on the unit circle.
' 'I'd.
4. The absolute value of Fi(z) for z = eiO is equivalent tn the .lbllClluto
value Fj-(e iO), that is,
IF Aell) I = IF:'(e") I (A.6)
The preceding can be easily shown from the relationship
F:'(z) = zra-IF,(~) (A.7)

When z = eiO, we get
F:'(e i') = eilra-i"FI(e-;') (A.H)
Since le i (ra- 1II1= 1, then
IF:(e I = IFI(e- II ) I = IFI(ell) I
j
,) (A.9)
To illustrate the mentioned properties, we shall consider the following
third-degree (n = 3) example.
F(z) = Z3 - 2.5z1 - 7z +4 = (z - 0.5)(z + 2)(z - 4)
Here
0.=1, a8=-2.5, 01=-7, ao=4
From (A.3), we obtain the inverse polynomial F*(z).
P(z) = Z3F(~) = z3(Z-· - 2.5z-8 - 7z-1 + 4)

=1-
2.5z - 7z· + 4z3 = 4(z - 2)(z + 0.5)(z - U
To obtain F 1(z), we use (A.I) for j = 0
FI(z) = aoF(z) - oraP(z) = _3zt - 25.5z + 15
= o~llza + a~llz + 0~1I
It is noticed that FI(z) is in this case one degree in z less than ,.~:). Ity
noting (A.2), we readily identify, O~l) = IS,! oP)
= -25.5, a~1) - -3.
These values can also be calculated from (A.4) for j = 0 as follow!!:
Oo
U) -
- at0 - a 3
l - 15
-
~Jll = 0oa 1 -
aaa. = -25.5
o~" = aoaa - a.o. = -3
which are the same as those obtained from (A. I). From (A.5), we ellll
calculate 61 as
61 = 0: - a: = J5 _ a~1I
IIAIUlII\' I'UNIIUUAIIUN tn.. UNIIAII UIIIC-1U1I ""IIIIMIl 12.1
Th. "mil .. I" N.) .r. 'I - +n.'. 'I - -2. 'I - 4. lhl' IcrUII uf
,.'.,.) IIro).. . •• I. which A." the hwcrliClN Clf Ihe I.crUII 01' f(:) with
",,,,,,,,'1 III the unit "lr~I,·. l,'urthcrmClrc. it hi noticed Ihal /-l::) hus one
If,., hull"e Ih~ unit circle Ilntl ""(z) hus " - I 11:1 2 zeros inside Ihe unit
,'lr~I",
Nlmllllriy. We emn culculutc /'.(:.), FaC:.), ,sa. and 6a following the same
,,,'m·rdurr. Iluwever, II systematic: method for obtaining the Fs(z) and
",11 I" Iu "'uIINlruel the following table for the discussed example.
,-"11,,,,,,,,,,,1 :0 zl zB I %3
,.'(:) 4 -7 -2.S I
,.'.(,) I -2.S -7 4
,",(1) ~I - IS -2S.S -3
,.•
"'r(:)
(:)
Ft(I)
d.... 216
-3
-4S9
-2S.S
-4S9
216
IS
Fo(z) d. ~ -164,000
.. 1_ noticed from the table that F3(Z) is a constant and thus has no roots
Inllhtr Ihl' lIlIil circle. In addition, from the table we obtain 61 - IS > 0,
I~. - ]U, > n, Ilnd ~3 ~ -164,000 < O. It should be noted from the
lilllilwillf( dhicussiuns that the sign of 6H1 is important in determining
1I111h11lly ur the lIumber of roots inside and outside the unit circle.
IItrlClIlI'M: If "~(:.) has PI zeros in the unit circle C and has no zeros on
(. And II' "111 "
0, then FIH(z) has
I'" I - Un - j - [en - j) - 2pj] sgn 6H1 } (A.IO)
' 'IU'' in (' And has no zeros on C.·

'I'll rruvc Ihis theorem, we assume first that 61+ 1 > O. From equation
(~.I) we CllII write
"'",..<:1- _ 1 _ a~'IF~(z) (A.lI)

1I!"I",(2) a~/'",,(Z)
IIIIIII~ e'lulltinn (A.S) und the property of equation (A.9), we ha¥e
"", , '- (:.) I

• II .. l'IlR
' n "
"~" ,.',(:)
<I
whl'1I ,.'(" hlllller"!1 UII

(A.l2)
the unit circle will be discussed in Appendix 2,

J24 THEORY AND APPLICATION OF THE z-TRANSFORM
Now if we apply the mapping theorem to equation (A. I I), we notice

that FH1 (z)/aV'F;(z) does not enclose the origin when % moves on C.
Therefore we deduce from Rouche's theorem (or from Cauchy's mapping
theorem) that the zeros of F1+1(z) are of the same number PI as the zeros
of F/(z). Since sgn 151+1 = I, this number is in agreement with (A.10).
Furthermore, inequality (A.l2) makes it impossible for FmCz) to have
any zeros on the unit circle.
If we take the case 151+1 < 0, we have, from (A.I) and (A.19), the
following:
F H1(Z) = a~IIFI(z) _ 1
(A.13)
a~.!.IF~(z) a~.!.IF:(z)
and
(z) I
Ia~.!.iF:(Z)
a(IIF
<I
(A.14)
By the same reasoning as before, this case shows that the polynomial
F/+l(Z) has in C (noting property 3) the same number (n - j - Pi) of
zeros as a!!~jF:(z). Since now sgn ,)i.tl = -I, this number is also in
agreement with the formula (A. 10). Likewise inequality (A.14) makes it
impossible for FiTl(Z) to have any zeros on the circle C. Thus we have
proved that the theorem of (A. I0) is valid in both cases.
If we apply the formula (A.lO) to each Fi(z) in equation (A.J), we
readily obtain
PI = Un - (n - 2p) sgn 61 ] (A.IS)
where po = P = number of zeros of F(z) inside the unit circle.
P2 = ~[(n - 1) - (n - 1 - 2Pl) sgn 62]

= H(n - I) - [en - 1) - n + (n - 2p) sgn 6d sgn ~z}
The expression for P2 is the special case of the formula
Pi = U(n - j + I) - (n - 2p) sgn (~1~2 ••• ~;)
+ sgn (~263 ..• ~j) + sgn (~3~1 ..• IIi) + ... + sgn 1i,1
(A.17)
Let U5 assume thnl Ih~ prl'l'~cJilll~ formula ha!t hl'l'lI Vt'rilit'cJ nlso fill
j = 3,4, ...• k - I, ami We' will !thow hy illllllC'lIClII Ihal il hold .. lot
all k.
Letj =k- I in equations (A.IO) and (A.l7) to obtain
PI< = Hn - k + I - [(n - k + I) - 2p,,_d sgn o,,}
= H(n - k + 1) - [(n - k + 1) - (n - k + 2)
+ (n- 2p) sgn (0 102 ••• bk - 1 ) - sgn (b 2b3 • , • 15k _ I )
- sgn (bab•... b"_I) - ... ~ sgn 15,,_1] sgn b,,}
= Wn - k + I) - (n - 2p) sgn (0 1152 , , • 15k )
+ sgn (b 2b3 ••• 15k ) + sgn (153 15• •• • 15,,) + ... + sgn b,,}
(A.I8)
It is noticed that (A.IS) is exactly equation (A.17) for j = k. Thus by
mathematical induction we have shown that (A.17) holds for all j.
2 ~j ~ n.
If we let j = n in (A. 17) and note that Pn = 0 because Fro is a constant
I,~rm (see Fa for the case n = 3). we have
o= I - (n - 2p) sgn (15 1152 bn ) + sgn (0 2153 ••• 0,,)

• , ,
+ sgn (153 15•••• an) + ... + sgn 15'1 (A.19)

Solving this equation for p, we obtain
p = !(n - !sgn Pk ) (A.20)

2 "=1
wh,'rc
P" = 01152 , •• 15k • k = 1,2,3, ... ,n (A.21)
rhc formula (A.20) is the key relationship which yields information on
Ih,' Ilumber of roots inside the unit circle. To interpret (A.20). let us
'!c'lIole: by 'II the number of negative Pk , k = I, 2, ...• fl. Then as (11 - 'II)
I'. Iht.' Ilumber of positive Pk • we may write (A.20) as
P = Hn + '11- (n - 'II)] = '/I (A.22)

whirh Icads us to express finally this important theorem.·
'11 for thc polynomial
F(::) = 00 + 0 1% + . , . + On%n
/' III Ihl' products P4' defined in (A.2I) are negative and the remaining
" I' OIf" positive. then F(z) has p zeros in the unit circle 1%1 I, no =
/t., .... 1111 Ihis drcJe. and 11 - P zeros outside the unit circle.
1,"111 Ihis J.:cllcral thcorem, we can deduce that the necessary and
·.,,'hlll·1I1 "olldilillns for F(:;) =
0 to have all its zeros (or roots) inside
,h.' 1111,1 l·ird,· is
".' II, b2 > n, ":1 > n, ... (i,. > 0 (A.23)

"It" It ~~'vc'!t, illlhi!t "11"",/' = II.
12Cl 'rIIltUkY ANn A""I,U'AIlIIN CII' 1'111' .-IRANII'UU.
A convenienl wily In nblllin~, .md I'~(') (aullluIlL'C'd 111111111", Ihlrd ,1"Mrl'"
case) is to construct Tllble (A.24)
TABLE A.24 A STABILITY ItROCEI)URL FORM
Poly"omial Row
~. _.. ,.
F(z) I ao al til" , II• •
".
F*(t.) 2 an an I ...
-_._. . I" "-
FICz) 3 <S, = a~lI all'
0
alii
a .. "
, tI~ll,
Fa·(z) 4 alII
n-a alII
n-I
..
... II~II
01 = a~·'
F._a(t.) 2n - I "n-a = a~n-U a~·-II

-_ ..
a:n-al
_- -- ----
...
F:_a(z) 2n a:n- II a~n-lI

--_..
F.Cz) 2n +I <Sn = a~'" a~nl
F:(z) 2n + 2 a~'"
CA.24)
In Table (A.24)
a l2'
o -
_I a~ll III
a~I~1 1
aol1l ' •••
(A.2S)
Q .. -l
Furthermore, it should be noted that the elements of row 2k + 2 consist

of the coefficients of the row 2k + I written in reverse order (k = 0, I,
2, ... , n).
The number of roots inside the unit circle if lJk " 0 is equal to the
negative number of the products of
k = I, 2, 3, 4, ... , n (A.26)
where
PI = 151, P 2 = 6 16 2 , • • • (A.27)
The necessary and sufficient condition for all the roots to be inside the
unit circle is that the negative number of the products Pk be n. This
condition is satisfied if and only if the following exists:
61 < 0, 62 > 0, 63 > 0, ... , 15" > 0 (A.28)
1114HIIII\' CUNII""MAIIIIN IIIIl I INUit 111'11 MI'II '1V!lIIMli 127
III 11'1111111 'llI't'llIl tll~t" III III1I1"'II~'1I1 p"IVllllllllll" wh~'11 "u III l' III' the
I I II' \It. "'III', 111 I I '1111' " ..... I,I ) III Inhk (A,~4) I" III' tle~ree less thull
" I, \\ hll h 11I1.,.hl II II lim Il' Ihlll "llIlll' III' the '~A.''' arc Illissing, Ilowcver.
III III...... Ul .... ~ \\t' 11111,1 IId.1 1111 the rC1luired zerus in thc tablc for F1(:::)
_II Ihlll 1111 Iht' ,\" "ppl'lIr IIl1d thus Ihe dctermination of stability or root
11 ... 1111111111111 1'1111 jlron'l·tI ill the usual munner.!1tI Similar procedures
~hllllill h,· 1I~,·tI ",lwlI the!o.e cascs ()ccur for any of the intermediate poly-
1111111 hi h III ',.1,) III Ih~' elll rics of thc table, Exam pies of such special cases
fill' 1111".\'111\',1 ill I'rllhlelll 3.9,
"I~II'I 1111 A III IN or 1111: STAllIl.ITY CRITI:RION. 24 To simplify the criterion
\\ 1111 h \\ III "lIable liS to wrilc Table (A.24) for the stability test, we relate
Ihl' 1111111\,,'111111 betwcclI thc Ilk's and Schur-Cohn determinants /).t' The
Ikll'llIlIll:llIl" '~A arc discusscd earlier, and they offer an alternate method
1111 IIhlalllill~ Ihe numbcr of roots inside the unit circle. The connection
hd \\ ''I'll 1\ alltl D.k was first established and proved by Marden 4 and is
1"\"11 hy Ihe fullowing relation:
h" 9)
~"
A _
- \k ,2 -'1<-3 .I: (A.2
111 U2 ••. flk-2
I hi, 1'llll:1Iiun could be used to calculate the /).,,·s from the b,;s, thus
11\ IIlding evuluation of higher-order determinants. When k = n. equation
IA ..")) yields
!:i
"
= -,n-2 -,,,-3 .l:n-4
CJ n
.I:
(A.30)
U1 (J2 liS ••• U,,_2
III
(A.3!)
I:rom the discussion of equation (3.22) it is indicated that
~n = A~ - B~ = (An-l - B n_ I )2F(l)· F( -I) (A.32)
Suhstituting this expression in (A.31), we obtain
n" = b~·-2 b;-3 15;-4, ... , CJ n _ 2(A /I-I - Bn_ 1)2F( I) , F( - I),
n ~ 2. (A,33)
Since (A n _ 1 - B,,_1)2 is always positive when, 15 1, b2 , • • , , bn - 1 yI: 0, and
~incc for the roots to be inside the unit circle b2 , b3, ••• , CJ n _ 1 are all
positive, the last condition ~n > 0 can be equivalently replaced by
15" > 0<=>( -l)nF(I)' F( -I) > O. (A.34)*
• This equivalence can also be shown from geometric consideration of the arrangement
of the zeros without the mathematical verification using equation (A.29).
12M IIIWllY AN.. 10'1'1." AllIIN I'" 11111 , IMANlltllMM
If we Ict II" > u, "n > U ..edu~·,"" In

> O. ,,-cvrll
f~l) > O. /-'( -I)
< 0, II-odd
It should be noted that the last constraint in CllulUinn (A.2M) \'lllIhl I",
replaced by equation (A.3S) by letting an > O.
We may utilize the relationship (A.33) to obtain the siMn nl' I'" wllhnlll
having to calculate 6" as follows. We multiply (A.33) hy ,\.,\•... ,," I
and use (A.26) and (A.27) to obtain
If the number of negative PIt (k = 1.2•...• n - I) is even,
sgnPn = sgn [F(I)F(-I)]
and if the number of negative PIt (k = 1.2, ... ,n - I) is odd,
sgnPn = sgn [-F(I)F(-I)] (A . .1~1')
Since the connection between 6,. and 6,. = A~ - Bi is obtained, we Ciln

apply all the significant simplifications obtained on the A,.·s and B,.'s In
6,. or to the stability problem using this criterion.
Utilizing equation (A.3S). we can rewrite (A.24) in the form of Table
(3.77) by noting the following from (A.24) and (3.77):
bIt -- all)
,., C - ,. - a,., (2 )
,. ,.,
d - a lS) ,,- "....
e - a (4 ) .
(A 36)
and
.I
Ul -
_ a(1I -
0 -
b0 -- as0 -
at
n'
.I
U n-l -
_ aln-1) -
0 -
,a _ ,1
0 O•
.I
un -
_ (
'0 -'a - '0'1 - '1'2)2 =
2 I)I! ( Inl
ao (A.38)
Appendix 2. Singular cases in

determinant and table !orms2.31.35.38.40
Sometimes the kth determinant 6" or the kth entry in the table 6" vanishes.
making the method of determining the zeros of F(z) inside the unit circle
appear inapplicable. Such cases of vanishing 6,. (or when A,. = B,.) or
6,. are called singular cases. One such case of the table form has been
briefly discussed in Section 3.12. and the purpose of this appendix. is to
present in detail the modifications required to enable us to determine
'11\1111 " ' '11"/'111111""111' 1111' IINIAII 111'1'111 II '1''111'''' 1."1
I 1IIIIh IIII' IIlIIlIh"1 .. , 111111'. 1I1·,I,h·. 1111. III 1I111·."h' Ih,' 111111 "11'1'1,·. I he
" ,"11. ,III· ·.I.,It't! \\ /Ihlllli 11/11111',.111111 11'''''''11'',·, III 1111' ""l'iValillll will be
I"" II I .. '.\"'."'11111111" II,,· '"·.'·11·.·.11111 lin' 1I11I"ili",1 IIIclhods to he applied
11111 .. '.1111'111111 'i".,". ,II,· 11/\111,'" 1IIII1IwlIllIailll'alagorics, the perturhation
IIlI"lh,"IIIIIIIIII,' ,,",·,·1 1Ill'lhml.
I 1'11'11'1111 ..\ "liN MI II II III. I his method is useful in certain cases, but
/1. ,11'1"11.11/01/1 lIIay /1111 hl' a!l !ltraightforward as the direct method. The
"U·.I· ... , 1111'0 1I1'1'1t"alillll lie~ in the fact that since both ~k and ()k are
I'"h IIIIIIII./h i/l Ilw rll"lIkiellts of F(~) of equation (3.9), a coefficient of
'I I 1I1.IV hi' \O,,·i,·11 ill ,"ch a way that ~k and 15k become nonzero (if they
\ .1 II 1·.It ... , I",f. In' Ihe variation). To he specific, let 0 •• 02' Q3• ••• ,0" be
It,," 1111'11 ,\,. alld 1\. arc nonconstant polynomials in 00' Now if we
"·1".11" "" h,v tilt we IIbtain the following new polynomial
".= "2=
/..'(.;) .. ~ II"* + + 2 + ... + 0,,2 II (
A. 39)
I III Ih" polynomial ~A. and /)k are nonvanishing. Furthermore, assuming
\, .11111 ,~! arc the first vanishing clements, the signs of o'i and OJ for j < k
\\ III hi' the same for F t (=) as for F(z) provided 100 - a~1 is sufficiently
'.1I1i111.
IIIC' ,Iill'crenee between the zeros of F(z) and F'(z) is also small provided
Ih.11 Illn .-. /I~I is small. Therefore if F(z) has no zeros on the unit circle
I '( ) \\ ill have the same number of zeros inside ·the unit circle as F(=}.
I hI' f"rc1!uing mcthod of perturbing 0u is useful in principle for handling
'.111111' !'.ingular cases. Therefore
has no zeros on the unit circle, consider r(z) = a~ + 0lZ + ...
,,. 1-"(::)
, tI,,~".Tlh.re exists an a~ (ncar ao) such that ~r and do not vanish or
1111" ,...(.~), and the latter has the same number of zeros inside the unit
!"Ill'll' as F(::;).
It should be noted that it suffices to compute ~j and for only j ~ k. h;

Tu determine whether F(z) has any roots on the unit circle or not,
which is important for this application, the following useful theorem
which is stated in Section 3.12 is reintroduced.
necessary condition for F(z) to have a zero on the unit circle is for
1\
to vanish for some I ~ m ~ n and for the next (that is, m + Ist) row
,~ '"
in the table to be all zeros.
Thus when I)k vanishes, we can sometimes ascertain whether F(z) has
:lny roots on the unit circle by computing the next row of the table. The
determination of the exact number of roots on the unit circle is presented
in the next section in a discussion of the direct method. The criterion
for determining whether F(z) has a root on the unit circle when using
UO 1I110ltV AN., AI'I'IUAIIiIN (110 1111' I IMAN!IIOIIM
the dctcl'minunt IIlcthml ili given hy Jury"': A lIl'l'l'~"1I1 Y \'1111111111111 hll

F(z) to have a root on the unit circlc is 111111 1\.,," n.
For a polynomial with real cocllicients. il is indknll',1 111111 1\,,-
(An-l - B,,_1)2F(I)F( -I) and thus thc vanishing of All ill SUIlI,' l·U!.l'1I
could be easily determined from F(I) and F( -I), In gel\t~ral. All vanishl~s
only if F(z) has roots on the unit circle or reciprocal pairs. This is evidenl
from equation (3.70), where we notice that A n- 1 - B .. .1 = 0 when Z,Zk"'"
Iz;12 or Z; = ..!... It should be noted that the computation of A l i i - /J" I
Zk
is rather lengthy in comparison with that for A k • k n. «
If A"_l - B"_1 = 0, we can factor out the polynomial that contains
the roots on the unit circle or the reciprocal pairs by obtaining D(z), the
greatest common divisor of F(z) and F*(z). Sometimes this can be easily
achieved by the division process (Euclidian algorithm) by using the
stability table form as shown later. The remaining polynomial G(z) =
F(z)/ D(z) can then be studied by using the perturbation method if required.
Closely related to the perturbation method is the method used by
Marden 4 to determine the number of zeros inside an ,-circle, that is,
I
,
Izi ::;; -. Consider the following polynomial if F(z) has roots on the unit
circle:
(A.40)
It is evident that G(z) of equation (A.40) has as many roots in the ,-circle
as F(z) has in the unit circle. Therefore, by examining the number of
roots of G(z) in a I + e or in a I - e circle for e smail, we can. in principle,
determine the number of roots of F(z) in and on the unit circle. An
example has been indicated in Section 3.12.
In concluding this section, it might be mentioned that the success of the
perturbation method depends on the choice of a*. In some rare cases it
might happen that one choice of a* would not make A; and Di nonzero;
thus another choice is required.
Therefore this method tells us that the perturbation procedure does
work and nothing further. However, the direct method. which will be
discussed. seems to offer a straightforward (although lengthy) way of
dealing with the singular cases and thus appears useful for application.
DIRECT METHOD. This method is only applicable to the table form and
offers a convenient way for obtaining information on the roots of F(z)
when 61: = O.
As indicated in Appendix I, Table (A.24) represents a sequence of poly-
nomials of descending degree. Any polynomial in this sequence may be
'II ... UII II' I ""''' 1111 II ... III IN 11111 I I NI AII 111'11 III" '0 \ 'oil ,...1 I \I
11'\,111'1'11 ltv ,1111111,,'1 \\llh IIII' •... 1111· IIIIUlIt"1 "I "'1'" 1I1',,,h' Ih,' 111111 l'irl'le
11'111\1111'111111' 1,".1 III IIII' '.1"1111'1111' 1', "111111"111'" hlllll Iht' IIl'W JllllYIIUllli:l1
'1lIlIllhlll'. hI IIII' ',111111" 1I1h-") \\llhlllll dlalll'.ing Ihl' validity "I' the tablc
1111'1111111. 1111' Ind ,'. IIIl" Im~i!> "I' Ihl' dirl'l:1 melhod.
SIII'I''''''' .~" ",,' i, lilt' I'Ir!>1 .\ {with k. = " - m + I} to vanish and that
tl,,· 1"1"'I'dilll~ I'IIW~ ill the tahle :lrc
/'n /"
"", ".11'
/)11 '"1' = b~ - h~1I = 0,

bob, - b .. _I"III' ... , bob",_1 - btb", (A.41)
We notice thut (~" ""'" = 0 iff Ib",1 = Ibol. The polynomial to be asso-
t'iatcd with the first row of Table (A.4I) is
(A.42)
Our objective in the following is to try to replace g(=) by a polynomial

with the same number of 'Zeros inside the unit circle and ·for which none
of the Ilk vanishes. Two cases are distinguished.
I. Some clement in the row of 15"_'"1) does 'not vanish. This is so
itT h", = ±bo but for some q < m/2, bill_II ~ ±bll • We shall let q denote
the smallest integer such that h "'_11 ~ ±bq is satisfied.
For this case the following theorem due to Cohn 2 •4 enables us to
construct a new polynomial having the same number of zeros inside and
on the unit circle as g(=).
THEOREM: If the coefficients of g(z) satisfy this case, g(z) has as many
zeros in and on the unit circle as the polynomial
"'+'1
Gl (:) = BoG(z) - B," l.qG*(z} = I B~"Z1 (A.43)
1-0
where
G(z) = (zq + 2..!?.) g(z) (A.44)
Ibl
b = (b,"_,,, - (bm/bo}b q ) (A.45)
b,,,
k> In (A.46)
and
(A.47)
1.12 IIIIClMV ANI) AI'I'IIC'AIIUN UI' 1m ,·lItANIIIClItM
hiliutiull (A.47) gUlIl'lllItccs tlmt rur Ihe new plll,Ynlllllllll f;.(:) Ihl
replace }:(z)J. ,)" -Inll ducs nut vlInish IIl1d thus we 1.'1111 I.'lIlltlllll~· tht' Inhll'
method to determine the cxact numbcr or zcros insidc the ullit I.'il'l'ic rill
g(z).
EXAMPLE
Let g(z) be given as
g(z) = I + 2z + 3z 2 + ;:3 (AAK)
To obtain the number of roots inside the unit circle, we form the fnll()will~
table:
g(z) 2 3
g*(z) 3 2
(A.49)
This represents the singular case discussed previously. Hence to form

Gl (;:), we first obtain from equations (A.43) to (A.47)
q = \, m = 3, bo = \, b", = b3 = 1 (A.SO)
b2 -(b)/bo )b.3-2
b= =l.lbl=l (A.51)
b) 1
G(z) = (z + 2)g(z) = 2 + Sz + 8z 2 + 5z3 + Z4 (A.52)
From equation (A.52) we have
Bo = 2, B4 = Brn = I (A.S3)
and
GJ(z) = 2G(z) - G*(z) = 3 + Sz + 8z2 + SZ3 (A.S4)
Now we can continue the table form starting from GJ(z).
3 5 8 5
5 8 5 3
G2(z) ~l = -16 -2S -1

Gi(z) -1 -25 -16
G3 (z) ~2 = 255 384

G:(z) 384 255
--------
<0
<53 (A.55)
.'AIII'''\' 'UNII.Uf:IIAIIIIN 111M IINIAII 1tllllRIli 11V!lIIM" I.U
I h, 1111111111.'1 "I 11111'" III ,tt("11II11 fil(z) III!!Idt.! Ih~ 111111 dn:lt.! IIl'e the sume
IIIIIIIIH' ,1~'lel'llllm',1 hy Ih,' II 11111 her ,,\" nC~lIlivc I'". Itere
I', '. . '~I < U,

I h,'n'l'lIl'l' Ihe number of' roots of ~(z) inside the unit circle is two,
1I11t1 111l~ Ihil'" remuining root should be outside, because if it were
1111 Ih,' IIl1il circle. the singular case as discussed in item 2 would have
m·~·urred.
We can readily establish for this example that no root exists on the
IInil circle from the fact that the multiplication of the roots is unity
Ifrom clJuution (A.48)] and g(l)g(-I);e O. Hence we can apply the
perlurbation method for the sake of illustration as follows.
In T~lblc A.49, replace the zero term by a small quantity (positive or
negalive) and then proceed with the table.
F. -1
I -1
(e 2 - 1) -(e - 1)
-(e - 1) (e 2 - I)
In this case if we consider e > 0, we have

h, =e > 0, b2 = (e 2 - I) < 0, 153 = (e2 - 1)2 - (e - 1)2> 0
As established before, two roots are inside the unit circle and one
outside the unit circle. We obtain the same distribution of the roots if
we consider e < O.
In conclusion, we may mention that if the singular case occurs, we
readily know that at least one root exists inside the unit circle because of
the arbitrary choice of the sign of e. Thus finding that g(l) > 0 and
g( - I) > 0, we readily' ascertain that two roots are inside the unit circle.
This simplification can also be obtained for higher-degree polynomials
by checking the sign of the constraints gel) and g( -1) at the start of
the test. S9
2. The entire row of which c5"_"'H is the first element vanishes. This
is so iff b... = =
±bo, b",_l = ±b1••••• bo ±bm or F"_"'+1 == O. For
this case the following theorem due to CohnM gives the form of a new
polynomial that has the same number of zeros inside the unit circle as
F,,_ ...(z).
1.'4 nll'IIMY ANn AI'I'lle'AIIIIN ell' 1111' .·IIlANIII....MM
TIU'.oRI!M: If the cucflicie:llhl ul' /.'" ",(:) !"Itj'll'y tla" 1''''''''llln" 1'1111111111111,
then f'n_m(z) has as many 1-ems in the ullit "Ird,' It!! th~ !,lIlynn",".I.
f,n-m+l(z) -- [F'(n-m)"
(,,)]* -_. (I) _""'
..~'" 11"" 'n -.
.:
_
1-0
I ,,111.1
I· (A ~/)
where
F 'n-m() _ dF n-m(z)
z --=~
dz
Thus for this case, we replace the row (bo• b10 "2' ...• "111) Ily ("l,".
bP) .•• b~:~l)' If (btu, hpj, ... , b~~l) still satisfies the: preccdill~ elise, it
can be shown that F,,_m(z) is of the form (z + ej~)'I."'. If it sntislies the
first singular case, then we proceed in the method outlined carlier.
EXAMPLE
Let
F{z) = -0.5 + 1.65z - 0.8z2 - 0.35z3 - 1.8z. + Z6 11=5
(A.59)
Form the following table:
F(z) -0.5 1.65 -0.8 -0.35 -1.8 I
F*(z) I -1.8 -0.35 -0.8 -1.65 -0.5
F.(z) ~1 = -0.75 0.975 0.75 0.975 -0.75
F~{z) -0.75 0.975 0.75 0.975 -0.75
F'l.(z) <5 2 = 0 0 0 0 0
(A.60)
From ~~ = 0 and F2(z) == 0, it is evident that the second singular case
occurs: therefore we obtain/2(z) from equation (A.57)
h(z) = [F;(z)]* = [0.975 + 1.5z + 2.925z2 - 3z3 ]*
= -3 + 2.925z + 1.5z2 + 0.975z3 (A.6J)
Now we continue the table for the new polynomiaI/2(z)
liz) -3 2.925 1.5 0.975
12*(Z) 0.975 1.5 2.925 -3
13(Z) 152 = 8.06 -10.24 -7.4

13*(Z) -7.4 -10.24 8.06
~(z) 153 = 9.24 -152
I.*(z) -152 9.24
(A.62)
IIII' !,lllvIIIIIIII,,1 1,1 I 11,,'1 IIII' '111111' 111111111,'1 III Ilioh 11I',lIk 1111' 111111
IIII11 11'1 '~I I 1111111 IIII' pll·,,·dlll~l. ,~( ) 1111 .. 1111" 1'11111 (/1'1'11) inside the
111111 1111 k (1"'lIIl1"" I~_ • II. I~:I • II, I~~ " II). Thlls FI(z) Iws also one root
Imltll' Ihl' 111111 1'111'1". A~ 1IIIIIrl'" 1'1"11111 a Ih~orel11 on p. 119 for this form of
',,11/'.11111, nl·,I·. II 1'11111 i~ ~I .. il-Ily inside the unit circle if and only if there
I " ..... II 1I"'ljlllIl'/l1 111'1 hili rool outside the unit circle, and, in fact, the two
1.... 1·. 111" Ihl' n'l'ipl'ocals or caeh other. Therefore, because F1(z) is a
'1IIIIIh link .. pulynolllial, the remaining two roots lie on the unit circle.
11111111'1111111'1'. I'WI1I F(::) we found that 15 1 < 0; therefore the remaining
1,,"1 III 1'(.) lil'S inside Ihe unit circle. In summarizing, F(z) has two roots
111'.11 It· IIII' IIl1it circle, Olle' roof outside the unit circle, and the remaining
'1111 ''',,(.\. arl' Oil the unit circle. The roots of equation (A.59) are i, 2, i,
II /1 I ;O,K.
Allhllll!'.11 Ihe ahove procedure is straightforward, in some cases it
1.. ·1'11111· ... 'I"ite lahorious to calculate all the (jk'S from the table. Therefore
II,' ,'all IIlilil:C the following two facts to simplify the determination of the
111111 di!'.lrihlltioll.
I, "'rolll cquation (A.35c). we can determine the sign of Pit from the
"'t"'" III' F I ( I) and
F1( - I).
,I. 1·'01' rcal polynomials complex zeros appear only in conjugate pairs.
Ba"'l·d on Ihese facts we have for the example discussed
FI(-I) <0
lienee F 1(::) should have only one real root between (-I, I); another
lI'al root (its reciprocal) is outside the unit circle and the remaining two
~holild be on the unit circle. It should be noted that three real roots
rallllot exist between (-1, I), for this requires F I (::) to be at least a sixth-
dl'!~rce polynomial. Therefore, by determining only the sign of FI(I) and
FI <-I) we can obtain the number of reciprocal roots with respect to the
IIl1it circle and roots on the unit circle, thus avoiding the use of the table
I'llr this example. Similar simplifications can also be obtained for other
n:ciprocal polynomials. It should be noted that in some of these singular
cases we can also obtain information on the number of roots on the unit
eircle or the number of reciprocal roots with certain multiplicity.2,31.39
In concluding the discussion of the singular cases, one may mention
that, in general. cases I and 2 could occur in combination for the same
polynomial F(z), In this situation, we may count in the regular way for
the sign of hk (or for the negative PA.) even though some of the interme-
diate polynomials are replaced by different ones, particularly for case 1.
This becomes evident if we would consider the counting using equation
(A. I?).
136 TltllOIlY ANIl APPI.IC'IITmN ft, TH' '·?lANlI'OIM
App"IICII:c 3. SU",,,,tlrJ' 'if 'hI' ",(,M")' (·rl,,.rlll

In this appendix. we summarize the threc form II nf tho IItnhlllly crl,erla
discussed in this chapter. For practical UllC. the lIummarl7.cd It"",. unly
are of importance.
DETERMINANT METHOD. This method, which is discussed in Seetiul1 l.5
and presented in equations (3.28) to (3.35). is presented here in 1\ mnr.
convenient form so that the sign of the constraints for stability ill himllllr
to Hurwitz constraints. This can be readily achieved if the columnli uro
interchanged in equations (3.28) and (3.29) and also a minor change or
sign is made by noting that (_I)k(k+l)f2 = (_I)k I IH'/2( _I)k,
If we denote by X~-1 and Y~-1 the following matrices.
aft 13,,-1 13,,-11 • I •• I as

0 a" 13 11_1 ..... 138
0 0 all ..... at
X~_l = 0 0 0 ..... as
0 0 0 0 Q II
Q,,_B ..... Qs Dl Do
aft_a I ••••• a1 Do 0
Q,,_t ..... Do 0 0
Y~-l =
ao 0 0 0 0 0 0
the stability constraints for n-even and n-odd are presented for the
following polynomial:
f{z) = 130 + a1z + a,.zll + ... + a,;zk + ... + a"z",

I. n-even
F(t) > 0, F(-I) >0
IX';"'1 ± Y~ll = A~ ± B~ > 0, k = ].~5.7., . .• n - I
....A.n "" C'aNIII.IIU IIUN I'nll UNUM UIIIC In II' IIVIIII Mil 137
III Itlll'rnllllvlly rur Ihl" d."rll1ll11""
A~ t ,,~ > n, k - 2.4 •...• n - 2
A~ I - IJ:, I > 0
,. I/-mld
F( 1) > O. F(-I) <0
IX; ± Y;I = A; ± B; > 0, k = 2, 4, 6, ... , n - I
III IIlh·rt1nlivcly for this determinant
k = I, 3, 5, ... , n- 2
A;, I - B~_l >0
I h,' IIhcfnate constraints are useful for design in order not to solve for
Iwn Ctllllltiuns of degree 11 - 1. It should be noted that all the IX; ± Y~I
Itn' nhluined ffom IX~_I ± y;'-ti as indicated in equations (3.30) and
( I. 'I). As an illustration, let us consider the stability constraint for the
1"lIowing polynomial:
Q s >0
I,'Oflll IX;' . I T Y~-.1 as follows:
De TO. "5 T D3 0. T a2 as T a, a. T au
k=3 A; T Ba
TO, a. T 01 T a,
01 a. T 00 "3
k=1 A, T B{
Tal TU,
I ". T "0 I a5
'"
=Fo, '=F°o 0 a. a5
=Fao 0 0 0
". Ai =F B;
F(I) > 0, F(-I) > 0

A; T B~ > 0, A~ T B3 > 0,
TABLE FORM. The table is useful mainly for numerical test of stability
Qr for the root distribution. It is presented in two forms in Tables (3.77)
and (3.121). The latter form is particularly useful for design purposes.
13K 'l'I"'U!tV ANn APPIWA'I'IUN III' nut '·I'UN.I't •• M
We summarile mllillly ill thill ApJlCII'Ux th" furm Itly,," In !'ahll' (Un
Row Zo zJ Z3 • •• .& I Ir ••• ,II' ."

0 0:
a~ as aa ., ,--..an Ir
"n 1 JI ..
2 Qn an - l °n-2 Q!:¥2:~~ D, ""'0
3 bo b1
bll _ 2 bn- s
b2 ba
"tI I
4 bll- bn-t
1
"0
5 Co C1 C2 Ca cna
6 Cn _2 c n- 3 Co
2n- 5 So SJ
rs~"1 Sa
2n -4 S8 32
'.- .... !,
1 SI So
2n -3 To
r- --,
: TJ :
~-- .. I
r2
2n- 2 to
where
, ao a" __ I_ , bo
=, Co
hk =
aft ak Ck =
b._1
b.- I _ k
bk
'
'
dk
C.. _I
C"-.-k' , ...
Ck
ro = I So
S,
s. ,_
So
r, =, So
s.
SI
s. I, to = I ro
r.
r.
ro I
F(l)> 0, (-WF(-I) > 0

bo < 0, Co > 0, do > 0, ... 30 > 0, To > 0, to > 0
(71-1) cnn.trainl.
It is noticed from the table that the dotted entry in row 2n - 3 is

redundant. Therefore the corresponding second-order matrix to be
calculated using the dotted entries in rows 2n - 4 and 2n - 5 is also
redundant. Finally, an easy way to remember the rule to form the table
is indicated by the arrows of the first two rows.
"IAIIIIII\' • IINNIItI'U,IIIN IIIIl IINIAM 11111. 11111 nil II MlC 1.19
III"""IIN MI'IIIIIII, II", ,llvl"""I lIu'lhml i, ,imilu!' III Ihc tllblc form,
hili II 1'1 \\'1'1111'11 III II 1111111 Ihlll l'lIl1 hI.' rcmemhcrcd by using simple
,IIVI'IIIII III' plllylllllllilll", II i, IIIM,'usscd ill Section 3.10 and a summary
1"lIl1wo,.
1'\11'111 Ihe illvcrsc pulynomial F*(z) as
,...(::) = ::" FG) = a" + O,,_lZ + 0,,_2z2 + ... + O,,_,.zk + ...

+ 002", a" > 0
()hlilin the stability constraints (XI: by simple division:
Hz) F (z}
1 2 F {z)
- - · ..~:lto
/:*(::)
+- --
Fi(z)
txo + +-
tx1- - ... -
F:(z)
F"_l(Z}
(xo + OC1 + (7.2 + . , . (X,,_2 + -*--
F,,_l(Z)
whcre 1-'\(::), F2(z}, . .. , F.. _2{z) are the remainder polynomials obtained
l"nllll the division and Ff(z), F!(z), ... , F:_ 2(z) are their inverses.
F(I) > 0, (-I)"F(-I) >0
l(Xltl < I, k = 0, I, 2, ... , n - 2
REFERENCES
I. Jury. E. I., "A Simplified S.ability Criterion for Linear Discrete Systems," Uni-
versity of California, Berkeley, ERL Report Series No. 60, Issue No. 373, June 1961.
Also, IRE Pror(!('cli'Ws, Vol. 50, No.6, June 1962, pp. 1493-1500.
2. Cohn, A., "Ober die Anzahl der Wurzeln einer Algebraischen Gleichung in einem
Kreise," Malh. 2, Vol. 14, August 1922, pp. 110-148.
.\. Fujiwara. M., "Ober die Aigebraischcn Gleichung deren Wurzeln in cinem Kreise
mler in einer Halbebeneliegen," Mall!. Z('ilscilr., Vol. 24, 1926, pp. 160-169.
4. Marden, M., "The Geometry of the Zeros ofa Polynomial in a Complex Variable,"
Amer. Malh. S"r., New York, 1949, pp. 152-157.
5. Samuelson, P. A., "Conditions That the Roots of a Polynomial Be Less Than Unity
in. Absolute Value," Am. Math. Slati.t" Vol. 12, 1941, pp. 360-364.
6. Tschauner, J .. "Die Stabilitat der Impulse Systeme," R('gelungslechnik, Vol. 8,
1960, pp. 42-46.
7. Jury, E. I., "Discussion on the 'Stability of Sampled-Data Systems' by J. Tschauner,"
R~f('III'!lfsle(·hllik. Vol. 8, Munich, Germany, August 1961, pp. 340-342.
8. Gantmacher, F. R., Tire Th('ory of Matrices, Vol. II, Chelsea Publishing Co., New
York, 1959.
9. Tsypkin, Y., Theory of Pulse Sysl('ms, State Press for Physics and Mathematical
Literature, Moscow (in Russian), 1958, pp. 423-427.
10. Brown, B. M.• The Malh('maliral Th('ory of Lillear Sysf('ms, John Wiley and Sons,
New York, 1961, pp. 119-152.
II. Rouche, E., "Memoire Sue la Serie de Lagrange," J. Ecole Polytech., Vol. 22,
1862, pp. 217-218.
12. Jury, E. J., and J. Blanchard, "A Stability Test for Linear Discrete Systems in
Table Form." Proc. IRE, Vol. 44, December 1961, pp. 1947-1948.
13. Romanov, M. I., "Algcbraic Criteria for Aperiodicity of Linear Systems," Soviet
Physirs Duklady, Vol. 4, March-April 1960, pp. 955-961.
14. Jury. E. I., "A Stability Test for Linear Discrete Systems Using a Simple Division,"
Proc. IRE. Vol. 44, December 1961, pp. 1948-1949.
15. Jury, E. I., "Additions to Notes on the Stability Criterion for Line Linear Discrete
Systems," IRE Trans. on AutOn/otic Control, Vol. AC-6, September 1961, pp. 342-343.
16. Jury, E. I., and S. C. Gupta, "On the Stability of Linear Discrete Systems," Rf.lfel-
IIngstet"l",ik, Munich, part 4, No. 10, April 1962, pp. 157-159.
17. Jury, E. I., Sampled-Data Control Systems, John Wiley and Sons, New York, 1958.
18. Jury, E. I., "Proof of a General Relationship Used in the Stability of Test of Linear
Discrete Systems, Addendum to Reference I.
19. Jury, E. I. and Pavlidis, T., "A periodicity Criteria for Linear Discrete Systems,"
IE££ PGCT, Vol. CT-9, No.4, December 1962, pp. 431-434.
20. Jury, E. I., and T. Pavlidis, "Stability and Aperiodicity Constraints for System
DeSign." IEEE PGCT, Vol. CT-IO, No. I. March 1963, pp. 137-141.
21. Wilf, H. S., "A Stability Criterion for Numerical Integration," J. Assoc. Compo
Marh., Vol. 6, 1959, pp. 363-365.
22. Arzhanykh. I. S., "New Stability Inequalitie~," AUlOmatics and Remote Control,
Vol. 22, No.4, April 1961, pp. 436-442.
23. Jury, E. I., "A Stability Chart for Linear Discrete Systems," Proc. I.R.E., Vol. 50,
No. 12, December 1962.
24. Jury, E. I., "On the Roots of a Real Polynomial Inside the Unit Circle and a
Stability Criterion for Linear Discrete Systems," presented at the Second IF A Congress
held August 27-September 3, 1963, Basle, Switzerland, published by the CUt!l{re.l.f
Proct't'ditWs, 1964.
25. Thoma. V. M., "Ein einfaches Verfuhen Zur Stabilitatsprufung Von Lincllrcn
Abtastsysteme," Rt'gt'ltmgstedmik, Part 7, Vol. 10, 1962, pp. 302-306.
26. Jury, E. I.. "Discussion of Reference 25. Regt'lungslt'C'hnik, December 1962.
27. Jury. E. I., "On the Generation of the Stability Constraints in Linear Discrete:
Systems," IEEE PGAC. Vol. AC-8, No.2. April 1963, p. 184.
28. Schur. I., Uber Potenzreihen. die im Innern des Einheitskreiscs beschrankt sind,
JOI//'nal fiir Mathematik, Vol. 147, 1917. pp. 205-232, and also Vol. 148, 1918, pp.
122-145.
29. Chow, T. S., and H. W. Milnes. "Numerical Solution of a Class uf lIypc:,.hnlic-
Parobolic Partial Differential Equations by Boundary Contraction," S.I.A.M. i",,,,IIII,
Vol. 10, No. I. March 1962, pp. 124-148.
30. Jury, E. I .• "On the EVlIluatioll of the: Stllhility De:tcrmi'llIl1ts ill l.ill~·i"· J)i'~''''h'
Systems," IRE PGA\, Vul. AC-7. Nil. 4, July 11)(,2. pp. 'il 'i~.
31. Jury, F.. I., "A Nute Oil the Rcdpl"ll~'ill Z~·nt\ III" a Itl'all'uIY"lllllial \\ilh 1h"I"'l"t It,
the lJnit Circ:le," /1-:/-:/': "(ie;-(·. til hi' pllhli!ihl'lI .111111' I'IM.
32. I'crrull, 0 .• A~~"I"'II, /1",,,/ II, "II,,'''' i,' .I," ."~,:",,, ,111'-/"." ("""'I''''~I:'''', W'III,', III'
<iruyter Ve,·I"I:, IIl'rlin, 1'''1,1'1'' I,' II.
:\3. Knopp, K., Theory of Function, Dover Publications, New York, p. 68.
\4. Fuller, A. T., "Conditions for Aperiodici~ in Linear Systems," British J. App/.
/'/lys., June 19S5, pp. 19S-198•
.l5. Thoma, Von M., "Ober die Wurzelverteilung von Linearen Abtastsystemen,"
H,:It('/uII"fsrechnik, Part 2, Vol. 11, )963, pp. 70-74.
\rl. Li~nard, and Chipart, "Sur la signe de la partie r~lIe des racines d'une equation
1I11:~hrique," J. Math. P""es Appl. (6), Vol. 10, 1914, pp. 291-346.
17. Jury, E, I., and S. C. Gupta, "A General Procedure for Obtaining Tschauner's
SIIIhility Constraints," Regel"'''Kstechnik, November 1962.
lH. Tschauner, J. "A General Formulation of the Stability Constraints, for Sampled-
D,II:1 Control Systems," IEEE Proc. Vol. SI, No.4, April 1963.
1'1, Jury. E. I. Further remarks on the paper of Reference (40), to be published by the
H':~"/II'!ltstechnik No.2, February 1964.
·111 nlUma, M., "Ober die Wurzelverteilung Von LinearenAbtastsystemen," Regelungs-
I,·,'/mik, Vol. II, 1963, Part 2, pp. 70-74.
·11 .I II I'y , E. I., "Proof of the Generation Rule for the Stability Constraints in Linear
D".,·n·le Systems," Technical Memorandum M-39, Electronics Res. Lab., Univ. of
c ,.111 • Uerkeley, California, December S, 1963.
·1.' .Jury, E. I., "A Memo on Marden's Investigations of the Zero-Distribution within
IIII'I hili Circle," Technical Memorandum M-38, Electronics Res. Lab., Univ. of Calif.,
1I,·,~,·lt·Y. California, December 4,1963.
4
CONVOLUTION z-TRANSFORM
The convolution z-transform is an extension of the z-transform from which

a certain theorem called the complex convolution theorem is obtained.
This theorem yields the z-transform of the multiplication of two (or
more) discrete sequences in terms of the z-transform of the individual
sequences.
This important theorem of the z-transform has many applications. such
as in summation of series. in obtaining the z-transform of special functions,
in extending the z-transform method to the solution of unmixed difference-
differential and sum equations, and others which are discussed in detail
in this chapter.
The significance of the convolution z-transform lies in the extension
of the z-transform to nonlinear discrete systems. Certain types of non-
linear difference equations could be solved by this method. Thus the
z-transform method which is usually considered applicable to \incur
discrete systems is extended to cover the nonlinear case too. This study
is presented in detail in the next two chapters.
4.1 Complex convolution theoreml-~
Assume we have the sampled function [f(t)h(t»)· with F·(s) and 11·(.\') liN
the Laplace transform of [·(/) and h·(/) respectively. Tu ubtllin Ihe
Laplace transform of the product we may use the complex convolution formulll
in the a-plane (see Appendix for derivation) 1'5 fullows.
(4.1)
1-17.
CONVOLUTION z-TRANSFORM 143
It should be noted that both F*(s) and
1I*(s) have an infinite number of poles in 1m
the s-plane because of the periodic form x x
of' these functions, s-plane
x x
Assume as an example that F(s) has
a pole in the left half plane at s = sa.. ~[X X
lind lI(s) has a pole in the same plane at T X x
S "': .1",: (both S7. and sri are assumed for
~ake of illustration to be real). then the Re
pille-zero pattern in both the s- and q- x x
plancs of F*(s) and H*(s) is as shown
e~
Sf3
III rigs. 4.1 and 4.2. x x

I he strip of analyticity in the q-plane 2"
T x x
I', ~IIown in crosshatch and the path of
IIlh'/'.ration as a dotted line, It should be sa

lIellt'd that because of the periodicity . FIGURE 4.1 Pole-zero pattern of
III' hoth F*(s) and 1I*(s), the path of F'"(s) and H·(s) in s-plane.
IIlh'/',ralion is taken along one strip and
IIIe' illlcWIlI is divided by the period 21Tj1T as required.
/lulll F*(s) and H*(s) are functions of e' " only; we can replace e T &
II.,. and ("/',/ by p. so that F*(q) becomes a function of p. that is • .fF(p)
1111.1 II +(.~ - q) becomes a function of zip. that is • .:f( (zIp). As required by
IIII', 111:1 ppi ng
X
2..
T
x
Re
X
x
I.
~
111,1110 ,I,' l(c','.1I1II1I1 '1IllIlytlnty III thl", plll1l1' 1111" pllih "I' 1IIII'I:ratillll.
___I_m-r-__ z-plane
FIGURE 4.3 Poles of ~(z) and Jf"'(z) in z-plane.
The mapping of the s-plane and q-plane into the z- and p-planes is shown
in Figs. 4.3 and 4.4. Note that the strip of analyticity in the q-plane maps
into an annulus of convergence in the p-plane. With this substitution
equation (4.1) becomes
9' ([[<t)h(t)] 6T (t)} == 1[[(t)h(t)]
~f P == I
==
217) Jr p-l§"(P).Y(>(=)d
p ,,=0
[flhll'l.-n
(4.2)
where r is the dotted contour in the p-plane as shown in Fig. 4.4.
Equation (4.2) represents the convolution 'I.-transform theorem which
gives the 'I.-transform of the product of two time functions. knowing the
'I.-transform of each of these functions separately, (for 1[[(/)] ~ ~(Z),
Radius ,'"
(contour r)
FUIlIMF ~.4 1'llIh 111'1111"1\1111111" III/'-I,IH"".

CONVOLUTION Z-TRANSFORM 145
J[h(t)]= .;f.,(z». Extension of this theorem to multiplication of more
than two discrete time functions is evident.
Em/ualion of the complex COnl'o/ution inlegra/4 ,.

Hy Cauchy's theorem on residues, we can evaluate residues at the singular-
ities of rl~(p) (if :F(p) has no branch points as assumed here). In
addition if the value of the integral along the infinite circle tends to zero,
we can evaluate residues at the singularities of .;I(z/p) (provided the path
~,r integration is taken in a negative sense). The choice between the two
llIethods depends on the number and complexity of poles of each of the
fUllctions .<F(p) and .Yf (p). Therefore
J[f(t)h(t)] = ~ f p-l:F(p).Tt'(~) dp
2TTJ )1' P
= ~f
Jr p- :Yf'(P):F(!) dp
1
2TT) p
= I residues at singularities of p-l~{p)
= -I residues at singularities of K(;) (4.3)
It J (I') hils simple poles only,
Jf/(t)h(t)] = ± A(~n)
,,~1 p,,8 (PR)
,Yt.(~) + :F(Po)J{'(-=-)/
P.. Po 1>0=0
(4.~)
"h"'I'
I' ' .. ~(p) = ~~:~, and 8(p) = 0,
for P = Po. Ph P2 • ...• P... with Po =0

till' I
/1'(/',,) ::;:: II,! / (4.5)

,III /' /'n
,I..' ('ulllplcx cOllvolution theorem

rllr Ihe.' modified .:::-tmnsform4
'11111,1,11 III l"II"'ilr",, we IlIlIy "hlain the ee.luivalcnt theorem for the
IIIII.ltlh',1 11,111',1""11 hy tlhlnil1i"~ firsl the I.nplllcc tmnsfurm of
lIlt. ", }hlt, ",) J It,,( t). U"';,,,:<I

This is given by
~({f(t, m)h(t, m)] <ST(t)}
T iC+ihtlTl
= -. F*(q, m)H·(s - q, m) dq, o~ m ~ 1
21J') c-/(lIITl
(4.6)t
By using the same mapping as for the z-transform case, we readily
obtain
.1m[f(t)h(t)] =~
21J') Jrr p-1.F(p, m)~ (~,
p
m) dp
=~
21T) Jr
r p-1Jf'(p, m).F(~,
p
m)dP, 0 ~m~ 1 (4.7)
For the case f(/) = h(t), we get

.1m[f(t)2] = ~ rp-1.F(p, m).F(~ , m) dp (4.8)
21T) Jr p
When h(t) = u(t), equation (4.7) readily yields .F(z, m).
EXAMPLE
Find the modified z-transform of the following function:
Im(if(t)] (4.9)
In this case
.1 [t] = mTz - mT + T (4.10)
m (z _ 1)2
and
.1m[f(t)] = .F(z, m) (4.10a)
From equation (4.7),
Im[if(t)] = _1
21Tj Jr
r mTp(p- - mT+
1)2
T !.F(~, m)dP
P p
(4.11)
The integrand in this integral equation has a multiple pole at p 1 =

(and a pole at the origin; however, since its residue is zero in most case!!
=
because .F(z/p, m) jp=o 0, it is ignored). The residue of equation (4.11)
yields
.1m[lf(t)] =.! [mTP - mT+ T .F(~, m)JI
op p P ,,-1
= :F(z. m>rf(m - I») + '/' -~- :F(~- m) 1
. (4.11.)
ai' P 11_1
tit mlly be noted thill/V. til) - /'(t •.• (I - m)1·).

We can easily verify that
i.~(~,m)1 = -z!~(z,m) (4.13)

op P 11_1 iJz
Therefore
Im[I/(I)] = T[(m - l)F(z, m) - z:z F(z, m)]

=§ I(t, m) wbere II (t) = tl(t) (4.14)
Similarly,
Imrt~(t)] = T[(m - 1)F (2:, m) - z:z :F (2:, m)]

1 1 (4.15)
lIy iterative extension of the preceding we can obtain

;1 ".[1'1(/)], where k is a positive integer
4.3 Applications of the convolution

modified or z-transform method 4 ,2
III Ihb section we shall introduce a few applications which are useful in
1II',ade system analysis as well as design.
Im'use Zm - !P trallsformatioll
III n l)J"l~ecding
discussion* we obtained the modified z-transform from
Iht, I.lIplace transform without indicating the inverse process. Now we
',hullllhiain /'1s), knowing ff(z, m)
,..(.~) = Z;,lF(z, m) (4.16)
IIII' 1.lIplace transform of /(1) can be written as
,..(.~) = L ob
o
/(t)e- SI dt = ! f'''+1IT
00
,,~OliT
/(t)e- SI dt (4.17)
Nm\ II Wt'lcl, = (1/ - I + m)T, 0 ;S; m :::;; I, II = integer, equation (4.17)

1.111 III' wl"illcn
n,v) "'" ."

'J' }; I' .,,'1'. 11 /(nT - T + mT)e l1- m )$T dm (4.18)
" II 0
III
(4.19)
• " •.•. 'H'.II.1I1 I \,

However. by definition
eo
~ l(nT - T + mT)z-n = ~(z. m) = .1m(j(/)] (4.20)
n=O
f:
Therefore
F(s) = Z-".' ~ (t. m) = T Z-I+", sr(%. m)dm It = en (4.21)
It should be noted that this transformation is unique. that is. for each
.~·(z.m) there exists only one form of F(s).
Im'erse .~(z) - F(s) or (z-s) translormation2,4

This transformation indicates how to obtain F(s) = 2[[(1)] when we
know F*(s) = ..P[J*(t)J. To obtain the z-s transformation, we first find
l(nT). This can be achieved by recognizing that F*(q) is a periodic
function of q with the imaginary period 21TjIT; hence the value l(nT)
is the coefficient of e 1[11'1' in the Fourier series representation Z of F*(q).
Hence,
T fC+J(or!TI
l(nT) = -. F*(q)e qnT dq, for/(O) = 0 (4.22)
21TJ c-i(~I7'1
where the path of integration lies to the right of the poles of F*(q).
In order to obtain a certain I(t) whose sampled values coincide with
l(nT), we change nT into t in equation (4.22) and then apply the Laplace
transform to obtain, after interchanging the order of integration,
F(s) = I-. [CHlatTI F*(q)[ [eo e-(··-qlt dtJ dq,

21TJ JC"J(/rITI Jo
for 1(0) =0 (4.2.\)
We note thatioor(S-qll dt = I/(s - -q); hence equation (4.23) reduces 10
T fC+iCorlTI F*(q)
F(s) = -. - - dq,
for/(O) = 0
21TJ c-j(IIITI S - q
It should be noted that the right-hand side of equation (4.24) relm'sellts
the Laplace transform of f*(/)u(t). Helice pes) can he also ohlairwd
directly from ,27[f*(t)u(t)] using the complex convolution theorem disl,'\I!\st'd
in this chapter.
If F*(s) is written in terms of z = (,'I'M, the function /0'.(.\.) ht'l'OIllI'~
.:F(z), the z-transform of F(s). Thus equatioll (4.24) is considt'J't'd liS lit,·
inverse z-s transformation, Ihal is,
It may be noted that this inverse transformation is not unique, that is,
many forms of F(s) have the same Z-transform F(z). This is evident
since f(t) is forced to coincide in values with f* (I) only at sampling
instants.
Skip sampling theorem 1 ,17,18
This theorem signifies that, knowing the z-transform with respect to the
pcriod TIn (when 1/ is an integer is larger than unity), we can obtain from
it the z-transform with respect to the period T. This theorem is important
ill obtaining the limit cycle (or periodic modes of oscillation) of nonlinear
discrete systems and in the study of multirate sampled-data systems.
The fundamental equation that describes this theorem is
f(kT) = f(k:T) u(kT), where k' = nk (4.26)
Applying the z-transform to this equation, and using the complex convolution
flll'lllula, we get
,:F(z) ~ J[f(kT)] = _1 f .?'(Pn)p;;l 1 dp (4.27)
2'TTj Jr 1 - (zn/Pn)-n
wlll'n~ ;:" = e(T{II)s = ZI/".
WI' ~':111 evaluate this integral in a positive sense for the singularities of
. Ii (1' .. )1'" " to obtain one form, or, alternatively, integrate in a negative
'.1'1\'0(' with respect to the poles of ( 1 I) to obtain
1- Pn z" -n
(4.28)
1 IIllhl'l' application of the preceding theorem is in the analysis of

. 1I1111111111tl'd discrete signals appearing in some systems,
'"tl'gmtli",/' fUI/Nioll 'I'onsjorm 4•1l
J ... I.Cf(t)II(t) till

. . T . (' f p- 1§(p, m)£(=-, m)d Pdm
~/~/(:: - I) Jo J1' P
I , '/' J'"'J' . I' I:F(p, m).Jf" (:.,p )

h~1 u I
m (Ip (1m (4.29)
This relationship can be obtained from writing the integral in a sum-
mation form.
l o'f(t)h(t)dt = TLi aO
i1f(k - 1
0
+ m)Th(k - 1 + m)Tdm
+i m
fen - 1 + m)T hen - 1 + m)T dm ]
(4.30)
The modified z-transform of the preceding, noting that when m = 0,
JJ"ff(k - I)T] = _1-1 ~(z), yields
~-o z-
1m [L'f(t)h(t) dtJ
=Z~1 f Jl(n - 1 + m)T hen - 1 + m)Tz- n dm
+ Tim n~/(n - 1 + m)T hen - 1 + m)Tz- n dm (4.31)
but the infinite summations are the modified z-transform of f(t)h(t), thus
1m [ltf(t)h(t) dtJ =
o
~
i1§"1(Z, m) dm
z-lo
+ Tim~l(Z, m) dm
0
(4.32)
where
§"l(Z, m} = 1m[f(t)h(I}} = ~ f p-1F(p, m)Jf'(-= , m) dp

21TJ Jr. P
(4.33)
For the special case when her) = u(t) = unit step, equation (4.33) yields
1m [itf(t) dtJ
o
=~ ll§"(z, m) dm + Tlm~(Z' m) dm
z-lo 0
(4.34)
where
§"(z, m) = 1m[/(t») (4.35)
Equation (4.34) also yields the modified z-transform of £(s)/s. Extension
to obtain Zm[£(s)/sk) by iterative process is straightforward.
z- Tran.iform of special fUlIcticm.v4,9,lO
One of the several applications of the ,,~onvoluti()11 z-trnllsfnrm is to cllublc

us to tlbtnin the z-trnnsforrn (If spcdnl function" such ns .1V'U)/t"). k > n.
This can be achieved by obtaining first the z-transform of (III) for t > O.
The z-transform of I/t for t > 0 can be easily obtained by using the
definition of the z-transform as follows:
<X>
~(z) 1%_1> = L f(nT)p-n (4.36)

n-O
Multiplying equation (4.36) by;-1 and integrating both sides from z to
co, we obtain
! f.<O rl~(p) dp + lim f(t) = 1 [/(t)J (4.36a)

T • 1-0 t t
If we let f(t) be a unit step for t > 0 whose ~(p) = l/(p - I), we have
t
[1J 1
1- =- f."" p
T z
-1 - dp-
p - 1
for t >0 (4.36b)
Choosing T = I, we have
1[!J t
= In-z
z- 1
t > O. (4.37)
For any f(t) we have, applying formula (4.2),
l[f(t)J = ~ f p-l~(p) In _z_ dp + lim fn (4.38)*

I 21TJ Jl' Z - P n-O n
The function In z/(z - p) has a branch cut in the p-plane extending from
I' = z to p = co. Since we are free to take the branch cut anywhere in
the p-plane, we shall take it along the positive real axis. This is illustrated
in Fig. 4.5. As before, the crosshatched area shows annulus of convergence
and we have
LI= -fEA
1.:", + J.w + fIleD + fm; = 0
-i.:. . = f.w + fV1-: + fBeI>

1.1 = f.,,/ LI'; + fReD
• III Ihl' 1'lIl1l1willl: It~rivali\ln we shall show the equivalence between this equation
hn~I'a1 lin 1'lIlIvllhllinll .··trallsrorm and equation (4.300) obtained directly from the
, 1...11.,1'111111.
/ ' - - - --........ p-plane
/ Poles of 1m "
_; p-'F'(p) '" "\
~
~___ PJJranch cut from
C , p=ztop_oo
~~~~ E -~-};7
1,0 7
/
'- _/
FIGURE 4.5 Branch cut in p-plane and path of integration.
Along AB, P = Ipl ei .,

Along DE, p = Ipl ei(21r-·>,
In _z_ = In (_~) + In _.;;,.1_ (4.39)

Z - P P 1 - zIp
Expansion for In I l L is given by the following:

-zlfJ
In
I-zip
1 = -Z
P
+ -2pB
z2 + . .. (I I )
-zp < 1 (4.40)
Therefore
In _z_ = In (_~) + ~ + L + ... (4.41)
z- p p P 2p2
Now r
JBCD
= O. Hence
(4.42)
CONVOLUTION z·TRANSfORM 153
Now In k (where k is a complex number) = In \k\ + arg k. It is then
easy to see that equation 4.42 reduces to
(4.43)
We can generalize this result for higher powers of t i'n the denominator.
~ b z-n =frr> p-l§(p) dp.

,,~1 n •
multiplying by Z-l and integrating from z to 00,
f ~n
n~1
z-n =f."'q-Il'" p-I§(p) dp dq
• q
(4.44)
Continuing this, we obtain
~ V~)} = J.rr> w-1L" v-11<O u-1 ••• i'" p-l§'(p) dp ... du dv dw

. .Himes· (4.45)
For k > I this equation is defined for t > 0; for k = 1, equation (4.38)
is valid.
Although the p'receding results appear to yield ~[f(t)ltk] for k > I in
II closed form, actually it is not so because of difficulty in evaluating
intcgrals in a closed form of the type S[ln (a + bx)jx]. Therefore it is
not possible to obtain closed expressions for z·transforms of functions
Iikef(t)/tk for k > 1. However, the usefulness of the result lies in evaluat·
'ing infinite series since numerical evaluations of definite integrals of the
preceding type are available. This is discussed in the next section.
To obtain the modified z·transform of [j(t)!t] for limf(t) = 0, we use
1-·0
unother method which avoids the complex integration, based on the
following derivation.
Let :F(z, m) be presented (notingf(nT)ln=o = 0) as follows:
tr)
:F(z, m)/._" = ~ f(nT - T + mT)p-li

"
.,. II
= ~ f(nT - T + mT)p-n (4.46)

11- I
Where we multiply this equation by p-m and integrate from Z to infinity,

roo p-m!F(p, m) dp _ roo I f(nT - T + mT)p-< ..+ml dp
J& J. n~l
_ ~1(nT- T+ mT)z-<n+m-lI
.. -I -n - m +1
0< m < 1
or
-1 zm-If.oop-mdr( )d ~ 1(nT - T + mT)z-"
or p, m p - ~ (4.47)
T. (nT- T+ mT) ..-1
If ~~n ~ / : ;>; z-O is added to both sides of equation (4.47), this term
has a contribution only at m = I, which is the lim f('); thus the right
rf(')J ,-0 t
side of the above equation is I m L-t- .
Im[f(I)J = ! zm-lf.oo p-m!F(p, m) dp + limf(t) ,
t T • ''''0 t
o < m ~ 1 (4.48)
For the case of z-transform we obtain, by substituting m = 1 in equation
(4.48)
l[l(t)J - ! foo p-1!F(p) dp + Iim1(')
IT. '-0 t
where
!F(z) = !F(z, m)lm_l (4.49)
Extension of equation (4.48) to obtain Im[f(t)/lk] is evident.
Special cases that stem from this theorem are indicated in the following.
1[ let)
t
] _1 i'f.oo p-l1+A)!F(p) dp
+ hT T &
(4.50)
where h is a positive constant larger than zero. Equation (4.50) is obtained

from equation (4.36) by multiplying both sides by p-(1+A) and integrating
from z to 00.
If we assume in equation (4.50), T - l,f(l) = unit step and if we denote
h as a (integer),
J
1 [ -1- = I00 -1- z-" - t"f.oo p-(1+a) -p- dp
n+a ..aO n + a • p- 1
_ zaJ.Z[.!. + _1_ + ... +! __J -J dp
00 pa p p- J
pa-l
= ZCl
Z
{In - - fZ 1"'-1) Z.(..
- - ---- +
II
-0.' - + ... + z.-..Il}
:-1 a-I a-2 I.
(4.51)
or
.1 [ --
1 ]
n+a
=Z4
[ In---2
Z 4-1 1
-z-1I
z - 1 1'~1 P
] (4.52)
EXAMPLE
Obtain the following z-transform:
~{(n + a;(n + b)J

From equation (4.52) we can write
"[(n + a;(n + b)J = "[b ~ aC~ a- n~ b) ]

= _ 1 [(In _z)(z~ _ Zb)
b-a z-J
4-1
,,~ -b
1 + b-l)
b ~ -/J
]
-z~-z z~-z (4.53)
1·=1 P 11=1 P
If we let a = I, b = 2,
.1[(n + 1)(n1 + 2)] = Z[1 + (1 - z) In _ z J (4.54)

z- 1
Similarly, we can develop the z-transform of
+ a)(n + Ih)(n + c) .. .J
(4.55)
"[(n
Summation of infinite and finite series 5- 7 ,9,10
The convolution z-transform readily yields the summation of series, where

Ihe sum exists. These summation forms are discussed as follows:
. I f we let z = 1 in equation (4.2), we readily obtain
i
" n
fllhn = ~ J.:~(P) ,;Yf(p-l) dp
27f} I P
(4.56)
where the contour r could be taken in a positive sense along a circle

co
whit'h encloses the poles of ,~(p)lp. If the summation I fnh" exists, then
n ·.0
Ihi .. l'011101lr could be along the unit circle in the p-plane. As a special
l'IIS,' or "'I"aliol1 (4.56). when II" = /". we get
.... ~ 1 ~ ::&"(p) "L"( I) I
-'\ 1'" '." ··.r I' I P (4.57)
" II 'J.rr', I' I'
where the contour r in this case is along the unit circle where the poles
of ~(p)lp are enclosed. A method for obtaining equation (4.57) without
having to obtain the poles of fF(P)/p is given in Table III of the Appendix
by letting m = 1 in I". The proof is presented in Appendix 2 of this
chapter.
Another important summation could be obtained by letting Z = I in
equation (4.45), which gives
I l; = lim f.oo w- 1oo 1 V-I • •• Coo p-1fF(p) dp . .. du dw (4.58)

n-on &~l ,J III Jq
k.th;;ea
00
To obtain the summation I n,zn, z < I, which often appears in the

,,~O
analysis of discrete filters, we can also apply the convolution z-transform as

follows:
I n'z" = _1
01=0 2.".j
I. (p - p z)p .Jt"(p-l) dp
I'
(4.59)
where r encloses the poles of l/(p - z) in a positive sense, and
(4.60)
The list of Table IV in the Appendix presents this summation for
values of r up to 10. As a special case of equation (4.59), we can also
obtain the finite sum of the following serieso,lO
,,-1
S, =I j'zi, r = 0, 1,2, ... (4.61)
j-O
for which
dS,_l
S,=z--, z"-I
with So = - - - (4.62)
dz z-1
k
Finite summation of the form If" can also be obtained from equation
(4.56) by letting ,,-0
h(t)I,=n = u(t)lt.n - U[I - (k + 1)]1,=" (4.63)

Therefore equation (4.56) reduces for this case to
I J, = _1 C!F(p) p-l [1 _ pieHl dp

,,-0" 2.".j JI' P p-l - 1
=-
1
2.".j
I.
I'
,~(p)(pIcH -
--
P P- 1
1)
dp (4.64)
where the contour I' cnclllsl.'s the poles (If the illtc~rnnd,
EXAMPLES
Obtain the following summation:
(1)
Using equation (4.58) for this particular case, we have

ao
! -2 I
=
iaorlln~dp
,,_tn 1 p- 1
With the change of the variables of integration from p-l = x and using
tables of integral, we finally obtain
.i n12 = Jofix-lin (_1_)

.. zl I - x
dx = "II
6
Summation of the form

obtained. n-I n
I (lk) for even values of k > 2 can be similarly
(2)
This summation can be obtained from equation (4.59) as follows:

GO
!nx"
n-O
= -1
21Tj
! I"
P P4
(p - x)p (p-l _ 1)2
dp
where r is the contour in the p-plane which encloses the pole of I/(p - x).
Evaluating this integral, we get
00 :t
!nxt'l = ,x< 1
II~O (1 - x)'I.
I<
.(3) In
.... 0
From equation (4.64), we have
I<
In
.. -0
= -
It
2"j r (p - 1)2(p - 1)
pl<+1_1
dp
This contour integral yields
~ n = .!.. ~ [ k+l _ 1]/ = k(k 2+ 1)

L
,,·0 2'. d pII P p_l
(4) l:'''. . __(-1)"

......... _-
,,·-0 (II + 1)(" + 2)
This summation can be obtained from equation (4.54) by letting %==

-I, which gives
i
n-II (n
(-I)"
+ I)(n + 2)
==2In2-1
«)
Finally, summations of the form! fnhn+m, 0 ~ m ~ 1 can be similarly

n=O
obtained from the convolution modified z-transform. Furthermore, from
the integrand form we can ascertain the conditions on convergence of the
series or the possibility of obtaining a closed form.
Unmixed difference-differential and sum l'quations1 ,B

The convolution %-transform is readily applicable to the solution of
equations of the following form:
.11 K n
I !ocremY~~m(t) +re=o
m-O re .. o
!fJreYre(t) == x..(t) (4.65)
Assuming zero initial conditions, we proceed to solve this equation

by applying first the Laplace transform and then the %-transform to obtain
M K
~OTc~OOCremsre.1[Yn+m(S)] + 1" Ln~!reY1o(S)] == .1[X ..(s)] (4.66)
where ~[y~Tc~."I(/)] == sre Yn+m(s).

Using the convolution %-transform and noting that
.1 f fre] == ~I §"(%),
rl..t=o %-
we obtain for
.1[IfJ/cY/c(S)] == (_%_)~ r 8{p) q1j(~, s) dp (4.67)

10-0 %- I 2."., Jr p p
where B{%) == .1[B.. ] and qv(%, s) == .1[Yn(s)] as defined in equation (2.120a).
Using equation (4.67) in obtaining the %-transform of equation (4.66) we
get
qv(%, s) I Iockmbm+ _%_~
m-OTc-O %-
r B(p) qv(~,
1 2."., Jr p p
s) dp
= .f."(%, s) (4.68)
This linear integral equation cannot he sulved in terms of elclllcnlluy
functions, but it cnn he solved hy Il l'ul1Vl'rgcl1t ilerntivc process.
If we examine the special case where {J,. = {J, when {J is a constant,
equation (4.68) reduces to
M K (J
tpj(z, s)! !OCkn/'z'" + _z_ </9'(z, s) = ~'(z. s) (4.69)
mcO/c~O Z - 1
This can be written
~'(z, s) .f'(z, s)
tpj{z, s) = (4.70)
~ £..
~ OC,.",s/czm
£..
+ -z{J- = £(z, s)
mmO/c=O Z - 1
To obtain y.,(t), it is necessary to know the zeros of the denominator.
Knowing these zeros, we can apply the inverse Laplace transform and
then the inverse z-transform as follows:
(4.71)
or
(4.72)
El'aluQtion oj illtegrals oj time Juncliolls'~,'·11.22

We shall use the convolution modified z-transform to evaluate the following
integral:
LO)J(t)h(t) dt (4.73)
We obtain this integral from equation (4.29) by using the final value
theorem for the modified z-transform
io O)J(I)h(t) dt = I... Jo(1 J.r §(P.p m) .YI'(p-l, m) dp dm

2?r)
(4.74)
The contour r can be taken in a positive sense which encloses the poles
,if :F(p. m)/p or in a negative sense that encloses the poles of .%(,-1, m)
ill the p-plane.
As special cases of the preceding, let h(t) = J(I); then
s.
"
trJ
[J(tW dt = -T.
2fT)
lIt
0 I'
p-l.:F(p, m)§'{p-l, m) dp dm (4.75)
whl'rc the contour I' is taken along the unit circle since all the poles of
:,.. (1',111)11' ure inside the unit circle if the integral is finite. It should be
IIlIh'd thut in the evaluation of the integral
I" _ I
1rrj
J' I' 1:IJ(p, m):F(/'
I'
I, m)tlp (4.76)
we do not need to fin8 the poles of ~(p, m) for the integral can be
calculated as the ratio of two determinants in the coefficients of the
numerator and denominator of ~(p, m). The evaluation of this integral
is presented in Table 1Il and calculated up to values n = 4.
It should be noted that the evaluation of the integral in equation (4.75)
is simplified if the order of integration in p and m is interchanged. This is
possible since the integrand is continuous along r and in the interval
O~m~l.
Furthermore, if we let h(/) = tf(t)
Lo'" t[f(t)]2 dt = T2.
2~)
(1 ( p-l[<m - l)§(p. m)
JoJr
- p :p .fF(p, m)Jl.~(p-\ m)] dp dm (4.77)
Similar relations can be obtained for higher-order moments.

In conclusion, by utilizing equation (1.178) of the modified z-transform
which states
Tf~(Z' m).)t"(z-l, m) dm = Z[F(s)H( -5)].
we can rewrite equation (4.74) in the following simplified form:
("'f(t)h(t) dt = _1. ( ZlF(s), H( -s)]I%~" p-1 dp (4.78)

Jo 2~) JI'
Similar relationship could be obtained for evaluating finite integrals
of the following form:
(II ('I
Jo fl(t) dt =Jo f(t)h(t) dt (4.79)
where 11 is finite integer.

The evaluation of the integral of the square of the time function is
very important in the design of sampled-data feedback system that is
based on the minimization of the total square value of the error function.
It is also important in the statistical design of a system that is based on
the minimization of the mean square error function. The entries of Table
III are important in obtaining the noise power gain in a discrete filter.
Two-sided z-transform 12.1~.16.17.18

In the preceding chapters, the z-transform of f(II) is defined fur positive
values of n, that is. II ~ n, which we may refer tn as the olle-sided z-
transform. III sOllie npplil'lltinns stich liS in tilt' stlitislit:1I1 study of discrete
1m
z-plane
Re
Ring of
convergence
of 6(z)
FIGURE 4.6 The contour C and the ring of convergence of .:F(z).
systems, the statistical functions may also be described for negative n,

thus the z-transform definition should be modified to cover such cases.
The two-sided z-transform is defined as follows:
'Ie""
~(z) = J[J(n)] == I /(n)z-lI (4.80)
where .~(::) is analytic in the ring shown in Fig. 4.6.

The inverse of ".(:) is unique and is given by
f(n) = .rl[§'(Z)] == ~
2fT)
r
Jc
§'(Z)Z"-l dz (4.81)
where thc closed contour C lies in the ring of convergence of 9"(z) with
the point:: = 0 in its interior. The definition of the two-sided z-transform
requires that ~(:) converge in some ring (of nonzero area) about z = 0
in the complex plane. The contour C is shown in Fig. 4.6.
If the functionf(lI) is bounded, the contour r of Fig 4.7 enclosing the
unit circle will belong to the ring of convergence. Here we write
/(11) = rl[.~(z)] = _1
2fTj
r§(Z)Z..-l dz,
Jr
(- 00 < n < 00)
(4.82)
1m
---r----~----~--x~
A pole 01
[J'(z-l,
I It il IIU .\ 7 I h., fllllhllil I' ill ,hl' • plmll"

16~ THEORY AND APPLICATION OF THE z-TRANSFORM
If the only singularities of .F(z) are poles of finite order, the inverse
can be evaluated as usual by the residue method, that is,
fen) = I residue .F(Z)Z"-l, (-00 < n < (0) (4.83) ,..,
nil poles
Inside unit
circle
We may note that we could also evaluate fen) by considering the poles
outside the unit circle; then we will have
fen) =- I residue .F(z)z..-t, (-00 < n < (0)
all poles
uutsidethe
unit circle (4.84) "
Although both equations (4.83) and (4.84) are valid for all n for
computational purposes, equation (4.83) is preferable for n ~ 0, and
equation (4.84) is preferable for n ~ O.
Since, in statistical study of signals in discrete systems, both boundedness
and symmetry around the time origin are assumed, we could then write
o 00
.F(z) = I f(n)z-" +I f(n)z-" - f(O) (4.85)

n--oo n=O
If we let m = -II in the first term of the right side and impose the
condition fen) = f( -n), we will have
~(z) = ~l(Z-l) + ~l(Z) - f(O) (4.86)
where §l(Z) has all its poles inside the unit circle and it is defined for
positive II, and ~l(Z-l) has all its poles outside the unit circle and is
defined for negative n.
The inverse of equation (4.86) gives
fen) = ,rl[§l(Z-l)] + ,rl[FJ(z)] - f(O){)"o (4.87)
where 6" o = I for n = 0
= 0 otherwise
We may note that here we might use equation (4.83) for n ~ 0, sub-
stituting only ~l(z) for $I"(z). Similarly, we may use equation (4.84) for
n ~ 0, substituting only F 1(z-J) for F(z).
The general case where fen) is not symmetrical and boundedness
condition does not exist can also be treated, However, in this situation
care should be exercised in defining the ring of convergence of F(z). If
such a ring is nonexistent the z-transform does not exist.
For instance, if a pole exists inside the unit circle, then this indicates a
decreasing exponential for positive n, or 1I I1OI17.em value for the timc
function for ne~Hlivt~ II. The decision liS to which time function is rep-
resented hy the .:-lrunsfol'lII (It'pends Oil Iht, riny, of ctlllvcrgl'nl'e ill Ihc
z-plane in which ~(z) is defined and on the manner in which the ring
divides the poles of .9"(z) within or outside this contour.
Let !F(z) be given as follows:
!F(z) = -z(1 - q2) , with q <1

q(z _ q-l)(Z - q}
To obtainf(n) for n ~ 0, we use equation (4.83) for roots inside the unit
circle. In this case, we obtain
fen) = residue of [~(z)z'a-J] Iz ~'1 = q", n~O
For n ~ 0, we obtain from equation (4.84) the following:

fen) = q-n, n~O
The total function fen) is represented by

len) = qft, II ~ 1; n s- 1
and
f(O) = I
Applications to correlation functions and spectral analysis l6 •17

One of the applications of the two-sided z-transform is in obtaining the
power spectrum density of the discrete random signal. This discrete
power spectrum is defined as the two-sided z-transform of the auto-
correlation function of the signal. For instance, if the signal is represented
as f<k), its correlation function is defined as
1 N
cf,(n) = lim I f(k)f(k - 11), (4.88)
.v-«> 2N + 1 /c",-lV
where n is an integer.
The power-spectrum density of the random signal is defined as
n=co
<I)(z) ~ I cf,(n)z-n ~ 1[cf,(n)] (4.89)
Since 1,(11) is an even function of time and bounded, it can be obtained

from equation (4.89) as
.p(II) = J 11'11(::)1 = _1- r

2nj ) •.
(I)(z)z"-I clz (4.90)
wh,-n' I' i~ th,- huundllr'y "f till! unit d ..de in the z-plam!.
It is seen from equation (4.88) that when n = 0, we get

I Ic-N _
+(0) = lim ~ {f(k)P = r
N-- 2N + 1 k--N
~ mean square value of the signal (4,91)

Substituting for n = 0 in (4.90) and using (4.91), we obtain:
- = _1
/2
21Tj ('
Is
(Z)Z-1 dz (4.92)
Since <1>(%) is a symmetric function with respect to z, it can be written as

(4.93)
If we substitute (4.93) in (4.92), we can evaluate f2 readily by using the
tables of total square integrals which we discussed earlier.
/2 = I
n
= -' r $(Z)jF(Z-l)Z-l dz.
21Tj )1'
(4.94)
Furthermore, if the stationary signal Ik having the power spectrum

(z) is sent through a linear, time-invariant system with discrete stable
transfer function ~(z), the output mean square value of the signal is
given as
(4.95)
where
EXAMPLE
1. We shall calculate the mean square value of the output of a discrete
system shown in Fig. 4.8.
Given the power spectrum density of the input
<1>,(.;:) = N2
where N is a constant (this represents white noise). To obtain the system
transfer function (,,9(z) from Fig. 4.8, we use
_ _c_' [1 + (blc)Z]
~9(z) =.~o(z) = z- I %- I
.~,(z) 1+ _"_[I + (b/C)%]
z- 1 z- I
= ::(1) + c:) -- C' (4.9(,)
z:& ... (II ,. c' -- 2): I- (I ... c')
Input+ Output
fin fon
FIGURE 4.8 A feedback discrete system.
The mean square value of the output is given as
f~n = ~ f !J(Z)!J(Z-l)<l>j(Z) z-Idz

27T) Jr
N2 f z(b + e) - e
=:"27Tj Jr Z2 + (b + e - 2)z + (1 - e)
z-I(b+ c)- c
x z-Idz (4.97)
Z-2 + (b + e - 2) Z-I + (I - c)
This integral can be easily evaluated by using Table III for the second-
order case and noting the following constants:
bo = 0, bi = b + e, (4.98)
ao = I, a1 = (b + e - 2),
Substituting these values in the 12 formula of the second-order case to

obtain finally
j2 = N 2 [c(1 + b/c2 ) + b/2l (4.99)
on 2 - e - (bt2)
When b = 0, this reduces to

- eN?
1on
2 -
- 2-- (4.100)
. -e
Extension of two-sided z-transJorm to continuous outputI2.18.2S

III Ihis application the two-sided z-transform is applied to the continuous
c:IIl'rclation function as follows:
x .f.,(k + III .- I)/,,(k - '1 + m - I), o~ In ~ 1 (4.101)

If we denote the coefficient of z-n as tPo(n, m), the continuous spectral

density is
n-«>
~o(z, m) = ! z-ntPo(n, m) (4.102)
nM.-CO
To obtain the mean square value of the outputf~(/), we use the following
form:
-
f~ = lim
1
!N 11f!(k + m - 1) dm, o~ m ~ 1
.v~«> 2N + 1 k~-.V 0
(4.103)
This equation can be obtained from tPo(n, m) by Jetting n = 0 as follows:
f~ = f tPo(O, m) dm (4.104)
On the other hand, tPo(n, m) is obtained from ~o(z, m) as inverse z-

transform relationship, that is,
<po(n, m) = _1-
27Tj
r<l>o(z, m)zn-l dz
Jr'
(4.105)
where r is taken along the unit circle for stable tPo(n, m) as is usually
the case.
Therefore
tPo(O, m) = _I.
27TJ
r<l>o(z, m)z-l dz
Jr
(4.106)
Inserting equation (4.106) in (4.104), we finally obtain
f: = _1_
27Tj
r
J,' Jo
r\f)o(z, m)z-l dm dz = _1_
27Tj
r<l>o(z, m)z-J dz
Jr
(4.107)
Evaluation of equation (4.107) can be readily obtained by using the tables
of integrals which were derived in Table Ill. It should be noted that
~o(z. m) can be always written in the form
(4.108)
which is readily amenable to table form.

Finally, we may note that the mean square value of the ouput for a
discrete system could also be obtained from the Laplace transform
theory. An illustrative exam pic of optimization and filtcring using Laplace
and z-tmnsform methods is dis~'ussed in d~~tllil in Section K.J.
Appendix I.. Proof of complex COnt'olulion formulatll.14.20
Suppose F*(s) and H*(s) are the Laplace transforms of sampled function
f*(t) and h*(t), respectively, defined by
pes)
1
= 2'[f*(t)] = -.
27T}
1 00
Qo-;W
+1<0
F(q}
I - e
I
-TI.-o) dq + If(O)
(A.I)
with Re rs1 > qo > max [real part of poles of F(s)], and F(s) = 2'[/(/)],
To find It'{[f(t)h(t)] . !ST(t)}, we use the following definition from (A.I):
It'{[f(t)lr(t)] . bT(t)} = -I.

1- e
27T}
I i
-TIs-w)
wo +JW
dw
100-1«>
2'[f(t)h(t)]
(A.2}t
From the convolution theorem of Laplace transform,3 equation (A.2) can
be written as
It'{[f(t)h(t)] . bT(t)} = -1.

27T}
f. WO
+ IW
100-;'"
X [_1_. ~T($-W) dw (A.3)

iC+.J"; F(q)H(w - q) ci q]
27T} C-}a:;I - e
In equation (A.3) we interchange the integration with respect to wand q.
Therefore
I
= -.
27T}
1
Z{[f(t)Ir(t)] . DT(t)}
c+;«>
c-;«>
1
F(q) dq - .
27T}
J.WO+i«>.
100-;'"
H(w - q)
I - It
1
-Tis-Io) dw
(A.4)
Noting equation (A. I), we obtain for the right side of equation (A.4)
-
I f<+I 00
F(q)H*(s - q) dq (A.S)
27Tj '-17>
It should be noted that H*(s - q) is a function of eT(.-q) only, In

\'\luation (A.5) we divide the range of integration into intervals of length
"'11 = 27T/T so that equation (A,S) becomes
t k~<o [fe+Ck+t);CIIO ]
.!I'{(J(t)h(t)]· ~T(t)} = -. 2 F(q)H*(s - q) dq
27T} k--a> <+(k-t);Cllo
(A.6)
I AlIlllht'r tlt'rivalillll 1>1' Ihe cunvl>huiun formula can be directly obtained from the
. 111111 ..1'''1'111 tlt'lillilioll .... 11 (See also l'rnhlclll 4.1.)
I I'''' silllpikily WI' as,,, II It' 111111 ((lI)iI(O) . ,n. Howeyer, the sallle re~1I1t is Ilhlained if
11111/'(1) tI- II. TIll' nlll"l .1""YllliulI i~ Icflils 1111 l""c,'dse Ii,,· Ihe rl';ulc,·.
Changing q to q + jkwo, and since H*(s - q) is a function of e7'('-Q) only,

we get
9{[f(t)h(t)] . c5 T (t)}
1 le-oo ic+/fllO/I 2fT
= -. I F(q + jkwo}H*(s - q) dq, wo = -
2fT} 1<--00 e-/(A)o/8 T
(A.7)
Equation (A.7) is further simplified as follows:
9'{[f(t)h(t)] . c5 T (t)}
= ~ fO+lhlITIH*(S _ q} qd [.! leICX) F(q + j 2kfT)] (A.8)

2fT} o-/hl/TI Tt--oo T
Noting that the bracketed term is P*(q),
T fc+/h,ITI
9'{[f(t)h(t)] . c5 T(t)} = - . H*(s - q}F*(q) dq (A.9)
2fT} <-/IIIIT)
Similar derivations could be used to prove the formula for the modified
z-transform convolutions.
Appendix 2: Derivation of total square integrals formula ll •22

The solution of the integral I" given as
/" = !h(kya = ~.1 .;f(Z).;f(Z-l) d% (A. 10)

11:-0 2fT} hDlt %
circle
is presented in this Appendix. The derivation is based on obtaining the

right side of equation (A.lO) as a sum of residues. However, instead of
using the explicit expressions for the residues, we obtain a set of 2n + I
linear equations, with the sum of the residues as one variable. Because
of symmetry this set reduces to n + I equations.
Assuming Jf'(z) is given,
(A.11)
The integral of equation (A. 10) is solved by the method of the residues;
thus we are interested in
zA(z) ~ ,
- - = ~ (',."z. k = n. I. 2•...• " (A.12)
z - Pit ,-II
Since we assume po to be the location of the pole at the origin of the
z-plane, we have
Co,,, == 0 when k F 0 (A.l3)
co,o == aft
We also have
ZA(Z-l) ZA(Z-I)p Z-I A(Z-I) "_
-1 == ( _~) == -zp" -1 == -zp" I ci,,,z i
21 - p" 21 p" - 21 21 - PIc i-O
(A.14)
and
(A.1S)
By partial fraction expansion,

B(z)B(z-l) ~ Rpt ~ R;!
-"-'---'-..;... == " - - + "
zA (z)A (Z-l) "-021 - Pic "-121 _ p;l
(A.17)
where RpA equals residue atp,,; R;"l equals residue at Pi" 1.

In the preceding we have assumed only simple poles. Later we shall
indicate that the results are also valid for multiple poles.
For k F 0, we have
zA(z)A(z-l) = z(z - p,,)(Z-l - p,,)[A'(p,,)A'(p;l)]Pk=Z
== -p,,(z - p,,)(z - p;l)[A'(p,,)A'(Pkl)]"k~Z (A.IS)
lind
R = B(p/e)B(p;;l) == _R- l (A.19)
PI: _ p,,(P/e _ p;;l)A'(p/e)A'(p;;l) Pt
('lIl1lbining equations (A. IS), (A. 16), (A.17), and (A.19), we obtain
ft ft ..
+I I Iz-IRptc""aft_i-I
/e-O 1=1 t-O
ft " ft
I I I zIR""p/eci,/eaft_i-J
+ zlr-,I/OI-O
•• ,. n
+ z~ ~ ~z·
'·-1/-1,-0
IR".Prrl"II/,lra" i (A.20)
We can also write

n n ft ft
B(z)B(z-l) = !zl! b"_Hlb,,_j + Iz-; I b,,_j_;b,,_j (A.21)

i-O j.O i-I j-O
Let
Mj =
Ie-O
!" Cj.kRpk (A.22)
and
Qj = I" Cj.lcRptplc
1e~1
(A.23)
From equation (A.12), we have

k = 0, I, ... ,n and Clk
a",
= --
Pic
k = 1,2 •... ,n (A.24)
Then
n
Mn = ao!Rpt
k=O
= I"ao (A.2S)
In addition,
Ql = -an!" Rpt = -a,,1 + a"Ro

1c=1
n (A.26)
Since
COIc = 0, when k:F 0
then
QO= 0 (A.27)
Combining equations (A.20) and (A.21) and using the preceding relation-
shi ps, we have
PI tI ft fJ
!Zi! bn_l+lbn_1 + !z-I Ib"-Hb,,-,

1-0 leO 1=1 i-O
fa fa n ft
= !Zl I Ml+ja,,_j + !z-I I Mjan_i-I

1=0 i-O 1=1 i-O
n n n n
+ ;I.. 0zl+1 i-I
I Qja,,-f-i + ! 1=1
Z-I+l I
i=O
Q/+jQ n-I (A .28)
Equating the coefficients of ZO in equation (A.28), we get

"n n
!
i-O
b~ = !
1'.0
M n_jaj + i-O
I ai Qn-HI (A.29)
We notice from equation (A.28) the coefficients of Zl is equal to the·
" n
• Note that I
i-'O
h,h" / - !
/-11
h,h/ -I'
CONVOLUTION %-TRANSfORM 171
coefficient of %-1. Thus, by summing the equations for j and -j,
n " "
2 !btbHJ = !M H7I- t t + !M 7I _.aH
Q
t=o i~O i~O
n-I "
! Q,,-t QH+l + i=O
+ 1=0 ! Q,,-Hi+1 Qi
" "
~ Qj(M J+n- i + Q,1-i+Hl) + ! Q;_;(M 7I-i + Q,,-i+l)'
= 1-0 i-O
for j = J, 2, 3, ... , n (A.30)*
where 0i = 0 when n < j and 0 > i.
Equation (A.30) can also be written as
"n II
2 ! bibi+1 = ~ at_/(M ,,+i + Q,,-i+l) ~ Qi_/(M n-i + Q,,-i+l)

+ i-O
i-O i~O
n
= ~ (ai-l + Qi+I)(M n-i + Q,,-I+l)'
/=0
for j = 1,2, ... ,n (A.31)
The n + 1 simultaneous linear equations (A.29) and (A.31) can be
written as the vector equation
Om =d (A.32)
where m and d are the vectors
Mn !b~
M n-l + Qn 2 ! bi bi +1
M n-8 + Qn-l 2! bi bi +2
and (A.33)
Mo + QI 2bob n
and Q is the matrix
Qo 01 Q2 03 (In
01 00 + Os 01 + 03 °2 + 0. 0,,-1
Q= Q2 Q3 Qo + 0, al + 06 0,,-2
(A. 34)
a" o o
• 1\ ~h()1I1d be: nlllt'" Ihlll Ihi~ cqlllliinn 1I1~o yield~ equal ion (A.29) when j .... O.
172 THF.ORY AND APPLICATION OF THE Z-TRANSFORM
Since, from equ.ation (A.25)
Mn
/,,=- (A.35)
00
Hence
I" = J.!hl
QolOI
(A.36)
where 0 1 is the matrix formed from n by replacing the first column by the
vector d. Since I" = In(ou, 01' •.• , 0,1' bo, bl> .•. , bll) is an analytic function
of the coefficients Q j and b,t 0 ~ i ~ n, the result is also valid for multiple
poles.
Table III presents the values of In for n = 1,2,3,4.
REFERENCES
J. Jury, E. r., Sampled-Data C(lntrol Sy.vtems, John Wiley and Sons. New York, 19S5.
2. Tsypkin. Y. Z., Theory of P"lse Systems (in Russian), State Press for Physics and
Mathematical Literature. Moscow, 19S8.
3. Gardner, M. F., and J. L. Barnes, Transients in Linear Systems, Vol. '1, John Wiley
and Son~, New York, 1942.
4. Jury, E. I., "A Contribution 10 Modified z-Transform Theory," J. Franklin Inst.,
Vol. 270, 1960, PI" 114-129. GO
S. Lawden, D. F., "The Functions! nrz" and Associated Polynomials," ProCl't'din,1{1

,,~t
o,'tl/e CUI/Ibrid..~e PI/i{os('phical Societ)', Vol. 47, Part 2,1954,1'1'.309-314.

6. Whcclon, A. D., "On the Summation of Infinite Series in Closed Form," J. Appl.
Ph)'.f.• Vol. 25,1954, pp. 113-IIS.
7. Ku, Y. H., and A. Wolf, "Laurent-Cauchy Transforms for Analysis of Linear
Systems Described by Differential Difference and Sum Eqllations," PrOf. IRE. Vol. 48,
1960, PI'. 923-431.
S. Di~cussion of Reference 7 by E. I. Jury in Pr(lf. IRE, Vol. 4S, 1960,l>p. 2026-27.
9. Jury, E. I., and M. A. Pai, "On the Summation of 1: .q and Its Associated Ir.te-
.. on
grals," J. Franklin In.f/., Vol. 271, No.2, February 1961.
10. Stone, W. M., "A List of Generalized Laplace Transforms," Iowa Slal(' Col/eKe
JOIITl/a{ (If Sc:iI'IlU, Vol. 22, April 1945, pp. 215-22S.
11. Whalen, B. A.. "Noise Power Gain of Discrete Filters," Master of Science Thesis,
University of California, Berkeley, June 19S9.
12. Mori, M., "Statistical Treatment of Sampled-Data Control Systems for Actual
Random Inputs." T,.ans. ASME, Vol. SO, No.2, February 1958.
13. Kaplan, Wilfred. Operatio"al Melhods /I" Lim'ar Sy.fl(!/1/l, Addison Wesley Pub-
lishing Co., Reading. Massachusells, 1962.
14. Pai, M. A., "The Analysis of Non-linear Feedback Silmpled-()i1ta Systems,"
Ph.D. Thesis. Dept. of F.lectricill Engineerin!~. University u«'alifllrni", Ilcrkeley. 1')1II.
IS. Thellicr, P. L., Ooctnr of Science Thesis, lInivcr\ity III' (il"cnllblc, ....alll:'.. 1%2.
CONVOLUTION %-TRANSFORM 173
16. Nishimura, T., and E. I. Jury. "Contribution to Statistical Designing of Sampled-
Data Control Systems." Electronics Research Laboratory Report Series 60, Issue No.
210, August 5. 1958, University of California, Berkeley.
17. Ragazzini, J. R., and G. F. Franklin, Sampled-Data Control Systems, McGraw-
Hill Book Co., New York, 1958.
18. Tou, J. T., Digital and Sampled-Data Control Systems, McGraw-Hili Book Co,
New York, 1959.
19. Monroe, A. J., Digital Processes lor Sampled-Data Systems, John Wiley and Sons,
New York, 1962.
20. Montel, Paul, Lerons sur les Recurrences et Leurs Applications. Gauthier-Villars,
Paris, 1957.
21. Friedland, B., "Sampled-Data Control Systems Containing Periodically Varying
Members," Automatic and Remote Control, Proceedingf 01 the First I.F.A.C. Congress,
Moscow, 1960. Butterworths, London, 1961, p. 369.
22. Bharucha, B. A., "Analysis of Integral-Square Error in Sampled-Data Control
Systems," Electronics Research Laboratory Report Series 60, Issue No .. 206, June 1958,
University of California, Berkeley.
23. Nishimura, T., "On the Modified z-Transform of Power Spectral Densities,"
Trans. IRE, PGAC, Vol. AC-7, No.4, July 1962, pp. 55-56.
24. Sawaragi, Y., and A. Fukaw, "On the Sum of Squares Values of Sampled-Data,"
'frans. 01 Japanese S.M.E., Part I, No. 155, Vol. 25, July 1959.
5
CONVOLUTION z-TRANSFORM
APPLIED TO NONLINEAR
DISCRETE SYSTEMSl,3,6
In the preceding discussions, the z-transform method has been mainly

applied to linear discrete systems. For time-invariant systems, the z-
transform offers some advantages as has been explained in detail. How-
ever, for time-varying systems this advantage is no longer apparent
except in a few particular cases. Nevertheless, the method is still
applicable as shown earlier.
In this discussion, the convolution z-transform will be applied for the
solution of certain nonlinear discrete systems described by nonlinear
difference equations. It will be shown that the method proposed is
general in solving such difference equations. The form is
r
Z ht>x(n + p) + F[x(n), x(n + 1) ... x(n + r - 1)] =0 (5.1)
JJ=O
where " is the discrete time variable and F[x(n), x(n + I) ... J is the
nonlinear term containing terms in x(n + p) of degree two and higher.
This form of the difference equation is fairly general in the sense
that it describes many nonlinear discrete systems with saturation-type
nonlinearities. This type is quite common in feedback sampled-data
systems.
The objective of this chapter is to obtain an approximate closed-form
solution of equation (5.1) in a systematic manner using the convolution
z-transform; a few examples arc; also discussed to explain in detail the
method of solution and to compare it with the exact method using the
difference equation as a recurrence relationship.
In order for the method of solution to he applicahle, certain assumptions
are to be imposed on the: tlilrercnce l'llulltion. These IIssumptions lire nol
174
2-TRANSFORM APPLIED TO NONLINEAR DISCRETE SYSTEMS 175
restrictive, for in practical situations they are usually satisfied. It should
be noted that there is no general method of solution for any nonlinear
difference equation, .,6-10 but certain techniques are applicable to certain
types of difference equations or noniinear discrete systems. In this chapter
the application of only one such method is discussed in detail. This
method is analogous to the application of the convolution Laplace
transform 2 to certain nonlinear differential equations. Rigorous mathe-
matical justification of this method was proposed by Wasow. 11 Similar
justification for the method has been indicated$ and by applying Wasow's
techniques a rigorous mathematical validity has been obtained. 12
5.1 Assumptions
I. The nonlinear difference equation in (5.1) has a steady-state or
equilibrium value Ao given by
r
! h~Ao + F(A o• ••.• Au) =
11-0
0 (5.2)
2. The nonlinear function F[x(n) •.• z(n + r - I)] is analytic in its

variables X(II), x(n + \) ... X(1l + r - \) around the equilibrium point.
Only nonlinear functions involving polynomials are to be considered.
3. The equilibrium point is asymptotically stable in the sense that x(n)
converges to the equilibrium point Ao as n -+ co. The initial values of the
dependent variable must lie "close" to the equilibrium point in order that
II convergent solution may be obtained. However, in some cases, as' will
be explained later, by shifting the time axis (U!;ing the difference equation
~IS a recurrence relationship) by one or two sampling periods, quicker
convergence can be expected.
The basis of the solution of the difference equation is to assume
:'1'(::) = J[r,,] to be a meromorphic function of the form
GO A
.~(2) =! II _I (5.3)
II~ 01 - a,.%
Whl'I'C (I ..'s are distinct with modulus less than unity, and ao = I.
Ikl'OlIIse of Ihis assumed basic form of the solution, we have to obtain

IIII' .:-Intnsform of functions of the form
JI/(" -/- k )'1 or 'yU(II + r),,/(n + q)I],

111111 ~,II 1111. '1lwll'flln' Wl' !tlmll prm'l'ed first to derive this ::-transform by
Ihl' nllwllllllillll ·111111 ... 1"111'111 IIl('lhllllll' di""IIS!tl'" ill ('hilpll'" 4.
5.2 Convolution z-transforms of certain functions

Assume fen) is a time function defined for discrete values of n = 0, I, 2•
. . . , and its z-transform is of the form
00 A
=
1[/(n)] ~(z) ! =«=01 - « -1
Q,.z
(5.4)
where 0 0 = I and the other Q'I:5 are distinct with absolute values less than
unity. Using the convolution integral,
1[f(n)2] = _1
. 2~)r
r
p-l!P"(P)~(=)dP
p
(5.5)
where r encloses all the singularities of p-l!P"(p) in a positive direction
in the p-p[ane. Therefore
1
[/(n)2J = _1 l -1~ A" ~
~ 1 - Q,.p-1 ,-0
27rJ. [. P ,,=0
A, d
~ I - Q,z -1 P P
(5.6)
Similar[y,
I[J(n)3] = III A,.ApA y

,.=0,.0 y-o 1 - Q"O,QyZ
-1 (5.7)
The extension to ,,[f(n)l] is evident.

It is also easy to obtain the following transforms in a similar manner.
1[J(n + k)2 = :~:[/(q)'lz-q]

zk[1[J(n)l] - (5.8)
For the nonlinear terms of the form!(n)!(" + k), we use the following:
1[/(n)J(n + k)J = ~
27rJ
rp-lpk
)1'
X [~(P) - ~>(q)P-q]~(;)dP (5.9)
where r encloses the singularities of p-lpk[$'(P) - kf!(q)p-q]

Equation (5.9) is evaluated as 11=0
(5.10)
Following the procedure outlined. we can, in general, find the ;:-

transform of nonlinear terms of the type fCIl + ,)It/<II + if)', where
" k, q, I are integers> O. pruvidillA .JII<Il)l ill I(iven hy etlulltiun (5.4).
Z-TRANSFORM APPLIED TO NONLINEAR DISCRETE SYSTEMS 177
5.3 Method of solution for second-

and higher-order equations
Consider a second-order nonlinear difference equation
x(n + 2) + bx(n + I) + cx(n) + kx(n)S = {3u(n) (5.1I)
where u(n) is a sampJed unit step function and the coefficients b, c, k are
constants. The constant Pis the magnitude of the step forcing function.
I. Assume the z-transform ,q[(z) of zen) to be a meromorphic function
of the form
.q(z) = f
II~O 1-
All
OIlZ-1
= Ao + Al + A2
1 - Z-I 1 - a 1z- 1 1 - a2z-1
+ As 1+ A41 +... (5.12)
I - a3z- I - 04Z-
The term O<z for IX ~ 3 is of the form aro~ (y and lJ are positive integers,
y + lJ ~ 2) arranged in decreasing order of magnitude so that
\ > lOll > 10 1> 10 1> 1041 > ...
2 3 (5.13)
2. Take z-transforms of all the terms in equation (5.11) to yield
2"(Z)(Z2 + bz + c) + kJ[x(n)S] = ~ + (Z2 + bz)x(O) + zx(\).

z- 1
(5.14)
Dividing by Z2 + bz + c = (z - a;)(z - a;) gives
.~"(z) + k J[x(n)3]
(z - a;)(z - a~)
= . t I [~+ (Z2 I + bz)x(O) + ZX(I)J (5.15)

(z - al)(z - U2) z - I
Now if .f(z) is written in the form
.'?f(z) = ~ + Al + A2 + A 20
I- Z-1 I - a l z- 1 1 - a22- 1 1- a~z-l
A30
+t . 3 -I + ... + 1 - A02U2Z2 -I
- a1z
+ -~:-I
A03
-I + . . .
+ All -1
I - {liZ t - a 1a22
Alii A. 2
., a'.+ -~-I+'" (5.16)
I . "."11: 1 .- lIllla:
178 THEORY AND APPLICATION OF THE ;::-TRANSfORM
the z-transrorm or X(II)3 can be written as

.\ A30
J[x(n)' J =
I - %-1
+ I 3A2oA 1
- al%-I
3A2oA2
+ -..::.....:=-
I - Q2Z-1
+ 3A~A20 + 3AoA~
I - a~z-I
+ 3A~A30 + A~ + 6AoAtA20
I - a~z-'
+ 3A~A40 + 3A~oAu + 6AoA,A30 + 3A~A20
1 - a~z-'
+ ... + 3A~A02 +2 3_~oA~ + ...

1- Q2Z
+ 3A~A21 + 3A~Az + 6AoAzoAz
I - a~azZ-l
+ 3A~AII + 6Ao~lA2 + . . . (5.17)

1 - o,azZ-
It is assumed that a;, o~ in equation (5.15) are distinct from the a's in
equation (5.12). Comparison or coefficients is now possible in equation
(5.15).
3. By multiplying both sides of equation (5.15) by % - I and letting
:: --.. I. the steady-state relation involving Ao is obtained. Ao can also be
obtained by letting x(n) = x(n + I) = x(n + 2) = Ao in equation (5.11).
4. By multiplying equation (5.15) by z - 0Ot. IX = 1, 2 and by letting
in each case z - 0",. we obtain
AI,2(a~.2 + ba,,2 c + k3A~) = 0+ (5.18)
For nontrivial solutions of AI, A2 in equation (5.18).
a~.2 + bal,2 + c + k3A~ = 0 (5.19)
°
which determines 1 and 02 in terms of steady-state amplitude Ao. All the
3) can then be evaluated as explained earlier in equation (5.12).
Ofl.·s (IX ;;::
5. By multiplying equation (5.15) by % - 0fl.' IX ;;:: 3 and by letting in
each case.: - a,.. the following relations are obtained:
A3 = !1(A o• AI' A 2)
A4 = !2(A CI• AI. A 2 • A:I)
( .....'(1)
;;-TRANSFORM APPI.IF.D TO NONLINEAR DISCRETE SYSTEMS 179
where fl' ./~, .... etc., denote functional relationships. The process by
which any A,(1. ~ 3) in equation (5.20) is found in tcrm~ of previous
A;s is possible because of the manner in which the a;s were ordered
in e'luation (5.12). Thus all A~'s (~~ 3) can be expressed ultimately in
terms of Au. AI' and A 2 •
6. Finally. by multiplying equation (5. I5) by z - a~, (1. = I, 2 and
letting in each case:; -> a~. two nonlinear relation~ are obtained.
k ('-'--I
Ai: 3A~AI + 3A~A2 ) iJ
+ , , + . .. = -'--I
Q a- ao - a l a. - "2a. -
+(a;+b).r(O)+:r(l) ,,;=1,2
(5.21)
The convergence of these nonlinear relations depends on the initial
values of .r(lI) and the magnitude of the nonlinearity. If the nonlinearity
is "small." then for a fairly large range of initial values convergence could
be expected. On the other hand, when the nonlinear terms are comparable
to the linear terms, then the range of initial values is necessarily restricted.
At any rate, convergence is best judged from the particular problem on
hand. With first-order difference equations, this does not pose a serious
problem. However, for higher-order systems the numerical computations
become too involved and it is also sometimes difficult to determine the
cnnvergence. The procedure just detailed can be easily extended to
higher-order equations.
5.4 Illustrative examples
To illustrate the method discussed in the preceding section. two examples

are introduced. The first example represents the solution of a certain
nonlinear first-order difference equation. This equation can be solved by
other methods which are limited only to first-order case. 4 The second
l'xample represents the analysis of a first-order saturating sampled-data
Il'cdbad system. 1 We limited these examples to first-order for ease of
1'l1l11putalion. However, as outlined in Section 5.3. the method is general
and call he applied to second- and higher-order nonlinear difference
\'qllatiolls or discrete systems. A study of a second-order system is
1I1l'IIIdni ill Chapter 1<.
I'XAMI'I F I
("'II~id\'r Ihc soilltioll of the following difference equation.
"!lIII I I)
H.'I(II)~ +1 (5.22)
!/(" f I I
1111I.llitlll (', .'.') 1.111 III' \C'\\Tilh'1I as
H'/(I/ I II I H'/(I/ I 1)1/11/)" H!lIIIf l'i.~~.\)
180 THEORY AND APPLICATION Of THE %·TRANSfORM
Assume
l!V(%) = i: All -I
II~O 1- Q,,%
Ao
= ---1
1- %
+ 1 - Alalz -1 + 1 - AeQl%e -I + . . . (5.24)
By taking the z·transform of equation (5.23),

8%[!V(%) - yeO)] + 8J[y(n + 1)y(n)2] - 8J[y(n)2] = ~
% - 1
(5.25)
But, as shown in Section 5.2,
(5.26)
and
+
J[y(n - 1)y(n)2] = -1.
21TJ ['
1. p-Ip LeoI.. 1 - AO"p..
0
-I - yeO)]
.
(5.27)
where r encloses the singularities of
p-'pLi:=0 1 - AllO"p -1 - yeo)]

•
This equation becomes
(5.28)
Substituting this transforms into equation (5.25) and dividing by z, we get
8 f
".0 1-
A"
a,.%-I
+!f I
zL.,,,,Y-G 1 -
A .. A,tA 7o y
o"O,tOyZ-1
3
Ieo 1 - Aa"a,z-I
- ",/1=0 A
"" ]
= - - + 8y(0) (5.29)
%- 1
Comparing coefficients of 1/(% - l)andwriting%/(z - ex) = [ex/(% - ex) + I],

we obtain
8Ao + 8A~ - 8A: = 3 (5.30)
The equilibrium solution Ao can be obtained, which is
A.. =0.5 (S.31)

The solution around this point is obtained as follows. Comparing the
coefficients of I/(z - Ql),
8Al + !. (A~AIQl + 2A~Al) - !. (2AoAl) = 0 (5.32)

al 01
Rearranging this equation,

AI(ol + A~al + 2A~ - 2Ao) =0 (5.33)
For the nontrivial solution of At> the term inside the parentheses must be
zero.
(5.34)
Substituting Au =! in this equation, we obtain Q I = §. Comparing
coefficients of I/(z - aI), the following relation is obtained:
8a~A2 + 8(A2A~0~ + 2A2A~ + 2A~Aool + A~Ao)

- 8(A~ + 2AoA2) = 0 (5.35)
which, when simplified. yields
A2 = -~A~ (5.36)
In the preceding manner. by comparing coefficients with the other terms,
A 3, A 4•••• can be determined in terms of AI' To obtain the nonlinear
relation for AI' compare the constant term on both sides of equation
(5.29). This gives
<tl
I
a~O
A.. = yeO) (5.37)
Thus
II¥( ) _ 1 + AI _ ! A~ + ... (5.38)
, z - 2(1 _ Z-I) 1 - ~Z-I 3 (1 - 246 Z- 1)
which. upon inverse, yields

yen) = ~ + A (i)" -
1 iA~a5)" + .,. (5.39)
If !I(O) is specified. Al can be determined from this equation. providing
(5.39) converges for the given value of yeO).
EXAMI'I.!: 2
III this cXlIIlIple u sumpled-data system with a small nonlinearity in the
f,'cdhal'k path is considcred. The nonlinearity is generally incidental and
IIll1y rl'prcscnt a sli~ht dcviution from otherwise normal linear feedback.
'I'll" prllhll'llI is (if sllch a systcm is stable) how to gct an cxplicit solution
for Ihc IIIIIPllt whcn Ihc furcin~ function is II stcp input, The stahility of
the system is determined from the stability of the "linearized" system,

with nonlinearities neglected. By referring to Fig. 5.1, the difference
equation relating the output to the input is first derived. An explicit
relation is possible between the sampled output and input. The input is
assumed to be a unit step function. The nonlinearity is represented as a
finite power series, as indicated in Fig. 5.1.
Let ret) = u(t), where u(t) is a unit step function
E(s) = R(s) - .!l'(m1c + Imllcv) (5.40)

11-2
where .!l' indicates the operation of taking the Laplace transform of the
function in the parentheses. The sampled form of equation (5.40) is
E*(s) = R*(s) - Z[.!l'(m1c +o-a

fmoc lI ) ]
I~.f'·
(5.41)
where Z indicates z-transform operation. In equation (5.41) the .!l'-

transform operation can also be omitted. The output transform is given
as
C*(s) = G*(s)E*(s)
= G*(S){R*(S) - Z[.!l'(m1c + Imllc lI ) ] } (5.42)
CJ-=2 .".".
Now
C*(s)/.=er• = <;fez) (5.43)
Therefore in z-transform notation,
<;f(z)[1 + m 1~(z)] + ~(Z)Z[.!l' C~2mvc")] = fR(z)~(z) (5.44)
It is easy to see that in the absence of nonlinearity and ml = 1, we obtain

the linear sampled-data system with unity feedback.
Zero-order
hold
c(t)
Gi.)
N
1II1C + ~ 1119C9,..----:--:---,
,-2
FIGURE S.I Ntlnlinelu' NIIIllI,lcd-dlllll r«dhm.:k lIy~lcll1.

Choose a first-order linear plant with G1(s) = 1/(s + 1),
~(z) = <[_1_
s+1
1-se-'I'S]
1 - e-'I'
- Z - e-'I'
(5.45)
With T= I
~(z) = 0.632 (5.46)
z - 0.368
and
~(%)[Z - 0.368 + 0.632mlJ

+ 0.632Z[2(fmqcq)] = 0.632% (5.47)
q~B z- 1
This corresponds to the nonlinear difference equation
N
c(n + 1) + (0.632m. - 0.368)c(n) + 0.632 I mqc'l(n)
= 0.632r(n) (5.48)
If mq = 0 (for all q), in equation (5.48) and m. = I, it reduces to the
linear system with unity feedback, as explained before. Equation (5.48)
then reduces to
~(%)(z + 0.264) = 0.632z (5.49)
%- 1
Taking the inverse z-transform or making use of tables, the solution for
linearized system is obtained as
c(nT) = HI - (-0.264)RJ (5.50)
By inspection it is seen that for n approaches infinity, c(nT) tends to ~ and
hence the linearized system is stable.
For the nonlinear system, assume that mq = 0 for all q in equation
(5.4K) except q = 3 and m. = I, and ma is "small" compared to J. Since
Ihe linearized system is stable, assume*
<r,'(z) =i All (5.51)

I - a,.z-·I
II 0
wilh till ,.." I (from steady-stute conditions), 02 = of. 03 = a~• ... , aq =

/I'f . . .. With '(,'(::) ~Issumed ;IS curlier, it readily follows from the discussion
• TIJI~ rllnll IIr ~lIhlllllll ire the bllsi~ IIr Ihe CUIIYlllutiulI %·trllllsrllrlll liSt'" ill Ihi~ CXlIl1lplc.
184 THEORY AND APPLICATION Of THE Z-TRANSFORM
of the previous section that

[c(n)3] = ( A~ + 3A~AI + 3A:Az + 3A~Ao
:1 1- Z
-1 1- Ql%
-I 1- I -1
Ql%
+ 3A~A3 + 6AlAaAo + A~
1- Q~Z-l
+ 3A~A4 + 3A~A2 + 3A:Ao + 6AIAaAo

1- Q~Z-1
+ 3A l A: + 3A~Aa + 6A.A4AO + 6AIIAaAo + 3A5A.s

1- a~z-1
3A~A6 + 6AIAsAo + 3A~A,
+ A: + 6AIAIAs + 6AoAtA, + 3A:Ao
+------~~~~~~~~~~~~~
1 - a:z- 1
+ ... + .. .) (5.52)
Take z-transform of equation (5.48) assuming zero initial conditions and
use equation (5.52) to get
f{(z) + 0. 632ms ( A~ + ...) = 0.632z
z + 0.264 1 - Z-1 (z - 1)(z + 0.264)
(5.53)
Comparison of coefficients is now possible. Comparing the coefficients
of z/(z - I), z/(% - 0 1), %/(% - a~) ••• z/(z - a1), and finally z/(z + 0.264),
the following relations are obtained:
A
o
+ 0.1.264
632m s A3 =
0
0.5 (5.54)
A + 0. 632m s (3AtA) = O' (5.55)

1 OJ+ 0.264 0 1
A
2
+ a~0.+632m s (3A 2 A + 3A t A ) =
0.264 0 2 1 0
0 (5.56)
0.632m 3 ( - ~~ ___ ~~~.__ - (3-1.~~~±.~_~_~.!!1_ ... )

1.2M I'I + O.2M (II + 0.264
....- ().~ (~.~K)
Z-TRANSFORM APPLIED TO NONLINEAR DJSCRETE SYSTEMS 185
Equation (5.54), when solved for Ao, gives the steady-state value of
c(nT). Substituting this value of Ao in equation (5.55) gives
A 1(al + 0.264 + O.632m33A~) = 0 (5.59)
For nontrivial solution of AI' it foHows that
al + 0.264 + 0.632m33A~ = 0 (5.60)
from which 0 1 can be readily computed. If lOll < 1, the system is locally
stable around the equilibrium point. The other equations can be succes-
sively solved for A 2 • A 3 • A4 ..• in terms of AI' Using equations (5.54),
(5.55), (5.56), and (5.57), equation (5.58) can be written in the form
Ao+AI+Az+ •.. =O (5.61)
This is merely the initial condition requirement (assuming zero initial
condition) which can be written by inspection of the form of ('6'(z) in
equation (5.51). By knowing Ao and A 2 , A 3 • •• in terms of AI> equation
(5.61) can be solved for Al for whatever accuracy desired. A 2 , A3 , ••• can
also be then computed. By taking a few trial values of Al convergence
can be easily ascertained. It is now a simple matter to take the inverse
z-transform of equation (5.51) to get the time domain solution. The actual
numerical calculatio.ns are now carried out.
Assuming* rn3 = 0.1. solution of equation (5.54) yields
Ao = 0.494 (5.62)
With these values of Ao and rna, ° is calculated from equation (5.59) as
1
a1 = -0.31 (5.63)
Since lad < I, the system is stable for a unit step function input.
From equations (5.56) and (5.57), A z• A 3 • •• are obtained as follows:
A _ _A _Z
1 A ___ A 31_ A _ A1
2- 4.35' 3- 13.95' 4- 6. 14 '
AS A~
As =- 10.;5' Ae =- 26.7' .. (5.64)
Suhstituting equation (5.64) in (5.61) gives

Ai A~ At
0.494 + Al - 4.35 - [3.95 + 6.14
A~ A~
- -10.25 - -26.7 + ... = 0 (5.65)
• hll III. (1.1. whkh i,"lkates a Salllnltiull type nonlinearity, it will be found that
,u"vl"I:I'IJ('I' 1'1' l'll"illi"" (~,44) ;~ 1'''''''' lIowever. hy shiftill!! time axis (lIsin!! the
,hlll''''''''(' (,IJlllllill" I" " "'1'111 n'm'(' !l'IIII;IlII) hy 1IIIt' '1IIIIplill/: Pl·ri,,&!. 111Ik-ker cunver·
1:(,111'1' 1""' II(' 1'~IlC'I'I.',1 'J hi' i, tllmllnl.',1 Inl .. r.
Solution by trial and error of equation (5.65) yields

Al ~ -0.461 (5.66)
Hence
AI = -0.0489 Aa = 0.007025 A. = 0.00735
A:; = 0.002015 A. = -0.000359 (5.67)
Therefore
Iit(z) = 0.494 0.461 0.0489 0.007025
1 - Z-l 1 + 0.31z- l 1 - 0.096z-1 + 1 + 0.0298z-1
+ 0.00735 + 0.002015
1 - 0.0092z- 1 1 + 0.00285z- 1
0.000359
1 _ 0.000884z-1 + ... (5.68)
Corresponding to
c(nT) = 0.494 - 0.461(-0.31)" - 0.0489(0.096)"
+ 0.007025(-0.0298)" + 0.00735(0.0092)"
+ 0.00201 5( -0.00285)" - 0.000359(0.000884)" (5.69)
It should be noted that some of the modes are oscillatory decaying unlike
the continuous case which are exponentially decaying. The nonlinear
difference equation of this system is
c(n + 1) + 0.264c(n) + 0.0632c(n)3 = 0.632r(n) (5.70)
With c(O) = 0, this can be solved as a recurrence relation to get an exact
solution. A comparison of this exact solution with the approximate
closed form solution in equation (5.69) is given and it is seen that agreement
is good up to the third decimal place. The output between sampling
instants could be easily obtained by minor modification of equation (5.51).
Sampled Exact Closed Form
Output Solution Solution
c(0) 0 0.000131
c(l) 0.632 0.6323
c(2) 0.4491 0.44925
c(3) 0.50779 0.50767
c(4) 0.48970 0.48975
c(5) 0.4955 0.4953
c(6) 0.4935 0.4936
c(7) 0.4942 0.4941
EXAMPLE 2a
If we take in Example 2 m3 = -OJ, the difference equation that describes
the system is now
c(n + 1) + 0.264c(n) - 0.0632c(n)3 = 0.632 (5.71)
For this equation, the coefficients are
Ao = 0.5065 A. = 0.93A~
At = 0.356A~ As = J.7A~ (5.72)
Aa = O.64A~ As = 3.74A~
It is noticed from the preceding that the convergence is rather poor.
However, it can be improved if we shift the time axis by one sampling
period so that the difference equation is the same but the new initial value
is 0.632 (instead of zero); this can be done by using t.he difference equation
as a recurrence relationship. For instance, if we let c(n)I,,~o = 0 in
Equation (5.71), we readily obtain the new initial value as 0.632.
The amplitude determining equation is
0.632 = 0.5065A 1 + 0.356A~ + O.64A~ + 0.93A~
+ 1.7A~ + 3.34A~ (5.73)
which gives
At = 0.1191 AI = 0.00506
(5.74)
Aa = 0.001 A. = 0.0001
The solution is then
c(nT) = 0.5065 + 0.1191(-0.2155)"-1 + (0.00506)(0.0465)"-1
+ (0.001)(-0.01)'1-1 n ~ I, for n = 0, c(O+) = O. (5.75)
. It is evident that this solution is rapidly convergent. Comparing the
output amplitude with the exact solution, we notice
Approximate Exact
c(1) = 0.632 c(1) = 0.632
c(2) = 0.4811 c(2) = 0.4809
('(3) = 0.512 c(3) = 0.512
c(4) = 0.5053 c(4) = 0.5054
III cCllldusiun. we ,'(lIIld deduce th:lt bcttcr convergence is expected if

Ih,' m~w inililll vllh,,' i ... IWIIC tlw \'(luilihriulll pnint. The new initial vullie
can be easily obtained for the discrete case by using the difference equation
as a recurrence relationship. Such an advantage is not possible in the
continuous case. A compromise between using the difference equation as
a recurrence relationship and the labor involved in the convergence is
required in any problem amenable to solution.
REFERENCES
J. Jury, E. I., and M. A. Pai, "Convolution .:-Transform Method Applied to Certain

Non-linear Discrete Systems," IRE Trans. PGAC, Vol. AC-7, No. I. January 1962,
pp.57-64.
2. Weber, E., "Complex Convolution Method Applied to Non-linear Problems;'
Proceedings (1' lire Symposium on Non-linear Circuil Ano~l'sis, Polytechnic I nstitute of
Brooklyn, Brooklyn, New York, Vol. 6, 1956.
3. Montel, Paul, Lerons sur Its Recurrences el Lel/rs Applications, Gauthier-Villars,
Paris, France, 1957.
4. Levy, H., and F. Lessman, Finile Dijference Equations, Macmillan Co., New York,
1961.
5. Pai, M. A., "The Analysis of Non-linear Feedback Sampled-Data Systems," Ph.D.
Dissertation. Dept. of Electrical Engineering, University of California, Berkeley,
September 1961.
6. Torng, H. C, "Second-Order Non·linear Difference Equations Containing Small
Parameters," J. Franklin Insl., Vol. 269, February 1960, pp. 97-104.
7. Norlund, N. E., "Nichtlineare Differenzgleichungen," Encyklopcdie dcr Mathe-
matischcn Wissenschaflen, Leipzig, Germany, Vol. II, No.3, 1927.
8. Trijinsky, W. J., "Non-linear Difference Equations," CO/llPO.I·iIO MOlh('/IIotic/(, Vol.
5, 1937, pp. 1-66.
9. Tsypkin, Y., Theory of Pulse Systems, State Press for Physics and Mathematical
Literature, Moscow, 1958.
10. Strodt, W., "Analytic Solutions of Nonlinear Difference Equations," Anl/ols Math.,
Vol. 44, No.3, July 1943, pp. 375-396.
11. Wasow, W., "Solution or Certain Nonlinear Differential Equations, by Series of
Exponential Functions," "'il/oid. Malh., Vol. 2, No.2, June 1958, pp. 254-260.
12. Dewey, A. G., "On the Convergence of a Series Solutiun of Certain Nonlinear
Difference Equations," Electronics Research Laboratory Technical Memo No. TM-18,
University of California, Berkeley, July II, 1963.
13. Alper, P., "Higher-Dimensional Z-Transforms and Nonlinear Discrete Systems,"
Technological University, Electronics Laboratory, Delft-Netherlands, August 1963.
14. Alper, P.:,"A consideration of the discrete volterra series," IEEE Trans. Automat.
Contr., Vol. AC10, July 1965, pp. 322-327.
IS. Rault, A. and Jury, E. I.: "Nonlinear sampled-data systems and multldimensional
z-transform," Electm. Res. Lab, University of California, Berkeley, Memo 1121,
February 1965.
16. Lavi, A. and Narayan, S.: "Analysis of a class of nonlinear discrete system using
multidimensional modified z-transform," IEEE Trans. Automat. ConlT. (shorl papers)
Vol. AC-13, February 1968, pp. 90-93.
6
PERIODIC MODES OF
OSCILLATION IN NONLINEAR
DISCRETE SYSTEMS
In this chapter a study is made of periodic modes of oscillation of auton-

omous discrete systems. The types of nonlinear systems to be studied are
pulse-width modulated, relay, nonlinear gain amplifier, and quantized
level systems. The method given is also applicable to obtaining the
periodic modes of oscillation for nonlinear difference equations as indi-
cated briefly toward the end.
In addition, the stability study of limit cycles as well as the limitation on
the period of limit cycles for relay mode of oscillations is presented.
Finally. a certain analytical technique for examining the existence and
stability of forced oscillation in a sampled-data system is discussed.
A~ain the method presented is based on the z-transform, and the examples
arc chosen from the widely used field of sampled-data and digital control
lIystcms.
6.1 Limit cycle analysis of

nonlinear discrete systems
Wltl'n a periodic forcing function is applied to a nonlinear discrete system
III' a spontaneous oscillation is excited within the feedback system, the
'.y~Il'1ll will frequently move into a limit cycle, that is, sustained periodic
IIwilial ions then arise in the system.
()I\l"l' lite feedback system shown in Fig. 6.1 is forced into a limit cycle,
1111" olltput fmm the nonlinear component e;.,h(t) will have a periodic
It'allln' ~illlilar 10 I he other variables of the system such as the output
/',,(1) til tlw (:olltml ('rmr (',,(t). This constitutes the basic assumptions for
IlIc' nllllly~i .. whidl ill l"()IllIllCIIl III all ollll'r sludies or periodic oSl"illatiol\s.
I Hel
190 THEORY AND APPLICATION Of THE z-TRANSFORM
In our study we shall also assume that these oscillations are periodic with
period an integc:r multiple of the sampling period T. With the pulse width
modulated feedback system, as will be discussed later, the signal e~it)
will actually be a train of finite pulses with periodic features as shown in
Fig.6.2. Therefore the application of the theory of the finite-pulsed system
shown in Fig. 6.3 is immediately suggested for the analysis. To obtain
the basic equations that yield the limit cycles, we shall review first some of
the pertinent concepts of finite-pulsed systems. 15 - 11
Let e~.Ia(t) be the output of the finite pulse width sampler as indicated
in Fig. 6.4c during the (n + I )th sampling period. This output is repre-
sented mathematically by the product of e~.Ia(t) and a unit pulse uli(t)
as indicated in Fig. 6.40, b.
e~.Ii(t) = e~(t)uli(t) (6.1)
where
for 0 S; t S; h
elsewhere (6.2)
The incremental response transform, ~C:'(s) is defined as the product
of the transfer function KG(s) and the Laplace transform of the pulsed
input to the linear system Y.le~.h(t)]
AC~{s) = KG(s)£~.II(s) (6.3)
FIGURE 6.2 SlImrling rnllcrn rllr II glll""1 rerill" '/~/

PERIODIC MODES OF OSCILLATION IN NONLINEAR DISCRETE SYSTEMS 191
E~ r---i
hIt .
KG(s) ~S)
FIGURE 6.3 Open loop finite pulse-width system.
The term llC;'(s) represents a fraction of the response transform effected

by the input which is supplied to the linear system during the (n + I )th
sampling period. This incremental response is shown in Fig. 6.4d. Let
n = Mk, Mk + I, ... ,Mk + I . .. , M(k + I) - I (6.4)
,-'
in equation (6.3) where M is given by To = MT. We then add AC;'(s)
from n = Mk up to n = M(k + I) - I, multiplied by the relative delay
L
o -------~
til
(b) Uh (t)
pulsed function
along t-axis
.(e) e~. h (t)

pulsed input
(eI) 6(',,'(t)
Incremental
-~.~
re!>llUnsl!
........
•. ~t
I Ie ;111(1· "."
factor. The relative delay is the delay with respect to the Mkth sampling
instant, and the delay factor is given by
where 1=0, 1.2•... , M - 1 (6.5)
Then equation (6.3) gives
M-l
I1Ck(s) = Ie-liT I1C MkH(S) (6.6)
'~O
Thus I1Ck (s) gives the incremental response resulting from the input
during the (k + l)th global period T G • If this incremental response
I1Ck (s) is superposed for all k from 0 to ro with proper time lags. the
summation will give the total response C(s). Hence
00
C(s) = Ie- kSTu I1Ck(s)
k=O
00 .v-I
= Ie- k• Tu I e- I• T I1C Mk+l(S) (6.7)
k=O lEO
The z-transform of this equation is obtained by using the notation
z = e'T and Z = e'Tu = zM.
where the function Zd[ ] represents an operator that may be described as

the sequence z-transformation and it is performed by multiplying the
function of k by Z-k and summing for all k from zero to infinity. Simi-
larly. the response between sampling instants is given with the aid of the
modified z-transform as follows:
<if(z, m) = IZ-k l1<ifk (z, m) = Zd I

<10 [,\(-1z- Z I m [KG(s)E;.lfk+ll.h(S)]
]
kaO 1=0
(6.9)
The general form of (6.8) and (6.9) can be readily applicable to the
analysis of finite pulsed sampled-data feedback systems. It is used in the
following to aid us in obtaining the fundamental equations for the limit
cycles.
After the system enters the limit cycle. the incremental response of each
sampling period will be repeated for every global period Ta of the limit
cycle, and thus I1Ck (s) of equation (6.6) will become identical for every
k. Thus the incremental responses of each limit cycle become independent
of k. Hence Mk at the subscript of (" and H' muy be removed. Iinwevcr,
we add the subscript .~ tn indicate the skady-state (althnll~h pcrimlically
varying) (If the variuhle. Thl'n the illl'n'lIIt'ntnl respnllsl' in (1m~ limit t'ydc
is denoted as t1Cs(s) and is given by the sum of t1C IS(s) from 1= 0 to
I = M - 1 with proper delay factors (where I indicates the relative
location of the specified sampling period within the global period T () of tbe
limit cycle). It is given as
,\1-1
t1CN(s) =I e-I,T 6,C I.(s) (6.10)
1-0
Using this notation, ~(z) of equation (6.8) will be divided into two
parts, the transient-state and the steady-state (limit cycle)
.\"-1 cc
<tf(z) = l: Z-k Mfk(z) + I Z-k 6,~.(z) (6.11)
k~O k-.""
where
t1~$(z) = <7.[t1C.(s)] and t1~k(Z) = <T[aCk(s)]
Equation (6.11) can also be written as
'\"-1 Z··.\·+!
~(z) = ~ Z-k 6,lifk (z) + - - a~s(z) (6.12)
k~O Z - 1
At th.is point the skip sampling operation (introduced in Section 4.3)
is lJSed to obtain the z-transform with respect to the period Tn = MT
from '?J'(z). This operation is described as <.[ )
and its mathematical
form is given as
1f(Z) = <.['if(z») = -1. ~ ~(p) 1II - 1dp

- (6.13)
21TJ r 1- p" Z P
Applying the skip sampling operation to equation (6.12) yields
X-I -X+l
'ifo(Z) = 1 Z-k U ~Ok(Z) + ~ t1~o.(Z) (6.14)
k~O Z- 1
where
(6.15)
If the response at the next set of sampling instants, that is, T. T + Ta•
T + 2T(; . ... is desired, it is obtained by two different ways as follows:
;\'-1 Z-.\·+1
~'I(Z) = Zs[zi{(z» = l: Z-k.('.[Z t1~k(Z») + - - a~ls(Z)
k=D Z - 1
(6.16)
where
(6.17)
or hy the IIlternate and equivalent form (since no discontinuities at the
·.alllplillg inslanls exist)
.\' I Z :v II
~ /. l·~.It\'(.;.(zU I ' N(.' (Z) (6.IH)
J II Z 1 h
where
(6.19)
and
(6.20)
The responses at other sets of sampling instants can be obtained in a

similar fashion. The relationships (6.16) and (6.17) with 6.18 through 6.20
will be utilized to arrive at two equivalent forms of the fundamental
equation for the limit cycles.
At this point. the assumption is made that the transfer function does
not contain any integrator and that all its poles possess negative real parts.
This restriction on KG(s) will be relaxed and KG(s) with a single integrator
will be discussed later. Then it is observed that neither 6.~Ok(Z) nor
6.~o.(Z) possesses poles at Z = I because neither KG(s) not E~is) has
poles at the origin of the s-plane. Hence the first term under the summation
in equation (6.14) has no pole at Z = I. and the last term has a single
pole at Z = I.
Now the final value theorem is applied to equation (6.14) to obtain the
output co. as follows:
Co. = lim (Z - 1) ~o(Z) = lim Mfo.(Z) (6.21)

Z-l 7.-1
In general. the final value of <if;(Z) is denoted as Cis and is given by
CiS = lim (Z - 1) ~I(Z) = lim 6.~i'(Z)

Z-1 7.-1
i=0.1.2..... M-1 (6.22)

This is one form of the fundamental equation to be used for the limit
cycle analysis. Another form which is very convenient consists of
equations (6.18) through (6.20). Following the same manipulation used
in (6.14) and (6.15), we finally obtain
i = 0, 1.2•...• M - I (6.23)
It may be noted that H; I ,.,,(.f) = I~'; I i AI.,,(s). "hen I -i ; > M.

The response between sampling instants is derived by applying the
final value theorem to fI;(Z, m):
;= 0, \, 2, ... , M - \ °: ; m ::;; \ (6.24)

When no discrete processing units are included in the plant, equation
(6.23) can be more conveniently and simply written as
; = 0, \, 2, ... , M - I (6.25)
We note that the skip sampling process is removed, and instead we use
the regular z-transform with respect to the global period Ta = MT.
The evaluation of equation (6.25) can be performed by using the integral
theorem which gives
'·i. = lim
1._r·"r·_l
<Ttl { e,\!$T [ !
,U-l
,~O
e-lsTKG(s)E;+i.h(S)
]}
Z-1
= t·"r'~1.-1
lim -
Z-l
21Tj
i I'
,\1-1
e,l1,,7'KG(p) ~ e-'"TE' . (p)
~
';0
1 _ e-.1ITI<-",
I+ •• h
dp,
.-0
i = 0, 1,2, . , , ,M - 1 (6.26)
The multiplication of the expression within the bracket by e·lh7' is done
to insure the convergence of the integral along the infinite semicircle in
the left half plane, This does not cause any change in the final result; we
compensated for this by multiplying by Z-I.
, We can interchange the limiting process and the integration if we take
the path of the latter sufficiently close to the imaginary axis of the p-plane.
lienee equation (6.26) reduces to
,\1-1
\ f;',llpTKG(p) ! e-I'PTE!+i.h(P)
(./. -__ r
21Tj J..
,~O
I - e!l1'PT
dp
'
; = 0, \,2, ... ,M - 1 (6,27)
Furthcr simplification uf this equation is possible, when the oscillations
lll'l'sYlllllIt'trinli. SYIIIIIlt'trical (lsciliuliull rcsults whcn the same shape or
IIsl'iliatioll:o. b rqwatcd lilr cwry 11lI1!' pcriod or thc limit l:ycle with an
"PII'I',ill' ~il',II,
If we let M = 2f.t (f.t = number of samples in the half period of limit

cycle) and
for I = 0, I, 2, ... , f.t - t (6.28)
then, since E/+II+U'(s) = -E/ H.is). equation (6.27) can be written as
or
1'-1
_I el,uTKG(p) ~e-">T£i_;.II(P)
c - -
II - 27Tj
r
)1'
1=0
1 + eP PT
dp (6.29)
We note that half the zeros of the denominator of the integrand which
include the zero at the origin are canceled by the zeros of (I - rIll>'/') in
the numerator by introducing the symmetry condition. In some cases the
symmetry condition requires that the summation of e;.h(O) over one global
period is equal to zero. It should be noted that the elimination of the
pole at the origin simplifies consideration of the analysis of the system
when KG(s) contains an integrator. The symmetry condition is a valid
assumption for relay and PWM systems when the plant has an integrator
and low-pass filter characteristic as is usually so.
In summarizing the preceding, equation (6.26) is a general form to obtain
the response at the sampling instants when the system is in the limit cycle.
Tt could be used for either free oscillation or forced oscillations (when
the forcing function is periodic with the period an integer multiple of the
sampling period). Equation (6.29) is a simplified form of (6.26) when the
symmetry condition exists. It yields the oscillations with zero doc com-
ponents* (because the constant value of the limit of the summation in
equation (6.18) is ignored).
6.2 Application of the fundamental

equation to specific examples 22
In this section we shall apply the theory to three examples representing a
pulse-width modulated feedback system, a nonlinear gain amplifier, and
a quantized level system. In the first two examples, symmetry condition
is assumed, whereas, in the last, asymmetrical oscillation is considered.
• Thc d·c component implies the averagc vaillc (lr Ihe response at the s:lmplinl: ilhwnl'
during onc period or Ihe limil cycle.
PERIODIC MODtiS OF OSCI LLA liON IN NONLINEAR DISCRETE SYSTEMS 197
r(t) + e(t)
FIGURE 6.S Pulse-width modulated system.
Pulse-ll'idth modulated feedback system

In this system as shown in Fig. 6.5, the signal pulse-width modulation is
indicated in Fig. 6.6.
I - e-&hr
E:'h(s) = y(1) s = 9'[e;.I.(t)1 = 9'[y(l)u,,(t)1 (6.30)
where
el(O)
1'(/) = le,(O)1 (and this takes only the values +I and -I)
and
h, = oc le,(O)1 = ~ Ir,(O) - c,(O)I, for oc le,(O)1 ::s; T (6.31)
Substituting for E':h(S) in equation (6.29), we readily obtain for this case
-I
= -27rj
1. el'PTKG(p) I'
I
1 y(l + i)e-",7'(I-e' "h, .. )
dp (6.32)
1 + ell"T
C
IS I' 1=0 p
,,(t). ". (t): input and sampled input

10 controller
",. A(t): oulpul uf I:olllroller
Ht ilIIU f •. f, 1111'"1 111111111111'111 IIrJ'WM '·IIIII'ull,· ...

T=1
To = 4T= 4
KG(s) = s(s~J)
oc = 0.5
rn: +1. +1, -I, -1
FIGURE 6.7 Phase plane analysis of PWM system (M = 4).
en'(O)
(Jd = 1
ell (0)
-----'-------- -1
FIGURE 6.8 Chlll'llctcrillic!lf IlIlurllllng IIl1lplillcr,

We assume for this example that KG(s) == Kls(s +
b), and the input has
a period of four examples, ro, r l , -ro, -r1 • Since the input is symmetric
and KG(s) has an integrator, we can assume that the output is also
symmetric. So we have
p. == 2, yeO) == y( I) == + I, ,,(2) == ')'(3) == - 1 (6.33)
Using the equation for ell and noting that
,,(1 + i) == + I, for I == 0, I, 2, ... , p - I - i
== -1, for I == p - i, P - i + I, ... , p -
1
(6.34)
we obtain for the response at the sampling instants on the limit cycle,
c = _ K{ho + hi _
o. b 2
e- bT
b(l + e- 2bT )
[e-1J7'(~/)lIo _ I) + (e blll - l)]}
(6.35)
C == _ Kfh l - ho _ e-b7' [e-bT(ebhl _ I) _ (ebhO _ I)]}
10 bl 2 b(1 + e-2bT )
(6.36)
From equation (6.31) we know that
(6.37)
These equations can be solved for ho and hI and consequently for cos and
(' ...
When b == I, K == I, ex == 0.5, T == I and ro == r 1 == + I, eO. and e..
can be solved numerically to yield
co. == -0.238 CIs == -0.162 (6.38)
The limit cycle obtained in the phase plane of Fig. 6.7 verifies these
calculated values.
Nonlinear gain amplifier21 ,92

For the nonlinear gain amplifier, a saturating amplifier with linear zone
is chosen, as shown in Fig. 6.8. The use of this nonlinearity in a discrete
l't'cdback system is shown in Fig. 6.9.
The description of the nonlinear gain is the following:
(1;.(/) == I, < e (0)
d 7l
= (lell(O), -d < e.. (O) < d

... -I, (1,,(0) < -d (6.39)
We cXlll11inc the mode M .... 4, where the output of the saturating
IllIIpliUer iN the Ne'luence I, k, -I, -k. Fur this cUlle the resrnl1se falls in
FlGURE 6.9 Saturating gain sampled-data system.
the saturated region at the first sampling instant and in the linear region
at the next sampling instant. The same feature is repeated with opposite
sign. No input is applied.
The plant is
KG(s) = t (6.40)
s(s + 1)
The sampling period is T = 2, (J = 2, and d = 0.5. .
The outputs of the saturating amplifier are assumed as follows:
e~iO) = 1, ei,(O) = k, e;,(O) = -1,
e;'(O) = -k, Ikl <1 (6.41)
For a saturating amplifier the pulse width h is equal to T, since there is
no pulse-width modulation and therefore the pulsed outputs from the
hold denoted as ej,h(t), j = 0, 1,2,3 will have the following Laplace
transforms:
t - e-,T
EOh(s)
, = s ,
-IT
(6.42)
I -e
E2h(S) = - ,
. s
This sequence of oscillations is symmetrical and equation (6.29) can be
used for the solution provided T is substituted for h. We ultimately get
the following response:
k[ , t - e- bT -bT
Co. = - -2b T(eo. + e lO ) +
I I
2bT (eo,e
b(l + e- )
+ e ts, )]
(6.43)
Substituting the given numerical values. we get

Co.= -0.885 - O.ISOk «(l.4S)
Ch = 0.1 SO - O,HHSk «(I.4ll)
Since it is assumed that the system is operating in the linear region at
the sampling instant corresponding to Cis' we get from equation (6.39),
(Jels = -pc IS = e;.(t) = k (6.47)
or
k
Clr =- p (6.48)
Substituting this into equation (6.46), we obtain

k = 0.390 < 1
Using this value of k, Co. and Cis can be calculated.
Co. = -0.943 < -d (6.49)
Cis = -0.195> -d (6.50)
The results show that co, falls in the saturating region producing unity
as the output of the saturating amplifier and ('10 falls in the linear region
producing k (= 0.390 in this case) as the output. Thus the assumption of
the sequence I, k, -I, -k is verified and the existence of such a limit
cycle of the mode M = 4 is proved.
Quantized levt'l amplifier

In this example, the nonlinear gain amplifier has a characteristic of two
tJlHlntized levels as shown in Fig. 6.10. Such a quantization appears in
"i~ital control systems where analog-to-digital converters are used.
e,,'(O)
Jolt illlt! ',III ('hllllll'lrl'~lIl' IIr 1111111111/1 •• 1 1,'\1'1,11111'11111'1

i:
co
-1
KG(a) =O(.~ 1)
T=2
a=3
+ 1. O. -1
ell':
FIGUR.E 6.11 Asymmetrical limit cycle of quantized level amplifier system.
The characteristic of this quantization is represented as follows:
e~(t) = 1 d < e,,(O)

=0.5 -d < e..(O) < d
2
=0 - -d2 < e"(0) <-d2

d
= -0.5 -d < e..(O) < - '2
= -1 e,,(O) < -d (6.51)
wher~ «d = 1 and the linear plant is the same as the previous example.
The existence of an asymmetrical oscillation with M = 3 can be
ascertained for this case,
We let M = 3, and assume that the output of the nonlinear component
+
has a sequence of I, 0, - I: hence ,,(0) = I, ,,(2) = -I. SlIhstitliting
these values in equation (6.26) with the preceding characteristics and
letting h" - T, we obtain
K[2T 1 - e-II~7' ]
co. == - b "3 - b(l + e-b7' + e-2b7') (6.52)
Cli ==
K[T"'3 -
b b(l
1 - e- b,.
+ ,-/}7' + e-267')
]
(6.53)
CIs
K
= b "3 -
[T b(1
e- b7' - e- 26 7'
+ e-b7' + e-2b7,)
]
(6.54)
Letting T == 2, b = I. K = I, and d = ! (0( == 3), these values yield
co. == -0.482, Clo = -0.083, Ch = 0.565 «(,.55)
The outputs of the quantized level amplifier corresponding to ('1,•• ( ....
and C2. are + I, 0 and -I. which satisfy the original assumption of Iht'
sequence. This limit cycle is shown in Fig. 6.11.
In another example of asymmetrical oscillations, if we let d = A (IX ,,~ ;")
and assume the sequence + I, 0, -~, o. A. 0, -I as the outpul of Ih,'
nonlinear component. then a limit cycle of M = 7 is sustained. us shown
in Fig. 6.12.
Extension of the method to other types of nonlinearities such liS n'lny
sampled-data systems is feasible and could be handled by Ihe lillm"
procedure. A discussion of the maximum period and the stability of Iht'
limit cycles obtained follows.
6.3 Limitation on the period of limit

cycles of relay mode oscillations22
In this section, a derivation of the fundamental equation that yicldli Iht'
limits of modes of oscillation is obtained. It is based on the USSUl11plioli
of symmetry of the oscillations and is valid for (autollomuus) I'dll)'
discrete systems. The equation for the response during the limit l'yd,'
c:ln be obtained from equation (6.32) of the previolls sect inn hy replllt'i"",
all the hili by T. Thus
(', = __1_
,. 2'".1'
f. I'
e,·,,7·KG(p) ,. " y(1
1 + (''"''/' Il II
+ ;k 1117'(1 -
P
,: ,':")" ,
I
(fl, "Itl)
From the liymmctry cunditiun. (';.(0) is positivc fur Ihc lin,t half 111'1'1",1
alld I1c~ativc fur the olhcr hillf pCl'iml. This implies Ihlll Ih,' 1'1'\1'"11'0"
(',,(0) is nCAalivc fur Ihe lirsl hull' periml 111111 pm.ilivc ful' Ihe swuml hlill
o;illl'l' 1111 illplll is lIpplit·d. Th,'rt'forc
)'(/) I I ,.. II. I. 2•...• I'
I , - 1', /' , I,.... At
KG(a) = a(,~ 1)
T=2
a=2
e,,': + 1,0, -0.5, 0, +0.5,0, -1
FIGURE 6.12 Asymmetrical limit cycle of quantized level amplifier system.
and
c" <0 I = 0, I, 2, ... , I" - I
(6.57)
C,. >0 I = ~, I" + I, ... , M - 1
By investigating the polarity of C,,_l,. which is the response at the last
sampling instant contained in the first half period of the limit cycle, we
can obtain the maximum number I""'"X or M IIlIIX • This is done by letting
i = I" - I in equation (6.56).
c _ = - -
pi,.
I
f.
e
I'pTKG() (I
p
2TTJ. r 1 + ~..MilT
- e
P
-liT) p-l
'" (I + _ 1)e- 11I7' d
~Y
1-0
I" P
(6.58)
Since Co. is negative by the original assumption, the polarity of C,•.. l,. is
tested starting from I" = 2, that is, from C II • The response CI , 1 •• may be
negative up to a certain I" = 1/' but it may become positive f()f all I" ~
~' + I. Then this critical 1/ is tuken us /"11111.'
This general formulation and procedure is applied to the following
examples.
EXAMPLE I
The plant transfer is of first order, that is, KG(s) = X/(s + b). The
response C,._I •• at the margin of the first half period is obtained by letting
i = P, - I in equation (6.34) to give
,,(1 + p,) = +I for I = 0
= -I =
fori 1, 2, .•. , p, - 1 (6.59)
and
Co' C1,cp _ 1
C2 • •••• <0 (6.60)
Equation (6.58) becomes
. _ 1 J.
e""TKG(p) J - e-"T(J
-k e
p~1 -'fiT) d
p
21Tj (. I + e""T
C -I
I' o· - - -
p 1=1
K 1 -liT
= _ - e [I + e- IIT + ... e-lp-2111T _ e- lp - lIbT ]
b 1+ e-pllT
(6.61)
It is evident from equation (6.61) that C,._I.R is positive for all p, ~ 2.
thus violating the assumption of equation (6.60). Therefore it can be
O
concluded that the first-order relay sampled-data system can maintain

only the relay mode of oscillations of two sampling periods (M = 2 or
I' = I) and that longer periods cannot be sustained. This holds for any
K. b, and sampling period T.
EXAMPLE 2
We shall consider a second-order plant with integrator in this example.
that is, KG(s) = K/s(s + b). Letting hi+; - T in equation (6.29) yields a
set of solutions Co,. Ch • ••• , c,,_1.. for the responses at all sampling instants
ur the limit cycle.
C = _ K {p-I-l [I _
e-1p-I-lIIIT(l - e-bT )]
i. b ,~ 2 b(1 + e-pll ',)
_ 1'-1 [I _ e-lp-l-mTo - e- IIT )]}
,J-i 2 b(1 + e-pl'T)
(6.62)
I,'or I' = I and i = P, - 1 = O.

('II. = - h 2" -
K[T I - rb'r ]
h(l + eO h7')
_ _ 0 1\ ,.".n I + I' h7') - 02( 1 -:'. I! 0"7')1

2/,- I + I' ,,',.
It can be proved that the numerator inside the bracket is positive for
all bT > O. Hence it may be concluded that COs is negative for all T.
for I-' = 2 and i = I-' - 1 = I,
K (I _ e- bT )2
Cls = - b2 1 + e- 2bT < 0, for all bT> 0 (6.64)
Therefore Cb is negative for any sampling period T.

for /l = 3 and i = /l - 1 = 2
C
28
= K
b 2
[I _ (1 - e- bT ) (t
b(1 +e- 3bT )
+ e- bT _ e-2bT)] (6.65)
Simple trial shows that c2• is negative for small T and becomes positive
for larger T, thus violating the original assumption. We can find the
critical Tc by solving equation (6.65) for T when C28 is equated to zero.
Therefore the oscillations of /l = 3 or M = 6 can exist for T < To, but
it cannot be sustained for T ~ Te' A similar observation is obtained for
larger /l.
To derive the fundamental equation that relates the maximum /l and
the sampling period T for any second-order system, we let i = /l -1 in
equation (6.62).
C =_K {! _
e-(p-UbT(I - e- bT )
p-l .• b 2 b(1 + e-PbT )
_"-l[I _ e-(p-l-l)bT(l - e-bT)J}
. ~1 2 b(l + e- PbT )
= K[bT( _ 2) _ ebT 1 - e- IlbT + ebT - IJ (6.66)
b2 2 '" 1 + e- IlbT
By the original assumption, e,l-l. s < O. Hence /In,,,x is obtained by
finding the maximum integer /l which satisfies the following inequality:
bT t -pbT
- ( - 2) - e- bT - e + e"T - 1 < 0 (6.67)
2 /l 1 + e- IlbT
To conclude, we may state that limit cycles of two and four sampling
periods can exist in the relay system for any sampling period T, but that
for '" ~ 3 the maximum number of sampling periods or the mode of
oscillations is restricted by the system parameters band T. figure 6.13
indicates /lmax as a function of bT.
These discussions can be applied to any order plant and to other types
of nonlinearities. 20 - 22 An important application of the maximum number
of modes of limit cycles is in the stability study of nonlinear discrete
systems. It yields the boundary stability curves for various nonlinear
discrete systems.
PERIODIC MODES OF OSCILLAnON IN NONLINEAR DISCRETE SYSTEMS 207
10
9
8
7 KG(s) = af.!61 Relay mode oscillation
6
5
4
3
2
1
O~__~~~~~~~~~____~~__~~~~~~
0.1 4 6 8 10 bT
FIGURE 6.13 Maximum number of sampling periods contained in half period of
limit cycle.
6.4 Stability study of limit eyeles S,so,II7

In the preceding section a method was presented for obtaining the periodic
oscillations (or limit cycles) of a system without information as to their
stability. We shall now discuss the stability by first defining a stable
limit cycle.
A limit cycle is considered to be asymptotically stable if any small
perturbation around the limit cycle approaches 0 as n approaches 00.
Dy following this definition, the following two theorems can be applied
for testing the stability.
TnEOREM 1: Assume that the nonlinear difference equation in vector
form
z ..+11 = F(z..)1 (6.68)
has a periodic solution s..1of period M, and that the functions F(z..>are
single valued and possess continuous first partial derivatives. The first
approximation of the difference equation for small perturbations about
the periodic solution s.. ] is
Y.. I1] = [A .. ] Y.. ] (6.69)
lII,d the solution of this equation determines the stability of the periodic
solutiol1 ''',,) if [A II] ill nonsingular at all the solution points. In this equation
r,,1 is the pcrturhatiol1 ;lhout ..... ] and the components of [An] are
"".11 -
iW,(;l' .. ) I
o.r, ... 1 •• 1 (6.70)
THEOREM 2: The system of equation (6.69) is asymptotically stable if all

the eigenvalues of the matrix
.11-1
[Ao] = n [Ata] = [A,1I-1][A.v - a] ... [Ao] (6.71)
lie inside the unit circle. Then all the solutions tend to 0 as n becomes
large. If one of the values lies outside the unit circle, the solution is
unstable. Finally, if all the values lie on the unit circle, the linearized
solution Y ta ] of equation (6.69) can be considered as stable. *
Combining the two theorems, the condition for the asymptotic stability
of limit cycles is reduced to the following statement. All the eigenvalues
of the matrix [Aa] which is the product of all the matrices which consist
of the first partial derivatives of F(xn) lie inside the unit circle at all the
periodic solution points sn].
The application of this theorem is shown in the following discussion.
6.S Forced oscillations in nonlinear

discrete systems4.23.27
In this section, a particular example of the nonlinear sampled-data system
discussed in the preceding chapter is analyzed for a sinusoidal forcing
function. Since only harmonic oscillations (oscillations having the same
frequency as input frequency) are considered, it is necessary to put suitable
restrictions on the parameters of the system in order for such oscillations
to exist. Furthermore, the ratio of the sampling frequency to signal
frequency is assumed to be an integer.
EXAMPLEu
Consider the sampled-data system shown in Fig. 6.14, where G1(s) =
K/s(s + I) and r(/) is a sinusoidal input. The nonlinear characteristic is
described by
(6.72)
where c' is the output of the nonlinearity and
~(z) = <[I - e- T • 01(5)J = K(0.368z + 0.264) ,

s (z' - 1.368z + 0.368)
T= 1 (6.73)
• It should be noted that even if the linearized solution of the perturbation is .fluhle-,
the limit cycle is not asymplolirQlly IIIQbl, since the perturbation will not decay in thill
case.
T..-
r ........ ~ ---<)
r(!) + I ~(z)
crt)
G(8)
Nonlinearity
F[G URE 6. [4 A nonlinear discrete system with nonlinearity in feedback path.
The difference equation relating output to input is easily derived as

follows:
e(n + 2) - 1.368c(n + J) + 0.368e(n)
= K[0.368e(n + I) + 0.264e(n)]
= K{0.368[r(n + I) - e'(n + I)] + 0.264[r(n) - e'(n)]}
(6.74)
Choosing rn3 = -0.1 and substituting for c' in terms of e, the following
difference equation is obtained:
('(n + 2) + ('(n + 1)[0.368K - 1.368] + e(n)[O.264K + 0.368]
- K[0.0368e(n + 1)3 + 0.0264e(n)3]
= O.368Kr(n + I) + 0.264Kr(n) (6.75)
If the peak value of r(n) is not chosen very large, it can be assumed that
the harmonic oscillations will be the dominant term in the periodic
response. Let r(n) = B sin (17/3)n. For the periodic steady-state output
the following form can be assumed:
c(n) = pen) sin:! II + q(n) cos:! 11 (6.76)

3 3
where pen) and q(n) are slowly varying discrete functions of n. We sub-
stitute for r(n) and c(n) in equation (6.75) and equate coefficients of
sin (17/3)1/ and cos (17/3)n, noting the fact that sin 3 () = 1(3 sin () - sin 3()
;tnd l'OSa (J = .\(3 cos 0 + cos 30). Since pen) and q(n) are slowly varying
wit h II, it can he assumed that .
+ 2) = 21'(n + I) - pen) (that is, !J.2p(n) = 0)
p(" (6.77)
and in knlls invlllviuA "'3' 1'(1/ + I) could he replaced by pen). These
"'lIIplitkatilllls arc It~J~itillwtc so lunA as t."onsidcration is restricted to the
1lIlI'IlIOllil~ IIM·illatillns. With these as.,lIl11ptillns. the comparison of
coefficients of sin ('TT13)n and cos ('TT13)n in equation (6.75) yields the
following two relations:
pen + I )(0.184K - 1.684) + q(n + 1)( -0.547 - 0.319K)
= p(n)( -0.868 - 0.264K + 0.0336A2K)
+ q(n)( -0.866 - 0.0239KA2) + 0.448B. (6.78)
pen + 1)(0.547 + 0.319K) + q(n + I )(0. I 84K - 1.684)
= p(n)(0.866 + 0.0239KA2) + q(n)( -0.868 - 0.264K
+ 0.0336A2K) + 0.3185B (6.79)
where A2 = p2(n) + q2(n). These are the key equations in the study of
periodic solutions and their stability.
After the transients have disappeared, the solution will tend toward
one of the periodic states given by the substitution of pen +
I) = pen) =
Ps and q(n + I) = q(n) = q.• in equations (6.78) and (6.79).
Define
0.184K - 1.684 -0.547 - 0.319KJ
[ =C (6.80)
0.547 + 0.319K 0.184K - 1.684
and let
[pen)] = yen) (6.81)

q(n)
Equations (6.78) and (6.79) are then rewritten in a simplified notation as
Cy(n + I} = H[y(n)] (6.82)
where H[y(n)] is the nonlinear term appearing on the right-hand sides of
equations (6.78) and (6.79). Equation (6.82) is further simplified as
yen + I) = C-IH[y(n)]
= F[y(n)] (6.83)
The equilibrium points of this equation are given by
; = F(;) where; = [;:J (6.84 )
which reduces to the following two equations:.

P.,( -0.816 + 0.448K - O.0336KA2) + q,(0.319 - 0.319K
+ 0.0239KA2) = 0.448KB (6.85)
p,(0.319K':"- 0.319 - 0.0239KA2) + l/.(-O.KI6 + 0.44KA·
- ().O:n6KA2) ,..~ O.JIII5KII (6.IIlI)
0.20 .---r-r-..--r--..----,r--....,...---,r--.....---,r---r---.
FIGURE 6.15 Amplitude characteristic curve ror the harmonic oscillation.
These equations when solved for p,. q. and using the fact that p! + q! =
A2 yield the amplitude characteristic
A2[(-0.816 + 0.448K - 0.0336KA2)2 + (0.3 19K - 0.319
- 0.0239KA2)2] = 0.448K2A2 + 0.3 I 85K2B2 (6.87)
For a given value of K the functional relation A2 versus B2 can be plotted.
It will be seen that for some values of K the system has more than one
periodic state for a certain range of values of B2. Figure 6.) 5 shows B2
plotted as a function of A2 with K = 2.395. Computation shows that for
values of K greater than 2.25 multivalued regions are obtained.
. Let us choose K = 2.395. This will make the poles of the linear part
IIf the difference equation just lie on the unit circle. The characteristic
"'Illation for the linear portion is
Z2 - 0.488z + 1= 0
which ~ives
z= 1)(' 1;0, where p = 1 and () = 217 (6.88)
5
rhi .. I!,iws th~ period as 271'/()= 5 secnnds. The period of the sinusoidal
11111111 is (, s""ulld!o alld since the dillerencc between the natural frequency
IIlId "11~1I111 rn·'IIIt·n,·y h. !;mall, II Inckin~ phcnnmcnnn or 1I frc'lucncy
"111 rni II II Il'II I IlIk,·., plm',' 1\ plW1l1l1UC1I1I1I !;illlilar 10 lhlll in cnlllillllnlls
!o'ystems. Calculation of the response using the difference equation as a

n'currence relation indicates a response of period 6 as expected.
The nmplitude characteristic has a region, for values of A2 between
1..25 and 4, where dB2/dA2 < O. From the theory for nonlinear continuous
'y~tl'IlIS, in particular, works of Hayashi,38 and more recently, that of
KIlIttcr and Pinney,2. we would expect the periodic solution in this region
III h,: unstable. That this is true is proved by considering the variational
"'Illalilln around the periodic state.
1,'1 the components of the vector F[y(n)] in equation (6.83) be denoted
!ly I/~
I:!
I If the eigenvalues of the matrix
[~ ;~l]
of2 of2
(6.89)
op oq If("1 ~~
III' III\j,k the unit circle, then the equilibrium point is stable; otherwise
II 1\ IIn,lahlc. For the region in Fig. 6.15 where dB2/dA2 < 0, the equilib-
1111111 point is ullstable and if the system is slightly disturbed at this point,
II Will IIltimalely ~ettle down to one of the other two stable periodic
·... 111111111',. III Ihc numerical example discussed in this section, and for
"V. 111.' 'i from Fig. 6.15, three periodic states with A2 = 1.495, 3.25,
111111 oI.M art' IIhtaincd. From equations (6,85) and (6.86), p" q, are com-
I'"h'" 1111 "u\'h III' these values and then the characteristic equation for the
\'111111111111111 "'lll:Iliull at each point is computed. Stability of this character-
I~III" "'11111111111 is thcll examined. A summary of the results are given.
". 2.395
/I .. O.J54
Chara(teristic Equation
I i(IIIII", illlll I·"i"'.~ for Variational Equation Stable or Unstable
/', I.UI.! 'I. -, -0.69 ).2 _ 1.793). + 0.809 Stable

/', LUI ~ 1/•.. -1.5 ).2 - 1.738). + 0.73 Unstable
1'•. IUI'IO'i 1/.. : -2.2 ).'1. _ 1.688), + 0.706 Stable
I hl'~I' ,·,'M.lls Wl~rc arrived at through lahorious numerical computation.
II I', '.1"'11 that t h(: milts (If the characteristic equation lic very ncar the
111111 1"111·1,', IICIIl't: it is til he cxpccted that the periodic solution will setth:
.1"wlI III .1\ !lIt'ally-stOlte vOllllc aftcr II vcry ImlA lime. Thc c\ulIcness uf the
1II11II(',"'al valill's i~ lllll'ihlltCtl tn thl' sllIall slope tlrtht~ amplitude churactcr-
."1",, 1;111' "XlIlIIl'lc, fill" Ilw inl'"t mllfl,llitllde of II ..,. U.294, which IlIlI; Ullly
one stable periodic solution. the characteristic equation is A2 - I.816A +
0.86 which is clearly stable.
6.6 Direct z-transform for

determining true oscillation 23
In addition to what has been discussed in the preceding section. other
methods exist for determining oscillation in nonlinear discrete systems.
These methods constitute both graphical procedures such as the phase
plane (space) and the describing function. and analytical procedures such
as the use of orthogonal discrete frequency functions or Fourier poly-
nomials. the state-vector approach. and the direct z-transform. In the
following. we shall discuss only the latter method, which is closely related
to the contents of this book; however. the other techniques are important
and have been extensively elaborated in the literature.
, In the direct z-transform method, the type of nonlinearity assumed is
either a relay or a saturating gain amplifier with straight-line characteristics,
as shown in Fig. 6.8. The method is based on specifying a certain mode
of oscillation for zero input and obtaining the conditions that sustain
these oscillations. This technique will now be explained and later applied
to a specific example.
Let the output of the nonlinearity be assumed as in Fig. 6.16, where
the period of oscillation, for sake of illustration, is assumed to be four
sampling periods, the sequence being mo, mil m2 , m3 • In the steady state
the z-transform of this repetitive sequence can be defined as
mo + m1z- 1 + m~-2 + m az- 3
(6.90)
1- Z-4 1- Z-4 1 - Z-4 1 _ Z-4
The output of the system with a repetitive sequence Co. Cl l C2' C3 has
similarly a steady-state z-transform of
(6.91)
"I( iI 11(1, f, If, 1111'111 11111 pili 'It-~clil)liun u" Iim'lIr 1'111111 IInder limit cycle npcrntinn.
FIGURE 6.17 Mathematical equivalent of system in Fig. 6.18.
Next take the z-transform of the difference equation describing the linear
plant and substitute the z-transform expression (6.90) and (6.91) for the
input and output, respectively. Comparison of coefficients of like powers
of z will yield four linear relations involving Co, Cl. C2. C3. It is found that
not all these relations are linearly independent. For the limit cycle with
period four. only three relations are linearly independent. The fourth
relation appears as a constraint which should be satisfied so that the
specified output of the nonlinearity is maintained. The solution of these
equations will yield bounds on the initial values of the output which are
the conditions that are necessary (but not sufficient) to sustain the specified
mode of oscillation. The following example will illustrate the procedure.
EXAMPLE
Let Fig. 6.17 be the mathematical equivalent of the sampled-data system
shown in Fig. 6.18. Let Gl(s) be
Gl(s) = 0.3055 + 0.695 (6.92)

s(s + 0.695)
and the nonlinearity be an ideal relay as shown in Fig. 6.19. The z-
transform of Gl(s) is
~l(Z) = <r - se- TS Gl(S)] where T= 1
0.5z
=------ (6.93)
(z - 1)(z - 0.5)
Hold
circuit
~~----~-----------
eft)
FIOURI'tl.IM A !lUtllilll"lI!' dl\l·l'l"ll' ~y!lh·1I1.

Thus the difference equation for the linear portion can at once be written as
c(n + 2) - ) .Sc(n + 1) + O.Se(n) = O.5e'(n + 1) (6.94)

where n takes on integer values 0, I, 2, ....
It is required to find initial conditions corresponding to periodic
solutions of period two.
The z-transform of equation (6.94) in the steady state is
(z2 - I.Sz + O.S)C/f(z) - Z2C(0) - zc(l} + I.Szc(O)

= 0.5z4"(z) - O,5ze'(0) (6,95)
In the steady state, let the relay output
be -I, + I, and output of the system be +1
Co, c 1 • Then the z-transform of relay output
in the steady state is given by
-1 + Z-1 -z I'() (6.96)

1- Z-2 1- Z-lI =Z+I= %
The z-transform of the output C(/) is

-1
(co + C1Z-1)/(1 - Z-2) = tjf(z),
FIGURE 6.19 Ideal relay.
We substitute these expressions in equation
(6.95), clear the fractions, and equate coefficients of like powers of z.
Coefficients of Z3 and Z2 drop out and the coefficient of z and the constant
term yield identical relation
(6.97)
If the relay output is to be maintai ned -I, + I , it is necessary that - Co < 0

nnd -C 1 > 0; that is, co> 0, C1 < O. Combining this with the relation
(6.97) between Co and c" it is immediately seen that - ~ < C1 < 0 and
O· < Co < l, which agrees with the results obtained using other methods,
6.7 Periodic solution of certain

nonlinear difference equations 27 ,37
111 the preceding sections, the z-transform method has been applied to
nhtain the pcriodic oscillations of nonlinear discrete systems. The systems
d iscusscd lIrc those lIsually encountered in control engineering applications.
Thc mathclllatical description ()f these systems represents the field of
1I11111illl'ar dif1i.on~lll·c t:tluatinlls. Hcnce the ::-transform as shown in this
dmpler call he readily lIsed to ohtilin the pcriodic solution of certain
nnnlilU'l\r tlilll'n'lIt't' t"IIIOIlillll!l. The Iype nf dil1'crence Ctl"ution that cun
be solved by this method is

, I,
'I,ptc(n
t-O
+ k) - 'I,«t!(n
t-O
+ k),
with 11 ~ I and n - 0, 1,2, . .. (6.98)
In equation (6.98), the nonlinear term/en) represents
fen) - 4>[c(n)] (6.99)
where 4> is a nonlinear single-valued symmetrical function and «t' PIt are
constants.
In applying the z-transform method, we can establish a set of nonlinear
algebraic equations which give the general solution for any period. In
particular, solution for the periodic oscillation c.(n) can be determined.
If we restrict the symmetrical solution to that for even periods only, we
can obtain the exact solution for periodic modes with half-periods less
than or equal to two. For higher harmonic oscillations approximate
solutions could be obtained. A detailed discussion of the solution of
equation (6.98) is beyond the scope of this text; however, the method that
can be used is similar to that applied in the preceding sections. Further-
more, the stability study of the periodic solution of equation (6.98) can be
readily studied as discussed in Section 6.4.
In concluding this chapter, it should be indicated that the z-transform
method is also applicable for obtaining the periodic modes of oscillation
in continuous nonlinear feedback systems. As mentioned earlier, other
methods exist for detennining the periodic modes of oscillation. The
purpose of this chapter is not to discuss and compare all these methods
but to propose in detail the application of one particular method, the
z-transform for obtaining a solution to these problems.
REFERENCES
1. Kalman, R. E., "Nonlinear Aspects of Sampled-Data Control Systems," Pro~eetJ
ings of the Symposium on Non-linear Circuit Analy.fis, Polytechnic Institute of Brooklyn,
Brooklyn, New York, April 1956.
2. Nease, R. F., "Analysis and Design of Nonlinear Sampled-Data Control Systems,"
WADD Technical Note 57-162, Servomechanism Laboratory, M.I.T., June 1957.
3. Mullin, F. J., "The Stability and Compensation of Saturation Sampled-Data Sys-
tems," AlEE TraM., Vol. 78, Part 1, July 1959, pp. 270-278.
4. Mullin, F. J., and E. I. Jury, "A Phase Plane Approach to Relay Sampled-Data
Feedback Systems," AlEE TraM., Vol. 78, Part 2,1959, pp. 517-524.
S. lzawa, K., and L. E. Weaver, "Relay-Type Feedback Control Systems with Dead
Time and Sampling," AlEE TraM., Vol. 78, Part 2,1959, rr. 49-54, and discus.~i()n by
K. lzawa in reference (4).
6. Cypkin, J., Thrnry of Pulsr S,I'slt'm., (in RUlIslnn), Slnte "re!!s fllr "hysic!l and Mathe·
matical Literature, Moscow, lISSR, 19!'iK.
7. Nelson, W. L., "Pulse Widlh Control of Sampled-Data Systems," Technical Report
T-35/B, Columbia University, July 1959.
8. Kadota, T. T., "Asymptotic Stability of Some Nonlinear Feedback Systems,"
ERL Report Series No. 60, Issue No. 264, University of California, Berkeley, January
1960.
9. Pyshkin, I. V., Self-Oscillation in Pulse-Width-Modulated Systems, Theory and
Application of Discrete Automatic System~ (in Ru~~ian), Academy of Science, Moscow,
USSR, 1960, pp. 134-1 SO.
JO. Andeen, "Analysis of Pulse Duration Sampled-Data Systems with Linear Ele-
ments," IRE Tran.t. Automatic Control, Vol. AC-5, No.4, September 1960, pp. 306-313.
11. Andeen, R., "Analysis of Sampled-Data Systems with Nonlinear Elements,"
Automatic and Remote Control, Proceedings First Int. Con.r:ress IFAC, Butlerworths,
London. 1961, pp. 314-320.
12. Torng, H. C., "A Study of Nonlinear Sampled-Data Systems Containing a Relay,
a Saturating, or a Backlash Element," Ph.D Thesis, Cornell University, February 1960.
13. Torng, H. C., and W. C. Meserve, "Determination of Periodic Modes in Relay
Servomechanisms Employing Sampled-Data," IRE Transacliems 011 AutClmatic Cmllml,
Vol. AC-5. No.4, Septemhcr 1960.
14. Meserve, W. E., and H. C. Torng. "Investigation of Periodic Modes of Sampled-
Data Control Systems Containing a Satul'ating Element," TrailS. ASME. Series 0,
J. Basic En./f., Vol. 83, 1961, pp. 77-81.
IS. Jury, E. I., and T. Nishimura, "Analysis of Finite Pulsed Systems with Periodically-
Varying Sampling Rate and Pulse Width," AlEE, CP 60-866, June 1960.
16. Nishimura, T., "Operational Analysis of Finite Pulsed Sampled-Data Systems,"
ERL Report Series No. 60, Issue No. 279, University of California, Berkeley, May 1960.
17. Jury, E. I., Sampled-Data Cemtrol Systems, John Wiley and Sons, New York, 1958,
Chapters I and 9.
18. Jury, E. I., "A Note on the Steady-State Response of Linear Time-Invariant
Systems to Genercll Periodic Inputs," IRE Proc. Vol. 48. No.5, May 1960.
19. Jury, E. I., "Contribution to the Modified :-Transform Theory," J. Franklin Inst.,
Vol. 270, No.2, August 1960. pp. 114-129.
20. Jury. E. I. and T. Nishimura, "On the Periodic Modes of Oscillations in Pulse-
Width-Modulated Feedback Systems," ERL Report No. 60, I~~ue 328, University of
Culifornia, Berkeley, November 29, 1960; also, Trans. ASME, Series E, J. Belsic E'\I{.,
March 1962, pp. 71-84.
21. Nishimura. T. "Stability Analysis of Pulse·Width-Modulated Feedback Systems,"
LRI. Report Serics No. 60, Issue 353, University of California, Berkeley, April 4,
1%1.
22. Jury, r. I., and T. Nishimura, "Stability Study of PWM Feedback Systems,"
I're~el1lc:d Ht the JACC mceting in Minnesota, June 1963. To appear in the 1'raIlJ· •
.-ISMI:, 1%4 .
•~ I. I'Hi. M. A., "Osc:iIlHtinns in Nllnlincilr Sampled-Datil Systems," AlEE T,.all.t.,
V,.I. HI. l'illI IIIAl'pli~'illinn' and Industry), JallIlilry 19(.3, pp. 350-355 .
.'.\, ('hnl\'. ('. K., "( 'nnla~'I'" S~'I VII"'t'chnni'l1I I "'pinyin!! SlIIlll,lc:d-[);ItIl," A lEE
1111'" , VIII. II, 1'111 I II, 1''',1, PI', 'I M.
25. Klouer, K., and E. Pinney, "A Comprehensive Stability Criterion for Forced
Vibrations in Nonlinear Systems." J. of Appl. M~ch .• Trans. ASME. Vol. 75. 1953. pp.
9-12.
26. Oa-Chuan, Shao, "On the Possibility of Certain Types of Oscillations in Sampled-
Data Control Systems," Automation and Remot~ Control, Vol. 10. June 1959. pp. 85-89.
27. Tsypkin. Y. Z.• "Periodic Solutions of Nonlinear Finite Difference Equations and
their Stability," International Union of Theoretical and Applied Mechanics, Institute of
Mathematics. Academy of Science of the Ukranian SSR. Kiev, 1961.
28. Delfeld, F. R., and Q. J. Murphy. "Analysis of Pulse-Width-Modulated Control
Systems," IRE Trans. A.C., Vol. AC-6, No.3. September 1961. pp. 283-292.
29. Tsypkin. Y. Z.• "Elements of Theory of Numerical Automatic Systems." Automatic
alld Remote Control. Proceedings First Inl. Congress IFAC, Butterworths. London,
1961, pp. 286-294.
30. Jury. E. I.. "Contribution to the Modified z-Transform Theory." J. Franklin Inst .•
Vol. 270. No.2. August 1960, pp. 114-129.
31. Monroe, J., D~f(ital Processe.f for Sampled-Dala Systems. John Wiley and Sons.
New York, 1962.
32. Torng, H. C., "Complete and Exact Identification of Self-Sustained Oscillations in
Relay Sampled-Data Control Systems," AlEE Trans. Vol. 81, part II, 1962. pp. 355-36\.
33. Jury, E. I.. Discussion of Reference (32).
34. Quichet, P. L., "Osci11ations Periodiques dans les Systemes Echantiltonnes Non-
Lineaires," AUlomatisme, No.6, June 1963, pp. 229-239.
35. Kuo, B. C., Analysis and Synthesis of Sall/pled-Data Control Systems. Prentice-
Hall, Englewood Cliffs, New Jersey, 1963.
36. Hayashi, C., Forced Oscillations in Nonlinear Systems, Nippon Printing and
Publishing Co., Japan, 1953.
37. Pasteur, Franceline, "Self-Sustained Oscillations in Nonlinear Sampled-Data Sys-
tems," Master of Science Project, Dept. of Electrical Engineering, University of Cali-
fornia, Berkeley, June 1963.
38. Kaplan, W., "Stability Theory." Proceedings of th~ SympOSium on Non-linear
Circllit AnalYSis, Polytechnic Institute of Brooklyn. Brooklyn, New York, Vol. 4, April
1956.
39. Kalman, R. E., and J. E. Bertram, "Control System Analysis and Design Via the
Second Method of Lyapunov," Parts I and II, Trans. ASME(J. Baliic Eng.). Vol. 82,
June 1960, pp. 371-400.
40. Tsypkin, Y. Z., "Investigation of Stability of Periodic States in Nonlinear Pulse
Automatic Systems," AUlomatic and Remote Control, Academic Press, New York.
Vol. 22. No.6, June 1961, pp. 614-623.
7
z-TRANSFORM METHOD IN
APPROXIMATION TECHNIQUES
The z-transform method discussed in the preceding chapters was applied

for the most part to the exact solution of difference equations or to discrete
systems which are describable by difference or difference-differential
equations. One of the basic advantages of the z-transform method is that
the time response can be obtained by power series or long division method.
This procedure circumvents the evaluation of the poles of the system
transfer function or the characteristic roots of the difference equation.
This advantage could and has been utilized for the approximate solution
of differential equations or continuous systems. Furthermore, the %-
transform could also be used for time domain synthesis of networks to
yield certain responses, or for identifying the linear plant transfer function.
The material of this chapter deals mainly with the various approximation
techniques which utilize the z-transform theory.
7.1 Approximation methods

Approximating a differential equation by a difference equation is not new.
Mathematicians have been in the past and are at present engaged in doing
extensive work toward this approximation, which is called numerical
analysis. With the advent of digital computers, approximation is becoming
increasingly important for programming the solution of differential
equutions. In the following discussion this subject is only briefly indicated
hy the irnportunt techniques for approximating a continuous system by a
discrete mudel.
Tn illustrate nne aspect nf the approximation procedure, assume that
thl' l'nntilillUII!o. !o'y''''1II "hown in Fi~. 7.1 ili represented by a continuous
219
220 THEORY AND APPLICATION OF THE :t-YRANSFORM
input and a transfer function G(s}, and that the output is obtained at all
times. An approximate system using sampled-data techniquesl - 3 is
presented in Fig. 7.2. By using the z-transform or the modified z-transform,
the approximate output of the system of Fig. 7.1 can be obtained. It is
noticed from Fig. 7.2 that an interpolator
x(t)
0x(s) ,
.
I G(s)
.
)0
Iy(t)
y(s)
0
(or a mathematical hold) is used. The pur-
pose of the interpolator is to reconstruct as
closely as possible the continuous function
FIGURE 7.1 Continuous sys- which has been fictitiously sampled.
lern.
Shown in Fig. 7.3a to e are the various
forms of the interpolator function commonly used. The approximation
x,,(t) is the sum of a number of impulse responses, that is,
at.
xG(t) = ! x"q(t - nT) (7.1)
.. =0
where q(t) = .2'-l[Q(S)].

For the cases a, b, d, and e, equation (7.1) applies, whereas for case c,
in which the values of x(/} at midpoints of the interval are used, the
following is used:
CD
xG(t) = ..!x..+lq(t -IT- nT)

~O
(7.la)
If we impose the condition xG(nT) = x(nT) [or N'G(z) = N'(z)] in a, d,

and e, by applying the z-transformation to equation (7.1) we obtain
1[q(t)J = J (7.lb)
The same relationship also holds for case c with the imposed condition
x,,(nT + T12) = x(nT + TI2). These conditions are not imposed on b
where the pulse heights are the average of the ordinates x«() on either
side of the pulse.
If we put .
.2'[q(t)] = Q(s) = Fk+l(Z-l) (7.le)

Sk+l
Interpolator
x(t)
o X x· (tJ I Q(s) I Sa (t) G(s) I
YaW
0
FIGURE 7.2 A slIl1l(lled-dlll8 1I1'I'roxilllllliul1 "r l'ig. 7.1.

Z-TRANSFORM METHOD IN APPROXIMATION TECHNIQUES 221
1.0 q(l)
(a)
q(t)
-T 0 T 2T -T
(b) (e)
(d)
1:I(jtJRE 7.3 Approximalion of a lime funclion.

222 THEORY AND APPUCATION OF THE z-TRANSFORM
Fk+l(Z-l) (a certain function of (Z-l» can be determined by applying the
z-transformation to equation (7. Ie) and using the results of7.lb to obtain
(7.ld)
and
(7.1 e)
The interpolator transfer functions Q(s) which result from equation

(7.le) will be termed normal interpolators of order 0, I, 2, by putting
k = 0, I, 2. For instance, the zero-order normal interpolator transfer
function can be readily obtained by letting k = 0 in equation (7.le) to
obtain
(7. I/)
Similarly, higher-order normal interpolator transfer functions can be

readily obtained.
To summarize the various interpolation schemes shown in Fig. 7.3, we
use Table 7.1, introduced by Cruickshank,' to indicate the transfer function
of each interpolator.
TABLE 7.1 CLASSIFICATION OF INTERPOLATIOW
Interpolator Tran.ifer Function Name of Approximation
(J - z-l)
Zero-order normal (a) Qo(s) = s
Rectangular
21(1 - 21-1)2
First-order normal (d) Ql(S) = TS2
Triangular
2%(1 - %-1)3
Second-order normal (e) Q.t<s) = 1'2(1 + z-l)r Parabolic
z!-{(l - 21-1)
Zero-order midinterval (c) Q(s) = s
Midinterval rectangular
z(1 - %-2)
Zero-order mean (b) Q(s) = 2$
Mean rectangular
~-- ......--..
It should be noted that the second-ordcr norntlll intcrpoilltiol\ C:I!iC (.

gives exact representlltion of plImhnllls, whcreas case tI gives cxact filling
Z-TRANSFORM METHOD IN APPROXIMATION TECHNIQUF..5 223
of straight lines. The mean rectangular and midinterval rectangular
approximations give a relatively coarse fit to any curve except stepped
functions.
Having obtained the transfer functions of the interpolators, we can
write the approximate output z-transform from Fig. 7.2 as follows:
f{!f ,,(z) = !l'(z)Z[ QG] (7.1 g)
For integrating operators of any order n, we can readily obtain from
tables of z-transforms the transform Z[Qr"]. These transforms are
given in the second column of Table 7.2.
By using the division method, the approximate output sequence yo(nT)
can be readily obtained. The normal interpolators may be combined to
represent accurately functions with given initial conditions, which is
explained in the next section.
7.2 Initial conditions nonzero·

The problem of initial conditions can be incorporated by combining the
normal operators. Thus we can have linear or quadratic interpolation,
depending on the number of terms used in the Taylor's series expansion of
x(t) of Fig. 7.1 around the origin. This is illustrated as follows.
Let the Taylor's series expansion of x(t) be presented
+ IX'(O) + -t
2
x(t) = x(O) x"(O) + ... + xm(t) (7.2)
2!
where xm(t) is the remainder after m terms.
Taking the z-transform of this equation, we obtain
~(z) = x(O) + x'(O)Tz- 1
1- Z-1 (1 _ Z-I)2
+ x"(O)T2z- 1(l + Z-l) + ... + ~ (z) (7.2a)

2(1 _ Z-I)3 m
We can replace '?V,,(z) from Fig. 7.2 by the following approximation:
(!VaCz) = X(O)_l Z[QoG]

1- z
+ x'(O)Tz- 1 "[Q G]
(1 _ Z-I)2 '" 1
+ ~"(~)T2Z \(1, ~I Z-'l) Z[Q2G] + ... + ~m(z)Z[Q",G]

2( I - z )'
(7.3)
N
~
TABLE 7.2 INTEGRATJNG OPERATORS

;!
Function ~
G(s) Zero-Order Normal Interpolation Linear Interpolation ~
C
TZ-l T(1 + Z-1) Tx(O) >
"II
s J - %-1 '2 (l - %-1) - 2(1 - %-1) "II
!:
~
i TZZ-1(1 + %-1) 1'2(1 + 4z-1 + Z-2) 1'2(1 + 2z-1)x(O)
~ 2(1 - %-1)2 6 (l - %-1)2 6(1 - %-1)2
z
1 T3z-1(l + 4z-1 + z-2) T'(I + llz-1 + Ilz-2 + z-3) T'(I + 8z-1 + 3z-2~0) o
."
:;a 6(1 - z 1)3 24(1 - z-I)3 24(1 - Z-1)3 ..oj
:c
III
T'z-l{1 + Ilz-l + J Iz-2 + z-3) T'(I + 26%-1 + 66z-2 + 26z-3 + Z-4)
:;i 24(1 - Z-1)4 120(1 - z-I)'
T'(l + 22%-1 + 33z-2 + 4z-3~O)
~
>
z
(I>
120(1 - z-I)'
T5Z-1(l + 26%-1 + 66z-2 + 26z-3 + z-') T5(l + 57r-l + 302%-2 + 302z-3 + 57z-' + z-5)
~3:
1
~ 120(1 - z-I)5 720(1 _ ;-1)5
T5(1 + 52z-1 + 198z-2 + 104;-3 + Sz-')x(O)

720(1 - %-1)5
Using only the first three terms on the right side of equations (7.2a) and
(7.3) and neglecting the remainder terms, we obtain the following relation-
ship:
As an illustration. if we let G(s) = I/s, then from (7.4) and from Table
7.1 we obtain
':"1 (z) = .'¥·(z) T(t + 4z- 1 + Z-2)
.. 3(1 + z-I)(I - %-1)
+ x(O) [~_ T(I + 4z-' + z '2)J

I- Z-I 1- Z-I 3(1 + Z ')(1 - Z-I)
+ x'(O)Tz- 1 {T(I + Z-I) _ TO + 4%·-1 + Z"2)}
(I - Z-I)2 2(1 - Z··I) 3(1 + z-I)(I _ Z··I)
(7.5)
Combining thesc expressions. we finally obtain the quadratic operator
of lIs with initial conditions as follows:
!/f (z) = .(·(z) T( I + 4z- 1 + Z-2)

1/ 3(1 + z.. I)(1 _ Z-I)
x(O)T(l +
2z- 1) + x'(O)T2z- 1 (7.6)
3(1 + z-I)(1 - z .1) 6(1 - Z-I)(1 + Z-I)
which is seen to be identical to the entry in Table 7.3 for I/s.

. Similarly. we can obtain the linear or higher-order interpolation as the
situation rcquires. Table 7.2 gives zero-order normal and linear inter-
polations [using only Qo and Ql terms in equation (7.3)] for G(s) up to
IIso. These are also known as point slope and trapezoidal rule inter-
polations. Table 7.3 lists these forms ofG(s) for quadratic (Simpson's rule)
interpolation.
EXAMPLE: I
In this cXlImplc we apply Table 7.1 to a linear system des.cribed by the
fllllnwing differential cquution with constant coefficients:
tll.l· t. ., ,I.r
r I· 2./' = 411( t) (7.7)
rl/~ /11
TABLE 7.3 INTEGRATING OPERATORS (quadratic interpolation-)
T(l + 4z-1 + 21- 2 ) TO + 2z-1):I:(O) T2z-1x'(0)

s 3(1 + % lXI - 2:-1) - 3(1 + % 1)(1 - z 1) + 6(1 + z 1)(1 - z 1)
T2(1 + IOz-1 + z-2) TI(I + 5z-1)x(O) T 3z-I,r'(0)

-~:-:---":":"':"'......;.
12(1 - 21-1)2
- 12(1 - 21-1)2
+ -:'7":~--=
12(1 - 21- 1)3
1 TaO + 26z- 1 + 66%-2 + 26%-3 + 21- 4)

sa 60(1 + 21- 1)(1 - % 1)3
T3(1 + 17.2:-1 + 332:-2 + 9z-3)2'(0)

60(1 + z 1)(1 - z 1)3
T4 z-I(3 + 14z-1 + 32:- 2 )x'(0)
+ (I +z 1)(1 - z-I)3
Tf(l + 56%-1 + 246:1:-2 + 56z-3 + z-t)
~ 360(1 - 21 1)4
T4(1 + 422:- 1 + 123z-2 + 14%-3):r(0)
360(1 - 21-1)4
Toz-IO + I3z-1 + z-2)x'(0)

+ 180(1 - z-I)4
1 TOO + 120z- 1 + 119lz- 2 + 24162--3 + 119lz-- 4 + 120z--5 + z-6)

j5 2520(1 + z-I)(1 - 2:-- 1)5
T°(l + looz- 1 + 724z- 2 + 1208:1:-3 + 467z- 4 + 20z-5}x(O)
2520(1 + z-I)(1 - z-I)5
T6z--1(5 + 17621-1 + 47821- 2 + 176z-3 + Sz-4)x'(O)
+---~-~~~-~~-~--~~
5040(1 %-1)(1 _ Z-I)5 +
where U(/) is a unit step and the given initial conditions are x(O) = 3,
x'(O) = -4. The following steps are indicated for the solution.
I. Apply the Laplace transform to equation (7.7) and divide by S2 (in
general, by sft, where n is the order of the equation) to get
( 1 + -3 + 2)
- X(s) = -4S3 + -S25 + -3S (7.8)
S S8
2. Substitute integrating operators for the left side. In this case we

choose quadratic operators from Table 7.3. We note equation (7.11:) and
apply the z-transform to the right side of c(luatioll (7.8) by using Table II
or Table I of the Appendix.
3. Carry out the algebraic manipulation to bring the form of .q'o(z) to
a ratio of two polynomials in Z-l as follows
(18 + 18T+ 3Tt) + (-18 + 30T+ 39T 2 + 4T3)z-1
+ (-18 - 54T - 3T2)z-2
q' + (18 + 6T + 9T2 - 4T3)z-3
. ,,(z) = (6 + 6T+ T2) + (-12 + 12T+ 101'2)z-1
+ (-36T)z-2 + (12 + 12T- tOT2)z-3
+ (-6 + 6T- T 2)z-4
(7.80)
4. Choose a certain T; let T = 0.1. The initial choice of T will be
discussed later. Divide the numerator by denominator to obtain an
expression in powers of Z-l. We then obtain
~t1(Z) = 3.00 + 2.646z-1 + 2.373::- 2 + 2.1 64z- 4 + . . . (7.9)
The coefficients of z-n yield the time response at the sampling instants.
By use of the modified z-transform, we can extend the preceding methods
to obtain the response between the sampling instants if desired.
EXAMPLE 2
To illustrate the application of the z-transform method to the approximate
analysis of continuous feedback system,1 consider the system shown in
Fig. 7.4 where
GJ() _ s + 0.3
S - , (7.10)
S2
and
e- 6 ' 4 = t = e-' (7.11 )
This example has been analyzed exactly for ramp input in Section 2.3.
In this discussion we shall compare the approximate response with the
exact one at the sampling instants.
If the ramp response is determined by analyzing the sampled-data
system shown in Fig. 7.5, the first step is to choose an approximate
sampling period. The Laplace transform of the output is R(s) times the
LG".l<-'" 1:"'·
1111111( I "1.4 ("UUIIlIlIIlIlS dust'II"lnll)' :,yslt'JIl wilh IIt'lay"
O(s) = O. ("e - {loS
FIGURE 7.S Sampled-data system approximation or Fig. 7.4.

overall transfer fUl'\ction; therefore, for a unit ramp input,
C(s) = + 0.3)e-·
(s (7.12)
S2[S2 + (s + 0.3)e-~]
The relative real-frequency composition of the output can be obtained
by replacing s by jw in equation (7.12), which yields
' )
C(}W =
(jw + 0.3)e- Jw
_(/I2[ _(1)2 + (j(l) + 0.3)e- . J"']
(7.13)
As will be explained in Section 7.5, for obtaining an initial value of T,

we let I C(j(l»I be approximately 0.01 or about -40 db. Thus equating
the absolute amplitude of C(j(l) to 0,0] yields
1j(JJ + 0.311 e-i"'l
An exact solution of equation (7.14) could be difficult, especially in

complicated systems. Fortunately. an exact solution is not necessary since
any convenient (" yielding an absolute value nearO.OI will be satisfactory.
Such a (IJ can be obtained by recognizing that for large w equation (7.14)
becomes approximately
IC(j(l)1 ~ ~= 0.01 (7.15)

CIJ
from which an upper frequency limit of 4.64 radians per second is obtained.
Evaluation of equation (7.14) for II) equal to 4.64 yields an absolute value
of C(j(l» of 0.0128, and therefore the approximation made in equation
(7.14) is satisfactory. As explained in Section 7.5, a sampling frequency
of twice 4.64 radians per second (twice because of the sampling theorem)
corresponds to a sampling period of 0.675. This sampling period can be
used. but a T of 0.5 seconds is more convenient (also more accumte) since
the delay is I second. It is not necessary for the delay to be nn integer
value of the sampling period since a sampled-data systcm can bc lInlllyl.cd
for Hny ~lrbitrary delay.
Z-TRANSfORM METHOD IN APPROXIMATION TECHNIQUES 229
With a period of 0.5 second, the approximate transient response at
the sampling instants to a ramp input can be found from the following
expression, if we use the fictitious triangu.l~lr h91d J?.<s) in the feedback
path as shown in Fig. 7.5.
~/I(z) = .~t6(Z)[ I ] (7.16)

1 + ~lt6(Z)
where
&l~(Z) = Z[R(s)G(s)] = 0.125(2.1z2 + O.4z - 1.9) (7.17)
2z(z - 1)4
and
~lrl(z) = ZlQl(S)G(S)] = 0.25(2.tz2 + O.4z - 1.9) (7.18)*
2Z2(Z _ 1)2
Substitution of equations (7.17) and (7.18) into (7.16) yields
tt' ( ) (1.05z2 + 0.2z - 0.95)z
(7.19)
II Z = (z _ 1)2(8z4 _ 16z3 + lO.lz11 + 0.4% - 1.9)
Long division of the denominator into the numerator yields the power
series
<if,,(z) = 0.131z-3 + 0.550z- 4 + 1.26z-5 + 2.18z-6 + 3.17z- 7
+ 4.lOz-8 + 4.86z- 9 + 5.4lz- 10 + 5.78z-11
+ 6.05z" 12 + . .. (7.20)
The response at the sampling instants is given by the coefficients of
equation (7.20) and it is noticed that it compares favorably with the
exact response obtained in equation (2.60). By using the modified z-
transform we may readily obtain the continuous response if desired. The
accuracy may be increased by reducing T and repeating the process. A
negligible error between the first and second choice makes certain that
the result is within engineering approximation. This is true provided the
division process is performed exactly without any approximation.
7.3 Integrating operators 2 ,3,27

The interpolating schemes indicated in Fig. 7.3 represent also the
integrating operators used by mathematicians in numerical analysis.
These integrating operators can also be classified as follows.
(a). INTEGRATION BY RECTANGULAR RULE. Here the interpolator form is
as shown in Fig. 7.30.
.'
)
(b). INTEGRATION BY THE TRAPEZOIDAL RULE. This is similar to linear

interpolation given in Table 7.2.
(c). INTEGRATION BY SIMPSON'S ONE-THIRD RULE. This is similar to
quadratic interpolation shown in Table 7.3.
(d). INTEGRATION BY SIMPSON'S THREE-EIGHTH RULE. In this case the
interpolation transfer function for I/s is
--+-
3T I + 3z- t + 3z- 2 + Z-3 (7.21)
S 8 I - Z-3
(e). INTEGRATION BY WEDDLE'S RULE.

I + 5z- 1 + Z-2 + 6z- 3 + Z-4 + 5z- 5 + Z-6
-
S
-.~ -3T
10
----'-----'----'--......;.-----
1 - Z·6
(7.22)
Comparison of the relative amplitude-characteristics curves of cases b,

e, d, e, and the ideal integrator is shown in Fig. 7.6. It is seen that b is of
simple form and can be used effectively because it also introduces high
attenuation at high frequency, thus minimizing the effect of an external
disturbance. The discussion and derivation of these integrating operators
are available in any standard text on numerical analyses,27 and thus only
a brief mention for comparison is described in this section.
3.0
2.5 I I
(d)1 V«)
2.0
1.5 /
1.2
./
/ .........
V (0)-
-- .- .~ '"""'- -....
1.0
Ideal trIlegralor..;!"
N"
0.8
0.6 -- ---'- 1 -~ ~- ... - -.--- ....

--
0.4
0.3
M
'" \ (b)
\
\
o w,/6
2..
w'='T
FIGURE 7.6 Relative amplitude-charaeteristie curves of the integration operators.
(From J. M. Sal7.er, "Frequency Anillysi~ of Digital Computers Operating in RClil
Timc," p,.o('. IR/~, Vol. 42, I'chnHlry 11)54, p. 4C!.1,)
7.4 z-Forms and modified z-formsl,6,l1,23
The z-form method of analysis is similar to the integrating operator

method discussed in the preceding section, This numerical method is
useful in obtaining the responses of continuous linear or nonlinear systems.
It is very useful for digital-computer approach to the solution of automatic
control problems. This method was proposed by Boxer and Thaler and,
in recent years, Wasow22 has put this
method on a rigorous mathematical joo
s-plane
basis. We shall discuss this method
both as proposed originally and with
the justifications indicated by Wasow. - - - - H - - - - -__ I1-axis
A closely related method to the z-forms
is that proposed by M. Cuenod. 7 • u
The latter algorithm deals directly with
the sequence f(nT) as an approxi- -joo
mation to f(t), whereas the z-form FIGURE 7.7 Path of integration in
deals directly with the Laplace trans- s-plane.
form, ff.[f(t)] = F(s). Wasow has
shown that the resulting two methods have the same accuracy as far as
the order of magnitude of the error of approximation is concerned.
z-Forms relation.~hips
The inverse Laplace transform of a rational function of s with higher-

dcgree denominator in s than the numerator is given by the following
contour integral:
1 i<+;oo
/(t) = .z-llF(s)] = -. F(s)e 'S ds (7.23)
271') <-joo
We assume for simplicity that c is zero (that is, all poles of F(s) lie in
the left half of the s-plane with the permissible location of a simple pole
at the origin). This condition assures that the response of a physically
realizable system is bounded. The contour integration can then be
performed as shown in Fig. 7.7.
Equation (7.23) can be written as the sum of the following three
integrals:
1
f(l) = - .
f'Ir,'T)
F(s)e tl ds
271') /1,,1'1')
(7.24)
232 THEORY AND APPLICATION Of THE z-TRANSFORM
If T is chosen sufficiently small, the first integral will yield a good approxi-
mation for J(/), and the second and the third integrals may be ignored.
The error involved in this approximation will be discussed later. Provided
T is chosen very small and letting t = nT, we obtain
1(1) :::::./u(nT) = -1.

211')
i llw17')
-iI_IT)
F(s)e"T6 ds (7.25)
By substituting for 5 = (lIT) In z (with z = eTI ) in this equation and

noting that by this change of variable the imaginary axis becomes the unit
circle, we obtain
1,.(nT) = _1 r 1. F(1.1n
211'j JI' T T
Z)Z"-1 dz (7.26)
The contour r is the unit circle in the z-plane. Equation (7.26) will be
recognized as similar to the relation for the inverse z-transform, with the
exception of the factor liT. Indeed, Wasow has shown that if F[(IIT) In z]
is considered as the z-transform of J(nT), the difference between the
I.aplace transform F(5) and the z-transform F(z) times T (which is
referred to as the Riemann sum corresponding to 2 f) can be expressed
by the Euler-Maclaurin formula. A pertinent form is given. 2•
fiJI - T11 =! Pv! BO) {d:

v~l dt
1
1 (/(t)e-"l}
t~O
+ R", (7.27)
where m is an arbitrary integer, BO) is the vth Bernoulli number, and R",
is the remainder that can be expressed by a Bernoulli polynomial. Further-
more, it can be shown that there exists a positive number A such that
I!l'f - T;yJI ~ AT (7.28)
Thus the error involved in the discretization of the Laplace transform
is proportional to the sampling period T. A better approximation is
ubtllined if a refinement of the z-transformation is used as follows. Let
(7.29)
Then Wusow has IIhown that
(7.30)
The transfurmation .1* IlmOuntli to approximating the Laplace trans-
""!'lIIatiun hy means of the tnapezoidal rule.
!la!>t'" un this ll1otivatiun, lIuxer and Thaler hllvc pmposcd still another
IIIl1llilkatioll whkh a!> will he shuwn later, yicldsevell better appruximlliinn.
It i!> nntkt~tI I'rlllll t~tJlmtiulI (7,211) that /0'1( 1/7') In :1 is a lransc~ndentlll
fllm'liCl" uf :, whkh IIlllkt· .. it illlpll ..sihlt' til t'xpancl in puwt'rs IIf liz by
z-TRANSFORM METHOD IN APPROXIMATION TECHNIQUES 233
synthetic division. Thus an approximation is made for In z. Of the
several possible, the following approximation will be chosen:
In z = 2(u + tU3 + ius + ...) (7.31)
where
J- Z-1
u= (7.32)
I + Z-1
The series in (7.31) converges rather rapidly and the approximation
yields a minimum phase error.
fn the final application of this method, F(s) is expanded in descending
powers of s, that is,
F(s) = Q1S- 1 + Q 2S-2 + Q~-3 + . . . (7.33)
Thus a substitution for S-k is required. This can be obtained if we rewrite
equation (7.31) as
1
- = -T = ---~~--- TI2
(7.34)
s In z u + u3/3 + u6/5 + ...
By synthetic division, the Laurent series is obtained.
S-1 = f(~ _~ _~~3 - : ' ••• ) (7.35)
Series expansions of s-" can be obtained by raising both sides of this

equation to the kth power. Retaining the principal part and the constant
term gives
S-k ~ N,,(Z-I) = F (Z-I) (7.36)
(1 _ Z-I)" "
where Nk is a polynomial in powers of Z-I. These approximate expressions

for S-k are called z{orms and are given in Table 7.4. If we denote the
::-forms as Tl**, they are related to the z-transform as follows:
It-I t"-I B
T'¥**'
t? (k _ I)!
= T'¥
t7 (k - 1)!
+ T".2
k!
(7 37)
.
Based on these notations, Wasow has shown that
I !R'
It-I
(k _ I)!
- T **
1 (k _ I)! - k
,k-l
< A Tk+l I (7.38)
where Ak is "constant. Hence the z-forms or (Tl**) leads to a still better

approximation to the I.aplace transformation for these special functions,
provided k > I.
In applying Ilw :-1'111"111 IIIclhod 10 th~' solutioll of lincar dill'crcntial
1'111111 t iOIl!> wit h '"ClII,llInl l'Cll"lIidt·llh. Wa 'ClW 22 ha!t shown I hili I hl~ crror
TABLE 7.4
z-Forms
Ft (Z-I)
T 1 + z-1
2" 1 - z-1
T2 1 + )()Z-1 + :-2
12 (I - z-I)2
T3 Z-1 + .r2
2" (I - z-I)3
T4 Z-1 + 4:- 2 + z-3 T4

6 (I - Z-I)4 720
T5 (z-1 + Ih- 2 + llz-3 + Z-4)
24 (1 - z-I)5
T6 (z·-1 + 26%-2 + 66z- 3 + 26:- 4 + z-5) T6

120 (1_2- 1)6 +30,240
T7 (2-1 + 572-2 + 3022-3 + 302:-4 + 572-5 + 2-. 6)
720 (l - 2-1)7
where !Ft (:) = Z [~J t
magnitude between the exact solution at sampling instantsf(nT) and the

approximate solutionf,,(nT) for (n > I, or t> 0) is proportional to T2,
that is,
I~nl = If.. (nT) - f(nT)1 ~ kT2 (7.39)
Furthermore, if F(s) is written in the following form:
F(s) = P(s) (7.40)

1 + Q(s)
where
...
pes) =~ QTS-T (7.41)
T~1
"" b,s-·'
Q(5) = ~ (7.42)
r -I
and if per) = ,fR-IP(S) and q(t) = g.>-lQ(S) have their first k (k > 0)
derivatives equal to zero at t = 0, then the error of approximation is
proportional to T2+k. This indicates that by using the z-forms a better
approximation is obtained than with z-transform or its modified form 1*.
z-Form procedure
The following steps are involved in obtaining the approximate response
Ja(nT) from F(s) = !e[J(t)].
I. Express the function F(s) as a rational fraction in powers of I/s.
2. Substitute for each S-k a rational fraction in Z-1 obtained from Table
7.4 and arrange ~a(z) as a rational fraction in Z-I.
3. Divide the resulting expression by T. The initial choice of T will be
discussed later.
4. Expand the fraction as a power series by using the synthetic division
or other methods to obtain
Ja(OT) +Ja{l T)Z-1 + JR(2T)z-2 + ... + J,,(nT)z-n + ...

(7.43)
The coefficientsJa(nT) are the approximate value ofJ(r) at the sampling

instants. To obtain the approximate response between the sampling
instants, if required. the method to be discussed on p. 236 can be
used.
If F(s) has one degree higher denominator than the numerator in s. we
can improve the degree of approximation using the z-forms by writing
F(s) in the following form:
A
F(s) = - + F 1(s)
s
By obtaining first the inverse Laplace transform of A/s, AU(f). and then
applying the z-forms into F1(s) and carrying out the procedure given.
we can improve on the approximate response to J(t). Finally, if initial
values do exist, the z-forms could be easily modified to cover this case as
well.
t See Table IV for higher values of k. The values of B•. are the Bernoulli numbers.
lIome of which are given below: 21
Bo = 1,8. = -A. B~ = "~' B. = -.'0. B. = *. B. = - ...
"," II"'., ".,.: -N.•. B•• =+.... with B~ • • =O.
for k > O.
Modified z-forms l1
As with the z-forms, we can also obtain approximate relations for the
continuous response by using other relations; these relations are called
modified z-forms and are derived as follows. Assume the Laplace trans-
form of a continuous output is given as
F(s) = Co + CIS + c'Jil.a+ ... + c"s: (7.44)

do + dis + d,.s + ... + dlts
If this equation is multiplied by e-'1', the inverse Laplace transform will
be /(1 + mT). By letting m vary between zero and unity. we obtain the
response between the sampling instants if /(/) is approximated by /o(nT)
as discussed earlier.
The use of these relations requires rewriting equation (7.44) as
F( ) = coO/sit) + cl(I/SIl- 1) + ... + c,,(l/slI-") (7.4S)

do(l/s ) + d1(1/s - ) + ... + dk
S It It 1
The numerator of equation (7.45) is multiplied by the Taylor series

expansion of e· mT ; that is, by
em.'1' -_ I + sm T+ (smT)t + S3 m 3T3 + ... (7.46)

2! 3!
If each term in the numerator is multiplied by the series and we carry
terms that yield only the principal part plus a constant term, then
F(s)e· m '1'
_ go(1/SIl, s, mT) + gl(1/SIl-1, s, mT) + ... + gll(I/i" ", x, m'l')
- do(l/Sk) + d (I/S + ... + "It
1
II - 1 )
(7.47)
Substituting the z-forms for I/slI, we obtain the modified z-forms.* These
modified z-forms are tabulated in Table 7.5.
• For instance, ir F(s) = I/s, we have from equation (7.46),

I CmT)1
FCs)e,"·T = -s + mT+ .\-2'
.
+ ... ('1.-1'1,,)
To obtain the mudified :·furl11s, we: sllh~lillll(' Ihe .~·I'III'III' 1'111' th,' lir,t IWII II'IIII~ hi
oblain
T: .. I 'I'll I 2m'- I (I 2""
1,'(11/, I) ,. .•. I ", I - .,
2, - I I. I
TABLE 7.S MODIFIED z-FORMS"
F(m,z)
sft
TO + 2m)z + (I - 2m)
s 2 z- 1
T2 (I + 6m + 6m2)z2 + (10 - 12m2)z + (l - 6m + 6m2 )
12 (z _ 1)2
(m + 3m2 + 2m3)Z3 + (6 + 9m - 3m2 - 6m3)Z2

ra + (6 - 9m - 3m2 + 6m3)z + (-m + 3m2 - m3)
12 (z - 1)3
If m - 0, the modified z-forms reduce to the z-forms. The accuracy

of the calculations is improved if m is allowed to vary in the range
... ~ ~ m ~ ! to obtain the total response.
The procedures to be followed using the modified z-forms are
I. The modified z-forms are substituted into the numerator of the
I.aplace transform expression.
2. Thc z-forms are substituted into the denominator of the Laplace
1IIIIIsfnrm expression.
J. Divide the resulting ratios of polynomials in Z-1 by T. To obtain
II,,· limc function, carryon the synthetic division method by first choosing
II I'"rticular value of T.
II.I.USTRATIVE EXAMPLE FOR z-FORMS

III Ihis example we consider the feedback control system shown in Fig.
I K. A!>SIlIllC the input as a unit step, R(s) == I/s; the output transform
e '(.,) i!> ,~ivcn by
('(s) = I (7.48)
.~3+.~2+S
1"111' rnllnwill~ procedure is used to obtain the approximate response

.... 111'·,IIII' .... I'nI'IIIS,
eft)
C(.,
I Ie IURI' 7.N ('UIIIIIIIIIIII'I rl't'dhllC."k "Ylll(,lI1,

I. Divide the numerator and denominator by the highest power of s;

then
(7.49)
2. Substitute for each term the ~-form from Table 7.4 and divide by T
to obtain
6T2(z-1 + Z-2)
~(I(z) = (12 + 6 T + T a) - (36 + 6T _ 9T2)z-1
+ (36 - 6T- 9rt)Z-2 + (12 - 6T+ T 2)z-3
(7.50)
3. Choose the sampling period T on the basis of frequency response.
Hcre T is chosen as 0.5, based on the criterion that frequency components
below -30 db are negligible. Therefore ~(I(z) becomes
~( ) 1.5z- 1 + 1.5z-2 (7 51)

(I z = 15.25 _ 36.75z- 1 + 30.75z-2 _ 9.25z-3 •
4. Dividing the numerator by the de!,\ominator, we obtain the time

licquence response c(I(n) as the coefficient of z-n. Hence
~(I(Z) = 0.0984z-1 + 0.335z-2 + 0.610z-3 + 0.853z-· + ...

(7.52)
In comparing the approximate with exact response, we would find that

Ihe difference is negligibly small.
7.S The choice of the sampling period 1,10

The crror of approximation depends to a large extent on the sampling
period. For instance, if the sampling period T is zero in the limit, the
;Irproximate system is exactly equivalent to the original system before
;Irrmximntion. On the other hand, if Tis large, the error of approximation
is increased although the number of computations required to obtain the
lime response is considerably less than for small T. Furthermore, if a
di~ilal computer is used in connection with z-forms, as the number of
cnmrutnlillns increases the round-ofT error also increases. Thus thc choice
Clf Ihc sampling rcriod rCllllircs " cmnpromisc betwecn the crror of
;Ipprnxill1;I(ion "lid thc lIul1lht~r nf l'nmpulatiolls relluircd.
The critcrinn fur Iht" illiliul dlOk,! nf T is rathcr difficult. Scvernl
nih'rill IIfC "dvm'nlctl, nl\lClI\~ whit'h is Ihe olle cnnllidcred hcrc bascd on
Z-TRANSfORM METHOD IN APPROxiMATION TECHNIQUES 239
the frequency response of the output. For instance, if the output mag-
nitude as a function of real frequency is written as
Icejw) I (7.53)
we can choose a value of frequency W k that gives a small value of lC(jw)l,
that is, around -30 db or about 0.01. By obtaining such a frequency,
then from the sampling theorem, the value of T is given byl
T= !!... (7.54)
WI<
To obtain an approximate value of W k , lcejm)/ is approximated for large

(I), which avoids any exact calculations. This procedure is explained in
Example 2. Another procedure for choosing T (as used in obtaining the
transfer function of a system or network when both the input and the
output /;(1), /,,(1) are given) is based on the fact that both /,,(1) and /;(1)
should change on the average by about 10% of their maximum values
within successive intervals of the chosen sampling time.
7.6 Analysis of the error6,22,~,2l;
The error entailed in such an approximate analysis, discussed in the

preceding sections, cannot be formulated exactly. The formulation of the
error bound which can be obtained for linear systems is generally quite
elaborate. The error bound becomes significant only when many samples
are considered, as is done in the analysis using digital computers. However,
the error can be readily estimated if the sampling period is halved or
subdivided to calculate a few transient terms. If by so changing the
sampling period the original solution is essentially unchanged or the change
is small for engineering approximation, the original solution is satisfactory.
This, of course, is based on the fact that round-off error is not considered,
that is, when the computation is carried on without digital computers.
If the initial choice of the sampling period is unsatisfactory. this procedure
has to be repeated until the change in the response is insignificant. Both
WlIsnw 22 ,and Bridgland 24 have discussed the error of approximation
invulved in the use of the z-forms and other rules indicated earlier. The
\'olllplete discussiun of this subject is beyond the scope of this text.
In some cases Simpson's (quadratic) rule is numerically unstable when
u!.t'd fur Ihe intc~mti()n of the dilTercntial equation d!lldx = AY when the
n'al parI uf A is Iwgalive, This is because the churucteristic equation in z
II:!!. a 1'11111111' IIllldulus ,',n':llt'r 111:111 unity. The stnbilily criterion estahlished
III ('hlll'tt'" \ \':111 Ill' n':lIlily lISt'cllu ch'h'l'lIlim' sudl II situation. To :Ivuid
Mle'h II tlillh'ulty, wC' IIIlIy liSt' II !.1II1111 inh'l'vlIl IIVI'I' tI", initinl alld sle\'I'
part of the response and continue with a much longer ihterval over the
more slowly changing part. Thus the adjustable sampling interval plays
an important role in minimizing the error of approximation.
7.7 Low-pass transformation for z-transforms2o ,21

(n certain applications, in which the output of a linear time-invariant
system !o(I) as well as its input h(l) are given as experimental data. the
z-transform method could be easily used to obtain the system transfer
function ,"'(z) as follows:
Jf(z) = fFo(1.) (7.55)

fFi(1.)
Using the synthetic division method .Jt"(z). the impulsive response at
discrete intervals of time, can be easily determined.
The data of the input and output are usually not exact in practice but
are subject to error or noise and the division method of the z-transform
will yield erroneous results. Error in the series for Jf'(z) is introduced
if the leading (or initial) term of !Fo(z), that is'/D(O) is much smaller than
the subsequent terms.
To avoid this difficulty. the z-variable may be replaced, before perform-
ing the synthetic division, by the variable A, where
A+ I = Z-l '(7.56)
It is done by simple substitution of A + I for I/z, in the expressions
for !Fo(z) and .~iz), followed by expansion and regrouping of terms
to form a power series in it; thus 'o(z) and !Fi(Z) become .'!io().) and
.#i().)' By this process, known as low-pass transformation. the leading
term of :j;D(it). that is, .#o(A)IA~O is much larger and thus the division
method is less sensitive to the errors in the individual terms. By dividing
/f;o(A) by !F,(A) and transforming back to the z-plane by this low-pass
transformation, we readily obtain J(.'(z) or the system transfer function.
It should be noted that only when we consider noisy data do the advantages
of A-domain become apparent. Solution of Problem 7.3 illustrates the
procedure and the advantage of this transformation. The A-transform
leads to a match for
H(s) = Fo(s) (7.57)

Fi(s)
at low frequency, whereas the z-transform leud!l to a mulch for the !ly!ltem
transfer function ul higher frequencies und therefore ex:tg~er:lte!l the en'cct
::-TRANSFORM METHOD IN APPROXIMATION TECHNIQUES 241
of the noise. Another kind of transformation. known as the two-sided
:-transform, has also been advocated for this particular case.!1 However,
we shall not elaborate on this technique.
7.8 Application to time-varying differential equations6 •1l

Boxer and Thaler have advocated use of the z-forms method for the
approximate solution of time-varying or nonlinear differential equations.
Wasow has shown in a rigorous way the validity of this method and the
underlying assumptions required to obtain a fairly good approximation.
In this section, we shall briefly indicate the procedure to be applied to the
following:
d" d n- 1
dl" [!,,(t)y] + dl"-I l!"_I(t)y] + .. ·!o(t)y = f(r) (7.58)
where 1,,(/), I" _1(1), .•• ,/"(t) are functions of the independent vari-
able I.
The procedure for the approximate solution is
I. Obtain the Laplace transform of (7.58) after substituting for /,,(1) ==
• • • ,/;(1) = c, and inserting the initial condition.
C II •
2. Proceed with the division as in the constant coefficient case. How-
ever. change the values of C '" C .. _ I •••• , Cn at each step in the division
process to the values of the functions at the corresponding times.
The minimum limitation on the form of the functions in (7.58) is that
they be Laplace transformable. The procedure for inserting the initial
conditions is illustrated in Problem 7.4.
EXAMPLE
To illustrate the procedure, we choose the following example:
(II/
-'- + t1/ == I, with yeO) == 0 (7.59)
tit
hllllltinn (7.59) is written as
..,/1/,- + ('1/ == I, where (. == I == nT (7.60)

,II '
Wh('11 Wl' lakl' Ihl' I.aplacc Iwnsfurlll. cnn!oilicring c' as constant and
!I( II) n,
l'( \ I (7.(,1)
or
S-3
Y(s) = 1
1 + cs-
By substituting the z-forms of Table 7.4 and dividing by T, we obtain
T (z-1 _ Z-2)
2
rr?la<z) = -------~-'-------':..-..____.,____.,----
(2 + en - (6 + eT)z-1 + (6 - CT)Z-2 - (2 - CT)Z-3
(7.62)
Letting T = 0.4,
0.16z- 1 + 0.16z- 2
~
o
(z)=-------------------------------------
(2+0.4c)-(6+0.4c)z-I+(6-0.4e)z-2_(2-0.4c)z-J
(7.63)
The long division process is carried out letting c = 0.4 during the first
step of the division, 0.8 during the second step, and so forth, yielding
9i,,(z) = 0.074Iz-1 + 0.266z- 2 + 0.503z-3 + 0.714z- 4 + ...

(7.64)
The approximate solution can be compared with the exact solution for this
example. The latter solution is
Y = I - e- t2 / 2 (7.65)
This procedure is also applicable to differential equation of the

following form if the j;(/) are slowly varying as in some physical systems.
tin dn - 1
- Y + ... + In-l-l
In dIn dt n-
y + '" + loY =/(1) (7.66)
This can be justified if we notice that
(7.67)
where (Z) are the binomial coefficients. If In is slowly varying, the con-
tributions of the derivatives ofIn are negligible, so that the approximation
dnUnY) '" d"y (7.68)

dIn -In dtn
can be justified. It should be cautioned that with this procedure infor-

mation on the stahility of these equations l'anno\ he ohtained.
7.9 Application to nonlinear

differential equations 8 ,l1-13,18
In the preceding sections the methods of integrating operators and the
z-form have been applied to obtain the approximate solution of linear
differential equations with or without initial conditions. In this section,
the method of sequential computation, which consists in approximating
a given differential equation by its discrete version, is employed. Further-
more, the method of integrating operators and the z-transform will be
employed to obtain an approximate solution of a nonlinear differential
equation.
EXAMPLE I
Consider, for illustration purposes, the nonlinear differential equation
-di + I. +'21 =1, with i(O) = 0 (7.69)

dt
We approximate this equation by means of a difference equation as follows.
Let
where T = sampling interval (7.70)
Then,
ill+l - in + Tin + Ti! = T. (7.71)
If we choose T = 0.2, we get
5i,.+1 - 4i" + i~ = I (7.72)
The series solution of this equation for T = 0.4 by the method of
complex convolution discussed in Chapter 5 does not converge. However,
we obtain a convergent solution for T = 0.2. The solution as a recurrence
relCltionship compared with the exact solution of the differential equation
is shown in Table 7.6.
TARLE 7.6 COMPARISON OF EXACT AND APPROXIMATE
SOLUTION FOR F.XAMPLE I
Exact Solution of
t = 117' Rc'currt'1Icc' Equation Sollllion Differential Equation
_ _ _ _ •._._. _ _ _ _._00
() () o
~ o.:mo 0.217
III o..mm 0.370
l'i 11,4710 0.467
,In II. 'i IH.I II. 'i:'I)
-\I) lI.t.I)!! I lI.tlllt.
EXAMPLE 2
In this example we employ the linear interpolation operators to solve the
following nonlinear equation:
-d 2x + 2 -dx + x = -0.9 -dv ,

2
with x(O) = 0, x'(O) = 1
dt dt dt
(7.73)
where L' = f(x) and 1~0) =0
I. Apply the Laplace transform to equation (7.73).
(S2 + 2s + I)X(s) = -0.9[sV(s) - L~O)] + 2.r(0)
+ x'(O) + sx(O) (7.74)
2. Divide by S2 and insert initial conditions
(.1 + 2-s + -1) X(s) ~

V(s)
= -0.9 -
s
+ -~t (7.75)
3. Substitute linear interpolation operators from Table 7.2 and the

z-transform for l/s2 in the right-hand side.
!?l" (z)
n
[I + 2T(l + Z-I)
2(1 _ Z-I)
+ T2 (I + 4z- 1 + Z-2)J
6 (1 _ Z-I)2
= -0.91""(z)TI + Z-1 + Tz- 1 (7.76)

21-z- 1 (I-Z- I )2
4. With T = 0.4, the regression (or difference) equation of the above
transform is
8.56x n - 11.36xn _ 1 + 3.76xn _2 = - 1.08vn + 1.08vn _ 2 + 2.4,
for n = I, = -1.08v" + 1.08vn_ 2 • for n > I (7.77)
If we let l'n = x~, we can solve this equation as a recurrence relationship
for each value of n starting with n = I and putting lIn = = 0 , for r;
n < 1. The response is shown in Fig. 7.9.
In conclusion it should be pointed out that the division method using
the z-forms can also be used to obtain an approximate solution of such
0.6 r---,--,---,---,--...----.
0.4
4.8
!'I(HIRE 7.1) Rt'Spllll~C III' I ~lIl1ll'lt' :' I'll' lIIililll Vt'llIl'ily .• ··(111·- I
equations. This method is based on adjusting the divisor at each step of
the division, as illustrated by the following example.
EXAMPLE 3
dY + y2 =1 with y(O) = 0 (7.78)

dt '
This equation is written as
dy
dl + cy = I
where c is considered constant in this operation and adjusted at each step
of the division.
The Laplace transform in powers of S-1 is given as
S-I
Y(s) = I + cs- 1 (7.79)
By substituting the z-forms and dividing by T,

TO + lOCi + Z-2)
~ (d = - - - - - - - - - - -
II (12 + 6cn - 24z- 1 + (12 - 6cT)Z-z
Letting T = 0.1,
o}j (z) _ 0.1(1 + 10%-1 + Z-I) (7.80)
• R - (12 + 0.6c) - 24z- 1 + (12 - 0.6C)Z-1
Carrying out the long division and letting c =0 for the step first and
adjusting it after each division, we get
tW'/J(z) = 0.00833 + 0.100z- 1 + 0.199z-1 + 0.295z- 3

+ 0.390z- 4 + . .. (7.81)
The exact solution of the original equation is
y = tanh t (7.82)
The approximate solution compares favorably with the exact one. This
method can be used for the approximate solution of the following differ-
ential equation:
d"
dt"
[J., (y, citdy , ...) yJ + clt"-
d"-1 [/,,-1 (y, dy , ...) yJ
1
cit
+ ... + lo(Y, dy , .. .)y = I(t) (7.83)

,It
wilh
'( ,/I,
,', . ./. /I. ' .'" ) (7.R4)
. ,It
We IlI'lII"'l'" I"m Ih,' III'PI'U!(illllllinll liS ill Ih,' "1111:.111111 ctll'lIkicIII case.
""Wl'VI'I, ill 1"'lflll'lIl1l1l! Ih,' IIIIII! tlivisiull PI'III""", WI' 1'111111,." Ihl' vlllm's
246 THEOIlY AND APPLICATION OF THE Z-TRANSFORM
of CII , ••• , Co at each step in the division process to the most recent
values available of Ci = /;(Y, ~~ ,.. .).
7.10 Other numerical techniques 8 •9 .l&.J7-19
To obtain an exhaustive study of all the numerical techniques is beyond
the scope of these discussions, and can be readily studied in any of the
standard books on numerical analysis. In this chapter, we have used
several of the integrating operators which are closely related to the
subject matter, that is, the z-transform method. The main emphasis has
been in using these to solve systems (linear or nonlinear) which are usually
encountered in engineering applications. Rigorous application of the
theory to approximation problems in mathematical terms is not attempted.
In trying to limit the discussion, it was necessary to ignore other numerical
techniques. Among those not discussed in detail are the works of Madwed,9
Cuenod,7·14 Tustin,' and Heizman et al. 13 However, the techniques dis-
cussed are closely related to these and other methods.
The main emphasis of these numerical techniques is to obtain approx-
imate solutions for systems in which exact solutions in a closed form are
not possible. This is often true in a complicated nonlinear or time-varying
continuous system. It is for solving such problems that this method could
be advantageously used. Furthermore, the use of digital computers to
solve differential equations requires the formulation of the problem in
terms of difference equations. Hence the z-transform method plays an
important part in obtaining an approximate solution.
REFERENCES
I. Jury, E. I., Sampled Data Contf'fll Systems, John Wiley and Sons, New York, 1958.
2. Ragazzini. J. R.• and G. F. Franklin, Sampled-Dala Control Syslems, McGraw-Hili
Book Co., New York. 1958.
3. TOll, J. T., Digital and Sampled-Dala Conlrol Syslems, McGraw-Hili Book Co.,
New York, 1959.
4. Cruickshank, A. J. 0., "Time Series and Z-Transform Methods of Analysis of
Linear and Nonlinear Control Systems," Aulomali~ Qnd Remole Control, Pro~. of firSl
/",. IFAC Congress, Moscow, 1960, Butterworths, London, 1961, pp. 277-285.
5. Robertson, H. H., Discussion of Reference (4), p. 285.
6. Boxer, R., and S. Thaler, "A Simplified Method for Solving Linear and Nunlinear
Systems," Prod. Insl. Radio En.t:ineers, Vol. 44, 1956, p. 89.
7. Cuenod, M., "Contribution a L'etude de Phenomenes Transitllires :\ I':lide des
Suites de Temps," BIIlI. Tech. SlIi,fse ((1m .• Vo\. 75, 1949, p. 201.
8. Tustin, A.• "A Method (If Analysing Ihe I'Irccl or Cerlain Kinlls of Nlllllilll:atity ill
Closed Cycle Control Systcm!>," J. I".v/. IJ/'/'I. I':'l~"'v., Vul. "4. 1'.u1 II. 1')47, p. I 'i~.
%-TRANSFORM METHOD IN APPROXIMATION TECHNIQUES 247
9. Madwed, A., "Numerical Analysis by the Number Series Transformation Method,"
Proreedin,gs Symposi"m on Nonlinear Cirrllit Analysis, ",ew York. 1953, Edwards
Publishing Company, Ann Arbor, Michigan, p. 320.
10. Truxal, J. G., "Numerical Analysis for Network Design," Proceeding InSlillile of
Radio En,lfineers, Professional Group on Circuit Theory, CT-I, 19S4, p. 49.
11. Boxer, R., and S. Thaler, "Extension of Numerical Transform Theory," Report
RADC-TR-S6-115, Rome Air Development Center, New York.
12. Naumov, B. N., "Eine Niherungmethode zur Berechnung der Obergangsprozesse
in Selbsllatigen Regelungssystemen mit nichtlinearen Elementen," Rep. Con[. Alllo,
m.llie Control, Heidelberg, 19S6, Oldenbourg, Munich, 19S6, p. 184.
13. Heizman, C., J. Millman, and A. Vagant, "Comparison of the X· Transform Method
with Other Numerical Methods," Technical Report T-8/C. E. R. L., Columbia Uni·
versity Engineering Center, New York, September, 1956.
14. Cuenod, M., "Principe de L'analyse impulsionelle et de son application ala Theorie
des Servo·mechanismes," Onde elerl. Vol. 37, 19S7, p. 723.
IS. Jury. E. I., and F. J. Mullin, "A Note on the Operational Solution of Linear
Difference Equations," J. Franklin Insl., Vol. 266, 1958, p. 189.
16. Mishkin, E.• M. S. Goldstein, and J. G. Truxal, "Nonlinear Feedback Systems with
Sampling." Research Report R-4SO-5S, PIB-380, Polytechnic Institute of Brooklyn,
New York.
17. Hellman. 0 .• "On the Iterative Solution of the Equations of a Single· Loop System
with One Non·linearity," J. Eleclronics and Conlrol, Vol. 6. 1959, p. 186.
18. Brown, B. M., The Mathemalical Theor)' of Linear SYltems, John Wiley and Sons,
New York. 1961.
19. Raymond, F. H.• "Regimes Transitoires et Techniques des Impulsions," aI/de
EIC'tri(l',es. Vol. 28. 1948, p. 222.
20. Huggins, W. H .• "A Low Pass Transformation for Z·Transforms," IRE Trans.
Circllil Thellr)" Vol. I, September. 1954, pp. 69-70.
21. Thomson. W. E.• "On Two Sided. Z-Transform," IRE Trans. Circuit Theory,
Vol. 3, Jure 1956. p. 156.
22. Wasow, W., "Discrete Approximalion to the laplace Transformation," Z. a/~few.
MUIII. II. Php., Vol. 8, 19S7 pp. 401-407.
2:1. noxc:r, R., "A Note onNurnerical Transform Calculus," Pror. IRE, Vol. 45, 1957.
24. Bridgland. T. F .• Jr .• "A Note on Numerical Integrating Operators," J. Soc.
IlUlllstri,,1 and Appl. Math .• Vol. 6, No.3, September 1958, pp. 240-256.
2~. (jihson. J. f.., Nonlinear Alilomatic Conlrol. McGraw·HiII Book Co .• New York,
1%.1.
2". Jllnhlll, C. Ca/cull/.r (if Finile Diffrrenc(!s, Chelsea Publishing Co., New York (2nd
111.). 11)(,f).
n. Sl·:ll'hnrllllgh. J. n.. N,,,I/("'irtl/ Malilematical Analy.ri.f. The Johns Hopkins Press,
lIallilllllrc. III~II. Chapler 7.
,III Salll·I' • .I. M., "Fl'ctlllcncy Analysis of Oighal Computers Operating in Real Time,"
,., ..... flU. VIII. 4~. Nil. 2. 11)54. pp. 4~7 461•.
•"1 WI'''~. I .• ;11111 It. N. Md)ullnlll:h. "1'ruIlY's Methnll. Z-Tmnsforrns, and Pad~
~. Nil. 2, Aplil 1'1t•.\, 1'1" 145 141).
"1'1 ....'"111111 .. 11 ....'i/"M H,',·., VIII.
8
APPLICATIONS TO VARIOUS
AREAS OF SYSTEM THEORY
In the preceding chapters the z-transform theory has been extensively

discussed and developed. Several examples from physical systems were
considered to indicate the applications of the theory.
This chapter is concerned mainly with the application of the material
developed. The examples chosen show the wide spectrum of application
of the z-transform method. To limit the material, only six examples are
discussed in detail. The problems section at the end of the book will give
the reader a greater indication of the extensive applications.
The examples are chosen from the following areas of system theory:
sampled-data feedback systems,I.II.5 antenna theory,7-' economic sys-
tems,n-II information and filtering theory,87-aI sequential circuits,lt-17
and finite-state Markov processes. II-eo The examples are formulated
mainly in terms of difference equations, both linear and nonlinear, and
the z-transform with the other theorems developed earlier are used to
obtain the required solutions.
8.1 Nonlinear sampled-data feedback systems3.6.1.23

Because the later examples are concerned for the most part with linear
theory, it is advisable for the sake of completeness to choose a saturating-
type, nonlinear sampled-data system as an illustration. In the following
discussion the convolution z-transform theory discussed in Chapters 4
and 5 will be applied to a second-order nonlinear system. 6
Consider the sampled-data system shown in Fig. 8. J. where the non-
linearity consists of a saturating amplifier in the fecd·forwIlrd p"th. We
will obtain the response of this system lIsing the cnnvnJutiun :·tmnsfnrm.
24K
APPI.ICATIONS TO VARIOUS AREAS OF SYSTEM THEORY 249
T.l. .... /l6'(z)
Zero-order Saluratins r-'''' ---0
hold I e(,1)
I
I e(l)
T=l
FIGURE 8.1 Nonlinear sampled-data reedback system.
We now derive the difference equation of the system. Since hen} is

constant over a sampling interval (in view of the zero-order hold), we
obtain the pulse transfer function for the linear part as
~(z) = <[I - e- T• G1(S}] = 0.716z - 0.284 = 'if(z}

S Z2 - +
1.135z 0.135 ,;;f(z)
(8.1)
This equation gives the following difference equation:
c(n + 2} - I. I35c(n + 1) + 0.135c(n) = 0.716h(n + I}

- 0.284h(n) (8.2)
Now hen) is given as
40 e(n) ~ 40 }
hen) = { e(n) -40 ~ e(n} ~ 40 (8.3)
-40 e(n} ~ -40
The power series representation of this nonlinearity over -65 ~ e(n) ~ 65

is
hen) = l.llle(n) - 0.1234 x 1O-3e(n}3 (8.4)
From Fig. 8.1 we note
e(n) = r(n) - c(n) (8.5)
Uy substituting for e(n) in equation (8.4) from equation (8.5) and inserting
the results in equation (8.2), we obtain
c(n + 2) - + I) + 0.135c(n)
1.1 35c(n
= 0.795['(11 + 1) - ('(n + I)] - 0.0884 X 10- 3
x [,en + 1)3 - C'(n + 1):1 + 3r(n + I)c(n + 1)2
-- 3r(I1 .. 1)2C"(1I + I)] - O.J 16[,(11) - c(Il)] + 0.035
x J() :1 If(II):1 1'(11):1·1 .1,(,,)1'(11):1 _.. -"(11):.11'(11>1 (H.6)
If we assume that r(n) = 60u(n), where u(n) is a unit step function, then
r(n +
k) = r(n) and equation (8.6) becomes
c(n + 2) - + 1) + 0.197c(n) + 10-3
1.295c(n
x [-6.3c(n)* + 15.9c(n + 1)* - 0.0884c(n + 1)3
+ 0.03Sc(n)3] = 0.2865 x 60u(n) (8.7)
If the system is stable around the equilibrium point as assumed, then
<if(z) can be written in the form
00 A
<if(z) = I
11-01 - a,.z
-1 ' (8.8)
(8.9)
where a,. for at ~ 3 is of the form ar
<4, where ". lJ are positive integers
(including zero) and" + lJ ~ 2.
Taking the z-transform of equation (8.7). we obtain
~(z) + (15.9z - 6.3) {,1[c(n)'] _ 0.00555,1[c(n)]3}
Zl - 1.295z + 0.197
= (0.2865) 60z
z- ]
c(O)(z· - 1.295z) + zc(l) + (IS.9)10-azc(0)8
+ - 1O-3(O.0884)zc(0)3 (8.10)
z'l. - 1.295z + 0.197
The demoninator of the last term of this equation can be written
ZB - 1.295z + 0.197 = (z - ai}(z - a~) (8.11)
where
ai = 1.12, a.
= 0.175 (8.11a)
That la;1 > 1 should not be of concern because it is due to simplification
of terms of the form ,(n + l)c(n + 1)2 to 60c(n + 1)2 and r(n + I )2c(n + I)
to 3600 c(n + I). This stems from the fact that ,(n + k) = ,en) = 60.
From equation (8.8) we assume the folJowing solution for <if(z):
C/f(z) = Ao + AI + As + A20
1 - Z-I 1 - alz- I 1 - a.z- I 1- a~z-l
+ Aao + ... + AOB + Aoa

1- O~Z-l I - a:Z- 1 1- O:Z-1
+ ... + Au -1 + A:I .. 1
1 - ala.z 1 - ala2z
(IU2)
APPLICATIONS TO VARIOUS AREAS OF SYSTEM THEORY 251
Similarly, from equation (8.8), the %-transforms of c(n)i and c(n)3 can be
written as
[c(n)'] = A: + 2AoAI + 2AoAa + 2AoAao + A~
;y 1 - Z-1 1 - OIZ-1 1 - Qr- I 1- a~z-I
+ 2AoAao + 2AIA20 + 2AoA2 + A: + ...
1 - a~z-I 1 - a~-I
+ 2AoAn + 2AIAa + 2AoA~n + 2AIIA. + 2AIAu
1 - 01Qr- 1 1- a~Qr-1
+ ... (8.13)
and
;y[c(n)3] = A~ + 3A:A 1 + 3A:A.
1 - %-1 1 - az1- 1 1 - ar- I
+ 3A:A20 +. 3~oA~ + ... +

1 - 0IZ
3A:A o2
1 - Or
+.3~oA: + ...
+ 3A:A 21 + 3A~A2 + 6AoA20Aa
1- o~ar-l
+ 3A:An + 6Ao~IAa + ... (8.14)

1 - ala,.,r
Substitute equations (8.12 through 8.14) into equation (8.10) and compare
coefficients of like terms of z/(% - ar a:>,
y, 6 = 0, I, 2, . . •. The steady-
state amplitude is given by
A + (lS.9 - 6.3)10-3(A: - 0.OOS55A~) = (0.2865)60 (8.15)

o 1.197 - 1.295
which gives Ao = 60. This is expected since the system is stable and the
steady-state error is zero. Comparison of coefficients of z/(z - al)'
z/(% - aJ yields
A1,.{a:,. - 1.295Ql,I + 0.197 + (15.901.8 - 6.3)(6O)10-3} = 0

(8.16)
For a nontrivial solution of Al and All we must have the term inside the
parentheses in equation (8.16) equal to zero, that is,
0:,2 - 0.34at.2 - 0.181 = 0 (8.17)

or
a 1 = 0.6275
" t = -O.2K72 (8.18)
Comparison of coefficients of other terms gives
Aao = (0.0645)10-aA I ; Aoa = (0.215)10-3A~.
Aio = (0.0249)10-8A~
A10 = (0.0143)10-t A:. Ao, = (O.t265)10-8A:.
All = (1.04)IO-3A~A2 (8.19)
ASI = (0.6l)IO- 9 A:A 2 ;
All = (0.465)10- 3AIA:
Au = (0.511)10- 8A I A:•...
Next. we consider the coefficient of terms %/(% - a;). %/(% - a~). where
a~ and a~ are given in equation (8.lla). We then obtain the two relations
20 + 60A I + 60A 2 _ A30 + ASD + AD3 + Aos + Au

492.5 1407.2 33.4 60.5 33.5 35.4 28
+ An + Au + A14 + A70 + Ani + ...

31.7 44 36.3 39.8 34.2
= l(0.2865)60 + c(0)(1.I2 - 1.295) + c(1)

0.12 } I
+ [15.9('(0)2 - (0.0884)c(O)3JI0-3 - (8.20)
and 11.5
-2.91 _ 60A 1 + 60A 2 + A3D + Aso + A03 + A21

452.5 462.2 2.76 4.6 5.81 6.55
+ A41 + ... = {(0.2865)60 + c(0)(0.175 _ 1.295)

5.97 -0.825
+ c(I)[15.9c(0)2 - 0.0884C(O)3JI0- 3} _1- (8.21)
-3.52
Using relations (8.19). we reduce the preceding two equations into two
nonlinear relations in Al and At. The series on the left side of equations
(8.20) and (8.21) must be absolutely convergent. For this problem. we
know c(O) = 0; therefore e(O) = 60 and h(O) = 40. Also we know that
c(-I) = h(-I) = 0, so from equation (8.2) with n = -I we get ('(1) =
28.6. A few trial values of Al and A2 (AI and A2 initially chosen around
Iintar solution values) indicate divergence of the series. This indicates
that the initial values are far away from the steady-state value (equilibrium
point). However. if the time axis is shifted by one integer vlllue (sampling
interval). we must use the new values of £'(0) = 28.6 lind ('(1) = 43.6. the
latter being calculated from equution (8.2) by using the dilTerencc ctlunti(ln
as recurrence relation. It sh(luld be p(linled Clul th,,' since this shiftin~
can always be used in such ,·i\ses. :1 mnn' rapid l'unvl~r~em'e is nhlainahlt'.
A similar modification does not exist in the application of Laplace trans-
form convolution to nonlinear differential equations. By this time axis
shifting, convergence is obtained with the right-hand side of equations
(8.20) and (8.21) now equal to -3.26 and 2.5, respectively. Numerical
values obtained for the various coefficients are
Al = -26, At = -1.4, Aa = -1.13, Aao = -0.298,
A70 = -0.115, AZ1 = -0.985, A41 = -0.54. ASI = -0.265,
An = -0.0239 .. , (8.22)
The z-transform of c(n) for n ~ I can now be written from equation (8.12),
using the calculated A's and Q's:
60 26 1.4
~(z) = -,- - - 1
-z 1 - 0.6275z- 1 1 + 0.2872z- 1
1.13 0.298 0.115
1 - 0.247z- 1 ] - 0.0974z- 1 I - 0.038z 1
0.985 0.54 0.265

1 + O.l131z-1 1 + 0.0445z- 1 I + 0.0175z- 1
0.0239
(8.23)
1 - 0.0518z- 1 •••
The output c(n) for n ~ I is readily obtained from the inverse z-
transform of equation (8.23).
c(n) = 60 - 26(0.6275)" - 1.4( -0.2872)" - 1.13(0.247)"
- 0.298(0.0974)" - 0.115(0.038)" - 0.985( -0.113 I)"
- 0.54(-0.0445)" - 0.265(-0.0175)"
- 0.0239{0.OSI8)". . . n ~ I (8.24)
A comparison between the response of equation (8.24) and that obtained

by using equation (8.7) as a recurrence relation is tabulated in Table 8. J.
Agreement is satisfactory and could be improved if higher terms are
considered.
This example has demonstrated that the convolution z-transform can be
systematically applied to the solution of higher-order discrete systems.
Ilowever. the computation becomes exceedingly involved as the order
increases. In practical cases for obtaining the response for a few sampling
intervals, the difference equution yields the response if used as a recurrence
rl'lilticmship. Whcre "'onvcrgence is mpid hy choosing the appropriate
initilll vllhlt~s, the l~(lllIltiun fur ,.(,,) l'oJ1silds only of /I fcw tcrms or response
IUndcli. This flldlitlltCli the lIynthl~lIill pfell'cdure by ()btllinin~ II lIystcm
modification to cancel the troublesome modes in the response. It is from

this viewpoint and for obtaining the response for larger n that the con-
volution z-transform procedure becomes very important. Thus the
system designer should possess all the possible methods of solutions so
that he can use his judgment as to the method best suited to his problem.
TABLE 8.\ COMPARISON OF EXACT AND APPROXIMATE RESPONSES

n Cont'olution t- Tran~rorm Recurrellce Relationship
o 28.6
I 43.88 43.31
2 49.56 49.32
3 53.596 53.43
4 55.96 55.88
5 57.47 57.45
6 58.415 58.43
8.2 Analysis of discrete antenna

array by z-transform method 8 ,It
In this section, a discrete linear array will be treated as a sampled-data
system and thus the z-transform method is readily applicable. Using this
method we may express the array polynomial in a closed form, even for
nonuniform arrays. Furthermore, the analysis of the important character-
istics of the radiation pattern and other useful characteristics can be
conveniently performed using the z-transform theory. This is explained
in the following.
Consider the linear array of n equally spaced, identical elements in
Fig. 8.2. The polynomial associated with such an array can be expressed
,,-I
G(z) = 00 + 01Z- 1 + oaZ-a + ... + a.._1z-(,,-1I = I a,.z-1c
k~O
where (8.2S)
'P = fJd cos ~ + ex

fJ = 21T
- = phase constant
,t
(8.26)
and ex represents the progressive phase lead of the excitation in an clement

compared to that with its neighbor on the left.
!II
~
Jdk- --0--10---0---0------------0
1 2 3 n-I
"
FIGURE 8.2 A linear array of II equally spaced elements.
The coefficients Qu, Ql' ••• , Qn-l represent the excitation of the elements.
For most practical arrays, a progressive phase shift (l exists along the
array and the a-coefficients are used. When all the a's are equal, a
uniform array results.
Suppose that the envelope of the amplitude distribution of the excitation
in the n-elcment array can be described by a continuous function fez)
within the range 0 ~ % < (n - I)d. Then the a-coefficients in equation
(8.25) can be written as
ao =/(0)
Q 1 = fed)
an_I = f[(n - I)d] (8.27)

nnd the array polynomial of equation (8.25) becomes
t."(z) = 1(0) + f(d)z-l + .. .f[(n - l)d]z-ln-ll
..-1
= I I(kd)z-k (8.28)
k~O
This can be written also as

CJ) GO
1,'(1.) = I f(kd)z-It - I f(kd)z-It (8.29)
k-,O It-n
E<luation (8.29) can be expressed as
1.'(;:) = .~(::) - z-'''~9(z) (8.30)
whl'rc
~q(::) = :1rl(1It1 + :r))
1111(1 "is thl' lilllllplill~~ Iwriutl.
For most practical arrays the amplitudes of the excitation in the two
end elements are equal and thus
I[(n - I)d + x] == %[(x) (S.31)
By using equation (S.31), fI(z) [from (S.30» becomes
co
fI(z) == 3[f(nd + x») == I I(nd + kd)z-k
k-O
co
== I [/(n - I)d + (k + I)d)z-Ic
k-O
co
== :±: I J[(k + l)d)Z-k (S.32)
/e.O
From the shifting theorem, we have

3[/(x + d)] == z[~(z) - /(0)] (S.33)
where /(0) is the excitation in the first element.
Substituting (S.33) in (S.30),we finally obtain
8(z) == [I T Z-("-II~(z) ±/(O)z-(II-1I (S.34)
where the plus and minus sign follow from (8.31). Since the z-transform
of many useful nonuniform envelope functions/(x) exist in a closed form,
&(z) in (S.34) is expressible as a closed function of z instead of a polynomial
of n terms (if this problem was treated in the classical way). Note further
that increasing the number of elements n in an array does not increase the
complexity of the expression for C(z).
ApplicQtions
UNIFORM ARRAY OF n ELEMENTS. The amplitude of the excitation can
be assumed to be unity with no loss of generality. Then,
1.4.(x) == u(x) (S.35)
I.~{(n - I)d + x] == I~l(x) (S.36)
1
!F..t(z) == 3[/....(x)] == I -z
-I (S.37)
In this case C(z) in (S.34), using the proper sign, yields

1 l - tl
~ (z) == -
-(11-11
z + z-(n-l1 == - Z (S.3S)
.... 1- Z-I 1- Z-1
LINEAR ARRAY OF n ELEMENTS WITH SINUSOIDAL AMPLITUDE DISTRI-

BUTION. For the distribution shown in Fig. S.3,
/ll(X) == sin fJx (8.39)
"' ....... - - - - - -............. (s(%)

/' "
/ / ",
"--~--o--o--o- -------~ ~ )C
OJ 1 2 3 (II - 1)
.~ d ~
FIGURE 8.3 linear array with sine amplitude distribution.
In this particular case, (n - l)fJd = 7r, and

lu[(n - I)d + x] == -IB(x) (8.40)
The .:-transform oflu(x) is
_
•.1-" u(z) == J[[/I(x» = -2---=-~~
z sin IJti (8.41)
% - 2% cos fJd + 1
Substituting in equation (8.34) and using the proper sign, we obtain
.
Iff,,(z) ==
[1 + %-11I- 1I JZ sin fJd (8.42)
%2 _ 2z cos (It/ + 1
In practical problems other properties of the urray puttern are required,

and among these the array factor thut describes the power pattern of the
array is important. The array factor 1«9(;:)1 2 can be obtained from 6(.:) by
noting that
(8.43)
Hence
1~'(z)12 == C(z)t'(Z-l) = C~:QkZ-k) C~:a~) (8.44)
Using equation (8.34), we can obtain the general formula

16(:)12 = [2 T ::111-11 T z (II lIJ.:F(.:).~(.:-l)
- /(0):[1 + .:111 IlW(:) + [I T ;;. (II lI).F(.:-I)] +.(2(0) (8.45)

To determine the array fuctor for the eX<lmples discussed earlier, we
suhstitute for I, ..,(=) and I, If(:) in expression (8.45) to get
1 z"
1,.,(% I) ==~ (8.46)
1-;:
(8.47)
(HAH,
For the linear array of n elements with sinusoidal amplitude distribution

with
[I + Z-(,,-lI]% sin (Jd
t!H(z) = ~.:....--~~~ (8.49)
Z? - 2z cos {Jd + 1
we get
[2 + zhl-lJ + z-, .. -1I] sin? fJd
It!II(Z)1 2 = --:--''--~----'---~--'-----:-
(2 + Z2 + z-2) - 4(z + Z-I) cos fJd + 4 cos2 (1d
(8.50)
Other properties such as nulls, maxima, and beamwidth can also be
obtained from the z-transform approach to this problem.
8.3 Application to information

and filtering theory 21-32
In some practical cases, for efficiency and economy, It IS necessary to
transmit continuous signals over a sampled-data link. For instance,
accurate transmitting of signals by short wave is reliably done in pulse-
code modulation. Here the sampling operation is part of the pulse-code
modulation. The sampling action causes frequency aliasing and mixing
of high-frequency noise into the signal band. This effect can be remedied
by using a low-pass filter F(s) before sampling to filter out most of the
high-frequency noise. At the receiving end, a wave-shaping filter G(s)
(or hold circuit) is used to reconstruct the continuous signal from the
sampled signa\. Such a sampled-data transmission link is shown in
Fig.8.4.
In this section. we shall "indicate the method of optimizing the system
in Fig. 8.4 by applying both the Laplace and z-transform theory. We can
also use the modified z-transform, but for this case the former procedure
is used because of ease of application.
n(l)
ret)
--+f G(8) : dO)
!I
L _____ -----~--\
,..--....,
L(s) r------------JI c'lm
~
FIGURE 8.4 Block di;lgmm ur IIptimi/alillll ur Nmnpl\."(I.llata u';I/lNmissilln lInk.
APPI.IC'ATIONS TO VARIOUS AREAS OF SYSTEM THEORY 259
Tilt' 0p,imi=alion procedure 27

In this procedure, given the spectral density ¢Jr",(jM), the networks F(s)
and G(s) are determined to give the minimum value of (e(/)]2, where the
input signal 1'1(1) = 1'(/) + n(l) and the desired output is cI(l). Three cases
can be considered; (I) F(s) is fixed and G(s) is to be optimized. (2) the
contrary. (3) the general case where F(s) and G(s) are to be optimized
simultaneously. For brevity. we shall discuss only the optimization
procedure where F(s) is fixed and G(s) is to be optimized. The other cases
can be handled in the same fashion.
We shall assume that the reader is familiar with the optimization
procedure for the continuous and the discrete cases, and therefore many
of the detailed derivations and explanations are omitted; references,
however. will be made to standard texts on statistical design theory.
From Fig. 8.4. the mean-square error is given as
[(>(,)]2 = _I.
27T}
1i
«.
-if¥;
<l> er(.s) ds (8.51)
where the spectral density <l>rc(s) is obtained as follows:

(8.510)
In terms of z-transform the mean-square error at the sampling instants

is obtained [from equation (4.92)] as follows:
[e(nT)]2 = _I .!. ,,(Z)Z-I dz (8.5Ib)

27T} 1:11'1
('j", ...
We can also obtain the mean-square error at all times by using the
modified z-transform. However, as mentioned earlier, this procedure for
this example will not be used. but instead equation (8.51) will be utilized.
The various spectral densities of equation (8.510) can be obtained from
Fig. 8.4 by following certain statistical definitions and concepts, which are27
(8.52)
(8.520)
(8.53)
In equations (8.52 through 8.530) the following symbols are used:
L = L( -s), etc. (8.54)

(FFct>rlr)* = Z(FF<1>Tlr)lz=eT. (8.540)
Substitutil1g equations (8.52) and (8.54) in equation (8.510), we obtain
<1>,. = LL<1>TT - ! FGLcP rTI - J. FGL<f>rlT + 1. (FF<l>r.,)*CG

T T T
(8.54b)
If we let F be fixed, we can obtain G such that the mean square error is
minimized. To find this optimum G, we can use a theorem from calculus
of variations which states the following.
The function G must be such that the partial derivative of <f>" with
respect to G is analytic in the left half of the s-plane (LHP). Hence we
obtain from equation (8.54b)
(8.54c)
where Xl is an unknown function which is analytic in the LHP. From

equation (S.54c) and the requirement on Xl' we notice that the partial
fraction expansion of a<f>,.• joG corresponding to poles in the LHP is
identically zero. that is,
[-FL<f>T1T + (FF<l>rlrl)*G]+ = 0 (8.55)
Because of symmetry, we can write
(FF<1\T,)* = (FFcJ>rlr)*-(FFcJ>rlr)*+ (8.550)
where ( )*+ is analytic in the RHP (or outside the unit circle). Since
G is analytic in the RHP (due to physical realizability), we can obtain a
function Q defined by
(8.55b)
which is also analytic in the RHP.

Using equations (8.550) and (8.55b) in (8.55), we have
[-FLct>r 1r + (FFcJ> .. I rI )*-Q]+ = 0 (8.5Se)
This equation requires that

Utilizing this equation, we can write from equation (8.53a)
<l>ce = .! (FF<1>'lrl)HG{ FF<1>"r,}*-a = .! QQ (8.56a)

T T
To obtain a relationship for the mean square error, we have from
equations (8.53), (8.53a), and (8.54c)
~ (ia) (<l>cci _ <l>cc) ds = ~ (ia) !

J-ia)
21TJ 27rJ L,» T
x [FGL<1>r,r - (FF<1>rIT.)*GG] ds
= _1 (iCO
21Tj J-ia)
(-G 8<1>,.)
iJG
ds
= _ ~ (ia) GXlds (8.56b)

21TJ L,co
Since both G and Xl are analytic in the LHP, this integral vanishes.
Furthermore, changing s into (-s) in equation (8.56b) we obtain the
following expression:
_I. (i"" (<1>C1C _ <l>cc) ds = 0 (8.56c)

21TJ J-i<t;>
From equations (8.51) and (8.51a) we have
[e(t)J2 =~ [iQ)<l>•• ds
21TJ Lia)
= ~ [iCO (el<l - <1>Clc - <1> cc l + .c) ds (8.57)
21TJ Li""
This integral can also be written as
- ~ L;:;o
(ioo ( eel
21TJ
- cc) ds (8.57a)
Noting equations (8.56a) and (8.56c), we finally obtain
[e(/)]2 = _1
21Tj
ji<'- (<1>
j' rlrt
_ QQ) ds
T
(8.57b)
1I.1.IIS ntA'IIVE EXAMI'L);

Thi" l~xall1plc illustrate)' a prohlclll of optimulll filtering and prediction in
thc prl'loCl1l'C "I' IlIll'orrl'latl'd l1oisc.
Referring to Fig. 8.4, assume that

e1(t) = r(t + IX) (8.s8a)
L(s) = e"' (8.58b)
20
Cl>r.r(s) = 2 2 (S.sSe)
a - s
2a u2(b2 - Sll)
Cl>r,rl(s) = II
a -s
II + a'l. = 2
a -s
'I. (8.sSd)
where b2 = 2a/a2 + 0 2•
b-a
F(s) = - - (8.sSe)
s+b
We shall determine the optimum G(s) and mean square value of the
error as well as the error due to sampling.
From equation (S.56) and the expressions (S.sSa-e), we have
Q(s) = [(b - a)/( -s + b)e"'(2a)/(a l - S2)] (S.s9)
[ (b - a)2a2]*-
a2 _ S2 +
From tables of z-transforms (see Table I) we obtain
[ (b 2- a)la
a _
S
S2
]* ~ Z[(b - a2 _
a)2a']
S2
= a 2(b _
2a
0)2
22 -
z(e- aT _ e+IIT )
(e"T
I
+ e-aT)z + 1 z=eT.
(8.s9a)
This equation can also be written in the following form:
(8.59b)
Substituting the last bracketed term of equation (S.s9b) into the

expression for Q(s) in equation (S.s9), we get
Q(s) = [
2ae'" ,J2a 1- e-aTeTI]
a(s - b)(s - ales + a) JJ - Ir'laT I.
J2a(1 - e 2<.7')" 1111

== •. ....... - .
,r(CI + /J)(.~ + t')
O(s)
FW T I ,
ret) + n(t) ..
I~Prefilter
~ ~1-e-aTz-1H 7ii
Digital filter Shaping
I
filter
FIGURE 8.S Block diagram for the optimized sampled-data transmission link.
From equation (8.55b) we finally obtain the optimum G(s) as follows:
Gopt(s) = a (b 2a-
I II B(1 - e-a e-2")--2' e-tl.a (8.60)
a) S +a
but 2a/a2 = b'l. - all; hence

-tJa
GuPI(S) = (1 _ e-tl.2'z-l>!.•••". _e__ (8.60a)
s+a
It is noticed from the preceding that G(s) is composed of a digital
filter and a shaping filter_ The block diagram of the optimized system is
shown in Fig. 8.5.
The mean-square error is given by equation (8.57b), which gives
-- (b )(1 -taT) -'l.tl.R
[e(I)]2 = 1 _ - a - e e (S.60b)
2aT(b +a)
As T approaches zero, equation (8.60b) gives the result of the Wiener
filter. that is.
- [
e(l) ]21 T.,O = I - -
b --a e-'I.tI. (8.60c)
b+a
The error due to sampling is given by
[e(1)]2 _ [e(I)]2IT~O = (1 _ 1-;:;'l.tl.2')(b ; :);-2za (8.61)
8.4 z-Transform method applied

to problems of economics
National income, consumption, and investment are the three basic
fundamentals of any economy_ These can be expressed at any given time,
t = nT, by mcans oCa difference equation and certain constants, Solution
uf Ihese C(IU:llions is Ihen found by the ::-trunsCorm method, Three models
of cl'unulIlil' systems havc hecn considercd and they involve first-. second-
and fuurth,"nler dint~n'l1l'" '~'IUUliClllli with l'nnstunl l'ucflicicnts, ('on·
wr.~'·IU:l· nf 111,· snlulinn IIlId slllhility IIrt~ nlsn disl'usscd in Ihe fllllnwin~,
Model I
Let y, e, i denote the national income, consumption, and investment,

respectively. Considering that time is divided in periods of equal length,
say years, we define that at any time t = nT (where n can have integer
values 0, 1,2, ...) y, e, i have respective values Yn' en' in.
Since national income is made up of investment and consumption, we
can write
(8.62)
Now we assume that consumption depends linearly on national income
but has a constant value even when there is no national income. This is
quite logical; therefore we can write
n = 0, ],2, ... (8.63)
where K ~ 0 and 0 < m < I.
We must also consider the effect of investment on national income which
can be expressed in terms of what the economists call the growth factor r.
We can assume the following relationship:
Yn+1 - Yn = rin (8.64)

where r > O.
I{clations (8.62), (8.63), and (8.64) enable us to determine the desired
clill'crence equation. Eliminating in and en, we get
(8.65)
01'
Yn' J + (mr - r - J)Yn + rK = 0 n = 0, 1',2, . . . (8.66)
Suilltioll of this difference equation can be easily found by taking the
,·-tmnsform
,Y[!ln ,d + (mr - r - I )J[Yn] + J[rK] = 0 (8.67)
Ill'
rKz
z[!V(z) - 110] + (mr - r - I)lt?/(z) + -- = 0 (8.68)
z-l
whit'll Aives
rKz
,!I/(::)(o: + mr - r - I) = zYo - - - (8.69)
%-1
Ill'
:I/(Z) :-: • Z?J.u - - -.. -- .

rK:::------
-.,,,,~.
Z - (I +r - ",r) (::: - I)f:: - (I + r _.. ",rll

(H.7())
Now I + r(l - m) > O. So let I + r(l - m) = a.
~(z) = Z1Io + rKz

z- a (z - 1)(z - a)
Z1l0 - rKz/(a - 1) rKz (8.71)
= -
z- a (1 - a)(z - 1)
Taking the inverse z-transform of equation (8.71), we get
rK )"
11,,= ( 110--- a - -rK- (8.72)
a-I I-a
or
K
11" = 110[1 + r(J - m»)" - - - ([I
I-m
+ r(1 - m»)" - I}
(8.73)
Rewriting,
11" = [1 + r(l - m»)"(1IO - ~)

I-m
+~
I-m
(8.74)
This gives the national income for any n = 0, I, 2, ...•
Model/l
In this modeJ, we define at any time t = n (n = 0, I, 2, ...),
11" = income received

Sn = total savings
;n = desired investment
We make the following assumptions:
I. Savings during any period is proportional to the income anticipated
during the next period which can be mathematically expressed as
(8.75)
where K is the proportionality constant
2. Desired investment during any period is equal to a constant multiple
X of the increase of the income of that period over the income of the
preceding period. This can be expressed mathematically
;,,11 = g(tl" I 1 - 1',1) (8.76)

J. Savin~s in any perind is cllLIll1 to desired investment
(M.77)
Equations (8.75), (8.76), and (8.77) can be combined to yield the following
difference equation:
g g
1In+1I - K 1In+l + K1In = 0, n~O (8.78)
Taking the z-transform of this equation, we obtain
ll9'(z) = 1.1110 - (gl K)zyo + 1.111 (8.79)

1.'1 - (g/K)z + glK
Using the inverse z-transform integral or tables, we obtain
1In = 1Io{G~ + J(~J- ~r+1

- ~~ -J(2~J- !T+1}/ J(g/K)2 - 4(g/K)
+ (Yl -! yo){[ii + J(gI2K)2 - glK r

- [2~ - J(g/2K)1I - glKT}/J(gIK)1 - 4(gIK),
n ~ 0 (8.80)
For stability or convergence we obtain from the denominator of
equation (8.79) '-(1.) the following conditions:
1. '-(1) > 0, or 1>0 (8.81)
2. '-(-1) > 0, or 1+ 2g >0 (8.82)

K
3. '-(0) < I, or g< K (8.83)
Mode/Ill
In this model we obtain, as shown by Tinbergen,13 the numerical equation
for the analysis of economic fluctuations. The numerical equation which
is determined from statistical consideration is a fourth-order difference
equation, given asla
YnH + DYn+a + byn+l + CUn+! + dun = 0 (8.84)

where
a = -0.398 b = 0.220 c = -0.013
d = -0.027 (8.85)
The characteristic equation for (8.84) which can be obtained by taking
the z-transform can be expressed as
F(z) = Z4 - 0.398z3 + O.220Z2 - O.Ol3z - 0.027 = 0 (8.86)
To determine stability for the system of Model III, we require for the
fourth-order case
I. F(l) > 0, that is, I - 0.398 + 0.22 - 0.013 - 0.027> 0
F( -I) > 0, that is, I + 0.398 + 0.220+ 0.013 - 0.027 > 0
(8.87)
which are both satisfied.
2. (0.027)2 - 12 <- 10.027 X 0.398 + 0.013 x II,
that is, -0.999 < - 10.023751 (8.88)
which is satisfied.
3. (-0.027)3 + 2( -0.027)(0.220) + (-0.013)(-0.398)
- (-0.027) - (0.220) - (-0.027)(-0.398)2 - (-0.027)2
- (-0.027)2(0.220) - (-0.013)2 + I
+ (-0.027)(-0.013)(0.270) > 0 (8.89)
which is also satisfied.
Therefore the system as described by equation (8.84) is stable.
8.S Linear sequential circuits14-17.37

In this section the z-transform method is used to analyze the behavior of
linear sequential circuits. Such devices find application, for example. in
coding theory, radar detection systems. and computer theory. The under-
lying mathematical field for analysis is not the field of real or complex
numbers; thus the z-transform theory is carefully applied in the following.
This discussion is limited, for the most part, to the autonomous behavior
of such circuits.
Sequential circuits are devices whose inputs are sequences of numbers
belonging to a finite field. When they are linear and have a single input
and output. the sequence x(n), its input, and yen) the output, may be
related by the following difference equation:
yen + k) + (J.11f(n + k - I) + ... + (J.oJ/(n)

= /10:.'(" + k) + (1.:r(I/ + k - I) + ... + /f...r(,,) (8.90)
In clillatioll (H.II0) IIIl1ltiplil'ation and additiull arc 1II(1(Iulo,I' alld the

vah,,'!.Ill' rt" {I,. !I. nlld .•. 1'I\1I'~l' OWl' tl1I' illh'I~"I'~ n. 1,2 •. l ..... " I.
268 THEORY AND APPLICATION OF THE z-:TRANSFORM
where p is a prime number. The set of numbers 0. 1.2•...• p - I with

modulo-p arithmetic operations is called a modular field.
The following shifting property of the z-transform remains valid when
arithmetic operations are carried out modulo-p, that is,
Tc-l
.1[x(n + k)] = zt,i"(z) - zt Ix(l)z-& (8.91)
1-0
Using this equation in obtaining the z-transform of equation (8.90). we

get
(z'" + «IZ"'-1 + ... + «... ]W(z)
= fPoZ" + (J1Z"-1 + ... + fJ. . ]!l'(z) + /(z) (8.92)
where I(z) is a polynomial in z resulting from the initial conditions. Its

exact form is not important for the following discussions.
To calculate yen), we solve for '?V(z) and then calculate its inverse. The
usual inversion integrals or tables are not directly applicable in this case
because the modular field is not the field of real or complex numbers. Thus
it is more convenient to resort to the power series method of finding the
inverse z-transform. *
The autonomous behavior of the circuit is determined by letting x = 0
in equation (8.90). Thus from equation (8.90) with % = O. it is easy to
notice that any k consecutive value of y(n). say (Yo, YI. Y2.··" Yk-I),
completely determines Y for all subsequent n. We may observe that the
number of distinct k-tuples (Yo, Yl •.•.• Yk-I) is p".
Consider the sequence of output symbols Yo. Yl' Y2' . .. Since the
number of distinct k-tuples is finite. there exist indices m and m + L such
that (y m' Y mU ••••• Y m+k-l) is identical to (y m+1.' Y." ~ L+I' ••• , Y rn U.+k-l)'
Since k consecutive values of y determine all subsequent values of y, the
output following Y m+L+k-l is identical to that following Y m+.l:-l; hence the
output sequence is periodic after Y m' The smallest integer L is called
the period of the sequence. The exact period of the sequence may be deter-
mined from equation (8.92) with ~(z) set equal to zero.
11/(z) _ I(z) I(z)

- z. . + «IZ"'-1 + «aZIt-B + . . . + «" =--
P(z)
(8.93)
Three cases of P(z) may be distinguished: (I) P(z) is irreducible,

(2) P(z) factors into distinct irreducible polynomials, and (3) PC;:) factors
but not into distinct polynomials. Each of these cases is discussed in the
following.
• However, with due care in interpretation, the inver~i()n integrals and tahles II1I1Y niNO
be used. Thus a solution in clo~d form cnn he.found.
P(z) Irreducible
A polynomial with coefficients belonging to a field is said to be irreducible
if it is not factorable into polynomials of smaller degree with coefficients
belonging to the same field. In this case the coefficients of P(z) belong to
a modular field of characteristic p. It is well known that an irreducible
polynomial with coefficients in a modular field is a factor of a polynomial
zL - I for some L that must be a divisor of pk - I, k being the degree of
the polynomial.
Using this fact,
ZL - 1 = P(z)R(z) (8.94)
where R(z) is a polynomial of degree L - k in z. The smallest L for

which equation (8.94) is satisfied is called the period of P(z). Solving for
P(z) in (8.94) and substituting in (8.93), we obtain
·!p(z) = z-L /(z)R(z) = z-LI(z)R(z)(J + z-L + Z-2/. + ... )

l-z- L
(8.95)
From (8.95) it is clear that y(k) = y(k + L) for all k, therefore the period
of yen) is at most L. (It can be shown that the period must be exactly L.)
When pl- - I is prime, L = pk - I. Then, the circuit is said to have
maximal period, for pI.; - I is the largest value L may assume. Except for
the all-zero k-tuple, every possible k-tuple appears once and only once
during each period.
PC:;;) factors into nonrepeated faclors

Suppose P(z) factors into distinct nonrepeated factors
Then the period of yen) is a divisor of the least common multiple of the
periods of p.(::), P2(z), ... , P,C::).
Tn sec this, make a partial fraction expansion of equation (8.93),
:1/(::) = 9'(z) + qz(z) + ... + qlz) (8.96)

1'.(::) 1'2(z) P,Cz)
1::II:h It'rlll "'(")/1',('), i .. 1,2, J" . , ,I in ~'(Iu:lti(ln (R.9(,), corresponds
III a Sl'llUt'm'(' wilh p,'riml Ihal is a divisor Ill' Illt~ pl'rilld "I' 1',(;:). The
!'t'riml Ill" !/(,,), whlt'h i~ I hl' !lIIIII (If' I ht' !i("IIJ('lIl'l'!I ""rr('s!,lllldin~ tn
q;(z)/P,(z), as a consequence, is a divisor of the least common multiple of

the periods of P 1(z), Pa(z) • .••• P,(z).
The maximum period of yen) if P(z) factors is always less than pk - I.
This is so because the period L of yen) (supposing the degree of P;(z) is k i )
satisfies the inequality
L ~ (pka _ I)(p'" _ I) ... (Pkl _ I) <pkJ+kl+ •.• +1c1 - 1
= pk _ 1 (8.97)
P(z) Has repeated factors
To compute the period of yen) for this case. the following lemma which is
proved by Elspas14 is needed.
LEMMA: If [q(z)J is irreducible. [q(z)]m has period kpTm where k is the
period of q(z) and r m is the smallest integer satisfying pTm ~ m.
Now, suppose
(8.98)
Thus. just as for the nonrepeated factor case. the period of yen) is a
divisor of the least common multiple of the periods of each factor pNz).
In the preceding. it has been tacitly assumed that in equation (8.39) /(z)
and P(z) have no common factors. This is always so when P(z) is irre-
ducible. In the last two cases discussed. only upper bounds on the period
were established. The reason more precise statements. in general. are not
possible is that the periods depend on I(z). that is. the initial conditions.
8.6 Application to discrete Markov processes18 - 21

The z-transform's application in probability theory is well known and
often used. This method is referred to as the generating function method
and was used by De Moivre and Laplace as early as the nineteenth
century. In the foHowing discussion, the z-transform method is used
in the solution of the vector difference equation arising in discrete
Markov process. Certain features of the process are discussed using this
method.
We shaH consider the problems of finite-state. discrete Markov process
and restrict the discussion to an irreducible. regular Markov process in
which there are neither transient states nor absorbing states and the state
transition of all persistent states is aperiodic..
We define a set of states
S={1.2.3, ...• m} (8.99)

APPLICATIONS TO VARIOUS AREAS OF SYSTEM TBEORY 271
Also, we define a column vector Pen) and a square matrix p(u, I) such that
PIC,,)
P3(1l)
P(n) = (8.100)
P",(n)
p(u, n) = [... P'J(u, II) •.. ], (8.101)
where P,(n) is the absolute probability of state i at time nand Pij(u, II) is
the trunsilion probability from a state i at time II to a state j at timc n.
(Here u and n are taken to be intcgral numbers.) They have the following
properties:
'"
!Pi/(u, II) = I, for all ; s and for II,
1=1
" integers with II ~ :t ~ 0 (8.102)
p;;(u, n) ~ 0 (8.103)
and
(8.104)
Using the matrices of equations (8.100) and (8.101), we can write

pen) = p'(u, n)P(u) (8.105)
where the prime indicates the transpose of a matrix. If the process is

time-homogeneous, the stochastic matrix p(u, n) is time-invariant and is
given by
p'(II, n) = p'(n - u) (8.106)

Since the process is a Markov one, the Chapman-Kolomogorov equation
must be satisfied, that is,
p'(u, n) = p'(u, u ..... T)p'(U + T, n), n~ T +U~ u (8.107)
This clJuation for II time-homogeneous process reduces to
p'(,,) = p'(r)p'(n - r) n ~ r ~ 0,' T = integcr (S.I08)
wilh Iht' kllnwll·t1'~I· Ihal

1,'(11) "kill/IV Illall'i\ (H. lUI))
(fwe define p(l) = p, by using equation (8.108) as it recurrence relation

we have
pen) = p", or ,'(n) = p'" (8.110)
Therefore, from equations (8.105) and (8.106), we obtain
pen + '7') = P'(T)P(n) = p"P(n) (8.111)
In particular. letting T = 1 in equation (8.111), we obtain the following

first-order vector difference equation:
Pen + 1) = p'P(n). n = 0, 1,2, ... (8.112)
Application of the z-Iransjorm

We shall apply the z-transform to the vector difference equation in (8.112)
ns follows (with the understanding that the z-transforms of vectors and
Il1ntriccs are obtained by taking the transform of each component in the
army).
."
1[P(n)] = D'(z) ~ IP(n)z-" (8.113)

,,=0
By the time-translation theorem,
,},[P(n + 1)] = z[&'(z) - P(O)] (8.114)
Tht'rl'lilrc, substituting equations (8.113) and (8.114) into equation (8.112),
(:I - p')D'(z) = zP(O)

D'(z) = z(d - p')-lP(O) £ ~(z)P(O) (8.115)
whl'rc ~(.:) is the fundamental matrix of the stochastic matrix p' such that
~(::) = z(d _ p')-1 (8.116)
I.nter we shall show the relation between ~(z) and pen).

We assume here that there is no multiple pole. Then by the partial
fraL'linll cxp~tnsion of equation (8.116), we have
~(::) =I"I --G",

II:
(8.117)
k I z - Ale
whcre the il,.'s lIrc the eigenvulucs ofp', that is the roots or the characteristic
l'tillal inn
1}.1 - p'l ,,= 0 (8.11 8)
Then Wl' haw Ihl' filllnwinl.~ thcnrclII.

APPLICATIONS TO VARIOUS AREAS Of SYSTEM THEORY 273
THEOREM I: For any regular stochastic matrix p of a finite-state, discrete-
lime Markov process
(i) There exists an eigenvalue of p', which is equal to one.
(ii) No eigenvalue of p' exists outside of the unit circle in the complex
z-plane.
Proof: Let us consider a determinant
Pll - I P21 PfI,l
PI~ P22 - I P,"2
Ip' -II =
Plm P2m ... Pmm - I

If we add all the rows from 2 to m to the first row, all the elements of the
first row become zero because
711
IPii - I = 0, Vi E S
i-I
Therefore we have
Ip' - II = 0 (8.119)
Comparing this with equation (8.118), we see that there exists an eigenvalue
which is equal to one, which thus proves (i).
The proof of (ii) is shown as follows. Let X be an eigenvector of p and
M be the absolute value of the component of X of greatest magnitude.
Then from the relation
AX =pX
we have
1ft
l..ll M ~ M. I;=1 Pii = M (8.120)

Therefore we get
(8.121)
Thus, no cigcnvulue of equation (8.118) exists outside the unit circle.
In fact, the term in equation (8.117) corresponding to .ill = I is the
stationary solution of the Murkov process. We can verify it in many
ways.
II" 111l~ prm:l'!oos i!oo slalillllary,
VII; illh',.,t'J"
and equation (8.112) can be rewritten as

(I - p')P(n) = 0 (8.122)
where P(n) is no longer dependent on n. From equation (8.119). we can
say that there always exists a nontrivial solution for P(n) corresponding
to the eigenvalue .ill = I.
Fundamental matrix
We will now study more about the fundamental matrix defined by equation
(8.116). If we assume that .ilk·s are the simple roots of the characteristic
equation, W(z) from equation (8.117) is expanded by
m z
<9(z) = I -- G k
k~l Z - .ilk
where G k is the residue of <9(z) and is
Gk = lim (z - Ak)(zI _ p,)-l (8.123)
z-At:
We further assume that

for all k = 2, 3, ... , m (8.124)
From the final value theorem of the z-transform
= lim
G(n)I .. _a:
.-1 (z - l) <9(z)
(8.125)
We see here again that the first term G 1 corresponding to A1 = I is the

limiting fundamental matrix of the stationary Markov process. Since
1.il,,1 < 1 for k > I, the state transition is always stable and the terms
other than the first converge to zero as n increases. This is quite natural
because Pen) itself is bounded by one.
By the inverse z-transform of W(z), we can find the fundamental matrix
in time domain
G(n) = r1[~(z)]
= _1_ f ~(Z)Z"-l dz
27Tj 11'
=1-
27Tj
f m z"
I--Gkclz
I'k I Z -,t"
(!U26)
where r is taken to be a counterclockwise closed contour on a unit circle
in the complex z-plane. This is given by the sum of residues
m
G(n) = l:'~:Gk (8.127)
kal
Since IAkl < I for all k(2, j, ... , m), the terms from 2 to m converge to
zero as n - .. 00. When Ak is a complex root, the kth term is an oscillatory
attenuating term, that is, denoting
Ak = ak + jbl< = e-I«r.-iUlk). (8.128)
the attenuating constant OCI< and the frequency (I),. are given by
(XI< = - ~ In (a: + b:) = -In IA"I
(I)" = tan -1 (01<)

b,. = ang Ak 1 (8.129)
Since A.k is in the unit circle, (Xk should always be positive.

By the inverse z-transform of equation (8.115), we have
Pen) = G(n)P(O) (8.130)
In general,
Pen + T) = G(T)P(n)
Comparing this with equation (8.111), we note that

G(n) = p'(n) (8.131)
~(z) = J[p'(n)] (8.132)
The initial value of G(n) is
G(II)ln=o = lim ~(z) = I (8.133)
,"co
Furthermore, from equation (8.125),
G1 = lim G(n) = lim p'" (8.134)
Therefore G 1 is the limit of the nth power of the transpose of the transition
matrix.
REI'ERI:NCI:S
I. Jury. L I.. SIII"I,/cIC/.I)CIICI Clltum/,C,)·.rlt'",." John Wiley and Sons, New York. 1958.
2. Tsypkin. Y. Z. 17,c'(,ry IIf' I'ul.w' S,'s/('/II.\·, .·hysicn·M:llhc,muiclIl '.ilcrlIlurc. Muscow,
I'J'iK.
I. .Iu,.y, I. I., "C '11111111mlilln 141 II,,· Mmlilh',I:·TI'I1I1,1'1I1'111 Ml·thllll," .I. ,.hlt/H,)' III.~/.,
VIII. UII, Nil. J, AII/:II',I 1%11,1'1" II"J I.t'J.
4. Jury, E. I., and M. A. Pai, "Convolution z-Transform Method Applied to Certain

Nonlinear Discrete Systems," IRE, PGAC, Vol. AC-7, No. I, January 1962.
5. Pai, M. A., "The Analysis of Nonlinear Feedback Sampled-Data Systems," Ph.D.
Thesis, September 1961, University of California, Berkeley.
6. Montel, P., LefonS SlIr les Recurrences et lellr Applications, Gauthier-Villars, Paris,
1957.
7. Schelkunoff, S. A., "A Mathematical Theory .of Linear Arrays," Bell System Tech.
J., Vol. 22, January 1943, pp. 80-107.
8. Cheng, D. K., and M. T. Ma, "A New Mathematical Approach for Linear Array
Analysis," IRE Trans. Anlennas and Propagation, Vol. AP-8, No.3, 1960.
9. Cheng, D. K., and M. T. Ma, "A Critical Study of Linear Arrays with Equal Side
Lobes," IRE llIternatioflol COllt'l.'IItion Record. Part I. 1961, pp. 110-122.
10. Stone, W. M., "A List of Generalized Laplace Transforms," Iowa SIOI(' College
Journal of Science, Vol. 22, April 1948, pp. 215-225.
11. Goldberg, S.,llItrodllctioll to Difference Equalions, John Wiley and Sons, New York,
1958.
12. Samuelson, P. A., "Conditions that Roots of a Polynomial Be Less than Unity in
Absolute Value." Allnals of Math. Slat., Vol. 12. 1\141, p. 360.
13. Tinbergen, J., Business Cycles in the United States, 1919-1932, League of Nations,
1939, pp. 126-150 (Economics Intelligence Service, Gcneva, Switzerland).
14. Elspas, B., "The Theory of Autonomous Linear Sequential Nctworks," IRE TrailS.
Cirmit Theory, Vol. CT-6, No. I, March 1959, pp. 45-60.
15. Friedland. B., "Lincar Modular Sequential Circuits," Ibid.• pp. 61-69.
16. Harlmanis, J., "Linear Multivalued Sequential Coding Nctworks," ibid., pp. 69-74.
17. Huffman, D. A., "The Synthesis of Linear Sequcntial Coding Networks, Proc('ed·
i'\~s
of the Symposium Oil b!formation Theory, London, England, 1955, pp. 77-95.
18. Feller, W., All Introdllctio" to Probability Theory and its Applications, John Wiley
and Sons, New York, Vol. I, 1960.
19. Sillier, P. W., "Systcm Analysis of Discrete Markov Process," IRE PGCT, Vol.
CT-3, No.4, December 1956, pp. 237-266.
20. Howard, R. A., Dynamic Programming and Markov Processes, MIT Press and
John Wiley and Sons, New York, 1960.
21. Bellman. R., Introduction to Matrix Analysis, McGraw-Hili Book Co., New York,
1960.
22. Van der Waerdon, B. L, Modern Algebra, Vol. I, F. Ungar Publishing Co., New
York,1945.
23. Mcschkowski, H., Dijji.·rellzengleicllullgen. Gotlingcn, Vandcnhoeck and Ruprecht,
1959.
24. Murphy, G. J., and J. F. Egan, "The Analysis of Demodulating Compcnsating
Networks," IRE Trans. Automatic Control, Vol. AC-4, Dccember 1\159. pp. 71-79.
25. Brule, John D .• "Polynomial Extrapolations of Sampled-Data with lin Analog
Computer," IRE Trans. AUIO/llalic Comrol, Vol. AC-7, January 1962, pp. 76 77.
26. Cattermolc, K. W., "Enlcicncy and Recipf"llcity in l'ube-Alllplitmlc Mudulati,ltI,"
Proc. lEE, Part a, Septcmher 19511, PI" 449 4(.2.
APPLICATIONS TO VARIOUS AR.EAS Of SYSTEM THEORY 277
27. Chang, S. S. L.. Synf"('.~is of Oplilllllm COlllrol S.I'SIt!ms, McGraw-Hili Book Co.'
New York, 1961. Ch. 6.
28. Jury. E. I., "Optimization Procedure for Sampled-Data and Digital Control
Systems," 5c'it!nfiu Ekclrico, Zurich, Switzerland, Vol. VII, Part I, 1961, pp. 16-26.
29. De Russo, P. M., "Optimum Linear Filtering of Signals Prior to Noise." AlEE
Trolls., Part II. Vol. 79, 1960, pp. 549-555.
30. TheJlier. P. L., "Optimization et Auto-Optimisation des systemes de Commande it
Donnees [chantillonnces, en Presence de Saturations," Doctor of Science Thesis,
1962. Facuhc des Sciences de L'universitc de Grenoble, Paris Impremiere Nationale,
1962.
31. Nishimura. T., "On the Modified ,,-Transform of Powcr Spectral Densities,"
IRE TrailS. AIIICJfI1atic C()IIlrel. July 1962, pp. 55-56.
32. Nishimura, T. and E. I. Jury, "Contribution to Statistical Design of Sampled-Data
Control Systems," Institute of Engineering Research Report, Series 60, No. 210,
July 1958, University of California. Berkeley.
33. Budnicki. Zdzislaw, "Input Impedance and Transfer Function of Ladder Network,"
U:F.f:, PGCT, Vol. CT-IO. No.2. June 1963, pp. 286-287.
34. Urkowitl. H .• "Analysis and Synthesis of Delay Line Periodic Filters," IRE. PGCT,
No.2. Vol. CT-4, June 1957, pp. 41-53.
35. Rohinson, E. A., "External Properties of the Wold DecompOSition," J. Math.
AII(I~\'sis tlIIII Appliw(iolls, Vol. 6. February 1963, pp. 75-85.
36. Murphy G. J .. "The Analysis and Design of Production and Inventory Control
Systems". Proc. Ntuiollal Elec(ronic.l· Conferellc('. Vol. 19. 1963.
37. Iwens. R. P.. "Application of the :-Transform to Sequential Circuits," Master of
Science Pr(lject. Dept. of Electrical Engineering. University of California. Berkeley,
JHnuary 1964.
3!!. Higgins, T . .I. and J. K. Shah, "Basic Theory. and Application in Circuit Analysis
of the Finite Alternate z-Transform," IEEE Conference paper No. CP 63-1467,
Presented at Chicago. Illinois, October 28-30, 1963.
39. Higgins, T. J. and Robert E. Oslerlei, "General Theory of Finite z-Transforms
and Applications to Analysis of Discrete Systems," IEEE Conference paper No.
CP 63- 970, Presented al Toronlo. Ont., Canada, June 16-21, 1963.
40. Christiansen, Peter L., "On lhe Closed Form of the Array Faclor for Linear
Arrays," 1£££ T/'{/lls. Oil AIlIt!t1/1QS olld Propagalioll, Vol. AP-J J, No.2, p. 198, March
1963.
APPENDIX
TABLE I. z-TRANSFORM PAIRS 1.-Transform
.F(z) = 1[f(n)), 11.1 >R
GO
Discrete Time-Function
= If(n)1.-n
Number f(n), n ~ 0 n=O
n ~0 1.
un
() =0,Ifor
otherwise 1. - 1
1.
2 e-o.n
z - e-e&
z
3 n
(z - 1)2
1.(1. + 1)
4 n2
(z - 1)3
1.(1.2 + 4z + 1)
5 nS
(z - I)'
%(zS + 11%2 + liz + 1)
6 n'
(z - 1)5
%(%4 + 261.3 + 661.2 + 26% + I)
7
"' <_I)kD'«_1._); D
(z - 1)8
=z~
8
""* z - I dz
9 u(n-k)
z - 1
10 F(e"1.)
11 ,,(2) = lI(n - 1) z
2 "'"(1.-----,1""")3
z
,,(3) = nO - 1)(n - 2)
3! (z _ 1)4
1.
13 nCk ) = n(n - 1)(11 - 2) ... (11 - k + 1) k! (1. _ l)k+1
14 n(kr(n),I1(k) = n(n + l)(n + 2) ••. (n + k - 1) dk

( _1)"zk d1. k [.F(~)]
= d1.d k .F(z)
k
15 (-l)kll(n - 1)(n - 2) ... (n - k + lif"-k+lt %.Fc'")(z), 3OCI:I(1.)
16 - (II - lifn-l .F(l)(t)
17 (-I)k(n - l)(n - 2) ... (n - k)fn-k .FCkl(t)
18 nf(n) -z.F(1)(z) .
19 n"l(n) . z2'{lr)(~) +
z.Fci)(~) .""
~o nSl(n) '..:. ~ly(3)(z) ":";-3z 2,Fi2)(z) _ z.FU)(z)
n!
.,., (In c)"
n!
• Table IV represents entries for k up to 10.
+ It may be noted that In is the same as (n)
t-.)
TABLE I (Continued) z-Trans[orm 00
0
§(z) = J[[(n»), Izi > R
=
<Xl
1 [(n)z-n '~"'
:t
Number fen), n > 0 n=O
+ el
">
-<
23 (' n)ena k - n, (~) =
k!
(k - n)! n!'
n~k
(a z
zk
:z:
0
zk+l >
"Cj
"Cj
24 (n : k)cn C
(z - c)k+1 (')
>
cn ::!
25 I
n. '
(n = 1,3.5.7, ...) sinh (;) 0
:z:
0
en
26 I (n = 0,2,4,6, .•• ) 'cosh (;) ""
n. ' '"'
:t
III
%sin oc to
I
27 sin (IXn) ~
z2 - 2% cos ex +1 >
:z:
z(z - cos ex) '"~
28 cos (exn)
z2 - 2z cos ex + 1
"3:
z2 sin'll + z sin (ex - '1')
29 sin (IXn + lJI) z2 - 2z cos 0: + 1
z(z - cosh oc)
30 cosh (exn)
z2 - 2z cosh a. +1
z sinh a.
31 sinh (a.n)
z2 - 2z cosh a. +1
z
n>O In--
II z - 1
z - e-7.
33 a. + 1n---. «>0
n z- 1
sin 'Xn sin x
n ex + tan- 1 , 0(>0
z - cos 0(
cos exn z
35 n>O In -;=~~===",:,
n
.Jz2 - 2% cos ex + 1
+ + 2) ... (II + k
36
(II l)(n
(k - I)!
- 1)
(1 - -1)-'-,%
k = 2,3, ...
" I z z
37 !-
lII=lm
--In--
%-1 z-1
n -1 1 ell:
38 I,
m=om. z - 1
( -1 )(11-1»/2
39 for n ~ p and n - p = even
2nt ; p)! (n ; P)!'
= O. for n 

~
III
n ;06 mk Z
S!
41 (lIIp (x) ~
= 2nn!
n -)
dx
(d'"
(.t'2 _ 1)" z ><
n
.JZ2 - 2xaz + a2 t-.)
00
N
TABLE I (Continued) z-Transform co
N
9>(z) = "[f(n)1, Izi >R
CD
-i
= If(n)z- Sl0
Number fen), n ~ 0 ,,-0 ~
z(z - az) >
42 anT,,(z) = a" cos (n oos-1 z) z2 - 2zaz + a2
2:
c
>
-=I (n)<-_z)r
L,.(z) _Z_ e-IC a-ll ""C
CIO 'U
43
n! r=O r r! z -I (')
H,.(x) ["/2) (_I),,-kx,,-ak g

=k~ok!(n -2k)!2A:
44 e-ta-l/hl 2!
7 0
'II
45 a"P::(x) = a"(1 -z2)m/l(!fP,,(x), m = integer

(2m)! z"'+1(1 - zl)"'/la'"
2"'m! (z2 - 2zaz + a2)"'+1/2
-i
:c
111
lit
46 L::(x) = (~\". L,,(z) m = integer ( -1 )"'z e-lf/l:o-ll ~

>
n! dX} n! ' (z - 1)"'+1
31't:I
·47 --,-I
II
1
Z
[~'( z)
.!"(z)
{I'(z) ]
---z--
{I(z)
.
,where '-(z) and !fez) are rational
9>(z)
In ~(z)
0
lIII
~
polynomials in z of the same order
48
m(m + I)(m + 2) ... (m + n)
(m - I)! z'" [ ell" - I
... -1
k=O
1
I"""'i
k. z
J
49
sin (exn)
n!
in «)
eCOB fl./z • sin -z- e
cos (~n)
SO /foa fl./z • cos ('in
-z-«)
n!
51 I"
k~O
!i.gn-k 9><z)~(z)
n
d9>(z)
~: I k!kg,,_A: - 9>1l)(z)~(z), 9>(I)(z) = _ _
k=O dz
to
!3 I
k~O
k'f,.g"_k 9>111(z)~(z)
ex" + (-«)"
~4
2«1 z2 _ «2
ex" - pn z
~5 ---
«-fJ (z - ocXz - fJ)
56 (If + k)lkl k Z
k! z (z _ I)k+l
57 (If - k)lkl ,-k z

k. z (z _ J)k+l
(n T k)'ml zl =F kemfl.
58 e"',,-kl
m! (z - efl.)tII+1
1 >
ii!2:""
59
'IT
-sin-n 'IT 1
n 2 - + tan-I-
2 z
cos ex(2n - 1)
!a
1 z+2v';cosoc+1 ><
60 n>O -_In _
2n -1 N
4v'z z - 2v'zcos« + I co
c...o
N
TABLE I (Continued) z- Transform 00
~
F(z) = .1[/(n»),lzl > R
ro -!
:t
Discrete Time-Function = 2: f(n)z-n
Number I(n), n ~ 0 ·n=O ~
0(
61
"n + -n- -
---:: ---
Z >
Z
(" - ])2 1- " (1 _ ,,)2 (z - ,,)(z - ])2 C
>
,,+ao l+ao (. 1 aO +l) z(z + ao) "'\I
"'\I
62 (y _ ])2 yR +~ n + 1 _ " - (1 _ y)2 !:
(z - y)(z - 1)2 (')
>
-!
z (5
63 a'i cos."n z
%+ a
z(z - e-ex cos a)
..,0
64 e-"'" cos an z2 - 2ze-ex cos a +e 2ex
-!
:t
m
z2 sinh 'P + ze-ex sinh (a - 'P) '"
I
65 e-"'R sinh (an + 'P) z2 - 2ze-'" cosh a + e 2« •

-!
;:Q
>
0= tan- 1 ~ ...,z
CI>
yR (~2 + fJ2)n/2 sin (nO + 'P) « Z 0

;:Q
·66
(y. - ox)2 + f/l + (I[( a - ')1)2 + fJ2]V2 fJ (z - y)[(z - «)2 + P2] 3:
'P = tan-l--
«-I'
ny"- 1 3')1" 1 [1I(n - I) 4n 6 ] z

67 (y - 1)3 - (y _ 1)4 +2 (1 - 1')2 - (1 - ')1)3 + (1 _ 1')4 (z - y)2(% _ 1)3
68 Lk (-1)" (k)(n +k - v)lk)

eexln- .1
%(z - 1)1.-
(z - eot)k+J
.~O v. k!
69 i l
«-
p-1/F(p) dp + lim
n-O
fen)
-
11
70
10 = 0 ziCX).F(P)dP
11 =0
1 + ao (" + ao>yR z(z + DO)
7]
(1 - ,,)[(1 - «)2 + P2] + (" - 1)[(" _ at)2 + P2] (z - 1)(z - y)[(z - at)2 + P2]
[a 2 + p2]"/2[(ao + «)2 + {J2]1/2 .
+ P[(at _ ])2 + {J2]112[(a _ ,,)2 + P2]112 sm (nO + 'II + A),
(J P
'P = 'PI + 'P2, 'PI = -tan· J --1
at-
I {} = tan-1 -IX
A = tan-1 _P- I fJ
ao +«
'P2 = - tan-J - -
at-"),
72 (n + 1)e«n - 2ne"'(n+l) + e<x(n-21(n - 1) ( ~y
z - e<X'
cos «n %
73 (_1)R - - 11>0 In 1-7=======
n ' .Jz2 + 2z cos « + I
(n + k)! dl.:
74 n! 11I+k. In = 0, forO" 11 <k ( -1 )',%2k dz k [.F(z)]
75
fen)
n + h'
h>O Zh i" p-IHh).F(p) dp
." 2a2z2
-nan cos- n
2 (z2 + a 2)2
TABLE 1 (Continued) z- Transform
3O(z) = J[f(n)], Izi > R
00
Number fen), n ~ 0
= ~f(n)%-n
n=O
nan
1 + cos 7Tn 202z2
77 (z2 _ a2)2
2
I + cos 7Tn
7T a 2z 2
78 an sin - n . -"""'"""""--
4 2 z4 + 0 4
79 an (
') + cos 7Tn -cos-n
7T ) 202z2
2 2 z4 - a4
P,,(:t·)
80
n!
P::"'(x)
81
(n +- m)!' m > 0, P~" = 0, for n < In
1
82 (n +ex)i1' ex > 0, Re fJ > 0 where 41(1, p, ex) = Wi, ex)
= generalized Rieman-
Zeta function
83 an (
I + cos rrn + cos- n 7r )'
2 2
c"
84 (II = 1,2,3,4, ... ) In z - In (z - c)
85
II
n = 2,4,6,8, ...
cz(z + c)
56 ./ n2c"
(z - C)3
5- cz(z2 + 4cz + c 2)
113C"
(z - C)4
/ n~c"
,W(zle)
85
-~
.fIO(z) = ,1[nl.:-1J
5;' -. £("-2'14(
- n .1112 ) a,,-2-4i(a
2 ;=02/+1
~ 4 _ b4 )i
z4
z2
+ 2a2z2 + b4
g,:. ,;yc;,), k > 0 and integer
d
-z dz Ft(z), Ft(z) = ;,-[nl;-l j'(n)]
(II - IXII - 2)(11 - 3) ... (n - k + 1) 'I-I;

91 ./
(k _ I)! a
9: J
k(k - l)(k - 2) ... (k - n +
n!
I) (z - a)1;
( Ir 1 +-z
[(z/a)3 + z/a] cos b - 2(z/a)2
93 ./ nan cos bll
[(z/a)2 - 2(z/a) cos b + 1]2
(z/a)3 sin b - (zla) sin b
94 lIa n Sin bn
"./ [(z/a)2 - 2(zla) cos b + IJ2 ;I>
"a n z(a - 2z) ( a) 2
"Cl
"Cl
95 J' (II + I)(n + 2) a2
In I - - - - Z
z a
tTl
z
( -a)n
52
)(
96'
/ (n + 1)(2n + I) 2 \',z/a
- tan- 1 val
- z - z In ( 1 a +;a) N
00
-...J
N
TABLE I '(Continued) z- Transform oc
oc
9"(z) = }[f(n»). 1z1 >R
-l
co :c
Number fen). n ~ 0
= If(n)z-n
n=O
a
~
-<
z cos :I: a sin a. >
97 - - tan- 1 - - - - z
a z-acos:I: o
% sin a. z2 - 2az: cos :I: + a2 ?;
+--In ~
2a z2 C
n
an cos ( ../2) n sin a.(n + 1) Z z2 + 2az sin a. + a2 >
-l
98 ./ n+ I
- In "7""--:::---:----:
4a z2 - 2az sin a. + a 2 oZ
1 o."
99 cosh (%-1/ 2)
. / (2n)! -l
:c
rn
100 ~O)n. \z/(z - a) N
~>
(~
- . "'co~n
z z
(/I
.101
2
-, 2 o
~
3:
eZ' :
102 ~.:ii:~'BernOulli
n.
polynomials
103 Wn(X) .~ Tchebycheff polynomials of the second kind

Z2 - 2.t·z + 1
+ z-m
104 / Isin : \ . m = 1.2•... Z2 -
1r In
2z cos ../m + I 1 - z-m
105 Q ..(x) = sin (n cos- 1 x)

.;' Z2 - 2xz + 1
TABLE II. PAIRS OF MODIFIED z-TRANSFORMSt

G(s) ~(::. m)t
Simple Forms of G(s)

(A.I)
z - I
e- alliT
(A.2)
I-a z - eaT
Quadratic Denominators of G(s)
~
. mT
z - 1 + (z
T
- 1)2
(A.3)
Te-IJIIIT[ m + e- aT ] (A.4)
z - e aT (z - e- aT )2
_1_( .e- amT _ e-bIllT )

(A.5)
4'; ~ a)(s + b) b - a z - e aT z - e- bT
s + ao _1_[(ao- o)e- alllT _ (00 - b)e- bIllT J (A.6)
I j .... a)(s+b) b - 0 z - e- aT z - e- bT
e":x,,,T z sin mf3T + e-:xT sin [(1 - m){JT]
(A.7)
f3 z2 - 2ze orT cos {JT + e- 27.T >
~
~
z cos (mpT + t/» - e-'%T cos [(1 - m)f3T - t/>] rn
e-"'''' T sec t/> ---""";;"'-~~;;;--=____=:---- (A.S) z
%2 _ 2ze "'7' cos PT + e-2:xl' S!
x
a. - 0u
tan 4> = - - N
(J oc
>D
t-.)
TABLE II (Continued) \0
0
G(s) ~(z, m) -I
::t:
1 z sin mpT + sin (1 - m)pT
(A.9)
"'
0
:0
S2 +p 2 {J z2 - 2z cos {JT + 1 -<
1 z sinh m{JT + sinh [(1 - m){JTJ >
1 z
(A.10) c
S2 - p2 p z2 - 2% cosh {JT + 1 >
"Ij
S z cos mpT - cos [(1 - m){:JTJ "II
S2 + p2 z2 - 2z cos {IT +1
(A.l1) c
(")
>
s % cosh mpT - cosh (1 - m){JT ::!
(A.l2) 0
l - {J2 z2 - 2% cosh PT + 1 z
0
S + ao vQf+P2 z cos (m{JT - 4» - cos [(1 - m){3T + 4» (A.l3)
"Jl
z2 - 2z cos {3T + 1
-I
s2 + (J2 {32 ::t:
ao
tan", = - "'
N
I
P -I
:0
Some special forms of G(s) >
Z
CI>
. hiT sin mk1T Cl
s2 + (k1TjT)2 z + (-l)k+l :0
3:
1TjT sin m7r
z + 1
e-<xmT sin m7r
(s + :x)2 + (7rjT)2 z + e-rr.T
(s + :x) cos m1T
e-a.m T
(s + %)2 + (1T/T)2 z + e-a.T
s+a e-rr.'lIT [T(a - CIt) ]

z + e-xT 1T sin m7r + cos m1T
Cubic Denominators of G(s)
T2[ m2 2m + 1 2 ]
(A.14)
"2 % - 1 + (z - 1)2 + (z - 1)3
(A.15)
1 [e-
;;: o:(z - e- aT )
XmT
+ (z
T
- 1)2 +
mT - 1/I1.J
(z - 1)
(A.16)
.; - ao 1 [ao - CIt e-ulIIT aoT ( ao) ] ]

~..;.; ~ x) -
CIt
--(
CIt z-e-:1<'1') + (-1)2
z- + aomT +] - -
()( - -
z-1
(A.17)
".;;. - a1s + ao
(A.18)
n.s + :x)
1 e-o. mT e- bmT
+ +-----.....,.-::- (A.19)
ab(z - 1) o(a - b)(z - e- a'1') b(b - a)(z - e-
bT )
ao a e-amT a _ - _
b _e- bmT >
___ + 0
0
-
+ -:--:-,-0 _----.'"
(A.20)
"II
"Ij
S'S - a)(s + b) ab z - 1 a(a - b) z - e- uT b(b - a) z - e- bT
"'z
s-. -:-.' als + ao
S!
a I a + a 0 e- alllT a 1b + a o _ x
___
ao 1 + a 2 - + b2 - bmT
e-_ -:-=.
(A.21)
s(s + a)(s + b) ab z - 1 a(a - bl z - e-tlT b(b - a) z _ e-bT t-.)
\0
N
TABLE II (Continued) -c
N
G(s) ~(z, m)
Cubic Denominators of G(s) (Col/ti"ued)

'0:z:m"'
1 e -alllT 1 e-'m.T 1 e-~,.. T "
0(
(A.22) >
z
(s + a)(s + b)(s + c) (b - a)(c - a) z - e liT + (a - b)(c - b) =- e bT + (a - c)(b - c) z - e- rT c
°o -a e-"mT ° -b e-h",T ° -c e tmT
>
."
.. + 0 • + :--~o-:-:-_ _ ----:c.:;; (A.23) ."
r-
(s + a)(s + b)(s + c) (b-a)(c-a)z-e a1 (a-b)(c-b)z-e- b1 (a-c)(b-c)z-e- r1' (;
>
a 2 -aa +a
_ _--"1_---'-::-0
e-""·1'
+ b2 -ba1 +a
0
e bm7'
+ c2 -ca 1 +a0 e- rm7'
•
::!
(A.24) 0
(s + O)(S + b)(s + c) (c-o)(b-a)z-e aT (o-b)(c-b)::-e b1' (a-c)(b-c)::-e- c7 z
1{ 1
-rt.2 - - - e '''"
Z - 1
7.[1 + rt.mT + :------::=
z - e «1'
rt.Te 27'
(z - e - ~T)2
J} (A.2S)
0
."
;!
s(s + 0)2
"'';'
s + 00 ~ {~ + e <1.IIlT[m'%T(Oo - :It) - ao + (ao - ot)«Te -<1.TJ}
s(s + :1)2 «2 z _ 1 :: - e -,,1' (z - e 21')2
(A.26) 'z"'
~
(I)
s2 + 0IS + ao 1
-02lzf-ao- + e _ ~m 1'[ot2 - ao + mTx(a1ot - «2 - ao) (XT(al~ - -a.TJ}
+ ----"--:------::m:-::-- x2 - 0o)e
(A.27)
~
::u
s(s + ~)2 -I z _e- a1' (z -e- "1')2 ~
1 _r e
- -........ + e
bmT
XIIlT
[",T(X - b) - 1 (<< - b)Te- 2 TJ}-
+ -:--~""'::- (A_28)
('%-b)2l.::- -b7 e =_e- x1' (z_e- xT)2
s + On 1 (00 - h)e- bm7' 21111.[(b - 00) + (ao - ~)(b - x)mT + (uo - 'X)(b - «)TIt a.TJ 1
(b - «)2l z - e b7' +e z: - e- 27' (z _ e 2T)2 f
(A_29)
r - als + 00
IS - bl(S + x)2
+ (<<2 - 0 1 :1 + oo)(b - ot)Te-,,"·T [ ", 7' + e <1.1' l' ]}
z - e- 2 (z - e- x )2 (A.30)
) ()
- - - - - - -e "Ill sec</>
_ T Z COS (mflT + </» - r"r COS [(I - m)fJT- - "']1
,%2 + p2lz - 1 z2 - 2=e xT cos fiT + e- 2'XT J
-!It
tan." = -{J- (A.3l)
00
---l-- f 1 - e
--:",,7'..L
sec '+'
Z cos (m{JT + "') - e-:x.T cos [(1 - m){JT - "']}
«2 + {J2 Z - 1 z2 - 2z:e-- 2T cos fiT + e-2x7'
«2 + fJ2 - «00
tan'" = ---:--~
aofl
(A.32)
1 { e- bll.7' -21117' ..L z cos (m{JT + .,,) - e- 2 7' cos [(1 - m){JT - "'l}
: - bl[(s + x)2 + (J2] (<< - b)2 + {J2 z - e-bT - e sec't' z2 _ 2:;:e aT COS {JT + e 2xT
b-cc
tan." =-{J- (A.33)
;::. - 0I S + 00 __1_ {~ + (<<2 + {12 _ a )e-2I1lT sec ." :: cos (m{lT + .,,) - e- .. r cos [(1 - m){JT - cf>]}
'::'! - x,2 + fJ2] «2 + fJ2 z - 1 0 =2 _ 2::e 21' cos fiT + e- :!d'
:1 - 01 0o(h - 0 1)
tan cf> = -- - -:-:-~--::,..--:~ (A.34)
fJ (J(~2 +
p2 - 0 0)
>
s + 00 ao - b (e- bIllT T:: cos (mfJT + cf» - e-"1' cos [(I - m){JT - ..L]1
."
."
---r= - r 211• sec '" --~_.~=__::.;;_____,~:..:...---.".__;,..:-=-_~'I' m
(cc - b)2 + p2lz - e- bT z2 - 2ze--"T cos PT + e--;!27' J Z
C
(00 - :1)(b - «) - p2
x
tan ." = --'---;----:-:-=---'-
(0 - b)P
(A.H) N
0 -c
w
N
TABLE 11 (Continued)
~
G(s) !fez, m) --l
==
~
1 { 2 e-blllT
+ 00) z-e-bT + (<<2 + fJ2 + alb - ao)e- XIIIT sec ~
+ p2
( b)2 (b - alb - 2rxb lIG
(s + b)[(s + «)2 + P2 ] (X- -<
>
z cos [mpT + ~] - e- rxT cos [(1 - m)pT - ~]\ z
c
z2 - 2ze- rxT cos PT + e 2.. '1' J >
"II
"II
C
(1
(A.36) ~
~
.!. {_l__ z COS m{JT - cos [(1 - m)PI1}
(A.37) 0
fJ2 z - I Z2 - 2z COS fJT + 1 "II
--l
X
00 I 1 z cos (mpT - ~) - cos [(1 - m)pT + ~]} 111
,
co
p2 \z _ 1 - sec ~ z2 - 2z cos PT + 1 --l
lIG
aD >
tan 4> = - (A.38) 2:
P fIl
"II
0
1 { ao z -rx T z CPS (m{JT - 4» - e-«T cos [(1 - m)pT + .p]} lIG
3:
Pz -
z-- I + (fJ - ao)e III sec 4> z2 - z2e «T cos {JT + e 2rx'1'
alfJ
tan+=-p:l (A.39)
-aD
S +Qo 00 - b { e- b•n2' z cos (mfJT + ~) - cos [(1 - m)PT - ~)}
-PZ-+-b2 z _ e- bT - sec ~ ------:zZ:----::2z:--c-os-p'::':T::--+---:-.- - -
Doh - p2
tan += (00 - b)P
(A.40)
~..I..__.~··~··~<. . . ·.. .$.'....I.··.F. .I.J.t.~ljS~"....~.-.~...·~·.'5~~. .,.~~~·~·le~G.r_·~·n~'·_~~~·i~~

t~·rn , ......... . ...::: ...;..:
~-.-
r + 01S + 00 1 { e-bIll'1'
(s + b)(r + PZ) pz + b2 (bZ - alb + 00) z _ e bT
1l2 b ) z cos (mpT + +) - cos [(1 - m)pT - +]}

+ ( P- + °1 - ao sec + z2 - 2:clCOS PT + 1
(A.41)
Biquadratic Denominators of G(s)
I T 3 [ m3 3m2 + 3m + 1 6m + 6 6 ]
(A.42)
s' 6" z - 1 + (z - 1)2 + (z - 1)3 + (z - 1).
I ~[T2m2/2 - mTla + l/az + T2(m + 1/2) - Tla +~ _.!. e-4mT ]

(A.43)
~s+a) a z - I (z - ])2 (z _ 1)3 a2 z _ e-aT
1 [mT - 2/a T e-aIllT(mT + 2/a) Te-GT(m+U]

r(s + a)2 -
02 z - I + (z-- -
1)2
+ Z - e-4 7' + :----=:;:;:-::
(z _ e-4T)2 (A,44)
-1- e (-II '1.[a ~

iii
2 p2 + mT + Te- aT J+ _____
1 cos m)pT + .pI ---:-::-----:-_
[(1 - ---::---=-__ - z cos (mpT ~»)
_ _---...,;",;,.
02 + p2 Z -e aT (z - e 4'1')2 fJ z2 - 2z COS PT + ]
>
"II
(A.4S) ~
!2
2. {mT + __T_ _ ~ z sin pmT + sin [(1 - m)PI1}
(A.46)
><
p2 z - 1 (z - 1)2 P 2:2 - 2z cos fJT + I ~
va
TABLE II (Continued)
G(s) ~(z,m)
s + DO
-~1 {P~ +z DofPmT
- 1
+ (z-Dop:r COS (mpT + .p) - e-«T COS [(1 - m)pT - .pJ}
- - - p21e-a.m 7' sec.p z----;:---:::--::';;;--=:-----:=;----
- 1)2 z2 - 2ze "'T cos {JT + e 2",T
P: = oc 2 + p2; p~ =~ - 2OCQo
ocp~ + DOP~
tan.p =- pp~
20cP
When Do = oc, tan .p = - -p2
-oc
2
E. I. Jury, "Additions to the Modified z-Transform Method," I.R.E. Wescon Convention Record, Part IV, August 21,1957,
pp. 136-156. Similar tables of modified z-transform are presented by R. H. Barker, "The Pulse Transfer Function and Its Appli-
cations to Sampling Servo Systems," Proc. I.E.E. (London), Vol. 99, Part IV, 1952, pp. 302-307.
t A few entries of Barker's table (British Crown Copyright) in a modified form have been included in the table above for which
the permission of the Controller of Her Britannic Majesty's Stationary Office has been obtained.
: It should be noted that ~(z) can be obtained by multiplying ~(z, m) by % and letting m go to zero.
.....
•• <
TlF." :-~-
III ;;;.
:l
Q.
~
:T
-16-
II> ~
:T > >-J
n
d :T ..
n
d ~~ '" tD
2-
.P n :T
_.n -1l
n c: r-'
tTl
iii'
;. ::=>
::=>
iii·
;. ;so..
:l n
(to
(7Qn
~ 31 ~
~
~
:l
~
Q.
:-.
S·
::I
...,
0
-=
t ... ~
N
N N
M M ii"M.. ...III
t1>
3
...
III>
~
n II - !l.
;so.. g '-' ~ .to
,!
...
S·
:T
II ...
:T
-J
Q
~I- ~I- S· >
t1>
~
'!;. ~
:::-
0- 0- ;:C. IQ
Jl ~ ~I 0' -::=> c;;.;:
;;;
...:.
00- :0 t1> Q:O'=-- 1M;! aM" II t"""
-:: ~ Q. S
::
~
T
to ...
.j:"
w"to
0'
... ..
~ :i"
(7Q
..c":l
E.O
00- 0Q
1.;s 1.;s ~I~
S! ,,~ ~
!D
...
as:
.
tI)
!. t:)
3
n
Q.
::;>
0 ; + ...Q
..Q ~
3
...;:c.
~ A
n
::I
.. .. I
~ .
I
~~ c:
~ >
~
0 :IA II> :! tTl
0 .!I ::I
3 ~ ~
:0 0 + + Jl
-III
:T:l
Q.
~ :$ ~
t:T
'<
... ~ Jl
t1> Co
.....
~o-
:;.
t1>
'II.
0 .'1 ~
n
-0 ~ ..Q
ere .0-
",'
,!
to
~
>
ii>
n
:r
(7Q
0 + +
1:1 ..1:1
Jl
i
2-1A
°IA
.. II>
n
t:T
~
.!. r-'
tI)
;. '" fi";z
0
c
t1> (til> ::I
Q.
:n
.., ... (i t1>
~.
-
Q.
~ ::I
Q 1:1 :l 0
n ..1:1 ::s
I
::s ~
::s III 2'
0 .!. :l ::I :l
e- to
iA~ a
3
::I
Il
III o·
:l
>-
'" cn "CI
t:T '"
~ ~. 02- "CI
'<
0-< "'z
~ :l
0
:::J
IA -:!
:!
~
~
.., IA N
\C
? ~
298 THEORY AND APPLICATJON OF THE Z-TRANSFORM
TABLE III (Conlinut'd)
Tobll/ott'd vollies of the integrol I ..
1 ~ B(z) B(z-l)
I =- ---z-Idz
n 2."j. unit A(~) A(z-I)
circle
boZ + bl B(z)
I. .9"(z) = =-
00'l. + 01 A(z)
I =
(b: +
b~)Qo - 2bob1o l
""':'--=-"":"'-::-:-"'-'-":
1 oo(o~ - o~)
boZ'I. + bIZ + hi
2. .9"(z) = Qot-2 + 0lz + Q2
Boooe, - Blooo l + B2(0~ - O~I)
12 = 0 0 [(0 01& - 2
0a)el - (0 0 -
0 1 0 10 2) 0 1]
Bo = b~ + b~ + b~
Bl = 2(bohl + bl b2 )
B2 = 2bohz
bot-a + bIZ! + bIZ + ba

3. .~"{'l.) = -=-=a:--""'::"""2 --=--..:::
QoZ + 0lz + Or + 03
ooRoQo - QOB 1Q 1 + ooBIQI - BaQs
13= ~:-----::~~:""""--=;"":"~-:-7-":";;;'~-'::"":"=--""""'-=-
[(Q~ - o:)Qo - (0 00 1 - OeDa)QI + (00"2 - 010a)Q 2lao
Bo = b~ + b~ + b~ + b:
Bl = 2(bohl + bib. + b2bs)
BI = 2(bohz + blba)
B3 = 2boha
Qo = (Ooel - 0ael)
QI = (0001 - 0ffla)
Q2 = (oJel - 02el)
Q3 = (al - 03)(e: - e~) + 00(00e2 - 0ael)
eJ = 00 + 02
e2 = 01 + 03
APPENDIX 299
TABLE III (Continued)
b~ + biz! + b"z2 + baz + b4
4, '-(z) = are + D1ZS + DeZ• + DSZ + D.
14 = ao[(D~ - o:)Qo - (D oa1 - aaa.)Ql

+ (0002 - D.D.)Q2 - (0008 - DJa.)Q81
80 = b: + b~ + b: + b: + b:, B3 = 2(boba + bib.)
Bl = 2(bobl + bJb2 + b2ba + bab4),
B2 = 2(bob2 + b1bs + bab.)
Qo = Doele4 - Do03e2 + a.(Dl e. - eae.)
QJ = aoOJe. - 000aDa + a4(DID2 - osee)
Q2 = DoOJea - DoOsel + D.(aae, - Dsea}
Qa = a1(aJell - eae.) - D2(D1e J - D3ea) + Da(ele. - D3e2)
0.· = Do[e 2(D Ja4 - a003) + e6(D~ - 0:))
+ (e~ - e:)[DJ (01 - 0 3 ) + (Do - D.)(e. - O2 )]
el = DO + 0'1.
ez = aJ + 03
e3 = a2 + 04
e. = ao + D.
es = DO + O2 + 04
• For elise of calcuilitions Q. and 0 3 of 13 can be respectively written

0., = - (14Q'1 + (I:lQ, - (l2Q2 + (I.Q:I
0:1 11:100 ":!('II (II (J~
1'11\' :l\lIh.1I " /:11111'1'111 I .. 1.1, 1'11111 I" I "VI" III' 1';11 i~, I nll1n' 1'.... p.. inlin~ oUI-thesc:
II'lalulII,hll"
00
TABLE IV CLOSED FORMS OF THE FUNCTION I nP:rf', :t; <1
,.~O
o 1 -x
(I - x)2
xl + a:
2
(I - x)3
:r3+~2+X
3
(I - :l~)4
;lA + Ilr + Ilx\! + :r

4
(l - x)5
:1:6 + 2~ + 66x3 + 2fix2 + :I:
S
(I - x)'
",6 + 57 x' + 302",3 + 302x· + 57 Xl + X
6
(I - X)7
X' + 12Ox' + 1191.1:3 + 241fix' + 119la.3 + 12OJ~\! +X
7
(l -:1:'1
.t.8 + 2473:7 + 4293.t:' + 15619'):6 + ... + ,I:
8
(l - ;r)8
.1,9 + 502z8 + 14608.r7 + 88234.r1 + 156190.,05 + , .. + ,I'
9
(l - X)10
,t,10 + 1013.1,9 + 47840.1.8 + 4SS192x' + 13103S4.I,8 + ... + ,I:
10
(l - :r)1l
The missing terms are apparent since the numerator polynomials are sym-
metric in the coefficients.
PROBLEMS
Chapter 1
PROBLEM 1.1 Prove equations: (LSO), (1.51), and (1.52).
PROBLEM 1.2 Prove equations: (1.97)-(1.105) and (1.107)-(1.109).
PROBLEM 1.3 The basic Fibonacci sequence In has the following recurrence
relationship
where 10 = O. II = 1.
Let "'n = /in and 'I.. = Izn+l' Obtain the z-transforms of
(a) ~(%) =
J["'.. ), and (b) 'fez) = "['I.. ]
(e) Find the relationship between 'fez) and ~(z).
PROBLEM 1.4 Show that
Z [ +I] =
(s u)n
r n- 1
(n - I)!
..r[1 + .,.n 2 + u~(.r. + J:n 3) + Q~(.r.1 + Xn 4) + ... + u:(;/:" + x: ~ £0)1

. (I - x)n .
where
.,' _, c' 11'1':. I
;I"d
II;: _":(/' I I)" I ,.!"" I I ... '( II~"~(" q' II" I
, ... ,t II",·:
"I
,"'-
" ,,1111 I'"
1111
PROBLEM 1.5 A sampled-data control system is shown in Fig. 1.5. Given

G(s) = _1_ e-1.35I7'
$+1
H(s) = _1- e-2.1I.2'

$+2
1
T = I, R(s) =-
$
(a) Obtain '6(%, III).
(b) Find the continuous response.
(e) Indicate how to obtain from (b) the response at sampling instants.
Rts} =1/. + ~(:, m}
FIGURE PROB.I.S
PROBLEM 1.6 For the delayed sampler operation of system in Fig. 1.6, ob-
tain the modified a-transform of the output and the actual output c(n, m)T ... c(t).
1
i(iibj
FIGURE PROB.I.6
PROBLEM 1.7 If
BII(x) _ eZ /l where B,.(z) are the Bernoulli polynomials
J - - - a[ell• - 11'
show "I
(a) BJ.z) = O! (I) - 1

B1(z) - I! (a: - i) .. a: - i
Ba(a:) = 3'• (-6 - -4

~ ~ Z)
+ -12 - ~ - -3:x;
2 :.;1 + 2
-
PROBLEMS 303
(b) B2n : 1 = 0, " = 1,2,3, ... , where B,. == B..(O) = Bernoulli numbers
n = 0, 2, 3, 4, ..•
B,,(:I:)
(d)-,- =".11
I'
IBn '( _
X"..... I ",
)' = - I (p) B,.XII-
If Ii
p. n. P n. p! 11-0
.V -I 1
!:r;1J = P--1
z-I +
[B" ....(N) - B"+I(O)
when p = S, we get
.\' ··1
!xS = UN· -
:1'=1
3N6 + tN4 -IN! + ,,12 - .\)

co I (271')1!p( _1)"-1
,,~nzp = 2(2p)! BIP' P = I, 2, 3, ...
Hint: (a) Show first that 2y coth y = 2y/(e211 - I) + 2y/(1 - e-2v)

(b) Using the generating function of Bernoulli's numbers, obtain·
co (2y)!"
11 coth y = ! BIP (2:p)'.
,,=0
B..(I - x) = (-I)"B,,(x)
PROBLEM 1.11 Obtain the steady-state response v" of the voltage across the
resistor for the initially unexcited circuit shown in Fig. I.Ila in response to a
periodic input shown in Fig. \.lIb.
(II' (hi
FI(illlt F I'ROII. 1.11
."
• Tho ,Ollorl" II,. runllllnll ur ,fn) I. ".dll.d by I 'fll't
It-II
ll •
f"[(/) dtJ = T f.F(Z, m)12= 1 dm
1""[(/)2 dtJ = 2~j frfp-l$ m) (p, $(p-l, m) dm dp
~kJ.Q)F(jk 2;) = .F(z, m)lz~l.m=I' for [(0';-) = 0
,(m[F(s)I,=ol = Tf :F(l,m)dm
PROBLEM 1.1 5 Show that
,(m[F(s)G(s)] = Tz Lm'(z, m - A)~(z, A) dA

+T f:,-'(z, 1 + m- A)~(z, A) d;'
and
,(m{F(s)G( -s)) = T Lm'(z, m - A)~(z-l, 1 - A) dA
+ TZ-1f 1,-(Z, 1 + m - ).)~(z-l, 1 - ).) d).

III
Chapler 2
PROBLEM 2.1 In the example of Section 2.1, let V be a d-c voltage source of
10 volts and R = 1 ohm. Find the current in the 10th loop (Le., n = 9), if the
end loop (Le., 10th) is short-circuited.
PROBLEM 2.2 For the ladder network with capacitors shown in Fig. 2.2,
assuming zero initial conditions,
(a) Write the equation of the (n + 1)st loop.
(b) Obtain the expression for /(z, s) ~ J[/(n, s)), and I(n, s) = ~t{;(n, m.
(e) Obtain I(n, s)
(d) Show how to obtain ;(n, I).
R R R R R
FIGURE PROB. 2.2. Ladder network with cup:lcilors.

PROBLEMS l()~
PROBLEM 2.3 Consider a series in which the first r terms are all positive.
the next r all negative, the next r all positive, and so on. This can be expressed
as follows
The sequence u.. is
1,1, ... ,1, -J,-I •... ,-I. 1.1 •...
r terms r terms
Obtain the factor u.. which must be introduced into the nth term to allow for
such ftuctuation of sign.
PROBLEM 2.4 Given the Fibonacci's numbers as follows:
0, I. 1.2,3, S. 8, 13,21.34•...
(Each of these numbers is the sum of the two numbers immediately preceding it.)
(a) Obtain the difference equation which describes the general terml...
(b) Solve the differ~nce equation in (D) by the z-transform method and show
that the solution forI.. can be presented by the following formula:
PROBLEM 2.S Given the following constant coefficient difference equation:
a,]/(n) + D'_l1/(n - J) + ... + a~/(n -I)

=b,/:(n - / + 'I) + b'l_l:c(n - / + 'I - 1) + ... + bo:r(n - I)
For zero initial condition show that yen) can be represented as follows:
PROBLEM 2.6 Find the solution %(/) for the following difference-differential
equation:
x'(/) + 2:c(1 - J) + :c(/) =/(1)
if:r; == 0 for 1 ~ 0, and/(I) = unit step and unit ramp.

PROBLEM 2.7 For a periodically varying sampling rate system shown in Fig.
2.7a obtain the response for a step input. T.. is represented as follows (shown
RCII':: l/s ,.. Cis'
!-IenlRI I'ROII.2.711
graphically in Fig. 2.7b):
O.S second n = 0, 4, 8, 12, ...

Tn = 0.5 second n ... I, S, 9, 13, .. .
1.0 second " - 2, 6, 10, IS, .. .
1.0 second n = 3, 7, ll, 15, .. .
The total period T = 3.0 seconds.
[II I lit LlL

o 0.5 1.0 2.0 3.0 3.5 4.0 5 6
1< T ,IE T--~~I
FIGURE PROB.2.7b The sampling pattern of system of Fig. 2.7a.
PROBLEM 2.8 For the time-varying sampled-data system shown in Fig. 2.8,
given the following values:
dn =l+e-n
. 1 z
'.J(z) = --1
z-
' T == I, BI(z) .. - - I
z-
(0) Obtain the equation for the z-transform of the output 't'(z).
(b) Show that the solution e" of the equation obtained in (a) is
c(n) = I - ( -1 )ne-(1/2)n(.1-11
~ --i
r~"J
d" <9(z)
c(n)
flGURE PROB. 2.8
PROBLEM 2.9 A square of side n units is subdivided into n2 (see Fig. 2.9)
unit squares by parallels to the sides. The diagonal OA is drawn and also the
diagonals which are parallel to OA of all the unit squares.
Obtain the number of paths by which one may travel from 0 to A kccping to
PROBLEMS 307
FIGURE PROB. 2.9 A square of side n units.
the lines of the figure. It is understood that at junctions such as P, a path may
proceed in one of the directions PQ, PR, or PS only.
PROBLEM ~.I0 Find the eigenvalues of the following matrix:
o o
010
o o o
A=
010
o 010
by uNing II recurrence rellltinnHhir in the comrutillion of the rcquircd-dctcrmi-

nllntl,
Chapter 3
PROBLEM 3.1 Check for the number of roots inside the unit circle for the
following polynomials:
(0) F&/:) = 122' + 1625 + 20z· + 14z3 + 922 + 32 + I

(6) F.(2) = 6z6 + 821> + 19:. + 19:3 + 1522 + 6: + 2
PROBLEM 3.2 If the determinant IXA• + Ykl is expanded as follows:
00 + bn _ k + J . oJ + bn _ ktl
bn _"+2 00 + 6"_"+8
111
Then if ~J is expanded. the expansion of the second determinant • .:1•• can be

written from the first. ~J' by replacing Ok by 6n•k and ble by On.'"
PROBLEM 3.3 Given the following fourth-order real pOlynomial.
with 04 =I
(0) Show that the conditions for the roots to be real are
Do = o~ - 2000 2 >0
DJ = I
o~ - 200"2
0 102 - 3000 3
ala. - 3000a
20. - 400 - 2010 a
I >0
DJ Ol Qa - 400
D2 = ................. .............. 2ozOa - 301 >0
Q10 3 - 400 2azOa - 3Ql 30~ - 20.
and
Ql
D. 202
Da = >0
30a
°1 20. 30a 4
PROBLEMS 309
(b) For the real roots to lie between zero and unity we require the following
additional conditions:
I + a3 + a2 + a1 + ao > 0
03 + 202 + 30 1 + 400 < 0
"t + 3al + 600 > 0
"I + 400 < 0
0u >0
PROBLEM 3.4 Show that for an aperiodic system (discrete or continuous), the
maximum number of extrema is n - I, where the order of the system is II.
Hint: Show this for a second-order system and by induction for the nth order
system.
PROBLEM 3.5 Obtain the number of roots inside, outside, and on the unit
circle for the following real polynomial:
F(z) = z8 + 0.7'& + 0.4,· - 8.5,3 - 5.6%2 - 3.2: +4
PROBLEM 3.6 Show that a necessary condition for the roots of F(z) =0 to
lie inside the unit circle is:
n!
(n _ r)! lanl > r! 10rl. (r = O. 1.2•... , n - I)
For a fifth-order polynomial this yields

with as =I
Hint: Show that every circle K (or polygon) encircling all roots of a poly-
nomial F(z) = 0 also encircles the roots of the derivatives FI,..'(z) = O.
PROBLEM 3.7 If the roots of F(z) (with On = I) are all inside the unit circle.
show that the maximum modulus of the roots, r, can be obtained from the
following relationship:
!l (A n_ 1 - B n _ 1)2F(l)F( -1)( _1)n(n-lI
r < 1- 2,,(n-l)
or
Hint: Use equations (3.68) through (3.70) to verify this inequality.

PRUIII.I:M 3.8 Obtain the root distribution of the following polynomial:
1-"(:) :II: z& - 3.5714z" + 5.7755z3 - 5.303z2 + 2.6405z - 0.5414

I'RClIlJ.l;M 3.') ()htllin the mill distribution of the following polynQmials:
(tl' /o"{:) - 4 I 3013 f 2:111
(h) /o'(1) - ··2 I 4. I ~.' .- ().rl of- 2,' of 6111 .- 4:' - 2,7 ... ,I
Chapter 4
PROBLEM 4.1 If §(z) = J[J] and (f(z) = J[g] are both analytic for Izl > Ro,
show that the function Jf"(z) = ][/g] is obtainable from .1"'(z) and ~(z) as
follows:
.)F(Z) = _1..1 ~m'Y(~)d'
2TTJ l(1 ~ R. , ,
for IZ, > RoRv where Rl > Ro. (The path of integration can be chosen as a
simple closed curve r.)
PROBLEM 4.2 For a multirate error-sampled feedback system shown in Fig.
4.2, obtain the modified z-transform of the output '6(zTI ' m); n is an integer
larger than one.
FIGURE PROS. 4.2
co
PROBLEM 4.3 Let F,. = ~ nr:r;"; show that
71~O
(a) Fr' = dFr = Fr+1 , . 1

dx x
with Fo = -I- - , x
-x
<1
(b) Show by induction that
S,(x)
F, = (I _ xy+l • x O
;=1
(c) Using Sr(x) in Fr of (b), differentiating and using (a), show that
.+1 r
- "
S r+l - '" Sr·t-l,ixi -- .~
, [ix
i + (r + 1 - i)xi+l]S r.,.
i-1 i=1
(d) From (c) by equating coefficients, show
S'+I,i = is,,; + (r + 2 - i)Sr,i_l

(e) If r = 3, obtain the value of
for x <I
PROBLEMS 311
.-1 -
PROBLEM 4.4 Let Fr,,,(x) = ~i rx' and Tr,II(x). DO ~i rx'; show from the
'-0 '-II
results of the preceding problem that
(Q) Fr,II("') - Fr(x) - Tr.,,(x)
(b) By a change of variable show that
Tr"(x)
'jaO
=I(n +Wx"+'= I I (~)ilnr-Ix"+'=.i
J
(~)nl'-i'l:"F/(X)
,=OJ-OJ. j"'O
(e) Show that
Fr ,,(x)
,
= Sr(:l') - f (~)nr-j:C"Sj('l:)'
j-O )
for all :l'
71-1
(d) Show that the finite sum I i' is a known result in terms of Bernoulli
numbers. i=1
zll + bIz zt + b1z
PROBLEM 4.5 Let .JF(z) = z2 + Ql% + Q2 = (z - P1)(% _ p,~.>
where
Using the residue method, show that
(a) N = _1 1.. Jt"(%).~(z-J) ~ = (b~ + 1)(1 + P.P2) + 2b1(P1 + P2)

• 27fj Jr z (I - pf)(I - p~)(l - PIP2)
(b) Show that N. in (Q) is the same as obtained from Table III of the Appendix
when n = 2 and bo = 1, bl! = 0, Q o = 1.
PROBLEM 4.6 Given the following expressions for the sampled spectral
density cJ)1.2(z).
1 z+I
(a) (1 + 0.5z- 1)(1 + 0.5%) , (b) (z + 0.5)(% - 2)
Determine "'J2(nT) for eath case.

PROBLEM 4.7 For the error-sampled data feedback system shown in Fig. 4.7.
(Q) Obtain the total squared error for a step input.
(b) Obtain the discrete square error (that is, the sampled square error).
27.8(s + 1)
G(.f) = s[(s + 3)2 + (27f)2]
T=l
ret) = u (I) + L~(t) v-
a JI. - ~.......,.
------------
He n IIU IIIU III. 4.7 1·lTllr 11111111,1"" liot'''hlll'k "YIII"IlI.
ChapterS
PROBLEM 5.1 A saturating amplifier sampled-data feedback system is given
in Fig. S.l; also given is the form of h(n) as follows:
4O e(n) ~ 40
(
hen) = e(n) -40 ~ e(n) S 40
-40 e(n) S -40
T Ty-
r~ -.Yf(z) .. .../, o~(z)
I
I
I
r(l) + 1
1-+1 - - - 0 cit)
FIGURE PROB. S.l Saturating amplifier sampled-data feedback system.
The power series representation of this nonlinearity over the interval -6S <
e(n) ~ 65 is
h(n) == 1.1IIe(n) - 0.1234 O.IO-a[e(n)p
r(n) = 60 volts
e(0) =60, e(1) = 32, T = 1
(a) Obtain the nonlinear difference equation which describes the error e(n).
(b) Solve the preceding equation using the convolution z..transform to obtain
e(n).
PROBLEM S.2 Given a nonlinear discrete system as shown in Fig. S.2. (f
hen) = m1e(n) + mae(n)8. with ma = 0.1, ml = I
and
r(n) = unit step. and T = 1
obtain
(a) The nonlinear difference equation which describes the output c(n).
(b) Solve the preceding equation using the convolution z-transform. and
obtain c(n).
r(l) =II(/) + 1-+---oc(tJ
FIGURE PROB. S.2 Nonlincllr di~rclc ~y5ICI1l.

PROBLEMS 313
(c) Compare a few values of c(n) from (b) with the exact solution using (a) as
a recurrence relationship.
Chapter 6
PROBLEM 6.1 Given a nonlinear sampled-data feedback system as shown in
Fig. 6.la and the type of nonlinearity shown in Fig. 6.1h.
(a) Show that the limit cycle +1, +1. -I, -1. exists.
(b) Show that the limit cycle 0, +1. -1,0, does not exist.
r,,(t)=O + c,,(t)
FIGURE PRO~ 6.ta
e,,'(O)
----~...j..:~---- ... e,,(O)
FIGURE PROB. 6.1b
PROBLEM 6.2 For the example in Section 6.6, if the relay output is -I,
-I, + I, + I, obtain the possible values of co. c.' "2, C3 to maintain a limit cycle
of period 4.
PROBLEM 6.3 The nonlinear difference equation describing a PWM system
which is in a limit cycle of period two is given as follows:
k 1
Cn H = y(n) IJ2 (bh" - e ·b(T -It.> + e -hT) + c" +
= f(c n• e.. )
.
r" II = yen) hk [t" b( T .It.> - t" •• hT] + e-bTc" = g(e", en)
If.
If" .. /I Ir" - t·"I. r" ~ I
;tnd
(a) Obtain the elements of the matrix [Ao].

(b) Find the characteristic equation of the matrix which can be tested for the
roots inside the unit circle.
(c) Obtain conditions on the stability of the limit cycle. 1 + e -117'
(d) Show that the solution for h ... 0, and a.k < (a.k)c = 2b 1 -liT' is
always stable. - e
PROBLEM 6.4 For a relay sampled-data system given in Fig. 6.1a and b, the
following are given:
b = 1, k = 1, T = 1
(a) Show that the continuous response at the limit cycle is given by
{!(1
I + e-22' + e-mT _ ,,-(1 + In)2'}
co(mT) = - - 2m)T + 1 + e- 2T '
o ~m ~ I
Hint: Note that when m = 0, the response at sampling instants is obtained.

(b) Show that the postulated sequence +1,0, -1,0 can exist if Co = -0.245
and Cl .. 0.166 and half the dead zone A - 0.2.
Chapter 7
PROBLEM 7.1 (a) Find the approximate solution of the following differ-
ential equation using the z-forms method and assuming T = 0.2 second.
day d2y ely
dIS + 2 dt 2 + 2 dl + Y - I, yeO) - Y(O) = yeO) = 0
R(s,:II lIs + 1-_..,-_-<> e(t,

Cis}
(0'
R(s'. lIs + t--,---<> Y(s}
(b,
FIGURE PROD. 7.2
PROBLEMS 315
(b) Compare this solution with the exact solution and comment on the choice
of T.
PROBLEM 7.2 Compare the exact step response of Fig. 7.2a with the approxi-
mate model of Fig. 7.2b for the following values:
(a) T = ls second, kl = 3.2, k2 = 555, Q = SO radians per second.
(b) Repeat (a) with T = ,Ill second.
PROBLEM 7.3 The z-transforms of the input and output of a discrete filter
are given as follows:
SOj(z) = I + 3z- 1 + SCi
SOo(z) = 4 + 142:-1 + 27z- 2 + 13z-3 + 5z- 4
(a) Determine the z-transform of the filter Jt"(z).
(b) If a small disturbance is made in the measurement of the output such that
the leading term of .'Fo(z), that is, 4, becomes 3, determine the z-transform of the
filter for this case .Jt"·(z).
(c) Using the substitution
z-I = l +1
obtain Jt"l(A) for (a).
(d) Determine .}f'tO) for (b).
(e) If we truncate the series for Jf"r(l) after the second (or after the third) term,
determine .}f'·(z) for these cases and compare with (b).
(f) Comment on the value of the transformation in c.
PROBLEM 7.4 To obtain the z-forms for a double integrator (l/SZ), the
following second-order difference equation is used to approximate iz(t) =
JIy(t) dt:
Aiz(t + 2T) + Bi2(t + T) + Ciz(t)
= Dy(t + 2T) + Ey(t + T) + Fy(t) (I)
where the coefficients A, B, C, D, E, and F are to be determined.

(a) Using the following relationship in equation (I),
.sf[i(t + nT») = e"T'9'[i(t)]

and
obtain the Laplace transform of the difference equation.

(h) If the exponential operators in (a) are expanded in a Taylor's series, and
we equate the same powers of s on both sides of the difference equation, show
that difference equation for a second-order integration is
;.J, + 2"1') - 2'2(1 + 1') + ;2(1)

.,.~ IIlT~ .,.2
-
I.' '
'/(1 I .''f') I.' ,lIlt I 1') I I.' !I(t) (2)
(e) Show that the z-transform of (2), neglecting initial conditions, is given by
the following z-form of l/s2
.I"W T"z"+10z+1
'dI(z) = 12 (z _ 1)2
(d) If initial conditions are not zero, and if yeT) and i,,(T) are expressed as
follows:
T2
y( T) = yeO) + Ty'(O) + 2! yW(O) + ...
T2
iz(T) = yH'(O) + T,II(-1'(O) + 2! yeO) + ...
show that Ja(z) is given by
T2 z2 + 10% + 1 z Tz
J 2(1.) = 12 (z _ 1)2&(%) + %_ 1 yH'(O) + (z _ 1)2,111 1'(0)
T2 z(z + 5) T3 % To z
-12 (1. - 1)2 !J(O) + 12 (z _ 1)2 y'(O) - 180 (1. _ 1)2 //"(0) + ...
(e) Similarly, for one integration, show that
.1 1(:) -+-I :f(z)

= -T:
2Z - 1
-
Tz
2(1. - I)
yeO)
z
+ _-!II -11(0),
z- I
where yl 1)(0) =J!I JII, CJ
PROBLEM 7.5 Using the :-forms obtained in (7.4) and the 1.-transfoflll, com-
pare the approximate solution of the following differential equation with the
exact solution:
d 2y (Iy
ill" + 3 dl + 2!/ = 0, yeO) = 3, y'(O) = -4
Choose T = 0.1 second.
PROBLEM 7.6 If the exact solution of the following time-varying equation
d2 d
~2
- [(/ 2 - 1)!I] - -~ [2/y] - 6y = 0, yeO) = -0.5, !I'(O) =0
is
yet) = 0.5(3/ 2 - 1), OS,IS,I
Obtain the approximate solution for T = 0.1 using the z-forms and the ,:-trans-
fOfm and compare with the exact solution.
Chapler 8
PROBLEM 8.1 In the analysis of one-way transmission ch.mnels. sampling uf

information is used. Figure 8.10 shows the elements of a connectiun cllllt:lin-
ing a unilateral repeater which isolates the two media (m()dern~) .lIld pmvidcs
each with a matching termination.
PROBLEMS 317
(II)
2 2
(b)
FIGURE PROB.8.1
(a) If the pulse length h is assumed much less than the sampling period T,
show that the circuit in Fig. 8.10 is equivalent to Fig. 8.lb, where
x
I, = c ~ ejUJtF,(jw)!J(t - nT)
11 -II
and Fr(jw) is an impedance ratio function of the frequency (w).

(b) If the voltage ratio 2v21v\ produced by insertion of the whole system is
denoted as T(jw) and we assume that the intervening network (1-2) is eliminated,
show that the modulus of transmission is
4exe-«(cosh « - cos y)
IT(J·w)1 = It 2
ex + y
where « = TI RIc, and y = wT.
(c) Plot the modulus of transmission IT(jm)1 for various values of « as a
function of !".If, = ratio between modulating and sampling frequencies.
Hilll; See reference 26 of Chapter 8.
PROBLEM 8.2 The circuit of a periodically switched network which produces
a phase lag is shown in Fig. 8.20. The actuation of the sampler is such that the
left-hand loop is closed for k T ~ t < (k + ,)T and open for (k + ,)T ~ t <
(k + I )T, and that the right-hand loop is closed for (k + O.S)T ~ t < (k +
0.5 + ,)T lind open for kT ~ t < (k + O.S)T and (k + 0.5 + h)T ~ I <
(I.. + I )1~ wherc () < ,) ~ 0.5 and k = 0, 1,2, .... The input to the network is
a carricl"-lIupprcssed iUnplitudc-mudulatcd sinusoid given by
('I(t) . (',(I) CII!> (UI , 'M,

318 THEORY AND APPLICATION OF THE 2-TRANSFORM
N
00
~
... 8:
~ e I.IJ
gc:
~
<:)
..... [i:;
.!
-:
~
~
I
~
..
~
~
"3
"0
0
::Ii
~I
e
8 ai»
PROBLEMS 319
(0) Show that the circuit system in Fig. 8.20 is equivalent to the block diagram
Fig. 8.2b. where I; is an infinitely small quantity
I I
G1(s) = R 1cs + I • G.Js) = R"cs + 1
(b) If et(t) is a step function of magnitude Eo. show that the response is
kEo[l - e-(k+1)6TIR~][e-(t-(k+O.ii)TJ/Rle]
ea(t) = -~----I:--_-e--T.6T;r,/~R~~c:-----~
(ult - (k + O.5)]T - lI[t - (k + ~ + O.5)]T)

where lI(t) is a unit step and
I
Rt = l/R) + I/R2
URIc sin (!MT + tf.) + cos (O~T + tf.)
- (ORlcsin tf. + cos tf.)e-6TIRI~
k=---------=-=-....,.....;,,.......;:....-----=.----=..;--~
02R~C2 + I
PROBLEM 8.3 Sometimes inventory control problems can be represented as

staggered sampled-data feedback systems. One such problem is that concerned
with the maintenance of a certain quantity of parts in a storage location. This
system can be represented in a simplified form as shown in Fig. 8.3. If
oT= 1.
Gain
eft)
FIGURE PROB. 8.3
(a) Determine the value of 61 (the sampler phasing) that yields the maximum
gain k for stability limit.
(b) Compare the gain obtained in (0) with the maximum gain for 61 = 0 and
"I = 1.
PROBLEM 8.4 THE TOYMAKER PROBLEM·
The toymaker is involved in the toy business. He may be in either of two
states. He is in the first state if his toy is in favor and in the second state if his toy
is uut of favor with the public. Suppose that when he is in state I there is a 50%
chance of II transition to state 2. When he is in state 2. he experiments with new
tuys und he Illuy return til stute I after a week with probability I or remain
unpmlitable in !ilute 2 with probubility g.
• Thill pl"llblcm I, uhlllillcd rmlll Itclcrolll."O (ZU). ('l1ul,lol' II.

(0) Find the transition matrix P which describes the discrete Markov process
P = [Ptil = [::: ::J

Define a state probability 'trt(n), the probability that the system will occupy
state i after n transitions if its state at n = 0 is known.
(b) Write the vector difference equation which describes the probability that
the system occupies each of its states after n moves; that is, the probability to be
in state 1 after n weeks if we know he is in state I at the beginning of the n-week
period, n = 0, 1, 2, .•.•
(e) Apply the z-transform to the vector difference equation of (b). Assume P
is given as in (0).
(d) Find H(n) the inverse transform of (I - zP)-I. I is the identity matrix.
(e) Obtain the state-probability vector at time n, knowing the initial state
probability vector.
({) If the toymaker starts in the successful state I, obtain ,,(0), the initial state.
Also obtain (e) for this particular initial state.
(g) Show that when n -+ 00, the state probability components approach the
limiting state probabilities of the process.
INDEX
Abscissa of absolute convergence. 20 Attenuating constant. 275

Absolute probability. 271 Auto-correlation function, 163
Absolutely convergent series. 252 Automatic gain control, 69
Absorbing st:lte. 270 Autonomous behavior, 268
Algorithm. 231 Autonomous discrete systems. 189
Alper. P.. 78. 188 AVe (automatic volume control), 70
Amplitude. 66
Amplitude characteristic. 211. 212 Backward difference. 4
Amplitude distribution. 255 Barker. R. H .• 43, 296
Analog-to-digital converters. 20 I Barnes. J. L .• 38, 42. 172
Analytic function. 10 Beamwidth. 258
Analytic. in the left half plane. 260 Bellman. R.• 77. 276
in the riglit half plane. 260 Bernoulli. number. 232. 235, 303, 311
in the ring. 161 polynomial, 232. 288, 302
Andeen. R_. 217 Bertram, J. E., 218
Annulus bf convergence. 144. 151 Bharucha, B. A., 173
Antenna theory. 248 Bifrequency system function. 67
Aperiodicity condition. 112 Bilinear transformation. 83
Aperiodicity criterion for linear discrete Binomial. 119
systems. 112 Bivariate generating function. 73
Aperiodic persistent states. 270 Boole. G., 76
Approximate solution. 69 Boundary condition. 52
Approximating a continuous function. Boundary value problems. 73
15 Bounded. 81
Approximation. mean rectangular. 222 Bounded function. 161
midinterval rectangular. 222 Boundedness and symmetry. 162
paraholic, 222 Boxer. R.• 231. 232. 241. 246. 247
rcctOlngulilr, 222 Branch cut. 13. 151
tri'lI1glllar. 222 Branch points. 9. 145
Approximation methods, 219 Bremmer. H .• 43
Army pCllynulIlilll. 2~~ Hl'idlli;md. T. F .• Jr .• 42. 247
Al'/hulIVkh, I. S.. 1·111 IImlllwitch Inll:IIr11I. 20
A!oc1Ii11l', J. A., .1.1, 71., '1'1 IIrnwlI, Ii. M., 411, 42, 44, 77, I W, 247
~21
322 INDEX
Brule, John D., 276 Convolution Laplace transform, 175

Bubb, F. W., 76 Convolution modified or z-transform
Budnicki, Zdzislaw, 277 method, applications, 147
Convolution modified z-transform, 158,
Cattermole, K. W., 276 159
Cauchy,.82 Convolution z-transform, 253,312
Cauchy's integral formula, 11 applied to nonlinear discrete systems,
Cauchy's mapping theorem, 124 174
Cauchy's theorem on residues, 145 of certain functions, 176
Chang, S. S. L., 277 method, 175
Chapman-Kolomogorov equation, 271 theory, 248
Characteristic equation, 81 Cooke, R. c., 76
Characteristic roots, 54, 219 Correlation function, 163
Cheng, D. K., 43, 276 Correlation functions and spectral anal-
Chow,C. K., 217 ysis, applications, 163
Chow, T. S., 140 Correlation functions, continuous, 165
Christiansen, P. L., 277 Critical aperiodicity constraints, liS
Classical method, 45 Critical constraints, 94, 115
Clockwise direction, 22 Critical To' 206
Closed-form solution, approximate, 174 Crosshatch, 143
Coding theory, 267 Crosshatched area, lSI
Cohn, A., 79, 131, 139 Cruickshank, A. J. 0., 220, 246
Commensurable time delays, 56 Cuenod, M., 231. 246, 247
Complex convolution formula, proof, Cypkin, J. A., 39, 42, 43, 77, 216
167 Set also Tsypkin, Y. Z.
Complex convolution integral, evalua-
tion, 145 [145 Da-Chuan, Shao, 218
theorem for the modified z-transform, doc component, 196
Complex convolution theorem, 142 Davis, H., 76
Complex differentiation, 7 Delay, 228
Complex integral formula, II fictitious negative, 16
Complex multiplication, 6 pure, 16, 51
Complex roots, 95 time, 51
Complex scale change,S Delay factors, 193
Complex z-plane, 275 relative. 19 I, 192
Computational rule, 79 Delay operator £-1,66
Computer theory, 267 Delayed sampler operation, 302
Computers, digital, I Delfeld, F. R., 218
Conjugate, 30 DeMoivre, I, 42, 270
Conjugate pairs, 135 Derivatives. 51
Consumption, 263 De Russo, P. M., 277
Continued fraction expansion, 85 Describing function, 213
Continuous feedback system, 227 Determinant method, 10, 136
Contour integration, 24 and table form, 101
Controlled process, 51 relationship, 101
Convergence, 179, 266 Determinant method and Hurwit7. crite-
Convergence of solution, 263 rion relationship, 97
Convergent infinite series, closed form, Determinant methods, 79
33 newey, A. G., HIli
INDEX 323
Difference equation, 264 Double-difference system function, 67

fourth-order, 266 Double z-transformation, 73
Difference equation with noncon5tant
coefficients, 68 Economic fluctuations. 266
Difference equations, 219 Economic systems. 45. 248
linear, I, 248 Egan,1. F., 276
nonlinear. 248 Eigenvalues, 208. 307
simultaneous (with periodic coeffi- Elspas. B., 270, 276
cients). solution of, 58 Equilibrium point. 2S0
Difference equ:.tions whose coefficients Equilibrium points, 2 10, 212
are periodic functions, 16, 48, 57 Equilibrium solution, 180
Difference-differential equation, 16, 219, Equilibrium value. point, 175
305 Erdelyi, A., 77
mixed,51 Error, analysis of, 239
and sum equations, 142, 158 approximation, 239
unmixed,64 Euclidian algorithm. 130
Differential equation, 243 Euler-Maclurian formula. 232
Differential equation in the variable z, Evaluations of the integral of the
59 square time functions. 160
Differential equations, approximate so- Extrema, 309
lution, 219
first-order, 52 Feedback continuous system with lime
time-varying. 241 delay, S3
Differentiation with respect to second Feller, W .• 78, 276
independent variable, 9 Fibonacci's numbers, 305
Digital computers, 79,219 Fibonacci sequence, 301
Digital-com.p.~!!!.r.approach, 231 Fictitiously advancing delay, 17
Digital control systems, 28, 189 Fi<:vet, M., 78
'Ufgii3i filter. 263· Filtering theory, 248, 258
Dirac (delta) function, 37 Final value theorem, 51. 194,274
Direct method, 130 Finite field, 267
Dirichlet series transform, 38 Finite pulse, train of, 190
Discontinuities. 27 Finite pulsed sampled-data feedback
Discontinuity, 3 system, 190, 192
Discrete antenna array, 254 Finite state, discrete Markov process,
Discrete filter. 315 248,270
Finite summation, 5
Discrete filters, 156
Forcing function (sinusoidal), 208
Discrete Markov process, 320
Fort. T., 38, 42, 77
Discrete Markov processes, 270
Forward difference, 4
Discrete square error, 311 Fourier polynomials. 213
Discrete systems, 36 Fourier series, 32, 148
higher-order, 253 summation, 32
time-varying, 81 Franklin, G. F., 42, 173,246
Discrete time function, 2, 278 Freeman, H., 43
l)i~til1ct im:ducihle (1olynomi:lls, 268 Frequency 1Iliasing, 258
Division, synthetic. 10 Frcqucncy entminment, 211
llivisiClII IIlI'thllll, 7'1, III'). I W Frieillami. It .. 4~, 76, 173,276
Hurlsdl. (i. "1 .., I. n Frnlt'll trlll1!1ft'I' function. (,'I, 71
324 INDEX
Fujiwara, M., 84, 139 Independent variable limit value, 9

Fukaw, A., 73 Infinite converging series, 16
Fuller, A. T., 141 Information theory, 248
Fuller's conditions, 112, 113 application, 258
Functional relationships, 179 Initial and final values, 5
Fundamental malrix, 272, 274 Initial conditions, 53, 185,223,268,
270,316
Gallieri, C. A., 43 Initial state probability vector, 320
Gantmacher, F. R., 139 Initial values, 235, 252
Gardner, M. F., 38,42, 172 Input-output relationship, 67
Gelfund, A. 0., 76 Integral time function transform, 149
Generating function, 1,270,303 Integrating operators, 229, 246
Generation rule, 105 Integration, rectangular rule, 229
Gibson, J. F.., 247 Simpson's three-eighth rule, 230
Global period Tn' 192, 195 trapezoidal rule, 230
Goldberg, S., 76, 276 Weddle's rule, 230
Goldstein, M. S., 247 Integration with respect to second inde-
Gradstein, E. S., 78 pendent variable, 9
Growth factor, 264 Integrator, 194
Guichet, P. L., 218 Interpolator, 220
Gupta, S. C., 140, 141 Interpolation, linear, 223, 224
point-slope, 225
H N (Hermite), 282 quadratic, 223
Hadamard, 82 second-order normal, 222
Hartmanis, J., 276 trapezoidal rule, 225
Hayashi, C .. 212,218 zero-order normal, 224
Heizman, C., 246, 247 Interpolators, first-order normal, 222
Hellman, 0., 247 higher-order normal, 222
Helm, H. A., 42 second-order normal, 222
Hermite polynomial, 13 zero-order mean, 222
Higgins, T. J., 277 zero-order midinterval, 222
Ho, E.C., 76 zero-order normal, 222
Hold,200 Inverse !J(z) - F(s) or (z - s) trans-
mathematical, 220 formr-tion, t 48
triangular fictitiolls, 229 Inverse Laplace and modified z-trans-
Hold circuit, 258 form, equivalence, 31
Howard, R. A., 276 Inverse Laplace transform, 23 I, 235
Huffman, D. A., 276 Inverse <m-,c-transformation, 147
Hufncgel, R. G., 77 Inverse z-transform, 232, 253, 265, 266,
Huggins, W. H., 247 274
Hurwitz cor,dition, 83, 136 tables, 46
Hurwitz-Routh, 79, 84 Inversion with respect to wand z suc-
Hybrid systems, IS cessively,74
Investment, 263
Impedance ratio, 317 desired, 265
Impulse function, 21 Irrational function of z, 13, 6 t
Impulse response, 66 Iterative methods, 69
Income received, 265 Iterutive process, 150
Incremental response transform, 190 convergent, ~ ~K
INDEX 325
Iwens. R. P .• 277 Levy. H .• 76, 188

Izawa. K .. 216 Lienard-Chipart. 79. 84. 141
Limit cycle, 149, 313
III (Bessel). 286 existence. 201
I", (Bessel>. 286 fundamental equation, 194
Ip (Bessel). 281 limitation on period, 189
James. H .• 42 period limitation. 203
Jordan. C .. 76. 247 sampling instants. 205
Jumps. 17 stable. 207
Jump-function method. 38 Limit cycles (stability study). 207
Jury. E. I.. 42. 43. 77. no, 139, 140, Linear array. 256
141. 172. 173, 188,216,217,218, Linear difference·differcntial equations.
246. 247, 275, 276. 277, 296 51
Linear differencc equations with con-
Kadota. T. T., 217 stant coefficients. 4S
Kalman, R. E., 216, 218 Linearized system. 182
Kaplan, W .. 44, 77. 172.218 Linear partial difference equations, 74
KloUer. K., 212, 218 Linear space. 3
Knopp, K .• 141 Linear time invariant circuits. 32
Kroneker delta, 66 Linear zone. 199
Ku, Y. H., 42. 73, 172 Linearity, 3
Kuo, B. Coo 44, 218 Load,46
Long division method, 219
),-transform. 240 Lowenschuss, 0., 43
Ln (Laguerre), 282 . Low-pass filter, 2~8
Ladder network, 46, 304 characteristics. 196
Lago, G. V.. 43 '
Laplace. P. S., 42, 270 Ma, M. T .• 276
Lapluce and modified z;-transform, rela- Madwed, A., 77,246.247
tionship. 26 Marden. M., 79, 127, 139
Laplace and z-transform, 64 Markov process, stationary, 274
relationship. 20 Matrix, 208
Laphlce-Sticltjes integral, 36 method,45
Laplacc transform, 1, 142, 141. 182, Maxima, 258
190.226,231.315 and minima points. 20
convolution. 253 Maximal.period, 269
convolution theorem. 167 Maximum value of the gain. 109
discrete, 39 Mean square error, 259. 261
discretization, 232 Mean square value, 164
generalized, 37 theorem, 29
of a sampled function, J67 Meromorphic fUnction, 175
one-sided. 20 Meschkowski. H.• 42. 77, 276
theory. 166.258 Meserve. W. E .• 217
I.attice network. 73 Mikusinski. Jan, 43,78
I.uurcllt-( 'illlchy's trunsform. 39 Mikusinski's operational calculus, 40,
I.anrcllt IOcl'ics. 11 41
I ",wllt·n. I). II .. 4 \,77. 172 Miller. K. S., 43. 78
I "FI'VII', 1' .• ~')') MillnlUn, J., 247
I '·'Mlllln. V. "". I KH Milnt· ·/lullm(III.I .. M., 76
326 INDEX
Milnes, Ho Wo, 140 Nonlinear difference equations, 179,

Minimax principle, 117 189
Mishkin, Eo, 247 second-order, 177
Mixed form, 32 Nonlinear differential equations, 175,
Modified l-forms, 231, 236 253
Modified l-lransform, 15,48, 192,220, application to, 243
259, 302 Nonlinear discrete system, 142, 174,
complex scale change, 18 189,206, 312
differentiation with respect to m, 18 forced oscillations, 208
division by r, 18 limit cycle, analysis of, 189
finnl value, 18 stabilily study, 206
initial value, 17 Nonlinear functions, 175
integration with respect to r, 18 Nonlinear gain amplifier, 199
inverse, 19 systems, 189
maximum or minimum points of Nonlinear sampled-data system, 208
/(1), 19 Nonlinear sampled-data feedback sys-
multiplicution by tAO, 18 tem, 248, 313
of a kth derivative, 20 Nonlinear system, second-order, 248
pairs, 289 Nonlinearity, incidental, 181
real translation, linearity, 18 magnitude of, 179
summation of series, 19 small, 179
Modular field, 268 Nonsingular solution, 207
Modulated discrete signals, 149 Nontrivial solution, 18S
Modulating frequency, 317 Nonuniform arrays, 254
Modulo-p, 268 Nonuniform envelope function, 256
Modulus, 17.5 Norlund, N. E., 76, 188
Monotonic conditions, 116 Nulls, 258
Monroe, Ao Jo, 77,173,218 Numerical analysis, 219
Montel, Po, 42, 77, 173, 188,276 Numerical equation, 266
Mori, Mo, 43, 172 Numerieallechniques, 246
Mullin, F. J., 42, 77, 216, 247 Nyquist diagram, 82
Multirate error-sampled feedback sys-
tern, 310 Operator, shifting, 37
Multiple poles, 272 Operators, 36
Operators exponential, 315
Murphy, O. J., 218, 276,277
Optimization procedure, 259
Optimum filtering, 261
National income, 263
Orlando's formula, 97
Naumov, B. N., 247
Orthogonal, 31
Nease, R. Fo, 216
Orthogonal discrete frequency, 213
Necessary and sufficient condition, 106
Oscillations, 196, 202, 203
Nelson, Wo L., 217 excessive, 112
Network theorl, 45 forced, 189, 196
Neufeld, J., 42, 77 free, 196
Nichols, N., 42 harmonic, 209
Nishimura, T., 173,217,277 periodic, 207
Noise power gain, 160 periodic mudes uf, 141), I HI)
Nonlinear difference equation, 207, 312 rch,y mude nf, I K9, 2() l, 2()~
periodic solution, 215 ~ymml'lrk·nl, 11)5
vector form, 207 Osh''''''i, R. K, 7.77
INDEX 327
p-phane, IcCt half of, 23 Probability theory, 73

right hllif of, 22 Programming, 219
P n (Legendre), 281 Pulse, of unit area, 21
Pnm (Legendre associated function), width of, 21
282,286 Pulse feedback, 69
Pai, M. A., 77,172, 188,217,276 Pulse-code modulation, 258
Partial derivatives, 208 Pulse-frequence modulator, 70
Partial fraction expansion, 10 Pulse-frequency modulation, 69
Partial difference equations, solution, Pulse transfer function, 249
73 Pulse width h, 200
Pasteur, Franceline, 43, 218 Pulse-width modulation, 200
PavJidis, T., 140 Pulse-width modulated feedback sys-
Period of the sequence, 268 tem, 190
Periodic coefficients, difference equa- PWM (pulse-width modulated) system,
tions,57 189, 196,313
Periodic forcing function, 189 Pyshkin, I. V .• 217
Periodic inputs, 16
Periodic regime, 79 Quadratic operator, 225
Periodic sawtooth voltage, 33 Quantization. 201
Periodic solution, 207 Quantized level amplifier, 20 I
Periodic states, 210 Quantized levels, 201
Periodically switched network, 317 Quantized level systems, 189
Periodically varying steady state, 192
Periodicity of I,(t) and g,(t), 48 Radar detection systems, 267
Perron, 0., 140 Ragazzini. J. R., 42, 43, 173.246
Perturbation, method, 129 Random signal. 163
small. 207 Randomly varying parameters, 79
Phase lead, 254 Rational coefficients, 59, 64
Phase plane, 198 Rational function. 10
space, 213 Raymond, F. H., 77, 247
Phillips, R., 42 RC circuit, 70
Physically realizable system, 231 Real convolution, 6
Physical systems, 45 Real polynomial, 105
Physiological systems, 45 Reciprocal pairs, 130
Pinney, E., 76, 212, 218 Rectified sine wave, 35
Poincare, H., 77 Recurrence method, 45
Polarity, 204 Recurrence relationship. 174. 175, 243,
Poles. 69, 194.219 252,272,307
multiple, 12 Relationship between coefficients, 116
simple, 11 Relay output, 215
Polynomial, array. 254 Relay sampled-data system. 314
in z. 60 Relay system, 196. 213
Polynomials in z-l, 10 autonomous, 203
Powc r pallern. 257 Relay systems, 189
I'uwcr licrics, 219 Residue, 274
finilc,IR2 method,311
IIIrlhllll ••, RCliitlues, II, 145
I'uwl'" ,.,r\" II 11111 t1l'lI~ily, It. I ~1I111 nr. I(IR, 27.~
"I
l'lt'llil'! lilli, I
1'1 illl" IIIIIUhrl, 'tlR
Rr~I"III~" IIf 1\ linrllr circllit for periodic
hllllll, "
328 INDEX
Riemann sum, 232 [286 Samuelson, P. A., 76, 139, 276

Riemann-Zeta function, generalized, Saturating amplifier, 199,248
Ring of convergence, 161 Saturating amplifier sampled-data feed-
RL circuit, 33 back system. 312
Robertson, H. H., 246 Saturating gain amplifier, 213
Romanov, M. I., 140 Saturating region, 201
Romanov conditions, 112, 113 Saturating-type nonlinear sampled-data
Root distribution, 79, 309 system, 248
Root locus method, 82 Saturation-type nonlinearities, 174
Roots, 308 Sawaragi. Y., 173
characteristic, 54, 219 Scarborough. J. G., 247
complex,9S Schelkunoff, S. A., 276
distinct, 61 Schur, 1.,79. 140
of the characteristic equation, 272 Seal. H. L .• I, 42
on the unit circle, 119, 120 Sequence, 2
maximum modulus of, 309 Sequence t-transformation, 192
number inside a circle of radius Sequential circuits, 248
.,<1,119 linear, 267
number of, 95 Sequential computation, method of,
reciprocal, 119, 120, 135 243
simple, 11,23 Shah, I. K., 277
Rouche, E., 97, 124, 140 Shaping filter, 263
Routh rule No.2. 108 Shift operator E, 39, 60
Ryzik, E. M., 78 Shifting the lime axis, 185
Shifting theorem. 3, 256
s-plane, left half of the, 79 Shifting transformation. 119
mapping of, 144 Sc:hur-Cohn determinants, 84, 85, 96,
Salzer, I. M., 230, 247 127
Sampled-data control system, 302 Signal band, 258
Sampled-data feedback system, 108, Signal frequency, 208
160,248 Simplification of equations before t-
configuration, 29 transform application, 64
with variable gain, 61 Simpson's rule, 225
Sampled-data system, 189,214.227 Single-valued nonlinear function, 216
multirate, 149 Singular case, 96
time-varying, 306 Singular cases in determinant and table
with a small nonlinearity, 181 forms, 128
Sampled-data systems, application, 28 Singularities, essential. 9, 12
Sampled-data transmission link. 258 isolated, I I
Sampled spectral density, 311 regular, 23
Sampler, actuation of, 317 Singularity, 80
Sampler phasing, 319 Sittler, P. W., 276
Sampling frequency, 208, 317 Skip-sampling,operution, 193
Sampling interval, 243 process, 195
Sampling period, 51, 190, 206, 228, 255 theorem, 149
choice of. 238 Slowly vnryinll discrcte funclions, 2111,
Sampling period T, 190 Slt.wly varyinM ~y'll'IIl, f,f.
Sampling rate system, periodiclllly S,'cl'lml ,ll'l\\iIV, .:!w
time-vuryinM, lOS St"hilily, f,I" !f.I, '2""
SamplinM theorC'IIl, 22K. 2111 hlllllllllil V \,111 vt'~, Jilt.
INDEX 3'29
SI:lhility, consideration, 79 Symmetric function. 164

con5111nls, 86 Symmetry condition. 196
dcfinition, 80 System theory, various area appliclI-
Ileterminants, evaluation of, 91 tions, 248
of limit cycles, 79 System transfer function. 67, 68
sufficient condition for, 71
test, J09 Til (Tchebychetr), 282
tests for, 82 Table form. 97. 137
directly applied in z-plane, 84 and determinant method, relation-
Stability and number of roots, theorems ship, 101
related to, 116 of stability. derivation, 121, 123
Stability constraints, 87, 88, 91 Table method, 79
critical, 94 Tamburelli, C., 44
for low-order system, 93 Taylor's series, 315
generation of, 105 expansion, 9, 223
simplification, 88, 100 Tchebychetr polynomials. second kind,
Stability criterion, alternate form, 92. 288
93 Thaler. S .• 231,232, 241, 246, 247
new, 88 Thellier. P. L. 172.277
simplification, 127 Thoma. V. M .. 140,141
summary, 136 Thomson, W. E., 247
Siable, asymptotically, 175, 207 Time-axis shifting, 253
state transition, 274 Time-homogeneous process, 271
system, 251 Time-invariant, linear, 240
Staggered sampled-data feedback sys- Time-invariant systems. 174
tem, 319 Time lags, 192
Siaircase function, 36 Time-varying ditrerence equations, 59
State probability, 320 Time-varying discrete system. 81
Stute vector, 213 stabilily condition, 81
Stationary process, 273 Time-varying equation, 316
Stationary signal, 164 Time-varying systems, 174
Statistical design, 160 Time-varying transform method, 66
theory, 259 Time-varying z-transform, 66
Stelldy state, the, 193 Tinbergen, J .• 266, 276
Stelldy-state amplitude, 178, 251 Torng, H. C., 188,217,218
Steady-stute error, 25 I Total savings, 265
Total squared error. 3 I I
Stelldy-st:lle output, periodic, 209
Total square integrals, derivation, 1(,8
Stcady-stute part, 32
formula, 168
Stcady-slate relation, 178
tables, 164, 297
Stcady-stllic I'esronse, 303 Tou, I. T., 42, 173,246
of a lincnr time-invariant circuit, 35 Toymaker, 319
St~':uIY'~latc valuc. 17!1
prohlem. 319
Shll'lH"li~' lIIalrix, 271 Tl'lInwclltlenlal equations. !l6
Slunt·. W. M .• 17.01.'. 7M, 172,27(, Tr:llls\'~'ml"1I1:I1 fum·lion. 232
Siril' IIf illlnlvlll'ilv. 1·11. 1·1·1 I. :1II,r,'" 1'I11l1'lillll •.~ I
SIIIIIII, W .• IKK II lin, ... " III II,,'lhllll,. 1(".1 ~
SlIlIIllIlIlIulI ur ~"I h·~. 1·1.' III",' VIII viII". fit,
1111111111' 1""llIlIlIr, I H IllImhllllllllloll., 11111 1111111111', 1·"1
SIIIIIIIIUlhlllllll'.II,·III. ~III,.IIII. " 111I1I~h'lIl ~h.h·, I'll
330 INDEX
Transition matrix, transpose of, 275 White noise, 164

Transition probability, 271 Widder. D. V.• 42
Translational form, 45 Wiener filter, 263
Transmission channels, one-way, 31"6 Wilf, H. S., 140
Transmission line, 73 Wilts. C. H., 43
Transportation lag, 5 I Wolf, A. A., 42, 78. 172
Trapezoidal rule, 232 Wloka,J .• 78
Tratakowskii, G. P., 69, 76
Trijinsky, W. J., 188 I-transform method, 40
Trudi's determinants, 112 z-forms, 23 I, 233
Truxal, J. G., 43, 247 illustrative example, 237
Two-dimensional z-transform, 73 method, 241, 314
Two-sided z-transform, 165,241 procedure, 235
extension to continuous output, 163 relationship, 231
Tschauner, J., 42, 77, 139, 141 table, 234
Tsypkin, Y. Z., 39, 42, 43, 139, 172, z-transform, 192, 220, 243. 248, 253,
188, 218, 275 278, 316
see also Cypkin, J. A. alternate, 44
Tustin, A., 246 application, to probability theory,
270
Uncorrelated noise, 26\ to vector difference equation, 272
Uniform array, 256 convolution, ]42
Unilateral repeater, 316 dofinition, 1
Unique inverse of !f(z), 161 direct, 213
Unit circle, 69, 232, 275 inverse, 9
inside, 79, 208, 308 low-pass transformation, 240
on. 21 I one-sided, 160
Unit pulse, 190 properties, I
Unit pulse of duration one, 38 regular, 195
UrkowilZ, H., 277 theory, 254, 2S8
time-varying. 66
Van Der Pol, Balth, 42 two-sided, 160
Van Der Waerden, B. L., 276 z-transform method, 174, 21S. 267
Vagant, A., 247 applied to economics, 263
Vector difference equation, 272, 320 approximation techniques, 219
direct, for determining true oscilla-
z-transform, application to, 320
tions, 213
Vector equation, 171
discrete antenna array analysis, 254
Volgin, L. N., 44
solution of linear difference equa-
tions,45
w-transform, 67
z-transform pairs, 278
inverse, 75 z-transform expansion, 41
Wasow, W., 175, 188, 231, 232, 233, z-transform of special functions. 150
247 zw-transform, 74
Weaver, L. E., 216 Zadeh, L. A., 43, 66, 76, 77
Weber, E., 180 Zero initial conditions, 304
Weiss, L., 247 Zero-order hold, 249
Whalen, B. A., 172 Zeros, in the unit Circle, 122, 123
Wheelon, A. D., 172 of .. polynomial, 121

Eliahu Ibrahim Jury - Theory and Application of The Z-Transform Method (1973)

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Eliahu Ibrahim Jury - Theory and Application of The Z-Transform Method (1973)

Загружено:

Авторское право:

THEORY AND APPLICATION

OF THE z-TRANSFORM METHOD

IINIVERSITY OF CALIFORNIA, BERKELEY, CALIFORNIA

I{(>BERT E. KRIEGER PUBLISHING CO.

In Chapter 2, the z-transform method is applied to the solution of linear

2 .:-TRANSrORM METHOD OF SOLUTION

3 STABILITY CONSIDERATION FOR

1'1« IIII.I:MS 301

1.1 Discrete time function and

== Tz-J [1 + 2z-1 + 3z- 2 + ... ] == Tz 2'

1.2 Properties of z-transforms5 ,14l,25,26

Linearity of the z-transJorm

Thus 1 is it linear operatur on the Iineur !lpace of ull ::-transformable

If" III 0' :4' (:),

.)I{(/' '1')1·-:1:1'(:) /(UI)I

1[f(t + T)] = I f[(n + l)T)z-n

where k = n + 1. By adding and substracting /(0+) term under the

1[/(1 + T)]" = zl~/(kT)Z-k - I(O+)J = z[§"(z) - 1(0+)]

1[/(1 + mT)] = zm[§"(z) - ],1/(kT)Z-"] (1.8)

COROLLARY: If 1[/(1)] = .F(z), then

1[/(t - nT)u(t - nT)] = Z-II§"(Z) where U(/)

.1[/(1 - nT)u(1 - nT)] =I [f(m - n)Tu(m - n)T]z-,n

= z-n I {t[(m - n)T]u[(m - n)T]}z-''''-"' (1.10)

.1[1(1 - nT)u(1 - nT)] = Z-II I f(kT)II(kT)z-"

= z-n I'" l(kT)z-k = z-"§"(z) (I. J I)

Complex scale change

~1(::) = J[i/(kT)] = _%_,:F(z), for Izl >1 ( 1.22)

We readily notice that the initial value is obtained as:

Iff(O) = 0, we can obtainf(T) as the lim %?(z).

if the limit exists.

Complex multiplication (real convolution)

!F1(z)!Fa(z) = 1~/I(kT)fl(n - k)T] (1.29)

but from equation (1.11)

but 12 (n - k) T = 0, for n < k and ~erefore

~l(z)j~'iz) = 1[~/1(kT)/2(n - k)TJ (1.33)

Complex differentiation or multiplication by I

l[tfJ = - Tz !!.. ~(z) (1,34)

The term in the brackets is a derivative of z-n with respect to l.

As II l'llrnllary In Ihis Ihcnrclll we can deduce

wlll"l' 11Il" rUIlction nUl iN lIivon by

"fA I _ 1/(1/ • 1)(11"- 2) •. ,(II ., k -j' I) ( 1.4U)

As special cases of the preceding.

1[( -1)"n 1lt1(n - k + 1)] = zdltF(z)

F(zlt) = I[V!t(n1)] (1.43)

and [n/k] denotes the largest integer in n/k.

IL!l/(mT)] == io ,,~,/(mT)z-" == l/(mT) 7I!ltz- n

Differentiation with respect to

Second independent ('ariable limit t'alue

.llim f(t. a)] = lim .'F(z. a) (LS1)

Integration wilh respect to

J[1:1f (t, a) daJ = f~1 :F(z, a) da (1.52)

if the integral is finite.

1.3 Inverse ~-transform and branch points 14 •a7

'J111' pm"c'r ,vc'r;c',I' mc·tlmt!

From equation (I.l) it is observed that

§l(z) = f(OT) + f(T)z-1 + .. .f(knz-k + .. .f(nnz-n + ...

§l(z) = Po + P1Z-1 + p"z-I + ... + P..z- n

PI = f(l T)qo + f(OT)ql

Pn = f(nT)qo + f(n - I)Tql + fen - 2)Tqa + ... + f(OT)q ..

Partial fraction expansion

We can easily identify the inverse of a typical !F/t;(z). from tablest or

t Sec Appendix, Tnblc I.

Complex integral formula

i$&"(Z)Z,.-1 dz = f(nT)' 2'1Tj (1.59)

The contour r encloses all singularities of .~(z) as shown in Fig. 1.2.

f(nT) = _1_. .1 $&"(z)zn-I dz = sum of the residues of §(Z)Z,.-I