- -~ '!

HE PlllSE TRANsffR FUNcTloN AND THE i-llANSfa 1UNC1cN

•. "! 125
Fiallloas Sampler

-~.•\.:'' .

Transfer Fu,i~~fo.ns,
.~
n,__::__J
• (I)
C(sf R(s) S
1 R*(s)
G(s) } · :
_ -~ -~ _
. (b)
,--X--~ti
J .
I
S2 C-(s)

c(I)•
C(s)

i,I
\\ Block Diag~iims,-- . llglrt'-1. (a) Linear system with continuous-data input. (b) Linear system with sampled-
daca input. ·· · · ·
i
and .Signal -F!~ip ~_.: ~' .

Graphs · ~-",.·./: . ;,.:-: :i;

assume that the initial conditions of the systems are all set to zero. If the same
system is now subject to a sampled or digital signal r*(t~ where the latter is '
simulated by the ideal sampler S 1 , as shown in Fig. 4-l(b), the objective now
, ..
is to Jind-Lnput-putput relations of the discrete-data sys!em an4, if possible, an
. , I ..-J, ·.· ' inpik>uYput transfer function tlt~ Laplace a;d the z-domairirnc Laplace
., !ransf<>nn of the output of the system in Fig. 4-l(b) is written as ·
KEYWORDS AND TOPICS C(s) = R*(s)G(s) (•2)
Where R*( ) 1· h ·
i' pulse transfer function • z-transfer function' • characteristic equation • block Alth s _s t e Laplace transform of r*(t) and G(s) is as defined in Eq. (4-li ·
I
diagrams • signal flow graphs • multirate sampled systems .. ~ugh this transfer relation is easy to achieve, it contains a mixture of
is d ~e and analog signals and is difficult to manipulate analytically. Thus, it
1
1tla:r;: e to e.xpress the system characteristics by a tran~fer relation that
4-1 INTRODUCTION introd (t) to c (t~ the sampled form of the output c(t). For this purpose we
4-i(b).;~ a fictitious sa_mpler S2 at the output of the system, as s~own in Fig.
Thus far our investigations or discrete-data and digital control systems hal'C the sa e sampler S2 1s synchronized with the sampler S1 ; that ts, they have
.focused on the mathematical treatment or the sampled signals and digital~!; have ~;hsampling period T and sample at the same instants of time. Now we _
Now we must consider the modeling and analysis of systems that are su.,,..- clirccu f c(t) and c*(t) at our disposal. Two useful expressions for C*(s) follow
to discrete and digital data. , , ·"·" , , . , .1•.·, , ; , edll, Y rom Eqs. (2-53) and (2-54~ and are written below. .
Just as in the analysis and design of linear analog-signal control sySt ,
I. the concepts of transfer function, block diagram, and signal flow graph can
be applied to linear discrete and digital control systems. •
' ' 1
· C*(s) :...-
·. . T•= -:- "'
"'
L
. .
C(s + jnw,)
.
' ..
.,,"
.
. c(O) = 0 · · ("3>
l... . ' '• ) ' I ;,

' .c•(s) = L"' c(kT)e-m

J ,. ,.
4-2 THE PULSE TRANSFER RJNCTION AND THE :-TRANSFER RJNCTION _.-/.
s· . ' (~
lln1lar e . t=o
, F Xpressions
·
It is well known that the transfer . continuoid'
function of the linear system with -·- ' , , r~lll Fig. 4-l(bc)an ~ritten for R*(s).
., • C (s) 1s written ' ·' '
data input r(t) shown by'thc block' diagram in Fig. 4-l(a) is defined a~ C*() 1 "' . , , ·,·
s ::::- 1 "' '
G(s) = C(s) .
,,.e Usin .. T •='="'
C(s + jnw,) = - L R*(s +J~o,J(,(s +jMJJ (A-5>
lhe1&thepe·. ' . r •.
a flod1 ·· . , ·
-1/l) ·. ·

. R(s) utf. st equation bee property of R*(s) of Eq. (2-76~ i.e., R•(s + jnOJJ • R*(s~
where R(s) ~nd C(s) arc the ~place transforms of the input ~t) an'thC
, ~ we ·:..d
tpr
0 comes • .
c(ti respectively. Note that, in defining the transfer function G(5 (U)
1
f24
C•(s) = R*(s)- L"' G(s +i'!°'J
Ta."".,
r
~ -:i

.'
, ..,
'
''
126 1INMI RINCll0NS, 11.oa DIAOIAMS, All) SIGNAi. R.ow GWHs

Now define G*(s) as

r
I I
1HE pULSE TIANSffl RINCTION NG H ,-1INMI IUNCll0N 127

1 "' · , f the fictitious sampler S2 is described by

G*(s) = - L
G(s +jnw,) .,, output o' '. , . -
·' -, ,,, '. ;, ,1 ,, ,: = L g(_k1)5(t -
' ' ,.,· • • (I)
T •"'-"' c•(t) =:= g•(t) kT) (4-12)
Substituting Eq. (4-7) into Eq. (4-6), we get - - I <'' . I ' 11:20

kD. for k = 0, 1, 2, .. . , is the weighting scqu~nce of the 5 t

'1 -, . .

C*(s) = R*(s)G*(s)
When the sampl ed ~1gna
where {J( . I r •(t) ts. ~PP1·ie.d as mput
. ys cm.
~o the linear system, the output
The Pulse Transfer Function G'lt(s) '· ' of the system is written as a senes of impulse r~pon~, each weighted by the
. 'dual values of r(kD, Thus, the output c(t) 1s wnttcn .
in d1v1 . ·
We can define the puJ,, transfe, function of lhe sys1em in Fig "-lib), c(t) = r(~)g(t) + r(T)g(t - + r{21)g(t - 2T) + .. : n
(4-13)
- . . C*(s)
,l •\, , G*(s) = - At t == kT, where k is a positive integer, the last equation gives
f
R*(s) c{kT) = r(O)g(kT) + r(T)g[(k - l)T] + ··· + r(kT)g(O) (,._1A)

i l
The ,-Trusler F•tioa Gti) - . _.

in Fig. 4-l(b):z = eT• in Eq. (4-9),

Substituting :r-. ·
we have
''
"I.

the z-transfer Junction oT the

The last series is truncated at the (k + l)st term since g(_t) is zero for t < o.
Equation (4-14) is written _
L• r(nT)g(kT -
.
(4-15)
' c(kn =·
a=O
nn

G(z) == Taking the z-transform on both sides of the last equation, which involves
_m~ltiplying both sides by z-• and summing from k = 0 to k = oo, we have
; R(z)
where G(z) is also defined as '~I

i l i
l
· .. «, •·I

G(z) == L g(_kT)z-•
•=o
In the last equation g{_k1) denotes the sequence of the impulse _respomc
,_,
(4-t'
C(z)
.
= L
co
L• r(nT)g(.kT- nT)z-•
•=-(1) •~o
· -·· - · Now, using the real convolution theorem of the z-transform of Eq. (3-130~
(,._16)

9(~ or the.,...,. daorihod by the tnn,fe, fonction G(,). In othe, """'-~

!
the last equation leads to

The-
I;,
= R(z)G{z)

,
I
0 the.._ 1.op1,o, bamfonu or G(,( The "'luen« o(k1), k : l,Z.,, C(z) (..,17)
.......... "the-.,- o, tho .,/ghth,g llquen,e ~rth, ,y,tom .:ii
O that the,........, """"on, or a lmea, system • "" where G(z) is as·defined in Eq. (4-11~
a- r...,.,. ;,, the "°'Pkd '°'"""-• whethe, it " fo the,."'. ·,. ' J

_..., "" - in the - way " ,ampl,d sign,Js Ba«d " . ., ' .
! discussions
the transfer given above,
relations we can make_
of discrete-data the following observations concemrDJ
systems. 4-2-1 Discrete-Data System with cascadee:f eements $epCJfOl8d
bya Sampler
L To,_"' ........t,., t,_,..,,. ,;;.tlmu 1n a u,,;fo,mly iampkd/.. When a discret~-data or digital control system contains cascaded elements.
.:,
) fa,,;,;,,,,, - - .. _,_,. _ , , tt,, OUlput •f the''""" ~re must be taken in deriving the transfer relations for the overall system.
-l 2. n,, •·•an,f~ Jt,,,,,t.., "'""'"•-data ,,,,,;,, respomn ,JJ
Figure 4-2 illustrates a discrete-data system with cascaded elemellts G, and G1-
.i
u .. -1
.,
bctwcco the sampling
"'-.,..,, - ~ -tnstan1s
The i~romw;,.
1s lost. ·. o, .,the ""'•"'
_ .or the sy,t,m dtl
1 . 8;,,,, 11,, b,p,,1 of II,, du,,,1o-do,o ,,.,_ In ,,,._ '-l(b) & .,,..,,, ,ksc,/bld
I /,i ; hy ,he ou1pu1 of ,1,, -,,,,,,fo, all '""''kai ,.,,_, ,. .., ,1n, ly ~""'
flclilioUIS...-

l
. ril
•1_o:1 _litesampler ar,d regard the input to the_system 08 r"'(r). , , . 'P ,--X--~~
I S2 ~ ) -
I ii Ao '1ton,.,;., w,y or.,,;,ing" Eq. (4-JO) ;, to_,,. 'b, lmp;,;. io,poo•
thod
1
. Consider that at t = 0 a unit impulse h . 11$ input th e system
_ - 1s applied
I. ,/ • ~ -- I
me
.{ In Fl& 4-J(bi Tho ootpot o/ tho,y,ton " •mp1YI «mp, " . . .10_ 9(,). Th< I{,) S G1(s) ,r(t) G1(s) - -
;, 1 I ..__ ____, S1 0-(J)
j Figure ,._2, Digital system with sa~pler-separated eJcmCIIIS-

--
 t28

,.,.,.~
11ANSf11MCIIONS,llOC1ow;aAMS.MJ SIGNAL ROW~
• fUNCTlOH Of THE ZERO-OROER HOU> Al«J IHI llllAllON llflWaN Gl•) AND Glz) 129
The two elements are separated by a second sampler
2
s . .

synchronized to the sampler at the input, S1 • To fa .th1ch is id . • (4-21) is written

method, a fictitious sampler S2 is introduced at the sycis1em
1tate the z-t·
Citt~ fquauo 0 .
C(z) = G1Gi(z)R(z) <•U>
I• in Fig. 4-l The z-transfonn of the output of the overall 0 utpu~ ·
st
;
the following. .. __ · .. . . sy ern is d"· , b ve analysis can be extended to systems with more than two
.l
ii
. The transfer relat,ion_s of the two systems GI ~nd G2 are . . The :n~ the extension is straightforward.
elelllents, complex system configurations and closed-lo~p systems, a block
D(z) = G1(z)R(z) for I bra similar to that of analog systems, using the fundamental
diagram al gt ns defined above, can be used to find the input-output transfer
and transfer r~ a i~owever, the sampled signal flow graph method presented in Sec.
C(z) ;;,, Gi(z)D(z) relationships(. ly more convenient to apply. ·
. 4-4 is genera .
where D(z) is the z-transform of d(t), the output of G1 • Substitution · i;]
(4-18) in Eq. (4-19) yields , . . ,. .. . . of\
ON OF THE ZERO-ORDER HOLD ·AND THE RElATION
i C(z) = G1(z)Gi(z)R(z)
·.J (~
Therefore, we can state that the z-transfer function of two linear . · the' zero-order hold (zoh) is an important element in discrete-data
'1 j
. sqarated by a sampler is equal to the product of the z-transfer funclio11J of lit
two systems. - Since · systems, its transfer function should be treated with special interest ·
contro 1 . of the zo h 1s
The transfer function . given
. m. Eq. (2-86) and is repeated

· here:

'
"2·2 D1screte-oata System with Cascaded Elements not Seporafed ,:' 1-e-T•
G.a(s)=--- (•25)
b't a Sampler_ . s .
Figure 4-3 shows a d~te~ata system that contains two elements DIX
-r,]
Taking the z:transfonn on both sides of the last equation, we get
ICparatcd by a sampler. The z-transform of the output of the overall system
should be written as I
G.a(z) == !l[ - : (4-26)
i
I
qz) = .!'[G1(s)G2(s)]R(z) ('-21)
I

Apparcndy, since G1(s) and Gi(s) arc not separated by a sampler, they should Since the term e-r, in the last equation represents a delay of one sampling
be treated as one function, and the z-transform operation must be taken on the period, from the right:shift theorem of the z-transfonn given in Eq. (3-95), G.a(z)
product of the two functions as a whole.
is written
We can introduce the following notation for this purpose:
f' .
I
J· .!'[G1(s)Gi(s)] ""G1G1(z) .. G2G1(z) , (4-22) G.o(z) = (I - z- 1 )!!G]
' { which reads as "the z-traosform of G (s)G., •"' N • · .
r 1 i,.s,- oticc that, m general, z (4-27)
' . G1Gi(z)" G1(z)Gi(z) (4-23) =(I -z- 1) - = I
except in some special situations. ' ' ,· z-1
\f
.,ii\' This r_esult is expected since the zoh simply holds the discrete signal for one

fl
f:
, I ,--X--~t'L
C•(6r
.
~phng period, and taking the z-transfonn of the zoh would revert it to the
onginal sampled signal again. However, the above exercise simply does not

l a<sJI {.--a-,,-,]--, / .
/ I T
•1 ,.,,, .,I tK.rJ .· · ~ve any physical usefulness, since a zoh is almost never found just situated by
s, •~If. A situation that is commonly found in discrete-data control systems is
1 R(I) R•(s) I / 0<.1) . _ 2 _ ',

..._
Rgu,e 4-J. Digital system with cascaded elements. 1 own in Fig. 4-4, that is a sample-and-hold (S/H) unit is followed by a linear
system
bl . wit· h transfer function
' . of the bnear
G(s). The transfer function • system
f lhcoc.k !5 G(s1 and we are interested in finding the transform relation between
...
' input r•(t) and the sampled output c•(ti Again, we introduce • fictitious
 mANSIIR IUNCll()NS. 11.0CK ~~SIGNAL ROW GWHs fllPICTION Of IHI ZfllO,OIIOB 1101D Nil) IHI 11BA110N lffil&N ~•) N#J ~,)
130 .m

,--.XT --£.-~l
1 C•(s)
' I
115
well as
lim H(s) = R(s)
I (4-33)
1-0
I c(I)
,.(I) zoh /,(1) G(s) . . nificance of the results in the last two equations is that if a
s I -
R(s) R•(s) ! H(s) C(s) 1hUS, th e .:gta signal rl,t) is sent through an SI H device with a sampling period
Flgllt '"'- A system with sample-and-hold. continuous of the latter can be reverted to r(t) by setting the sampling period
1, theoutpu
110 zero.

sampler at the system output. The z-transform of the output c(t) . .

. . IS W111tCJi
C(z) = ~ [G~ 0(s)G(s)]R(z) ' . ...dl'° 4
1

fill"'~ . strative example o f the_11m1tmg

· . . process of the S/H unit, consider that
c,_ AS ~n 111 r(t) in the system of Fig. 4-4 is .
where
•I
. the inpu . .
r(t) = e-"'u,(t) ciJ.C)
1 - T, ]
~[G-0(s)G(s)] = [ G(s) fbe 1,aplace transform of r(t) is .
1
= (1 z-)~[~s)J
1
('~
I
R(s)=-
.-
s+a
('"35)

fhe 1,aplace transform of the sampled signal r-(t) is [ using Eq. (2-64)]
Toe last equation represents the important z-transfer function of the combm.
tion of an S/H device and a linear system wit_h the transfer function G(s). N~
1
that the (1 - z- ) is factored out as a rcsu)t of the delay theorem, and the
%t : : -e_q fl-T -_(l_!:e,.,- QT)
~le k].
R•(s) = eT• -eT•e-•T (A-36)

z-transfonn of G(s)/s must be taken as one function.

, . Tfie
loU};-i.-1)
Laplace transform of the output of the zoh is
It was mentioned in Chapter 2 tliat, in theory, when the sampling frequCDt}
1 - e-r, e7'
reaches infinity, a discrete-data _system reverts to a corresponding continuo111- (4-37)
H(s) = G~(s)R•(s) =-s--
eT• --e-• 1
data system, i.e., the S/H can be eliminated. However, analytically, this dots
not mean that, given
Taking the limit as T approaches i.ero in H(s), we get
-"[ G(s)J = G(z) (~ 1 - e-. r, eT• (4-38)
then lim H(s) = lim - 71 -•T
r-o r-o s· e - e

I lim G(z) = G(z) The evaluation of the above limit requires the use of L'Hopital's theorem

'I•
(4-3i)
T-o
of calculus. The result is
Since the z-transfonn t~cory i~ based on the impulse-amplitude modula• . 1 ("39>
tion of a continuous-data signal with the sampling period of Ts setting T to 11m H(s) = - = R(s)
,:1 i
zero simply "bunched" all the impulses together, and so it docs n~t make much T-0 S +Q
physical sense. In other words, if the signal ,(t) is sampled by an ideal sampler This shows that the limit of the output of an S/H operation as T approaches
to give ,-(t), then setting the sarnpling_pcriod _T to zero or, equivalently, w, to
infinity does not revert r•(t) to ,(t). This explain~ why Eq. (4-31) is not true in zero is equal to the input of the S/H unit.
'i,'!'
,i
-1' ~'1
general. However, if we first send the sampled s1gna_l /"(1) through zoh with .
, 11 ., the output designated as h(t), and then take the bm1t of T ... o, then8
rI A more useful and important property of the z-transform is to show that
Jim h(t) = ,(t)
1
I r-o (4-32) lim ~(G.0 (s)G(s)] = G(s) (Ul>
r- o
·1
t, -~
ra
.
, ClOSBM.00, fflllMS 135
i3' · 1INlfll Ml(:ll0NS, llOCK w,GRAMS, N¥J SIGNAL A.OW GRAPHS
C(s)
For the system in Fig. 4-5, th~ output of the sampler e'(t . G(s)
c(I)
as an input to the system, and the mput of the sampler e(t) caj ta
as an output, in addition to the true output _c(r). Once the to~
variables are defined; the sampler can essentially be eliminat CS
H(s)
input e(r) and output e•(t) are well_ defined. To summarize, fo~h
in Fig. 4-5, the input and output vanables are defined as follows: c
l. ~put variables:
output variables:
·,{!), ~•(t)
~t), c(t~ c•(t).
f111Ult
,l-6. Discrete-data control system with sampler in the feedback path.

.,
,.

\.
.
. -- Now WC write the input-output relations of the system using the ca111c No•, ,.1,s1it,ti?g Eq.' (4-55) ioto Eq. (4-57\ tho '"""" fw,cti<m bet"""
. effect principle; i.e, · R•(s) and c•(s) 1s ~ntten as , : ,. . .. . ·.: , ,
output = gain x input C*(s) G~(s)
R*(s) = 1 + GH*(s) (""59}
From Fig. 4-5, the following cause-and-effect equations are written:
E(s) = R(s) - G(s)H(s)E*(s) (1
The last equation can readily be converted to the°z-transfonn fonn.
:1
\
C(s) = G(s)E*(s) . (' 'c(z) G(z) (UO>
-=
Notice that only the output variables are found on the left-hand side ol R(z) 1 + GH(z)
equations, and only input variables are found on the right-hand side. ~I
taking the pulse ~sform on both sid_cs of Eq. (4-51), we get · where GH(z) =Eq.
Although _q'(G(s)H(s)].
(4-56) gives the .transfer function relation between the con-
E*(s) = R*(s) - GH*(s)E*(s) (411 tinuous-data output c(t) and the input ~•(t), the expression has little practical
where value since the transfer function has mixed analog and discrete quantities,
,,
1;'
GH*(s) = [G(s)H(s)]* making it difficult to utilize analytically. ,
. Figure 4-6 illustrates a discrete-data system in which the sampler is
1 _,
I .• ', « .
i~cated in the feedback path. In this case; the signal r(t) and the output of
. = - L G(s +jnwJH(s + jnw,) (414 the sampler, c*(t), are the input of the system; e(t) and c(t) are regarded as
·.. .T ••-.,
· · the outputs of the· system. Writing the Laplace transfonncd quantities of
Solving for E*(s) from Eq. (4-S31 we have
the output variables as functions of the inputs, we have the following equations:
'
(U1)
E*(s) =
l
l ·
+ GH*(s)
· R*(s) (4461 E(s) = R(s) - H(s)C•(s)
("42)
C(s) = G(s)R(s) ..'.. G(s)H(s)C*(s) ·
Substituting the iast equation in Eq, (4-52) the transfer relation betwco
R"(s) and C(s): ·;; , ,,.
· The reader should examine these equations closely and notice that, in addition
, G(s) . .· • to the system's transfer functions, the right-hand sides of the two equations
11 C(s)"'.1 +GH°WR*(s) . (A-66)
contain only the input signal, whereas the left-hand sides contain only the
, ';.

i Taking the pulse transform on both sides of Eq. (4-52), we get output variables. Taking the pulse transform on both sides of Eqs. (4-61) and
I' I C*(s) = (G(s)E*(s)]• (4-62), we have (4-63) .
·: I i · =G0(s)E•(s) E*(s) = R*(s) - H*(s)C*(s)
:.1 , , (~7J (~)
where it has been rccogni7.ed that the pulse transform of E*(s) is still E*(s), since c•(s) = GR*(s) - GW(s)C•(s)
' . 1 ., .. '
[E"(s)]• "'.' T•• .. E*(s +}lkOJ = E*(s) .: where ("65)
("'61l
GR*(s) = [G(s)R(s)]*
;i
1~-
,a -.
--=-- ,.;:._ ·-.- ·.==
llf
r ·J
ClOSID-l.oo, fflllMs 137
11ANS111 fUNC1l0NS, llOCIC DIAGRAMS. AND SIGNAL R.OW GRAPHS
t36

· and ·
GH•(s) = [G(s)H(s)]• 1

- ~ 1-3 · 'der t ha t the forward-path transfer function G(s) of the system h . ·

s own m
Solving for c•(s) from Eq. (4-64), we get coos• , ·,
fig, 4.S is
1 10
C•(s) = l + GH•(s) GR•(s) G(s) = s(s + 5) (..73)

(4ii
The z-transfonn of the output c•(t) is written directly ftom the la •' aod the feedbac~-path transfe·r function is H(s) =1 .
1 f • o.t s Applym8. Eq. (4-60), we """' · The ,..,plmg period q
1 s equatiai;
C(z) = 1 + GH(z) GR(z) G(z) == GH(z) == ~[G(s)H(s)] =· ~(~) = 2(1 - e-,r)z
' ' . . ' '. (~ s(s+ 5) (z-lXz-e-51)
Of interest in this case is the fact that we can no longer define the input
transfer function of this system by either C•(s)/R*(s) or C(z)/R(z). Since th-o~ir.
= 0.7869z
(..74)
r(t) is not directly sampled by a sampler, the sampled signal r*(t) sim
not exist · pY
tn~ (z - lXz - 0.606~)

,. substituting Eq. (4-74) into Eq. (4-60), we have the closed-loop transfer
To express the continuous-data output C(s) in terms of the inpu~ 1
,oW
1 .
substitute Eq. (4-67) into Eq. (4-62); the result is 111 function as
C(z) 0.7869z (A-75)
C(s) = G(s)R(s) - G(s)H(s) GR*(s) -R(z) = z2 - 0.8196z + 0.6065
1 + GH*(s) · (~
Thus, the characteristic equation of the system is
I 4-4-i 1he Characteristic EquaHon . z 2 - 0.8196z + 0.6065 = 0
(..76)

.
The characteristic equati~n plays ·an important role in the study of linear
,I systems. It can be defined from the_basis of the difference equation or the
transfer function. · · ·
Consider that a linear ti~e-in~ariant discrete-data system is described 4-4-2 causality and Physical Realizability
by the 11th-order difference equation,
A causal system is one in which its output does not precede the input. In
c(k + 11) + a.- 1c(k + 11- l) + a0 _ 2 c(k + n - 2) + ··· + a c(k + 1) + aoc(k) other words, a causal system produces an output at any time kT depending
= b.,r(k + m) + b.,_ 1r(k + in - I)+ .. . + b r(~ + 1) +1b r(k)
1
only on input values that have occurred for times up to and including kT. The
,,
j
.
0
(4-70) transfer function of a linear causal system is also said to be physically realizable,
, r" '~
where r(k) and c(k) denote the kth input and output, respectively. The dilferena: · since the system must be composed of physical components.
It is shown in Eq. (4-11) that, by expanding. the transfer function G(z) into
equa!ion can be so!ved by_ means of t~e z-transform. The input-output transfer
functJon C(z)/R(z) 1s obtamed by taking the z-transform on both sides of Eq. · · a power series in z - 1, the coefficients of the series represent the values of the
(4-70) and assuming zero initial conditions. Thus, . · . , . . weighting sequence of the system. The coefficients of the z-t term correspond
to the value of the weighting sequence g(kn at t = kT. Clearly, for the
.. : C (z)~
b.,z"'+b.,_1z"'-1+--•+bz+b
G(z) == - = ,z"+a._ 1z" - I +···+az+a
R(z)
I O
·•· (
A-71) discrete-data system to be causal, or physically realizable, the power-series
I · O .
~xpa_nsiim of G(z) must not contain any positive power in z. Any positive power th
The characteristic equation of the system is defined as the eq . ' b . ed · ~n z m the series simply indicates "prediction," or that the output precedes e
by equating
. the denominator
. of·G(z) to zero; that is,
. , . ua11on o tam
mput. Therefore.far the G(z) in Eq. (4-7 /) to be a physically realizable transfer
,,I
z" +
1
a _,z"- _ + .. , + a 1z + a0 == O
0 ( • 2) · function , n must be greater than or equal to m. When m = n. G(z) is referred to
; as a proper transfer function. This condition exists rarely in physical 11systemsthe
as
We shall show later that the roots of the characteristic cquar 4 .7
the absolute stab1hty
. . of the system. , · ton determine th e output responds to a jump discontinuity in the input. When > 111,

1!, 1 iransfer function is referred to as _strictly proper. ·

t31 1IANRI fUNC1l0NS. ll0CK DIAQRH.tl, SIGNAi. R.OW
•• J. , .. - ~--.. 11ew-...
Quite of'ten;in the analysis and design of discrete-data ·. . . 139
. . al . ed .
function of a digit process 1s express m terms of negat·
. syste"'
.. ,s, the, G{1) " C{,)
. .. . ive Pow'
I Ji ., . 'I'. ;.•) · •' .,· ·· ·'·· ·"'·, ···b"' +'b111~,z
·, "" 1·+ •·.• + b z:-.'I' ersor t
G(z) = o ,,.c, .· -
a. + a._,z -1 + .. . + a0 z-n , . .1-, .';:
.r
·,, ,,-1; .• ','. J ,l • •
-1 l'~"

., where n and m are positive integers. Notice that no matter Wh

(1
-: ' f~ ,
' .'c~ I' I .
'·;t.•,'_~

of mand n are, G(z) has· the ,1

same number of poles and zeros. In t~t. 1hc C(I)
.;l
is physically realizable if _i1c_~,oes not ,~ontain any factor z-t (k 1:
denominator, or a. ::/: 0. . ,. ·. ·· . 0) tQ •
(b)

Block diagram of. a closed-loop digital system. (b) Equivalent signal flow
4-S . THE SAMPLED SIGNAL R.OW GRAPH _.,.1,,1. (a)
. ,aP
,_h of the
.
system. . , '
.
.
I • • • •• I • " 1 t l

In the preceding section we demonstrated the algebraic method of ana) .

closed-loop discrete-data control -systems. _F or systems with multiple sani>;::
signal flow graph will have_ loops that are not broken by the sam I d
, and loops, block diagram algebra may become unwieldy. It is well known that
the determinant of the signal flow graph would not be equal 1: ers._:;
transfer functions of linear continuous-data systems can be determined fJQa
3, Take the pulse transform on both side_s of the cause-and-effect e;,~io°;u
signal flow graphs using Mason's gain formula (1], (2]. In this section
111 obtained ;~ Step _2- The purpose of this step is to convert all the system
extend the signal flow graph method to the analysis of discrete-data syslenis,
variables mto discrete-data form. For example, the pulse-transformed
For reference purposes, a review of the Mason's gain formula for signal flow
graphs is given in Appendix D. .. . . form of Eqs. (4-51) and (4-52) are Eqs. (4-53) and (4-57), respectively.
. 4. Draw a sampled signal flow graph using the equations with only discrete-data
Since most discrete-data controi systems contain analog as ~ell as discrete
. ... . vari@les obtained in Step 3. For Eqs. (4-53) and (4-57), the ~pied signal
signals, Mason's gain formula cannot be applied directly to the system's ori~oal
signal flow graph -or block diagram. The bottom line is that Mason's gain • , :·-.:, flow graph is drawn as shown in Fig. 4-8.
5. Once .the sampled signal flow graph is drawn, the trans/er function relation
formula can be applied to a signal flow graph with node variables that are
either all analog or all discrete but not a mixture of the two. Therefore, the first ,•~. ;· between any pair of input and output nodes, defined according to the signal
. step in applying the signal flow graph to discrete-data systems is to express the •.·:" flow graph terminology, can be determined by use of Mason's gai}lformula.
system's equations in terms of discrete-data variables only. The steps for the :; :s ·' In the present case, applying the gain formula to the sampled signal flow
. ' . sampled signal flow graph method for discrete-data systems are outlined 81 ., ( graph of Fig. 4-8 with C*(s) and E*(s) as output nodes, we get
"·', , follows. · · ·· · ' ..
C*(s) = G*(s) R•(s) ('"71)
l. With the system's block diagr~ as ~he ~;;;,ting p;in;, c~ns~ruct an equivaltlll 1 + GH•(s)
as
_signal flow graph for the system. This is also known the signal flow graph
of the system, since it is entirely equivalent to the block diagram represen~· · and
tion. As an illustration, Fig. 4-7 shows _the block diagram of the' system in ('"79) •
> , \f,..'
Fig. 4-5 and its equivalent signal flow graph. •·. , , , ,,. . 1 R•(s)
E*(s) =
· 2. Following the definitions of the input and output Dariahles of.the system g/ven 1-+- -··•·(s)
in the block diagram algebra in Sec. 4-4, write the cause-and-effect equations
.J•i
of the system from the equival~n{ signal fl.ow_ graph. For instance, for the

. .,_ .,.; .0
,1)¥ ,
system in Fig. 4-7 th~ eq~ations an: already written in Eqs. ( -51) and
··~ ~- 1~·. '\"
4 signal
.: (4-52). Note that Masons ga"! formula should he applied lo the original
-Glr(s)

.,. ,. flow graph in accordance _with. all the defined input and output nodes. The
. . aJ flow graph shown m Fig. 4-7(b) docs not have a loop in the sense G'(•) .
·' sr.he conventional definition, so th~t Eqs. (4-Sl) and !'4:-S2) are obtained I I ,oc-{I)
. !ith the trivial application of the gam formula. In complex •itllati~ns, the
- ~(I) 7
4-1. Sampled signal flow graph of the system shown in Fil, + .

_ __,.,,,,..~;:.. ~·..
..•. ;:':·::·- -,~-: ~--::----= I
 .., 1IAN5fllfUNCIIONS, aoacDIAQIAMS. ANO SIGNAL R.OW GRAPHs ,i7 ' •
· IHIIAMftml!Q,w.llOWatNH
1.a1
6. ThL compositt sig'!°I flow_ ~raph J5] is obtaJned by co,n .. . is more proficient with the application or the sampled signal
tJlld tht sampled signal flow gra?hs, according to the 'Ire ,}'"~g .When. on~hod Step 2 can be bypassed, and the writing or the input-output
ow graphs. 11,e transfer f. tunction .relationships betWee,,latio,,,be t'. , flo .;. graph mde. the taking or the
. , s an . ...Pulse transform .in Step 3 can be combined.
fl :....,.us-data outnuts are determmedfrom the co- . the i11n,_·,q ho..·
equauor_, ::"-.:._.- -- - - .. - . . . - -- . . .. - . . . ..
conl.,...., r . ." . .' "'Posue . ,,.,, , . . ,.
,,. .,,
by use of Mason's gain formula. In fact, all the out s,gl!Qlfio,,,""i1
8
system, •discrete and analog, can be determined from ih~t / riab1es •,

"'rii
0
flow graph. .· ,. ,• .
rnllosj~ •1
. I, , block diagram of a multiloop discrete-data control 5 t · .
eO( ) The equivalent signal flow graph of the system 1·syshem ,s_shown in Fig.
More specificall~, the composite signal flow graph is·formed· 4-i a• f s own m Fig 4-IO(b
· regarding the output o the samplers as inputs to the sy t h .· ).
the output nodes of samplers on the sampled signal flow graph ~y C01 BY utput variables of the system are identified as follows~ em, t e input and
branches to the same nodes on the equivalent signal flow graphWllh u~· thcl O •
the composite signal flow graph of the system in Fig. 4-7 is show : Fo~ . ul variables: r*(t), e*(t), c*(t) ·
Application of Mason's gain formula to the composite si;~"t'8.4-9.
in Fig. 4-9 yields the input-output transfer relations for all the
continuous-data outputs. Thus, from Fig. 4-9 we get
di; 80•
retc-111
:!put variables: e(t), c(t).
. , fhe reader should not be puzzl~ _by the fact that c*(t) is treated as an input
,
1
f the system.. The above defimt1ons are based on the signal flow a h
.· ~nninology, and the roles ~f c(t) and c*(t) ~hould be distinguished. Also, ifa:y
. G(s) R*(s) , ;:,, signal flow graph, any nonanput node vanable can be regarded as an output.
.. ... .qs).= 1 + GH*(s) • f fbus, the two outputs d~fined ~bove are onl~ the ones that are essential for

and ,.. -: ·· ., \) , , the·present system. If desired. we can also designate the output node variable
·'• . · r a (s) as an output.
I '
"l'
' . ·· - , , - .:•
_ . ., , G(s)H(s) ·R*(s) 0 The 1 cause-and-effect equation~ between the inputs and the ,outputs
•..i:t' -:·· ·1•E(s) =. R(s) - 1 + GH*(s)
:· '· - -- •., designated above are written by applying Mason's gain formula to the original
. 'i'', .· ·. signal flow graph of F~g. 4-IO(b). We can 1ake the pulse transform or these
The discrete-data output relations in Eqs. (4- 78) and (4-79) can also
· determined from the composite signal flow graph in Fig. 4-9 by using lhej'
a,~
equations the same tim.e. The results ar,e '
. · E*(s) = R*(s) - G1 GJ(s)E*(s) + G2 H*(s)C*(s) (U2}
. formula. • .• ,,., ,. · .. , ..;, ·,i ,,, . . =' ·
,,, , · · The sampled signal flow graph method outlined above can be applit.d C*(s) = G GJ(s)E*(s)- G2 H*(s)~*(s)
1
(U3}
linear multiloop multisampler systems, provided that the samplers ares I'
'?. -
nim1 and or the same sampling frequency. The method can also be appli~ C(.r)
the modifled z•transform :relations so that responses between the sam G1(s)
instants can be recovered.
·•' " ... ~ I ' .J,

T ::.
If ; , .•

. ' ! l' j

.~. ; ,r!';·' • (1)

·~. ~ J G-<,) ll1l.S/ ·' Gi(.s) ,• -ctr)
R-(1)0 I

E"(s) ,~
• oc•c,) 11 .• , 1,.I
0 • • /

G(s) I
R(s}O I
-cer>
E ( s } ~- - -~

...Jf(s) i '.\ ·-·. flgln.,a. ' . (b) -

u lignal ir:wo. (a) Blo~k diagram of a multiloop discrete-data control system. (b) Original
·--.
a...- Comn1>site signal flow graph of the digital system in F" ·
. . graph.

- - - ,..-, -~ ---....~~.:I::!::: II
 .
_..IUNCIIONl.1.0CK~NIJSIQNAI. ROW~
fA2 ,i
llll w.tPta, IIQ,w, ltOW91A111
-GiGi(I) 143

,-6 ,,
.,,, .
arnple we illustrate a discrete-data system th ' .
P(l}O i 2'!( • Jn 1'115 ::rnplers. The bl~ck diagram of the system is s:~ has_ a 1<>,<>P not broken
- bY tile_ steps are earned out. .. .., , . wn m Fig. 4-12(a). The
following . ._ . . .
Jte original signal flow ~raph of the system is drawn in Fi
1, ~put and the ~utp~t v~nabl~,s -~f the s~gnal flow graph are g. 4-_l2(b). The
R(r). • " . 'ut variables: r(t), e*(t) ·· " · ·· · ·
C(r) 111 ~put
variables: e(t), c(t).
oU .
l ·Notice that the signal flow graph has an analog loop with a 1 .
Flglft._H, Composite signal fl~w graph of the system in Fig. 4-10. _ G (s). The D(s) block with samplers at its input and out t _oopbgalm
of l . f d' . pu 1s a ock
\i diagrarn representation o a 1~ta1 controller that has discrete-data input
I
and output. The transfer function D(s)_ and the sampler at the output can
The sampled signal flow graph is drawn as the top part of Fig. 4-11 llS' be replaced by the ~ulse transfe! function D~(s1 as shown in Fig. 4-ll.
last two equations. The composite signal flow graph is obtained by diing .. 2, The foJlow!ng equation: are ~ntten from Fig. 4-12(a) using Mason's gain
branches with unity gains from E*(s) of the sampled signal flow graph ~I formula, with r(t) and e (t) as mp~ts and e(t) and c(t) as outputs, and taking
same node on the original signal flow graph and the same between C•(s; the piJlse transfonn at the sam~ llme:
. ·. _..., -·: · .. . . . _,.. '
Applying Mason's gain formula to the composite signal flow graph on·,
4-11, we get the following input-output transfer function relations:
.'J ;'. . ·E•(s) =( R(s)
. 1 + Gi{s)
)* - v•clGi(s)Gi(s))*E•(s)
'\l + G2(s)
(A-89)

,,.,~\ . . ' \
' . ~s),,;, ' ' 1 + G2 H*(s) R*(s) C*(s) = (R(s)Gi(s))* + v•clGi(s)Gi(s))* E*(s) (A-90) .
-~ . 1 + G1Gf(s) + G2 H*(s) (~ . , 1 + Gz(s) '\ 1 + Gz(s)
,
1 C*(s) = · G1G!(s) ·. R*(s) ("I . Notice that in this case, similar to the system in Fig. 4-6, the input signal
l + G1GJ(s) + G2H*(s) ·does not appear as an independent entity, since it is not sampled directly.

1i; , ·• / J··
.
5 G1(s)G 2(s)[l + G2 H*(s)] - Gi(s)H(s)G 1 G!(s) R*(s) (~
l + G1GJ(s) + G2 H*(s)
'

C( )

In determining E*(s) and C~(s) in Eqs. (4-84) and (4-85), respective_ly; only ~(s~

-the sampled signal flow graph portion of Fig. 4-11 is of concern, since 00
;_;_ ·, r·~r·
if ..-~(s)--t _.,.,,.
forward paths exist between R(s) and these two variables. In obtaining~
fi analog output C(s) in Eq. (4-86), there are two forward paths between R (~
E(s)
/ ·; ; fl
! and C(si and one of th~ ~as a nontouching loop with a loop gain of -G2W(4
Since no forward paths be between R(s) and C(s~ the analog input do~s nol
appear in Eq. (4-86).
;;, If ncccssary0 ~ ..z-t~a~sfo~ ~uiva~.nts of Eqs: !4-84) and (4-85) are writlell
(1)
i: as ~ -·-- · . •· (t_i'! ,,. ' .·,--- ~- ·>~

6
'f ,. • '· I + G2H(zf
, £(: ) = 1, + G1 G2(z) + G ii(;jR(z) (A-87)
t 2
G1 Gz(z) ·
0
R l E~ ~
Yi lC •C
I C(z) =
10,
. R(z)
.. _ ., J + G1 Gz(z) + G2H(z) . ,.,.~ %
,.,_,,,
:1 •. r,,
' ' -1
(b)
Figwu. 12· (a) Block diagram of a discrete-data control system- fb) The signal flow graph.

. --~:;;....:;--=- -------;
 tU llAIISRI MICIK)HS, llOCX DIAGRAMS, AND SIGNAi. R.OW Gt4.PHs Ill M00IIID :•IIAHlffl 1VNC110N 1'5

. the gain formula to the composite signal flow graph in Fig.

· AP?1Y1~s) as the o~tput node, we have - · · -
4.I 3 11s1n 8 (G 1(s)Gi(s))v• )( R(s) )•
;, :
I
• R(s)Gz(s) 1 + G2(s) (s 'i"+GJsi (4-94)
l C(s)::::: t+Gz{s) + . 1 + v•(/G 1(s)G 2(s))•
. .. , , . '\ 1 + G2(s) .

1 \ ' G2 ' The composite signal flow gra~h in Fig. 4-13 also includes information
Ro ,- E°'Z... 1
Y1 ·, /( ode variables f1 and ff, tn case these are of interest.
on t he n .
-1

Flgllt .;13. Composi_

1c signal _flow ,grap~ of the digital system in Fig. 4-J2, led signal flow graph is not the only signal flow graph method
1
'
/·
•
, \·
. •
_.r ! , t ,

"t
•

---n.
' I
1
.'fhC
bl Samp .
for discrete-data systems. The direct .
signal-ffow graph [3] .s an
availa te method which allows the evaluation of the input-output transfer
l ·The composite sigrial flow graph ef the system is drawn as sho'IIQ · atter?a : of discrete-data systems by direct inspection. The method depends on
4-13. . . ·: . . • .1 retauot~rely different set of terminologies and definitions than those of Mason's
4. Applying Mason's gain_ formula to the c?mposite signal flow Sfaphdl an en 1
l · 4-13, the transfer relallon between the 1~put and the analog outpa
•gnat flow graph.
SI . ___________ • -
•

J
and the discrete output c•(s) arc determmed as follows. FUNCTION
i The composite signal flow graph has a total of three inputs. Theselt DIAED z,fRANSFER

,_
.
.I
i/)
l (~ ) •
I + Gi(s) ,
(R(s)Gi(s))*
I + Gi(s)
R(s) 4-6_
t, s, 3-8-2 w< intrndu"'1 tho modill<d ,-t-fonn m,thod th, '°"riptio•
' ,Ith<0,,,po,so t,,two,, the s,mpling ;,,,ran~ ofa ,..,plod ,ig,,,11• thl, ,e,ti<,n
ro,
I
.l
' . '!
'1
The analog input R(s) does not affect the output C*(s), and in fac~ al: we •PP'Y the method to di=te-d>I• ,y,ton• .,d dofine the mod/fed ,-t<mu/"
the original signal flow graph has an analog loop that is not affected~
samplers, that part of the flow graph will not contribute to the Junction. ·:
· Consider that we are interested in determining ·
the response be~ween the
signal flow graph. Thus, for the output C*(s), we have ' sampling instants of the system shown in Fig. 4-14(a). First, we insert a fictitious
time delay /J,. T, 0 < /J,. < 1, at the output of the system. and the delayed output
I ·.
C*(s)=(~)• +- \I~}
D*( jG 1(s)Gi(s)\•
( R(s)
.
)*
I+ Gi(s) . I+ D • ( i ~ \ • J + Gi(s)
\l+Gi(s)j · l __ _x._:j~
or, in the z-transfonn fonn, - ,- T C(z)

I
G1(sJGz(s)}z) C(z)

C(z)= ( - RGz(s)
I + Gz(s)
-- D(z{ --:--=:
z+_ J)• I ~-2. .
r..~
:!:S(s); . ,(_
l + D(zl~\, I + Gi(s)
_ z)
R(s) J ('-II
t(I) T
(a)
,_.x ..:..<~
.
where . . ., .
'\1 + Gi(s)]Z) 1
T C(z,111)

I
t(rl T i.-..J---c(-,..t,7)-

(~\,z) = ~(~) FictitiOUS

I + Gz(s)/ I + G (s) (A-9ll TimeDelaY
2 (b)
and the other z-trahsform quantiJies are defined silllilarly.
::,-:; ..., l•I A di~""'"' •>'"m (bl A o=ood•• •Y""" with • O«i<i""' Om<
 1IANSfH fUNC1l0IIS, llOCK ~ S . NIO SIGNAL Fl.OW GRAPHS

c(t - t,. n is sampled by a fictitious sampler, as shown in Fig. 4-l4(b

As Sho..,,
in Eq. (3-194~ the modified z-transform of the output c(r) is defin ed). as
-1Hl ~ l - ~ Mtc:1loN
The modified z-transform of c(r) is written

C(z, m) = G(z, m)E(z)

m Il
I
"' (4-102)
fl .,[c(t)] = C(z, m) = z- L c(kT + mnzt
1
The z- transform of e*(r) is
t=O
z
The modified z-transform in the last equation can be used t (~ E(z) = -z-1 (4-103)
mination of the input-output transfer relation of the system shor the_dct~
4-14. Substituting C(s) = 0s)E*(s) i,n Eq. (3-201), v:e get own 1n F~ an d the modified z-transform of G(s) is already determined in Eq. (3-203),
1 . "' . . . . e- ... r
C(z, m) = - L G(s +jnwJE*(s +jnw,)e-(1-.,)(,+ i•o,,)TI G(z, m) = z -e • --=--r (4-104)
T .s-oo . . • ~•" (4-!ii
Since E*(s + jnw,) = E*(s), the last equatio!1 bec?mes ThUS,
C(z, m) =' E(z)-
1 L"' G(s + jnw ')e-(1-m)(i+ i•w,)TI e-... r _z_
. T,.=-oo s .. : i=eT, C(z, m) = z _ e-•T z _ 1 (4-105)

= E(z)G(z, m) ( which can be expanded into an infinite series of z-1,

• . 4i1)
In the last equation G(z, m) denotes the modified ,z-transform of G(s) and is C(z,m) = e-'""Tz-1 + e- ... r(l + e-•T)z-2 + e-... r(l + e-•T + e-2a1)z-1 +· ..
expressed as
(4-106)
G(z,m) = -I "L."' G(s +1·nw,)e-O-•tt•+j,t.,,)rl . . ('-91) It can be shown that the coefficient of z-t is
T ==•''
•= -ao
· Therefore, the modified z-iransform of a system with the transfer funciion
c[(k +m- l)T] = e--r(l + e-•T + e- 2• 1 + ... + e-u1)
'G(s) is defined in exactly the same way as that of a signal. e--r(l _ e-ur)
(4-107)
The conclusion of the result in Eq. (4-97) is that for the system in Fig. 4-1~ . 1 -e -•T
which has a sampler at the input only, the modified z-transform of the output
c(r) is the product of the modified z-transform of the system transfer function which gives the output response for the time duration kT S t < (k + t)T for
G(s) and the z-transform ,or the input E(s). Alternativ_ely, k <!'. 0 and when m is varied between zero and one.
The z-transform of the output c(r) is written
!l .,[C(s)] = C(z, m) = !l .,[G(s)E*(s)] = G(z, m)E(z) (~
. z
Further, the modified z-transform of a sampled signal e*(t) is just the z-transform C(z) = G(z)E(z) = - - _z_
of the signal, or (l ·'
, Z - e-•T Z - 1

· !l.,[E*(s)] = E(z) (4-100)

I = I + (1 + e-•riz- 1 + (1 + e-•T + e-Z.1)z- 2 + ··· (4-108)

Comparing Eq. (4-108) with Eq. (4-105) we see that

xample-4-6 C(z, m)l.,=o = z- 1C(z) (4-109)

Consider that the system shown in Fig. 4-14 has the transfer function which is derived earlier in Eq. (3-198~ Since the input e(t) and the impulse
response of G(s) both have jump discontinuities at t = 0, C(z, I) -t, C(z~
G() I I
s = --
1,1-(111)
'
:igure 4-15 shows the response of c(t) and the sampled values at I".' kT.
s+a ~otice that the z-transform result is misleading, since the output has Jump
discontinuities at the sampling instants kT. By varying the value of m~tween
where a 1s · a constant. Th e input
· · a unit-step
to the system 1s · function.
·fled
t(t) = u,(t). The output of the system is to be ev:iluated by the mod 1
zero and one, all the data between the sampling instants can be spcctfied by
z-transform method. .
 1"
... _..fUNC110NS, 1&.0CK~ ,w,·SIQNAL p.<1#1GWHS

r-t1111Sform
-solution · nizATE Dl~SC~R~ET:E-~D~AT~A~S~YS~TE=M:.:::S~----------
MUI.IIIAJI DISCllrfl.oMA ffllEMs
1•9

A-7 t . 4 ~
..-·he_ dA•i•srcrete-data
- · systems w considered in th e preceding . ·-
((lT,111)
1
iform samp1mg rate. hen the data are d' . I sections all have a
un d fd ' . I igtta' the rates Of th .
the output ata o _ig1ta processors and controllers thr e _mput and
identical and um form. For sampled-data syst · houghout the system are
:he11 system a!I have the sa~e uniform sampling :a7:.' t e samplers throughout
In practice, the bandwidths of various portions f d'
rnaY be far apart. Thus, it is more appropriate to samp~ a Iisc~e_te-da~a sy_stem
111•1
t a low sampling rate, while a signal with high-freq ea s ow Yvarying signal
0 T 2T 3T 4T 5T 6T 1T
a · I h' h uency content should be
RgUre 1-15. Output response of the discrete-data system in Example 4-5. S
ampled . at a re Iat1ve ·y 1gI rate. • When a discrete-dat a system has different .
, samphng rates at vanous . 4ocallons,
17 ·1 the system is referred t·o as a multirate .
iscrete-data system. F 1gure · 1 lustrates the block diagram ofa t • 1d' . -
d issile autopt·1ot contro I system. Due to the difference between thyp1ca 1g1tal
dynam1cs .
m · · d • e
in the pos1t1on an the rate feedback loops, the samplers in the system h
two sampling periods, ~1 ~nd : 2 - Note that the two data holds in the syst:
Oosed-Loop Systems
have different char~ctenst1cs, smce they hold the sampled data for periods of
The modified z-transform method can be applied to' closed-loop disc~ete-d r 1 and T 2 , respectively. ,· . ,
systems without complications. Figure 4-16 shows a closed-loop system Wit~ta . Due to the variety of sampling schemes, the analysis and design of multirate
fictitious time delay positioned at the system output. The following cause-anda discrete-data systems are usually quite complex. It would be helpful first to
effect equations are written from the block diagram by treating the output; , . establish the basic types of multirate sampling. Figure 4-18 shows three basic
t~e sampler, e*(t), as an input, in addition lo the inpu~ r(t), and the output
n: ·
types of open-loop discrete-data systems with multirate sampling. The system

,'
'
signals of the system are e(t) and c(t -'- !J. shown in Fig. 4-1 S(a) has a slow-rate-sampler at the input and a fast-rate sampler
C(z, m) = G(z, m)E(z) · (<l-11~
at the output. This type of system is referred to as the slowfast multirate-sampled .
system. Figure 4-IS(b) shows a system in which the sampler at the input has a

I 'J
E(z) ':" R(z) - GH(z)E(z) (<l-111) faster rate than that of the output sampler, and the system is referred to as a
Solving for E(z) in terms of R(z) and substituting the result in Eq_- (4-110\ ~-:'fast-slow multirate-sampled system. A digital control system whose controllers
. have different input and output data rates can be modeled by the block diagram
we get

I
·, in Fig. 4-IS(c). In general, the sampling periods T1 and T2 may not be related

I'
. G(z, m) • R(;) (4-11~ 1 by an ·integral factor. For analytical purposes, the mathematics are greatly
C(z, m) = 1 + GH(z) · simplified if we assume that T1 and T2 are integrally related; that is, with
which is the modified z-transform of the output c(t).
reference to Fig. 4-IS(a), T1 = NT 2 , where N is a positive integer greater than f ii1.
:. fl
Although the modified z-transform was originally devised for the purpose one. ,I
of describing signals between sampling instants, we shall show later that the
method is also useful for the analysis of multirate sampled-data systeips.

/,
Hold

l
,--X
1 T
..:~<~in
C(z.m) •
I

c(t--.il)

Figure 1-17. A digital missile autopilot control system with multirate samplers.
Figure '-16. A closed-loop di5".rete-data system with a fictitious time delay.
------ ·l,
MUI.IIIAJI DIICIIIIOi\TA MllMI m
I
~~---
----•OCI----NIJll9NAlFlOWGINHS
·
'ii
flt
50 ,..
-eT•/N

(a} '
--------------
-~ r;;-y_ OutpUt.,
et.T,/N
Ti~
(b)
, Ti
----
. Output., e<N-l)T,/N

Ti~ - T2 ._ . Slow-fast multirate-sarnpled system with the fast-rate sampler decomposed

(C)
20
~-N slow-rate samplers with time-advance and time-delay units.
f9llt ••a.
Open-loop multirale discrete-data systems. (a) Slow-fast sam 1 ~w I

(T > TJ. (b) Fast-slow sampled system (T, >. T2), (c) All-digital system. p ed
1
· and C(z)N is defined in the same way. As in Eq. (H87), G(z)N can be found
;," i )
__ from G(z) through the following transformation: . .. .
' ('-115l
A-7-1 SloW-fGlt Muntrat.Sampled Systems G(z)N = G(z)l,=,''"'· r~rtN
Consider the slow'-fast sampled system shown in Fig. 4-18(a), with r, •
NT , where N is a positive integer greater than one. Several methods Ill .' z.'' Sampler-Decomposition Method
available for the analytical study of tliis type of system. Each of these methods A versatile method of analyzing the slow-fast multirate system of Fig. 4-lS(a)
is useful for a particular situation, and the methods are described as follows. is to decompose the fast-rate sampler into N parallel-connected slow-rate
1
! i I '"'l
' ~, 1 • f,
· . >', samplers with time-delay and time-advance units, as shown in Fig. 4-20. Since
1. Ficlitioas-Sampler Method ,: now the samplers are all at the same sampling rate, the ordinary z-transform
· method ofanalysis may be applied. The z-transform of the output of the system
Since the input sampler samples at a lower rate, or a larger sampling perioo
T,, we caq insert a fictitious fast-rate sampler with the sampling perioo ,, c*(t)N is written N- 1 .
. T2 • T1/N at the input, as shown in Fig. 4-19. Since the input to the fictitious C(z)N =L z~k/N~[em/NG(s)]R(z) ('-116)
sampler, ,.(t), is a pulse train with sampling period T 1 , the output of the i=O •
fictitious fast sampler is still r•(t). The situation is very similar to the submultipk
sampling method described in Sec. 3-8-l. The input-output transfer relation ci : j' . where, the z-transforms of the time delays at the output are written as
the system shown in Fig. 4-! 9 is written • . ~[e-kT•IH] = z·l/N ('-117)

C(z)N = G(z)NR(z) (4-11~

= l, 2, ..'° .Eq.
for kComparing , N (4-ll6)with
- l. : - .. Eq. (4-113~ we seet_~at G(z)Ncan be written as
where G(z)iv is as defined in Eq. (3-186), (,6-118)
N·I
00
G(z)N = L z•ttN~[ITIIHG(s)]
G(z)N = L g(,kT/N)z·ktN ,. (4-H~ t=O '
k• O The z-transform of e1rs/HG(s) can be determined from the modified ,z-
transform of G(s). By definition of the z-transform. ·
,: ,~[emtN~s)]~t~[(n+Mr}-•
. . I
~ ~ ~ « , , . -x·~ <•119>
, T R(i) TIN . C(z)N

.. Aclllious
Sampler ..·
•· . -- - ... -- s;,,._ k/N ii ;iway•"" th"' .,ity'in tho,-,,
oppliati•• ·~pori"'
e nght-hand side of the last equation with the defining cquauon or the
figure '-f9. Slow-fast multirale-sampled system with a fictitious fast-rate sampler. th
 111fOON1,-.0Cl~NGSl8NAL FtOWGINHS .d ; :. ' ~ (',
. Mll111AJ1 ~ - - 163
...... 1-uansf~rm.in Eq. _(3-19~), we ~av~Jh_e following rel
ar
• .•· - 1 · · 4 ~•(t)z and f•(t) of the Multirate-Sampled System in
I ; . ~r,m!NG(s)] ;;[zG(z,m)].I IOnship, ,_.. 4.1 . r · ,, · · ·
· · ,I: ·· · · . m•l/N ' ~ 8,nple ,
• 1__ ........ ,,. I
·,-- ··-- --- -•··• I, •r· c•(t)~ · .. · i c•(t)
or
~[eiT~NG(s)] = iG(i, k/N) ,• (
I '< i
. •·, ~ o•.·d
pie pen . .. ·, values
sr-· ·: Sample Periods · ' Values
.; ·•· • • I ' 11 • 4 ..._. , __ 1

. . f

Substituting Eq. (4-121) into Eg. t4-118), we get ~ - - · . . 0.000000 · ; ,,r :

'• • i' i,•••· 0 •Ii·• 1·. , 0 '.! • · : - , ,,, 0,000000
N-1 • --· -• - - , ;I . ., 0.393000 ..... ; - ; • 1 , . l , • i . 0,632000
\..~-: G(z)N=
' . ....
L t•-•tNG(z~' k/N)
l•O . ••·· -·
• ~-
1 1' "
;· r

•
J ('
2
I I 0,631551 , . , . ;,,_ . 2 '· . '
.. . , · 1.496576
.
.. " . ' . "'> : ·: ·•:.: : ( " . ! . ": •. '• 11 :1 "... ; C., .
,
• ,. I
I \ ,
1. 3 ,', r , :
'
11
1.169351 . . 3 I ' •• , • 2,446740
4 " . ' . · ' 3.428401
1~
·, •]".-...' .·:•·_''. '4_ / · ' 1.4957,96
I '! . ' ; \ '

Therefore. the qutput of the system in Fig. 4-20 is_written · · . ·:::,, ,.t'
·'
2.086948 5 4.421651
N-1 . '' 5
2.445778
C(z)N = R(z) L z' -lfNG(z, k/N) 6
3.056587 i • 7 ;
6 5.419168 .
6.418253
l•O .7
3.427348 8 7.417916
..,. ,·.· ,. . ; .> --:-, ':J n! :'!.t !,:·.,.; •·./ 8
1 4.045400 9 . i. _ 8.4)7791 . -
1' )i:
:' 9
,•of'.,:' 10 4.420558 . .10 . 9.417746
.5.041279 . 11 (, 10.417728
1111111N ..7 " :: I 11 ·. 5.418056 .. O\ 12 ,. ) _ ; ~ ; 11.417722
12
I
. i'
Consider that the transfer function of the slow-fast· multirate-sampled s 6.039761 13 ' ,, -. ·., 12417719 ''.
in f'ig. 4-18(a) is .. · · · 13 ·
6.417135 · · 14 · - - · - 13.417717
14 ·
• -f<-~ - ~:~),:'!·j:t.;A •,·.~ --· ..Y. .!c ~;-::\v:JL 1 '!·~~ 'q f! ~:.-:"!, 'l L.ci-; , 'A,. 7.039202 . 15 14.417716
1
15
· · .. _· . _ s(s + 1r
G(s)=- .i; 7.416796 16 15.417716 .
·-
· ·' ·, r" " , . 16
8.038996 . · , ! ,. ' 17 .. ,, '~- 16.417715
,• ;._1;, . ,-., . . 17
•. 18 · 8.416671 18 .' 17.417713

-
The input of the system is a unit-step function u.(t). The sampling :,' :, •I; 'J,)l 19 :'. , :. ·_ 19 ' ' . 18.417711
· " ire T1 •ls and T2 = 1/2s. , ;,:: •. ·; ,_., . 9.038919 j- · : ·t i' ,"f' H..

.... _:· ! .' 20 ·· 9.416624

The modified z-transfonn of G(s) for the sampling period of 1 s is , 21 10.038890
\· , '; ·f'•

1 . e-• 10.416605 .
= z - 1 - z ----,.
J , . 22
"\ G(z,/m)
.
e 11.038877
23
--~,.
.

, 11.416596 ,,' ,! i •
Using Eq. (4-123), the z-transform of the outpu't wiih the fast-rate sampling'
,:: . ,,«· ' : :, -- ;~
12.038872 ' I

I z•-•12[_Z· -_1_1 - Z-e-•12 26 12416593

'./ I - ' ! ,',. •

qz)z:;_, ]-z- .27 13.038869

,.1:., . . l•O . 0.368 Z - 1 28 . 13.416591
: .. ,.: ' ' · , ... i · -- -1
The last equation is expanded tJ iud 29 14.038868
.
··1
' I 30 1.4.416591
. . :·. .: , .,. ·, .,_[z + z•t2 . -z + e- i12z112] z -~. f:: , IJ J<', : • 31
' 15.038868
:,t

I
. .
qz)z= - - - - - - - -
z - 1 ·, z - 0.368 z - 1
- ,._121JI I·· : ·· /i 1\\ 32 · _15.416590 · 1,
·~
: 1•,. )

I
IJ'·

33 16.038866
=z112, then zi =z; ·the. last equation is written
' . ' N •, :· .' ; • I >' > ,. · • , )' • _- • ;~ • , .. , · '

·' · · "· Letting z2 34 16.416588

. , \ •.

35 17.038864
' q) -. Z2z + Z2 __z2z + 0.607z2] ·Z22 36 17.416586 I
z2 - [ --:::---- - :,r
. . , Zz ,_ 1 · · zj - 0.368 zj - 1 37 18.038862
, , ..,. • a : ·,i 1.: - -. ·,: f =-[0.393zf + 0.632zi + 0.239.z ] .A_ . 38 18.416584 · ,, ! ·.;,,:

·• :.'}·,_, ,,:•, .,, .. •.:. ,: . . z1"":"1.368zi+0.368 .,2 zi-1 (f42'1 39 . 19.038860 . ' !!I
1
!I!!!!
 Mll11AJ1 ~ - - -
·-NNCflONl,ILOCl~,_SIGNM,R.OWGWHI 155
tM ) ct(t)N ~-ct(~
Cancclinoe the common ,factor in the last equation, .
C(z)i is reduced to ~ 'N C(z)N T C(z)
0J93z2 ' " ] Zi . fl(Z)// ' Flculious
Sampler ·
-~
··. • ·, • C(z)a • .....
[ (z 2 - 1)(z 2 - 0.607) - z22 - I
- •· '{fl . ..
.-: . \
'
A-21, Fast-slow
. · multirate-sampled
· . · "· system with a .fic1t1tous sampler.
f!QII'• ''
The expression inside the brackets, in the last equation is-rec 0801.
G(z)z which can be obtained from G(z) using Eq. (4-11_5). , le\l 10
The list equation is expanded into a power series in z by 2 1
1 · . . us-Sampler Method
using the computer program INVZ of the DCSP software pactng dilisi._ 1 ficUUO .
response of t'(t) at the sampling instants of kT2 is tabulated in T~e, allil·i · hod of analyzin~ ~he fast-slow multirate-sa_mpled sy~tem sh~wn in Fig.
le-~ to 39. The reader can check that if the output sampler in th ble one me~ insert a fict1t1ous fast-rate sampler with samphng pcnod T/N in
8 4.1 ~
1

F'l8, 4-JS(a) were with

. the sampling period, T 1 = 1 s, then the z-tran Yst"-
e 81r
4-tS(b) isthto low-rate sampler at the output. The result is shown in Fig. 4-21.
front of e : ut of the system is not affected by the ficti'tious fast-rate sampler,
the system output 1s • . ,..· -~-~ . 0111i•i
Since the out p s in Figs. 4-18(b) and 4-21 are the same from the analytical
0
C(;) = [ · 0.632z : the dp
tw omt.
. syS Fem
ro m Fig. 4-21 the z-transform of the output of the fictitious
stan . - .
. -: z2 - l.368~ -f- 0.368 _fz=l . (4'I sampler 1s
The ~ponse of c(t) a_t the sa~pling instants of kT 1 is obtained b e C(z)N = G(z)NR(z)N (~33)
C(z) into a power senes of z 1, and the results are tabulated i { xpalld~1
I *• 0 10 19. Notio: thatlhe values of <(kT,) agree with e,eryn here G(z)N is defined in Eqs. (4-114) and (4-115), and R(z) is similarly

t
;~• <10 ;efined. Once C(z)N is determined, which corresponds to the z-t;nsform of the
t'(kT2~ starting from k = O.

-
· ot er value
fast-rate sampled ou:J'ut c*(tJN, ~he output of the fictitious slow-rate sampler
c*(t) is obtained by extracting the samples at t = kT, k = 0, 1, 2, ... , from
c*(t)N, or obtaining c(kn from the set t'(kT/N). Then, C(z) is simply the
• '1. ' 1
z-transform of c*(t). - - . .
3. Infinite-Series_ Repr~tation We presenl three different_~e!hods of extracting C(z) from C(z)N.
The input-output transf~r function relation in Eq. (4-113) for the slow.fail
sampled system can be written as a pulse transform relation in the Lapiaii (a) The Frequency-Domain Method
' domain; that is, . · . ,, .
First, we write
C*(s)N = G*(s)~R*(s) (4-131) · 1(1){ 2,rn]
• .. 'r
C*Cs) = - I s +i - (•13')

Substituting T/N for T in the infinite-series representation of G*(s), -as i r T

Eq. (2-53) and assuming that g(0) = 0, G*(s)N is written as
. (,:
Letting n = pN + k, where -oo :s; p :s; oo, and k = 0, 1,2, ... , N-1, the last
equation becomes .
,, .
) i
N 00 • [ . •·- 2,rnN]
G*(s)N =- L' G s + j - - (4-1~
1 21tpN 2,rk] (•135}
T •= - oo I L . s +j T +j T
Cl) {.
T C*(s) = -
TpN+la-CI) .
Thus, C*(s)N can be expressed in the sam~ ~a/ These pulse transform relati~
are useful for frequency-domain analysis. Another application of Eq. (4-13 2)• which is then written as
demonstrated in the following development.
C*(s)=-
1L --
TN
N- 1
L
00
{ 2npN
s+j-+i-
2d] <•136)

, / •· ·
T l•O N T ,. -oo T T
In view of the relation for G*(s)N in Eq. (4-132i the last equation is
~7-2 fast.Slow Mufflrat•Sampled svit~~
The discrete-data system with the fast-slow sampling shown in f ig.
can be analyzed by several alternate methods. The~e methods are descri
written
' I N-1
Ct{s) =- L
[
s+ir
2.kl (•'37)

the (ollowing. ·· N l•O

., --flNCIIONl,&OCl__.,,_,lllfWROW_,.. -111A11~-- 117
Now substituting Eq. (4;131) in the last equation, we get _. fhC z-tra nsform of the output
, or "'the system in Fig. 4-21 is
· C(z) • L c(nT)z-•
~s) • -
N ••O
t G* s +J-.
I· N- 1
T
R* s +/1tk] ·
2nk] [ ·
[

T
·· ,.,. 11 '9•0
(4-1..,
.
'N . , ' , ''
Since le ii an integer, R*(s + j2nk/1), = R*(s); '. Eq. (4-138) becomes
(41
t,ctttn8
.
, · •· ., 1
C(z) = , L
C(zN)t;,N-t dzNz-•
f ··
ri == k/f'(and substituting Eq. (4-143) into Eq. (4-145), we get

C*(s) = G*(s)R*(s) ' • a O 7tJ r ,! .

f
. •
. i, , :: (4-i~
or ·1 ,{_ = -i C(zNh:. Nz- 1 -dzN
(4-146)
C(z) = G{z)R(z) . ,. 21tj r ••O ZN
• ,• ' I ' (41
Equation (~137) gives the slow-sampled signal in terms of the fa t . ·st equation; is simplified by writing .the infinite sen·ca · l
fc
h 13 ., macosedfonn.
lignal. Using the complex translation th~rem of ,the z-transform in ; ·sail\~
thcz-domaincquivalcntofEq.(4-137)18 ·, ! ,,: ,:,· ;.
'.', 1 N-, ·
, . , q.(3-J~
.L, :•;. •_p,··. '.Lf .: :i1' i-.'
. we have the result ,) .
1
. : ,,;
C(z) ·=2-' Cz
7tJ r
. ,
1 .. d
( N) } N -·I - ZN ,
• - ZNZ ZN ·
·f ·. ,
(4-1•7)
C(z) =- L C[zNe'2Kt/N3N ·1- i ~--: r.J
(4-1411
N t•O The contour integral of Eq. (4-147) is carried out along the path r in the
Thus, given C(z)", the z-transform of the fast-sampled fuhction, C(z) -plane. Assuming that the contour r encloses all the poles of C(z )z-1 using
the residues theorem, Eq. (4-147) ts
. wntten
. ..
0 ' ZN N N '
o ~ by replacing z" by f"'_ 2 1
_" N in C(z)N ,and substituting it : ~ · -
summabon of Eq. (4-141). . , . 1, 1 , _, · -1
• ·1 _'I
;·- '' _C(z) = r. restdu~ ~f C(zN) 1 - zzz-1 at ~he poles of C(zN)z;
• ZN 1
(4-1.a)
(b) 71it Time-Domain Method
It should be pointed out that in .the last equati~n, although zZ = z, the terms
Another more comprehensible method offjnding C(z) C(z)~-is to make fr~m z" and z - 1 in the denominator should not cancel each other.
use of the equivalent time response of c(kT) and c(kT/N). The method is outlined N • ·., .. .·
as follows.
l. Take the inverse z-transform of C(z)N to give c(kT/N).
2. Replace T/N by Tin c(kT/N) to give c(kT). bafflple 4-8
3. Take the :-transform of c(kT) to get C(z). Consider that the ·transfer function G(s) of the system in Fig. 4-21 is as given
in Eq. (4-124), and T = 1 s, N = 2. The input ,{t) is a unit-step function. The
(c) 11ie Residues Method z-transform of the output of the fast sampler is written as C(z)N = G(z)"R(z)".
Thus,
1be z-traiisform of the outpu~ of the fast-rate sampler in Fig. 4-21 is written
0.393z
= ___ __:;-- -
] , ;2
. ., C( )
Z 2 2 (•1~
C(z)N = L c(kT/N)z-•tN (4-112) . [ (Z2 - l)(z2 - 0.607) Z2 - 1
k=O
We shall use the three methods de~ribed .above to find the z-transfonn of
Using the inversion formula, the inverse z-transform ~f C(z)N is the output C(z). The z-transform of the output ofthe slow sampler c-<t) is
) obtained by substituting the last equation in Eq. (4-141~,We have •
, \ • · c(kT/N)=-!__\!q;N)zt'.. 1dzN (4-113)
2nJ Jr l I
· where C(z) = - .L C(z e,i2d/2)
', , 2 k ;0 2 2 ; ' I , l ' . " r
<•150)
C(z") = C(z)N (A-1 ..)
.\.,

·L' ''
'

·, _ l[ 0.393zi ,- . 0.393:~ :]
with z 11"
'
replacerl by zN. - 2 (z 2
-
2
2
1) 2(z - 0.607) + (-z2 -1} (-z;:-0,607)

1111111
 IINtllll fUNCIIONS, llOCIC l)IAGIAMS. Al«J SIGNAL A.OW QWHS MlllllAJl ~IYIIIMI
51 119
After simplification and substituting z = zt the last equation 1cads t0 !)le A-2 C•(t) 2 and c•(t) of the Multiratc-Sampled System in
TO le 4-8
1.0246z + 0.2386 ] sxarnP

--
z
Crj= - ---- --
( [ z2 - 1.368z + 0.368 (z - 1)2 c•(t)z c*(t)
Using the time-domain method, we start again with Eq (4 149 (4-,~ Values ·

'
sarnple Periods Sample Periods Values
out the partial-fraction expansion, C(zh is written ·· • ). Carri•
0 0.000000 0 0.000000
_ z2. . • 1.5445z 2 ' 1.5445z
'

0.393000
Z
C( ) 2 ----+ 2 1 1 1.024600
(z2 - 1)2 Z2 - 1 Z2 - 0.607 (41~ 2 1.024551 ·- 2 · 2664853
1.800902 .3.
3 4.531666
The inverse z-transform of the last equation is . 2.665147 , 4 . 6.481853
4
c(kT/2) = 2(kT/2) - 1.5445.+ 1.5445e-Ht2
J L
s 3.582742 s 8.462722
6 4.532722 6 10.454881
for k = 0, l, 2, .. .. The equivalent c(kn is obtained from the la t (4-1~ 7 5.502357 7 11451195
by replacing T/2 by r. Thus, ., s equa~ 8 6.483922 8 14.449038
9 7.472728 9 16.447441
c(kn = 2kT- 1.5445 + l.5445e - kT 8.465928 10 18.446051
(4-1~ 10
11 9.461794 11 20.444735
fork= 0, l, 2, .. .. The z-transform, of c(kn is simply .' 1; ,'
12 22443445
12 10.459277
· 2z l.5445z' ·. ·l.5445z (l.0246z + 0.2386)
)~
11.457741 . 13 24.442163
C(z) = - -2 - - - +~-- =-~ . -c---.:..:. ('-151)
13
12.456799 14 26.440884
(z - 1) . z - l z - 0.368 (z - 1)2(z - 0.368) 14
13.456215 15 28.439604
IS
which is the correct result. · , 14.455849 16 . 30.438324
16
The last method to be illustrated is the residues method of Eq. (4-148) 15.455612 17 31437046
17 ·. 34.435768
Substituting Eq. (4-149) into Eq. (4-148), we get . 18 16.455454 18
J ,, 17.455341 19 36.434490
1 19
-~ . f • 0J93zj - Z2 20 . 18.455256
z - i.. residues o 2
C() -- • (4156)
(z 2 - l) 2(z2 - 0.607) 1 - z22 z- 1 21 19.455183
22 20:455116
at the poles z; = l and z2 = 0.607. These residues are evaluat~d as follows. 23 21.455050
24 22454979

.1: •
Residue at : 2 = 0.607
· 0.393z 2 I·
(z 2 - 1)2(1- zjz- 1) ,,=o.607
1.5445z
=z - 0.368
l•1sn
25
26
27
23.454903
24.454821
25.454733
28 26.454639
Residue at : 2 -= I 29 27.454538
30 28.454430
[· . 0.393z2 ] -1.5445z + 3.5445z
2
(4-l51) 31 . . 29.454315
dz 2 (z 2 -0.607)(1-z~z- 1) ,,=t (z - 1) 2 32 30.454193
33 31.454063
• .. . Thus, 32.453926 .
2 34
. l.544Sz . l.544Sz - 3.S44Sz (l.0246z + 0.2386)z . (4-1591 35 33.453777
C(z) =z- --0.-36-8 - --(z---1-)2- - = -'-(z---1)72(-z-- --:0--:.3:-:-6:::--8) 36 34.453617
37 35.453445 .·
which is the expected result. . , · and 36.453262
38
The responses of c*(th and c*(t) are determined by expanding ~(zh bk :w 37.453068 1111!!!!

C(z) into power series of z2 and z- 1, respectively, and are tabulated in Ta

1

4-2.
..~•.".':
r .

.11r"i

d.
, P••

.
,..
• tion rnetbod.
Th •
.
Transform Design 0 r . • ·
.
""'e trial and error uncertam11es are elim'n led •
I a
e point of departure in th
to th
i D1g,raf Controls 9S

e Parameter opti-
1111:edure is that the performance specifications :t~ameter ?Plimization
t
r
pr nee index, Integral square error performance . d ns1~t of a single Pcrfor-
p1a er performance m · d'ices can be used as well Fm ex 1s ve ry C011l11Jon but
ot h th t . . . . or a fixed sy t
tioD parameters a mm1m1zc the performa . s em configu.

3 ra The ready availability of the digital com~~et::te selected. .

to e
ot of modern optimal control theory which h be
f .
ed to the develop-
as comethem ·
Transform Design of Digital Controls .
e for the design o automatic control systems Th' aJor tccboi-

----~-----.. qU · h .
the second design approac mentioned above.
· 1stheory'1 b
for continuous-time control systems• the meth Ods of des1g
s ased on
. bo h
classical an d mo dern, have be en well developed and ar . n- t
roented in numerous books. c extensively docu-
ODUCTION . . . · d' . t·
3.1 INTR . bl encountered 1n 1screte• 1me systems are essenf .. Two situations may arise in case of design of digital controls:
Tbc design P:_" ed~s the design of continuous-time systems Bas· ial1ly .
• -
um•1ar to
those ,oun 1n
aplantwic
•
h. h needs to be controlled so that the output f •
1ca ly
0th
(i) An existing
. continuous-time control system has beco weII dcsigncd • •
! I!
h
we av~ beh
nt will ave a
ccording to some preset performance specifications
• h
°
• 1n
however m order
. to replace
.. it with a digital version , Ira ns,ormatlon
, • '
pla
the classical design philosophy, we _decide at t e outset that there should of analog design to d1g1tal domain is required. I

(ii) A control system is to be designed in digital domain • .·. '

be feedback from the output to the tnput reference so that an error he-
signals is formed; the controller then operates on the err Discretization of the analog design has already been discussed in Sec-
twceD t hC tWO 'fi . . fl or
signal in such I way that the design spect ~at1ons arc salts td . The appr- tion 2.7. In this chapter we shall be mainly concerned with the design of
oach relics heavily on Laplace and Foun~r tr~nsform representations of control systems in digital domain. We assume that the reader is familiar
be tem dynamics and the control spec1ficat1ons. System dynamics i with the classical design techniques applied to continuous-time systems.
;ud; and compared under application of st:indard test sig~als-impu)~ Here, we shall discuss modifications of these techniques to make them
step, ramp, parabolic. The n~ture of t~e transient response 1s revealed by directly applicable to the design of digital controls. It is found that the
any of these test signals as this nature ts dependent upon system poles only root locus technique of s-plane can be transferred unchanged to the z-plane,
and not 00 the type of the input. It is therefore sufficient to analyze the but that for Bode techniques to be useful, we need bilinear transformation,
transient response to one of the standard test signals-a step is generally which we will call the w·transform.
used for this purpose. Steady-state response is then examined with respect Our discussion will be limited to single-input/single-output (scalar) sys•
to this particular test signal as well as the other test signals. tems. The problems arising due to interaction in the design ofmultivariablc
The desired transient response to step input is specified in terms of peak controllers are highlighted in Section 3 8. The extensions of classical design
overshoot, settling time, rise time etc., and the de~ired steady-state response procedures for the design of multivariable controllers, have been reported
in terms of steady state errors to test signals (or error constants). The in the literature ; these topics, however, are kept outside the scope of this
frequency-domain specifications are phase margin, bandwidth, resonant book.
peak, resonant frequency etc. J"hese design specifications are selected be· An appendix on Bode plots has been focluded to refresh. the _memo'!
cause of convenience in graphical interpretation with respect to root locus or of the reader. Some simple process controllers and mcth°<l:s of scttmg theu
frequency plots. Compensators are selected that give, as closely as possible, parameters are descdbed with the help of unsolved exerC1ses at the end of
the desired system performance. In general, it may not be possible to satisfy this chapter.
all the desired specifications. Thea, through a trial and error procedure, an
acceptable system performance (within certain tolerable .error) is achieved. ESIGN EXAMPLE: POSITION SERVO .
3•2 D It m t to illustrate digit11l control theory with the
There are generally 1Djlny designs that can yield this acceptable perfor-
m~nce, i.e'., the solution is not unique. This trial and error procedure works
t
In this book, we_a :ntrol problems from various engine~ring ~isci-
he~p of some practt7 one such control problem is derived in this section.
satisfactorily ro_r scata~. (single-input/single-output) systems. The gap be- plmes. Plant model o lectromagnctic pulse is radiated from an antenna
tween the classical design procedure and its application to multivariable In a radar system, an c
(multi-input/multi-output) system has been bridged recently.
I 96 DIGITAL CON'l'ltOL BNGINEIIRING

r
into space. An echo pulse is received bac~ if a conducting surfaee Tramform Design of Digital Contro/J 91 !
ID ll'rplaoe appears in the path of the signal. When the rada . • such a
r isisl
•n searchs uc developed by the motor is assumed to be proportional to
ofa target, the antenna is• con!inuous
• Iy rotatcd. . When the target
1hC to_rq ut voltage to the motor, so that
the antenna is stop~ and point~ towards the target by -varying its Ocate<1,
direction until a maximum echo 1s heard. an&lllar ,(t)-tbc ,np T,.(t) = k u(t)
The position coordinate 8(t) of the antenna can be controlled -v· orameters, /
train by a d.c. servomotor. The torque of the motor can be 'Varied•~ agear for tb0 P B = 50 newton-m per rad/sec
magnitude and direction by means of a control voltage obtained r111 both
output • The controI problem 1s
of an amplifier. . to command the lllotrolll the J = 10 kg-m1
that .
orsuch k = 10 newton-m/volt
8(t) oc 6,(t), a reference angular position. 3 l) 1,ecomes,
c4n.(· 1Mt50d=10u+r.
A possible method of forcing the antenna to point toward the rcfcrc
position is shown in Fig. 3.1. 8(t) and 6,(t) are converted into clcctr~ or + Se = u + w; == 0.1 T., (3.2)
signa)s, the difference is amplified which serves as input voltage to th
motor. As a result. when 8,(r) - 6(t) is positive, positive input voltage i: .. .. . d . 'berl by the block diagram shown in Fig. 3.2.
and is escr1
aPPlied to the motor which makes the antenna rotate in a direction so that
the difference between 6,(1) and 8(t) is reduced. A negative input voltage •
makes the antenna rotate in the OPl)Osite direction. This scheme obviously
tepresents a closed-loop controller. e
GIii=_!_
1l1•il
Tw
j
Fig. 3.2 Plant model of antenna stabilization system
:,"
1,ltl
Tran,ciuc•r Controller
lAmp;.ritrJ Moto, Qnd
antenna
8(t,
Suppose that the plant of Fig. 3.2 is part of a control system thlt is
commanded by a digital com;,uter, This leads to a block diagram of the
Ii
form shown in Fig. 3.3, '
•
l
Tronsd<:cer !i
'!
I I•
i
Jllc. 3.t A control 11:beino for ot an antenna FIi, 3.3

l
0

Antenna and motor constitute the Plant of the 1yste111. The motion of 3.3 We
DESIGN SPECIFICATIONS t
the antenna can be described by the difl'erentia1 equation shall discuss design specifications in the con(cxt of the specific con-
trol problem-the p0sition servo, described in the earlier section.
Jl(t) + BtJ(t) - T. + T. . , . (3.1) The specifications for this system1. which are characteristic of manY others
where J = moment of inertia of motor and load; B .,. COeflicient of
viscous friction; T. = torque developed by tho !Dolor; IDd r,,.., Wind ~ t Is that tho reaclor 11 famlliat with tho desisn of unity and _1111,IY
torque (disturbance). · · foc,dbaCk continuous-time syttoms. With this t,ackground. tllO resutts presented
In this chapter for unity fccdt,aclc discrct~tlmo systomS can easily b.: utcndcd

...
for nonunltY fec:dbaCk case-
 98 DIGITAL coNTJlOL BNGtNBBRING
Tra11J/or111 DeJtg,r 0, Digit C
may be listed as follows: 1<tl 1
Q O"tro/J !19

t. Transient accuracy
2. Steady-state accuracy
3. Disturbance rejection
4. Control effort constraints
5. Sensitivity to parameter changes 1.ot:-:...r---11----'

We will discuss each of these in turn.

translt11t Accuracy
Jn the design of continous-time systems, the specifications of th t 5.
re~ponse are usually made in the time domain and then transl ran icnt
frequency
. domain either in terms of characteristic pole location sa.10eds-pl
to the
or m.terms of frequency response ,ea tu res such as resonant peak anc l-. tr ...,i
frequency, phase margin and bandwidth, ' resonant l-i--
r' -- - -----·ts ----..J t

Tune-domain specifications: Consider a second-order unity feedb k

F)g. 3,5 Unit-step response of undcrdan)l)ed ICCOnd-order IYSlt1II p
d
(Fig. 1,4), ac system \i
(J)!
,r - tan-1 [ V\- ~ 1
] \!
iI
Y(s)
R(s) == s1 + 2C(J),.s + w! (3.3) ,, ... (l).v1 - ~ 1 (3.4b)

where C== damping factor; and (J)• = undamped natural frequency. _ (for 2% tolerance band)
t• 0t ;'-(a)• 11
!I
(3.4c)
AsseenfromFig.3.6;thepcak ovenbootisam0001 • n . \
2 ., function of damping~- · omca Y decreasing Ii
!~
The given time-domain specifications can ·be transl tcd ·
Wn
· · 11 · · a 1ntcrmsor
charactenst1c po e ocattons ID the s-plane (refer eqIL (3.3)): '1
s(s+2~Wn)
I
;
S1, Ss = - t(I)• ±i"'•~'l,- ca (3.S)
Figure 3.7 shows the pole locations ins-plane.
It may be noted that the results established in (3.4) are valid only for
the second-order closed-loop .transfer function of the form (3.3). It can
Fig. 3.4 A unity-feedback system easily be verified that the effect of a zero1 at s added to the transfer
,3 -

function (3.3) is to contribute a proaounced early peak whereby the peak

Control systems are usually designed with damping factor C< 1 and overshoot may increase appreciably. Further, smaller the value of i.e.,
therefore only this case will be considered here. The transient behaviour"of the closer the zero to the origin, the more pronounced is the peaking
the underdamped second-order system ( 3.3) can be characterised by (Fig. 3.5), phenomenon. On account of this fact, the zeros on the real axis near the
origin are g~nerally avoided in design. However, in a sluggish system, _ibe
:1
(i) damping factor Cor peak overshoot to step input, Mp; artful introduction of a zero at proper position can improve the lraDJ1Cnt
(ii) settling time t,. response. II
If the rise time t, is also specified, it should be consistent with the
l'(s) + ~)w1/~
(S r I/~ cnS1JR1 lhal 1,. IO Ullit•~CI!
Ii
specifications oft, as both these depend-Upon Cand (,) 0 , as is obvious from 2. - = . . • . ; the scale actor w. r
the following relations (refer Nagrath an.<! Gopal (1981)) : R(s) s• + 2~(1).,r + <o1!
input Is equal tn unity.
i
M, = exp ( - irt/v' 1..: t•) (3.4a)
\

L
 t~
· Transform Dengn ·'D· .
o, 1g1tal Contro/1 101
(11,~.-1!'10 .der now the addition of a pole• at s,.,, _ t h
Co'fosh1 effect of this pole will depend on th/ o t .e transier function
100 pl(JITALcOtfl"
(3,;,~)· ves e cl,1ser to the origin,
. . the system responmagnitude
. of 11• As th'1s
i I
. ptile the effect of the two complex poles at:"t!ominated II
:I: lw.v'I - t•
~7
11-
pd~11015bes On the other hand as this pole moves a 1-1
diL 5 on~e is dominated by the complex poles at rom the ~gi~ Il ll
the te !erved that for ~tt(J). = 6, the response of the th~d ± }CJ'J,Vl - t1• I

Jt is ractical purposes, the response Y(t) of the se~r dorder system Is,
for a ·'opll (3 3), Therefore, the results presented ear11·e ·order trans fer
f nct1
u .., 5 can•be extended to h'1gher-order systems with ard1or .second-ord . er
5yste.... . h ommant pair of
lex poles, 1,e., as long as t e real part of these polesi
·s m h h
colllP h , uc 1ess t an
tbOSe of the other po1e~, t e_t,me response o~ higher-order systems can be
.,I a proximated_ by that obtamed by neglecting all the system poles other
t{an the dominant poles. _ _
\··
speclllcatioos !o t~rms of roo~ locatioo_s in z-plane: For the system of 1
Fig. 3.4, we will aim to consider transient response specifications in terms !~
0 f characteristic root locations in z-plane. The approach will be to first l
btain the specifications in the a-plane (Fig. 3.7) and then use the relation
0,6 0,8 == e'r to map the poles and zeros in the a-plane to the z-plane.
o.4 1,0
The relationship between s-planc and z-planc was established in Section
f-
2.4 (refer Fig. 2.8). Figure 3.8 shows some specific patterns on the .r-plane
and the z-plane under the transformation z = e'f. :
FIi• 3,6 M, versus t for a _second-order system In the z-plane, curves of pole locations for constant t are logarithmic
spirals (Fig. 3.8a); the curves of constant arc lines drawn at right angles i
to the constant-t spirals (Fig. 3.Sb) and the curves corresponding to con-
stant tCJln (real part of the dominant roots) arc circles (Fig. 3.Sc). In Fig. 3.9,
the patterns are traced for various frequencies and various damping factors. i
The transformation z = e-r maps the pole ,l
Ill

I
s-plon•
s=- tw. + j.._yl - ti = - a +jb,
in the s-plane to the location
\
wnfi-12 !
111n z = reJ' (3.6) i

. r = exp (-6t/vl-t1).,. r7'

tr
1 8=tt1.T~=-bT
ecc0i· :r
-t~
IiIi
3_ Y(s) _
61:« ;tbuc:alo factor 61!« ensures that 1,. to
\t
R(1) (, + «)l,' + + 111:>_
unit-stop Input Is equal to unit)'.
fls.17 Pole locati0111 for an undcrdamped s: cond-ordcr system

It... \'
I
102 DIOft'AL (.lOll'BOL IJGNIDINO
1'ra11Jforrri Dei/g11 of Di"'t I Co
60
Ill Im ii;:j-i----._ a "''ob 103

••plartt

. _,
(

t
''
(a) Colmant clampins-loci for l;-cos-1 8 (c.i,-2n/T,
the 11111plin1 frequency) .
jw
lm ii
r
, • plant 1
"'"
--t-----1-~-•
r 1,l
Re ":!
,.,
\
I •I
1'1

i ·~1

·\'11,,
\-t
(b) C.onstant Coln•loci ,,
T~

Fig. 3.9 Constant i:- and constant w.-Ioci in ,-plane;

llm
I z=e'1', S=-l;w.±j w.yt-~1
jw -~wnT
2•. ,, ' \1'..
'-
,-plQ,.. :1
The time-domain performance specifications can thus be traaslatcd to
the following closed-loop transfer function in the z-planc: \ .J

-\'111n1 I . -1
Rt R(z) = (z
l
- rei6) (z - re-JO)= z1 -
1
2r cos 6z +rs
(3.7)
1
I
Ii
In the z-domain design, our attention will be restricted tothcslcprcsponse
of th e discrete-time system given by (3.7).
J
It can easily be established by simulation that the addition of a pole or

l
l.
zero to.the transfer function (3.7) has only a small effoct on 1hc step res-
ponse if the added singularities are in the range from Oto - 1. Howc'/Cf,
l
(o) Constant (~Col.)-Joci
l'I&. 3,8 a zero moving toward z = + 1 greatly increases the syst~m ovmhoot. A
pole placed toward z = + l causes the response to slow down and tbus

_........_
" 116 01orrAL co
!lfllOL ENGINSJ!RINO
r BT
"' Traneform Dti/gr, of Digital Controls 117

S, == 80 X f
a be selected (Katz 1981). Another rough guide is to select a sampling period
(3,28)
that is much shorter than a_ny of the time-constants in the contlnuou1-time
• d r the sensitivity of r with respect to H. plant to be controlled digitally, A 'practical' rule given by Oustovsson
Let us aoft' 11n 5•• . (19 75) la to choose the samplina Interval equal to the smallest time-constant
T H I+ H(z) - H(:) X H(z) = ·1 or tho lnvme orthe laraest real polo (or real part of complell pole) or
fJi
s~"' XT == H(z)/[l + H(z)] 1+1l<z) · lotore1t , Howovor, th l, may prova to be a very conaorvatlve choice and hair
(3.29) thl• 1amplln1 l!lterval may •uffice, Another rulo of thumb alvcn by
ouatov11on Is to use one-tenth the major timc-con,tant 11 tho 1ampllna
Interval,
At the frequeocY .,, I . Like the trial and error method, these rules too aro without much or a
S~ .. ~==TT D(el•T)G,..G(eM) theoretical bas!,, _Some cff'cctive 1nalysi1 tool, to auide the de1lJner In
makln11 tho_samplina rate choice are &ivcn by Franklin and Powell (1980),
. h. the system bandwidth, we require that lli(el•l)
For Sar to be small
r
wit in • h I .
be reduced by increasing t e open- oop gain. Howcve
3 4 DESIGN ON THE w-PLANE
• s can .
be large, 1.e., R rl' •A-asing the open•1oop gam can cause stab'il't
r, · The design of analog compensation network using the frequency res-
·oned ea ier, i....... . f d •h IY ponse properties of t~o plant and of the compensation network is well
as we men b again in design we arc ace wit trade-offs
problems. Therefore, once . established and extensively dccumcnted in several books (refer, for example,
Nagrath and Oopal (1981)). Elaborate graphical methods have been deve-
"'-pie Rate Setectlon · loped and since a generation of designers arc used to them, the method,
"""' . difficult to formulate a uniform method",or th e selecuon
·
of an arc still widely used in practical design, , .,I
It is ~ery 1· g rate 1/T (or sampling frequency ..,, 21t/T) The
appropnate samp•tng theorem (Section
well. known 58tDth
• }.6) wh'1ch says th at •sample at· least
P110highest frequency conta1De
• d · th · I' , 1s
• not always
Unfortunately, pulse transfer functions are typically not rational func-
tions, but rather the frequency appears in the form z = el•T and the simpli• -~
twiCe as ,ast as e
ID
• h · I
c s1gna
city of frequency• domain design techniques is altogether lost in the z-plane. i. 11
applicable for digital control ~ystems, wherein t c s1gna s. are ~ot accurate-
The cure is to make a transformation to a different plane (called w-plane)
ly delloed and further, a step mput to the system has an mfimte frequency
where the simplicity is regained. As a matter of fact, we have already used
content 'bl • · · the bilinear transformation (3.8) to investigate the stability of discrcu-- ~.
1be designer wants to use the lowest p~ss1 e 5:1mp1mg rate. The basic time systems. With this transformation, called the w-transfonn of z-transfcr l
motivation for this is the cost. A decrease ID sampling rate means more time function, frequency-domain techniques of design of analog compensation I'
is available for control calculations, hence slower computers arc possible can directly be applied5• With reference to the block diagram of Fig. 3.3, a.
for agiven control function, or more control capacity is available for a given '
l
the frequency-domain design procedure may be outlinedasfollows:
computer. For systems with .A.JD converters, there will be less demand on
1. Evaluate G1,oG(z). Apply the w-transform defined in (3.8a) to obtain
conversion speed resulting in low cost. However, system performance is G1,oG(w).
degraded as we sample more slowly (refer Example 2.12). Thus economi- To obtain
cally, the best choice is the slowest possible sampling rate that meets all the b
performance specifications. G.,G(w) = G.,G(z) \,. i+• (3,30a) i.
We are usually satisfied with the trial and error method of selection of
WDpling rate. We compare the response of the conti:iuous-time plant with
models discretized for various sampling rates. Then the model with the
•-·
the following ready-to-use formula may be used.

n• (z + a,)
l
slowest sampling rate which gives response within tolerable limits is select- 1 1 ' I
ed for fu_ture work. However, the method is not rigorous in approach. G,..G(z)= • • (3.30b)
Also a wide .variety
. of 1·nPUts must be given
· to each prospective . model to (z - I)' n (z +b1)
,-1
ensure that it is a true representative of the plant.
There
• t h
are a number
. .
of rules of thumb available
. . . .
111 literature. The required
t rans1en c aractenstic m be . S. It is not possible to review the design of analoa compensation ll<IWotb adequ-
:ddth and then a s a! translated m terms of closed-loop band· ately in this brief account. It is assmcd that the reader is wdJ conversant with the!o
amphng rate ten times larger than the bandwidth may topics of classical eontrol theory (Nagratb and Gopal 1981).

_A__J
1·
t t 8 DIGITAL CONTllOL B!WINf!l!lllNG·-
Transform De11lgr, of Digital Cor,tro/1 119
0 (I + a,)(I - w)f-+A 1ii1 [1 + o ·.+-a,)/(1-a)
~ ----]
1-1 - 1
: [1 - e-T, ( ..,...!S_)1- l6(z
_ K(zI)+(z0.76)
G..a(w)- - ~
D (l + b,)(2w)'.
,..
P-1
• [
l +(1
w
+ b J )/tl ~
- b
(3.30c
1)
G,..G(z) = fE · s s(s + 2) _ 0.45) (3.31)

• oslru
ct the Bode plot of G,,.G(w) in db (decibel) mag •
. ) T n,tud
) Let
. · us' set t~e ramp error
1 constant K, to be at 4. From cqn. (3.19b),
2 Co
1iaseve"us .,,(6c:tittous frequency on w-p1ane . ransfer the d e and
rbe K, = -T lim [(z - 1) GAoG(z)]
Bode plot to the Nichols chart if necessary. Determine th ata from
maoce characteristics of the uncompensated system by find in
margin, bandwidth aod resonant peak from the Bode pl~t he Phase
Perfor. t or
•-+1 '

Nichols chart. and the 4=~

• If the ped'onnance is unsatisfactory, the system needs co 2
3 Tbe open-loop transfer function of the digital contro~~ensation. 5t
'(bus K = 8 may be 1selected
[D(z)G11oG(z)) or in w-dom~in, [D(w)G1,oG(w)]. Using the m Y ern is
desi,n of analog compensat1on, the transfer function D(w) . et hods of
+ w to satisfy steady-state performance requirement.
w and K = 8 into eqn. (3.31) yields (alternatively,
4. Ooce D(w) is determined, D(z) is obtained by using
form defined in (3.8b).
i:
so that the desired system performance specifications ar is ~etermined
satisfied.
verse W-trans.
substi•tution of z = 1- -
use eqn. (3.30c)),
- -

Ii
S. The ftnal step In the design pro..:edure is to convert D( )
cal algorithm (discussed in Section 4.S). z to a numeri• G1ioG(w) = O.S(- w +
l) (w/7.35 1) + f'
_ w(w/0.38 l~) - + (3.32a)
EnmpJ:, 3.1: Consider the digital control system shown tn
• Fig.
. 3,18.
G1,oG(jv) = 0.8(- fa+ l)(j,/7.3S + I}
_ . jy(j,/0.38 1) + (3.32b) l
t
lllal . Gl'lolll
O!Sl•~I
K Yttl The Bode plot of G,..G(jv) (i.e •• the uncompensated system) is shown in
Fig. 3.19. We find from this figure that the uncompensated system has a I
Co"'Pen- ION cross-over frequency "ci = 0.52 and a phase margin ,f,~ = 14°. It is desired
Plont to raise the phase margin to the specified value ,f,, = 40" without altering
K,, the error constant.

Fil- 3.18 A digital control system It is noticed from the Bode plot of the uncompensated system that
the desired phase margin can be attained by modifying the db-log .,, plot
so as to lower the cross-over frequency, while the phase plot is not allowed
The system is required to meet the ~ollowing specifications: to alter significantly in the region of the new cross-over frequency. This can .
(i) velocity error constant. K, ;;a,, 4 sec be achieved by a lag compensator with the transfer function,

(ii) phase margin, ,f,,.. ;;a,, 40" +-

D(w) = 11 + f:lw-c; > 1, T > O (3.33)
(iiJ) bandwidth "'• = 1.5 rad/sec

sampling frequency equal to 10 times the bandwidth, we choose

Takinginterval Bode plot of the lag compensator is shown in Fig. 3.20. From this figure it
sampling is_ obvious that the lag compensator has a d.c. gain of unity, while it offers
high frequency gain of 1/~. Since~> 1. it means that the high frequency
2,r noise is attenuated in passing through the compensator; signal to noise
T=z Ts°"0.4 sec.
ratio is thus improved.
The high frequency attenuation (- 20 log ~) Is utilized to lower the
. ~be
tion z-transform
is (for T- 0.4), of the uncompensated open-loop sys h!m transfer func- cross-over frequency while the two corner frequencies of the compensator
(1/~T and 1/T) are placed sufficiently lower than the desired cross-over fre-
quency so that the phase-lag contribution of the compensator at this cross-
over frequency is made sufficiently small. Usually the upper comer frequency
 -~

r-r, Tran1/or,n Dt1ig11 of Di""tal C

~[TJJJ
H-
J :rn b compensator is plaecd one octave ,to one d
oC t :ver frequency .,,ca of the compensated system
059
d11• cntrot., 121
e Il~wer than the
so~I\.. '7"" cr 01 au phase-lag contribution of the compen' to nuh11:Y the effector
tbe sent
s at .,,,,, the u~cornpensated system must contribute
sa or wan•ch glWill} till "-

'
C
7""
IQ..-

o~,...0-011
I
~-
h.\.,
\
N~

...._ ,
-~1m-i I 1111111 -+-i..
. rt-
pre
towards phase margin where,.

' ,. - tf,,+ f;
an e .,., at ,
11

-10-
I'""'
vr • o:~.'\1 - \
. ..
"""~ 1-
11 allowed a value S-lS°.
• 0-
-10
1
1 ,I,
1
Ufticlff'P•"•olld 1'111,,,_ .__ .
V
I m

I\
~"':~. '"f:I:,,
t - - ,-.!";,~te-
"':· ·-
-,-~J
:::!?:~ ' It ,nay bo notod tb1t tho
rrequcncy
""''""'°' Contril,utos o db lo 1h I
region so that the K, remains unaltered by lntroduclna th:
I'
,.... - - ,-~ · ··•~ ....J.,Jj oipensator.
.,o -~mp,r,sotfd IYI~~"'. , ·- • ~· ~-.
,...--, ,At - I-
/ ,-. I- . •·- - .
~r- . " From the Bodo plot or tho '"'"''°"'lod ,,.,,. .,,., ;. 3. ,
It 1, observed that neglecting the phase-lag contribution of the compen-
11
• t00'- V ._,_ - --.. tor the specified phase margin of 40" 11 obtained it the cro11-over frequency
•f'/0
1
•~ ,, I ,-:
1
0.3 rad/aec, Since thi1 Is fairly low, the upper corner frequency ot tho
'•!J.0'
... ,
i·'',;?~
..-.,,11 1:
,_r
- ;- mpenaator cannot be taken far to its left In order to avoid larae time-
:~stants, This Indicates that phase-tag contribution ot tho- compensator
-160 r.- 1
1
11 1~ .11' 11 'I. T at tho new cross-over frequency will be considerable and may be auelled
-110' II
Q,11
\ - ~· at 15", The uncompensated system must, therefore, make a phue IDll'lin
contribution of
-200' I " ,..
-zzo' I ' ·-- - ,. • 40' +, == 40' + 15' = ss-
0·01 O•I 1·0 10 ..
v- It~ at "ct which is found to be 0.18 rad/sec.
Placing the upper corner frequency of the compensator two octaves below

II
Jlls.11' ~ t a r design ror & sampled-data control system in the W-dornaiu 'ct• we have
o--,.... ' 1/,: = vcJ(2)1 = 0,045 rad/sec.

t To bring the log-magnitude curve down to O db at v"' the lag com-

s,.-2odb/~ca pensator must provide an attenuation of 12 db (Fig. 3.19),
4~
IDl,jY)I
I Therefore,
20 log~= 12
or
o' ~=4
The lower corner frequency of the compensator is then fixed at
I
t-45'
/J:,ejYJ
1/~T • 0.011 rad/sec.
The transfer function of tho tai compensator Is then
I
ij
\•

f
1 +22.2w
D(w)= - ----
-to'~ 1 + 88.8w
1/jT '
\rm• 1//pT
Phase lag Introduced by the compensator at "•• .. tan·1 (22.lv.J-
1/t
tan-1 (88.8 .,,1) - tan-1 3.996 - tan-115,994 - 76- 86.5 • - 10.S-,
logv-
3.20 Bode plot ol the pbue-laa network

....-...