2 The z Transform 2-1 INTRODUCTION ‘A mathematical tool commonly used for the analysis and synthesis of diserete-time control systems is the z transform. The role of the z transform in discrete-time systems is similar to that of the Laplace transform in continuous-time systems. Ina linear discrete-time control system, a linear difference equation character- izes thie dynamics of the system. To determine the system's response to a given input, such a difference equation must be solved. With the z transform method, the solu- tions to linear difference equations become algebraic in nature. (Just as the Laplace transformation transforms linear time-invariant differential equations into algebraic equations in s, the z transformation transforms linear time-invariant difference equations into algebraic equations in z.) “The main objective of this chapter is to present definitions of the z transform, basic theorems associated with the z transform, and methods for finding the inverse 7 transform. Solving difference equations by the z transform method is also dis- cussed. Discrete-Time Signals. Discrete-time signals arise if the system involves a sampling operation of continuous-time signals. The sampled signal is x(0),x(7), x(27),..., where T is the sampling period. Such a sequence of values arising from the sampling operation is usually written asx(kT). Ifthe system involves an iterative \ process caried out bya digital computer, the signal involved isa number sequence ) (0),x(1),x(2).., The sequence of numbers is usually written as x(le), where the argument k indicates the order in which the number occurs in the sequence, for example, x(0),x(1),x(2).... Although x(k) is a number sequence, it can be con- sidered as a sampled signal of x(¢) when the sampling period T'is 1 sec Scanned with CamScanner28 The z Transform Chap. 2 The z transform applies to the continuous-time signal x(t), sampled signal x(KT), and the number sequence x(k). In dealing with the z transform, if no confusion occurs in the discussion, we occasionally use x(k7) and x(k) interchange- ably. [That is, to simplify the presentation, we occasionally drop the explicit appear- ance of T and write x(&T) as x(k) Outline of the Chapter. Section 2-1 has presented introductory remarks. Section 2-2 presents the definition of the z transform and associated subjects. Section 2-3 gives z transforms of elementary functions. Important properties and theorems of the z transform are presented in Section 2-4. Both analytical and computational methods for finding the inverse z transform are discussed in Section 2-5. Section 2-6 presents the solution of difference equations by the z transform method, Finally, Section 2-7 gives concluding comments. 2-2 THE z TRANSFORM The 2 transform method is an operational method that is very powerful when working with discrete-time systems. In what follows we shell define the z transform of a time function or a number sequence In considering the z transform of a time function x(i), we consider only the sampled values of x), thatis,x(0),x(T), x(27),..., where Tis the sampling period The transform ofa time function x(), where fis nonnegative, or of asequence of values x(K7), where k takes zero or positive integers and Ts the sampling period, is defined by the following equation: Dake" 4) X(z) = For a sequence of numbers x(k), the z transform is defined by 2 [xe@)] = ZEN) HG) = Zl] = Eateye* @2) ‘The z transform defined by Equation (2~1) or (2-2) is referred to as the one-sided z transform. io The symbol 2 denotes “the z transform of ” in the one-sided 2 transform, we assume x(t) = Ofors 1. In finding the tianstoni, the variable. ‘Zacis as a dummy operator. fesnotne ‘nesessary to specify the region of z over which 1 X(z) is convergent. It suffices to know that such a region exists. The z transform X(z) of a time function x(t) obtained in this way is valid throughout the z plane except at poles of X(z). It is noted that 1(k) as defined by = 0,1,2, 1, k=0,1,2, is commonly called a unit-step sequence Unit-Ramp Function. Consider the unit-ramp fuinction Ost 0, <0 Notice that ‘ XQN)=AT, = 0,12)... , Figure 2-1 depicts the sampled unit-ramp signal. The magnitudes of the sampled values are proportional to the sampling period . The 2 transform of the unit-ramp function can be written as t= Sees = Seret = 7S tet = Dz? hace (ea + 2278 + 3273 4---) ) poe \ \ KBE baw Bl doe | fen tot t ‘fe Veg iae et) Zeta «Tele a8 497 peor, eh fy | fe eepeess Fn ot fe Kod, wan \be fle Teles Sap a 7 1 a'Fl) 2 T's oT8 we 1) -@) 94-2) FB) 2 TAY TE ua ant) dae mar thply Lad Side ef 6 ¥ | ina Te reTeis Tee ®D NOS 1f@), Tate Te te ee stot > Te Ally -#'- aje Fey — x()= XQ) of tere AT u) xo -¢ Daa d Scanned with CamScannerSec. 2-3 2 Transforms of Elementary Functions 2 Note that itis a function of the sampling period 7. Polynomial Function a Let us obtain the z transform of x(k) as defined by where a is a constant. Referring to the definition of the z transform given by Equation (2-2), we obtain X@) [a] Sw Zar tarttatzt tarts Exponential Function. Let us find the z transform of ose <0 Since x(kT) =e", k= 01,2, we have Xa) = Zle*] = Daler et = Dersts SLtemt teem tte My Sinusoidal Function, Consider the sinusoidal function x0 (ee wt, O56 0. a) Noting that eM = coset + j sin at v eM = coset ~ j sin wt we have Scanned with CamScanner28 ‘Thez Transform — Chap. 2 Since the z transform of the exponential function is a 1 2e1- a we have XG) = Z{sinat] = zt (eH ~~] -iG-ba 1 ) aM ez 1l-ey Li emne “RIM sin wT, Example 2-1 Obtain the z transform of the cosine function won [e125 If we proceed in a manner similar to the way we treated the 2 transform of the sine function, we have X(2) = Z[eosut] = | Zl +e] 2= (eM + eh) Fe ee . wh 1 the’ etre fans off t)eat — Bxample2-2 (Obtain the z transform of teva, cmt the XG) 1 wD = SA Whenever a function in 5 is given, one approach for finding the corresponding 2 transform is to convert X(s) into x(0) and then find the z transform of x(1). Another approach isto expand X(5) into partial fractions and use a z transform table to find the + transforms of the expended terms Still other approaches will be discussed in Section 3-3 Scanned with CamScannerSec. 2-3 2 Transforms of Elementary Functions 29 “The inverse Laplace transform of XG) is Ava w | aeine, ose LO X@)=20-e4=72 ober w-rhire ner, = U-e @-1e--) Comments. Just asin working with the Laplace transformation, a table of z transforms of commonly encountered functions is very useful for solving problems in the field of discrete-time systems. Table 2-1 is such a table. 7) zfs (h) = 4 nble 21 uch 8 ME Saye BLS) ranger roicoremansroms ete, ¢ SO at 8 xo |) | eon) xe) rons dis 8) if = = rer 1 a kao Bin — k) # nh al = - wie . 1 uae al 1) i a 4 ita 7 . i sf} a [is a ae 2 A ; ; Prasey )T Ha) 13 fad ve Sart “Ga § ; 5 Pea aey x] § ’ ary Pegees 8) Gere \ oa | 9 arama ere | 1 0 or | w| ety | te | / ul gig [aca] o-anes | r Scanned with CamScanner30 TABLE 2-1. (continued) ‘The z Transform Chap. 2 XG) x) x(kT) or x(k) XG) 2 ren penn revaset | 2) aa ary ery ca [erate ye Goto B antter | akT-14ee" i a ec wl gta | sn sinwkT 3} zig | com cont wl arira| ae “sn akT weit | ete | eerconatr " ¢ ae > oe ; k22,3, 20 kat! a ee % are Ede der) 7 erry - Petes eres 3B wat"! Plas nets ners ee :y 4 cock ra 2 = 6 Bs end a n HED > WED Gm, (0 = 0. for <0 7 MKT) = (4) = 0,f0rk <0 Unless otherwise noted, k = 0.1,2,3, Scanned with CamScannerSec. 2-4 Important Properties end Theorems of the z Transform 31 2-4 IMPORTANT PROPERTIES AND THEOREMS OF THE 2 TRANSFORM “The use of the z transform method in the analysis of discrete-time control systems may be facilitated if theorems of the z transform are referred to. In this section we ‘present important properties and useful theorems ofthe z transform. We assume that the time function x(¢) is z-transformable and that x(t) is zero for ¢ <0. ‘Mubiplication by a Constant. If X(z) is the z transform of x(#), then Zlax(t)] =o Z(x@] = aXe) where a is a constant ‘To prove this, note that by definition 2 [exe] = Zaxtkre* = 05, x(kT)e* = aX) Linearity ofthe « Transform. The z transform possesses an important prop- erty: linearity. This means that, if f(k) and g(k) are z-transformable and a and 8 are scalars, then x(k) formed by a linear combination x(k) = afk) + atk) has the 2 transform X(z) = aF(z) + BG(z) where F(z) and G(z) are the z transforms of f(fe) and g(k), respectively. “The linearity property can be proved by referring to Equation (2~2) as follows: X(2) = 2[x(6)] = Zlaflk) + AB) Ldegunttine of 2 5 fia) - Zia = Zlerte + astiole rk te? = adie + Beto a2 (fk) + BZ (800) aF(z) + BG(z) Multiplication by at. _1f X(z) is the z transform of x(k), then the z transform of atx(k) can be given by X(a"'z): [Bley = Xz) / 2-6) This can be proved as follows: Zfete(e] = Sete Sxuyety Y yf Freer boa = X(e"2) v ced / Shifting/Theorem. ‘The shifting theorem presented here is also referred to as the real translation theorem If x(t) = Ofore < Oandx(t) has the 2 transform X(z), thei Scanned with CamScanner® a beg (apap resem Chap. 2 =a) = 2"X(6) ee and & wee __ fete + 079) = 2 x02) - Sacer 3) ge lee tea < ~ where 1 is zero or a positive integer wh To prove Equation (2-7), note that ay = ! e Z[x(e— nTy) = D x(k - nT) v xT nT 2-9) By defining m = k ~ m, Equation (2-9) can be written as follows: 2[x@ - nT) = © xemT)" VA Since x(mT) = 0 for mt < 0, we may change the lower limit of the summation from m= —nto m= 0. Hence, Z [x(t ~ nT] = 2 Dxene =r"X@) (210) (Thus, multiplication of a z transform by 2"* has the effect of delaying the time ; function x(¢) by time nT. (That is, move the fynction to the right by time nT.) jae 2% alles To prove Equation (2-8), we note that (Line. xdvawsce) cle deg pp - Zee + aT = Deer + net seSurenees So, oy at = aane] v = 5 kT + nye +S xtarye* = Be Rajlanae, * o[Sxene* - Seanes] = »fxer - Zens] 4 For the number sequence x(k), Equation (2-8) can be written as follows: zt =mi=e[xe- Sea] ¥ From this last equation, we obtain Zix(k + D) = 2X2) - 2x0) 2-4) Elae +2 = ez IMR + 1] sx) = PX) ~ Zx(0)~ 2x1) 12) SL plee)-x00-2009|% = PXQ)- # ufo) 2nd) Scanned with CamScannerSec. 2-4 Important Properties and Theorems of the z Transform 33 Similary, Zla(k +n) = P-X(z) ~ 28x(0) — 2 1x(t) = 27x) - where m is 2 positive integer ¢ TAN gemember that multiplication of the = transform X(z) by z has the effect of |) advancing the signal x(K7) by one step (I sampling period) and that multiplication # Y ofthe: transform X(z) by 2~" has the effect of delaying the signal x(k7) by one step [ Gsampling peri), Example 2-3 Find the z transforms of unit-step funetions that are delayed by 1 sampling period and 4 sampling periods, respectively, as shown in Figure 2-2(a) and (b) Using the shifting theorem given by Equation (2~7), we have = X22") wat oie Ter - nettae pete} feat = | Scanned with CamScannerSec 2-4 Important Properties and Theorems of the z Transform 35 Noving that 27 sinwT 2 bina] = pT BT eAT FE yay the Covy', ‘we substitute ae for z to obtain the z transform of e~* sin at/jas follows: “ let ined) = ge SRE coal +e Similarly, for the cosine function, we have tas" coset 2 eos] = Ta ase +2F By substituting ze*" for z inthe z transform of cost, we obtain Ze" cosu) = a ee Example 2-7 ‘Obtain the z transform of we ‘Notice that 20> Goa xe) Thus, rer Xe") = ery Initial Value Theorem, —YEx(¢) has the z transform X(z) and iflim X(z) exists, then the initial value x(0) of x(¢) or x(&) is given by a x(0) = limX(@)_, 415) ‘To prove this theorem, note that X(z) = D x(kpet = x(0) + x(A)zt + x(2)z7 + Letting 2» in this lst equation, we obtain Equation (2-13). The behavior ofthe signal in the neighborhood of = Gor & = Ocan thus be determined by the behavior of X(z) atz = © "The intial value theorem is convenient for checking z transform calculations for possible errors. Since x(0) is usually known, @ check of the inital value by lim.¥(2) can easily spot errors in X(z), i any exis. Example 2-8 ae(o) e fam 2 (*? Determine the intial value x(0) if the z transform of x(0 is given by ae ira rar By using the inital value theorem, we find ae Tet 2O= ingsSateey 8 Vv Scanned with CamScanner36 ThezTransiorm — Chap. 2 Referring to Example 2-2, notice that ths X(z) was the z transform of a= ie" and thus x(0) = 0, which ogres with the result obtained eatir Final Value Theorem. Suppose that x(k), where x(k rk <0, has the 2 transform X(z) and that all the poles of X(z) lie inside the unit circle, with the possible exception of a simple pole at z = 1 i the condition for the stability of X(z), or the condition for x(k) (k ) to remain finite.] Them the final value of x(K), that is, the value of x(k) as k approaches infinity, ean be given by Jim (K) = lim [1 ~ 2°)XG)] (16) To prove the final value theorem, note that ZEx()] = X@) Sewer (4 Zlxlk~D)=2'XG)= Dxke- yet yw (2) Hence, talong, (1) (2), Saxtiyet ~ Sate 3 = x@)- 21x) ‘Taking the limit as 2 approaches unity, we have _taBewe- Soxtk = 1)e" im [(1 ~ 2-)X(@)) Because of the assumed stability condition and the condition that x(k) = 0 for <0, the lefthand sie of this last equation begomes Zee — a(k ~ 1) = (xO) ~ toile +k +bxQ)~ x] + 1 Hence, fim x(k) = lim (1 ~ 2-)X(@)] which is Equation (2-16) The final value theorem is very useful in a the behavior of x(k) as k—»* from its z transform X(z). } ‘ 8) atHeat the weed 1 fee fab ! Example 2-9 Determine the final value x(=) of Xe) = a>o by using the final value theorem By applying the final value theorem to the given X(2), we obtain Scanned with CamScanner2 Transform 37 Sec. 2-5 The Inver (2) = fim[@ - 27).4@)] It is noted that the given X(2) is actually the 2 transform of sels" By substituting ¢ = «in this equation, we have x2) = ‘As a matter of course, the two results agree. ‘Summary. In this section we have presented important properties and theo: ems of the 2 transform that will prove to be useful in solving many 2 transform problems. For the purpose of convenient reference, these important properties and theorems are summarized in Table 2-2. (Many of the theorems presented in this, table are discussed in this section. Those not discussed here but included in the table are derived or proved in Appendix B.) 2.8 THE INVERSE z TRANSFORM ‘The z teansformation serves the same role for discrete-time control systems that the Laplace transformation serves for continuous-time control systems. For the z trans- form to be useful, we must be familiar with methods for finding the inverse = werse z transform is 2”. The inverse z transform of X(z) yields the corresponding time sequence x(k). It should be noted that only the time sequence at the sampling instants is obtained from the inverse z transform. Thus, the inverse z transform of X(z) yields a unigue x((e), but does not yield a unique x(é). This means that the inverse 2 transform yields a time sequence that specifies the values of x(¢) only at discrete instants of time, ¢ = 0,7,2T,..., and says nothing about the values of x(¢) at all other times. That is, many different time functions +(¢) can have the same x(K7). See Figure 2-3. ‘When X(z), the z transtorm of x(kT) or x(k), is given, the operation that determines the corresponding x(kT) or x(k) is called the inverse z transformation. An obvious method for finding the inverse z transform is to refer to a 2 transform table. However, unless we refer to an extensive z transform table, we may not be able to find the inverse z transform of a complicated function of z. (If we use a less extensive table of z transforms, itis necessary to express a complex z transform as a sum of simpler z transforms, Refer to the partial-fraction-expansion method presented in this section.) Other than referring to z transform tables, four methods for obtaining the inverse z transform are commonly available: / Scanned with CamScanner50 ‘Thez Transform — Chap. 2 and 2 sink =) say we have ke 4 2 (k-1): c sym [fH SDE, 21,233, 0, ks0 ‘Although this solution may look different from the one obtained earlier, both solutions ‘are correct and yield the same values for x(k). + Inversion Integral Method. This isa useful technique for obtaining the in- verse z transform. The inversion integral for the z transform X(2) is given by okt A 1-2 sifxiora oy where C is a citcle with its center at the origin of the z plane such that all poles of X(z)z*" are inside it. [For the derivation of Equation (2-23), see Appendix B.]-> SJ The equation for giving the inverse z transform in terms of residues ean be <6 derived by using theory of complex variables. It can be obtained as follows: 2KT) = (k) = Ky Ky to + Ky 4 27UXG)] = x(RT) = x(k) = [residue of X(z2)z!“ at pole z = zj0f X(z)z*“"] ¥ (2-24) where Kj, Kz, -.., Km denote the residues of X(z)z‘"' at poles 21, 22, - , Zn, TeSpec~ tively. For the derivation ofthis equation, see Appendix B.) In evaluating residues, note that if the denominator of X(z)z*“! contains a simple pole z = z, then the corresponding residue K is given by K = lim[(z = 2)X(z)z*"] ¥ (2-25) If X(z)z"" contains a multiple pole z, of order g, then the residue K is given by 1 tim XC) “ Ko Gapping a x@) (2-26) Note that the values of & in Equations (2-24), (2-25), and (2-26) are nonnegative integer values. If X(z) has a zero of order r at the origin, then X(z)2!"! in Equation (2-24) will involve a zero of order r + & ~ 1 at the origin. If r = 1, thenr +k —1=0 for k 2 0, and there is no pole at z = 0 in X(z)2""!. However, if r = 0, then there will be a pole at z = 0 for one or more nonnegative values of k'JIn such a case, separate inversion of Equation (2-24) is necessary for each such Value of k. (See Problem A-2-9.) > Tt should be noted that the inversion integral method, when evaluated by residues, is a very simple technique for obtaining the inverse z transform, provided that X(z)z*~" has no poles at the origin, > = 0. If, however, X(2)2*" has a simple pole or a multiple pole at z = 0, then calculations may become cumbersome and the Scanned with CamScannerSec. 2-5 The Inverse z Transform 51 pattial-raction-expansion method may prove to be simpler to apply. On the other hand, in certain problems the partial-fraction-expansion approach may become laborious. Then, the inversion integral method proves to be very convenient Bxample 2-16 Obtain x(K7) by using the inversion integral method when X(z) is given by xe) @= De =e) Note thet ate OST XO = Sesh For k= 0.1.2... X(z)2**" has two simple poles, z= 2 Hence, from Equation (2-24), we have $ (ire vod sence =KtK where residue at simple pole z = 1] inte - ngage] an) nt @ = Ye -e™). = [reside at simple pote = = lim, fe =: we Hence, . Fen Kee Example 2-17 ‘Obtain the inverse z transform of 40° Gee) by using the inversion integral method Notice that HO = GEE For k =0,1,2, .,X(z)z*"* hasa simple pole at z = 2 “T and a double pole at 2a a= 1 Hence, from Equation (2-24), we obtain 7 residue of -— ary 7 [ @-Ve= “hth xh o, Scanned with CamScanner52 ‘The z Transform — Chap. 2 where y= [residue at simple pole 2 = e~"7 Hence, XUN) = K+ Kee k mT-¢ 2-6 z TRANSFORM METHOD FOR SOLVING DIFFERENCE EQUATIONS Difference equations can be solved easily by use of a digital computer, provided the ‘numerical values of all coefficients and parameters are given. However, closed-form expressions for x(k) cannot be obtained from the computer solution, except for very special cases. The usefulness of the z transform method is that it enables us to obtain the closed-form expression for x(k) Consider the linear time-invariant discrete-time system characterized by the following linear difference equation: x(k) + aix(k — 1) +--+ + aga(k =n) = boul) + Buk 1) $+ bulk =n) (2-27) where w(K) and x(k) are the system’s input and output, respectively, at the Ath iteration. In describing such a difference equation in the z plane, we take the z transform of each term in the equation. Let us define Zx(K)] = X@) Then x(k + 1).x(k + 2).x(k +3)... and x(k ~ 1),2(k ~ 2),x(k ~ 3), canbe expressed in terms of X(z) and the initial conditions. Their exact z transforms were derived in Section 2~4 and are summarized in Table 2-3 for convenient reference. Next we present two example problems for solving difference equations by the 2 transform method Scanned with CamScanneryt ay Z (xe -2)| -2°X) ay Yle ve) vey 7 eth) P)eGand jer xw- Let) 3 Sec. 2-6 2 Transform Method for Solving Difference Equations 53 TABLE 2-2. 2 TRANSFORMS OF xlk + m) AND x(k ~ mb Diserete function 2 Transform x(k +4) 2 X(z) — z*x(0) — 2°x(1) ~ 27x@) ~ x) xe +3) PX(e) ~ Px(0) — F2(1) ~ 2) x(k +2) 2°X(z) — 27x(0) ~ zx(1) x(k +1) 2X (z) ~ 2x(0) #®) XG) xe-1) xe) x(k - 2) 27X(2) x(k — 3) 2X) xe~ 4) rx@) Example 2-18 Solve the following difference equation by use ofthe z transform method: Mk+2)+3x(k +1) 2K) =0, x)=, x()=1 First note that the 2 transforms of x(k +2), x(k + 1), and x(&) are given, respectively, by Zia(k + 2))= FX) ~ 2x(0 ~ x) fall + Y] = 2x(@) ~ 2x00) ZEx(h)] = X@) Taking ther transforms of both sides of the given difference equation, we obtain P(e) ~ Zx(0) ~ 2i(1) + 32X(z) ~ 3ex(0) + 2X(2) = 0 Substcuting the iil data an simpliying gives XG)" 35HFI" ETDETD FHT F427 etd Tee" 14a Noting that co ids we have x0) =(CI= (CI, k= 01,2, Example 2-19 ‘Obtain the solution of the following difference equation in terms of x(0) and x(1): x(k +2) + (@ + B)e(k +1) + abx(k) = 0 ‘where a and b are constants and k= 0, 1,2, Scanned with CamScanner54 ‘Thez Transform Chap. 2 The z transform of this differenee equation ean be given by [27X(2) ~ 27x(0) ~ ex(1)] + (@ + 6)[zX(z) ~ 2x(0)] + adxX(2) = 0. [22+ (@ + bye + abjX(e) = [2 + (a + b)z]x(O) + 2x(1) Solving this last equation for X(2) gives (2+ @ + ek + XQ) XQ) Sy ee be tab Notice that constants a and b are the negatives ofthe two roots of the characteristic ‘equation, We shall now consider separately two cases: (a) a # b and (b) a = b. (@) Forthecase where a + b, expanding X(2)/z into partial fractions, we obtain, (2) _ bx) +x0)_1 ae +xQ) 1 4, z b-a zta* a-b +b" from which we get X(e) -BO+x@) 1, eX 4x0) boa ita a-b +b ‘The inverse : transform of X(z) gives a(R) = s+ 20) (nyt SOO ry, ae where k = 0,1,2,-.. (b) For the case where ®, the z transform X(z) becomes 2) , sfax(0) + x01)) Feat @ray (0), [ax(0) + x(a))e"* Tea? Gray ‘The inverse 2 transform of X(2) gives x(k) = (O(a + far(0) + CQ K(-ay", a = where k= 0,1,2, 2-7 CONCLUDING COMMENTS In this chapter the basic theory of the z transform method has been presented. The 2 transform serves the same purpose for linear time-invariant discrete-time systems asthe Laplace transform provides for linear time-invariant continuous-time systems. ‘The computer method of analyzing data in discrete time results in difference equations. With the z transform method, linear time-invariant difference equations can be transformed into algebraic equations. This facilitates the transient response analysis of the digital control system. Also, the z transform method allows us to use Scanned with CamScanner60 The z Transform — Chap. 2 eat To 2 X(2) = zt em mate dy) + Brot + 1628 + 322 + EP ANE YEH GEE wwe find 0, x(k) = 41, hy, Problem A-2-8 (Obtain the inverse 2 transform of X(2) Solution The inverse 2 ransform of 277/(1~ 2" is not available from most z transform tables. It is possible, however, to write the given X(z) as a sum of z transforms that are commonly available in z transform tables, Since the denominator of X(z) is | (1 = 2") and the z transform of k? is 274(1 + 2~(1 ~ 27!)*, let us rewrite X(z) as Gary a-ey or ery rt G-ry) G-Fy O-7 from whieh we obtain the following partial fraction expansion: z afetate) oz TF PrAG j “The = transforms of the two terms on the right-hand side of ths last equation can be found from Table 2-1 Thus, la ¥ It is noted that ifthe given X(2) is expanded into other partial fractions then the inverse z transform may not be obtained —SS_Asanatteraive approach, the inverse z transform of X(z) may be obtained by “ise of the inversion integral method First, note that ey J (make, k= 012, x(a) Hence, for k = 0,1,2...X(2)s*“" hasa tripe pole at z = 1. Referring to Equation (2-28), we have sine [rime ee] =r Scanned with CamScannerChap. 2 Example Problems and Solutions a é a ninial ee 7 ings) 4 ggeny i ge (fe) im (aC ~ 12] 2 1 “ = 5k 1), k= 0,12, Problem A-2-9 Using the inversion integral method, obtain the iverse z transform of ___0 ¥0) = GE =H Solution Note that sos" XO Ga NE=D For k=, notice that X(2)z*" becomes a = Xen =Gajeame K9 Hence, for k = 0, X(z)z*~* has three simple poles, z= 2) = 1, z= 2: = 2, and z 2) = 0. For k = 1,2,3,-.. however, X(z)2*"" has only two simple poles, 2 and z= 2; = 2. Therefore, we must consider x(0) and x(k) (where k = 1,2,3,...) separately. Fork =O. For this case, refersing to Equation (2-24), we have : 10 (0) = 3 residue of —— 12 at pote « = (0) 3 idue of aye = ye | Kt GEK where K, = [residue at simple pole z = 1] -tal-ee—ae=ad--° ‘= [Fesidue ot simple pole 2 = 2} s -9— 7 vale Ie-Ne- dl y= residue at simple pole 2 = 0] 10 7 sales =Ne= | ? Scanned with CamScanner‘Thez Transform — Chap. 2 Hence, 20) = Ki + K+ Kye 10454520 Fork =1,2,3,... For this case, Equation (2-24) becomes <= 10z*“ 1th) = 3 ee of te ates = 2] = KK where residue at simple pole z = 1 : et “3 [e- Dez He=D, ae a= [resive at simple pote «= 2] 10z*-" - : tele - ~OE= NER ]= 12 , “Ths, x() = Ki + Ky -10 + 102") = 102" 1), 1,23, Hence, the inverse z transform of the given X(z) can be written i= [ognr ey, Eas, ‘An alternative way to write x(&) for k = Os x(k) = Sb(k) +1024" — 1), k= 1,2, where &.(K) is the Kronecker delta function and is given by aoo~ {i SETS (Obtain the inverse z transform of X(2) = en (2-30) by use of the four methods presented in Section 2-5 Solution Method 1 Diret division method We rst rewrite X(2) asa ratio of two polynomiats 142 xo) = 2 142s" Trae Dividing the numerator by the denominator, we get N(e)= 14427! #1277 +1027 Hence, x(0) 1 Scanned with CamScannerChap. 2 Example Problems and Solutions a x(=4 xQ)=7 xQ) = 10 ‘Method 2: Computational method (MATLAB approach). X(z) can be written as xe Hence, the inverse = transform of X(z) can be obtained with MATLAB as follows: Define num = (1 2 0] den={1 =2 1 Irthe values of x(k) for k = 0,1,2, input as follows: 30 are desired, then enter the Kronecker delta [1 zeros(1,30)) Then enter the command lter(num,den,u) ‘See MATLAB Program 2-3, (The screen will show the output x(k) from k = 0 10 4 = 30,] (MATLAB computations begin from column 1 and end at column 31, rather (MATLAB Program 2-3, pum ={1 2 ol; den=(l 21; zeros 30); x fiter'oumden,a) CCohimes 1 through 12 : 147.0013 6 9 ms mH Columns 13 trough 24 37 as a6 ay 82. 581g 7.70 Columns 25 through 31 7 76 79 62 85 oH Scanned with CamScannerThe z Transform — Chap. 2 than from column Oto column 30.) The values x(k) give the inverse z transform of X(z). That is, x0) =1 x)y=4 xQ)07 (30) = 91 ‘Method 3: Parttalfraction-expansion method We expand X(z) into the following partial fractions: wey Dare ‘Then, noting that we obtain 20-1 HG) =3k41, 2 1,2,3, which can be combined into one equation a follows: HQ) S341, k= 01,2, [Note that if we expand X(z) into the following partial factions ot4 ee Se eH X@)e erty ap ty then the inverse z transform of X(z) becomes x(0) =1 x(k) = 44 3(k = 1) = 3k +1, XA) 3K AT, k= 01,2, ‘which is the same as the result obtained by expanding X(z) into the other partial fractions. [Remember that X(2) can be expanded into different partial fractions, but the final result for the inverse 2 transform is the same ] Method 4 Inversion integral method. First, note that 2 + 2)2* xe = GARE Scanned with CamScannerChap. 2 Example Problems and Solutions 6 Fork = 0,1,2, . .X(2)*“!has a double pole atz = 1. Hence, referring to Equation (2-24), we have x(t) [esta or £2 ar dose pote = = ] Thus, a (2 +2) 10 gi age - ye taz] = tine +24 Bet, k= 0,1,2, Problem A-2-11 Solve the following difference equation: 2e(k) ~ x(k = 1) + x(k - 2) ‘where x(k) = 0 for k <0 and {i ke 0, <0 Ak) atk) Solution By taking the z transform of the given difference equation, 1 2X(z) ~ 2X) + FXG) Solving this last equation for X(z), we obtain 1 T-Pa-e set Ge Expanding X(2) into partial fractions, we get Xe) tee T* imei +2 Tw Fe Notice that the two poles involved in the quadratic term in this last equation are complex conjugates. Hence, we rewrite X(2) as follows: 1A _t=ose* 1 ase +052 Os xX) XG) = [By referring to the formulas for the z transforms of damped cosine and damped sine functions, we identify e~#" = 0.5 and cos wT = V2 for this problem. Hence, we get wT = id, sin wT = U'VG, and e*” = U'V2. Then the inverse z transform of X(z) can be written as x(k) = 1 Le cosukT + he sin ak from which we obtain x(Q) =05 x) =1 Scanned with CamScanner‘Scanned with CamScannerZl st mde a me _ ZH(0) - 2" X41) Za Plo D ye Pte cz KC) -\ cae Xe) #* Bes Boe eo (2+) (ar?) bs (kty= Kirke : r z > : ae ee aw is ae >t e K e Kie GI) : , . < —_ ‘ee Le d- aa GBA) ‘We shall define a train of unit impulses as 5,(¢), or Bre) = De — ‘The sampler output is equal to the product of the continuous-time input x(¢) and the train of unit impulses 3,(¢). Conseque the sampler may be considered a modu- lator withthe input x(0) as the modi n of w 8) as thie carrier, as shown in Figure 3 xin) o xt a & xe Figure 3-1 Impulse sampler Scanned with CamScanner6 .2-Plane Analysis of Discrete-Time Control Systems Chap. 3 carir ai Modulating ml i Ly eter i Figure 3-2 Impulse sampler 28 a modulstor Next, consider the Laplace transform of Equation (3-1): 28) = Be" (O] = x(OVLIBLO) + x(TIELEC- TH] +x(2T)¢[s(e — 27) + = x(0) + x(T)e"™ + x(2T)e% + «++ = DxkN¢ Vv G2) Notice that if we define ~~ / or then Equation (3-2) becomes £0) ne = DADE 6) ‘The right-hand side of Equation (3-3) is exactly the same as the right-hand side of Equation (2-1): Itis the z transform of the sequence x(0),x(7),x@27),... gener- ated from x(#) at = kT, where k= 0,1,2,.... Hence we may write =x) > Xo 2X) ‘and Equation (3-3) becomes , lecuryms x5) leas ne oa X*G)| —s Note that the variable zis a complex v be stressed that the notation X(z) does not si rather X*(5 = T-" nz) } period. [Itshould (5) with s replaced by z, but Scanned with CamScannerSec. 3-2 Impulse Sampling and Data Hold n ‘Summary. Let ussummarize what we have juststated. Ifthe continuous-time signal x(¢) is impulse sampled in a periodic manner, mathematically the sampled signal may be represented by #0 = Zxcost - kr) In the impulse sampler the switch may be thought of as closing instantaneously every sampling period and generating impulses x(k7)8(t ~ kT) Such a sampling pro- cess is called impulse sampling. The impulse sampler is introduced for mathematical convenience; itis a fictitious sampler and it does not “The Laplace transform ofthe impulse-sampled signal x") has been shown 10 bbe the same as the z transform of signal x(¢) if e” is defined as z, or e” = z Data-Hold Circuits. In a conventional sampler, a switch closes to admit an input signal every sampling period 7. In practice, the sampling duration is very short in comparison with the most sigan ie constant of the plant. A sampler ‘converts a continuous-time signal into a train of pulses occurring at the sampling instants ¢ = 0,727, ..., where Tis the sampling period. (Note that between any two consecutive sampling instants the sampler transmits no information. ‘Two signals whose respective values atthe sampling instants are equal will give rise to the same sampled signal.) Data-hold is a process of generating 2 continuous-time signal h(é) from a discrete-time sequence x(kT). A hold circuit converts the sampled signal into a continuous-time signal, which approximately reproduces the signal applied to the sampler. The signal h(?) during the time interval kT = 1 < (k + 1)Tmay be approx- mated by a polynomial in 7 as follows: MKT + 9) = 447" + ay where 0 = 7 at - wn] Feb bem s omy Since xo = a0) = [0S 50 - en] wwe see that X*(s) is the Leplace transform of the product of two time functions, x(0) and 3i-06(¢ — kT), Note that itis nat equal tothe product of the two corresponding Laplace transforms v ‘The Laplace transform of the product of two Laplace transformable functions {f@) and g() can be given by . spaatol= [Aastnenee can eae 1) = Ag Ftevee - pap [For the derivation of Equation (3-19), see Problem A~3~4,] Let us substitute x(t) and Si-o6(¢ ~ kT) for f(t) and g(¢), respectively. Then the Laplace transform of X"(3), where xe) geo S30 - xn plat itn i owe 3-7 tmpuse wpe Scanned with CamScannera ‘Plane Analysis of Discrete-Time Control Systems Chap. 3 1 w= Ab can be given by le) - d]glee |= We > % 13" X*(5) = £}x() D a(t - kT) LC ea a ‘ feo] OE oo) wv G-2) be ( Ande me SS where the integration is along the line from c — jo>to.¢ + joo and this line is parallel to the imaginary axis in the p plane and separates the poles of X(p) from those of {1 = e°7¥°P)], Equation (3-20) is the convolution integral. It is a well-known fact that such an integral can be evaluated in terms of residues by forming a closed contour consisting ofthe line from ~ j=toc + jeeand semicircle of infinite radius in the left or right half-plane, provided that the integral along the added semicircle is a constant (zero or a nonzero constant). That is, we rewrite Equation (3-20) as follows: x) ~Pal, MOT Bahr em P ~ i ee /G-2) or rele (c%- 2a) céctlu = Seateryele = 5 where Mis a semicircle of infinite radius in the left or right half p plane. ‘There are two ways to evaluate this integral (one using an infinite semicircle in the left-half plane and the other an infinite semicircle in the right half-plane); we shall describe the results obtained by the two cases separately. Tn our analysis here, we assume that the poles of X(s) lie n the let half-plane and that X(s) can be expressed as 2 ratio of polynomials in s, as) Ps) where q(s) and p(s) are polynomials in s. We also assume that p(s) is of a higher degree in s than q(s), which means that lim X(5) = 0 XG) = Evaluation of the Convolution Integral in the Left Half-Plane. We shall evalu- ate the convolution integral given by Equation (3-21) using a closed contour in the left half of the p plane as shown in Figure 3-8. Using this closed contour, Equation (3-21) may be evaluated as follows: Noting that the denominator of X(s) is of higher degree in 5 than the numerator, the integral along Ii, vanishes. Hence, 1 § x) 2a] | dp . (5) ‘This integral is equal to the sum of the residues of X'(p) in the closed contour. xw=d [resi of Lat pole of X( 2)| 6-22) Scanned with CamScannerSec. 3-3. Obtaining the z Transform by the Convolution integral Method 85 x plane fe Figure 3-8 Closed contour in the lett half of the p plane [Refer to Problem A-3~6 for the derivation of Equation (3-22).] By substituting z for e” in Equation (3-22), we have x)= eof KPI ac pote ot x¢p)| By changing the complex variable notation from p to s, we obtain 3 x@)=5 [residue ot 2G at pote of xe] (3-23) ze Assume that X(s) has poles si, 5:,..-45m. Ifa pole at 5 = sis a simple pole, then the corresponding residue Ki is = lim [« = 5): #0) (3-24) Ifa pole at s = sis a multiple pole of order m, then the residue K; is od img — ayn XO . Keg tinge [c sy Ze al @-25) Example 3-1 Given L 20-35 obtain X(2) by use of the convolution integral in the left half-plane Scanned with CamScanner86 zPlane Analysis of Discrete-Time Control Systems Chap. 3 Note that X(s) has @ double pole at s = O and a simple pole at s = ~1. Hence, Equation (3-23) becomes, x0) = 3 [rite of 22% wpa orx0)] wim TE, “ig ere eT Evaluation of the Convolution Integral in the Right Half-Plane. Let us next evaluate the convolution integral given by Equation (3-21) in the right half of the plane. Let us choose the closed contour shown in Figure 3-9, which consists of the 1e from ¢ — je to c + je and Tp, the portion of the semicircle of infinite radius in the right half of the p plane that lies to the right of this line. The closed contour encloses all poles of if ~ e~®"?)], but it does not enclose any poles of X(p). Now X*(5) can be written as, Figure 3.9 Closed contour in the right half ofthe p plane Scanned with CamScannera8 _z-Plane Analysis of Discrete-Time Control Systems Chap. 3 Henee, 1 XG Jat) = 2S x0 + jank + nk) # 2x00) Let k + = m Then this last equation becomes rotmted Say rmadt hone ‘Therefore, we have XG) = XS jk), k= 01,2, ‘Thus, X*(6) is periodic, with period 2n/ay. This means that, if a function X(3) has a pole at s = 5; in the s plane, then X°(s) has poles at s = s, = jayk (k= 0,1,2,...) Obtaining 2 Transforms of Functions Involving the Term (1 ~ e~")/s. We shall here consider the function (5) involving (1 — ¢~?)/'s. Suppose the transfer function G(s) follows the zero-order hold. Then the produit of the transfer function of the zero-order hold and G(s) becomes 6) G(s) “ In what follows we shall obtain the 2 transform of such an X(s). Note that X(s) can be written as follows: $ G-30) where Consider the function (5) G-31) Since X;(5) is a product of two Laplace-transformed functions, the inverse Laplace transform of Equation (3~31) can be given by the following convolution integral: anf) = [edt ~ eseddr where! edt) = Ee] = 8-7) a) = EG] Thus. w= [9-7 ~ meyer sae T) Scanned with CamScanner‘See. 3-3 Obtaining the z Transform by the Convolution Integral Method 89 Hence, by writing Z1g(0] = Ge) the z transform of x,(¢) becomes Zln(Ol = Zia ~ N= Ge) Referring to Equations (3-30) and (3~31), we have X(2) = Z(G) - eG) = Ze) ~ Ztel0] = Ge) = FG) = 0-2 96@) or X(@) = 21K@)) = 0-292] [2 6-22) ‘We have thus shown that if X(s) involves a factor (1 — e~™) then, in obtaining the 2 transform of X(s), the term 1 ~ e~™ = 1 — 2~* may be factored out so that X(z) ‘becomes the product of (1 — 2~) and the z transform of the remaining term. ‘Similarly, ifthe transfer function G(s) follows the first-order hold G(s), where Gls) = ( then the z transform of the function XG) = ( ‘can be obtained as follows. Since ane xu) =a - ey EtG. by employing the same approach as used in obtaining Equation (3-32), we have [e-em SHe0] X(z) = -(- eye [Be 2609] 6-33) Equation (3-33) can be used for obtaining the z transform of the function involving the first-order hold circuit. Example 3-3 ‘Obtain the 2 transform of Scanned with CamScanner90 2-Plane Analysis of Discrete-Time Control Systems Chap. 3 Referring to Equation (3-32), we have 3-4 RECONSTRUCTING ORIGINAL SIGNALS FROM SAMPLED SIGNALS Sampling Theorem. If the sampling frequency is sufficiently high compared with the highest-frequency component involved in the continuous-time signal, the amplitude characteristics of the continuous-time signal may be preserved in the envelope of the sampled signal. To reconstruct the original signal from a sampled signal, there is a certain minimum frequency that the sampling operation must satisfy. Such a minimum frequency is specified in the sampling theorem. We shall assume that a continuous- time signal x(0) has a frequency spectrum as shown in Figure 3-10. This signal x(t) oes not contain any frequency components above w, radians per second Sampling Theorem. . If w,, defined as 2m/T,, where Tis the sampling period, fs greater than 204, or 0, > uy where « is the highest-frequency component present in the continuous-time sig- nal x(¢), then the signal x(?) can be reconstructed completely from the sampled signal x°(0), ‘The theorem implies that if a, > 2w, then, from the knowledge of the sampled signal, itis theoretically possible to reconstruct exactly the original continuous-time signal. In what follows, we shall use an intuitive graphical approach to explain the sampling theorem. For an analytical approach, see Problem A~3-10. ct To show the validity of this sampling theorem, we need to find the frequency Daye Tay Oe To Figure 3-10 feequency spectrum Scanned with CamScannerSec 3-5 The Pulse Transfer Function 103 Note that G(z) is also the z transform of the system’s response to the Kro- necker delta input: 1 ke (KT) = (KT) = {o: fork eo ‘Since the z transform of the Kronecker delta input is, X(z) = Dx(kT)z* = 1 then, referring to Equation (3-44), the response ¥(z) to the Kronecker delta input is ¥(@) = G@) “Thus, the system's response to the Kronecker delta input is G(z), the z transform of the weighting sequence. This fact is parallel to the fact that G(s) is the Laplace transform ofthe system's weighting function, which isthe system's response to the unit-impulse function, Starred Laplace Transform of the Signal Involving Both Ordinary and Starred Laplace Transforms. Tn analyzing discrete-time control systems, we often find that some signals in the system are starred (meaning that signals are impulse sampled) and others are not, To obtain pulse transfer functions and to analyze discrete-time control systems, therefore, we must be able to obtain the transforms of output signals ff systems that contain sampling operations in various places in the loops. ‘Suppose the impulse sampler is followed by @ linear continuous-time element whose transfer function is G(s), as shown in Figure 3-23. In the following analysis sve assume that al intial conditions are zero inthe system. Then the output Y(s) is ¥(s) = G()XMs) (3-45) Notice that ¥(s) is a product of X*(s), which is periodic with period 27/u,, and G(s), which is not periodic. The fact that the impulse-sampled signals are periodic can be seen from the fect that XG) =X # fork) = 0,12). (3-46) (See Example 3-2.) In the following ve shall show that in taking the starred Laplace transform of Equation (3--45) we may factor out X*(s) so that ¥*(5) = [G()X*(6)]" = [G()]*X"G) = GSAS) G47) ‘This fact is very important in deriving the pulse traisfer function and also in simplifying the block diagram of the discrete-time control system. “To derive Equation (3-47), note that the inverse Laplace transform of ¥(s) given by Equation (3-43) can be written as follows: ia ar" fa or “ Figure 3-23 Impulse-sampled system Scanned with CamScanner18 Plane Analysis of Discrete-Time Control Systems Chap. 3 V9 = LOOX] [few - nerenar fee - DE x ~kI)dz = E [let - nxcnace— ener = Bae -ereuer) Then the z transform of y(1) becomes ¥@)=Z0]=3[Ze0r - ensan|e = 3 Sammreceneo where m ~ k. Thus, Ye) = E gare" Baer )e* = G(z)X(z) G-48) Since the z transform can be understood as the starred Laplace transform with replaced by z, the z transform may be considered to be a shorthand notation for the starred Laplace transform. Thus, Equation (3-48) may be expressed as ¥*(5) = G*5)X"G5) which is Equation (3-47). We have thus shown that by taking the starred Laplace transform of both sides of Equation (3~45) we obtain Equation (3-47). ‘To summarize what we have obtained, note that Equations (3-45) and (3-47) state that in taking the starred Laplace transform of a product of transforms, where some are ordinary Laplace transforms and others are starred Laplace transforms, the functions already in starred transforms can be factored out of the starred Laplace transform operation. It is noted that systems become periodic under starred Laplace transform operations. Such periodic systems are generally more complicated to analyze than the original nonperiodic ones, but the former may be analyzed without difficulty if carried out in the z plane (that is, by use of the pulse-transfer-function approach). General Procedures for Obtaining Pulse Transfer Functions. Here we shall present general procedures for obtaining the pulse transfer function of a system that has an impulse sampler at the input to the system, as shown in Figure 3~24(a). ‘The pulse transfer function G(z) of the system shown in Figure 3-26(a) is Ye) XG) = GG) = Z{G6)] Scanned with CamScannerSec. 3-5 The Pulse Transfer Function 105, on wT oa ve 4, ial Arie vied Fictitious sample ot xe) an ve. xia 7 Ye Figure3-24_ (a) Continuoussime system with an impulse sampler at the inp () continuous-time sytem, Next, consider the system shown in Figure 3-24(b). The transfer function G(s) is siven by Ys) XG) ‘The important fact to remember is that the pulse transfer function for this system is not Z[G(s)], because of the absence of the input sampler. The presence or absence of the input sampler is crucial in determining the pulse transfer function of 2 system, because, for example, for the system shown in Figure 3-24(a), the Laplace transform of the output y(F) is ¥(s) = G(s)X"(3) Hence, by taking the starred Laplace transform of ¥(s), we have ¥*(3) = G(s) X05) or, in terms of the z transform, ¥(2) = G@)X(z) while, for the system shown in Figure 3-24(b), the Laplace transform of the output HO is = G6) l YG) = GX) which yields ¥*6) = (COXON = (EXO Scanned with CamScanner108 2Plane Analysis of Discrete-Time Control Systems Chap. 3 cr, in terms of the 2 transform, ¥@) = Z1YG)) = Z1G(X()] = ZIGXW)] = GXG) # G)XW The fact that the z transform of G(s)X(s) is not equal to G(z2)X(z) will be discussed in detail later in this section. In discussing the pulse transfer function, we assume that there isa sampler at the input of thie element in consideration. The presence or absence of a sampler at the output ofthe clement (or the system) does not affect the pulse transfer function, because, if the ssmpler is not physically present at the output side of the system, i is always possible to assume that a fictitious sampler is present at the output. T means that, although the output signal is continuous, we can consider the Values of the output only at ¢ = kT (k= 0,1,2,...) and thus get sequence y(k7), Note that only for the case where the input to the system G(s) is an impulse- sampled signal is the pulse transfer function given by G2) = Z[G()] Examples 3-4 and 3-5 demonstrate the methods for obtaining the pulse transfer function. Example 3-4 Obtain the pulse transfer function G(z) of the system shown in Figure 3-24(a), where G(6) is given by 2 ra G6) = "Note that theres a sampler atthe input of G(s) and therefore the pulse transfer function is Ge) = 216) Method 1. By referring to Table 2-1, we have Henee, Method 2. The impulse response function for the system is obtained as follows: 8) = £1G(S)] =e" Hence, Therefore, GG) = Ear = Sew: Scanned with CamScanner| (9-7 ‘Sec. 3-5 The Pulse Transfer Function 107 Example 3-5 Obtain the pulse transfer function of the system shown in Figure 3-24), where G(s) is given by ny 60) = FD Note that there is a sampler at the input of G(s. Method I G(s) involves the erm (1 ~ e~"); therefore, referring to Equation (3-32), ‘we obtain the pulse transfer function as follows: Gia) = Z1G(91 =z] fo =e?) ns a -a- olay GF sl peseh From Table 2-1, the z transform of each ofthe partiat-raction expansion terms can be found Thus, coy Ge) =~ pol 49) ‘Therefore, by taking the inverse Laplace transform, we have the following impulse response function: ae -[r HT ee (ET) = AT = Le eM) = f= TOK NT) ee Petr, . a{eet) 7 fg kao @ Ten the pulse transfer fnstion (2) canbe obtained as flow bona Scanned with CamScanner108 Plane Analysis of Discrete-Time Control Systems Chap. 3 Pulse Transfer Function of Cascaded Elements. Consider the systems shown in Figures 3-25(a) and (b). Here we assume that the samplers are synchronized and have the same sampling period. We shall show that the pulse transfer function of the system shown in Figure 3-25(a) is G(2)H(z), while that shown in Figure 3-25(b) is Z(G(s)H()] = Z[GH(s)] = GH(z), which is different from G(z)H(2). Consider the system shown in Figure 3~25(a). From the diagram we obtain UG) = G6)X"@) ¥(s) = HU)U() Hence, by taking the starred Laplace transform of each of these two equations, we get UG) = GS) ¥*(5) = H°@)U"G) Consequently, Y*(s) = H*G)U"(5) = HG)GO)X"(S) or ¥*(6) = G*G)H)A*(S) In terms of the transform notation, ¥(2) = Ge)H@)X(2) ‘The pulse transfer funetion between the output y*(f) and input x*(t is therefore given by 22). Gene FOy7 S@He) ee an Lt a it Pigg vt io st vo Toy Te Le ria by br » Figure 3-25 (a) Sampled system witha sampler between cascaded elements G(s) and H(s): (}) sampled system with no sampler Botween cascaded elernems Gis) and #15) Scanned with CamScannerSec. 3-5 The Pulse Transfer Function 109 Next, consider the system shown in Figure 3-25(b) From the diagram we find ¥(s) = G(s)H(s)X"(65) = GH()X"(5) where i GH{(s) = G(s)HG) Taking the starred Laplace transform of ¥(s), we have ¥"(3) = [GH(s)PX"(5) In terms of the z transform notation, ¥(z) = GH(2)X(z) ‘and the pulse transfer function between the output y*(¢) and input x(t) is Y(z) YO) Gre) = Z[GH Fo 7 HE) = ZICH) Note that G(@)H(z) # GH) = Z1GH(s)) Hence, the pulse transfer functions of the systems shown in Figures 3-25(a) and (b) are different. We will now verify this statement in Example 3-6, Example 3-6 Consider the systems shown in Figures 3-26(a) and (b). Obtain the pulse transfer function ¥(zV/X(z) for each of these two systems. ain) so fo Lew vo fa we i La ty az Gia ar 0 xe) ate 1 1 ve 1 me me oie Wer o Figure3-26 (a) Sampled system witha sampler berween clement G(s) = U(s + 2) and H(a) = is + B};(b) sompled system with no sampler between elements G() and HU) Scanned with CamScanner10 z-Plane Analysis of Discrete-Time Control Systems Chap. 3 or the system of Figure 3-26(0), the two transfer functions Gs) and H() are sepucund oy ergs meer tel ints sengiey rears tears zeit same sampling pero. The pute transfer function for this stem i Ye) , Ye) UE) X(z) Ul) X@) (2)G(2) = G(e)HG) Henee, Ye 1 1 HB- oom =o) (a) eer For the system shown in Figure 3-26(0), the pulse transfer function Y(2)X(2) is obtained as follows: FB -ztowuen 2 fs tarts, i] oly (45-7h)} Henee, YO _ XG) ‘Clearly, we see thatthe pulse transfer functions ofthe two systems are different; that = GHz) = G(e)H(2) # GH) ‘Therefore, we must be careful fo observe whether or not there is a sampler between cascaded elements Pulse Transfer Function of Closed-Loop Systems. In a closed-loop system the existence or nonexistence of an output sampler within the loop makes a difference in the behavior of the system. (If there is an output sampler outside the loop, it will ‘make no difference in the closed-loop operation.) Consider the closed-loop control system shown in Figure 3-27. In this system, the actuating error is sampled. From the block diagram, E(s) = RG) ~ H(S)C@) C(s) = GWE") 0 w et cle la) eu 0 Fain mat Figure 3-27, Closed-toop contol system Scanned with CamScannerSec, 3-5 The Pulse Transfer Function m Henee, £(s) = R(s) ~ H)GS)EG) ‘Then, by taking the starred Laplace transform, we obtain EM(s) = RY) — GHY()E) or EX) = TF GHG) since CW = GWE we obtain or) « FNRS CO) TF GHG) In terms of the z transform notation, the output can be given by c() = SFO + GHG) The inverse z transform of this last equation gives the values of the output at the sampling instants. [Note that the actual output c(*) ofthe system isa continuous-time signal. The inverse z tcansform of C(z) will not give the continuous-time output c()] The pulse transfer function for the present closed-loop system is @)___ GG) RG) 1+ GHE) 6-5) Table 3-1 shows five typical configurations for closed-loop discrete-time con- trol systems. Here, the samplers are synchronized and have the same sampling period. For each configuration, the corresponding output C(z)is shown. Notice that “Somme discrete-time closed-loop control systems cannot be represented by C(z)/R(z) (that is, they do not have pulse transfer functions) because the input signal R(s) cannot be separated from the system dynamics. Although the pulse transfer function may not exist for certain system configurations, the same techniques discussed in this, ‘chapter can still be applied for analyzing them, Pulse Transfer Function of a Digital Controller. ‘The pulse transfer function of a digital controller may be obtained from the required input-output characteristics of the digital controller ‘Suppose the input t0 the digital controller is ¢(k) and the output is m(k) In general, the output m(&) may be given by the following type of difference equation: mk) + aymm(k ~ 1) + aym(k = 2) +--+ + agm(k - 1) = boe(k) + bye(ke~ 1) +--+ + Dyelk ~ n) Gs Scanned with CamScanner112 z Plane Analysis of Discrete-Time Control Systems Chap. 3 ‘TABLE 2-1. FIVE TYPICAL CONFIGURATIONS FOR CLOSED-LOOP DISCRETE-TIME CONTROL SYSTEMS, at aw Gah tt rete ; - ou, : tn TF Gel “4 a | a) cin ew | ‘B—"—| Gus) 1 1 = Olean | Cine TF etait v a a cn te 46,181] [sat _ Geile CTF qa |W ma Ley ma cst ee cus yen a a os na — wy cia “6 fou mA om of Ce = a Scanned with CamScannermy ch) tCB ‘Scanned with CamScanner; (4-3) Pulse “Nrensfer Funckios of Caxcooleol Elentk sf () | Key, x"CE) ult) yuo C& WG) KG) XG) — aot SO US). Gey. 6) YG) = Hts) - UG) XG) = AG). GG) ¥G) Ye GRO) XG) YW = GHG) ¥8G) Ye) = GHIZ) x%) Beet CH) ACE) Scanned with CamScanneree cy) GH Kirke \ ee K+ - |e (esas) Te ie ay ee Pow 04) Ge) : x ae Rel [6- couw =) 5 —3-b oe co ee aa a eek: i: aa ee Scanned with CamScannerPe eed= Ray Hen. te : Ween. Gi. 6°) a> C= ae E®=> Resy-Hey EG) EO ; Pio O- GAOe ae x es) = oe i | l+GH C$) a Bee Ce OCs). Baise IeGH CSy ers CO Ratis WWE is. SO % \-0tz) ~ 14 GH) Scanned with CamScanner182 Design of Discrete-Time Control Systems by Conventional Methods Chap. 4 plane The tne ls o Figure 4-11 (3) A desirable region inthe «plane fr closed-loop pole locations; (©) corresponding region inthe = plane ‘On the basis of the preceding discussions on mapping from the s plane to the = plane, the desirable region can be mapped to the z plane as in Figure 411(b) ‘Note that if the dominant closed-loop poles of the continuous-time control sys- tem are required to be in the desirable region specified in the s plane, then the domi- ‘nant closed-loop poles ofthe equivalent discrete-time control system must lie inside the region in the z plane that corresponds to the desirable region in the s plane. Once the discrete-time control system is designed, the system response characteristics must be checked by experiments or simulation. If the response characteristics are not satisfac tory, then closed-loop pole and zero locations must be modified until satisfactory results are obtsined Comments. For discrete-time control systems, itis necessary to pay particular attention to the sampling period T- Thisis because, if the sampling period is too long. and the sampling theorem is not satisfied, then frequency folding occurs and the effective pole and zero locations will be changed Suppose a continuous-time control system has closed-loop poles at 5 = —o; = jus inthe s plane. Ifthe sampling operation is involved in this system and if, > fo,, where us is the sampling frequency, then frequency folding occurs and the system behaves as if it had poles at s = —o, + j(an + may), where m = 1,2, 3, .... This means that the sampling operation folds the poles outside the primary strip back into the primary strip, and the poles will appear at s = ~o, + (a ~ w,); see Figure 4-12(a). On the 2 plane those poles are mapped into one pair of conjugate complex poles, as shown in Figure 4~12(b). When frequency folding occurs, oscilla- tions with frequency w, ~ a, rather than frequency a, are observed. 4-3 STABILITY ANALYSIS OF CLOSED-LOOP SYSTEMS IN THE z PLANE Stability Analysis of a Closed-Loop System. In what follows we shall discuss the stability of linear time-invariant single-input-single-output discrete-time control systems. Consider the following closed-loop pulse-transfer function system: Scanned with CamScanner‘Sec. 4-3 Stability Analysis of Closed-Loop Systems in the z Plane 193 0, oT OB na, =o) ye tl Lely “ita, ~ oy) ta w Figure 4-12 (a) Diagram showing s plane poles at ~oy = jen and folded poles appearing ato = fos 3m = lon = 2) (0) lane mapping a plane poles at =o, © fay, = 07 & oy 2 a4), ~ oy = He Gq) T+ GHE) ‘The stability of the system defined by Equation (4-3), as well as of other types of discrete-time control systems, may be determined from the locations of the closed- loop poles in the z plane, or the roots of the characteristic equation, P(z) =1 + GH(z)=0 (4-3) as follows: 1, For the system to be stable, the closed-loop poles or the roots of the charac teristic equation must lie within the unit circle in the z plane. Any closed-loop pole outside the unit circle makes the system unstable. 2, Ifa simple pole lies at z = 1, then the system becomes critically stable Also, the system becomes critically stable if single pair of conjugate complex poles lies on the unit circle in the z plane Any multiple closed-loop pole on the unit circle makes the system unstable 3. Closed-loop zeros do not affect the absolute stability and therefore may be located anywhere in the z plane. ‘Thus, a linear time-invariant single-input-single-output discrete-time closed- Joop control system becomes unstable if any of the closed-loop poles lies outside the unit circle and/or any multiple closed-loop pole lies on the unit circle in the z plane Scanned with CamScanner184 Design of Discrete-Time Control Systems by Conventional Methods Chap. 4 Example 4-2 Consider the elosed-loop control system shown in Figure 4-13. Determine the stability of the system when X = 1. The open-loop transfer function G(s) of the system is 1 1 60) ED Referring to Equation (3-58), the z transform of G(s) is [: mt J 0.36792 + 0.2642 GE) WE GD] ” G - 03675}: — 1) eo Since the closed-loop pulse transfer function for the system is, Ce) _ te) Re "T+ Ge) the characteristic equation is 1+6@)=0 whieh becomes (= = 0.3679)(2 ~ 1) + 0.36792 + 0.2682 = 0 B-2+06321=0 ‘The roots of the characteristic equation are found to be 5054/0681, z= 05~ 706181 Since lal <1 the system is stable Itisimportant to note that in the absence of the sampler a second-order system is always stable. In the presence of the sampler, however, a second-order system such as this can become unstable for large values of gain. In fact, it can be shown that the second-order system shown in Figure 4-13 will become unstable if K > 2.3925. (See Example 4-7) Methods for Testing Absolute Stability. Three stability tests can be applied directly to the characteristic equation P(z) = 0 without solving for the roots. Two of them are the Schur-Cohn stability test and the Jury stability test. These two tests Ast er fi ce cia) Figure 13 Closed-oop contol system of Example 4-2 Scanned with CamScannerSee. 4-3 Stability Analysis of Closed-Loop Systems in the z Plane 185 reveal the existence of any unstable roots (the roots that lie outside the unit citele in the 2 plane) However, these tests neither give the locations of unstable roots nor indicate the effects of parameter changes on the system stability, except for the simple case of low-order systems. (See Example 4-7.) The third method is based on the bilinear transformation coupled with the Routh stability criterion, which will be ‘outlined later in this section. (In Chapter 5 we discuss Liapunov stability analysis, which is applicable to control systems defined in state space.) ‘Both the Schur-Cohn stability test and the Jury stability test may be applied to polynomial equations with real or complex coefficients. The computations re- quired in the Jury test, when the polynomial equation involves only real coefficients, are much simpler than those required in the Schur-Cobn test. Since the coefficients of the characteristic equations corresponding to physically realizable systems ate always teal, the Jury test is preferred to the Schur-Cohn test. The Jury Stability Test. In applying the Jury stability test to a given chara teristic equation P(z) = 0, we construct a table whose elements are based on the coefficients of P(z), Assume that the characteristic equation P(z) is a polynomial in z as follows Uz) = ay2h + ayzt! + Fagan z + dad” (4-5) where a) > 0. Then the Jury table becomes as given in Table 4-1. Notice that the elements in the first row consist of the coefficients in P(z) arranged in the ascending arder of powers of z. The elements in the second row consist of the coefficients of P(z) arranged in the descending order of powers of z ‘The elements for rows 3 through 2n ~ 3 are given by the following determinants: TABLE 4-1 GENERAL FORM OF THE JURY STABILITY TABLE Row Foe of #8 ems 1 Cae 4 2 ee ee rrr 3 bas tyan bene be + & baat 5 on ee e an BP m4 Pe an =3 ew Scanned with CamScanner186 Design of Discrete-Time Control Systems by Conventional Methods Chap. 4 [te mele irl anit Pert, kena. -2 Ips - an [Pm k= 0,12 Note that the last row in the table consists of three elements. (For second-order systems, 2 ~ 3 = 1 and the Jury table consists only of one row containing three elements.) Notice that the elements in any even-numbered row are simply the reverse of the immediately preceding odd-numbered row. Stability Criterion by the Jury Test. A system with the characteristic equation P(z) = 0 given by Equation (4-5), rewritten as Pe) = 092" + att $+ + ages Z + ay where a, > 0, is stable if the following conditions are all satisfied: A. aa] < a0. 2. Player > 0 3. Polen 24 orm en 4 |b anil > lool en-al > leah lal > laud Example 4-3 Construct the Jury stability table for the following characteristic equation: Plz) = aa! + a2? + a2? + asz +a where op > 0. Write the stability conditions Referring to the general case ofthe Jury stability table given by Table 4-1, a Jury stability table for the fourth-order system may be constructed as shown in Table 4-2 ‘This table is slightly modified from the standard form and is convenient for the eompu- tations ofthe b's and c's The determinant given in middle of each row gives the value of b or ¢ written on the right-hand side of the same row The stability conditions are as follows: 1 Jas < ay 2. P()= ae +a, +0: +0 +a >0 3. P(N) =ay— a+ a= a+a>0, n= d= even 4. Io > loa led > lel Scanned with CamScannerSec. 4-3 Stability Analysis of Closed-Loop Systems in the z Plane 137 [TABLE 4-2. JURY STABILITY TABLE FOR THE FOURTH-ORDER SYSTEM [: th? ~ bo 2 o—— 4 bs be . bs a bo bs =| 3 (. | 4 bo—— 4 a e e e ‘tis noted thatthe value of cy (or, inthe ease ofthe th-order system, the value ot qi isnot used in the stability test, and therefore the computation ofc (Or 4s) may bbe omitted. Example 4-4 ‘Examine the stability of the following characteristic equation: Ple) = 24 = 122? + 007s? +032 ~ 008-0 Notice that for this characteristic equation Cleatly, the frst condition, Ja] < op, i sais ied Let us examine the second condition for stability: 1.24007 +03-008=009>0 Pay Scanned with CamScanner188 Design of Discrete-Time Control Systems by Conventional Methods Chap. 4 ia The second condition is satisfied. The third condition for stability becomes PC1)=1412+007~03~008=189>0, Hence the third condition is satistied jWenow consiuct the Jury stability table. Referring to Example 4-3, we compute Ing Yale of, bay and bandos anda The resuitisshown in Table 4-3. (Although {he value of cis shown in the table, cis not needed in the stability test and therefore ‘eed not be computed.) From ths table, we get lbs] = 0.994 > 0.208 = jbl les] = 0.946 > 0.315 = Jeg Ths both parts ofthe fourth condition given in Example 4-3 are satisfied. Since all 0 BAI) <0, ne3= odd [bal > I “The first condition, |as| < ao, is clearly satisfied. Now we examine the second condition of te Jury stability test p(y =1-11-01402=0 ‘Tisindicates thatat last one rootisatz = 1, Therefore, the systems at best rtcally roote ‘The remaining este determine whether the system is critically table or unstable {itt piven characterintie equation represents a contol system, ciical stability will fot be desied. The stability test may be stopped at this point) Te third condition ofthe Jury test gives P(-1) = -1- 1140140. -18<0, 4 “The third condition is satisfied. Now we examine the fourth condition ofthe Tar test Simple computations give by = ~0396 and by = ~0.12. Hence, Wl > Ib “The fourth condition of the Jury ts is satisfied Tom the above analysis we conclude thatthe given characteristic equation has one rooton the unit etele (2 = I) and its other two roots within the ult izle in the Siplane. Hence the system eicallystble Example 4-6 ‘A control system has the following characteristic equatio Pz) =~ 1324 = 0.08 +024 =0 Determine the stability of the system We first identify the coefficients: = odd Scanned with CamScanner190 Design of Discrete-Time Control Systems by Conventional Methods Chap. 4 Cleatly, the first condition for stability, |a,| 0 3 P(-1)>0, \We shal now apply the frst condition for stability. Since = 03679 + 02612K and ay= 1, the fis condition for sailiy becomes (03679 + 0.2642K| <1 or 23905 > K > -5.1775 (4-6) ‘The second condition for stability becomes PCI) = 1 + (0.3679K ~ 1.3679) + 0.3679 + 0.2642K = 0 6321K > 0 which gives K>o (7 The third condition for stability gives PC“) = 1 ~ (0 3679K ~ 1 3679) + 0 3679 + 0 2642K = 2.7358 - 0 1037K > 0 which yields 26 382 > K (4-8) For stability, gain constant K must satisty inequalities (46), (4-7), and (4-8) Hence. 235>K>0 ‘The range of gain constant XK for stability is between U and 2.3925 Scanned with CamScannerSec. 4-3 Stability Analysis of Closed-Loop Systems in the z Plane 181 Ifgain Kis setequal to2 3925, then the system becomes critically stable (meaning that sustained oscillations exist at the output) The frequency of the sustained oscilia- tions ean be determined if 2.3925 is substituted for Ki the characteristic equation and the resulting equation is investigated, With K = 2 3925, the characteristic equation becomes #0487? +1=0 ‘The characteristic roots are at z © 0 2439 + j0 9698 Noting that the sampling period Tis equal to 1 see, from Equation (4~2) we have 9698, 02039 The frequency of the sustained oscillations is 1 3244 radlsee Stability Analysis by Use of the Bilinear Transformation and Routh Stability Criterion. Another method frequently used in the stability analysis of discrete- time control systems is to use the bilinear transformation coupled with the Routh stability criterion. The method requires transformation from the z plane to another complex plane, the » plane. Those who ate familiar with the Routh-Hurwitz stability criterion will find the method simple and straightforward. However, the amount of computation required is much more than that required in the Jury stability criterion The bilinear transformation defined by 13244 radisec paett wel which, when solved for 1, gives +1 weg-t maps the inside of the unit circle inthe z plane into the left half ofthe w plane, This ccan be seen as follows. Let the real part of w be called ¢-and the imaginary part «, so that weotjo Since the inside of the unit circle in the z plane is b= eal = Al «+ or ee IP tat coe we get (otters (o- bP + which yields a<0 Scanned with CamScanner192 Design of Discrete-Time Control Systems by Conventional Methods Chap. 4 ‘Thus, the inside of the unit circle in the z plane (|z| < 1) corresponds to the left half of the w plane. The unit circle in the z plane is mapped into the imaginary axis in the w plane, and the outside of the unit circle in the 2 plane is mapped into the right half of the w plane. (It is pointed out that, although the w plane is similar to the s plane in thatit maps the inside of the unit circle to the left half-plane, itis by no means ‘quantitatively equivalent to the s plane. Therefore, estimating the relative stability of the system from the pole locations in the w plane is difficult.) In the stability analysis using the bilinear transformation coupled with the Routh stability criterion, we first substitute (w + 1)(w — 1) for z in the character- istic equation PU) = ag?" + az ot agz ta 0 as follows: +a,=0 ‘Then, clearing the fractions by multiplying both sides of this last equation by (w ="), we obtain Q(w) = byw" + byw. + baw +b, =0 Once we transform P(z) = 0 into Q(w) = 0, it is possible to apply the Routh stability criterion in the same manner as in continuous-time systems. tis noted that the bilinear transformation coupled with the Routh stability criterion will indicate exactly how many roots of the characteristic equation lie in the right half of the w plane and how many lie on the imaginary axis. However, such information about the exact number of unstable poles is usually not needed in control systems design, because unstable or critically stable control systems are not desired. As mentioned earlier, the amount of computation required in this approach is much ‘more than that required in the Jury stability test. Therefore, we shall not go any further on this subject here We refer the reader to Problem A~4-3, where the present method is used for stability analysis. A Few Comments on the Stability of Closed-Loop Control Systems 1. If we are interested in the effect of a system parameter on the stability of a closed-loop control system, a root-locus diagram may prove to be useful MATLAB may be used to compute and plot a root-locus diagram 2, It is noted that in testing the stability of a characteristic equation it may be simpler, in some eases, to find the roots of the characteristic equation directly by use of MATLAB. 3. Its important to point out that stability has nothing to do with the system's ability to follow a particular input. The error signal in a closed-loop control system may increase without bound, even if the system is stable. (Refer to Section 4-4 for a discussion of error constants.) Scanned with CamScannerSec. 3-5 The Pulse Transfer Function 113 The z transform of Equation (3-51) gives (M(z) + a 2°'M(z) + a2? M(z) + 6+ + a2 MC2) = by E(2) + by2" E(z) + = + baz BZ) (Lt aye + az Pt + a2 *)M(2) = (bo + bz! + + Baz E(2) ‘The pulse transfer function Go(z) of the digital controller may then be given by _M(z) _ bot biz + bya" Gol) = FG) ~ Tarte Fm orn, ‘The use of the pulse transfer function Go(2) in the form of Equation (3-52) enables tus to analyze digital control systems in the z plane. Closed-Loop Pulse Transfer Function of a Digital Control System. Figure 3-28(a) shows a block diagram of a digital control system. Here, the sampler, ‘A/D converter, digital controller, zero-order hold, and D/A converter produce continuous-time (piecewise-constant) control signal u(t) to be fed to the plant Figure 3~28(b) shows the transfer functions of blocks involved in the system. ‘The transfer function of the digital controller is shown as G(s). In the actual “0 cone en omer Peerless aa ew a oie rm a 7 et Jaren, yee we cia na an 7 00 om pe LG a =f ——— oo o Figure 3-28. (a) Block diagram of a digital contol system; (6) equivalent block diagram showing transfer functions of blocks. Scanned with CamScannerm4 2-Plane Analysis of Discrete-Time Control Systems Chap. 3 system the computer (digital controller) solves a difference equation whose input output relationship is given by the pulse transfer function Ge (2) In the present system the output signal c()s fed back for comparison with the input signal (0) The error signal e(t) = r(0) ~ e(1)ssampled, and the analog signal is converted to a digital signal through the A/D device. The digital signal e(JT) is fed to the digital controller, which operates on the sampled sequence e(&7) in some desirable manner to produce the signal m(kT). This desirable relationship between the sequences m(kT) and e(KT) is speci- fied by the pulse transfer function Go(z) of the digital controller. (By properly selecting the poles and ze10s of Gp(z), a number of input-output characteristics cay be generated] Referring to Figure 3-28(b), let us define G(s) = G(s) From Figure 3-28(b), notice that Cls) = G(s)GB(S)E*(s) or (5) = G*(S)GR(S)E*(S) In terms of the = transform notation, Cl) = G(z)Go(z)E(z) Since E(z) = R(z) ~ C2) we have Cz) = Go(z)G(2)[R(2) - CY) and, therefore, GB) __ Golz)G@ RG) ~ 1 Go@)GG) Equation (3-53) gives the closed-loop pulse transfer function of the digital control system shown in Figure 3-28(b). The performance of such a closed-loop system can be improved by the proper choice of Go(z), the pulse transfer function ofthe digital controller. We shall later discuss & variety of forms for Gp(z) to be used in obtaining optimal performance for various given performance indexes In the following, we shall consider only a simple case, where the pulse transfer function Go(z) is of the PID (proportional plus integral plus derivative) type Pulse Transfer Function of a Digital PID Controller. The analog PID control scheme has been used successfully in many industrial control systems for over half a century. The basic principle of the PID control scheme is to act on the variable to be manipulated through a proper combination of three control actions: propor. tional control action (where the control action is proportional to the actuating error (3-53) Scanned with CamScannerSee. 3-5" The Pulse Transfer Function 115 signal, which is the difference between the input and the feedback signal), integral control action (Where the control action is proportional to the integral of the actuating ervor signal), and derivative control action (where the control action is proportional to the derivative of the actuating error signal). ‘Where many plants are controlled directly by a single digital computer (as in ‘a control scheme in which from several loops to several hundred loops are controlled by a single digital controller), the majority of the control loops may be handled by PID control schemes. The PID control action in analog controllers is given by mo) = [ee +E [etna 160 6-54 ‘where ¢(:)is the input tothe controller (the actuating error signal), (is the output of the controller (the manipulating signal), K is the proportional gain, T; is the integral time (or reset time), and Zy is the derivative time (or rate time) ‘To obtain the pulse transfer function for the digital PID controller, we may discretize Equation (3-54), By approximating the integral term by the trapezoidal summation and the derivative term by a two-point difference form, we obtain meer = x{ecer) + F[O42) , AD see DaeGN .. pies e((k = DT) + elkD)] , 5 (kT) — ek - DT) sont MES DD EED « sshD=SB ND or : m(kT) = x{ean + oe Betery - eck ~ on} 6-55) Define Ha OD FH an, yoye0 Figure 3-29 shows the function f(/i7). Then & o((h = 1)T) + eT) 2 2 “iar Taking the z transform of this last equation, we obtain 2[3¢ = an, es sn) Scanned with CamScanner116 Plane Analysis of Discrete-Time Control Systems Chap. 3 0, eile etn) Ota «ry oT a ar Figure 3-29 Diagram depicting function f(h7) Hence, El) & el - TN) + eT] +e 22 2 “Key ‘Then the z transform of Equation (3-55) gives exis Zit mey= K+ i+ ‘This ist equation may be rewritten as fellows: where K, = XZ = integral gain Kp = “= derivative gain Y a ree) 4 Notice that the proportional gain K; for the digital PID controller is smaller than the proportional gain K for the analog PID controller by K,/2. ‘The pulse transfer function for the digital PID controller becomes Golz) -H- Ke+ Ky 7 Kp(l ~ 27 (3-56) ‘The pulse transfer function of the digital PID controller given by Equation (3-56) is commonly referred to as the positional form of the PID control scheme Scanned with CamScannerAtel Sec. 3-5 The Pulse Transfer Function 7 The other form commonly used in the digital PID control scheme is referred to as the velocity form. To derive the velocity-form PID control equation, we ‘consider the backward difference in m(K7), that is, the difference between m(kT) and m((k ~ 1)T), With some assumptions and manipulations, we obtain R@)~ Ce) M(z) = ~KpO(z) + Ky ~ Koll = 2)Cz) 3-57) [For the derivation of Equation (3~57),see Problem A-3-17 ] Equation (3-57) gives, the velocity-form PID control scheme Figure 3-30 shows the block diagram realiza- tion ofthe velocity-form digital PID control scheme. Notice that in Equation (3-57) only the iftegral control term involves the input (z). Hence, the integral term ‘cannot be excluded from the digital controller if the velocity form is used ‘An advantage of the velocity-form PID control scheme is that initialization is not necessary when the operation is switched from manual to jc. Thus, if there are sudden. es in the set point or at the sartof the process operation, the velocity-form PID control scheme exhibits better response characteristics than the positional-form PID control scheme. Another advantage of the velocity-form PID control scheme is that its useful in suppressing excessive corrections in process control systems Linear control laws in the form of PID control actions, in both positional form and velocity form, are basic in digital controls because they frequently give satisfac- tory solutions to many practical control problems, in particular, process control “problems. Noie that, in digital controllers, control laws can be implemented by software, and therefore the hardware restrictions of analog PID controllers can be completely ignored. (For a comparison of the frequency-response characteristics of analog and digital PID controllers, see Problem A~3 Example 3-7 Consider the control system with a digital PID controller shown in Figure 3~31(a). (The PID controller here is in the positional form.) The transfer function of the plant is assumed to be BQ & Figure 3-30 Block diagram realization of the velocity frm digital PID control scheme Scanned with CamScannerPlane Analysis of Discrete-Time Control Systems Chap. 3 cur, [ogee [2147 Fae ete, ate] corer rd Pant = (93670"-1 + 02642¢-? cut + ket) tose = o Fie 3-31. (3) Block diagram ofa contol stem; (0) equvalen block disgram oo-— OD and the sampling period T is assumed to be 1 see. ‘Then the transfer function of the zero-order hold becomes Gals) Since s(5 +1). oy (= 036792") = 27) on, swe may redraw the block diagram of Figure 3-31(3) a8 shown in Figure 3-31(b) Tet us obtain the unitstep response of his system when the digital controller i PID controller with Ky = 1, Rs = 02, and Ko = 0.2 The pose transfer function of the digital controller is ven by [ 1 } 0.36792" + 0.26422"? 5 4m 1s" + 022" Gotz ‘Then the closed-loop pulse transfer function becomes RO Teste RG)” T+ Ga@GG) 0.51512"! = 0.14522"? ~ 0.29632"? + 0.05282 T= 1 85a + 1 S906 — 0 65422"? + 0.05282 We shall use the MATLAB approach to obtain the unit-step response Scanned with CamScanner‘Sec. 3-5 The Pulse Transfer Function 19 Obtaining Transient Response with MATLAB. Let us assume that we want the uunitstep response up to k = 40. Then the input r() may be written as nest 1,41) ‘A MATLAB program for obtaining the unit-step response for this system is, shown in MATLAB Program 3-1. The resulting output c(k) versus k is plotted in Figure 3-32 ' ‘MATLAB Program 3-1 ‘% ——~ Unit-step response ——— rum {0 05151 -0.1452 -0.2963 0.0526I: den=[1 -1,8528 1.5906 0.6642 0.0520); = ones(,4i); velo 40.0 21; axis); plot <0 kc, rid ttle Unit-Step Response’) label) vlabel ch) 7 UnivStp Response 1g 1 of 02 oS 0 k Figure 3-32 Unitstep response Scanned with CamScanner120 e-Plane Analysis of Discrete-Time Control Systems Chap. 3 ‘The response of this system to a unit-ramp input can be obtained by entering MATLAB Program 3-2 into the computer. The resulting plot is shown in Fig- ture 3-33. Note that this system exhibits no steady-state error in the ramp response. (MATLAB Program 3-2 che % Unitramp response —— num = {0 0.5151 -0.1452 -0.2963 0.0528); den =[1"-1.8528 1.5905 -0.6542 0.0528]; velo 16 0 16) axis = fiernurn den, plttk rote hk! sid ‘ile Unie-Ramp Responée? label) label) Unit Rarnp Response Scanned with CamScannerSec. 3-§ The Pulse Transfer Function 121 Comments. PID controllers for such process control systems as temperature systems, pressure systems, and liquid-level systems are usually tuned experimen- tally. In fact, in the PID control of any industrial plant, where its dynamics are not well known or defined, the controller variables (Ky, Ky, and Kp) must be determined experimentally. Such determination or tuning may be made using step changes in “The Fetérence or disturbance signal A few established procedures are available for such a purpose Basically, tuning (determining the constants Kp, K;, and Kp) is accomplished by systematically varying the values until reasonably good response characteristics are obtained For digital PID controllers used for process control systems, the sampling period must be chosen properly. Many process control systems have faicly large time constants. A rule of thumb in the selection of the sampling period (sampling period ~T'=Tal,, where «, is the sampling frequency) in process control systems is that for temperature control systems the sampling period should be 10 to 30 sec, for pressure contcolsysiems I to 5 sec, and for liquid-level control systems Obiaining Response Between Consecutive Sampling Instants, The z transform analysis will not give information on the response between two consecutive sampling instants. In ordinary cases this is not serious, because if the sampling theorem is satisfied, then the output will not vary very much between any two consecutive sampling instants. In certain cases, however, we may need to find the response between consecutive sampling instants. ‘Three methods for providing a response between two consecutive sampling instants are commonly available: 1. Laplace transform method 2. Modified z transform method 3. State-space method Hece we shall briefly discuss the Laplace transform method. The modified z trans- form methods presented in Appendix B. (Those readers interested in the modified transform should read Section B-4.) The state-space method will be discussed in Section 5-5 Laplace Transform Method. Consider, for example, the system shown in \ Figure 3-27. The output C(s) can be given by 5). CG) = GQ)E) = oo aH e() = ECGs) = ral Equation (3-59) will give the continuous-time response c(t). Hence, the response at any time between two consecutive sampling instants can be calculated by the use ‘of Equation (3-59). [See Problem A~3-16 for sample calculations of the cight-hand side of Equation (3-59) ] pile Scanned with CamScanner126 Plane Analysis of Discrete-Time Control Systems Chap. 3 The block diagrams in Figures 3-35 and 3-36(c) are equivalent, but the latter uses n delay elements, while the former uses n + m delay elements. Obviously, the latter, which uses a smaller number of delay elements, is preferred Comments. Note first thet the use of a minimal number of delay elements aves memory space in digital comrolrs. Also, te use of minimal mmbor of Summing poinis is desiral “Tn realizing digital controllers or digital filters, it is important to have a good level of accuracy. Basically, three sources of errors affect the accuracy: 1, The error due to the quantization of the input signal into a finite number of dis- crete levels. (In Chapter I we discussed this type of error, which may be con- Sidered an additive source of noise, called quantization noise. The quantization noise may be considered white noise; the variance of the noise is 0? = Q#/12.) 2, The error due to the accumulation of round-off errors in the arithmetic oper- ations in the digital system. 3. The error due to quantization of the coefficients a, and b, ofthe pulse transfer function. This error may become large as the order of the puke transfer function is increased. That is, in a higher-order digital filter in direct structure, small errors in the coefficients a, and 6, cause large errors in the locations of ‘These three errors arise because of the practical limitations of the number of, bits that represent various signal samples and coefficients. Note that the third type of error listed may be reduced by mathematically decomposing a higher-order pulse “transfer function into a combination of lower-order pulse transfer functions. In this “Way, the system may be made less sensitive to coefficient inaccuracies. For decomposing higher-order pulse transfer functions in order to avoid the coefficient sensitivity problem, the following three approaches are commonly used, 1. Series programming 2, Parallel programming 3. Ladder programming ‘We shall discuss these three programmings next A. Series Programming. ‘The first approach used to avoid the sensitivity problem is to implement the pulse transfer function G(z) as a series connection of first-order andior second-order pulse transfer functions. If G(z) can be written as a product of pulse transfer functions G,(z),G:(z),.. ,G,(2), or X Gz) = G(z)Gx2)-- Gy) then the digital filter for G(z) may be given as a series connection of the component igital filters G,(z), Ge(z),.._,G,(z), as shown in Figure 3-37 In most cases the G,(z)(i = 1,2, _. ,p) are chosen to be either first- or second- ‘order functions. If the poles and zeros of G(z) are known, Gy(z), Gilz),. -s Ga(z) “ean be obiained by grouping a pair of conjugate complex poies and a pait of Scanned with CamScannerSec. 2-6 Realization of Digital Controllers and Digital Filters 2 ne) vied ou Gye ote) xia vie Figure 3-37 Digital fer G(2) decomposed into a series connection of Gy=), G42)... Gyl2) conjugate complex zeros to produce a second-order function or by grouping real polesand real zeros to produce either first-or second-order functions. Itis, of course possible to group two real zeros with a pairof conjugate complex poles, or vice versa ‘The grouping is, in sense, arbitrary. It is desirable to group several different ways to see which is best with respect to the number of arithmetic operations required, the range of coefficients, and so forth To summarize, G(z) may be decomposed as follows: G(z) = Gi(2)Gz)--- Gz) The block diagram for and that for are shown in Figures 3-38(a) and (b), respectively. The block diagram for the digital filter G(z) is a series connection of p component digital filters such as shown in Figures 3~38(a) and (b) 2. Parallel Programming. ‘The second approach to avoiding the coefficient sen- sitivity problem is to expand the pulse transfer function G(z) into partial fract If GG) is expanded as a sum of A, Gi(z), Gx(2),...,Gy(2), oF $0 that Glz) = A + Giz) + Gz) +--+ + Gl) where A is simply a constant, then the block diagram for the digital filter G(z) can be obtained as a parallel con | +1 digital filters, as shown in Figure 3-39, Because of the presence of the constant term A, the first- and second-order funetions can be chosen in simpler forms. That is, G(z) may be expressed as Glz) = A + G2) + G2) + + GZ) “ ‘ =A4DG(2)+ D Gee) Kalb ee tt de The block diagram for 3-67) Scanned with CamScannerXD ee eva ye) :: Geeta aaa Fr Oy lag 128 Plane Analyse of Discrete-Time Control Systems Chap. 3 ra he wer bin peer’ h or x8 xia) 4-4aan AGrabawent | Maeiaf 2). i oy tad WD 10 Wi 8" coi” ial Wala) a) Figure 3-38 (a) Block diagram representation of Equation (3-65); (0) block diagram representation of Equation (3-66) ‘and that for ¥@)_ gt fiz XQ) Tez + det Ga) are shown in Figures 3~40(a) and (b), respectively. The parallel connection of g + 1 component digital filters as shown in Figure 3~40 will produce the block diagram for the digital filter G(z). 3, Ladder Programming, ‘The third approach to avoiding the coefficient sensitiv- ity problem is to implement a ladder structure, that is, to expand the pulse transfer Scanned with CamScannerSec. 3-6 Realization of Digital Controllers and Digital Filters, 129 outa, att 7 me xe is Sieh bf, sta Figure 3-39 Digital filter (2) ote! ‘decomposed asa parallel connection of LJ A, Gile), Coe). Gz) function G(2) into the following continued-fraci to this equation: form and to program according Giz) = Ag+ Bat Bz + ——_1__ 6-69) Ante Baty ‘The programming method based on this scheme is called ladder programming. Let us define " 1 OO) Bee OGY ia OM) = aM’ P2120 yn aiay=—+t B+ wo Then G(2) may be written as Gz) = An + Giz) Scanned with CamScanner130 Plane Analysis of Discrete-Time Control Systems Chap. 3 va Ye atl) sbyx a) ara 12) @). 20) yb) la Vv : a8 me ve Mar Hu) w Figure 3-40 (a) Block diagram representation of Equation (3-67); (b) block diagram representation of Equation (3-68) We shall explain this programming method by using a simple example where n= 2 Thatis, G(z) = Ag+ — i Bz + At Scanned with CamScannerChap 3. Example Problems and Solutions 163 Figure 3~60(b) shows plots of individual impulse responses given by Equation (3-109) [Observe that e(¢ consists ofthe sum of impulse responses that occur at = 0,¢ = 1, with Weighting factors 1,-03679, -0.6821,. From Equation (3-108) we see that for time intervals 05 ¢<1,1=0<2, 2513, .. the outpot (is the sum of impulse responses as follows: rn1te osrel (= 14°) ~ 03679[(¢ = 1) = 1 +e" 1er<2 et) =] (= 1 +e") ~ 0 3679{(¢ ~ 1) Le Te — 2) asrcs ~O6Rif = 2-1 +e Sample calculations for several values of ¢ follow: ()=0-1+1=0 (0.5) = 0.5 ~ 1 + 0.6065 = 0.1065 (1.0) = 0.3679 ~ 0 3679 x 0= 0.3679 (1.5) = 0.7231 ~ 0 3679 x 0.1065 = 0.6839 €(20) = 1.1353 ~ 0 3679 x 0.3679 = 1.0000 (2.5) = 1.5821 ~ 0.3679 x 0.7231 ~ 0.6321 x 0.1065 = 1.2487 (3.0) = 2.0898 ~ 0.2679 x 1.1353 — 0 6321 x 0.3679 = 1.3996 (4.0) = 3.0183 ~ 0.3679 x 2.0498 ~ 0.6321 x 1.1383 ~ 0.3996 x 0.3679 = 1.3996 (5.0) = 4.0067 ~ 0.3679 x 3.0183 ~ 0.6321 x 2.0498 ~ 0.3996 x 1.1353 +0. 0.3679 = 11469 {6 0) = 5.0025 ~ 0.3679 x 4.0067 — 0.6321 x 3.0183 ~ 0.3996 x 2.0498, 0.x 1.1353 + 0.2526 x 0.3679 = 0.8944 ‘The continuous-time output ¢(¢) thus obtained is plotted in Figure 3-60(¢). Problem A-3-19 Consider the digital filter defined by Gt = YQ = ME= MG +12 41 X@ "+0032 +08) Draw a series realization diagram and a parallel realization diagram. (Use one first: order section and one second-order section } Solution We shall first consider the series realization scheme, To limit the coefficients toreal quantities, we group the second-order term in the numerator (which has complex zeros) and the second-order term in the denominator (which has complex poles). Hence, we group G(z) as follows: pol zor dae! ere T+01e7 103277 + O82 Figure 3-61(a) shows a series realization diagram, Pronto Scanned with CamScanner164 2-Plane Analysis of Discrete-Time Control Systems Chap. 3 xe xiel 08 Le oy [igure 3-61 Block diagram realizations ofthe digital filter considered in Problem ‘A-H-19 (a) Series realization; (b) pallet realization Scanned with CamScannerChap. 3. Example Problems and Solutions 165 Next, we shall consider the parallel realization scheme Expansion of G(zW2 into partial fractions gives GW) 4 = We 2 He + OAK 12z +1) 3508) + +t 2 Fe01 F-03708 ‘Then G(z) can be written as follows: Gt) = -50 + Figure 3-61(b) shows a parallel realization diagram. Problem A-3-20 A slowly changing continuous-time signal x(¢) is sampled every T sec. Assume that ‘changes in signal x(¢) are very slow compared to the sampling frequency. Show that in the z plane (1 ~ z~'V/T corresponds to “differentiation,” just as 5 corresponds to “differentiation” in the s plane Solution Fora slowly changing signal x), the derivative of) can be approximated by v= 22 » Laan - xu - 91 ‘The 2 transform of this equation gives xy) = 2h = 29x) Vea) = EEX) - 2X) = FCI - 7) from which we obtain the block diagram shown in Figure 3~62(a) This diagram corresponds o thes plane differentiation shown in Figure 3~62(b). Note that the block diagram shown in Figure 362(a) can be modified to that shown in Figure 3-63. (For approximation of “integration” in the z plane, see Problems B-3-25 through B-3-27.) x [Tay Lee xt []|_ ma r . eo Figure 3-62 (a) Block diagram for “dtferentiation” in the 2 plane; (b) block diagram for “ferentiation” inthe + plane xo [4 vee Tr Figure 3-63 Approximate “aifferentistor” inthe = plane Scanned with CamScanner(5-4) t Glock ¢ Diagram Roiantion & PDosskonal from of Oi gyikied PID Gevol + Rie) s ECe) nA@) _ CHa \eray AS Kp Q-® 2 EQ) DANS! GS) sas) oss) @-2) Koxkst Ke) — © Cos OOS Kot se Go) = — G2") Scanned with CamScannerG Pe Prrbier Ges -\0)) Parallel ren \aabion e= Cte Aa Ce eag ETE CAxeN)C Sat) ExPan sion of [Gay /Z \ nro Parikich Frackens oir Co te ee (Gils FN FA Geae on) (Tost tod) aA Bg CES (%) Cea — ee ae fe ieee Qe ~LESOY) (Geese xo-3)) ZY 2. SSO Se 8 wie (ape) ie ee al Z(t) (Ee °o so, SC h eee = z See - Ho Sk Pie Goma a Be aes ee o3te OB) Ee ee Pa 8B Oe ay ce ‘° DW Scanned with CamScanner