Proceedings of the 36th Conference on Decision & Control San Diego, Califomla USA December 1997

FMll 1:30

Some Qualitative Properties of Multirate Digital Control Systems

Bo Hu and Anthony N.Michel Department of Electrical Engineering University of Notre Dame Notre Dame, IN 46556, U.S.A.
Abstract We consider multirate digital control systems which consist of an interconnection of a continuous-time nonlinear plant (described by ordinary differential equations), a digital controller (described by ordinary difference equations) which has quantizers (but is otherwise linear), along with the required interface elements (A/D and D/A converters). The input to the digital controller consists of the multirate sampled output of the plant. In the present note we show that when quantizer nonlinearities are neglected, then under reasonable conditions (which exclude the critical cases), the stability properties (in the Lyapunov sense) of the trivial solution of the nonlinear multirate digital control systems can be deduced from the stability properties of the trivial solution of its linearization. For such systems we also present a result concerning the existence and construction of stabilizing multirate-output digital controllers. In the present note we also show that the solutions of multirate digital feedback control systems with nonlinear plant and quanizers are uniformly ultimately bounded if the trivial solution of the corresponding linear systems consisting of the linearization of the plant and with the quantization removed from the digital controller, is asymptotically stable. We also provide a result which compares the response of multirate digital control systems with nonlinear plant and quantizers in the controller with the response of the corresponding nonlinear multirate digital control systems without quantizers in the digital controller.
1. Introduction A. Background Sampled-data control systems, resp., digital control systems, are hybrid dynamical systems which usually consist of a continuous-time plant (which can be described by a set of first order ordinary differential equations) and a digital controller (which can be described by a set of first order ordinary difference equations). When data in such systems are sampled at more than one rate, such systems are called multirate digital control systems, resp., multirate sampled-data control systems.
0-7803-3970-8197 $10.00 0 1997 IEEE

The analysis and synthesis of such systems (especially systems with one sampling rate) have been of continuing interest for several decades (refer, e.g., to [I], [5], [6], [8], [12] and the references cited therein). Although the plant is frequently nonlinear in such systems, for purposes of design and analysis, a linearization of the plant is usually used in the literature, without much justification. Furthermore, most of these results do not take into account quantizer effects in the digital controller and in the interconnecting elements (A/D and D/A converters). In several works (e.g., [2], [15]-[17]) quantization effects in sampled-data control systems with linear plants are studied. In a recent paper [7] we investigated the qualitative behavior of single rate digital control systems with nonlinear continuous-time plants and with digital controllers and interface elements (A/D and D/A converters) which include signal quantization. In this work we presented Lyapunov stability results for systems with nonlinear plants but no quantizers in the controllers and Lagrange stability results for systems with nonlinear plants and quantizers in the digital controllers. In the present paper we establish new qualitative results for multirate digital control systems with nonlinear continuous-time plants and with quantization in the digital controllers. For the special case when the plant is linear and there is no quantization in the digital controller, the class of systems considered herein reduces to the class of the systems treated in [6].

B. Multirate Digital Feedback Control Systems We consider multirate digital feedback control systems of the type depicted in Fig. 1. The nonlinear plant is described by equations of the form

{
4298

e ( t )= E(kTo), 2 = f ( z ( t ) ) Ke(t), + t E [kTo,(k + 1)To) y ( t ) = L z ( t ) , k = 0 , 1 , 2 , . ..

(1.1)

where TO > 0 is the frame period (see, [6]),and the digital controller without quantizers is given by
U(@

+ 1)TO) = Cu(kT0)+ Mv(kTo),k = 0 , 1 , . ..

(1.2)
\ I

P(kT0) = Nu(kT0)

Authorized licensed use limited to: SUNY Buffalo. Downloaded on August 30, 2009 at 23:24 from IEEE Xplore. Restrictions apply.

where x E R", y E !I?', U E R", e , d,p E Rm, and v(kT0) will be specified in the subsequent discussion. The output of the plant y ( t ) is sampled componentwise. The ith component of the plant output, gi, is given by
p = 0,. ..,N - 1 i (1.3) where Li is the ith row of the output matrix L , and Ti,To are related by Ti = Ni 9 i = 1 ,...,1, where Ni is a positive integer. Define

For s E [O,To],we have for system (2.2), z(kT0 s) = eA8x(kTo) [J," eAtdt].Bu(kTo).Using this equation and (1.3), we obtain tlhe output of the linearized plant as

Yi(kT0 pTi) = LiZ(kT0

+ PTi),

V(kT0)

which consists of all the sampled output values in [kTo, (Ic + l)To)* Finally, we will assume that in (1.1) and (1.2), f E C1[!Rn,!I?fl] f : 8 + 8" and f is continuously (i.e., " differentiable), f (0) = 0, and that K , L , C , M , N denote . . . constant matrices with appropriate dimensions.
,

G(kT0) [ ( y l ( k T O ) ) T , . 9*(Yl(kT0 + (NI * 1)T1)lT, * * , (Yl(kTo))T,. . , (YI(kT0+ (NI - l)T1)>TIT, .

' '
e(t)

where W1 = [ ( L I ) (LleA7"1)T,. ,(LleA(N1-l)T1 7 ~, .. 1 ,( L I ) ~ , ( L l e A z I T , . , ~2 = [OT, e ~ t d t l . ~.) ,~ ,L ~ [ J : ~ ~ t- d t )] ~ ~~) ~ , ..( e ~ . ...,oT, eAtdt1. B I T , .. .,( L ~ [(NI-1)Tl eAtdt] s, B)T]T. Substituting (2.3) into (2.2), we abtain for the linearization of (2.1) the alternative expression

...

..

(~11~2

5 = A z ( t ) + Bii(t), ii(t) = ~ ( k T o ) , t E [kG, (k 1)To) ~ ( ( k 1)To) = (C MW2)u(kTo) M W l z ( k T o ) ,

+-y

r(t)3 0

Y(t)
U

Fig. 1 Nonlinear sampled-data control system

2. Stability Analysis of Systems

(2.4) Note that the trivial solution = (OT,OT)T is an equilibrium of the single-rate digital control system (2.4). For system (2.4), the following results are well known (see, e.g., [4]). L e m m a 1. The equilibrium ( z Tu ~= (OT,OT)T of , ) ~ the linear digital control s,ystem (2.4) is ezponentially stable in the large if and oiily if the matrix

Without Quantizers In this section we assume ideal A/D and D/A converters and we assume infinite wordlength digital controllers (i.e., there is no quantization in the converters and in the digital controller). Under these assumptions, we have e'(kT0)= p(kT0) = Nu(kTo), and v(kT0) = d(kT0) = [(Yl(kTo))T, , ( Y 1 (kT0 + (Nl l ) T l ) ) T , * , (Yl(kTo))T, (Y'(kT0 W l - l)T1))TIT. * *, The multirate digital feedback control system of Fig. 1 is now described by the equations
f

is Schur stable (i.e., all the eigenvalues of H are within the unit circle of the complex plane). L e m m a 2. Assume that H has at least one eigenvalue outside the unit cixcle. Then the equilibrium ( x T , u ~= (OT, OT)T of the linear digital control sys) ~ tem (2.4) is unstable. When the eigenvalues of H satisfy either Lemma 1 or Lemma 2, they are said to be noncritical. Otherwise, they are said to be critical eigenvalues. (For these definitions and the usual definitions and results of the x = f ( z ( t ) ) Bii(t),ii(t)= u(kTo),t E [ W O , (k 1)To) Lyapunov stability of an equilibrium and the boundedU((k 1)To) = CU(kT0) Mjj(kTo), k = 0 , 1 , .. . ness of solutions (Lagrange stability), the reader may (2.1) want to refer, e.g., to [13] and [14].) where B = K N . We note that since f(0) = 0, the Theorem 1. The equilibrium ( x T , ~ T )= (OT,OT)T T trivial solution ( z T2, ~ = ~ ) ~ is an equilibrium (OT, of the nonlinear multirate digital control system (2.1) of the multirate system (2.1). is uniformly asymptotically stable if the equilibrium Associated with (2.1) is the system given by ( x T ,u ~= (OT, OT)T of tbe linear digital control sys) ~ tem (2.4) is exponentially stable, or equivalently, if the 2 = Ax(t)+ Biqt), C(t) = u(kTo), matrix H given in (2.5) is Schur stable. t E [kTo,(k + 1)To) Proof: Since f : Rn -+ 3Ifl is continuously differ~ ( ( k 1 T )= Cu(kT0) M g ~ ( k T o k, = 0 , 1 , ... )o ) entiable with f(0) = 0, we can represent f as f ( x ) = (2.2) A z g(z), where A = [%.(O)] n x n and g E Cl[%", Sn] where d,r,(kTo) represents the multirate output of the ._.._ . linearized system and A denotes the Jacobian off eval= 0, where 11 * 11 satisfies the condition liml,,o uated at 2 = 0, i.e., A = The linear denotes the Euclidean norm on W. The first equation nxn of (2.1) now assumes the form x = A z ( t ) g ( z ( t ) ) system of equations (2.2), is called the linearization of system (2.1). B i i ( t ) , G ( t ) = zc(kT0) for t E [kTo,(k 1)To). Its 4299

[g(O)] .

Authorized licensed use limited to: SUNY Buffalo. Downloaded on August 30, 2009 at 23:24 from IEEE Xplore. Restrictions apply.

+J[~L(

k = 0,11,.

+ ..

+ +

whenever Ilw(k)\l < 6, for any k E Z+ = (0,1,2,. ..}. The proof of the above proposition is omitted due to the space limitations. If we now choose a vo > 0 such that 2vollPHII vO2llPll < 1, then there exists a S(YO) > 0 such that

V(W(k + 1)) - V ( w ( k ) )I -llW(k>1l2

+2YOllPHll * llW(~>1l2 . 2 1 l llW(k)1l2 + 0 1 Pl = -U - 2vollPHII - vo211PII) llw(k>l12

(2.14)

whenever Ilw(k)II < ~ ( v o It. follows from the definition ) of V(w)and (2.14) that Xmi"(P)llw(k + 1 1 2 I v(w(k+ )1 1)) I V ( w ( k ) )5 Anax(P)llW(k)112, where X m i n ( P ) and A,,(P) denote the smallest and largest eigenvalues of

P , respectively. Let d ( ~ ) ~ S ( Y OIf) Ilw(k0)ll < . d for some ko E Z+, then the above inequalities yield Ilu(ko 1)11 < YO). Thus, (2.13) is applicable for k = ko 1, which yields IlV(w(k0 2))11 5 IlV(w(ko + 1))Il 5 IIV(w(ko))ll. Therefore L n i n ( P ) I l w ( k o + 2)112 5 V(w(ko 2)) L V(w(ko))I L"(P)llw(h)l12, which implies that Ilw(k0 2)11 < S ( v 0 ) . By induction, it follows that Ilw(k)II < S(v0) for k 2 k ~ . Hence, (2.14) is satisfied for all k 2 ko whenever Ilw(k0)ll < d. Therefore, the trivial solution of (2.10)is uniformly asympA totically stable in the region &(O) = {w E !J2(n+s) : llwll < d } (refer, e.g. to [13]). Next, recalling that G ( t ) = u(kT0) for t E TO, (k l)To), it follows that for t E [kTo,(k l)To), Il+O)ll < d , and L ko, Il4t)lI 5 eTo~'A"I14k~o)ll + T e o l l l B X IlU(kTo)II 11 oTlAlll l eA(t-r) X g((Z(T))dTll

+ JLTo

now in a pogtion to show that the trivial sblution of - eTo~~AllJ~~w(~)II < the system (2.1) is uniformly asymptotically stable if eTollAll119((.(T))11dT the trivial solution of system (2.4)is exponentially stable. Since H is assumed to be Schur stable, there ex< eTollAllJ1+ To2e2TollAllllB11211w(k>ll vollw(k>>ll + ists a positive definite symmetric matrix P such that - (eTollAllJ1+ TO2e2TollAllll~112 u o ) ~ ~ w ( ~ ) ~ ~ < H T P H - P = -In+,, where In+s the ( n + s ) x ( n + s ) is where in the second-last line, inequality (2.13) was identity matrix. Choosing V(w) = w T P w , where w E !IFs, and letting w ( k ) = [ ~ ~ ( k T o ) , u ~ ( k T o ) used. Therefore, z(t) converges to the origin simul]~~ taneously with w ( k ) = [zT(kTo), uT(kTo)lT whenever (2.10) can be written as lIw(ko)ll < d . Noting that d is independent of ko, we u ( k 1) = Hw(k)+ O ( k ) . (2.11) conclude that the trivial solution of system (2.1)is uniformly asymptotically stable if H , given by (2.5), is The first forward difference of V evaluated along the Schur stable. 0 solutions of (2.11) yields To accommodate hybrid systems, we have modified V ( w ( k + 1)) - V(w(k))= w T ( k ) [ H T P H- P ] w ( k ) the usual definition of asymptotic stability in Theo+2RT(k)PHw(k)+ R(k)PR(k) = -l/w(k)112 = rem 1. Specifically, the equilibrium w = ( z ~ , u ~ ) ~ (OT,OT)T of the hybrid system (2.1) is taken to be uni+2RT(lc)PHw(k) fl(k)*PO(k) -llw(k)112 formly asymptotically stable if the equilibrium w = 0 +2llQ(k)ll . IlPHll * Il4k)ll + I l ~ ( ~ > lI l11. . P2 of the corresponding discrete-time system (2.10) is uni(2.12) formly asymptotically stable and in addition Ilz(t))I Before proceeding further, we need the following intercllw(k)II for all t E [kTo, (k l)To), k 2 ko, where c is mediate result. a constant independent of ko. Proposition 1 For any given Y > 0, there exists a . Theorem 2. Assume that H given in (2.5) has at 6 = S(v) > 0, such that least one eigenvalue outside the unit circle in the com(k+l)TOeT plex plane. Then the equilibrium w = ( z ~ , u=~ ) ~ Iln(k>II5 JkT0 ollAll Il9(4t)) lldt (2.13) (OT, OT)T of the nonlinear sampled-data control sys+llMll * IlWdkT0)ll I Yll4k)II

JLzl)To

4300

Authorized licensed use limited to: SUNY Buffalo. Downloaded on August 30, 2009 at 23:24 from IEEE Xplore. Restrictions apply.

tem (2.1) is unstable. Proof: Omitted due to space limitations. 0 For single-rate digital control systems considered in [7],we have W1 = L, and W2 = 0. Clearly then, the results of the present paper constitute generalizations to the results given in [7].
3. Stabilization b y Multirate-Output Controllers In this section, we will present a result concerning the existence and construction of multirate-output controllers which stabilize nonlinear systems of the type considered herein. The nonlinear plant may not be stable at the outset. Definition 1. [8] The frame period TOis said to be nonpathological if whenever X is an eigenvalue of A , no point X + j .27rk/To is an eigenvalue of A when k is a nonzero integer. [6] (Observability Index Vector) Definition 2 . Consider an observable pair ( A , L ) where A E WXn and L E and express L as L = [LT,-..,LTIT. A set of 1 integers (121,. nr) is said to be an observability index vector (abbreviated OIV) of the pair ( A , L ) if Ci=lni = n and the vectors L1,..,L I A n l - l ; L a , ... ,L2An2-1;... ;Ll,. ..,LrAnJ-1 . are linearly independent. Here ni = 0 means that the vectors of the form LiA" do not appear in the above series. Note that an observable pair has at least one OIV, e.g., the set of Kronecker invariants [9] of the pair ( A T ,L T ) ,and may have more OIV's if 1 2 2. An algorithm to determine OIV's was presented in Luenberger [ll].Also, it is known that any OIV is preserved under similarity transformations. In the following, we use the triplet ( A ,K , L ) to denote the linearized system 5 = A z ( t ) K e ( t ) , y ( t ) = L z ( t ) corresponding to the nonlinear plant (1.1). Theorem 3. If the linear system ( A , K , L ) is a controllable and observable triplet, where A , K , L are the same as in Section 11, and if ( m l , . ,ml), the OIV of the augmented system A ,[L,oI) satisfy Ni 2 m,,i = l , . . - , Z , then for any nonpathological To > 0, there exists a multirate digital controller of the form (1.2) which stabilizes the nonlinear plant ( l . l ) , i.e., makes the nonlinear multirate digital control system (2.1) uniformly exponentially stable. Proof. This theorem is a direct consequence of the linear system result given in [6] and Theorem 1. In the the present case, we let N = I, E PXg, s x s identity matrix. Consider the linearized system. By z(kTo s) = eAuz(kTo) [J," eAtdt].Bu(kTo),we have

gain matrix F such that - BF is stable (its eigenvalues can be arbitrarilly assigned). By [6], it follaws that under the present assumptions, has full column rank, where 6' = [ ( L ~AT0 ) T , ( L,-A (To-TI 1)" ,. .., (LlevAT1),. ., e~ . ( L ~ ~ - A T O )(T , ~ - A ( " O - T I ) , . . ., re-^")^]^ L~ )T = (WleATO)T,& = [ ( ~ 1 [ ~ , - ~ 0 e A txd t ] BIT, ~[J,-'~O-~" ( 1 e ~ t d t ]B I T , ...,( ~ 1 e ~ t d tx ] BIT,. . , (L,[s,-~o . eA'tdt1 B I T , (LIIJo- (To-*)eAtdt]

[6',4

. .,(Lr[&* eAtdt] * B)T]T.For any given state transition matrix C in the second equation of (2.2), we solve - M [ C , G ] = [.F,C]to obtain one solution, e.g., M = -[~,C]([6',G:IT[6',G])-1[d,G]T. Now for the above choice of F,C, M , system (2.2) is uniformly exponentially stable (see [SI). In fact, in the present case, the multirate digital controller can be reduced to the equivalent form,
u(kT0) = -Fz(kTo),

eXn

(k 2 1)

(3.2)

e ,

regardless of the choice of the state transition matrix C. For a detailed discussion concerning the results for linear systems, refer to [6]. Now, by (3.1), (3.2) and 0 Theorem 1, we obtain Theorem 3.
4. Analysis of !Systems w i t h Quantization

Nonlinearities In the implementatj on of digital controllers, quantization is unavoidable. In [3], [15] and [16], details concerning the inclusion of quantizers into digital controllers are discussedl. In the present section, we study qualitative properties of the nonlinear multirate digital control system of Fig.1 with fixed-point quantization included in both the A/D converter and the digital controller. Fixed-point quantization can be characterized by the relation &(e) = 8 q(8) where lq(8)l < e for all 8 E 92 and t:, the quantization level, depends on the desired preci,sion. From Fig. 1, we obtain, assuming that r = O,v(kTo) = Q(Q(kT0)) = Q(kT0) qi(ij(kTo)),u((k 1:ITo) = Q(Cu(kTo) Mv(kTo)) = Cu(kT0)+Mv(kTo)i-472 (Cu(kTo)+Mv(kTo)), p(kT0) = Q ( N ~ ( k 2 ' 0 )= N~(lcT0) qS(NU(kTo)),k= 0 , 1 , 2 , * * * ) where the q1, 42 and q3 should be interpreted as vectors whose components contain quantization terms. By a slight abuse of notatton, we will henceforth write q1 (k) in place of q1(Q(kTo:l), qZ(k) in place of q2(Cu(kTo) Mv(kTo))and so forth. It is easily verified that there exist positive constants Ji which are independent of E such that Ilqi(k)II 5 Jie,i = 1,2,3, k = O,l,. In the presence of quantizer nonlinearities, we can no longer expect that the system of Fig. 1 will have a uniformly asymptotically stable equilibrium at the origin; in fact, ~ ( ( k 1)To) = A~(lcT0) Bu(kT0) there may not even be an equilibrium at the origin. In (3.1) view of this, we will investigate the (ultimate) boundwhere = eATo,B = [ eAtdt] K. For a nonpathoJ : edness of the solutioin of the system of Fig. 1, including the dependence of thle bounds on the quantization size. logical frame period TO> 0 , we know that B is conSimilarly as in Section 1 , we can show that the 1 trollable [lo] and thus there exists a stable feedback 430 1

([

...

a,

Authorized licensed use limited to: SUNY Buffalo. Downloaded on August 30, 2009 at 23:24 from IEEE Xplore. Restrictions apply.

system of Fig. 1 can be represented by the equations k ( t ) = A z ( t ) g ( z ( t ) ) Bu(kT0) Kq3(k), t E W O , (k WO) u((k 1)To) = MWlZ(kT0) (C MWz)u(kTo) +MW3(kTO) + Mql(k) q2(k)* (4.1) Letting w ( k ) = [ ~ ~ ( k ) , u ~ we have the equivalent (k)]~, representation of (4.1), valid at t = kT0, k = 0,1,. .-, given by w ( k ) = H w ( k ) n ( k ) , where H is defined by (2.5) and where

+ +
+

Theorem 4. For system (4.1), if the matrix H defined in (2.5) is Schur stable, then there exist 0 > 0, do > 0, and a constant ro > 0 such that

Remark. From the above theorem we can conclude that the bound of the solution of (4.1) can be made arbitrarily small by choosing the quantization level sufficiently small, provided that the initial state of the system is close enough to the origin. Indeed, replacing 60 by in Theorem 4, we obtain from (4.3) and the last statement in the above proof that llw(k)ll 5 9 6 , 2 ko;

Ilz(t)II 5

( ( e d aTOJ~~~KI~)~~O(II~~~+~ +

whenever E 5

and Ilw(k)ll I 6,.

whenever E 5 EO and Ilw(k0)ll 5 *e for some ko E Z+ and t 2 koT0. Clearly, these bounds are all linear in E , the quantization level. 0 In our final result we establish an estimate of the norm of the difference between the response of nonlinear multirate digital control systems without quantizers, given by (2.1), and nonlinear digital feedback control systems with quantizers, given by (4.1). For the 4302

Authorized licensed use limited to: SUNY Buffalo. Downloaded on August 30, 2009 at 23:24 from IEEE Xplore. Restrictions apply.

present purpose, we rewrite (2.8) and (2.9) as k ( k ) = H 3 ( k ) fi(k), where 3 ( k ) = [ZT(kTo),iiT(kTo)]T

(71 L. Hou, A.N. INichel and H. Ye, Some Qualitative Properties of Sampled-Data Control Systems, IEEE Runs. Automat. Contr., to appear.

-.

-,

z(k) = w ( k ) -:(IC),

we obtain

z(k + 1) = Hz(lc) + SkTo eA((k+l)To-t)[g(z) - g(2) + ~q3(1c)lcit (k+l)To M Q l (IC) + Q2 (k) (4.7) Suppose g ( - ) has the property that lim,,o,*+o 1 b = 0. Following a similar procedure as in Section 11, we can establish the next result. We will omit the details of the proof. Theorem 5. Assume that H defined in (2.5) is Schur stable and lim,,o,%,o j e = 0. Then there exist 0 > 0, 60 > 0, and a constant TO > 0, such that

[8] H. Ito, H. Ohmori and A. Sano, Stability Analysis of Multirate Sampled-data Control Systems, IMA Journal of Mathematical Control & Information, vol. 11, pp. 341-354, 1994.
[9] R.E. Kalman, Kronecker Invariants and Feedback, Proc. Conf. Ordinay Differential Equations, Math. Research Center, Naval Research Labs., W s ah ington, DC, pp 459-471, 1971.

[lo] R.E. Kalman, Y. C. Ho and K. S. Narendra, Controllability of linear dynamical systems, Contributions to DifferentialEquations, vol. 1, No. 2, pp. 189213, 1962.
[ll] D. G . Luenberger, Canonical Forms for the Linear Multivariable Systems, IEEE lkans. Automat. Contr., vol. 12, pp. 290-293, 1967.

[12] D. G. Meyer, A Parameterization of Stability Controllers for Multirate Sampled-data Systems, IEEE lkans. Automat. Contr., vol. 35, pp. 223-235, 1990. [13] A.N. Michel and K. Wang, Qualitative Theory of Dynamical Systems, ]Marcel Dekker, New York, 1995. [14] R. K. Miller and A. N. Michel, Ordinary Differential Equations, Academic Press, New York, 1982. References [15] R. K. Miller, ,4. N. Michel and J. A. Farrell, Quantizer Effects on Steady-State Error Specifications of Digital Feedback Control Systems, IEEE Runs. Automat. Coiatr., vol. 34, No. 6, pp. 651-654, 1989. [16] R. K. Miller, Ivl. S. Mousa and A. N. Michel, Quantization and Overflow Effects in Digital Implementations of Linjear Dynamic Controllers, IEEE Trans. Automat. Contr., vol. 33, NO. 7, pp. 698-704, 1988. [17] J. E. Slaughter, Quantization Errors in Digital Control Systems, IE:EE Runs. Automat. Contr., vol. 9, pp. 70-74, 1964.

[l] M. Araki and K. Yamamoto, Multivariable Multirate Sampled-data Systems: State-space Description, Transfer Characteristics, and Nyquist Criterion, IEEE Trans. Automat. Contr., vol. 31, pp. 145-154, 1986.
[2] J. E. Bertram, The effect of Quantization in Sampled Feedback Systems, Dans. AIEE Appl. Ind., part 2, vol. 77, pp. 177-181, Sept., 1988. [3] J. A. Farrell and A. N. Michel, Estimates of Asymptotic Trajectory Bounds in Digital Implementations of Linear Feedback Control Systems, IEEE Runs. Automat. Contr., vol. 34, No. 12, pp 1319-1324, 1989. [4] B.A. Francis and T.T. Georgiou, Stability Theory for Linear Time-invariant Plants with Periodic Digital Controllers, IEEE Dans. Automat. Contr., vol. 33, NO. 9, pp 820-832, 1988. [5]
*

of Dynamic Systems, Addison-Wesley, 1980.

G. F. Franklin and J. D. Powell, Digital Control

[6] T. Hagiwara and M. Araki, Design of a Stable State Feedback Controller Based on the Multirate Sampling of the Plant Output, IEEE %ns. Automat. Contr., vol. 33, No. 9, pp. 812-819, 1988.

4303

Authorized licensed use limited to: SUNY Buffalo. Downloaded on August 30, 2009 at 23:24 from IEEE Xplore. Restrictions apply.