Anda di halaman 1dari 6

STRUCTURE ASSIGNMENT OF RIGHT INVERTIBLE

IMPLICIT SYSTEMS THROUGH NONREGULAR


STATIC STATE FEEDBACK
S. Mondie, J.J. Loiseauy
CINVESTAV, Av. IPN 2508, Zacatenco, 07300, Mexico D.F.


fax:(52)57-47-70-89 e-mail:smondie@ctrl.cinvestav.mx
y
IRCyN, UMR CNRS 6597, BP 92101, F{44321 Nante Cedex 03, France
fax:(33)240-37-25-22 e-mail:loiseau@lan.ec-nantes.fr

Keywords : Invariants, linear systems, nonregular with or without internal stability, in terms of the invariant
feedback. lists [3, 12].
When G is not necessarily invertible, the feedback is
said to be nonregular, the structure of the system is mod-
Abstract i ed and the complexity of the problem signi cantly in-
Necessary and sucient conditions for the existence of a creases. A rst step toward a solution of the classical
solution to the problem of assigning the structure of a right control problems when the feedback law is nonregular is
invertible system through nonregular static state feedback the study of the limits in assigning a prescribed structure
are obtained. In addition, three results of interest in sys- using such a feedback law. Solutions of this problem for
tem control and design are obtained as corollaries. particular cases are available in the literature. These are
the theorem of Rosenbrock on pole placement [15], and
further generalizations of Zaballa [17], Baraga~na and Za-
1 Introduction balla [1], Zagalak and Loiseau [19], as well as Heymann's
theorem [5].
The study of the structure of linear systems is an im- The problem solved in this paper generalizes these re-
portant part of the algebraic theory of linear systems de- sults to the case where J is monic and both systems (1)
scribed by :
and (3) are right invertible.
J x= Ax + Bu ; (1) The solution is obtained as the generalization of the
where A; J 2 Rqn with Rank(J) = r, B 2 Rq(m+k) : result for explicit systems [1, 13], i.e. when the unique
The main fundamental results are those given by Kro- lists of invariants of the system are invariant factors and
necker [4], Brunovsky [2] and Morse [14] among other au- column minimal indices. This generalization is based on
thors. The structure of a system is given by lists of in- the statement of the problem as a matrix pencil comple-
tegers and of polynomials known as Kronecker invariants tion problem and on a mapping described by Loiseau [11]
or Morse lists. These lists characterize the fundamental which brings all the zeros of the system, either nite or
behavior of the system. in nite, into nite positions.
When a static state feedback, described by Three variations are derived. First, the regularization of
a right invertible implicit system through non regular feed-
u = Fx + Gv ; (2) back, i.e. when the closed{loop system unique invariants
are invariant factors and in nite zeros. Second, the struc-
where F 2 R(m+k)n and G 2 R(m+k)m , is applied, the ture assignment of a linear system with output equation
closed{loop system is given by through non regular feedback. Finally, the assignment of
: the structure of a strictly proper linear system through
J x= (A + BF)x + BGv : (3) the choice of the output equation, the state equation be-
When the feedback law is regular, i.e. G is invertible, the ing given.
structure of the system remains invariant. This permits to
express the existence of solutions to control problems, such
as model matching, disturbance rejection, or decoupling,
2 Problem statement
 The authors thank Conacyt, Mexico and MESR, France for sup- The statement of the problem as a matrix pencil comple-
porting this research through PCP 14 tion problem is an important tool for the solution.
Lemma 1 Consider an implicit system described by (1) 3 Preliminary results
with system matrix (sE H) = (sJ A; B) 2 Rq n m k ( + + )

and a pencil (sE 0 H 0 ) 2 Rq n m ; and assume that J The background is the solution of the problem in the case
( + )

is monic. Then the following statements are equivalent. of explicit systems, i.e. when J is square and invertible. In
(i) There exist matrices F and G; as in (2), such that the that case the matrix pencils (sE H) and (sE 0 H 0 ) only
closed{loop system matrix (sJ A BF; BG) has the have invariant factors and column minimal indices. Two
same Kronecker invariants as (sE 0 H 0). polynomials (s) and (s) being given, deg (s) denotes
(ii) There exists a constant matrix  2 Rqk such that the degree of (s), lcm( (s); (s)) denotes the least com-
mon multiple of (s) and (s), and (s)  (s) means
(sE 0 H 0 ; )  (sE H) ; that (s) divides (s) without remainder.
where  is the symbol for the equivalence of matrix pencils Theorem 1 [13] Let a pencil (sE H) have Kronecker
in the sense of Kronecker. invariants
{ invariant factors i (s); i = 1; :::; n
Proof Let (sE H) = (sJ A; B) and (sE H ) = { column0 minimal
0 0 indices ci ; i = 1; :::; m + k.
and (sE H 0) be a matrix pencil with Kronecker invari-
(sJ A BF; BG) be the system matrices in the sense
of Rosenbrock [15] corresponding to the open{loop system ants 0
and the closed{loop system described by equations (1) and { invariant factors i (s); i =0 1; :::; n
(3) respectively. Clearly, { column minimal indices ci ; i = 1; :::; m:
I 0 Then (sE 0 H 0 ) can be completed so that
(sE 0 H 0 ) = (sE H) Fn G : (sE 0 H 0; )  (sE H) ;
With no loss of generality, G is one to one and it can be if and only if the following conditions hold.
completed so that (G; Gc) is invertible. The transforma- 0i+k (s)  i (s)  0i(s) ; i = 1; :::; n;
tion I 0 0 
n ci+k  c0i ; i = 1; :::; m;
F G Gc
is then invertible. Then (ii) comes choosing the constant Xl k
jn+X k j
ci + deg(lcm( i (s); 0i+j (s))) 
matrix  = BGc . Reversely we can assume, with no loss i=1 i=1
of generality, that sE 0 H 0 is the system matrix of some lXj
system,
k
j nX+k

 ci + deg 0i(s) ; j = 1; :::; k;


0
sE 0 H 0 = sJ 0 A0 ; B 0 ; i=1 i=1
and
where J 0 ; A0 and B 0 are matrices of adequate sizes. Since mX
(ii) is satis ed, there exist matrices P and Q, +k X
n X
m nX
+k

ci + deg( i (s)) = c0i + deg( 0i(s)) ;


Q Q Q  i=1 i=1 i=1 i=1
Q= Q Q Q ;1 2 3
4 5 6 where
both invertible, such that ljk = minfi j c0i j +1 < ci g ; j = 1; :::; k;
P sJ 0 A0
 B0 B 00 = and, by convention, 0i(s) = 1, for i > n.
  It is possible to extend this result to the case of right in-
(sJ A; B) Q Q Q :
Q Q Q
1 2 3
vertible pencils. The key results to that end is the Lemma
4 5 6
2 of the Appendix, which permits to reformulate the ques-
Identifying the coecients in s on both sides of the tion in terms of explicit systems, and hence to apply The-
equality, we obtain that PJ 0 = JQ1 with P invertible, orem 1.
JQ2 = 0 and JQ3 = 0. Taking in account the assumption
that J is monic, it appears that Q1 is invertible, Q2 = 0 Theorem 2 Let (sE H) be right invertible, having for
and Q3 = 0. Now let F = Q4Q1 1 and G = Q5 , therefore Kronecker invariants
Q1 0 { invariant factors i (s) ; i = 1; :::; q,
{ in nite zero orders ni ; i = 1; :::; q n,
P(sJ 0 A0 ; B 0) 01 I = (sJ A BF; BG) { column minimal indices ci ; i = 1; :::; n+ m + k q,
and let (sE 0 H 0) be right invertible, having for Kronecker
which ends the proof. invariants
{ invariant factors 0i(s) ; i = 1; :::; q, and
{ in nite zero orders n0i ; i = 1; :::; q n,
{ column minimal indices c0i ; i = 1; :::; n + m q. e0i (w) = 0i( 1 +waw )wdeg (s) wn ; i = 1; :::; q; (6)
0
i
0
i
0 0
Then (sE H ) can be completed by a constant matrix
 so that where ni = n0i = 0, for i = q n + 1; : : :; q. Their col-
(sE0 0
H ; )  (sE H) ; umn minimal indices are the same as those of the orig-
if and only if the following conditions hold. inal pencils, that are ci ; i = 1; : : :; n + m + k q and
c0i ; i = 1; : : :; n + m q, respectively.
0 0
i+k (s)  i (s)  i(s) ; i = 1; :::; q; Since (4) is true, these invariant factors and column
minimal indices satisfy the conditions of Theorem 1.
n0i+k  ni  n0i ; i = 1; :::; q n; Hence the conditions of Theorem 2 are deduced from that
ci+k  c0i ; i = 1; :::; n + m q; of Theorem 1,(5), and (6).
Conversely, if these conditions hold, from Theorem
Xl Xq qXn (5), (6), the pencil (wEe0 He 0 ) can be completed
k
j
1,
ci + deg(lcm( i(s); 0i+j (s))) + max(ni ; n0i+j ) with a constant matrix e such that (wEe 0 He 0 ; e) 
i=1 i=1 i=1
(wEe H): e The matrix pencils are transformed back to
lXj
k
j
X 0 X 0
q q n (sE 0 H 0); (sE H) and there exists a matrix pencil
0
 ci + deg i (s) + ni ; j = 1; :::; k;  = (sE H 00) (not necessarily constant) and invertible
00
i=1 i=1 i=1 matrices P and Q such that P(sE 0 H 0; ) = (sE H)Q.
n+mX +k q
X q qXn By assumption, the pencils (sE H) and (sE 0 H 0) have
ci + deg i(s) + ni the same number of in nite zero orders, and of row min-
i=1 i=1 i=1 imal indices. The last condition of the theorem expresses
n+X m q X q qXn that they also have the same number of rows. Thus, ac-
= 0 0
ci + deg i (s) + ni ; 0 cordingly to the Lemma 3 of the Appendix, E and E 0
i=1 i=1 i=1 have the same rank. Hence it appears that the constant
where  = H 00 can be choosen as well.
ljk = minfi j c0i j +1 < ci g ; j = 1; :::; k;
4 Related results
and where by convention 0i(s) = 1 ; i > q, and n0i = 0 ; i >
q n. It would be clear from Lemma 1 that, if J is monic, the
conditions of Theorem 2 also characterize the freedom in
Proof modifying by static state feedback the Kronecker invari-
Assume that there exists P; Q invertible such that ants of (1), so that (3) is right invertible.
Next, three results that are closely related are pre-
P(sE 0 H 0; )Q = (sE H) : sented. The rst and the second ones are actually direct
De ning w = s 1 a , we obtain applications of Theorem 2, that may present some interest
in applications. The third one develop a fact that has been
1 + aw
P( w E 0 H 0; )Q = w:( w E H) ; 1 + aw rst pointed out by Rosenbrock [15], that pole placement
through static state feedback and zero placement through
or equivalently the choice of the output equation are in some sense equiv-
alent.
e e
P(wE H ; e)Q = (wE H) ;
0 0 e e
where the matrices Ee0 ; He 0; E,
e and He are de ned in Lemma 4.1 Regularization of right invertible sys-
2 of the Appendix, and tems through non regular feedback
The implicit system (3) is said to be regular if it is square,
(wEe0 He 0; e)  (wEe H) e (4) q = n, and (sJ A BF) is invertible [10]. In that case
According to Lemma 2, if a is neither a root of (sE 0 the system (3) admits a unique solution x for every input
H ) nor of (sE H), then the matrix pencils (wEe0 He 0 )
0 u, providing that u is derivable enough. The question of
assigning a given pole structure to system (3) by static
and (wEe H) e have no in nite elementary divisors. Since state feedback, in addition to regularize it, was answered
(sE H) and (sE 0 H 0 ) are both right invertible, they no by Kucera and Zagalak [9]. In the terms of the present
longer have row minimal indices. Their invariant factors paper, it corresponds to the case where G = 0 and q = n,
are respectively but J is not necessarily monic.
If J and A are not square matrices, which means that
ei(w) = i ( 1 +waw )wdeg (s)wn ; i = 1; :::; q; (5)
i i
the system is constrained by some static relations, it
makes sense to choose the free parameter u as in (2) so where F 2 R(m+k)n and G 2 R(m+k)m , one obtains the
that a unique solution (x; v) is de ned by (3), and describe closed{loop system
the freedom in assigning the dynamics of the closed{loop :
system. Hence we shall extend to this case the concept of x= (A + BF)x + BGu (10)
a regular system, de ning (3) as a regular system if the y = (C + DF)x + DGu (11)
pencil (sJ A BF; BG) is square and invertible.
Corollary 1 Assume that system (1) is right invertible, The question is to design the feedback (9) so that the
closed{loop system (10){(11) is right invertible with a pre-
rankJ = n, q = m + n,and let i(s); i = 1; :::; q; ni; i = scribed structure.
1; :::; m, and ci ; i = 1; :::; k respectively denote the invari-
ant factors, in nite zero orders and column minimal in- Corollary 2 A feedback (9) exists so that the closed{loop
dices of the matrix pencil (sJ A; B). Let be given a system (10){(11) is right invertible if and only if the open{
list of monic polynomials 0i(s); i = 1; :::; q;, and a list of loop system (7){(8) is right invertible. In that case, m 
nonnegative integers n0i; i = 1; :::; m. p. Let i (s); i = 1; :::; n + p; ni; i = 1; :::; p, and ci ; i =
There exists a static state feedback (2) so that the 1; :::; m + k p, respectively denote the invariant factors,
closed{loop system (3) is regular, with invariant factors in nite zero orders, and column minimal indices of the
0i (s); i = 1; :::; q;, and in nite zero orders n0i ; i = 1; :::; m, system matrix
if and only if the conditions of Theorem 2 holds.  sI A B

(sE H) = n
Proof According to Lemma 3 of the Appendix, the C D
number of elements of each list of invariants is related to
the size and rank of the matrices involved in the problem. of system (7){(8), and let be given 0i (s); i = 1; :::; n +
The right invertibility of the open{loop system, together p; n0i; i = 1; :::; p and c0i; i = 1; :::; m p. There exists a
with the regularity of the closed{loop system imply that feedback (9) so that 0i(s); i = 1; :::; n+p; n0i; i = 1; :::; p and
q = n + m and that their Kronecker invariants are as c0i ; i = 1; :::; m p are respectively the invariant factors,
stated in the corollary. in nite zero orders, and column minimal indices of the
closed{loop system matrix of (10){11)
Remark 1 The condition q = n + m is not restrictive.  sI A BF BG 
The system being right invertible, it is readily seen that 0
(sE H ) = 0 n
q  n + m + k, hence the value of kand m, and the size of C + DF DG
the nonregular feedback G, are actually determined by the if and only if the conditions of Theorem 2 hold with q =
regularity of the closed{loop system. Note that this one has n + p.
no column minimal indices as it is invertible (and indeed
n+ m q = 0). Finally note that, under the hypothesis, a Proof Considering the singular system
regularizing feedback allways exists. One can for instance I : A B 
choose n0i = ni ; i = 1; :::; m, 0i(s) = i(s); i = 2; :::; q;, n
and 01 (s) = 1(s) (s), where (s) is any monic polyno- 0 x= C x + D u ;
mial such that
X k it appears that the question comes to the previously con-
deg (s) = ci ; sidered problem of assigning by static state feedback the
i=1 structure of an implicit system of the form (1) when J
and verify that the conditions of Theorem 2 hold. is monic. The number of rows of the matrix J is here
q = n + p.
4.2 Structure assignment of systems de-
scribed by a state and output equa- 4.3 Assignment of the tructure of strictly
tions through static state feedback proper systems through the choice of
Next, the main result is applied to the classic linear system
the output equation
described by a state equation Theorem 2 is now applied to characterize the freedom in
: assigning the structure of an input{output system choos-
x= Ax + Bu ; (7) ing the output equation while the state equation is given.
where A 2 Rnn, B 2 Rn(m+k) , and an output equation This question makes sense in the context of integrated de-
sign. Karcanias [7], [8] suggested that adequate structural
y = Cx + Du (8) properties should be researched, in early design, when se-
lecting the outputs and instrumentation. Indeed this se-
where C 2 Rpn, and D 2 Rp(m+k) . Applying the static lection sizably a ects the structure of the system, and
state feedback hence it determines the ends and speci cations a control
u = Fx + Gv (9) may permit to reach. It is in fact a fundamental step of
control design. Theorem 2 permits to handle the question Under the hypothesis, in particular that n0i  1; i =
in the case of right invertible strictly proper systems. 1; :::; p, we can assume that
Consider the linear system  0 0 
:
x= Ax + Bu (12) (sE 0 H 0) = sInC 0 A 0B
where A 2 Rnn,and B 2 Rnm , and consider the output where A0 ; B 0 , and C 0 are real matrices of convenient sizes,
equation and B 0 is monic. One can then verify that P and Q have
y = Cx (13) the form
p  n
where C 2 R . The problem is here choosing the matrix P P  0Q Q Q 1
C in (13) so that the overall system (12){(13) is right P = P1 P2 ; Q = @ Q4 Q5 Q6 A ;
1 2 3

invertible with a prescribed structure. 3 4


Q Q Q 7 8 9

Corollary 3 Assume that (12) is right invertible and B that P1 = Q1 ; P3 = 0; Q2 = 0; P1B 0 = BQ4 , that
is monic, and let i(s); i = 1; :::; n+p, and ci ; i = 1; :::; m,
respectively denote the invariant factors, and column min- P1; P4; Q1, and Q4 are invertible, and nally that
imal indices of the matrix pencil (sIn A; B). Let be  sI A B 
0 0 0 (sE 0 H 0 )  n
given i(s); i = 1; :::; n + p; ni  1; i = 1; :::; p and ci ; i = C 0Q1 1 0
1; :::; m p. There exists a matrix C as in (13) so that
0i (s); i = 1; :::; n+p;n0i; i = 1; :::; p, and c0i ; i = 1; :::; m p C = C 0Q1 1 provides a conclusion.
are respectively the invariant factors, in nite zero orders,
and column minimal indices of the right invertible system
matrix  sIn A
B
5 Conclusions
(sE 0 H 0) = C 0 Necessary and sucient conditions for the existence of a
if and only if the conditions of Theorem 2 hold, with q = nonregular satic state feedback that assigns the structure
n + p, k = p, and ni = 0; i = 1; : : :; p. of a given linear system to some prescribed values are
obtained when both the open{loop and the closed{loop
Proof As noted by Rosenbrock [15] in the case of ex- implicit systems are right invertible. However, non regular
plicit and controllable systems, and by Syrmos and Lewis feedback does not preserve right invertibility of a system
[16], this problem can be stated in terms of static state so that row minimal indices can appear in the closed{loop
feedback. The existence of C 2 Rpn such that the sys- system. In addition, we used as a crucial hypothesis that
tem is right invertible with prescribed structure comes to J is monic. The general case, where J is any matrix and
the existence of a nonregular feedback F = C 0 2 row minimal indices are also present, remains an open
Rp(n+m) ; G = 0 such that the structure of the augmented problem.
implicit system
 I 0  x:   A B  x   0  6 Appendix
n 0
0 0 u: = 0 0 u + Ip u
Lemma 2 [11, 13] Let sE H be a matrix pencil which
is as precribed. The pencil Kronecker invariants are
  { row minimal indices     :::  r  0 ,
(sE H) = sIn 0 A 0B I0p
1 2
{ column minimal indices     :::  s  0 ,
1 2
{ invariant factors (s)  (s)  :::  t(s)  1 ,
1 2

has the same invariant factors and column minimalindices { in nite zero orders n  n  :::  nz  1 ,
1 2

as the pencil (sIn A; B), and has also in nite zeros with where (s)  (s) means that (s) divides (s) without
orders ni = 1; i = 1; : : :; p. Taking (sE 0 H 0) as above, remainder. Note that z  t. Then the invertible transfor-
and hence q = n+p, and k = p, it appears that the condi- mation
tions of Theorem 2 hold true. Lemma 3 of the appendix, sE H ! wEe He ;
together with the hypothesis, permits to ensure that the where a is not a root of 1(s), and
invariant lists are as prescribed. Reversely, if these condi-
tions hold, every pencil (sE 0 H 0 ) having invariant factors wEe He = w( 1 +waw E H) = w(aE H) + E ;
0i (s); i = 1; :::; n + p, in nite zero orders n0i ; i = 1; :::; p,
and column minimal indices c0i ; i = 1; :::; m p, can be
completed to get a pencil equivalent to (sE H), thus transforms the matrix pencil (sE H) into the matrix
there exist matrices P; Q, and , so that pencil (wEe H)e whose Kronecker invariants are
{ row minimal indices 1  2  :::  r  0 ,
P (sE 0 H 0 ; ) = (sE H)Q { column minimal indices 1  2  :::  s  0 ,
{ invariant factors f1(w)  f2(w)  :::  et(w)  1 , [7] Karcanias N., \Global process instrumentation { Is-
where sues and problems of a system and control theory
ei(w) = i( 1 +waw )wdeg (s) wn ;
i i framework", Measurement, Vol. 14, pp. 103-113,
(1994).
and ni = 0, for i > z .
[8] Karcanias N., \The selection of input and output
The following describes how the number of elements of schemes for a system and the model projection prob-
the Kronecker lists of a matrix pencil sE H are related lems", Kybernetica, Vol. 30, pp. 585-596, (1994).
to its size and to the rank of matrix E. [9] Kucera V., Zagalak P., \Fundamental theorem of
Lemma 3 [13] Let #columns; #rows; #if; #ied; #cmi, state feedback for singular systems", Automatica,
and #rmi respectively denote the number of columns, of Vol. 24, pp. 653{658, (1988).
rows, of invariant factors, of in nite elementary divisors, [10] Loiseau J.J., Ozkaldiran K., Malabre M. and Karka-
of column minimal indices, and of row minimal indices of nias N., \Feedback canonical forms of singular sys-
a matrix pencil sE H . The following equalities hold. tems", Kybernetica, Vol. 27, pp. 289-305, (1991).
#columns = rankE + #ied#cmi ; [11] Loiseau J.J., \Pole placement and related problems",
Kybernetica, Vol. 28, pp. 90-99, (1992).
#rows = rankE + #ied + #rmi :
It implies in particular that [12] Malabre M., Kucera V., \In nite structure and exact
model matching problem : a geometric approach",
#cmi #rmi = #columns #rows: IEEE Transactions on Automatic Control, Vol. AC{
29, pp. 266-268, (1984).
In addition, if sE H is right invertible, which is the case
for all matrix pencils considered through the paper, then [13] Mondie S., \Contribucion al estudio de la modi -
cacion de estructura por medio de retroalimentacion
#rmi = 0: no regular", Tesis, CINVESTAV, Mexico, (1996).
[14] Morse A.S., \Structural invariants of linear multivari-
References able systems", SIAM Journal of Control, Vol. 11, pp.
446-465, (1973).
[1] Baraga~na I. Zaballa I., \Column completion of a pair [15] Rosenbrock H.H., State{space and Multivariable The-
of matrices", Linear and multilinear algebra, Vol. 27, ory, Wiley, New York, (1970).
pp. 243-273, (1990).
[2] Brunovsky P., \A classi cation of linear controlable [16] Syrmos V.L., Lewis F.L., \Transmission zero asign-
systems", Kybernetica, Vol. 6, pp. 173-188, (1970). ment using semistate descriptions", IEEE Transac-
tions on Automatic Control, Vol. 38, pp. 1115-1120,
[3] Descusse J., Dion J.M., \On the structure at in nity (1993).
of linear square decoupled systems", IEEE Transac- [17] Zaballa I., \Interlacing inequalities and control the-
tions on Automatic Control, Vol. AC{27, pp. 971- ory", Linear Algebra and its Applications, Vol. 101,
974, (1982). pp. 9-31, (1988).
[4] F.R. Gantmacher, Matrix Theory, Vol. I and II, [18] Zagalak P., Kucera V., Loiseau J.J, \Eigenstructure
Chelsea, New York, (1974). assignment in linear systems by state feedback", in
[5] Heymann M., \Controllability subspaces and feed- Polynomial Metods in Optimal Control and Filtering,
back simulation", SIAM Journal of Control, Vol. 14, IEEE Control Engineering Series 49, Peter Pelegri-
pp. 769-789, (1976). nus, London, (1993).
[6] Kailath T., Linear Systems, Prentice Hall, (1980). [19] Zagalak P., Loiseau J.J., \Eigenstructure assign-
ment in linear and uncontrollable systems", in Proc.
MTNS'93, Regensburg, Germany, (1993).

Anda mungkin juga menyukai