1930 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO.

10, OCTOBER 2007

intuition. Finally, the quadratic D0stability of an antiwindup closed Static Output Feedback Control Synthesis for Linear
loop system was analyzed as an application example. Systems With Time-Invariant Parametric Uncertainties
Jiuxiang Dong and Guang-Hong Yang, Senior Member, IEEE
REFERENCES
[1] R. E. Kalman, “Lyapunov functions for the problem of the Lur’e in
automatic control,” Proc. Nat. Acad. Sci., vol. 49, no. 2, pp. 201–205, Abstract—This technical note is concerned with the problem of designing
1963. robust static output feedback controllers for linear discrete and contin-
[2] R. E. Kalman, “On a characterization of linear passive systems,” in uous-time systems with time-invariant polytopic uncertainties. Sufficient
Proc. 1st Ann. Allerton Conf. Commun., Contr., Comput., 1963, pp. conditions for static output feedback stabilizing controller designs are given
456–470. in terms of solutions to a set of linear matrix inequalities. Furthermore, the
[3] V. A. Yakubovich, “The solution to certain matrix inequalities in auto- results are also extended to static output feedback controller designs.
matic control,” Soviet Math Dokl., vol. 3, pp. 620–623, 1962, in Rus- Numerical examples are given to illustrate the effectiveness of the proposed
sian. design methods.
[4] V. M. Popov, “Absolute stability of nonlinear systems of automatic
control,” Autom. Remote Control, vol. 22, no. 8, pp. 857–875, 1961. Index Terms—Linear matrix inequalities (LMIs), parameter dependent
[5] J. Willems, “Least squares stationary optimal control and the alge- Lyapunov functions, robust control, static output feedback (SOF), time-
braic Riccati equation,” IEEE Trans. Autom. Control, vol. 16, no. 6, invariant uncertainty.
pp. 621–634, 1971.
[6] A. Lindquist and G. Picci, “A geometric approach to modeling and
estimation of linear stochastic systems,” J. Math. Syst., Estimation and I. INTRODUCTION
Control, vol. 1, pp. 241–333, 1991.
[7] C. Langbort, R. S. Chandra, and R. D’Andrea, “Distributed control de- Static output feedback (SOF) control is one of the most important
sign for systems interconnected over an arbitrary graph,” IEEE Trans. open problems in control theory and practice. It represents the sim-
Autom. Control, vol. 49, no. 9, 2004.
[8] A. Megretski and A. Rantzer, “System analysis via integral quadratic plest closed-loop control system, which can be easily realized in prac-
constraints,” IEEE Trans. Autom. Control, vol. 42, no. 6, pp. 819–830, tice. Therefore, the problem has been extensively studied in the past
1997. decades and for the SOF control problem of linear systems, there are
[9] C.-Y. Kao, A. Megretski, and U. Jönsson, “Specialized fast algorithms
for IQC feasibility and optimization problems,” Automatica, vol. 40,
various approaches to deal with it, see the survey paper [1] and the ref-
no. 2, pp. 239–252, 2004. erences therein. Among these existing methods, we distinguish them
[10] L. Vandenberghe, V. R. Balakrishnan, R. Wallin, and A. Hansson, based on Riccati equations [2]–[4], convex approaches based on op-
“On the implementation of primal-dual interior-point methods for timization techniques [5] (condition in term of min-max algorithm),
semidefinite programming problems derived from the KYP lemma,”
in Proc. 42nd IEEE Conf. Decision Control, Maui, HI, Dec. 2003, pp. [6], [7] (convex programming procedures). Because the SOF problem
4658–4663. is nonconvex, iterative linear matrix inequality (ILMI) approaches are
[11] A. Rantzer, “On the Kalman-Yakubovich-Popov lemma,” Syst. Contr. exploited after it is expressed as a bilinear matrix inequality formu-
Lett., vol. 28, no. 1, pp. 7–10, 1996.
[12] T. Iwasaki and S. Hara, “Generalized KYP lemma: Unified frequency
lation, see [8]–[12]. Recently, many efforts have been done to obtain
domain inequalities with design applications,” IEEE Trans. Autom. sufficient linear matrix inequality (LMI) conditions for SOF controller
Control, vol. 50, no. 1, pp. 41–59, 2005. design. Although only sufficient, the solutions have the advantage of
[13] M. Chilali, P. Gahinet, and P. Apkarian, “Robust pole placement being linear and, hence, easily tractable by standard optimization tech-
in LMI regions,” IEEE Trans. Autom. Control, vol. 44, no. 12, pp.
2257–2270, 1999. niques. As a result, some significant advances have been achieved, see
[14] G. Grimm, J. Hatfield, I. Postlethwaite, A. R. Teel, M. C. Turner, and L. [13]–[21], and the references therein.
Zaccarian, “Antiwindup for stable linear systems with input saturation: Many practical systems are commonly required to have good robust-
An LMI-based synthesis,” IEEE Trans. Autom. Control, vol. 48, no. 9,
pp. 1509–1525, 2003. ness so that robust control problems become important research topics
[15] A. Graham, Kronecker Product and Matrix Calculus with Applica- in control theory. Recently, [22]–[30] have studied various robust con-
tions. Chichester, U.K.: Ellis Horwood, 1981. trol problems for linear discrete or continuous-time systems with un-
[16] S. Boyd, L. E. Ghaoui, E. Feron, and V. Balakrishnan, “Linear matrix certainties. Based on parameter dependent Lyapunov functions, [31]
inequalities in system and control theory,” Soc. Indust. Appl. Math.,
1994, Philadelphia, PA. and [32] provide a simple and easy-to-test robust stability condition for
[17] V. Balakrishnan and L. Vandenberghe, “Semidefinite programming du- discrete- and continuous-time uncertain linear systems with polytopic
ality and linear time-invariant systems,” IEEE Trans. Autom. Control, uncertainties, respectively. [33] provides robust stability evaluation for
vol. 48, no. 1, pp. 30–41, 2003.
[18] P. Kapasouris, M. Athans, and G. Stein, “Design of feedback control
systems for stable plants with saturating actuators,” in Proc. Conf. De- Manuscript received April 12, 2006; revised October 1, 2006; April 18, 2007.
cision Control, Austin, TX, Dec. 1988, pp. 469–479. Recommended by Associate Editor X. Henrion. This work was supported in part
[19] P. Gahinet, A. Nemirovski, A. J. Laub, and M. Chilali, LMI Control by the Program for New Century Excellent Talents in University (NCET-04-
Toolbox. Natick, MA: The MathWorks, Inc., 1995. 0283), the Funds for Creative Research Groups of China (No. 60521003), Pro-
[20] B. Hencey and A. Alleyne, “A static antiwindup compensator design gram for Changjiang Scholars and Innovative Research Team in University (No.
technique for robust regional pole placement,” in Proc. ASME Int. IRT0421), the State Key Program of National Natural Science of China by Grant
Mech. Eng. Conf. Expo., Chicago, IL, Nov. 2006, IMECE2006-14653. 60534010, the Funds of National Science of China by Grant 60674021, and by
the Funds of Ph.D. program of MOE, China by Grant 20060145019.
J. Dong is with the College of Information Science and Engineering, North-
eastern University, Shenyang, 110004, P. R. China (e-mail: dongjiuxiang@ise.
neu.edu.cn).
G.-H. Yang is with the College of Information Science and Engineering and
the Key Laboratory of Integrated Automation of Process Industry, Ministry of
Education, Northeastern University, Shenyang, 110004, P. R. China, (e-mail:
[email protected]; [email protected]).
Digital Object Identifier 10.1109/TAC.2007.906227

0018-9286/$25.00 © 2007 IEEE

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 10, OCTOBER 2007 1931

uncertain discrete-time linear systems with convex polytopic uncer- Assume that C2i ; 1  i  N , are of full row rank, and let invertible
tainties, which can concludes some recent results as a special case. In matrices Ti ; 1  i  N , such that
particular, the methods of separating Lyapunov matrix with dynamic
matrix by adding a slack variable in [34] and [35] for discrete and con-
tinuous-time systems, respectively, are noticeable. Moreover, a method C2i Ti = [I 0]; for 1  i  N: (3)
of introducing multi-slack variables is given in [36] for the robust sta-
bility analysis for uncertain polytopic linear discrete-time systems. In Remark 1: For each C2i , the corresponding Ti generally is not
particular, by introducing a parameter-independent slack variable with unique. A special Ti can be obtained by
lower-triangular structure, a less conservative LMI-based method of
designing robust static output feedback controller for linear systems
with time-invariant uncertainties is proposed in [37]. However, the ro- Ti = T(
C2i C2i C2i
T )01 C2i
? (4)
bust design requires that the considered system output matrix is fixed,
i.e., without uncertainties. For the case of both system output matrices where C2?i denotes an orthogonal basis for the null space of C2i .
and input matrices with uncertainties, few methods have been proposed Our aim is to design a static output feedback controller
to robust static output feedback controller design in the literatures.
This note will be concerned with the robust static output feedback
controller design for uncertain linear discrete and continuous- time
( ) = Ky(k);
u k for discrete time case (5a)
systems, where the considered uncertainties are time-invariant, poly- u(t) = Ky (t); for continuous time case (5b)
topic. In contrast to the results of [37], the new condition is appli-
cable for linear systems with the time-varying polytopic uncertainties, such that the resulting following closed-loop system (6a) [or (6b)]
which may simultaneously emerge on system output and input ma- is robustly stable or simultaneously meets H2 performance bound
trices. Moreover, on account of using the properties of the null space of requirement.
output matrices, a parameter-independent slack variable is introduced,
which is helpful for obtaining less conservative results than [37].
This technical note is organized as follows. In the next section, ( + 1) = (A + B2 K C2 )x(k) + B1 w(k)
x k
system description and some preliminaries are given. In Section III, z (k ) = (C1 + D12 K C2 )x(k ) (6a)
a robust static output feedback controller design method is proposed, x_ (t) = (A + B2 K C2 )x(t) + B1 w (t)
and extended to H2 guaranteed cost controller design. Section IV
z (t) = (C1 + D12 K C2 )x(t): (6b)
presents three examples to illustrate the effectiveness of the proposed
design methods. Finally, Section V concludes the paper.
Notation: denotes the set of real numbers, C ? denotes the matrix The following preliminary lemmas will be used in this sequel.
with maximal rank such that CC ? = 0. Lemma 1: [34]If there exist a symmetric matrix Q and a matrix S
such that

II. PROBLEM STATEMENT AND PRELIMINARIES Q 0 S 0 ST 3 0

(A + B2 K C2 )S 0Q <
Consider a linear time-invariant system (1) with polytopic uncertain-
ties described by state-space equations: then the discrete-time system (6a) with w(k) = 0 is asymptotically
stable.
Lemma 2: [35] If there exist a symmetric matrix X and a matrix V
[ ( )] = Ax(t) + B1 w(t) + B2 u(t)
x t
such that
z (t) = C1 x(t) + D12 u(t)
y ( t) = C 2 x ( t) (1)
0V 0 V T 3 3
(A + B2 K C2 )V + X 0X 3 < 0
where the symbol  [1] represents the derived operator for continuous- V 0 0X
time and the forward operator for discrete-time system. x(t) 2 n is
the state vector, u(t) 2 p is the control input, w(t) 2 m is the then the continuous-time system (6b) with w(t) = 0 is asymptotically
disturbance, y (t) 2 r is the measured output, and z (t) 2 q is the stable.
controlled output. The matrices [A]n2n ; [B1 ]n2m ; [B2 ]n2p ; [C1 ]q2n ; Lemma 3: [37]Assume that the system (1) is with C2i = C; i =
[C2 ]r2n ; [D12 ]q2p belong to the following uncertainty polytope: 1; . . . ; N , and T is an invertible matrix such that CT = [I 0]. If there
exist symmetric matrices Qi ; 1  i  N and matrices

= f([A]n2n; [B1 ]n2m ; [B2 ]n2p ; [C1 ]q2n ; [C2 ]r2n ; [D12 ]q2p )j
= S11 0 = [L1 0]
([A]n2n; [B1 ]n2m ; [B2 ]n2p ; [C1 ]q2n ; [C2 ]r2n ; [D12 ]q2p ) S
S21 S22
; L
N
= ([ ]n2n ; [B1i ]n2m ; [B2i ]n2p ; [C1i ]q2n ; [C2i ]r2n
i A i such that
i=1
N
2 [D12i ]q2p ); i  0; = 1g: Qi 0 S 0 S T 3
i (2)
T Ai T S + T 01 B2i L
01
0Qi 0 for 1  i  N
< ;
i=1
1932 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 10, OCTOBER 2007

then the discrete-time system (6a) with w(k) = 0 is robustly stable. Proof: From the structure of L; Si and (3), (10), we can obtain
Remark 2: Lemma 3 is a direct consequence of the results in [37].
Lemma 4: [37] If there exist symmetric matrices Q; Z and a matrix S11 0
S satisfying the following LMIs L = [KS11 0] = [K i S22
0]
S21 i
0 1
0]Ti Ti Si = KC 2i Ti Si :
Q 0 S 0 ST 3 3 = K [I
(A + B2 K C2 )S 0Q 3 <0
(C1 + D12 K C2 )S 0 0I Substituting L for KC 2i Ti Si in (8), then (8) can be rewritten as
0Z 3 < 0; Tr(Z ) < follows:
B1 0Q 0Ti Si 0 SiT TiT 3 3
Ai Ti Si + B2j KC2i Ti Si 0Qi + Jij11 3
then the discrete-time system (6a) is robustly stable with H2 norm less 12 T
p. Qj (Jij ) 0Qj + Jij22
< 0 1  i; j  N:
than
Lemma 5: [35] If there exist symmetric matrices X ; Z and a matrix
(11)

V satisfying the following LMIs:

Let Wi = (Ti Si )01 and pre- and postmultiplying (11) by diag[WiT II ]
0V 0 V T 3 3 3 and its transpose, it follows that:
(A + B2 K C2 )V + X 0X 3 3
(C1 + D12 K C2 )V 0 0I 3 <0
0Wi 0 WiT 3 3
V 0 0 0X Ai + B2j KC2i 0Qi + Jij11 3 <0 1  i; j  N :
12 T
0Z 3 < 0; Tr(Z ) < Qj W i (Jij ) 0Qj + Jij22
B1 0X Multiplying (12) by i j and summing them, then we have
(12)

then the continuous-time system (6b) is robustly stable with H2 norm

less than
p. 0W 0 W T 3 0 3 3
Remark 3: Lemma 4 and 5 can be obtained from [37, Theorem 1] A + B2 K C2 0Q + 0 <0 (13)
and [35, Theorem 3.3], respectively. The proofs are omitted here. QW 0 0Q 0 J
where A; B2 ; C2 are same as in (2), and
III. MAIN RESULTS
In this section, first, two LMI-based sufficient conditions for de- N N
signing robust static output feedback controller for discrete and con- W= i W i ; Q= i Q i
tinuous-time systems are given, respectively. Then, the results are ex- i=1 i=1
tended to robust H2 control. N N 11
Jij 3
J = i j 12 T
i=1 j =1 (Jij ) J ij22
A. Robust Static Output Feedback Controller Design
N N
The following theorem presents a robust static output feedback con- = i j Jij :
troller design for discrete-time systems. i=1 j =1
Theorem 1: (Discrete-time case) If there exist symmetric matrices
Qi ; Jij and matrices Si ; 1  i  N; L with
Let S = W 01 and pre- and postmultiply (13) by diag[S T II ] and its
S11 0 transpose, then we can obtain
Si = i S22 i ;
S21
L = [L1 0] 0S 0 S T 3 3 0 3 3
11
Jij 12
Jij ( A + B2 K C2 )S 0Q 3 + 0 < 0: (14)
Jij = 12 T
(Jij )
22
Jij
(7) Q 0 0Q 0 J
satisfying the following LMIs: On the other hand, pre- and postmultiplying (9) by 1 I 111 N I
0Ti Si 0 SiT TiT
and its transpose, it follows that:
3 3
Ai Ti Si + B2j L 0Qi + Jij11 3 <0 N N N N
T
12 T
Qj (Jij ) 0Qj + Jij22 i j Jij + i j Jij
1  i; j  N (8) i=1 j =1 i=1 j =1
T
[Jij ]N2N + [Jij ]N2N 0 N N N N
i j Jij = 2J  0:
(9)
= i j Jij + (15)
i=1 j =1 i=1 j =1
where Ti ; i = 1; . . . ; N satisfying (3) and are nonsingular, then (1)
with w(k) = 0 and From (14) and (15), then we have

K =
01
L1 S11 (10) 0S 0 S T 3 3
(A + B2 K C2 )S 0Q 3 <0
is robustly stable. Q 0 0Q
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 10, OCTOBER 2007 1933

which is equivalent to iv) It should be noted that Corollary 1 is obtained based on the
technique from [32]. If the conditions in Corollary 1 hold, then
choose

Q 0 S 0 ST 3 < 0:
(A + B2 K C2 )S 0Q
(16) I I
Jii =
0
0
I
Jij = 0 N 10 1 0
0
I
From (16) and Lemma 1, we can obtain that the system (1) is robustly
stable. Moreover, from (1), we can deduce that the matrices Ti Si are furthermore
positive-definite (not necessarily symmetric) which implies that the T
[Jij ]N2N + [Jij ]N2N
matrices Si , and implicitly S11 , are invertible because Ti are invert-
ible. Then L1 = KS11 admits the solution (10). Thus, the proof is I 0 N01 1 I 1 1 1 0 N01 1 I
complete. 0 1
N01 I I 1 1 1 0 N01 1 I
By combining Theorem 1 and the technique of [32], we can have the =2 .. .. .. .. 0
following corollary. . . . .
Corollary 1: If there exist symmetric matrices Qi ; 1  i  N and 0 N01 1 I 0 N01 1 I 1 1 1 I
matrices Si ; 1  i  N; L with the same structures as in (7), satisfying
therefore, the conditions of Theorem 1 hold, which shows that
the following LMIs:
the conditions of Theorem 1 are less conservative than those of
Corollary 1.
0Ti Si 0 SiT TiT 3 3 0 3 3 It should be noted that for each C2i , there may exist different choices
A i Ti Si + B 2 j L 0 Qi 3 < 0 0 I 3 of Ti satisfying (3). The following theorem shows that the feasibility
Qi 0 0Qi 0 0 I of the condition of Theorem 1 is independent of the choices of Ti .
1iN Theorem 2: If the conditions of Theorem 1 are feasible for some
0Ti Si 0 SiT TiT 3 3 0 3 3 Ti satisfying (3), then they are feasible for any T~i satisfying (3), i.e.,
Ai Ti Si + B2j L 0 Qi 3 <
1
I 3 C2i T~i = [I 0].
Qj 0Qj N 01 0
I Proof: Since Ti and T~i satisfy (3), Ci Ti = Ci T~i = [I 0], which
implies that [I 0]Ti01 = [I 0]T~i01 . Postmultiplying it by Ti , then we
0 0 0
1  i 6= j  N have

where Ti ; i = 1; . . . ; N , satisfying (3) and are nonsingular, then (1) [I 0] = [I

01 Ti :
0]T~i (17)
with w(k) = 0 and (10) is robustly stable.
Proof: By the technique similar to the ones used in [32] and using Denote
the method of Theorem 1, the proof is easily obtained and omitted.
i i
Hi = T~i01 Ti =
Remark 4: H11 H12
i) Theorem 1 presents new sufficient conditions for designing ro- i
H21 i
H22
bust static output feedback controllers for discrete-time systems
which are of LMIs and can be effectively solved via LMI Con- i = I; H12
then from (17), it follows that: H11 i = 0. Consider
trol Toolbox [38].
ii) Since C2i usually are not square matrices, then the null space of I 0 S11 0
C2i are not zero. Therefore, the introduced slack variable S can Ti Si = T~i Hi Si = T~i i i i i
H21 H22 S21 S22
be dependent on the uncertain parameter i , which is helpful for
S11 0
reducing the conservatism of the new proposed design condition. = T~i i i (18)
S~21 S~22
Compared to the result of [37], the new result is applicable for
system output matrix C2 () with uncertainties. But, the system
output matrix C2 () must be fixed in [37], i.e., C2 () is indepen- i = H21
where S~21 i S11 + H22
i S21
i , S~22
i = H22
i S22
i . Let
dent parameter i . Even if the system output matrix is not uncer-
tain, i.e., C2i = C21 (1 = 1; . . . ; N ), the new design condition
S11 0
also provides less conservative results than the existing design S~i = i i
method based on the technique of [32] (Corollary 1), which has S~21 S~22
been shown in Example 1.
iii) By considering the relations of parameter i , a type of slack vari- then (18) can be rewritten as Ti Si = T~i S~i . Therefore, if (8) holds for
Ti , then (8) holds for T~i , which implies that the conditions of Theorem
ables are introduced for less conservative stability conditions for
polytopic linear systems in [33]. In Theorem 1, the type of slack 1 is feasible for T~i . Thus, the proof is complete.
variables (i.e., Jij ) are also exploited. However, in order to ob- For continuous-time case, we have Theorem 3.
Theorem 3. (Continuous-Time Case): If there exist symmetric ma-
trices Xi ; Jij ; 1  i; j  N and matrices Vi ; 1  i  N; L with
tain convex conditions, the technique is only used for one part
of the global matrix inequality, [i.e., the type of slack variables
only emergence on the blocks (2,2), (2,3), (3,2), (3,3) of (8) and
(13)]. If the type of slack variables in [33] is applied to the global V11 0
Vi = i V22i ; L = [L1 0];
matrix inequality, less conservative result can be given. But, the V21
11 12
corresponding conditions will become nonlinear, and, hence, not Jij Jij
Jij = 12 T 22
convex. (Jij ) Jij
1934 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 10, OCTOBER 2007

satisfying the following LMIs: From Lemma 2, it follows that (1) is robustly stable. Moreover, from
(19), we can deduce that the matrices Ti Vi are positive-definite (not
0Ti Vi 0 ViT TiT 3 3 necessarily symmetric) which implies that the matrices Vi , and implic-
Ai Ti Vi + B2j L + Xj 0Xi + Jij11 3 itly V11 , are invertible because Ti are invertible. Then L1 = KV11
12 T
Ti Vi (Jij ) 0Xi + Jij22 admits the solution (21). Thus, the proof is complete.
< 01  i; j  N (19) Remark 5: Similar to Theorem 2, it can be shown that the feasibility
[Jij ]N2N + [Jij ]N2N
T 0 (20) of the condition of Theorem 3 is also independent of the choices of Ti .

B. Extension to H2 Control
where Ti ; i = 1; . . . ; N; satisfying (3) and are nonsingular, then the
continuous-time system (1) with w(t) = 0 and Here, we extend the results given earlier to H2 control.
Theorem: (Discrete-time case with H2 performance) If there exist
symmetric matrices Qi ; Zi ; Jij ; 1  i; j  N and matrices Si ; 1 
K =
01
L1 V11 (21) i  N; L with
is robustly stable.
S11 0
Proof: By virtue of the structure of L; Vi and (3), (21), we have Si = i S22 i ; L = [L1 0];
S21
L = KC2i Ti Vi , then (19) can be rewritten as follows: 11 12 13
Jij Jij Jij
12 T
0Ti Vi 0 ViT TiT 3 3 Jij = (Jij )
13 T
22
Jij
23 T
23
Jij
33
Ai Ti Vi + B2j KC2i Ti Vi + Xj 0Xi + Jij 11
3 (Jij ) (Jij ) Jij
12 T
Ti Vi (Jij ) 0Xi + Jij22
< 0; 1  i; j  N: (22) satisfying the following LMIs:

Let Wi = (Ti Vi )01 and pre- and postmultiplying (22) by diag[WiT II ] 0Ti Si 0 SiT TiT 3 3 3
and its transpose, then we can obtain Ai Ti Si + B2j L 0Qi + Jij11
3 3
12 T
C1i Ti Si + D12j L (Jij ) 0I + Jij22 3
0Wi 0 WiT 3 3 Qj 13 T
(Jij )
23 T
(Jij ) 0Qj + Jij33
Ai + B2j KC2i + Xj Wi 0Xi + Jij 11
3 < 0: < 0; 1  i; j  N
12 T
I (Jij ) 0Xi + Jij22 0 Z i 3 < 0; 1  i  N
B1i 0Qi
Multiplying the above inequality by i j and summing them, then we Tr(Zi ) < ; 1  i  N
have
[Jij ]N2N + [Jij ]N2N
T 0
0W 0 W T 3 3 0 3 3
A + B2 K C2 + XW 0X 3 + 0 <0 (23) where Ti ; i = 1; . . . ; N; satisfying (3) and are nonsingular, then the
I 0 0X 0 J static output feedback control law (5a) with K = L1 S11 01 renders H2
norm of (1) less than
p .
where A; B2 ; C2 are same as in (2), and Proof: By using Lemma 4, it is similar to the proof of Theorem
1. The details are omitted here.
N N
Theorem 5: (Continuous-time case with H2 performance) If there
W= i W i ; X= i X i
exist symmetric matrices Xi ; Zi ; Jij ; 1  i; j  N and matrices
i=1 i=1
N N Vi ; 1  i  N; L with
J i j
11
Jij 3
= 12 T
i=1 j =1 (Jij ) J ij22 V11 0
N N Vi = i V22i ; L = [L1 0];
V21
= i j Jij : 11
Jij 12
Jij 13
Jij
i=1 j =1 Jij =
12 T
(Jij )
22
Jij 23
Jij
13 T 23 T 33
On the other hand, by (20) and the arguments similar to those for the (Jij ) (Jij ) Jij
proof of Theorem 1, we have J > 0. Combining this condition and
(23), it follows that: satisfying the LMIs at the bottom of the next page, where Ti ; i =
1; . . . ; N; satisfying (3) and are nonsingular, then the static output feed-
0W 0 W T 3 3 back control law (5b) with K = L1 V11 01 renders H2 norm of (1) less
A + B2 K C2 + XW 0X 3 < 0: (24) than
p .
I 0 0X Proof: By using Lemma 5, it is similar to the proof of Theorem
Let V = W 01 and pre- and postmultiplying (24) by diag[V T II ] and 3. The details are omitted here.
its transpose, then we can obtain
IV. EXAMPLE
0V 0 V T 3 3 In this section, three numerical examples are presented to illustrate
(A + B2 K C2 )V + X 0X 3 < 0: the effectiveness of the proposed methods. In Example 1, a comparison
V 0 0X between Theorem 1, Corollary 1 and Lemma 3 is given. The design
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 10, OCTOBER 2007 1935

methods given by Theorems 4 and 5 are illustrated by Examples 2 and TABLE I

3, respectively. CONTROLLER GAIN AND PERFORMANCE H
Example 1: Consider a discrete-time system which belongs to the
2-polytopic convex polyhedron in the form of (2) with w(k) = 0 and

1:2 1 :1 TABLE II
A1 =
1
2
0 0: 7 A2 =
0
1
0 0: 6 B21 =
1
1 :1
CONTROLLER GAIN AND PERFORMANCE H
1
B22 = C21 = C22 = [1 0]
1

For this example with the fixed output matrix, Lemma 3, Theorem 1,
Example 3: Consider a continuous-time system which belongs to
and Corollary 1 are applicable for designing robust static output feed-
the 2-polytopic convex polyhedron in the form of (2) with
back controllers. The LMIs of Lemma 3 are infeasible, and the ones
of Corollary 1 are also infeasible. However, by Theorem 1, a stabi-
lizing controller gain can be obtained as K = 00:7589. Therefore,
A1
1 2
; A2
2 01 ; B11
2
the method given by Theorem 1 provides a better alternative design for
this example.
=
0 04 =
0 05 =
1
1 1 1
Example 2: Consider a discrete-time system which belongs to the B12 = B21 = ; B22 =
1 0 1
2-polytopic convex polyhedron in the form of (2) with
C11 = [1 2]; C12 = [1 1] C21 = [1 0]

A1 =
1 :6 0
; A2 =
1 :1 0 0: 3 ; B11 =
0 :6 C22 = [2 1]; D121 = 1; D122 = 2:
1 0 :7 0 0 :1 0 :3
0 :5 1 1 :5 By (4), we can obtain
B12 = B21 = ; B22 = ; C11 = [1 0:3]
0:4 1 1
C12 = [1:2 0:5] C21 = [1 0 0:1]; C22 = [1 0:1] 1 0 0:4000 00:4472
D121 = 1; D122 = 0:9 T1 = ; T2 =
0 1 0:2000 0:8944

By (4), we can obtain satisfying (3).

The guaranteed H2 controllers are obtained by Theorem 5 with
and without the slack variables Jij , respectively, which are given in
Table II. The computation results also show that the design with the
T1
0:9901 0:0995
; T2
0:9901 00:0995 slack variables Jij can reduce conservatism.
=
00:0990 0:9950
=
0:0990 0:9950

satisfying (3). V. CONCLUSION

By applying Theorem 4 with and without the slack variables Jij , the
guaranteed H2 performance
p and the corresponding static output In this technical note, the problem of designing robust SOF con-
trollers for linear discrete and continuous-time systems with time-in-
feedback gain K are given in Table I, respectively. The computation variant polytopic uncertainties has been investigated. LMI-based suf-
results show that the design method given by Theorem 4 is less conser- ficient conditions for SOF stabilizing controller design are given, and
vative than the one given by Theorem 4 without the slack variables Jij the results are also extended to H2 SOF controller design. Compared
(i.e., Jij = 0). to existing results, the new proposed approach is applicable for linear

0Ti Vi 0 ViT TiT 3 3 3

Ai Ti Vi + B2j L + Xj 0Xi + Jij
11
3 3
12 T <0
C1i Ti Vi + D12j L (Jij ) 0I + Jij22 3
13 T T
Ti Vi (Jij )
23
(Jij ) 0Xi + Jij 33

1  i; j  N
0Zi 3 < 0; 1  i  N
B1i 0Xi
Tr(Zi ) < ; 1  i  N

[Jij ]N2N + [Jij ]N2N

T 0
1936 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 10, OCTOBER 2007

system with time-invariant polytopic uncertainties, which may simul- [21] C. E. de Souza and A. Trofino, “An LMI approach to stabilization
taneously emerge on system output and input matrices. In the technical of linear discrete-time periodic systems,” Int. J. Control, vol. 73, pp.
696–703, 2000.
development, the properties of the null space of system output matrices [22] P. Gahinet, P. Apkarian, and M. Chilaly, “Affine parameter depen-
are exploited and a parameter-dependent slack variable is introduced to dent Lyapunov functions and real parametric uncertainty,” IEEE Trans.
separate system matrix and Lyapunov matrix, which are helpful for re- Autom. Control, vol. 41, pp. 436–442, 1996.
[23] U. Shaked, “Improved LMI representations for the analysis and the
ducing the conservatism of the obtained conditions for designing SOF design of continuous-time systems with polytopic type uncertainty,”
controllers. The numerical examples have shown the effectiveness of IEEE Trans. Autom. Control, vol. 46, pp. 652–656, 2001.
the proposed design methods. [24] J. C. Geromel and P. Colaneri, “Robust stability of time varying poly-
topic systems,” Syst. Control Lett., vol. 55, pp. 81–85, 2006.
[25] J. C. Geromel, M. C. de Oliveira, and J. Bernussou, “Robust filtering of
discrete-time linear systems with parameter dependent lyapunov func-
tions,” SIAM J. Control Optim., vol. 41, pp. 700–711, 2002.
[26] Y.-Y. Cao and Z. Lin, “A descriptor system approach to robust stability
REFERENCES analysis and controller synthesis,” IEEE Trans. Autom. Control, vol. 49,
pp. 2081–2084, 2004.
[27] D. Peaucelle, D. Arzelier, O. Bachelier, and J. Bernussou, “A new ro-
[1] V. L. Syrmos, C. T. Abdallah, and K. Grigoriadis, “Static output feed- bust D-stability condition for real convex polytopic uncertainty,” Syst.
back-a survey,” Automatica, vol. 33, pp. 125–137, 1997. Control Lett., vol. 40, pp. 21–30, 2000.
[2] A. Trofino-Neto and V. Kucera, “Stabilization via static output feed- [28] L. Lu, R. Yang, and L. Xie, “Robust H and H control of dis-
back,” IEEE Trans. Autom. Control, vol. 38, pp. 764–765, 1993. crete-time systems with polytopic uncertainties via dynamic output
[3] V. Kucera and C. de Souza, “A necessary and sufficient condition for feedback,” in Proc. Amer. Contr. Conf., Portland, OR, 2005, pp.
output feedback stabilizability,” Automatica, vol. 10, pp. 1357–1359, 4315–4319.
1995. [29] M. C. de Oliveira and J. C. Geromel, “A class of robust stability condi-
[4] G. Garcia, B. Pradin, and F. Zeng, “Stabilization of discrete time linear tions where linear parameter dependence of the Lyapunov function is a
systems by static output feedback,” IEEE Trans. Autom. Control, vol. necessary condition for arbitrary parameter dependence,” Syst. Control
46, pp. 1954–1958, 2001. Lett., vol. 54, pp. 1131–1134, 2005.
[5] J. C. Geromel, C. C. deSouza, and R. E. Skelton, “Static output feed- [30] V. J. S. Leite and P. L. D. Peres, “A improved LMI condition for ro-
back controllers: Stability and convexity,” IEEE Trans. Autom. Control, bust D-stability of uncertain polytopic systems,” IEEE Trans. Autom.
vol. 43, pp. 120–125, 1998. Control, vol. 48, pp. 500–504, 2003.
[6] D. Henrion and J. B. Lasserre, “Convergent relaxations of polynomial [31] D. C. W. Ramos and P. L. D. Peres, “A less conservative LMI condi-
matrix inequalities and static output feedback,” IEEE Trans. Autom. tion for the robust stability of discrete-time uncertain systems,” Syst.
Control, vol. 51, pp. 192–202, 2006. Control Lett., vol. 43, pp. 371–378, 2001.
[7] J. C. Geromel, P. L. D. Peres, and S. R. Souza, “Convex analysis of [32] D. C. W. Ramos and P. L. D. Peres, “An LMI condition for the robust
output feedback control problems: Robust stability and performance,” stability of uncertain continuous-time linear systems,” IEEE Trans.
IEEE Trans. Autom. Control, vol. 41, no. 7, pp. 997–1003, 1996. Autom. Control, vol. 47, pp. 675–678, 2002.
[8] L. E. Ghaoui, F. Oustry, and M. AitRami, “A cone complemen- [33] S.-W. Kau, Y.-S. Liu, L. Hong, C.-H. Lee, C.-H. Fang, and L. Lee,
tarity linearization algorithm for static output-feedback and related “A new LMI condition for robust stability of discrete-time uncertain
problems,” IEEE Trans. Autom. Control, vol. 42, pp. 1171–1175, systems,” Syst. Control Lett., vol. 54, pp. 1195–1203, 2005.
1997. [34] M. C. de Oliveira, J. Bernussou, and J. C. Geromel, “A new dis-
[9] Y.-Y. Cao, J. Lam, and Y. X. Sun, “Static output feedback stabilization: crete-time robust stability condition,” Syst. Control Lett., vol. 37, pp.
An ILMI approach,” Automatica, vol. 34, pp. 1641–1645, 1998. 261–265, 1999.
[10] F. Leibfritz, “An LMI-based algorithm for designing suboptimal static
H =H output feedback controllers,” SIAM. J. Control Optim., vol.
[35] P. Apkarian, T. H. Duong, and J. Bernussou, “Continuous-time anal-
ysis, eigenstructure assignment, and H synthesis with enhanced linear
39, pp. 1711–1735, 2001. matrix inequalities (LMI) characterizations,” IEEE Trans. Autom. Con-
[11] D. Rosinova, V. Vesely, and V. Kucera, “A necessary and sufficient trol, vol. 46, pp. 1941–1946, 2001.
condition for static output feedback stabilizability of linear discrete- [36] J. Daafouz and J. Bernussou, “Parameter dependent Lyapunov func-
time systems,” KYBERNETIKA, vol. 39, pp. 447–459, 2003. tions for discrete time systems with time varying parametric uncertain-
[12] D. Huang and S. K. Nguang, “Robust H static output feedback con- ties,” Syst. Control Lett., vol. 43, pp. 355–359, 2001.
trol of fuzzy systems: An ILMI approach,” IEEE Trans. Syst., Man. [37] M. C. de Oliveira, J. C. Geromel, and J. Bernussou, “Extended H
Cybern. B, Cybern., vol. 36, pp. 216–222, 2006. and H characterizations and controller parameterizations for dis-
[13] U. Shaked, “An LPD approach to robust H and H static output- crete-time systems,” Int. J. Control, vol. 75, pp. 666–679, 2002.
feedback design,” IEEE Trans. Autom. Control, vol. 48, pp. 866–872, [38] P. Gahinet, A. Nemirovski, A. J. Laub, and M. Chilali, LMI Control
2003. Toolbox. Natick, MA: The MathWorks, 1995.
[14] E. Prempain and I. Postlethwaite, “Static output feedback stabilization
with H performance for a class of plants,” Syst. Control Lett., vol.
43, pp. 159–166, 2001.
[15] G. I. Bara and M. Boutayeb, “Static output feedback stabilization
with H performance for linear discrete-time systems,” IEEE Trans.
Autom. Control, vol. 50, pp. 250–254, 2005.
[16] D. Peaucelle and D. Arzelier, “Ellipsoidal sets for resilient and ro-
bust static output-feedback,” IEEE Trans. Autom. Control, vol. 50, pp.
899–904, 2005.
[17] R. E. Benton and D. A. Smith, “A static-output-feedback design proce-
dure for robust emergency lateral control of a highway vehicle,” IEEE
Trans. Contr. Syst. Technol., vol. 13, pp. 618–623, 2005.
[18] J. C. Lo and M. L. Lin, “Robust H nonlinear control via fuzzy static
output feedback,” IEEE Trans. Circuits Syst. I, Fundam. Theory Appl.,
vol. 50, pp. 1494–1502, 2003.
[19] D. W. C. Ho and G. P. Lu, “Robust stabilization for a class of dis-
crete-time non-linear systems via output feedback: The unified LMI
approach,” Int. J. Control, vol. 76, pp. 105–115, 2003.
[20] C. A. R. Crusius and A. Trofino, “Sufficient LMI conditions for output
feedback control problems,” IEEE Trans. Autom. Conttol, vol. 44, pp.
1053–1057, 1999.