IMA Journal of Mathematical Control and Information Advance Access published January 27, 2011

IMA Journal of Mathematical Control and Information Page 1 of 21

doi:10.1093/imamci/dnq035

Exponential stability and robust H∞ control for uncertain discrete

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
switched systems with interval time-varying delay

C HANG -H UA L IEN∗ , K ER -W EI Y U, Y EONG -JAY C HUNG AND H AO -C HIN C HANG

Department of Marine Engineering, National Kaohsiung Marine University, Kaohsiung,
Taiwan 811, Republic of China
∗ Corresponding author: [email protected]

L ONG -Y EU C HUNG
Department of Applied Geoinformation, Chia Nan University of Pharmacy & Science,
Tainan, Taiwan 717, Republic of China
AND

J ENQ -D ER C HEN
Department of Electronic Engineering, National Kinmen Institute of Technology, Kinmen,
Taiwan 892, Republic of China
[Received on 7 December 2009; revised on 30 November 2010;
accepted on 17 December 2010]

The global exponential stability and robust H∞ control for discrete switched systems with interval time-
varying delay under an arbitrary switching signal is considered in this paper. Linear Matrix Inequality
(LMI)-based conditions are proposed to guarantee the global stability and H∞ control for the switched
system with parametric perturbations. Some non-negative inequalities are used to reduce the conserva-
tiveness of systems. Finally, some numerical examples are illustrated to show the main results.

Keywords: global exponential stability; discrete switched system; interval time-varying delay; H∞
control; additional non-negative inequality.

1. Introduction
Switched systems are composed by several subsystems and use the switching signal to specify which
subsystem is activated to the system trajectories at each instant of time. Some examples for switched sys-
tems are automated highway systems, automotive engine control system, chemical process, constrained
robotics, power systems and power electronics, robot manufacture and stepper motors (Sun et al., 2008;
Xie et al., 2008; Zhai et al., 2006). It is also well known that the existence of delay in a system may cause
instability or bad performance in closed control systems (Gu et al., 2003; Hale & Verduyn lunel, 1993;
Kolmanoskii & Myshkis, 1992). Time-delay phenomena are usually encountered in many practical sys-
tems, such as AIDS epidemic, aircraft stabilization, chemical engineering systems, hydraulic systems,
inferred grinding model, manual control, neural network, nuclear reactor, population dynamic model,
rolling mill, ship stabilization and systems with lossless transmission lines. Hence, stability analysis and
stabilization for continuous or discrete switched systems with time delay have received some attention

c The author 2011. Published by Oxford University Press on behalf of the Institute of Mathematics and its Applications. All rights reserved.
2 of 21 C.-H. LIEN ET AL.

by many researchers in recent years (Han & Tang, 2007; Ibrir, 2008; Liu et al., 2008; Ma et al., 2006;
Mahmoud, 2008; Song et al., 2008; Sun et al., 2006, 2007, 2008; Zhang et al., 2008).
It is interesting to note that the stable property for each subsystem cannot imply that the overall sys-
tem is also stable under arbitrary switching signal (Liu et al., 2008). On the other hand, the H∞ control

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
concept was proposed to reduce the effect of the disturbance input on the regulated output to within a
prescribed level (Han & Tang, 2007; Ma et al., 2006; Mahmoud, 2008; Song et al., 2008; Sun et al.,
2007). Hence, the global exponential stability and robust H∞ control problems for uncertain switched
discrete systems with interval time-varying delay under arbtrary switching signal will be considered in
this paper. In the past, there were many approaches been proposed to guarantee the stability for switched
discrete time-delay systems. Free weighting (slack) matrix approach is applied to slove memoryless
state-feedback H∞ control problem in Han & Tang (2007) and filtering problem in Mahmoud (2008).
Lyapunov theory is provided to design the memory state-feedback H∞ control problem (Ma et al., 2006)
and memoryless state-feedback H∞ control problem (Ibrir, 2008) for switched systems with constant
delay. In Song et al. (2008), the average dwell-time method is provided to design the state-feedback
H∞ control problem for switched systems with constant delay. In Sun et al. (2006, 2007), the descrip-
tor system approach had been used to stabilize the considered switched discrete time-delay system. In
Zhang et al. (2008), the robust stability and stabilization for discrete switched systems with interval
time-varying delay had been considered. In this paper, some non-negative inequalities and Lyapunov
theory are proposed to guarantee the global exponential stability and memoryless state-feedback H∞
control problem for uncertain switched systems with interval time-varying delay. Delay-dependent suf-
ficient conditions are proposed to guarantee the global exponential stability and H∞ controller design.
Our proposed approach is a new method and easy to apply to more general switched systems and other
type controller design (Lien et al., 2010). Some numerical examples are provided to demonstrate the use
of obtained results. From the simulation results, our proposed approach will provide less conservative
and useful results.
The notation used throughout this paper is as follows. For a matrix A, we denote the transpose by A> ,
spectral norm by kAk, symmetric positive (negative) definite by A > 0 (A < 0), maximal eigenvalue
by λmax (A), minimal eigenvalue by λmin (A), n × m dimension by A(n×m) . A 6 B means that matrix
B − A is symmetric positive semi-definite. I denotes the identity matrix. For a vector x, we denote the
Euclidean norm by kxk.PDefine N̄ = {1, 2, . . . , N }, kxk ks = maxθ =−r M ,−r M +1,...,0 kx(k + θ)k, and
L 2 (α, 0, ∞) = w(k): ∞ 2k >
k=0 α w (k)w(k) < ∞ , for 0 < α < 1.

2. Stability analysis for uncertain discrete switched system

Consider the following discrete switched system with interval time-varying delay:

x(k + 1) = [Aσ + 1Aσ (k)]x(k) + [Bσ + 1Bσ (k)]x(k − r (k)), k = 0, 1, 2, 3, . . . , (1a)

x(θ) = φ(θ), θ = −r M , −r M + 1, . . . , 0, (1b)

where x(k) ∈ <n , xk is the state at time k defined by xk (θ) := x(k + θ), ∀ θ ∈ {−r M , −r M + 1, . . . , 0},
σ (k, x(k)) is a switching signal which is a piecewise constant function and may depend on k or x, φ(k) ∈
Rn is an initial state function, time-varying delay r (k) is a function from {0, 1, 2, 3, . . .} to {0, 1, 2,
3, . . .} and 0 < rm 6 r (k) 6 r M , rm and r M are two given positive integers. Switching signal σ takes
its values in the finite set {1, 2, . . . , N }. Matrices Ai , Bi ∈ <n×n , i = 1, 2, . . . , N , are constant. The
perturbed matrices 1 Ai (k) and 1Bi (k), i = 1, 2, . . . , N , are some perturbed matrices and satisfy the
 DISCRETE SWITCHED SYSTEMS 3 of 21

following conditions:

[1Ai (k) 1Bi (k) ] = Mi Fi (k)[ N Ai N Bi ], ∀ i ∈ N̄ , k = 0, 1, 2, 3, . . . , (1c)

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
where Mi , N Ai and N Bi , i = 1, 2, . . . , N , are some given constant matrices with appropriate dimen-
sions, and Fi (k), i = 1, 2, . . . , N , are unknown matrices representing the parameter perturbations which
satisfy
Fi > (k)Fi (k) 6 I, ∀ i ∈ N̄ , k = 0, 1, 2, 3, . . . . (1d)

L EMMA 1 (Lien et al., 2009). Let U , V , W and M be real matrices of appropriate dimensions with M
satisfying M = M > , then

M + U V W + W > V > U > < 0 for all V > V 6 I,

if and only if there exists a scalar ε > 0 such that

M + ε −1 · UU > + ε · W > W = M + ε −1 · UU > + ε −1 · (εW )> (εW ) < 0.

 
W11 W12
L EMMA 2 (Schur complement of Boyd et al., 1994). For a given matrix W = > W with
> , W = W > , then the following conditions are equivalent:
W12 22
W11 = W11 22 22
−1 >
(1) W < 0, (2) W22 < 0, W11 − W12 W22 W12 < 0.
T HEOREM 1 System (1) is globally exponentially stable with convergence rate 0 < α < 1, if there
exist some n × n matrices P j > 0, j = 1, 2, . . . , N , Q > 0, R1 > 0, R2 > 0, R3 > 0, S > 0, T > 0,
Vi11 ∈ <4n×4n , Vi12 ∈ <4n×n , Vi22 ∈ <n×n , i = 1, 2, 3, 4, positive constants ε j , j = 1, 2, . . . , N , such
that the following LMI conditions hold for all j, l = 1, 2, . . . , N ,

α 2r M R1 − V122 > 0, α 2r M R2 − V222 > 0, α 2rm R3 − V322 > 0,

α 2r M R1 − V422 > 0, V411 − V111 > 0, (2a)

     
V111 V112 V211 V212 V311 V312
V1 = > 0, V2 = > 0, V3 = > 0,
∗ V122 ∗ V222 ∗ V322
 
V411 V412
V4 = > 0, (2b)
∗ V422

 
Σ11 0 0 0 Σ15 Σ16 0 ε j · N A>j
 ∗
 Σ22 0 0 Σ25 Σ26 0 ε j · N B>j 

 ∗ ∗ Σ33 0 0 0 0 0 
 
 ∗ ∗ ∗ 
Σ̄¯ = 
 ∗
Σ44 0 0 0 0 

 ∗ ∗ ∗ Σ55 0 Σ57 0 
 ∗ ∗ ∗ ∗ ∗ Σ66 Σ67 0 
 
 ∗ ∗ ∗ ∗ ∗ ∗ −ε j · I 0 
∗ ∗ ∗ ∗ ∗ ∗ ∗ −ε j · I
4 of 21 C.-H. LIEN ET AL.

 
Ω(4n×4n) 0(4n×n) 0(4n×n) 0(4n×n) 0(4n×n)
 ∗ 0(n×n) 0(n×n) 0(n×n) 0(n×n) 
 
+
 ∗ ∗ 0(n×n) 0(n×n) 0(n×n)  < 0, (2c)
 ∗ ∗ ∗ 0(n×n) 

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
0(n×n)
∗ ∗ ∗ ∗ 0(n×n)

where

Σ11 = −P j + T + (r M − rm + 1) · Q,

Σ15 = A>j Pl , Θ = r M · (R1 + R2 ) + rm · R3 , Σ16 = (A j − I )> Θ,

Σ22 = −α 2r M · Q, Σ25 = B >

j Pl , Σ26 = B >
j Θ, Σ57 = Pl M j , Σ67 = Θ M j ,
Σ33 = −α 2r M S, Σ44 = −α 2rm · (T − S),

Σ55 = −α 2 · Pl , Σ66 = −[r M · (R1 + R2 ) + rm · R3 ],

Ω = r M · V211 + rm · V311 + r M · V411 − rm · (V411 − V111 ) + V112 Λ1 + Λ> >

1 V112

+ V212 Λ2 + Λ> > > > > >

2 V212 + V312 Λ3 + Λ3 V312 + V412 Λ4 + Λ4 V412 ,

Λ1 = [ I −I 0 0 ](n×4n) , Λ2 = [ I 0 −I 0 ](n×4n) ,

Λ3 = [ I 0 0 −I ](n×4n) , Λ4 = [0 I −I 0 ](n×4n) .

Proof. Define the Lyapunov functional

k−1
X −rm
X k−1
X
V (xk ) = α −2k x > (k)Pσ (k,x(k)) x(k) + α −2i x > (i)Qx(i) + α −2i x > (i)Qx(i)
i=k−r (k) j=−r M +1 i=k+ j

0
X k−1
X 0
X k−1
X
+ α −2i y > (i)[R1 + R2 ]y(i) + α −2i y > (i)R3 y(i)
j=−r M +1 i=k−1+ j j=−rm +1 i=k−1+ j

k−1−r
Xm k−1
X
+ α −2i x > (i)Sx(i) + α −2i x > (i)T x(i), (3)
i=k−r M i=k−rm

where Pσ (k,x(k)) > 0, Q > 0, R1 > 0, R2 > 0, R3 > 0, S > 0, T > 0 and y(i) = x(i + 1) − x(i). The
difference of Lyapunov functional (3) along the solutions of system (1) has the form

1V (xk ) = V (xk+1 ) − V (xk )

= α −2k · [α −2 · x > (k + 1)Pσ (k+1,x(k+1)) x(k + 1) − x > (k)Pσ (k,x(k)) x(k)]
k
X k−1
X
−2i >
+ α x (i)Qx(i) − α −2i x > (i)Qx(i)
i=k+1−r (k+1) i=k−r (k)
 DISCRETE SWITCHED SYSTEMS 5 of 21

k−r
Xm
+ α −2k · (r M − rm ) · x > (k)Qx(k) − α −2i x > (i)Qx(i)
i=k+1−r M

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
k−1
X
+ α −2k · r M · y > (k)[R1 + R2 ]y(k) − α −2i y > (i)[R1 + R2 ]y(i)
i=k−r M

k−1
X
+ α −2k · rm · y > (k)R3 y(k) − α −2i y > (i)R3 y(i)
i=k−rm

+ α −2k · [α 2rm · x > (k − rm )Sx(k − rm ) − α 2r M · x > (k − r M )Sx(k − r M )]

+ α −2k · [x > (k)T x(k) − α 2rm · x > (k − rm )T x(k − rm )]. (4)

By some simple derivations, we have

k
X k−1
X
α −2i x > (i)Qx(i) − α −2i x > (i)Qx(i)
i=k+1−r (k+1) i=k−r (k)

k−1
X k−1
X
= α −2k x > (k)Qx(k) + α −2i x > (i)Qx(i) − α −2i x > (i)Qx(i)
i=k+1−r (k+1) i=k−r (k)+1

− α −2(k−r (k)) x > (k − r (k))Qx(k − r (k))

k−1
X k−1
X
6 α −2k x > (k)Qx(k) + α −2i x > (i)Qx(i) − α −2i x > (i)Qx(i)
i=k+1−r M i=k+1−rm

− α −2(k−r (k)) x > (k − r (k))Qx(k − r (k))

= α −2k x > (k)Qx(k) − α −2(k−r (k)) x > (k − r (k))Qx(k − r (k))
k−r
Xm
+ α −2i x > (i)Qx(i).
i=k+1−r M

Assume σ (k + 1, x(k + 1)) = l ∈ N̄ and σ (k, x(k)) = j ∈ N̄ . Then we can obtain the following result
from (4)

1V (xk ) = V (xk+1 ) − V (xk )



6 α −2k · α −2 · x > (k + 1)Pl x(k + 1) − x > (k)[P j − T ]x(k)


+ x > (k)[(r M − rm + 1) · Q]x(k) − α 2r M · x > (k − r (k))Qx(k − r (k))

+ [x(k + 1) − x(k)]> [r M · (R1 + R2 ) + rm · R3 ][x(k + 1) − x(k)]
6 of 21 C.-H. LIEN ET AL.

 
k−1
X (k)−1
k−rX k−1
X
− α 2r M ·  y > (i)R1 y(i) + y > (i)R1 y(i) + y > (i)R2 y(i)
i=k−r (k) i=k−r M i=k−r M
 

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
k−1
X
− α 2rm ·  y > (i)R3 y(i) − α 2rm · x > (k − rm )[T − S]x(k − rm )
i=k−rm


−α 2r M · x > (k − r M )Sx(k − r M ) . (5)


Define
X > (k) = [ x > (k) x > (k − r (k)) x > (k − r M ) x > (k − rm ) ].
Pk−1
By the system (1), LMIs in (2b), i=k−r (k) y(i) = x(k) − x(k − r (k)) and similar equations, we
have
k−1 
X >   
X (k) V111 V112 X (k)
y(i) ∗ V122 y(i)
i=k−r (k)
k−1
X
= r (k)X > (k)V111 X (k) + 2X > (k)V112 [x(k) − x(k − r (k))] + y > (i)V122 y(i) > 0, (6a)
i=k−r (k)
k−1 
X >   
X (k) V211 V212 X (k)
y(i) ∗ V222 y(i)
i=k−r M
k−1
X
> >
= r M X (k)V211 X (k) + 2X (k)V212 [x(k) − x(k − r M )] + y > (i)V222 y(i) > 0, (6b)
i=k−r M
k−1 
X >   
X (k) V311 V312 X (k)
y(i) ∗ V322 y(i)
i=k−rm
k−1
X
= rm X > (k)V311 X (k) + 2X > (k)V312 [x(k) − x(k − rm )] + y > (i)V322 y(i) > 0, (6c)
i=k−rm
(k)−1 
k−rX >   
X (k) V411 V412 X (k)
y(i) ∗ V422 y(i)
i=k−r M
= (r M − r (k))X > (k)V411 X (k) + 2X > (k)V412 [x(k − r (k)) − x(k − r M )]
(k)−1
k−rX
+ y > (i)V422 y(i) > 0. (6d)
i=k−r M

From (4–6), we have

1V (xk )
 
k−1
X
+ α −2k · r (k)X >(k)V111 X (k) + 2X T (k)V112 [x(k) − x(k − r (k)] + y >(i)V122 y(i)
i=k−r (k)‘
 DISCRETE SWITCHED SYSTEMS 7 of 21
 
k−1
X
+ α −2k · r M X > (k)V211 X (k) + 2X > (k)V212 [x(k) − x(k − r M )] + y > (i)V222 y(i)
i=k−r M ‘
 

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
k−1
X
+ α −2k · rm X > (k)V311 X (k) + 2X > (k)V312 [x(k) − x(k − rm )] + y > (i)V322 y(i)
i=k−rm


+ α −2k · (r M − r (k))X > (k)V411 X (k) + 2X > (k)V412 [x(k − r (k)) − x(k − r M )]


(k)−1
k−rX
+ y > (i)V422 y(i)
i=k−r M

k−1
X
6 α −2k · [X > (k) · Σ · X (k)] − α −2k · y > (i)[α 2r M R1 − V122 ]y(i)
i=k−r (k)‘

k−1
X k−1
X
− α −2k · y > (i)[α 2r M R2 − V222 ]y(i) − α −2k · y > (i)[α 2rm R3 − V322 ]y(i)
i=k−r M ‘ i=k−rm ‘

(k)−1
k−rX
−2k
−α · y > (i)[α 2r M R1 − V422 ]y(i), (7a)
i=k−r M ‘

where
 
Σ11 0 0 0
 0 Σ22 0 0 
Σ =
 0
 + r M · V211 + rm · V311 + r M · V411 − rm · (V411 − V111 )
0 Σ33 0 
0 0 0 Σ44

+ V112 Λ1 + Λ> > > > > > > >

1 V112 + V212 Λ2 + Λ2 V212 + V312 Λ3 + Λ3 V312 + V412 Λ4 + Λ4 V412
   >    >
Σ̄15 Σ̄15 Σ̄16 Σ̄16
 Σ̄25  −1  Σ̄25   Σ̄26l  −1  Σ̄26 
−     
 0  Σ55  0  −  0  Σ66  0  ,
  (7b)
0 0 0 0

where Σ11 , Σ22 , Σ33 , Σ44 , Σ55 , Σ66 , Λ1 , Λ2 , Λ3 and Λ4 are defined in (2c),

Σ̄15 = (A j + 1A j (k))> Pl , Θ = r M · (R1 + R2 ) + rm · R3 ,

Σ̄16 = (A j − I + 1A j (k))> Θ, Σ̄25 = (B j + 1B0 j (k))> Pl , Σ̄26 = (B j + 1B0 j (k))> Θ.

8 of 21 C.-H. LIEN ET AL.

From (1c) and (1d), we can define and obtain the following result:
 
Σ11 0 0 0 Σ̄15 Σ̄16
 ∗ Σ22 0 0 Σ̄25 Σ̄26   
  Ω 0 0

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
 ∗ ∗ Σ33 0 0 
0  
Σ̄ = 
 ∗ + 0 0 0
 ∗ ∗ Σ 44 0 0 

 ∗ 0 0 0
∗ ∗ ∗ Σ55 0 
∗ ∗ ∗ ∗ ∗ Σ66
 
Σ11 0 0 0 Σ15 Σ16
 ∗ Σ22 0 0 Σ25 Σ26   
  Ω 0 0
 ∗ ∗ Σ33 0 0 
0  
= ∗ + 0 0 0
 ∗ ∗ Σ 44 0 0 

 ∗ 0 0 0
∗ ∗ ∗ Σ55 0 
∗ ∗ ∗ ∗ ∗ Σ66
   > >  >   >
0 NAj NAj 0
 0   N>   N>   0 
   Bj   Bj   
 0   0   0  >  0 
+       
 F j (k)  0  +  0  F j (k)  0  , (7c)
 0       
 Σ57   0   0   Σ57 
Σ67 0 0 Σ67

where Ω, Σ15 , Σ16 , Σ25 , Σ26 , Σ57 and Σ67 are defined in (2c). By Lemmas 1 and 2 with above
condition, the condition Σ̄¯ < 0 in (2c) will imply Σ̄ < 0 in (7c) and Σ < 0 in (7b). From Σ < 0 in
(7b) with (2a) and (7a), we have

1V (xk ) = V (xk+1 ) − V (xk ) 6 0, k = 0, 1, 2, 3, . . . ,

V (xk+1 ) 6 V (xk ), k = 0, 1, 2, 3, . . . .

This implies

V (xk ) 6 V (x0 ), k = 0, 1, 2, 3, . . . ,

α −2k · λmin (Pi ) · kx(k)k2 6 V (xk ) 6 V (x0 ) 6 δ1 · kx0 k2s ,

where

δ1 = max λmax (Pi ) + [r M + (r M − rm )2 ]· λmax (Q) + r M

2
· λmax (R1 + R2 ) + rm2 · λmax (R3 )
i∈ N̄

+ (r M − rm ) · λmax (S) + rm · λmax (T ).

By some simple derivations, we have

p
kx(k)k 6 δ1 /.λmin (P) · α k · kx0 ks , k = 0, 1, 2, 3, . . . .

This completes this proof. 

R EMARK 1 By setting α = 1 in Theorem 1, the global asymptotic stability of system (1) can be
achieved.
 DISCRETE SWITCHED SYSTEMS 9 of 21

R EMARK 2 In Theorem 1, some additional non-negative inequalities in (6) are proposed. Free weighting
matrices Vi , i = 1, 2, 3, 4, in (6a–6d), are determined appropriately by LMI toolbox of MATLAB. From
our numerical examples 1 and 2, we can obtain that our proposed additional non-negative inequality
approach is good enough. Hence, our proposed additional non-negative inequality approach will provide

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
some flexible and potential improvement. In Section 3, our proposed additional non-negative inequality
approach will play an important role in the design of robust H∞ controller.

3. Robust H∞ control for uncertain discrete switched system

Consider the following discrete switched system with interval time-varying delay:

x(k + 1) = Āσ x(k) + B̄σ x(k − r (k)) + C̄σ u(k) + D̄σ w(k), k = 0, 1, 2, 3, . . . , (8a)

z(k) = Ē σ x(k) + Ḡ σ x(k − r (k)) + H̄σ u(k) + I¯σ w(k), k = 0, 1, 2, 3, . . . , (8b)

x(t) = φ(t), t ∈ [−H, 0], θ = −r M , −r M + 1, . . . , 0, (8c)

where u(k) ∈ <m is the control input, w(k) ∈ < p is the disturbance input which belongs to L 2 (α, 0, ∞),
for some 0 < α < 1, and x(t) = φ(t), t ∈ [−H, 0] is the regulated output, x(t) = φ(t), t ∈ [−H, 0] is
an initial state function, time-varying delay r (k) is a function from {0, 1, 2, 3, . . .} to {0, 1, 2, 3, . . .} and
0 < rm 6 r (k) 6 r M , rm and r M are two given positive integers. Some uncertain matrices are defined
by

Aσ = Aσ + 1Aσ (k), B σ = Bσ + 1Bσ (k), C σ = Cσ + 1Cσ (k), D̄σ = Dσ + 1Dσ (k),

Ē σ = E σ + 1E σ (k), Ḡ σ = G σ + 1G σ (k), H̄σ = Hσ + 1Hσ (k), I¯σ = Iσ + 1Iσ (k),

where Ai , Bi , Ci , Di , E i , G i , Hi , Ii ∈ <n×n , i = 1, 2, . . . , N , are some given constant matrices.

1Ai (k), 1Bi (k), 1Ci (k), 1Di (k), 1E i (k), 1G i (k), 1Hi (k), 1Ii (k), i = 1, 2, . . . , N , are some
perturbed matrices and satisfy the following conditions:
" #  
1 Ai (k) 1Bi (k) 1Ci (k) 1Di (k) Mi
= Fi (k)[ N1i N2i N3i N4i ], ∀ i ∈ {1, 2, . . . , N },
1E i (k) 1G i (k) 1Hi (k) 1Ii (k) Ti
(9a)
where Mi , Ti and N ji , j ∈ {1, 2, 3, 4}, i = 1, 2, . . . , N , are some given constant matrices with ap-
propriate dimensions, and Fi (k), i = 1, 2, . . . , N , are unknown matrices representing the parameter
perturbations which satisfy

Fi> (k)Fi (k) 6 I, ∀ i ∈ {1, 2, . . . , N }, k = 0, 1, 2, 3, . . . . (9b)

The state-feedback control is proposed by

u(k) = K i x(k) if σ (k, x(k)) = i,

where K i ∈ <m×n , i = 1, 2, . . . , N , are some matrices which will be designed. The final state-feedback
control can be rewritten as follows:
N
X
u(k) = λi (k) · K i x(k), t > 0, (10)
i=1
10 of 21 C.-H. LIEN ET AL.

where λi (k) is defined by

(
1, σ (k, x(k)) = i,
λi (k) =
0, otherwise.

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
D EFINITION 1 Consider the system (8) with (9) and state-feedback (10) and the following conditions:
(i) With w(k) = 0, the closed-loop system (8) with (9) is exponentially stable with convergence rate
0 < α < 1.
(ii) With zero initial conditions (i.e. x(t) = φ(t), t ∈ [−H, 0]), the signals w(k) and z(kt) are
bounded by
∞
X ∞
X
α −2k · z > (k)z(k) 6 γ 2 · α −2k · w > (k)w(k), ∀ w ∈ L 2 (α, 0, ∞), w 6= 0,
k=0 k=0

for constants γ > 0 and 0 < α < 1. In above conditions, the system (8) with (9) is said to be
exponentially stabilizable by (10) with disturbance attenuation γ and convergence rate α. The
control law (10) is said to be an H∞ control for system (8) with (9). γ is said to be the H∞ -norm
bound of feedback control (10).
R EMARK 3 In Remark 1, we have shown that the asymptotic stability property of system with α = 1.
If we choose α = 1 in Definition 1, then this definition will reduce to standard H∞ control (Sun et al.,
2007).
R EMARK 4 For two symmetric positive-definite matrices P ∈ R n×n and R ∈ R n×n , we have

−R P −1 R 6 P − 2R.

Proof. This proof can be done from the inequality −(R − P)> P −1 (R − P) 6 0. 
T HEOREM 2 Consider the system (8) with (9) and the state-feedback control (10). Suppose for a con-
stant η > 0, the following LMI optimization problem:

minimize γ̄ , (11a)

subject to

α 2r M R̂1 − V̂122 > 0, α 2r M R̂1 − V̂422 > 0, α 2r M R̂2 − V̂222 > 0, α 2rm R̂3 − V̂322 > 0,
V̂411 − V̂111 > 0, (11b)

− η · R̂1 + r M · R̂1 + r M · R̂2 + rm · R̂3 < 0, (11c)

     
V̂111 V̂112 V̂211 V̂212 V̂311 V̂312
V̂1 = > 0, V̂2 = > 0, V̂3 = > 0,
∗ V̂122 ∗ V̂222 ∗ V̂322
 
V̂411 V̂412
V̂4 = > 0, (11d)
∗ V̂422
 DISCRETE SWITCHED SYSTEMS 11 of 21
 
Ω̃11 0 0 0 0 Ω̃16 Ω̃17 Ω̃18 0 Ω̃110
 ∗
 Ω̃22 0 0 0 R̂1 B >
j η · R̂1 B >j R̂1 G >j 0 Ω̃210 
 ∗ ∗ Ω̃33 0 0 0 0 0 0 0 
 
 
 ∗ ∗ ∗ Ω̃44 0 0 0 0 0 0 

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
 
 ∗ ∗ ∗ ∗ −γ̄ · I D> η · D> I> 0 Ω̃510 
Ω = j j j 
 ∗ ∗ ∗ ∗ ∗ −α 2 · P̂l 0 0 εj · Mj 0 
 
 ∗ ∗ ∗ ∗ ∗ ∗ −η · R̂1 0 εj · η · Mj 0 
 
 ∗ ∗ ∗ ∗ ∗ ∗ ∗ −I ε j · Tj 0 
 
 ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −ε j · I 0 
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −ε j · I
 
ˆ¯
Ω̄
 ((4n+ p)×(4n+ p)) 0 ((4n+ p)×n) 0((4n+ p)×n) 0 ((4n+ p)×n) 0((4n+ p)×n) 0 ((4n+ p)×n) 
 ∗ 0(n×n) 0(n×n) 0(n×n) 0(n×n) 0(n×n) 
 
 ∗ ∗ 0 0 0 0(n×n) 
+ (n×n) (n×n) (n×n) 
 ∗ ∗ ∗ 0(n×n) 0(n×n) 0(n×n) 
 
 ∗ ∗ ∗ ∗ 0(n×n) 0(n×n) 
∗ ∗ ∗ ∗ ∗ 0(n×n)

< 0 for all j, l ∈ N̄ , (11e)

has a solution P̂i > 0, i ∈ N̄ , Q̂ > 0, R̂1 > 0, R̂2 > 0, R̂3 > 0, V̂111 ∈ <(4n+ p)×(4n+ p) ,
V̂112 ∈ <(4n+ p)×n , V̂122 ∈ <n×n , V̂211 ∈ <(4n+ p)×(4n+ p) , V̂212 ∈ <(4n+ p)×n , V̂222 ∈ <n×n , V̂311 ∈
<(4n+ p)×(4n+ p) , V̂312 ∈ <(4n+ p)×n , V̂322 ∈ <n×n , V̂411 ∈ <(4n+ p)×(4n+ p) , V̂412 ∈ <t (4n+ p)×n ,
V̂422 ∈ <n×n , positive constants γ̄ , ε j and K̂ j ∈ <m×n , j = 1, 2, . . . , N , where
Ω̃11 j = P̂ j − 2 R̂1 + T̂ + (r M − rm + 1) · Q̂, Ω̃16 j = R̂1 A>j + K̂ > >
j Cj ,

Ω̃17 j = η · ( R̂1 A>j + K̂ > >

j C j − R̂1 ),

Ω̃18 j = R̂1 E > > >

j + K̂ j H j , Ω̃110 j = R̂1 N1>j + K̂ > >
j N3 j , Ω̃22 j = −α 2r M · Q̂, Ω̃210 j = R̂1 N2>j ,

Ω33 = −α 2r M Ŝ, Ω̃44 j = −α 2rm · (T̂ − Ŝ), Ω̃510 j = N4>j ,

ˆ
Ω̄¯ = r M · V̂211 + rm · V̂311 + r M · V̂411 − rm · (V̂411 − V̂111 ) + V̂112 Λ̂1 + Λ̂> >
1 V̂112 + V̂212 Λ̂2

+ Λ̂> > > > > >

2 V̂212 + V̂312 Λ̂3 + Λ̂3 V̂312 + V̂412 Λ̂4 + Λ̂4 V̂412 ,

Λ̂1 = [ I −I 0 0 0 ](n×(4n+ p)) , Λ̂2 = [ I 0 −I 0 0 ](n×(4n+ p)) ,

Λ̂3 = [ I 0 0 −I 0 ](n×(4n+ p)) , Λ̂4 = [0 I −I 0 0 ](n×(4n+ p)) .

Then the control (10) with the matrices√K j = K̂ j R̂1−1 , j = 1, 2, . . . , N , is the H∞ control of system
(8) with the disturbance attenuation γ = γ̄ and convergence rate 0 < α < 1.
Proof. Define the Lyapunov functional
k−1
X −rm
X k−1
X
V (xk ) = α −2k x > (k)Pσ (k,x(k)) x(k) + α −2i x > (i)Qx(i) + α −2i x > (i)Qx(i)
i=k−r (k) j=−r M +1 i=k+ j
12 of 21 C.-H. LIEN ET AL.

0
X k−1
X 0
X k−1
X
+ α −2i y > (i)[R1 + R2 ]y(i) + α −2i y > (i)R3 y(i)
j=−r M +1 i=k−1+ j j=−rm +1 i=k−1+ j
k−1−r
Xm k−1
X

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
+ α −2i x > (i)Sx(i) + α −2i x > (i)T x(i), (12)
i=k−r M i=k−rm

where Pi = P̂i−1 > 0, i = 1, 2, . . . , N , Q = R̂1−1 Q̂ R̂1−1 > 0, R1 = R̂1−1 > 0, R2 = R̂1−1 R̂2 R̂1−1 > 0,
R3 = R̂1−1 R̂3 R̂1−1 > 0, S = R̂1−1 Ŝ R̂1−1 , T = R̂1−1 T̂ R̂1−1 , y(i) = x(i + 1) − x(i). By the similar
derivation for proof of Theorem 1, the difference of Lyapunov functional (12) along the solutions of
system (8) with (9) has the form

1V (xk ) = V (xk+1 ) − V (xk )



6 α −2k · α −2 · x > (k + 1).Pl x(k + 1) − x > (k)[P j − T ]x(k)


+ x > (k)[(r M − rm + 1) · Q]x(k) − α 2r M · x > (k − r (k))Qx(k − r (k))

+ [x(k + 1) − x(k)]> [r M · (R1 + R2 ) + rm · R3 ][x(k + 1t) − x(k)]
 
k−1
X (k)−1
k−rX k−1
X
− α 2r M ·  y > (i)R1 y(i) + y > (i)R1 y(i) y > (i)R2 y(i)
i=k−r (k) i=k−r M i=k−r M
 
k−1
X
− α 2rm ·  y > (i)R3 y(i)
i=k−rm


− α 2rm · x > (k − rm )[T − S]x(k − rm ) − α 2r M · x > (k − r M )Sx(k − r M ) . (13)


Define

Z > (k) = [ x > (k) x > (k − r (k)) x > (k − r M ) x > (k − rm ) w > (k) ].

By the condition (11d), we have

 
Vi11 Vi12
∗ Vi22 (5n+ p)×(5n+ p)
   
R1 0 0 0 0 0 R1 0 0 0 0 0
 0 R1 0 0 0 0   0 R1 0 0 0 0 
    
 0 0 R 0 0 0   0 0 
= 1  V̂i11 V̂i12  0 R1 0 0 
 0 0  V̂i22 (5n+ p)×(5n+ p)  0 
 0 0 R 1 0  ∗  0 0 0 R1 0 
 0 0 0 0 I 0   0 0 0 0 I 0 
0 0 0 0 0 R1 0 0 0 0 0 R1
> 0, i = 1, 2, 3, 4. (14)
 DISCRETE SWITCHED SYSTEMS 13 of 21
Pk−1
From (14) with i=k−r (k) y(i) = x(k) − x(k − r (k)) and the similar equations, we have
k−1
X  >    k−rX
(k)−1  >   
Z (k) V111 V112 Z (k) Z (k) V411 V412 Z (k)
+
y(i) ∗ V122 y(i) y(i) ∗ V422 y(i)

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
i=k−r (k) i=k−r M
k−1
X
= r (k)Z > (k)V111 Z (k) + 2Z > (k)V112 [x(k) − x(k − r (k))] + y > (i)V122 y(i)
i=k−r (k)
+ (r M − r (k))Z > (k)V411 Z (k) + 2Z > (k)V412 [x(k − r (k)) − x(k − r M )]
(k)−1
k−rX
+ y > (i)V422 y(i)
i=k−r M

6 r M Z > (k)V411 Z (k) − rm Z > (k)(V411 − V111 )Z (k) + 2Z > (k)V112 [x(k) − x(k − r (k))]
k−1
X
+ y > (i)V122 y(i) + 2Z > (k)V412 [x(k − r (k)) − x(k − r M )]
i=k−r (k)
(k)−1
k−rX
+ y > (i)V422 y(i) > 0, (15a)
i=k−r M
k−1 
X >   
Z (k) V211 V212 Z (k)
y(i) ∗ V222 y(i)
i=k−r M
k−1
X
= r M Z > (k)V211 Z (k) + 2Z > (k)V212 [x(k) − x(k − r M )] + y > (i)V222 y(i) > 0, (15b)
i=k−r M ‘
k−1
X  >   
Z (k) V311 V312 Z (k)
y(i) ∗ V322 y(i)
i=k−rm
k−1
X
= rm Z > (k)V311 Z (k) + 2Z > (k)V312 [x(k) − x(k − rm )] + y > (i)V322 y(i) > 0. (15c)
i=k−rm

From (11c), we have

r M · R1 + r M · R2 + r m · R3 < η · R1 . (16)
From (13–16), we have
1V (xk ) + α −2k · [z > (k)z(k) − γ 2 · w > (k)w(k)]
 k−rX (k)−1
+ α −2k · r M Z > (k)V411 Z (k) − rm Z > (k)(V411 − V111 )Z (k)
i=k−r M
k−1
X
+ 2Z > (k)V112 [x(k) − x(k − r (k))] + y > (i)V122 y(i)
i=k−r (k)
(k)−1
k−rX 
+ 2Z > (k)V412 [x(k − r (k)) − x(k − r M )] + y > (i)V422 y(i)
i=k−r M
14 of 21 C.-H. LIEN ET AL.

 
k−1
X
+ α −2k · r M Z > (k)V211 Z (k) + 2Z > (k)V212 [x(k) − x(k − r M )] + y > (i)V222 y(i)
i=k−r M ‘
 

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
k−1
X
+ α −2k · rm Z > (k)V311 Z (k) + 2Z > (k)V312 [x(k) − x(k − rm )] + y > (i)V322 y(i)
i=k−rm
k−1
X h i
6 α −2k · [Z > (k) · Ω̄ · Z (k)] − α −2k · y > (i) α 2r M R1 − V122 y(i)
i=k−r (k)‘
k−1
X h i k−1
X h i
− α −2k · y > (i) α 2r M R2 − V222 y(i) − α −2k · y > (i) α 2rm R3 − V322 y(i)
i=k−r M i=k−rm
(k)−1
k−rX h i
− α −2k · y > (i) α 2r M R1 − V422 y(i), (17a)
i=k−r M

where
 
Ω̄11 0 0 0 0
 0 Ω̄22 0 0 0 
 
Ω̄ = 
 0 0 Ω̄33 0 0  + r M · V211 + rm · V311 + r M · V411 − rm · (V411 − V111 )

 0 0 0 Ω̄44 0 
0 0 0 0 −γ 2 · I
+ V112 Λ̂1 + Λ̂> > > > > >
1 V112 + V212 Λ̂2 + Λ̂2 V212 + V312 Λ̂3 + Λ̂3 V312 + V412 Λ̂4 + Λ̂4 V412
> >

   >
( Ā j + C̄ j K j )> ( Ā j + C̄ j K j )>

 B̄ >j

 −2

 B̄ >
j



+   
0  (α · Pl )  0 
 0   0 
D̄ j > >
D̄ j
   >
η · ( Ā j + C̄ j K j − I )> η · ( Ā j + C̄ j K j − I )>

 η · B̄ >
j

 −1

 η · B̄ >
j


+ 0  (η · R1 ) 
  0 

 0   0 
η · D̄ >
j η · D̄ >
j
 >  > >
( Ē j + H̄ j K j ) ( Ē j + H̄ j K j )
 Ḡ >   Ḡ >j 
 j  
+ 0   0  , (17b)
 
 0   0 
¯I > ¯I >
j j

where

Ω̄11 = −P j +T +(r M − rm + 1)·Q, Ω̄22 = −α 2r M ·Q, Ω̄33 = −α 2r M S, Ω̄44 = −α 2rm ·(T − S).
 DISCRETE SWITCHED SYSTEMS 15 of 21

Define
 
Ω̄11 0 0 0 0 Ā>j + K > >
j C̄ j η · ( Ā j + C̄ j K j − I )> Ē > > >
j + K j H̄ j
 ∗ Ω̄22 0 0 0 B̄ > η · B̄ > Ḡ >j 
 j j 

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
 
 ∗ ∗ Ω̄33 0 0 0 0 0 
 
 ∗ ∗ ∗ Ω̄44 0 0 0 0 
Ω̂ =  
 ∗ ∗ ∗ ∗ −γ 2 · I D̄ >
j η · D̄ >
j I¯j
>

 
 ∗
 ∗ ∗ ∗ ∗ −α 2 · Pl−1 0 0 

 ∗ ∗ ∗ ∗ ∗ ∗ −η · R1−1 0 
∗ ∗ ∗ ∗ ∗ ∗ ∗ −I
 
Ω̄¯ 0 0 0
0 0 0 0
+ 0 0
, (17c)
0 0
0 0 0 0

where

Ω̄¯ = r M · V211 + rm · V311 + r M · V411 − rm · (V411 − V111 )

+ V112 Λ̂1 + Λ̂> > > > > > > >
1 V112 + V212 Λ̂2 + Λ̂2 V212 + V312 Λ̂3 + Λ̂3 V312 + V412 Λ̂4 + Λ̂4 V412 .

Define
Φ = diag[R1−1 , R1−1 , R1−1 , R1−1 , I, I, I, I ].

Pre-multiplying and post-multiplying the matrix Ω̂ in (17c) by Φ > and Φ, we have

Ω̃ = Φ > · Ω̂ · Φ,
where
P̂l = Pl−1 , P̂ j = P j−1 , Q̂ = R1−1 Q R1−1 , Ŝ = R1−1 S R1−1 , T̂ = R1−1 T R1−1 ,

−1 >
K̂ >
j = R1 K j , R̂1 = R1−1 , R̂2 = R1−1 R2 R1−1 , R̂3 = R1−1 R3 R1−1 ,

ˆ¯
¯ · diag[R −1 , R −1 , R −1 , R −1 , I ] = Ω̄,
diag[R1−1 , R1−1 , R1−1 , R1−1 , I ] · (Ω̄) 1 1 1 1
 > >
Ω̃11 0 0 0 0 R̂1 Ā>j + R̂1 K > >
j C̄ j η · ( R̂1 Ā>j + R̂1 K > >
j C̄ j − R̂1 ) R̂1 Ē >
j + R̂1 K j H̄ j
 ∗ Ω̃22 0 0 0 >
R̂1 B̄ j >
η · R̂1 B̄ j >
R̂1 Ḡ j 
 
 
 ∗ ∗ Ω̃33 0 0 0 0 0 
 
 ∗ ∗ ∗ Ω̃44 0 0 0 0 
Ω̃ =  
 ∗ ∗ ∗ ∗ −γ 2 · I D̄ > η · D̄ > I¯j
> 
 j j 
 ∗ ∗ ∗ ∗ ∗ −α 2 · P̂l 0 0 
 
 ∗ ∗ ∗ ∗ ∗ ∗ −η · R̂1 0 
∗ ∗ ∗ ∗ ∗ ∗ ∗ −I
 
ˆ¯
 Ω̄ 0 0 0
 0
+ 0 0 0 
0 0 0 0
0 0 0 0
16 of 21 C.-H. LIEN ET AL.

 > >
Ω̃11 0 0 0 0 R̂1 A>j + K̂ > >
j Cj η · ( R̂1 A>j + K̂ > >
j C j − R̂1 ) R̂1 E >
j + K̂ j H j
 ∗ Ω̃22 0 0 0 R̂1 B > η · R̂1 B > R̂1 G >j 
 j j 
 
 ∗ ∗ Ω̃33 0 0 0 0 0 
 
 ∗ ∗ ∗ Ω̃44 0 0 0 0 
= 

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
 ∗ ∗ ∗ ∗ −γ 2 · I D> η · D> Ij> 
 j j 
 ∗ ∗ ∗ ∗ ∗ −α 2 · P̂l 0 0 
 
 ∗ ∗ ∗ ∗ ∗ ∗ −η · R̂1 0 
∗ ∗ ∗ ∗ ∗ ∗ ∗ −I

   >    >
0 (N1 j R̂1 + N3 j K̂ j )> (N1 j R̂1 + N3 j K̂ j )> 0
 0   (N2 j R̂1 ) >   (N2 j R̂1 ) >   0 
       
 0       0 
   0   0   
 0     
 0   0  >  0 
+ 
 0  F j (k)  >  + >  F j (k) 

 ,

   N4 j   N4 j   0 
 Mj       
   0   0   M j 
η · Mj      η · Mj 
 0   0 
Tj 0 0 Tj
 
ˆ¯
 Ω̄ 0 0 0
0 0 0 0
+ , (18)
0 0 0 0
0 0 0 0

By Lemmas 1–3, Ω < 0 in (11e) will imply Ω̃ < 0 in (18). Ω̃ < 0 will imply Ω̂ < 0 in (17c) . Ω̂ < 0
will imply Ω̄ < 0 in (17b). From (11b) and (14), we have

α 2r M Ri − Vi22 > 0, i = 1, 2, 3, 4.

From Ω̄ < 0 and α 2r M Ri − Vi22 > 0 in (17a), we have

1V (xk )|w(k)=0 6 0.

By the similar derivation in Theorem 1, we conclude that the closed-loop system (8) with (9) and (10)
and w(k) = 0 is exponentially stable. Summating the equation in (17a) from 0 to ∞, we have

∞
X ∞
X
−2k >
V (x∞ ) − V (φ0 ) + α · [z (k)z(k)] − γ · 2
α −2k · [w > (k)w(k)] 6 0.
k=0 k=0

With zero initial condition (x(t) = ∈ [−H, 0]), we have P

Pφ(t), t−2k
V (φ0 ) = 0, V (x∞ ) > 0, and ∞k=0 α · [z > (k)z(k)] 6 γ 2 · ∞
k=0 α
−2k · [w > (k)w(k)], ∀w(k) ∈

L 2 (α, 0, ∞).
By the Definition
√ 1, the control (10) is the robust H∞ control of system (8) with (9), the disturbance
attenuation γ = γ̄ , and convergence rate 0 < α < 1. 
 DISCRETE SWITCHED SYSTEMS 17 of 21

4. Illustrative examples
E XAMPLE 1. Consider the system (1) with the following parameters: (Example 1 of Sun et al., 2007)
 

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
0.3 0
N = 2, ẋ(t) = Ax(t) + Bx(t − h) + C ẋ(t − h), t > 0, A2 = ,
0 0.58
 
0.12 0
ẋ(t) = Ax(t) + Bx(t − h) + C ẋ(t − h), t > 0, B2 = ,
0.11 0.11
   
0.1 0 0.01 0
M1 = M2 = , N A1 = N A2 = N B1 = N B2 = . (19)
0 0.1 0 0.01

In order to show the obtained results, the allowable delay upper bound that guarantees the global asymp-
totic stability for system (1) with (19) is provided in Table 1.
E XAMPLE 2. Consider the system (1) with the following parameters:
 
0.4 −0.01
N = 2, ẋ(t) = Ax(t) + Bx(t − h) + C ẋ(t − h), t > 0, A2 = ,
−0.01 0.58
 
0.2 0.1
ẋ(t) = Ax(t) + Bx(t − h) + C ẋ(t − h), t > 0, B2 = ,
0.1 0.1
   
0.12 0 0.01 0
M1 = M2 = , N A1 = N A2 = N B1 = N B2 = . (20)
0 0.12 0 0.01

In order to show the obtained results, the allowable delay bounds that guarantee the global asymptotic
stability for system (1) with (20) is provided in Table 2.
E XAMPLE 3. Consider the system (1) with the following parameters: (Example of Zhang et al., 2008)
 
0.7 0
N = 2, ẋ(t) = Ax(t) + Bx(t − h) + C ẋ(t − h), t > 0, A2 = ,
0.08 0.9
 
0.14 0
ẋ(t) = Ax(t) + Bx(t − h) + C ẋ(t − h), t > 0, B2 = ,
−0.1 −0.1

TABLE 1 Some obtained results for our proposed results in this paper

The delay upper bound to guarantee the stability of switched non-linear system (1) with (19)
Results Sun et al., 2007 Our results, Theorem 1
α = 1, rm = 1 globally asymptotically stable rM = 5 rM = 6

TABLE 2 Some obtained results for our proposed results in this paper

The delay bounds to guarantee the stability of switched non-linear system (1) with (20)
Results Sun et al., 2007 Our results, Theorem 1
Globally asymptotically rm = 1, r M = 2 rm = 1, r M = 3 rm = 2, r M = 3 rm = 3, r M = 4
stable α = 1
18 of 21 C.-H. LIEN ET AL.

   
0.05 0.05
M1 = , M2 = , N A1 = [0.2 0.3 ], N A2 = [−0.1 −0.1 ],
0 −0.02

N B1 = [ 0 −0.1 ], N B2 = [−0.3 −0.2]. (21)

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
In order to show the obtained results, the allowable delay bounds that guarantee the global asymptotic
stability for system (1) with (21) is provided in Table 3.
E XAMPLE 4. Consider the system (8) with (9) and the following parameters: (Example 2 of Sun et al.,
2007)
" #
0.96 0
N = 2, ẋ(t) = Ax(t) + Bx(t − h) + C ẋ(t − h), t > 0, A2 = ,
0 1.03
" #
−0.03 0.01
ẋ(t) = Ax(t) + Bx(t − h) + C ẋ(t − h), t > 0, B2 = ,
0 −0.02
  " #
0.01 0
ẋ(t) = Ax(t) + Bx(t − h) + C ẋ(t − h), t > 0, C2 = , D1 = ,
0.02 0.01
     
0 0.08 0 0.03 0
D2 = , E1 = , E2 = ,
0.03 0 0.01 0 0.05
 
0.02 0
ẋ(t) = Ax(t) + Bx(t − h) + C ẋ(t − h), t > 0, G 2 = ,
0 0.04
     
0.01 0.01 0.02
ẋ(t) = Ax(t) + Bx(t − h) + C ẋ(t − h), t > 0, I2 = , H1 = , H2 = ,
0.06 0.03 0.03
     
0.1 0 0.1 0 0.01 0
M1 = M2 = , T1 = T2 = , N11 = N12 = ,
0 0.1 0 0.1 0 0.01
    
0.01 0 0.01 0.01
N21 = N22 = , N31 = N32 = , N41 = N42 = .
0 0.01 0 0
By Theorem 2 with η = 10, rm = 1, r M = 9 and α = 1, the stabilizing state-feedback control gains in
(10) are given by
K 1 = K̂ 1 R̂1−1 = [−2.1548 −1.6904], K 2 = K̂ 2 R̂1−1 = [−0.1212 −3.1238],
with the disturbance attenuation γ = 0.07178. Some comparisons are made in Table 3 to some the
improvement of this paper.

TABLE 3 Some delay bounds to guarantee the global asymptotic stability (α = 1) of system

The delay bounds to guarantee the stability of switched non-linear system (1) with (21)
Results Zhang et al. (2008) Our results, Theorem 1
Conditions rm = 2, r M = 6 rm = 1, r M = 10 rm = 2, r M = 10 rm = 4, r M = 11 rm = 8, r M = 12
 DISCRETE SWITCHED SYSTEMS 19 of 21

TABLE 4 Comparing the results of this paper with Sun et al., 2007

Results Delay H∞ -norm bound H∞ state-feedback control gains, α = 1

Results of
rm = 1, r M =9 γ = 0.08 K 1 = [ −6.03 −2.6857 ], K 2 = [ −0.7574 −3.3116 ]

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
Sun et al., 2007
Results of this paper rm = 1, r M =9 γ = 0.07178 K 1 = [ −2.1548 −1.6904 ], K 2 = [ −0.1212 −3.1238 ]
rm = 3, r M =9 γ = 0.07177 K 1 = [ −2.1645 −1.6800 ], K 2 = [ −0.1171 −3.1222 ]
rm = 5, r M =9 γ = 0.07176 K 1 = [ −2.1755 −1.6714 ], K 2 = [ −0.1136 −3.1201 ]
rm = 7, r M =9 γ = 0.07175 K 1 = [ −2.1897 −1.6636 ], K 2 = [ −0.1110 −3.1169 ]

Major improvements of our results can be summarized as follows:

1. Better disturbance attenuation effect is achieved (smaller values of γ ).
2. Bigger upper bound of time-varying delay for system to be stabilized is guaranteed (bigger values
of r M ).
3. Interval time-varying delay is considered in this paper.

5. Conclusions
In this paper, the global exponential stability and H∞ state-feedback control for uncertain discrete
switched systems with interval time-varying delay have been considered. Some additional non-negative
inequalities and LMI approach are used to guarantee the proposed results of the system. The obtained
results are shown to be less conservative via numerical examples comparing the proposed results with
Sun et al., 2007, the following improvements may be observed:
1. Instead of the asymptotic property in Sun et al. (2007), the exponential one is investigated in this
paper.
2. With R > 0 and X (i) ∈ R n , the following inequality is proposed in Lemma 3 of Sun et al.
(2007):
 >  
k−1
X k−1
X k−1
X
 X (i) R  X (i) 6 r M · X > (i)R X (i), k = 0, 1, 2, 3, . . . .
i=k−r (k) i=k−r (k) i=k−r M

From above formulation, some upper bounds should be derived for the following terms in (5) and (13):
 
k−1
X (k)−1
k−rX k−1
X
− α 2r M ·  y > (i)R1 y(i) + y > (i)R1 y(i) + y > (i)R2 y(i)
i=k−r (k) i=k−r M i=k−r M
 
k−1
X
− α 2rm ·  y > (i)R3 y(i) .
i=k−rm

In this paper, some additional non-negative inequalities in (6) and (15) are proposed. Hence, redun-
dant estimated upper bounds can be avoided. From the numerical comparisons, less conservativeness is
reached. This approach can be generalized to use in continuous or more general systems (Lien et al.,
2010).
20 of 21 C.-H. LIEN ET AL.

3. The following two terms:

V112 Λ1 + Λ> > > > > > > >

1 V112 + V212 Λ2 + Λ2 V212 + V312 Λ3 + Λ3 V312 + V412 Λ4 + Λ4 V412 ,

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
V̂112 Λ̂1 + Λ̂> > > > > > > >
1 V̂112 + V̂212 Λ̂2 + Λ̂2 V̂212 + V̂312 Λ̂3 + Λ̂3 V̂312 + V̂412 Λ̂4 + Λ̂4 V̂412

ˆ
are included in Ω of (2c) and Ω̄¯ of (11e), respectively. Above two terms will provide some negative-
definite symmetric matrices to improve the conservativeness of our proposed results.
For our future research, the proposed approaches in Li et al. (2009, 2010) will be investigated to
improve the obtained results in this paper and future work.

Acknowledgement
The authors would like to thank associate editor and anonymous reviewers for their helpful comments.

Funding
National Science Council of Taiwan, Republic of China (NSC 97-2221-E-022-009-MY2).

R EFERENCES
B OYD , S. P., G HAOUI , L. EL., F ERON , E. & BALAKRISHNAN , V. (1994) Linear Matrix Inequalities in System
and Control Theory. Philadelphia, PA: SIAM.
G U , K., K HARITONOV, V. L. & C HEN , J. (2003) Stability of Time-Delay Systems. Boston, MA: Birkhauser.
H ALE , J. K. & V ERDUYN LUNEL , S. M. (1993) Introduction to Functional Differential Equations. New York:
Springer.
H AN , Y. & TANG , H. (2007) Robust H∞ control for a class of discrete switched systems with uncertainties and
delays. Proceedings of the 26th Chinese Control Conference. Hunan, China: IEEE, pp. 681–684.
I BRIR , S. (2008) Stability and robust stabilization of discrete-time switched systems with time-delays: LMI
approach. Appl. Math. Comput., 206, 570–578.
KOLMANOSKII , V. B. & M YSHKIS , A. (1992) Applied Theory of Functional Differential Equations. Dordrecht,
the Netherlands: Kluwer Academic Publishers.
L I , H., C HEN , B., Z HOU , Q. & Q IAN , W. (2009) Robust stability for uncertain delayed fuzzy Hopfield neural
networks with Markovian jumping parameters. IEEE Trans. Syst. Man Cybern. B, 39, 94–102.
L I , H., WANG , C. S HI , P. & G AO , H. (2010) New passivity results for uncertain discrete-time stochastic neural
networks with mixed time delays. Neurocomputing, 73, 3291–3299.
L IEN , C. H., Y U , K. W., C HANG , H. C. & C HUNG , Y. J. (2010) Switching signal design for stability of switched
systems with time-varying delay. 2010 International Symposium on Computer, Communication, Control and
Automation. Tainan, Taiwan: IEEE, pp. 286–289.
L IEN , C. H., Y U , K. W., L IN , Y. F., C HUNG , Y. J. & C HUNG , L. Y. (2009) Exponential convergence rate
estimation for uncertain delayed neural networks of neutral type. Chaos Solitons Fractals, 40, 2491–2499.
L IU , J., L IU , X. & X IE , W. C. (2008) Delay-dependent robust control for uncertain switched systems with time-
delay. Nonlinear Anal. Hybrid Syst., 2, 81–95.
M A , Y., YAN , W., L IU , D. & Z HANG , Y. (2006) H∞ control of discrete switched systems with time delay via
state feedback with memory. Proceedings of the 6th World Congress on Intelligent Control and Automation.
Dalian, China. pp. 1328–1332.
M AHMOUD , M. S. (2008) Delay-dependent H∞ filtering of a class of switched discrete-time state delay systems.
Signal Process., 88, 2709–2719.
 DISCRETE SWITCHED SYSTEMS 21 of 21

S ONG , Y., FAN , J., F EI , M. & YANG , T. (2008) Robust H∞ control of discrete switched systems with time delay.
Appl. Math. Comput., 205, 159–169.
S UN , X. M., WANG , W., L IU , G. P. & Z HAO , J. (2008) Stability analysis for linear switched systems with
time-varying delay. IEEE Trans. Syst. Man Cybern. B, 38, 528–533.

Downloaded from imamci.oxfordjournals.org at Centro Federal de Educação Tecnológica de Minas Gerais on February 3, 2011
S UN , Y. G., WANG , L. & X IE , G. (2006) delay-dependent robust stability and stabilization for discrete-time
switched systems with mode-dependent time-varying delays. Appl. Math. Comput., 180, 428–435.
S UN , Y. G., WANG , L. & X IE , G. (2007) Delay-dependent robust stability and H∞ control for uncertain discrete-
time switched systems with mode-dependent time delays. Appl. Math. Comput., 187, 1228–1237.
X IE , D., X U , N. & C HEN , X. (2008) Stabilisability and observer-based switched control design for switched linear
systems. IET Control Theory Appl., 2, 192–199.
Z HAI , G., L IU , D., L MAE , J. & KOBAYASHI , T. (2006) Lie algebraic stability analysis for switched systems with
continuous-time and discrete-time subsystems. IEEE Trans. Circuits Syst., 53, 152–156.
Z HANG , L., S HI , P. & BASIN , M. (2008) Robust stability and stabilisation of uncertain switched linear discrete
time-delay systems. IET Control Theory Appl., 2, 606–614.