Automatica 43 (2007) 903 – 911

www.elsevier.com/locate/automatica

Brief paper
Sampled-data control of networked linear control systems夡
Li-Sheng Hu a,∗ , Tao Bai a , Peng Shi b , Ziming Wu a
a Department of Automation, Shanghai Jiao Tong University, Shanghai 200030, PR China
b Faculty of Advanced Technology, University of Glamorgan, Pontypridd CF37 1DL, UK

Received 20 May 2005; received in revised form 5 October 2006; accepted 13 November 2006
Available online 13 March 2007

Abstract
In this paper, the problem of synthesis and analysis for the networked control systems (NCSs) with time-driven digital controllers and event-
driven holders is considered. The NCS is modelled as a sampled-data system with time-delay in its discrete-time subsystem. This model is able
to capture many network-induced features, for example, time-delay and packet dropout. Moreover, the model allows different combinations of
the time-driven or event-driven mode of the devices, including the samplers, the controllers and the holders. By transforming time-delay in the
discrete-time subsystem into its continuous-time subsystem of the sampled-data system, we have also obtained a less conservative time-delay
dependent stability result for the NCSs, using a new Lyapunov function and a relaxed condition. Some limitations of the existing literatures on
network-induced time-delay and sampling period are removed in the proposed framework. Furthermore, a sampled-data control design procedure
is developed for the NCSs. Linear matrix inequality approach has been employed to solve the stability and control design problems. Finally,
numerical examples are included to demonstrate the effectiveness of the proposed stability result and the potential of the proposed techniques.
䉷 2007 Elsevier Ltd. All rights reserved.

Keywords: Networked control system; Stability; Sampled-data system; Feedback control; Linear matrix inequality

1. Introduction Recently, the network-induced time-delay has been received

widely attentions. In Halevi and Ray (1988) and Nilsson (1998),
Nowadays, networks play important roles in many industrial the authors systematically considered synthesis and analysis of
control applications, especially for example, complex control the NCSs with synchronization, message rejection and vacant
systems and remote control systems. Comparing with the con- sampling. In the case of periodic time-delay, the sufficient and
ventional point-to-point control systems, the networked control necessary condition for the systems with uniformly asymptotic
systems (NCSs) show some nice features, such as, flexibility stability was established. Using stochastic Lyapunov function,
of operation, ease of diagnosis and maintenance, small volume in Krtolica, Oguner, and Chan (1994), the sufficient and nec-
of wiring, low cost, etc. In the NCSs, all devices (the samplers, essary condition of mean-square exponential stability for the
the digital controllers and the holders) are distributed, and con- linear discrete-time systems with a time-invariant controller in
nected by the networks. Because of the limitation of the network the place and time-varying time-delay modelled as a Markov
resource, time-delay caused by data transmission and/or packet chain was set up. An exponential stability condition for the sys-
drop will inevitably degrade control performance of the NCSs, tems with random one-step time-delay was also obtained using
or even cause the control systems instable (Branicky, Phillips, the deterministic structural perturbation approach. In Nilsson
& Zhang, 2002; Halevi & Ray, 1988; Nilsson, 1998; Walsh, (1998) and references therein, the authors systematically inves-
Ye, & Bushnell, 2002; Zhang, Branicky, & Phillips, 2001). tigated the modelling and analysis problems for NCSs under an
assumption that the time-delay from sensor to actuator is less
than one sampling period. In the work, the author also consid-
夡 This paper was not presented at any IFAC meeting. This paper was
ered the NCSs with stochastic time-delay.
recommended for publication in revised form by Associate Editor Ioannis
Paschalidis under the direction of Editor Ian Petersen.
Besides the network-induced time-delay, the data trans-
∗ Corresponding author. mission mechanism is also an important issue for NCSs.
E-mail addresses: [email protected] (L.-S. Hu), [email protected] (P. Shi). Considering two different network protocols, the
0005-1098/$ - see front matter 䉷 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.automatica.2006.11.015
904 L.-S. Hu et al. / Automatica 43 (2007) 903 – 911

try-once-discard/maximum-error-first and the statically sched- 2004), this kind of sampled-data systems was considered. In
uled protocols, two globally exponential stability conditions that paper, the authors proposed a robust control method by
for the NCSs with respect to multiple-packet and one-packet transforming the sampled-data system into a continuous-time
transmission cases were presented in Walsh et al. (2002). system with control time-delay. However, this transformation
In the paper, the authors presented a conservative bound of the is under a condition: the sampling period is infinite small.
allowable maximum transfer interval using the perturbation In this paper, a sampled-data system with time-delay in its
theory. In Zhang et al. (2001), the stability problems for the discrete-time subsystem is proposed to model the NCSs with
NCSs modelled as hybrid or asynchronous dynamical systems time-driven digital controllers and event-driven holder devices.
with network-induced delay, packet dropout and multi-packet The configuration shows that this model is quite general, it
transmission between sensor and controller were considered. can capture many network-induced features, for example, time-
Considering that the power spectral density of the output is delay, packet dropout, multi-packet transmission. The sampled-
a function of the dropout probability, the upper bound of the data system formulation also provides a nature way to describe
packet dropout rates was obtained under which the NCSs the so-called “communication sequence”. Moreover, it allows
are stochastically asymptotically stable in Ling and Lemmon different combinations of the time-driven or event-driven mode
(2002). of the devices, including the samplers, the controllers and the
Basically, NCS is hybrid, which involves a continuous plant holders. It is also able to remove the assumption that the time-
and event-driven or time-driven devices (digital controller, delay from sensor to actuator is less than one sampling pe-
holder and sampler) and networks. Hybrid nature of NCSs riod used in Nilsson (1998). By transforming time-delay in
makes the synthesis and analysis problems for NCSs difficult. the discrete-time subsystem into its continuous-time subsys-
This kind of the systems is usually referred as the sampled-data tem of the sampled-data system, a less conservative time-delay
control systems which simultaneously contain continuous- dependent stability result is obtained for the NCSs by using
time and discrete-time signals (Boukas, 2005; Boukas & the new Lyapunov function and a special relaxed condition,
Al-Muthairi, 2006; Chen & Francis, 1995; Chen, Lam, & Xu, which removes the limitation on the sampling period imposed
2006; Shi, 1998; Shi, Boukas, Agarwal, & Shue, 1998; Shi, in Fridman et al. (2004). Linear matrix inequality (LMI) for-
de Souza, & Xie, 1997; Shi, Fu, & de Souza, 1996; Xie, Shi, mulation of a sampled-data control design procedure is then
& de Souza, 1993). Sampled-data control system provides a presented for the NCSs, which is solvable using Toolbox for
direct approach to design digital controller for the continuous- Matlab. Finally, numerical examples are given to show the ef-
time systems without approximations, see for example, Hu, fectiveness of the proposed method.
Lam, Cao, and Shao (2003). Sampled-data control system for- Notations: Throughout this paper, for symmetric matrix X
mulation has been recognized a modelling method for NCSs and Y, the notation X Y (respectively, X > Y ) means that
for years, see Walsh et al. (2002) and Zhang et al. (2001), for the matrix X − Y is positive semi-definite (respectively, posi-
example. In Lian, Moyne, and Tilbury (2003), the authors con- tive definite). I is the identity matrix with appropriate dimen-
sidered a modelling problem of NCSs for multivariable linear sion. {tk , k = 1, 2, . . . , } are the sampling instants satisfying
systems with distributed asynchronous sampling. In Zhang tk+1 −tk =Ts , where Ts denotes the sampling period with which
et al. (2001), the authors considered an analysis problem for the outputs of a plant are synchronously measured by ideal sam-
NCSs under hybrid system framework. In Walsh et al. (2002) plers. u(·) and ũ[·] are used to denote the continuous-time and
and Zhang et al. (2001), the authors set up a stability result for the discrete-time signal of the associated variable, respectively.
the NCSs with only one time-delay. In the paper (Halevi & Ray,
1988), the authors proved that two delays cannot be lumped 2. Modeling and problem formulation
together for the cases referring as the so-called message re-
jection and vacant sampling even though feedback controller Consider an NCS shown as in Fig. 1, in which a continuous
is time-invariant. Conventionally, the lifting technique (Chen plant P is controlled by a digital controller C. In the system,
& Francis, 1995) and the traditional discretized method are the output signals x(t) of the plant are synchronously mea-
used as tools to consider the synthesis and analysis problems sured with ideal samplers S at a sampling rate 1/Ts . The digital
for NCSs (Walsh et al., 2002). However, the lifting technique
only works for LTI systems, and the traditional discretized
method has approximation. In our previous works, a hybrid
system synthesis approach was set up for the sampled-data
systems. Moreover, a series works on stability, robust control,
H2 control, H∞ control (filtering) and model predictive control
problems for the sampled-data control systems (with time-
delay) have been investigated (Hu, Cao, & Shao, 2002; Hu &
Huang, 2005; Hu, Huang, & Cao, 2004; Hu et al., 2003; Hu,
Shi, & Frank, 2006). However, the methods developed (Hu
et al., 2002, 2003, 2004, 2006; Hu & Huang, 2005) do not work
for the sampled-data systems with time-delay in its discrete-
time subsystems. In a new work (Fridman, Seuret, & Richard, Fig. 1. Schematic diagram of networked control system.
 L.-S. Hu et al. / Automatica 43 (2007) 903 – 911 905

Recalling the fact, every control action ũ is held by a zero-order

holder and only valid over the interval (tk + kca , tk+1 + k+1
ca ],
we have

u(t) = ũ[tk ], (3)

for t ∈ (tk + kca , tk+1 + k+1

ca ], where x(t) ∈ R denotes the
n

state vector of the plant, u(t) ∈ R denotes the control in-

m

put vector. A, B and F are the real matrices with appropriate

dimensions.
Then the problems considered in this paper can be formu-
lated as:
Fig. 2. Timing mechanism diagram of networked control system.
1. set up a stability result for the NCS described by (1)–(3);
2. design a digital controller (2) such that the NCS stable with
possible maximum time-delay.
controller uses the information of the plant transmitted through
the networks to generate a digital control action ũ[tk ]. The Remark 1. The system formulated above is referred as a
control action ũ[tk ] is then transmitted via the networks again at sampled-data one which contains continuous-time and discrete-
the rate 1/Ts , held by a zero-order holder H to drive the plant. In time signals simultaneously. In our previous works, a hybrid
this formulation, the measurements x(t) and the control action systematic synthesis approach was set up for the sampled-data
ũ[tk ] may be delayed because of the network traffic. Here, k−1 sc systems. Moreover, a series works on stability, robust control,
and kca denote the network-induced delays in the control input H2 control, H∞ control (filtering) and model predictive control
channels and the plant input channels, respectively. problems for the sampled-data control systems (with time-
To well understand the NCS considered, its timing mecha- delay) have been investigated (Hu et al., 2002, 2003, 2004,
nism is shown as in Fig. 2. In this scenario, we assume that the 2006; Hu & Huang, 2005). In Walsh et al. (2002) and Zhang
digital controller and the sampler are working in time-driven et al. (2001), the authors set up a stability result for the NCSs
mode, while the holder is working in event-driven mode. More- with only one time-delay ksc . In the paper (Halevi & Ray,
over, each data transmitted can reach its destination within one 1988), the authors proved that two delays cannot be lumped
sampling period Ts (Nilsson, 1998). At the previous time in- together for the cases, which are referred as message rejection
stant tk−1 , the output of the plant x(t) was measured, we as- and vacant sampling, respectively, even though feedback con-
sume that its measurement x(tk−1 ) reaches the controller C at troller is time-invariant. For the message rejection case, if there
time tk−1 +k−1
sc , where sc is the network-induced time-delay.
k−1
are one more plant-measurements arriving at the controller
At the current time tk , the controller C uses the most recently over one control period (at which the controller switches), only
arrival measurement to generate a control action ũ[tk ]. The con- the latest arrival measurement is used by the time-driven con-
trol action is transmitted through the networks and will be re- troller to generate a control action. The other case is that, over
ceived by the hold device after the time lag kca . The holder will one control period, if there is no plant-measurement arriving,
hold this control action, until the new control action arrives at the controller will use the previous measurements to generate
the instant tk+1 + k+1ca , that is, any control action will be held control action.
over an interval (tk + kca , tk+1 + k+1
ca ]. Here, the superscript Different from the conventional sampled-data systems where
k of sc and ca is used to describe time-varying nature of the the controllers switch at the sampling instant tk and the con-
time-delay. trol action is held at the same time, the controller considered
In the configuration as in Fig. 2, the sampler and the con- switches at the sampling instant tk , but the control action starts
troller are synchronized, which is a common in industrial con- to take effect after a network-induced time-delay kca . Time-
trol systems, for example, FCSs (field-bus control systems). varying nature of the transmission delay ksc and kca makes
Clock synchronization which is also discussed in Nilsson the lifting technique (Chen & Francis, 1995) for the sampled-
(1998) is an essential function to make a control system as it data systems not working any more. In a new work (Fridman
would be. In the sequel, we will show that this restriction can et al., 2004), the authors proposed a robust control method by
be removed for the cases when the controller and/or the holder transforming the sampled-data system into a continuous-time
are/is event-driven. system with control time-delay. However, this transformation
In this paper, consider the plant described as is under a condition: the sampling period is infinite small. In
this paper, we will propose a new synthesis method by trans-
ẋ(t) = Ax(t) + Bu(t). (1) forming the time-delay in the discrete-time subsystem into the
continuous-time subsystem and keeping the sampled-data sys-
It is controlled by a digital state-feedback controller tem nature unchanged. In this framework, no limitation on
sampling period is introduced, which shows higher application
ũ[tk ] = F x(tk−1 ). (2) potential.
906 L.-S. Hu et al. / Automatica 43 (2007) 903 – 911

Fig. 3. Timing mechanism diagram of networked control system-case 1.

Fig. 4. Timing mechanism diagram of networked control system-case 2.

For simplification, we assume k+1 ca 2ca . Letting ū[tk +
k

ca ] = F x(tk + ca ) and û(t) = ū[tk + ca ], then we have

k k k

ũ[tk ] = ū[tk + kca − Ts − kca ], (4) The control action is still held at tk + kca up to tk+1 + k+1
ca .
Then the system can be rewritten as
u(t) = û(t − Ts − kca ). (5)
ẋ(t) = Ax(t) + Bu(t),
Therefore, the NCS (1)–(3) is equivalent to the following
closed-loop system: ũ[tk ] = F x(tk−j )

ẋ(t) = Ax(t) + B û(t − Ts − kca ), (6) u(t) = ũ[tk ], t ∈ (tk + kca , tk+1 + k+1
ca ],

û(t) = ū[tk + kca ], (7) Under condition of k+1

ca 2ca +(j −1)Ts , by the transforming
k

method and letting d = j T s + kca , the above system is rewritten

for t ∈ (tk + kca , tk+1 + k+1
ca ] and as follows:
ū[tk + kca ] = F x(tk + kca ). (8) ẋ(t) = Ax(t) + B û(t − d),
Actually, the above method defines a mapping (u, ũ) → (û, ū) û(t) = ū[tk + kca ], t ∈ (tk + kca , tk+1 + k+1
ca ],
to transform the time-delay in the digital controller (2) to the
continuous-time subsystem (6) and keep the dynamics of the ū[tk + kca ] = F x(tk + kca )
closed-loop system of (1)–(3) unchanged.
Letting x̃(t) = (x T (t), ûT (t))T and d = Ts + kca , then, the Case 2: Consider the NCSs with the time-driven sampler,
NCS (6)–(8) is rewritten as the event-driven controller and the event-driven holder. For this
case, it is not difficult to know that the control action is held
˙ = Ãx̃(t) + B̃ x̃(t − d), over the interval (tk + ksc + kca , tk+1 + k+1
sc + ca ] and the
k+1
x̃(t) (9)
equivalent delay d =ksc +kca , where ksc and kca are not limited
for t ∈ (tk + kca , tk+1 + k+1
ca ], and within one sampling period (Fig. 4). Then the system can be
rewritten as
x̃(tk + kca + 0) = (Ā + B̄F C̄)x̃(tk + kca ), (10)
ẋ(t) = Ax(t) + Bu(t),
where
      ũ[tk + ksc ] = F x(tk ),
A 0 0 B 0
à = , B̃ = , B̄ = ,
0 0 0 0 I u(t) = ũ[tk + ksc ], t ∈ (tk + ksc +kca , tk+1 + k+1
sc + ca ].
k+1
 
I 0
C̄ = [ I and Ā = . Under condition of Ts + k+1
sc + ca 2(sc + ca ), by the
0] k+1 k k
0 0
transforming method and letting d =sc +ca , the above system
k k

Remark 2. The model formulated for NCSs is general which is rewritten as follows:
also admits many other kinds of NCSs or the NCSs with the
ẋ(t) = Ax(t) + B û(t − d),
different working modes.
Case 1: In this case, the controller and the sampler are û(t) = ū[tk + ksc + kca ],
time-driven, the holder is event-driven, we assume that the
time-delay ksc and kca are not limited within one sampling t ∈ (tk + ksc + kca , tk+1 + k+1
sc + ca ],
k+1

period as shown in Fig. 3, for example. In this scenario, be- ū[tk + ksc + kca ] = F x(tk + ksc + kca ).
cause of time-driven nature of the controller, the controller only
uses newly available measurement x(tk−j ) of the plant reached Case 3: The NCSs with packet dropout can be described as
at the controller input buffer to generate its control action. the control action is not updated during the time interval of
 L.-S. Hu et al. / Automatica 43 (2007) 903 – 911 907

packet dropout. Hence, it is just a special case of case 1 or case is equivalent to that there exists a matrix Y such that
2 with longer delay than one sampling period.  
Case 4: For the limited network resource cases, the ef- P0 AT Y T
> 0.
fect of communication constraints on the controlled sys- YA Y + Y T − P1
tems was considered using the so-called “communication
sequences” (Hristu-Varsakelis, 2001; Hristu-Varsakelis & 3. Stability analysis
Kumar, 2002). If the network is ideal, that is, with the con-
stant packet transmission time and no dropout, a stability In what follows in this section, we will present a delay-
result was set up for the NCSs using the “communication dependent Lyapunov function to ensure the stability of NCS
sequences” (Hristu-Varsakelis, 2001; Hristu-Varsakelis & (9)–(10) with F = 0 and possible maximum time-delay d0 sat-
Kumar, 2002). Actually, an essential problem of the “com- isfying d d0 .
munication sequences” for the NCSs is to find a maximum
sampling period Ts such that the following system stable, Theorem 2. The NCS (9)–(10) with F = 0 is stable for delay
d d0 , if there exist matrices P > 0, Q1 > 0, Q2 > 0,  > 0,
ẋ(t) = (A + BF s(t) )x(t), T = [T1 , T2 , T3 , T4 ], N such that the following matrix inequal-
where Fs(t) ∈ {F o , F c }, F c and F o denote the feedback gains ities hold:
for s(t)=1 and s(t)=0, respectively. Here, s(t) ∈ {0, 1} denotes  
 N
the communication constraints at the instant t. Actually, this 0, (11)
system can also be modelled as a sampled-data system. N T Q2
Let tk be the time instants where the control action up-
dating, tk = {t : s(t) = 1, t 0}, for k = 0, 1, . . ., and Ts be M + T T Â + ÂT T + T N T + N  + d < 0, (12)
sampling period, Ts = maxk {tk+1 − tk }. Under the conditions
in Hristu-Varsakelis (2001) and Hristu-Varsakelis and Kumar  =: ĀT P Ā − P < 0, (13)
(2002), the “communication sequences” problem for the NCSs
can be restated as finding a controller where
⎡ ⎤
u(t) = Fs(t) x(tk ), t ∈ (tk , tk+1 ], Q1 0 P 0
⎢ 0 −Q1 0⎥
with sc = 0 and ca = 0 such that the closed-loop NCSs stable. ⎢ 0 ⎥
M =⎢ ⎥,
This is exactly the model considered in the paper. In the follow- ⎣P 0 dQ2 0⎦
ing, the established stability condition will provide a method to 0 0 0 0
estimate the maximum sampling period Ts , which shows that
the developed framework has a nice feature to deal with the  = [ à B̃ −I 0] ,
“communication sequences” problem.
 = [I −I 0 −I ] .
Remark 3. In the above formulation, the equivalent sampled-
data system (6)–(8) switches at tk +kca with time-varying time- Proof. Recall the fact that the control action u(t) is only avail-
delay d =Ts +kca . The time-varying nature of the sampled-data able over every holding interval (tk + kca , tk+1 + k+1
ca ]. Noting
system (6)–(8) makes it different from the ones considered in tk−1 < t − d tk + kca , then by Newton–Leibniz we have,
Hu et al. (2002, 2003, 2004, 2006) and Hu and Huang (2005).
However, it is naturally generalized from the models considered tk−1 +k−1
ca tk +kca t
in Hu et al. (2002, 2003, 2004, 2006) and Hu and Huang (2005), x̃˙ ds + x̃˙ ds + x̃˙ ds
that is, over every time-varying period (tk + kca , tk+1 + k+1 tk−1 +k−1 tk +kca
ca ],
t−d ca

the system behaves as a normal continuous-time one, while the = x̃(t) − x̃(t − d) − wk ,
controller switches at the time instant tk + kca . Interestingly,
over every interval (tk + kca , tk+1 + k+1
ca ], the time-delay d = where
Ts + kca is constant, which shows that the sampled-data system ⎧
formulation can simplify the analysis problem of time-varying ⎪
⎪ [0 ūT [tk + kca ] − ūT [tk−1 + k−1
ca ]]
T
⎪
⎨ for t − d > t
k−1 + ca ,
nature of the NCSs. k−1
wk =
⎪
⎪ [0 ūT [tk + kca ] − ūT [tk−2 + k−2
ca ]]
T
For self-completeness, an existing result is presented here ⎪
⎩
for t − d tk−1 + ca . k−1
for the sequel development.
 tk−1 +k−1
ca ˙
 t +k t
Lemma 1 (Boyd, El Ghaoui, Feron, and Balakrishnan (1994), Let yt (d)= t−d x̃ ds + k cak−1 x̃˙ ds + tk +k x̃˙ ds :=
tk−1 +ca
t ca
Hu et al. (2004)). For matrices A, P0 > 0 and P1 > 0, the fol-
t−dx̃˙ ds, then the system (9) can be rewritten as
lowing inequality:
AT P1 A − P0 < 0, ˙ = (Ã + B̃)x̃(t) − B̃yt (d) − B̃wk .
x̃(t)
908 L.-S. Hu et al. / Automatica 43 (2007) 903 – 911

Obviously, there exist arbitrary matrices T1 , T2 , T3 and T4 such Then we obtain

that
˙ + x̃ T (t)Q1 x̃(t) − x̃ T (d)Q1 x̃t (d)
V̇ (x̃) = 2x̃ T (t)P x̃(t) t
˙ + T 4 wk )
(T1 x̃(t) + T2 x̃t (d) + T3 x̃(t) t
+ dx̃˙ (t)Q2 x̃(t)
˙ − x̃˙ (s)Q2 x̃(s)
˙ ds
T T
˙ − B̃wk ) = 0.
× ((Ã + B̃)x̃(t) − B̃yt (d) − x̃(t) (14) t−d
˙ + T 4 wk )
+ 2(T1 x̃(t) + T2 x̃t (d) + T3 x̃(t)
Let x̃t (d) = x̃(t − d), T = [T1 , T2 , T3 , T4 ] and (t) =
˙ − B̃wk )
(x̃ T (t), x̃tT (d), x̃˙ (t), wkT )T . For t ∈ (tk + kca , tk+1 + k+1 × ((Ã + B̃)x̃(t) − B̃yt (d) − x̃(t)
T
ca ],
choose the following Lyapunov functional candidate, ˙ − B̃wk )
= T (t)M(t) + 2T (t)T T ((Ã + B̃)x̃(t) − x̃(t)
t
x̃˙ (s)Q2 x̃(s)
˙ ds.
T
V (x̃) = V1 (x̃) + V2 (x̃), (15) − 2T (t)T T B̃yt (d) − (16)
t−d
where
By inequality in Moon, Park, Kwon, and Lee (2001) and
tk−1 +k−1
ca
abusing the notation, we have
V1 (x̃) = x̃ T (t)P x̃(t) + x̃ T (s)Q1 x̃(s) ds
t−d − 2T (t)T T B̃yt (d)
tk +kca dT (t)(t) + 2T (t)(N − T T B̃)yt (d)
+ x̃ T (s)Q1 x̃(s) ds t
tk−1 +k−1
x̃˙ (s)Q2 x̃(s)
˙ ds
T
ca
+
t t−d
+ x̃ T (s)Q1 x̃(s) ds,
tk +kca = d (t)(t) + 2T (t)N (t)
T

− 2T (t)T T B̃(x̃(t) − x̃t (d) − wk )

−t+tk−1 +k−1
ca t
x̃˙ (s)Q2 x̃(s)
˙ ds d
T t
V2 (x̃) =
x̃˙ (s)Q2 x̃(s)
˙ ds,
T
−d t+
+ (17)
t−d
−t+tk +kca t
x̃˙ (s)Q2 x̃(s)
˙ ds d for any matrices  and N satisfying (11).
T
+
−t+tk−1 +k−1
ca t+ Therefore, by the equality (16), and the inequality (17), V̇
0 t can be described as
x̃˙ (s)Q2 x̃(s)
˙ ds d,
T
+
−t+tk +kca t+ V̇ T (t)(M + d + 2N  + 2T T Â)(t) < 0, (18)

P > 0, Q1 > 0 and Q2 > 0. Then we have On the other hand, from (13), we have
˙ + x̃ T (t)Q1 x̃(t) − x̃ T (d)Q1 x̃t (d),
V˙1 (x̃) = 2x̃ T (t)P x̃(t) V = x̃(tk + kca + 0)T P x̃(tk + kca + 0)
t

− x̃(tk + kca )T P x̃(tk + kca )

and
= x̃(tk + kca )T x̃(tk + kca ) < 0. (19)
−t+tk−1 +k−1
ca
V˙2 (x̃) = − ˙T ˙ + ) ds
x̃ (t + )Q2 x̃(t The conditions (18) and (19) show the closed-loop system
−d
(9)–(10) with F = 0 is asymptotically stable. We therefore have
+ (d − t + tk−1 + k−1 ˙T ˙ obtained the desired result. 
ca )x̃ (t)Q2 x̃(t)

−t+tk +kca
Remark 4. In the proof of the theorem, a relaxed condition
x̃˙ (t + )Q2 x̃(t
˙ + ) d
T
−
−t+tk−1 +k−1
ca
(14) which involves the derivative ẋ(t) of the state x(t) is used
to derive the sufficient stability condition for the sampled-data
+ (tk + kca − tk−1 − k−1 ˙T ˙ system with time-varying time-delay. The existing results, see
ca )x̃ (t)Q2 x̃(t)
Cao, Hu, and Xu (2004) and the reference therein, show that
0
x̃˙ (t + )Q2 x̃(t
˙ + ) d
T Lyapunov–Krasovskii functionals (15) has an ability to reduce
−
−t+tk +kca the conservativeness in the stability analysis and design of the
time-delay systems. Introducing the relaxed condition (14) is
˙
+ (t − tk − kca )x̃˙ (t)Q2 x̃(t).
T
used to avoid introducing extra dynamics of the system de-
fined in Gu and Niculescu (2000), when the derivative term
Abusing the notation, we have of the state in Leibniz–Newton equation is directly applied to
t
the system equation. This relaxed condition (14) involves some
˙ −
V˙2 (x̃) = dx̃˙ (t)Q2 x̃(t) x̃˙ (s)Q2 x̃(s)
˙ ds.
T T weighting matrices which can also provide extra freedoms to
t−d obtain the less conservative stability conditions. Actually, we
 L.-S. Hu et al. / Automatica 43 (2007) 903 – 911 909

can obtain the following delay independent stability condition Theorem 4 is obtained from a less conservative stability con-
as a special case of the theorem for the sampled-data systems dition stated in Theorem 2, which shows the controller designed
with time-delay by setting T = [P , 0, 0, 0], Q2 = 0, N = 0 and based on Theorem 4 will have better performance than that on
 = 0. the existing results.

Corollary 3. The NCS (9)–(10) with F = 0 is stable for delay 5. Numerical example
d d0 , if there exist matrices P > 0 and Q1 > 0 such that the
following matrix inequalities hold: Example 1. Consider the following system described (Zhang,
  2001)
P Ã + ÃT P + Q1 P B̃    
< 0, 0 1 0
B̃ T P −Q1 ẋ(t) = x(t) + u(t),
0 −0.1 0.1
and (13).
y(t) = [ 0 1 ] x(t),
4. Control design which is controlled by a controller u(t) = −[3.75 11.5]x(t).
If this controller is implemented with networks, the allowable
In this section, we present a method to design a control (2) maximum time-delay in the control loop is 2.7 × 10−4 s given
such that the NCS (1)–(3) stable. by Walsh, Beldiman, and Bushnell (1999). The results given by
Zhang (2001) are 4.5 × 10−4 s and 0.0593 s. However, using
Theorem 4. If there exist matrices P > 0, Y and Y1 , F̄ , T = Theorem 2, we obtain the allowable maximum time-delay guar-
[T1 , T2 , T3 , T4 ], N satisfying the following matrix inequalities: anteeing stability is 0.699 s. This result is much more closer
⎡ ⎤
−P +ĀT P Ā ĀT P B̄ C̄ T F̄ T C̄ T F̄ T to the true value 1.7 s of the allowable maximum time-delay
⎢ B̄ T P Ā −Y ⎥ given by Zhang (2001) than the existing results, which shows
⎢ 0 0 ⎥
⎢ ⎥ the lower conservativeness of the proposed stability condition
⎣ F̄ C̄ 0 −Y1 −Y1 +Y T 0 ⎦ given by Theorem 2.
F̄ C̄ 0 0 −Y1 −Y1 +B̄ P B̄
T T

<0 (20) Example 2. Consider the following system:

   
and (11), (12), then the NCS (1)–(3) is stabilized via a state −0.8 −0.01 0.4
ẋ(t) = x(t) + u(t).
feedback controller, where the gain can be calculated by 1 0.1 0.1
F = Y1−1 F̄ Set the sample period Ts = 0.5 s. Using LMI Toolbox for
Matlab, we obtain a feedback controller by solving the matrix
whenever the loop delay satisfies d < d0 . inequalities (11), (12) and (20),

Proof. Eq. (13) can be expressed as ũ[tk ] = −[1.2625 1.2679]x(tk−1 ),

−P + (Ā + B̄F C̄)T P (Ā + B̄F C̄) < 0, for the sampling interval t ∈ (tk + kca , tk+1 + k+1
ca ], which
stabilizes the system.
which implies Theorem 4 shows that, the allowable maximum network-
induced time-delay guaranteeing the closed-loop system stable
− P + ĀT P Ā + C̄ T F T B̄ T P Ā
is kca = 0.5 s, that is, the allowable maximum time-delay of
+ C̄ T F T B̄ T P B̄F C̄ + ĀT P B̄F C̄ < 0. the closed-loop system is d0 = 1 s. Fig. 5 shows the dynamic
response of the system with the initial conditions x1 (0) = 1,
Noting the fact x2 (0) = 1.
C̄ T F T B̄ T P Ā + ĀT P B̄F C̄
The simulation results well show that the feasibility of the
 C̄ T F T Y F C̄ + ĀT P B̄Y −1 B̄ T P Ā, design procedure formulated in Theorem 4 for the NCS.
by Lemma 1, we can obtain 6. Conclusion
⎡ ⎤
−P +ĀT P Ā ĀT P B̄ C̄ T F T Y1T C̄ T F T Y1T
⎢ B̄ T P Ā −Y ⎥ In this paper, a networked-control system (NCS) with a
⎢ 0 0 ⎥
⎢ ⎥ continuous-time plant, a time-driven digital controller and a
⎣ Y1 F C̄ 0 −Y1 − Y1T +Y 0 ⎦
event-driven holder device is considered. This system is mod-
Y1 F C̄ 0 0 −Y1 −Y1T +B̄ T P B̄ elled as a sampled-data one with discrete-time delay. The model
<0. shows some nice features to describe many cases of the NCS.
By transferring time-delay in the discrete-time subsystem to
Let F̄ = Y1 F , we therefore have obtained the desired result. its continuous counterpart, an equivalent sampled-data system
 with continuous-time delay is obtained with no limitation on
910 L.-S. Hu et al. / Automatica 43 (2007) 903 – 911
Dynamic Response of Closed loop System

Fridman, E., Seuret, A., & Richard, J. P. (2004). Robust sampled-data

2 stabilization of linear systems: An input delay approach. Automatica, 40,
x1
1.5 x2 1441–1446.
u Gu, K., & Niculescu, S. I. (2000). Additional dynamics in transformed
1 time-delay systems. IEEE Transactions on Automatic Control, 45(3),
0.5 572–575.
Hu, L.-S., Cao, Y.-Y., & Shao, H.-H. (2002). Constrained robust sampled-data
0
control for nonlinear uncertain systems. International Journal of Robust
−0.5 and Nonlinear Control, 12(5), 447–464.
Hu, L.-S., & Huang, B. (2005). Multirate robust control for fuzzy systems
−1
with periodic Lyapunov function. IEEE Transactions on Fuzzy Systems,
−1.5 13(4), 436–443.
−2 Hu, L.-S., Huang, B., & Cao, Y.-Y. (2004). Robust digital model predictive
control for linear uncertain systems with saturations. IEEE Transactions
−2.5 on Automatic. Control, 49(5), 792–796.
−3 Hu, L.-S., Lam, J., Cao, Y.-Y., & Shao, H.-H. (2003). LMI approach to robust
0 2 4 6 8 10 12 14 16 18 20 H2 sampled-data control for linear uncertain systems. IEEE Transactions
of System, Man and Cybernet, Part B, 33(1), 149–155.
Time(s)
Hu, L.-S., Shi, P., & Frank, P. M. (2006). Robust sampled-data control for
Markovian jump linear systems. Automatica, 42, 2025–2030.
Fig. 5. Dynamic response of the networked-control system. Hristu-Varsakelis, D. (2001). Feedback control systems as users of a shared
network: Communication sequences that guarantee stability. Proceedings
of 40th Conference on Decision and Control, 4, 3631–3636.
the sampling period. Moreover, a new conservative stability re- Hristu-Varsakelis, D., & Kumar, P. R. (2002). Interrupt-based feedback control
sult for the NCSs is established using a new Lyapunov function over a shared communication medium. Proceedings of 41st Conference
and a relaxed condition. Based on the stability result, a state on Decision and Control, 4, 3596–3601.
Halevi, Y., & Ray, A. (1988). Integrated Communication and Control Systems:
feedback sampled-data control algorithm such that the closed- Part I—analysis and Part II—design consideration. ASME Journal of
loop system with possible maximum time-delay caused by the Dynamic System Measurements and Control, 110, 367–381.
networks is considered. The numerical examples demonstrate Krtolica, R., Oguner, U., & Chan, H. (1994). Stability of feedback systems
the effectiveness of the proposed method for the NCSs. with random communication delays. International Journal of Control,
59(4), 925–953.
Ling, Q. & Lemmon, M.D. (2002). Robust performance of soft real-
Acknowledgments time networked control systems with data dropouts. Proceedings of 41st
Conference on Decision and Control (pp. 1225–1230). Las Vegas.
The authors are grateful to the anonymous reviewers for Lian, F.-L., Moyne, J., & Tilbury, D. (2003). Modelling and optimal controller
design of networked control systems with multiple delays. International
their valuable comments and suggestions that helped improve
Journal of Control, 76(6), 591–606.
the presentation of the paper. This work was partially sup- Moon, Y. S., Park, P., Kwon, W. H., & Lee, Y. S. (2001). Delay-dependent
ported by Natural Science Foundation of China (60474020 robust stabilization of uncertain state-delayed systems. International
and 60574049), and Natural Science Foundation of Shan Xi Journal of Control, 74, 1447–1455.
Province (20051020). Peng Shi also gratefully acknowledges Nilsson, J. (1998). Real-time control systems with delays. Ph.D. Thesis,
Department of Automatic Control, Lund Institute of Technology.
the support from Harbin Institute of Technology; Nanjing Uni- Shi, P. (1998). Filtering on sampled-data systems with parametric uncertainty.
versity of Aeronautics and Astronautics; and the LCSIS, Insti- IEEE Transactions on Automatic Control, 43(7), 1022–1027.
tute of Automation, Chinese Academy of Sciences. Shi, P., Boukas, E. K., Agarwal, R., & Shue, S. (1998). Robust control of
linear continuous time-delay systems with finite discrete jumps and norm-
bounded uncertainties. International Journal of Systems and Sciences,
References
29(12), 1381–1392.
Shi, P., de Souza, C. E., & Xie, L. (1997). Bounded real lemma for linear
Boukas, E. K. (2005). Stabilization of stochastic nonlinear hybrid systems. systems with finite discrete jumps. International Journal of Control, 66(1),
International Journal of Innovative Computing, Information and Control, 145–159.
1(1), 131–141. Shi, P., Fu, M., & de Souza, C. E. (1996). Loop transfer recovery for systems
Boukas, E. K., & Al-Muthairi, N. F. (2006). Delay-dependent stabilization of under sampled measurements. IEE-D, Control Theory and Applications,
singular linear systems with delays. International Journal of Innovative 143(4), 333–337.
Computing, Information and Control, 2(2), 283–291. Walsh, G. C., Beldiman, O., & Bushnell, L. G. (1999). Stability analysis of
Boyd, S., El Ghaoui, L., Feron, E., & Balakrishnan, V. (1994). Linear matrix networked control system. Proceedings of American Control Conference
inequalities in system and control theory. Philadelphia: SIAM. (pp. 2876–2880). San Diego: IEEE Press.
Branicky, M. S., Phillips, S. M., & Zhang, W. (2002). Scheduling and feedback Walsh, G. C., Ye, H., & Bushnell, L. G. (2002). Stability analysis of networked
co-design for networked control systems. Proceedings of American Control control systems. IEEE Transactions on Control Systems Technology, 10,
Conference, 2, 1211–1217. 438–446.
Chen, B., Lam, J., & Xu, S. (2006). Memory state feedback guaranteed Xie, L., Shi, P., & de Souza, C. E. (1993). On designing controller for a class
cost control for neutral delay systems. International Journal of Innovative of uncertain sampled-data nonlinear systems. IEE-D, Control Theory and
Computing, Information and Control, 2(2), 293–303. Applications, 140(2), 119–126.
Chen, T., & Francis, B. (1995). Optimal sampled-data control systems. Berlin: Zhang, W. (2001). Stability analysis of networked control systems. Ph.D.
Springer. Thesis, Department of Electrical Engineering and Computer Science, Case
Cao, Y.-Y., Hu, L.-S. & Xu, A. (2004). A new delay-dependent stability Western Reserve University, August 4, 2001.
condition and H∞ control for jump time-delay systems. Proceedings of Zhang, W., Branicky, M. S., & Phillips, S. M. (2001). Stability of networked
American Control Conference, 4183–4188. control systems. IEEE Control and Systems Magazine, 21, 84–99.
 L.-S. Hu et al. / Automatica 43 (2007) 903 – 911 911

Li-Sheng Hu received his B.Sc. degree in He worked in the Defence Science and Technology Organisation, Depart-
engineering science, and his Ph.D. degree in ment of Defence, Australia from 1999 to 2003, as research scientist, senior
industrial automation, both from Zhejiang Uni- research scientist and task manager. In 2004, he joined the University of
versity, China in 1986 and 1988, respectively. Glamorgan, United Kingdom, as professor. Dr. Shi’s research interests in-
He was a postdoctoral fellow in Automation clude fault detection and tolerant control, intelligent systems and information
Engineering, Shanghai Jiao Tong University, processing, robust control and filtering, and operations research. He has pub-
Shanghai, China, from 1998 to 2000, and in lished a number of papers in these areas. He is a co-author (with W. Assaw-
Chemical Engineering, University of Alberta, inchaichote and S. Nguang) of the book Fuzzy Control and Filtering Design
Canada, from 2002 to 2003. He is now a for Uncertain Fuzzy Systems (Berlin, Springer, 2006), and a co-author (with
Professor in the Department of Automation, M. Mahmoud) of the book Methodologies for Control of Jump Time-Delay
Shanghai Jiao Tong University. His current Systems (Boston, Kluwer, 2003).
research interests are in the areas of robust model prediction control, process Dr. Shi serves as Editor-in-Chief of International Journal of Innovative Com-
monitoring, control performance limitation and performance assessment, and puting, Information and Control, and as Regional Editor of International
sampled-data control. Journal of Nonlinear Dynamics and Systems Theory. He is also an Associate
Editor for several other journals, such as IEEE Transactions on Systems, Man
Tao Bai received her M.S. degree in industry and Cybernetics-B, and IEEE Transactions on Fuzzy Systems. Dr. Shi is a
automation, and her Ph.D. degree in control en- fellow of Institute of Mathematics and its Applications (UK), and a senior
gineering from Shanghai Jiao Tong University member of the IEEE.
in 2001 and 2005, respectively. She was with
the Department of Automation, Tai Yuan Heavy Zhiming Wu served as a full professor in the Department of Automation,
Machinery Institute, and is now with the Depart- Shanghai Jiaotong University from 1985. His main research interests include
ment of Automation, Shanghai Jiao Tong Uni- discrete event system, artificial intelligence, soft-computing and their appli-
versity. Her research interests include industrial cation in manufacturing engineering.
networked system modeling, control and opti-
mization.

Peng Shi received his B.Sc. degree in mathemat-

ics from Harbin Institute of Technology, China
in 1982, the M.E. degree in control theory from
Harbin Engineering University, China in 1985,
the Ph.D. degree in electrical engineering from
the University of Newcastle, Australia in 1994,
and the PhD degree in mathematics from the
University of South Australia in 1998. He was
also awarded the degree of Doctor of Science
by the University of Glamorgan, UK in 2000.
From 1985 to 1989, Dr. Shi was a lecturer in
Heilongjiang University. He held visiting fellow position in the University of
Newcastle, Australia from 1989 to 1990. He was postdoctorate from 1995 to
1997, and lecturer from 1997 to 1999, in the University of South Australia.