1314
7, JULY 2007
Design of Networked Control Systems
With Packet Dropouts
Jing Wu and Tongwen Chen
Abstract—This note is concerned with stability and controller design of
networked control systems (NCSs) with packet dropouts. New NCS models
are provided considering both single- and multiple-packet transmissions.
Both sensor-to-controller (S/C) and controller-to-actuator (C/A) packet
dropouts are modeled and their history behavior is described by different
independent Markov chains. In term of the given models, sufficient
conditions for stochastic stability are derived in the form of linear matrix
inequalities (LMIs) and corresponding control laws are given. Numerical
examples illustrate the effectiveness of the results.
Index Terms—Markov chains, networked control systems (NCSs), packet
dropout, stochastic stability.
I. INTRODUCTION
In networked control systems (NCSs), control loops are closed
through real-time networks. Such networked systems bring new
functionalities that were not available in the past, such as low cost,
reduced system wiring, simple system diagnosis and maintenance, and
increased system agility. However, the insertion of communication
networks in feedback control loops makes the NCS analysis and
synthesis complex; see [1]–[3] and the references therein, where
much attention has been paid to the delayed data packets of an NCS
due to network transmissions. In fact, data packets through networks
suffer not only transmission delays, but also, possibly, transmission
loss/packet dropout [4], [5]; the latter is a potential source of insta-
bility and poor performance in NCSs because of the critical real-time
requirement in control systems. How such packet dropout affects
stability and performance of NCSs is an issue focused in this note.
Prior work, examining the effect of dropouts on system stability
and performance, can be roughly categorized into three types based
on the resulting closed-loop systems: switching systems [6], asyn-
chronous dynamical systems (ADSs) [7], and jump linear systems with
Markov chains [8]–[12]. It is noticed that all the stability conditions
and controller designs given in the aforementioned references are
derived based on the assumption that packet dropout exists only in
the sensor-to-controller (S/C) side. The effect of controller-to-actuator
(C/A) packet dropouts is neglected due to the complicated NCS mod-
eling. Recently, some results were obtained in [13]–[16], where ADSs
were introduced to model NCSs with packet dropouts on both S/C
and C/A sides [13], [14], a switching system was used to model NCSs
[15], and a linear system with stochastic variables was discussed in
[16] to describe NCSs with both-side packet dropouts. Moreover, the
controller gain in [13] is obtained by solving bilinear matrix inequal-
ities (BMIs) with rate constraints on the occurrence of events, while
in [14] the controller gain is chosen in advance by a pole placement
method considering rate constraints on the occurrence of discrete
states of a dynamical system. In [16], linear/nonlinear LQG optimal
Manuscript received February 23, 2006; revised August 1, 2006 and January
12, 2007. Recommended by Associate Editor M. Fujita. This work was sup-
ported by the Natural Sciences and Engineering Research Council of Canada
(NSERC).
The authors are with the Department of Electrical and Computer Engineering,
University of Alberta, Edmonton, AB T6G 2V4, Canada (e-mail: tchen@ece.
ualberta.ca).
Color versions of one or more of the figures in this note are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAC.2007.900839
0018-9286/$25.00 © 2007 IEEE
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO.
Fig. 1. NCS with data packet dropout via state feedback.
controllers were designed to minimize a cost function according to
the transmission control protocol (TCP) and user datagram protocol
(UDP). Those references discuss packet dropouts in the single-packet
transmission. As to the jump linear system approach, to our best
knowledge, no work has been done at present for modeling NCSs with
both S/C and C/A packet dropouts histories simultaneously. Note that
packet dropouts defined in the aforementioned references have two
cases, dropped or sent successfully, which are modeled as a Bernoulli
or a two-state Markov chain process.
In this note, Markov chains are introduced to describe S/C and C/A
packet dropouts. The Markov chains in our note describe the quan-
tity of packet dropouts between current time k and its latest successful
transmission instead of only the information on if a packet is dropped
or not, which is different from the aforementioned references. By this
definition, the number of states of Markov chains is larger than two and
the history of packet dropouts can be seen clearly. Under consideration
of network packet size constraints, new models of NCSs with packet
dropouts are presented according to the single- and multiple-packet
transmissions. By augmenting the state vector, the resulting closed-
loop system can be transformed to a standard jump linear system with
time delays, which enables us to apply the results of jump linear sys-
tems to the analysis and synthesis of such NCSs. Sufficient conditions
for stochastic stability are given and corresponding controller design
steps are provided. Examples are finally given to show the effective-
ness of our method.
This note is organized as follows. Section II introduces the basic
preliminary of our setup in Fig. 1. Sections III and IV consider the
modeling of NCSs with packet dropouts in single- and multiple-packet
transmissions, respectively. According to the resulting NCSs, the sto-
chastic stabilities and controller designs are discussed. Section V pro-
vides two numerical examples to illustrate the effectiveness of our re-
sults. Finally, Section VI gives some concluding remarks.
II. PROBLEM FORMULATION
Consider the NCS setup with data packet dropouts in Fig. 1, where
sensors, controllers, and actuators are clock-driven. The linear time-
invariant (LTI) plant we consider here is
x(k + 1) = 8x(k) + 0u(k) (1)
where x(k) 2 <n is the state and u(k) 2 <m 
is the input. 8 and 0 are
known real constant matrices with appropriate dimensions. Suppose
buffers are long enough to hold all the packets arrived, which will be
picked up according to the last-in-first-out rule. For example, when a
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 7, JULY 2007 1315

sensor data x(k) is lost, the controller will read out the most recent where F (dksc ) is a set of controllers and will be designed based on dksc .
data x(k 0 1) from the buffer and utilize it as x
(k) to calculate the new Substituting (2) and (5) into the system in (1), we have the following
control input, which will be sent to the plant; otherwise, the new sensor closed-loop system:
data x(k) will be saved to the buffer and used by the controller as x
(k).
Thus, for the buffers, we have 8x(k) + 0F (dksc )x(k); if dkca = 0
x(k + 1) = : (6)
8x(k) + 0u(k 0 1); otherwise dkca >0
u(k); if transmitted successfully
u(k) =
u(k 0 1); (k) = x(k 0 dksc ), which can be easily derived by iterations
(2)
otherwise Note that x
x(k); if transmitted successfully based on (3). To simplify the expression of the closed-loop system, we
x(k) = : introduce a function ( 1 ) to combine the previous closed-loop system
x(k 0 1);
(3)
otherwise
as
Moreover, due to the bandwidth and packet size constraints of the net-
work, the packet transmission can be classified into two types, single- x(k + 1) = 8x(k) +  (dkca ) 0u(k 0 1)
and multiple-packet transmissions. By this classification, we will have + [1 0  (dkca )] 0F (dksc ) x (k 0 dksc ) (7)
u(k) =  (dkca ) u(k 0 1)
two new NCS models for the setup in Fig. 1, which are given in the
following sections.
+ [1 0  (dkca )] F (dksc ) x (k 0 dksc ) (8)

III. MODELING AND CONTROLLER DESIGN OF NCSS WITH where

SINGLE-PACKET TRANSMISSIONS
In this section, system modeling, stability analysis, and controller
design are considered for the NCSs with single-packet transmissions.
The single-packet transmission means that data is lumped together into
one network packet and transmitted at the same time. This type of trans-
mission is suitable for networks with large packet sizes, e.g., Ethernet
which can hold a maximum of 1500 B of data in a single packet.
A. Modeling NCSs With Single-Packet Transmissions
Assume that dsc
k is the quantity of packets dropped at time k on the
S/C side, which is calculated from the current time k to the last suc-
cessful transmission (happened at time k 0 dksc ), dkca is the packet quan-
tity dropped on the C/A side between the current time k and its last
successful transmission at time k 0 dkca , and both of them are bounded.
Thus, we have
1; dkca > 0
 (dkca ) = :
0; dkca = 0
Remark 1: The value of ( 1 ) depends on whether the designed
control signal is successfully transmitted or not (namely, dkca = 0 or
dkca > 0), instead of how many designed control signals are dropped
(the value of dkca ). This classification can simplify the modeling of the
closed-loop system since the control input u(k) will not be updated no
matter what value dkca > 0 will be. That is, the control signal u(k)
will be the same when dkca = 1; 2; 3; . . . ; d2 . Another advantage of
this classification is to avoid introducing this unknown dkca in the aug-
mented state vectors and controller design. Thus, we replace u(k) with
(2) instead of u(k) = u (k 0 dkca ), the deriving method being the same
as the iteration method for x (k).
Concatenating plant and controller state vectors to obtain a global
vector z(k) = [xT (k)uT (k 0 1)]T by (7)–(8), we can obtain the
closed-loop system for the NCS with single-packet transmissions in
Fig. 1 as

0  dksc  d1 0  dkca  d2 z(k + 1) =

8 0 (dkca ) x(k)
0  (dkca ) u(k 0 1)
where d1 and d2 are nonnegative integers. We model dksc and dkca as two (1 0  (dkca ))0F (dksc ) 0 x (k 0 dksc )
+
homogeneous independent Markov chains, which take values in S1 = (1 0  (dk ))F (dk ) 0 u (k 0 dksc 0 1)
ca sc
f0; 1; . . . ; d1 g and S2 = f0; 1; . . . ; d2 g with the generators 1 = = A (dk ) z(k) + B (dkca ; dksc ) z (k 0 dksc ) :
ca
(9)
(ij ) and 2 = (mn ), respectively. The transition probabilities of
dksc (jumping from mode i to j ) and dkca (jumping from mode m to n) Remark 2: The resulting closed-loop system in (9) is a jump
are defined by linear system with two modes (dksc and dkca ) and one mode-dependent
time-varying delay dksc , where their transitions are described by two
ij = Pr (dksc+1 = j j dksc = i) Markov chains, which give the history behavior of S/C and C/A packet
mn = Pr (dkca+1 = n j dkca = m ) (4) dropouts, respectively. This also enables us to apply the results of
jumping linear systems with time-delays to the analysis and synthesis
whereij  0; i; j 2 S1 ; mn  0; m; n 2 S2 , and dj=0 ij = of such NCSs.
d
1; n=0 mn = 1. It is obvious that the transition probabilities satisfy Before proceeding, we need the following definition.
Definition 1 [17]: The free nominal jump discrete-time system in
(9) is said to be stochastically stable, if for all finite zk = ' 2 <n +m

ij = 0; if j 6= i + 1 and j 6= 0 defined on k 2 [0d1 ; 0] and initial model d0 ; d0 , there exists a finite
sc ca

mn = 0; if n 6= m + 1 and n=6 0 number 4 ~ ('; d0sc ; d0ca ) > 0 such that
N
which can be derived by packet dropout definitions. Assume the state
feedback control law is
lim
N !1 E kzk k2 '; d0sc ; d0ca < 4~ ('; d0sc ; d0ca ) (10)
k=0
u(k) = F (dksc ) x(k) (5) holds, where E is the statistical expectation operator.
1316 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 7, JULY 2007

B. Stability Analysis and Controller Design of NCSs With where P (dksc ; dkca ) = P T (dksc ; dkca ) > 0; and R = RT > 0 are to be
Single-Packet Transmissions determined. Then, we have

In this section, a sufficient condition on the stochastic stability of E fV (z(k + 1); k + 1) j V (z(k); k)g 0 V (z(k); k)
the system in (9) with single-packet transmissions is derived. For nota-  zT (k + 1)P (i; m)z(k + 1) + zT (k)
tional simplicity, in the sequel, for dsc ca
k = i 2 S1 ; dk = m 2 S2 , we 2 [R 0 P (i; m)]z(k) 0 zT (k 0 i)Rz(k 0 i)
ca sc ca
denote A(dk ) Am ; B(dk ; dk ) Bi;m , and k0d d
0 (1 0 ) z T ()Rz() + ij
 = minfii ; i 2 S1 g d1 = minfdksc ; k 2 g  =k0d +1 6 i
j =0;j =
k01 k01
 = 1 + (1 0 )(d1 0 d1 )
2 z T ()Rz() 0 z T ()Rz()
K = (I + 0T 0)01 [ 0T I ] :  =k0d +1  =k0d +1

 (k) W (k; i; m)(k)

T
(13)
Then, we have Theorem 1.
Theorem 1: The system in (9) is stochastically stable if there exist where the equation shown at the bottom of the page holds. Let Xi;m =
Xi;m > 0; Q > 0, and 9i;m such that the following LMI: P 01 (i; m); 9i;m = R01 Bi;m
T
; and Q = R01 . By Schur comple-
ments and (11) and (12), we know that W (k; i; m) < 0. Thus, we
0Xi;m 0 Xi;m AmT 2 Xi;m have
3 0Q 9i;m 2 0
<0 E fV (z(k + 1); k + 1)jV (z(k); k)g 0 V (z(k); k)
3 3 0
0
(11)
 0min [0W (k; i; m)]T (k)(k)
3 3 3 0 1 Q
 0zT (k)z(k) (14)

holds for all i; j 2 S1 and m; n 2 S2 , where 3 denotes blocks that are where min [0W (k; i; m)] denotes the minimal eigenvalue of
readily inferred by symmetry and 0W (k; i; m) and  = inf fmin [0W (k; i; m)]; i 2 S1 ; m 2 S2 g.
From (14), we derive that for any N  1
2 = [ m1 i1 I; . . . ; mn ij I; . . . ; md id I] E fV (z(N + 1); dNsc+1 ; dNca+1 )g
d 1d N

Q = R01
= diag[X1;1 ; . . . ; Xj;n ; . . . ; Xd ;d ]: (12) 0E fV (z(0); d0sc ; d0ca )g  0 E kz T (k)z(k)k:
k=0

Here, d1 d2 is the number of matrices. The control law we have is

This yields, for any N 1
N
E kz T (k)z(k)k < 1 E fV (z(0); d0sc ; d0ca )g
u(k 0 1); dca > 0 k=0

u(k) = K9T Q01 [ I 0 ]T x (k 0 dsc ) ; dkca = 0 :
i;0 k k N
) Nlim
!1 E kz T (k)z(k)k
Proof: Let the Lyapunov–Krasovskii functional be k=0

< 1 E fV ('; d0sc ; d0ca )g = 4~ ('; d0sc ; d0ca ) :

k 01 
V (z(k); k) = z T (k)P (dksc ; dkca )z(k) + z T ()Rz() By Definition 1, the system in (9) is stochastically stable. In addition,
 =k d0 by 9i;m = R01 Bi;m T
, we have (15), shown at the bottom of the page;
0d k 01 then, the controller Fi in (5) is obtained by premultiplying [ I 0 ]Q01
+ (1 0 ) z T ()Rz() and postmultiplying [ 0T I ]T to both sides of (15). Note that 0T 0+I
0 +1  =k+
= d is of full rank. The proof is completed.

d d
(k) = [z T (k)z T (k 0 i)]T P (i; m) = mn ij P (j; n)
n=0 j =0

W (k; i; m) =
0P (i; m) + R + ATm P (i; m)Am ATm P (i; m)Bi;m
:
3 0R + Bi;m
T 
P (i; m)Bi;m

T
0Fi 0 I T T
9i;m = (1 0  (dkca )) Q = (1 0  (dkca )) Q F [0 I ]
Fi 0 0 i
I T T ca ca
) 9i;0 = Q 0 Fi [ 0 I ] ;  (dk ) = 0 , dk = m = 0 (15)
9i;m = 0;  (dkca ) = 1 , dkca = m > 0
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 7, JULY 2007 1317

IV. MODELING AND CONTROLLER DESIGN OF NCSS WITH
MULTIPLE-PACKET TRANSMISSIONS
From Section III, we know that the single-packet transmission is
usually used for networks with enough capacity for large-size packets.
However, some packet-switched networks can only carry limited infor-
mation in a single packet due to packet size constraints, e.g., DeviceNet
which has a maximum 8-B data field in each packet. Thus, in such net-
works, the multiple-packet transmission is widely used, where sensor
or actuator data is transmitted in separate network packets and may not
arrive at the controller and plant simultaneously.
A. Modeling NCSs With Multiple-Packet Transmissions
with dj=0 ij = 1; dq=0 pq = 1, and dn=0 mn = 1. Similarly,
as in the single-packet transmission case, the closed-loop system in this
case can be modeled by (18), shown at the bottom of the page, where
the control law to be designed is
F (dsc ) = F  2 (d1sc ) F  2( 0 ) (d2sc ) :
k m r k m n r k (19)
Remark 3: It is easy to see that the complexity of the closed-loop
models depends on M and N : The larger M and N are, the more com-
plicated models we obtain. However, we remark that this complication
will not affect our method for stability analysis and controller design;
only the mathematical calculation is complicated.
Definition 2 [17]: The free nominal jump discrete-time system in
(18) is said to be stochastically stable, if for all finite zk = ' 2 <n +m 

defined on k 2 [0 max(d11 ; d12 ); 0] and initial model d10

Due to the characteristic given previously, the NCSs are modeled
with M + N(M + N > 2) Markov chains. Split the plant state ; d20 ; d0 ,
sc sc ca

there exists a finite number 4~ ('; d10 ; d20 ; d0 ) > 0 such that
sc sc ca
into M separate packets as x(k) = [X1T (k); X2T (k); . . . ; XM
T
(k)]T
and the controller output into N separate packets as u (k) = N

[U1T (k); U2T (k); . . . ; UNT (k)]T , where N

lim
!1 E kz k2 '; d10
k ; d20 ; d0 < 4~ ('; d10
sc sc ca sc sc ca
; d20 ; d0 )
k=0
T (20)
X1 (k) = x1 (k); x2 (k); . . . ; x (k)
T T T
r
holds, where E is the statistical expectation operator.
..
. B. Stability Analysis and Controller Design of NCSs With
T
X (k) = x
M
T
r +1 (k); xr
T
+2 (k); . . . ; xn
 (k)
T
(16) Multiple-Packet Transmissions
In this section, the stability of NCSs with multiple-packet transmis-
and 1  r1 < 1 1 1 < rM 01  n ; then, we have the corresponding sion is investigated. Denote
input of the controller x
(k) as
 = minf ; i 2 S11 g ii

x(k) = X 1 (k); X2 (k); . . . ; X (k)

T

d11 = min fd1sc = i; k 2 g

T T T
M
k

X 1 (k) = x1 (k); x2 (k); . . . ; x (k)

T T T
T
1 = 1 + (1 0 )(d11 0 d11 )
 = minf ; p 2 S12 g
r

pp
..
. d12 = min fd2sc = p; k 2 g k

X (k) = x 2 = 1 + (1 0 )(d12 0 d12 ):

T

+1 (k); x
 +2 (k); . . . ; x
 (k) :
T T T
M r r 
n

Similar definitions are used for the signals u (k) and u(k), and are Then, we have Theorem 2.
omitted. For the sake of simplicity, in this note, we assume two-packet Theorem 2: The system in (18) is stochastically stable if there exist
transmission on the S/C side and single-packet transmission on the C/A Xi;p;m > 0; Q1 > 0; Q2 > 0; 91;i;m and 92;p;m such that the
side; i.e., M = 2 and N = 1. Thus LMI in (21), shown at the bottom of the next page, holds for all i; j 2
S11 ; p; q 2 S12 and m; n 2 S2 , where 3 denotes blocks that are readily
X (k) X (k 0 dsc
1 )
x(k) =  1 = 1 k inferred by symmetry and
X2 (k) X2 (k 0 d2sc ) k

u(k) = U1 (k) = U1 (k 0 dca ) k

21 = [  1  1  1 I; . . . ;    I]
i p m id pd md

1 1 d d d
where we set r1 = r; d1sck and d2sck reflect the quantities of S/C packet
dropouts in channels 1 and 2, and dkca reflects the quantity of C/A packet
Q1 = R101 Q2 = R201
dropouts. Their transition probabilities are given by
1 = diag[X1 1 1 ; . . . ; X ; . . . ; X ; ; ]: (22) j;q;n d ;d ;d

Here, d11 d12 d2 is the number of matrices. In this case, the control law
+1) = j j d1 = i
sc sc
 = Pr d1(
ij k k

  0; i; j 2 S11 = f0; 1; . . . ; d11 g

is
ij
I
 = Pr d2( +1) = q j d2sc = p F  2 (d1sc ) = K91 0 Q101
r
sc
pq k k m r k
T
;i; 0 2( 0 )
r n r

  0;
pq p; q 2 S12 = f0; 1; . . . ; d12 g 0
 = Pr (d +1 = n j dca = m)
ca 0( 0 )2
Q201
n r r
mn k k
F  2( 0 ) (d2sc ) = K92 T
I0 : (23)
  0; m; n 2 S2 = f0; 1; . . . ; d2 g
m n r k ;p;0 n r
mn (17) 0

8 0 (dca ) (1 0  (dca )) 0F  2 (d1sc ) [I 0 2( 0 ) ] 0

z(k + 1) = z(k) +
k k m
z (k 0 d1sc )
r k r r n r

0  (d ) ca
k (1 0  (dca )) F  2 (d1sc ) [I 0 2( 0 ) ] 0
k m r k r r n r
k

(1 0  (dca )) 0F  2( 0 ) (d2sc ) [0( 0 )2 I  0 ] 0

+ k m n r k
z (k 0 d2sc )
n r r n r

(1 0  (dca )) F  2( 0 ) (d2sc ) [0( 0 )2 I  0 ] 0

k m n r k n r r n r
k

= A (dca ) z(k) + B1 (dca ; d1sc ) z (k 0 d1sc ) + B2 (dca ; d2sc ) z (k 0 d2sc )

k k k k k k k (18)
1318 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 7, JULY 2007

Proof: Define a Lyapunov–Krasovskii functional as

V (z (k); k) = z T (k)P (dsc sc ca

1k ; d2k ; dk ) z (k)
k01
+ z T ( )R1 z ( ) + (1 0 )
 =k0d
0d k01
2 z T ( )R 1 z ( )
=0d +1  =k+
k01
+ z T ( )R2 z ( ) + (1 0 )
 =k0d
0d k01
2 z T ( )R 2 z ( )
=0d +1  =k+

where P (d1sck ; d2sck ; dkca ) = P T (d1sck ; d2sck ; dkca ) > 0; R1 = R1T >
0; and R2 = R2T > 0 are to be determined. Then, the proof follows a
similar procedure as that for the single-packet dropout case; hence, it
is omitted.
Remark 4: Theorem 2 gives a sufficient condition for the stochastic
stability of NCSs with packet dropouts, which are described by three
Markov chains. This method can also be extended to the case of NCSs
with both dropouts and network-induced delays.

V. NUMERICAL EXAMPLES
In this section, two examples are used to demonstrate that the
methods presented in this note are effective. We compare our method
with the traditional pole placement method, which shows that the
packet dropouts do affect the system stability. For simplicity, we just
do this comparison in the single-packet transmission case.
Example 1 (Single-Packet Transmission Case): Consider the fol-
lowing discrete-time system:
Fig. 2. State response with single-packet transmissions (Example 1): (a) and
(b) with zero packet loss by pole placement method; (c) and (d) packet dropout
history; (e) and (f) with packet loss by pole placement method; and (g) and (h)
with packet loss by our method.
converge finally but with poor dynamics, e.g., when the poles are
placed at f0:3; 0:3; 0g, the state trajectories are oscillatory; when the
poles are with magnitude less than 0.3, then the state trajectories start

8=
00:7 2
0=
00:03 :
converging but with poor ripples in our example.
0 0 1: 5 01 (24) By these two random serials, we can calculate their transition prob-
ability matrices as
It is clear that the discrete-time system is unstable. 0:5385 0:4615 0
By pole placement method, we have the controller gain 31 = 0:8333 0 0:1667
K = [00:9684; 3:6292] (closed-loop poles are f0:7; 0:7; 0g), 1 0 0
which is designed without considering the packet dropouts. Suppose 0:75 0:25
the initial condition is z (0) = [0:1; 0:1; 0]T , we have Fig. 2(a) and (b), 32 = :
1 0
from which we can see that the system in (9) can be stabilized by
this pole placement controller when there is no packet dropout. Then, by Theorem 1 of our method, we have the following controllers:
However, this method fails when there are packet dropouts, see
Fig. 2(c)–(f); Fig. 2(c) and (d) gives the changes in packet dropout F0 = [ 00:0451 01:5041 ]
numbers. dksc = i; i 2 f0; 1; 2g means i packets on the S/C side are F1 = [ 00:0902 03:0081 ]
dropped during the transmission. Similarly, dkca = m; m 2 f0; 1g F2 = [ 00:1354 04:5122 ]:
means that m packets on the C/A side are dropped. Fig. 2 (e) and (f)
are the state trajectories of the system in (9), by which it is easy to With same initial conditions, we have Fig. 2(g) and (h), by which we
see that packet dropouts destroy the system stability. In fact, if we know that the system in (9) can be stabilized by our designed controller;
place the eigenvalues closer to the origin, the state trajectories will so our method is more effective.

0Xi;p;m 0 0 Xi;p;m ATm 21 Xi;p;m Xi;p;m

3 0 Q1 0 91;i;m 21 0 0
3 3 0 Q2 92;p;m 21 0 0
<0
3 3 3 0
1 0 0
(21)

3 3 3 3 0 1 Q1 0
3 3 3 3 3 0 1 Q2
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 52, NO. 7, JULY 2007 1319

Fig. 3. State response with multiple-packet transmissions (Example 2).

Example 2 (Multiple-Packet Transmission Case): This example [2] R. Krtolica and U. Ozguner, “Stability of linear feedback systems with
demonstrate that our method is also feasible for the multiple-packet random communication delays,” Int. J. Control, vol. 39, no. 4, pp.
dropout case. The system is 925–953, 1994.

01 43 0 1 0 = 02 5
[3] L. A. Montestruque and P. Antasklis, “Stability of model-based net-
8=
: : : worked control systems with time-varying transmission times,” IEEE
0 00 3 : 00 15 :
:
Trans. Autom. Control, vol. 49, no. 9, pp. 1562–1572, Sep. 2004.
[4] D. Yue, Q. L. Han, and C. Peng, “State feedback controller design of
For simplicity, we suppose 1 2 f0 1 2g 2 2 f0 1g, and ca 2
sc
d k
sc
; ; ;d k ; dk networked control systems,” IEEE Trans. Circuits Syst. II, Exp. Briefs,
f0 1g. Their transition probability matrices for 1sc 2sc and ca are
; d k; d k; dk vol. 51, no. 11, pp. 640–644, Nov. 2004.
0:4615 0:5385 [5] L. Zhang, Y. Shi, T. Chen, and B. Huang, “A new method for stabiliza-
31 301 = 32 tion of networked control systems with random delays,” IEEE Trans.
1 0 Autom. Control, vol. 50, no. 8, pp. 1177–1181, Aug. 2005.
respectively, where 31 ; 32 are the same as those in Example 1. There
[6] M. Yu, L. Wang, G. M. Xie, and T. G. Chu, “Stabilization of net-
worked control systems with data packet dropout via switched system
are two channels, namely, X1 = x1 (k ) and X2 = x2 (k). Then, by approach,” in Proc. IEEE Int. Symp. Comput. Aided Control Syst. De-
Theorem 2, we can obtain the control law as sign, 2004, pp. 362–367.

= 00:3966 = 00:4231
[7] W. Zhang, M. S. Branicky, and S. M. Phillips, “Stability of networked
F10 F11 control systems,” IEEE Control Syst. Mag., vol. 21, no. 1, pp. 84–99,
F12 = 00:4495 F20 = 00:0265 Feb. 2001.
[8] J. Nilsson, “Real-time control systems with delays,” Ph.D. dissertation,
F21 = 00:0280 Dept. Autom. Control, Lund Inst. Technol., Lund, Sweden, 1998.

f0 1 2g, belong to the controller

[9] L. Xiao, A. Hassibi, and J. P. How, “Control with random communica-
where F1 i ; i = ; ; set tion delays via a discrete-time jump linear system approach,” in Proc.

Fm 2r (d1sck ) and F2 p ; p = f0 1g, belong to the controller
; set Amer. Control Conf., 2000, pp. 2199–2204.
n0r) (d2k ).
sc
 2(
Fm The state trajectories versus the time is shown [10] Q. Ling and M. D. Lemmon, “Soft real-time scheduling of networked
in Fig. 3, which illustrates that the controllers we designed can control systems with dropouts governed by a Markov chain,” in Proc.
Amer. Control Conf., 2003, pp. 4845–4850.
guarantee the stochastic stability of the NCSs with the multiple-packet [11] P. Seiler and R. Sengupta, “An h approach to networked control,”
transmissions. IEEE Trans. Autom. Control, vol. 50, no. 3, pp. 356–364, Mar. 2005.
[12] P. Seiler and R. Sengupta, “Analysis of communication losses in
VI. CONCLUSION vehicle control problems,” in Proc. Amer. Control Conf., 2001, pp.
1491–1496.
In this note, the problem of stability analysis and controller design [13] A. Hassibi, S. P. Boyd, and J. P. How, “Control of asynchronous dy-
has been proposed based on a new model of the NCSs with single- namical systems with rate constraints on events,” in Proc. 38th IEEE
packet transmission. In this model, S/C and C/A packet dropout his- Conf. Decision Control, 1999, pp. 1345–1351.
tory behaviors are described by two independent Markov chains, the [14] A. Rabello and A. Bhaya, “Stability of asynchronous dynamical sys-
definitions of which are different from those in existing references. tems with rate constraints and application,” Inst. Electr. Eng. Proc.
—Control Theory Appl., vol. 150, no. 5, pp. 546–550, 2003.
Moreover, NCSs with multiple-packet transmission are also investi- [15] H. Ishii, “h control with limited communication and message losses,”
gated. The derived control laws guarantee stochastic stability of the in Proc. Amer. Control Conf., 2006, pp. 5620–5625.
resulting closed-loop systems. Simulation results illustrate the feasi- [16] B. Sinopoli, L. Schenato, M. Franceschetti, K. Poolla, and S. Sastry,
bility and effectiveness of our methods. “An LQG optimal linear controller for control systems with packet
losses,” in Proc. 44th IEEE Conf. Decision Control/Eur. Control Conf.,
2005, pp. 458–463.
REFERENCES [17] J. Wu, T. Chen, and S. Xu, “Stochastic stabilization and h control
[1] P. V. Zhivoglyadov and R. H. Middleton, “Networked control design for discrete jumping systems with time delay,” Asian J. Control, vol. 7,
for linear systems,” Automatica, vol. 39, pp. 743–750, 2003. no. 3, pp. 223–230, 2005.