Professional Documents
Culture Documents
Hai Lin, G. Zhai, P. J. Antsaklis 2003 Robust Stability and Disturbance Attenuation Analysis of A Class of Networked Control Systems
Hai Lin, G. Zhai, P. J. Antsaklis 2003 Robust Stability and Disturbance Attenuation Analysis of A Class of Networked Control Systems
Robust Stability and Disturbance Attenuation Analysis of a Class of Networked Control Systems
Abstract-In this paper, stability and disturbance attenua- problem in NCSs is the packet dropout issue. Because of
tion issues for a class of Networked Control Systems (NCSs) the uncertainties and noise in the communication channels,
under uncertain access delay and packet dropout elTects are there may exist unavoidable errors in the transmitted packet
considered. Our aim is to find conditions on the delay and or even losst. If this happens, the conuptedpacket is dropped
packet dropout rate, under which the system stability and
71" disturbance attenuation properties are preserved to a and the receiver (controller or actuator) uses the packet
desired level. The basic idea in this paper is to formulate such that it received most recently. In addition, packet dropout
Networked Control System as a discrete-time switched system. may occur when one packet, say sampled values from the
Then the NCSs' stability and performance problems can be sensor, reaches the destination later than its successors. In
reduced to corresponding problems for the switched systems, such situation, the old packet is dropped, and its successive
which have been studied for decades and for which a number
of results are available in the literature. The techniques in this packet is used instead. There is another important issue in
paper are based on recent progress in the discrete-time switched NCSs, that is the quantization effect. With finite hit-rate
systems and piecewise Lyapunov functions. constraints, quantization has to be taken into consideration
in NCSs. Therefore, quantization and limited bit rate issues
I. INTRODUCTION have attracted many researchers' attention, see for example
By Networked Control Systems (NCSs), we mean feed- [3], [6], 141. It has been known [3], [6] that an exponential
hack control systems where networks, typically digital band- data representation scheme is most desirable under certain
limited serial communication channels, are used for the conditions.
connections between spatially distributed system components In this paper, we consider uncertain time delay and packet
like sensors and actuators to controllers. These channels may dropout issues of NCSs in the framework of switched sys-
be shared by other feedback control loops. In traditional feed- tems. The strength of this approach comes from the solid
hack control systems these connections are through point-to- theoretic results existing in the literature for stability, robust
point cables. Compared with the point-to-point cables, the performance etc. for switched systems. By a switched system,
introduction of serial communication networks has many we mean a hybrid dynamical system consisting of a finite
advantages, such as high system testability and resource number of subsystems described by differential or difference
utilization, as well as low weight, space, power and wiring equations and a logical rule that orchestrates switching he-
requirements [9], 1141. These advantages make the networks tween these subsystems. Properties of this type of model have
connecting sensorslactuators to controllers more and more been studied for the past fifty years t o consider engineering
popular in many applications, including traffic control, satel- systems that contain relays and/or hysteresis. Recently, there
lite clusters, mobile robotics etc. Recently modeling, analysis has been increasing interest in the stability analysis and
and control of networked control systems with limited com- switching control design of switched systems (see, e.g., [IO],
munication capability has emerged as a topic of significant [SI, [ l l ] , [12] and the references cited therein). Notice that
interest to control community, see for example [4], [3], [6], hybridhwitched system research has provided useful results
~ 4 1 [I].
. t91. and promising techniques to deal with NCSs with delay and
Time delay typically has negative effects on the Networked packet dropout effects. For example, in [14], the multiple
Control Systems' stability and performance. There are sev- Lyapunov function method [2] was employed to analyze the
eral situations where time delay may arise. First, transmission stability of NCSs with network-induced delay. In addition,
delay is caused by the limited bit rate of the communication the packet dropout effects in NCSs are closely related to
channels. Secondly, the channel in NCSs is usually shared the controller failures studied in the fault tolerance literature.
by multiple sources of data, and the channel is usually There are some recent work on the controller failure analysis
multiplexed by time-division method. Therefore, there are based on switched system techniques 11 I], [13]. By using a
delays caused by a node waiting to send out a message piecewise Lyapunov function, the author in [I31 showed that
through a busy channel, which is usually called accessing if the controller failures did not occur too frequently or last
delay and serves as the main source of delays in NCSs. There
are also some delays caused by processing and propagation, 'Error control coding andlor Automatic Repeal ~CQUCSI
(ARQ) mecha-
which are usually negligible for NCSs. Another interesting nism may be employed, but the possibility of e m r occurring still exisrs.
1183
111. MODELS FOR NETWORKED CONTROL Then, we may plug in the u[k]= K z [ k ] and get
SYSTEMS
+
z [ k 11
In this section, we will consider the sampled-data model of
the plant. Because we d o not assume the synchronization be-
+
= As[k] K B K z [- + +
~ 11 ( N - t c ) B K z [ k ] Ed[k]
tween the sampler and the digital controller, the control signal + +
= [ A ( N - n ) B K ] z [ k ] ~ t B K z [- +
k 11 Ed[k]
is no longer of constant value within a sampling period.
Therefore the control signal within a sampling period has to
If we let 4 k j = [ z[til1, then the above equa-
be divided into subintervals corresponding to the controller's tions can be written as:
reading buffer period, T = 3.
Within each subinterval.
the control signal is constant under the assumptions of the
previous section. Hence the continuous-time plant may be
discretized into the following sampled-data systems:
1
u l ~ k1~
"'.~ I
1
+ E d [ k ] (I)
where n = 0 , 1 , 2 , . . . ,D,,,. Note that K = 0 implies
r = 0, which corresponds to the previous "no delay"
case. And the controlled output z [ k ] is given by
where A = eAPT*, B = &%eAC"BCdq and E =
J,'. eA'qECdq. Note that for linear time-invariant plant and z [ k ]= [ 0 c ] ?[k]
constant-periodic sampling, the matrices A, B and E are where C = Cc.
constant. 3) If packet dropout happens, due to corrupted packet or
During each sampling period, several different cases may delay T > Dmo,x %, then the actuator will implement
arise, which can be listed as follows. the previous control signal, i.e. u'[k] = ~ ' [ k=] . . . =
u N [ k ]= u [ k - 11, then the state transition equation (1)
1) No delay, r = 0, u'[k] = u 2 [ k ]= ' . . = u N [ k ]= u [ k ] , for this case can be written as follows.
then the state transition equation ( I ) for this case can
be written as follows.
r ulkl 1
1 r'lk- 11 1
= Az[k] + N . Bu[k - 11t Ed[k]
Using the same variable transformation as in the above
+ Ed[k] case, we get
= Azjk] + N . Bulk] + Ed[k]
2) Delay 7 = K x %,
where n = 1 , 2 . . . . ,D,,,. For
this case u'[k] = u'[k] = . . . = un[k] = ~ [ -k 11,
uK+'[k]= u"+'[k] = . .. = u N [ k ] = u[k],and
Equation (1) can be written as:
The controlled output z [ k ] is given by
z [ k ]= [0 c ] ?[k]
z[k + 11 = A i [ k ]+ [ B B . .. B ] I "[:[kjll I where C = Cc.
In the next section, we will formulate the above NCSs as a
class of discrete-time switched systems.
IV. STABILITY ANALYSIS
+ Ed[k] Motivated by the above analysis of NCSs, we introduce
a family of discrete-time linear systems described by the
= Az[k] + K . B u [ k - 11 +(N- K) +
B u [ k ] Ed[k] following difference equations.
Let us assume that the controller uses just the time-
invariant linear feedback control law, u [ k ] = K z [ k ] ,
+
z [ k 11 = A , z [ k ] E,d[k], k E Z++ (2)
which may he obtained as the solution of a LQR prob- where z [ k ] E E%" is the state variable, and the disturbance
lem without considering the network induced effects. input d [ k ] is contained in D ' C .'%E A, E RnX"and E, E
1184
R””‘ are constant matrices indexed by q E Q, where the kl > 0, let N o ( k l ,kz) denote the number of switchings of
finite set Q = { q l . 42,. . . ,qn} is called the set of modes. u ( k ) on the interval [kl,kz). Let K , ( k l .kz) denote the total
Combine the family of discrete-time uncertain linear sys- period that the i-th subsystem is activated during [ k l , k z ) .
-
tems (2) with a class of piecewise constant functions of time
U : Z+ Q. Then we can define the following linear time-
varying system as a discrete-time switched linear system
Define K - ( k I , kZ) = xis?, i E Q K,(kl, kz), which stands
for the total activation period of the Schur stable subsystems.
On the other hand, K + ( k l , k?) = xi>v, iGQ Ki(k1, kz)
denotes the total activation period of the Schur unstable
+ +
~ [ kl ] = A , j k ] ~ [ k ] E,lqd[k], k E Z+ (3)
+
subsystems. We have K - ( k l , k2) K + ( k l l k2) = kz - k 1.
The signal u [ k ] is called a switching signal. For given NO 2 0, T ~ let . Sa(ra) denote the set of all
Associated with the switched system (3), a controlled switching signals satisfying
output r [ k ] is considered.
k
4kl = C,[klz[kl No(0. k ) 5 No +-
Ta
(6)
where Calk] E RPYnand t j k ] E RP.
For the NCS we considered in this paper, we may formu- where the constant ra is called the average dwell time and
late it as a switched system with D,, +
2 different modes, NO the chatter bound. The idea is that there may exist
which can he expressed as follows. consecutive switching separated by less than T,, hut the
average time interval between consecutive switcbings is not
t [ k + I] = Ar?[kj + E,d[k]
{ z [ k ] = C,?[k]
(4) less than ra. Note that the concept of average dwell time
between subsystems was originally proposed for continuous-
[0 I
w h e r e A K = KBK . A ~ + ( N - K ) B K ] ’ ~ ~ =
and C, = I 0 C 1 for K = 0 . 1.. 2., . ’ ’ D,,,. . N . And the
[ L] time switched systems in [8]. With these assumptions and
notations, we may apply the techniques and results developed
in [12] to the NCSs and get the following theorem for
set of moles Q is hiven by Q = { 0 , 1 , 2 , .. D,,,, N}. globally exponential stability. The proof of the theorem is
Note that K = 0 implies T = 0, which corresponds to the not difficult by using the technique of Theorem 3 in [121,
“no delay” case, while K. = N corresponds to the “packet and thus is omitted here.
dropout” case. Theorem I: For any given X E AI,^), the NCS (4) is
It is reasonable to assume that, for the cases of no delay globally exponentially stable with stability degree X if there
(n = 0) or small delay ( R 5 no), the corresponding exists a finite constant r,‘ and A’ E (X1,X) such that the
state matrix As’s are Schur stable, while, for the cases of K+(O,k ) and “(0, k ) satisfy the following two conditions
large delay ( K > K O ) or packet dropout ( K = N ) , the
Ax’s are not Schur stable. Therefore, in this paper it is
a
I ) infk>o K O k 2 1”A ,nA3-l,AL
-1nA. holds for some scalar
1185
V. DISTURBANCE ATTENUATION ANALYSIS where r ( j ) = z T [ j ] z [ j-
] $ d T [ j ] d [ j ] .Since V ( k i ) 5
In this section, we will study the disturbance attenuation pV(kty) holds on every switching point k;, we obtain by
induction that
property for the NCSs (4). Note that the Cz gain property
of discrete-time switched systems was studied in [I21 under "(k) 5 -
wNm(O,k)AZK-(O.L) A+2 K + ( O , k ) V (0)
the assumption that all subsystems were Schur stable. In this k--1
aillzllz I I o ~ z I I ~VZ
I I ~E, W", V i E Q (8) We sum (12) from k = 1 to k = +m to obtain
Here A,,(P,) and A,(P,) denotes the largest and smallest
eigenvalue of Pi respectively. There exist a constant scalar
/L 2 1 such that
4 - ( 1 - X Z ) V , ( z [ k ]-
-(1 - A 3 V , ( o [ k ] -
) z ~ [ k l z [ +rodT[kld[k]
k]
) r T [ k ] z [ k+?odT[k]d[k]
For a piecewise constant switching signal u[k]and any
]
Therefore, Cz gain 70is achieved for the switched system,
given integer k > 0, we let k1 < . . . < ki (i 2 1) denote the namely the NCS (4).
switching points of u [ k ]over the interval [0, k). Then, using For the case fi > 1, we multiply both sides of ( 1 1) by
the above difference inequalities, we obtain p-No(OJ-l) to get
~ k - 1 -iVe(o,j)(A*)Z(k-l-j) T '
OP 1314j1 (14)
7;X,=o w
@1 -N,(O.j)(A*)Z(k-l--j)dT[.
I 3ldijl
1186
Now, we choose a positive scalar X larger than 1 to consider VII. ACKNOWLEDGMENTS
the following average dwell time condition: for any positive The partial support of the National Science Foundation
integer j > 0, (NSF ECS99-12458 & CCROI-13131) is gratefully acknowl-
edged. The first author appreciates the support from Center
of Applied Mathematics Fellowship (2003-04), University of
Notre Dame.
Therefore f i - N ” ( o J ) > X-23 holds for any j > 0, where
X = P (zTc)-l. Then, from (14) we obtain VIII. REFERENCES
[I] P. Bauer, C. Lorand, M. Sichitiu, and K. Premaratne,
“Total delay compensation in LAN control systems and
implications for scheduling,” in Pmceedings of 2001
American Control Conference, pp. 4300-4305, Arlington,
Similarly, we sum both sides of the above inequality from VA, 2001.
k = 1 to k = +m to get 121 M. S. Branicky, “Multiple Lyapunov functions and other
analysis tools for switched and hydrid systems,” IEEE
Trnnsacrions on Auronmric Conrrol, vol. 43, no. 4, pp.
475-482, 1998.
[3] R. W. Bmkett, and D. Liberzon, “Quantized feedback
and thus stabilization of linear systems,” IEEE Transacrions on
Aurornaric Conrml, vol. 45, no. 7, pp. 1279-1289, 2000.
141 D. E Delchamps, “Stabilizing a linear system with quan-
tized state feedback:’ IEEE Transactions on Autosuric
3 4 3=1 Control, vol. 35, no. 8, pp. 916-924, 1990.
holds for any d[k] E &[O, +m). Following the notation in [5] R. A. Decarlo, M. S. Branicky, S. Pettenson, and B.
Lennmson. “Penoectives and results on the stabilitv
[IZ],we say that a weighted LZ gain ’yo is achieved. In and stabilizability bf hybrid systems:’ Proccedings of &
summary, we prove the following theorem. IEEE, vol. 88, no. 7. !m. 1069-1082. 2000.
77ieorern 2: The NCS (4) achieves a weighted Lz gain [6] N.Elia. and S. K. Miiter, ”Stabilization of linear systems
‘yo if the K + ( k l , k z ) satisfies (IO) and N,(O,k) satisfy the with limited information,” IEEE Transactions on Aulo-
condition of (15). malic Conrml, vol. 46, no. 9, pp. 1384-1400, 2001.
[7] P. Gahinet, and P. Apkarian, “A linear matrix inequality
Remark 4; Similarly, the condition (10) resuicts the num- approach to H , control,” Intemarional Journal of Robust
ber of the packet dropout and large delayed packet, while ond Nonlinear Conrrol, vol. 4, pp. 421-448, 1994.
the condition ( I 5 ) restricts the happening frequency of them. [8] I. P. Hespanha, and A. S. Morse, “Stability of Switched
Both of the conditions are given in the sense of average over Systems with Average Dwell-time,’’ in Proceedings of {he
38th Conference on Decision and Control, pp. 2655-2660,
time.
Dec. 1999.
VI. CONCLUDING REMARKS [9] H. Ishii, and B. A. Francis, Limited Dura Rare in Control
Sjsrerns wirh Network, Lecture Notes in Control and
In this paper, we modeled a class of NCSs under uncertain Information Sciences. vol. 275, Springer, Berlin, 2002.
access delay and packet dropout as discrete-time linear [IO] D. Liberzon, and A. S. Morse, “Basic problems in stability
switched systems. The stability and disturbance attenuation and design of switched systems,” IEEE Conrml Sysrerns
issues for such NCSs were studied in the framework of Magazine, vol. 19, no. 15, pp. 59-70. 1999.
[Ill G. Zhai, B. Hu, K. Yasuda, and A. N. Michel, “Distur-
switched systems. The strength of this approach comes bance attenuation properties of time-controlled switched
from the solid theoretic results existing in the literature of systems:’ Journal of rhe Franklin Insrirure, vol. 338, pp.
switchedhybrid systems. It was shown that the exponential 765-779, 2001.
stability and disturbance attenuation level might be preserved [I21 G. Zhai, B. Hu, K. Yasuda, and A. N.Michel, “Qualitative
for the NCSs under certain bounds on the amount and rate analvsis of discrete-time switched svstems:’ in Pmcecd-
ings of 2W2 American Conrml Confemmce, vol. 3, pp.
of the dropped and large delayed packets. Although we only 1880-1885. 2002.
consider state feedback control law here, the techniques and 1131 G. Zhai, ”Stability and H , Disturbance Attenuation
results developed here can be easily extended to the case of Analysis for LTI Control Systems with Controller Fail-
static output feedback control law. It should be pointed out ures,” in Pmceedings of rhe 41st Conference on Decision
that the conditions concerning the delay and packet dropout and Conrml. pp. 3869-3874, 2002.
rate for the preservation of the NCSs’ stability and ‘H” [I41 W. Zhang, M. S. Branicky, and S. M. Phillips, “Stability
disturbance attenuation properties were based on Lyapunov of networked control systems:’ IEEE Conrml Sysrems
theory. Therefore, the conditions are sufficient only and Magazine, vol. 21, no. I , pp. 84-99, 2001.
maybe conservative for some cases.
1187