IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, VOL. 55, NO. 7, JULY 2008 695
Optimal Filtering for Systems With
Multiple Packet Dropouts
Shuli Sun, Lihua Xie, Wendong Xiao, and Nan Xiao
Abstract—This paper is concerned with the optimal filtering
problem for discrete-time stochastic linear systems with mul-
tiple packet dropouts, where the number of consecutive packet
dropouts is limited by a known upper bound. Without resorting to
state augmentation, the system is converted to one with measure-
ment delays and a moving average (MV) colored measurement
noise. An unbiased optimal filter is developed in the linear
least-mean-square sense. Its solution depends on the recursion of a
Riccati equation and a Lyapunov equation. A numerical example
shows the effectiveness of the proposed filter.
Index Terms—Lyapunov equation, optimal linear filter, packet
dropouts, random measurement delays, Riccati equation.
I. INTRODUCTION
I
N recent years, the research on delayed systems has gained
lots of interest [1]–[3], particularly, with emphasis on net-
worked control systems. In networked systems, random time
delay and packet dropout are unavoidable in data transmission
through unreliable communication networks from sensors to
processing center (fusion center/decision maker) and from pro-
cessing center to end users (actuators). The data available in con-
trol and estimation may not be up-to-date due to stochastic de-
lays and/or packet dropouts. So estimation and control for net-
worked systems are very challenging [4].
In wireless networks, the systems with stochastic sensor de-
lays, packet dropouts and uncertain observations can be de-
scribed by a stochastic parameter system [5]–[15]. The optimal
estimation problem for systems with uncertain measurements is
investigated in [5], [6], where sensor data that are simply the
measurement noises at some samples are used for updating the
estimate, resulting in undesirable estimation performance. In
fact, previous measurements rather than noises should be used
in the absence of valid current sensor data for the case of un-
certain observations. The variance-constrained filtering has also
been considered in [7] for systems with uncertain measurements
and norm-bounded parameter uncertainties.
Manuscript received August 14, 2007; revised October 11, 2007. First pub-
lished April 16, 2008; last published July 16, 2008 (projected). This work was
supported in part by the Agency for Science, Technology and Research under
Grant SERC 052 101 0037, by the Natural Science Foundation of China under
Grant NSFC-60504034, and the by the Key Lab of Electronic Engineering, Hei-
longjiang University, Harbin, China. This paper was recommended by Associate
Editor A. Vicino.
S. Sun is with the School of Electrical and Electronic Engineering, Nanyang
Technological University, Singapore 639798, on leave from the Department
of Automation, Heilongjiang University, Harbin 150080, China (e-mail:
sunsl@hlju.edu.cn).
L. Xie and N. Xiao are with the School of Electrical and Electronic En-
gineering, Nanyang Technological University, Singapore 639798 (e-mail: el-
hxie@ntu.edu.sg).
W. Xiao is with the Media Processing Department, Institute for Infocomm
Research, Singapore (e-mail: wxiao@i2r.a-star.edu.sg).
Digital Object Identifier 10.1109/TCSII.2008.921576
Yaz et al. [8] discuss the least-mean-square filtering problem
for systems with one random sampling delay. The filters de-
rived, however, are not optimal since a nonwhite noise due to
augmentation is treated as a white noise. [9] gives a modifica-
tion on the minimum variance state estimator to accommodate
the effects of random delays in sensor data arrival at the con-
troller terminal. An extended Kalman filter is given for systems
with delayed measurements in [10] whereas [11] deal with the
recursive least-squares linear estimation for signals with random
delays by using a covariance information approach. Robust esti-
mation problems for systems with random delays and uncertain
measurements are also investigated in [12] and [13]. The op-
timal
filtering for systems with random sensor delays, mul-
tiple packet dropouts or uncertain observations is presented in
[14] where a unified stochastic parameter model is introduced.
In Sinopoli et al. [15] the stability of the Kalman filter in rela-
tion to the data arrival rate is investigated. It is shown that there
exists a critical data arrival rate for an unstable system so that
the mean filtering error covariance will be bounded for any ini-
tial condition.
So far, the research on the estimation problem for systems
with packet dropouts is mainly focused on single packet dropout
or multiple packet dropouts with possibly infinite number of
consecutive packet dropouts [14]. In the latter case, it may lead
to conservative results since the number of consecutive packet
dropouts cannot be infinite but bounded by a finite number in
practice. In this paper, we study the linear optimal filtering
problem for systems with multiple packet dropouts, where the
number of consecutive packet dropouts is limited with a known
upper bound. A mathematical model is first developed to model
this kind of packet losses. With the model, the system is then
converted to one with measurement delays and an moving av-
erage (MV) colored measurement noise. An unbiased optimal
filter is designed in the linear least-mean-square sense via com-
pletion of squares. Simulations show the better performance of
the proposed filter as compared to that of [14] in the case of
finite number of consecutive packet dropouts.
II. P
ROBLEM FORMULATION
Consider the discrete time-invariant linear stochastic system
with multiple packet dropouts
(1a)
(1b)
(1c)
where
is the state, is the sensor output
which is transmitted to estimator via unreliable networks,
is the output received by the estimator, , and
1549-7747/$25.00 © 2008 IEEE
Authorized licensed use limited to: IEEE Xplore. Downloaded on January 5, 2009 at 22:06 from IEEE Xplore. Restrictions apply.