with multiple sensors and random measurement delays is
given, where a model to describe multiple sampling
delays is also presented. The optimal H
2
filtering problems
associated respectively with possible delay of one sam-
pling period, uncertain observations and multiple packet
dropouts are studied under a unified framework in [19].
On the other hand, the optimal and steady-state
estimators in linear minimum variance sense are devel-
oped for systems with multiple packet dropouts in [20].
However, it may lead to conservative results since the
number of consecutive packet dropouts cannot be infinite
but bounded by a finite number in practice. A novel model
to describe the case of the finite packet dropouts is
developed and an optimal filter like Kalman filter is
designed in [21]. However, the random delays are not
considered in [20,21]. The mean square stochastic stability
for some kind of discrete and continuous systems with
stochastic delays or packet dropouts have also been
analyzed in [22–24]. So far, estimation problems with
bounded random measurement delays and packet drop-
outs are seldom reported. Furthermore, they are challen-
ging problems in networked control systems.
Different from the recent works [20,21] where only the
packet dropouts are treated, in this paper we investigate
the estimation problem for the systems with bounded
random measurement delays and packet dropouts, which
are described by some binary distributed random vari-
ables whose probabilities are only known. The original
system is transferred to a new system with the random
delayed measurements and moving average (MA) colored
measurement noise. The optimal filter, predictor and
smoother are presented based on the innovation analysis
approach. They are optimal in the linear minimum
variance sense. The designed estimators only depend on
the probabilities of the delay and packet dropout at each
instant but do not need to know if a measurement is
delayed or received at a particular instant. Furthermore,
they do not require the measurements to be time-
stamped. For the system with unbounded random delays
and packet dropouts, the proposed algorithm gives a
suboptimal estimate by taking a sufficiently large bound.
The rest of this paper is organized as follows. Problem
formulation is given in Section 2. The optimal estimators
are developed based on the innovation analysis approach
in Section 3. In Section 4, a simulation is given. At last, the
conclusions are drawn in Section 5.
2. Problem formulation
Consider the discrete time-varying linear stochastic
system
xðt þ 1Þ¼
F
ðtÞxðtÞþ
G
ðtÞwðtÞ (1a)
zðtÞ¼HðtÞxðtÞþvðtÞ (1b)
where xðtÞ2R
n
is the state, zðtÞ2R
m
is the output, wðtÞ2
R
r
and vðtÞ2R
m
are white noises and
F
ðtÞ;
G
ðtÞ; HðtÞ are
time-varying matrices with suitable dimensions.
In the networked system, the sensor measures the
output of the system at every time and transmits the
measurement to a data processing center (the estimator).
Delays and packet dropouts are unavoidable by the
unreliable network communication. To reduce the effect
of packet dropouts without overloading the network
traffic too much, each sensor measurement is transmitted
for several times consecutively, for example N+1 times.
We assume that the largest delay and the number of
consecutive packet dropouts in data transmission are not
more than N+1 and there is a packet arriving at the
estimator at each time. Here the following model for the
measurement received by the estimator is adopted:
yðtÞ¼
x
0
ðtÞzðtÞþð1
x
0
ðtÞÞ
x
1
ðtÞzðt 1Þ
þþð1
x
0
ðtÞÞð1
x
1
ðtÞÞð1
x
N1
ðtÞÞzðt NÞ,
NX1 (1c)
where yðtÞ2R
m
is the measurement received by the
estimator,
x
i
ðtÞ,0pipN 1 are mutually independent
scalar binary distributed random variables with the known
distributions Probf
x
i
ðtÞ¼1g¼
a
i
ðtÞ and Probf
x
i
ðtÞ¼0g¼
1
a
i
ðtÞ where the symbol ‘‘Prob’’ denotes probability and
are uncorrelated with other random variables. In the
following, we explain the model (1c) for N ¼ 2. From (1c),
we can see that at t time zðtÞ is received if
x
0
ðtÞ¼1, i.e.,
yðtÞ¼zðtÞ with the probability
a
0
ðtÞ, zðt 1Þ is received if
x
0
ðtÞ¼0and
x
1
ðtÞ¼1, i.e., yðtÞ¼zðt 1Þ with the prob-
ability ð1
a
0
ðtÞÞ
a
1
ðtÞ,andzðt 2Þ is received if
x
0
ðtÞ¼0
and
x
1
ðtÞ¼0, i.e., yðtÞ¼zðt 2Þ with the probability
ð1
a
0
ðtÞÞð1
a
1
ðtÞÞ. It is worth noting that the output
z(t)att time can be received on time by the estimator,
delayed or lost in transmission. The following Table 1 can
give the results:
From Table 1, we see that z(1), z(2), z(4), z(6), z(7), z(11)
and z(12) are received on time, z(3) and z (8) are delayed,
z(5), z(9) and z(10) are lost. Furthermore, z(2), z(6) and
z(8) are re-received since the output at each instant is
transmitted three times. We also see that the number of
the largest delay and consecutive packet dropouts is
N ¼ 2. So the model (1c) describes the possible bounded
random delays and packet dropouts.
Substituting (1b) into (1c), we can obtain the following
system equivalent to (1):
xðt þ 1Þ¼
F
ðtÞxðtÞþ
G
ðtÞwðtÞ (2)
yðtÞ¼
X
N
i¼0
a
i
ðtÞHðt iÞxðt iÞþ
X
N
i¼0
a
i
ðtÞvðt iÞ; NX1 (3)
where the new random variables a
0
ðtÞ¼
x
0
ðtÞ, a
i
ðtÞ¼
Q
i1
k¼0
ð1
x
k
ðtÞÞ
x
i
ðtÞ,0oioN and a
N
ðtÞ¼
Q
N1
k¼0
ð1
x
k
ðtÞÞ.
Models (2) and (3) are the systems with the stochastic
N-step measurement delays and the N-order MA colored
measurement noise. The work of this paper is carried out
based on the following assumptions.
ARTICLE IN PRESS
Table 1
Data transmission in network.
t 123456789101112
x
0
(t)
11010110001 1
x
1
(t)
10 010
y(t) z(1) z(2) z(2) z(4) z(3) z(6) z(7) z(6) z(8) z(8) z(11) z(12)
S. Sun / Signal Processing 89 (2009) 1457–14661458