Sequential Filtering for Linear Discrete-Time Systems with
Delayed Measurements
ZHAO Hongguo
1
,
1. School of Information Science and Technology, Taishan University, Taian Shandong 271021, P. R. China
E-mail: h
g zhao@126.com
Abstract: This paper investigates the problems of designing optimal sequential filter for discrete-time systems with measurement
time-delay. Two kinds of approaches are proposed to tackle such problems via the innovation analysis theory. The approaches,
to be presented, are to convert a delay optimal sequential filtering problems into a delay-free ones. Based on the minimum
mean square error estimation principle, the optimal sequential filter is designed by solving the recursive matrix equations. Two
examples are given to illustrate the effectiveness of the approaches presented.
Key Words: Sequential Filtering, Time-Delay, Innovation
1 Introduction
The optimal sequential filtering [1, 2] for linear delay-free
systems is one of classical estimation approaches used in
the fields of signal processing, communication, and wireless
sensor networks. Different from Kalman filtering, the central
idea in sequential filtering form is based on the sequential
processing of observation signals, i.e. the observation val-
ues obtained are processed one at a time, or sequentially [1].
Therefore, the sequential filter has important applications in
many engineering fields such as communication and sensor
fusion, see, for instance [3–7].
However, many systems with time-delay can be found
in the process engineering, chemical industry, communica-
tion, etc. Accordingly, it is necessary to effectively handle
the time-delay in systems to improve the estimator and con-
troller performance. Hence, the filtering and control prob-
lems for time-delay systems have received much attention
in the literature, see, for example [8–12]. However, to the
best of the author knowledge, there do not exist the optimal
sequential filtering schemes for time-delay systems and the
contribution of this letter is to develop sequential filter for
linear discrete-time systems with time-delays in the obser-
vations.
In this paper, two kinds of effective approaches will be
presented to deal with the optimal sequential filtering prob-
lems by using innovation analysis theory [1]. In Section 3,
by using state augmentation approach, the optimal sequen-
tial filter is designed by solving l augmented Riccati matrix
equations. However, the approaches lead to much expensive
calculation burden. In Section 4, the sequential filter can be
given by using new approach proposed where the l Riccati
matrix difference equations with the same dimension as the
original systems are involved.
2 Problem Formulation and Preliminaries
The sequential filtering presented in this paper is devel-
oped for the following discrete stochastic systems with mea-
This work is supported by National Natural Science Foundation
(NNSF) of China under Grants 61273124 and 61074038, Doctoral Foun-
dation of Taishan University under Grant Y11-2-02, and A Project of Shan-
dong Province Higher Education Science and Technology Program under
Grant J12LN90.
surement delay:
x(k +1)=Φx(k)+Γu(k), x(0) = x
0
, (1)
y
(i)
(k)=H
i
x(k
h
)+v
(i)
(k),i=1, 2, ···,l, (2)
k
h
= k − h,
where x(k) ∈ R
n
is the state signal, y
(i)
(k) ∈ R
p
i
,i =
1, 2, ···,l, are the output signals. The u(k) ∈ R
r
and
v
(i)
(k) ∈ R
p
i
represent the system noise and the measure-
ment noises, respectively. The measurement delay h is a
known positive integer.
Assumption: The u(k), v
(i)
(k),i =1, 2, ···,l,
and the initial state x(0) are mutually uncorrelated
white noises with zero means and covariance matrices
as E[u(k)u
T
(j)] = Q
u
(k)δ
kj
, E[v
(i)
(k)v
T
(i)
(j)] =
Q
v
(i)
(k)δ
kj
, and E[x(0)x
T
(0)] = P
0
, respectively, where
δ
kj
is the Dirac Delta function, the superscript T indicates
the transpose of a matrix, and E[·] denotes the mathematical
expectation.
Setting
Y
(l)
(1)
(
s
)=
{
y
(1)
(
s
)
,
···
,
y
(l)
(
s
)
}
,h
≤
s
≤
k,
(3)
then, the problem in this paper can be stated as: Given obser-
vation sequence {Y
(l)
(1)
(h); ···; Y
(l)
(1)
(k)}, find a linear mini-
mum mean square error filter
ˇ
x(k|k) of the state signal x(k)
for k ≥ h.
It is easy to know that the optimal sequential filter can be
designed by
ˇ
x(k|k)=E[x(k)|Y
(l)
(1)
(h); ···; Y
(l)
(1)
(k)]. (4)
In view of the idea of sequential processing from [1], we
have
ˇ
x(k|k, i)=E[x(k)|Y
(l)
(1)
(h); ···; Y
(l)
(1)
(k − 1);
y
(1)
(k), ···, y
(i)
(k)]. (5)
It follows from (4) and (5) that
ˇ
x(k|k)=
ˇ
x(k|k, l). (6)
In the following, we aim to design the optimal estimator
ˇ
x(k|k, i),i=1, 2, ···,l, of the state signal x(k) for the
time-delay systems (1) and (2) based on the measurement
sequence {Y
(l)
(1)
(h); ···; Y
(l)
(1)
(k −1); y
(1)
(k), ···, y
(i)
(k)}.
Proceedings of the 34th Chinese Control Conference
Jul
28-30, 2015, Han
zhou, China
2384