Chinese Journal of Electronics
Vol.20, No.1, Jan. 2011
Universal Delayed Kalman Filter with
Measurement Weighted Summation for the
Linear Time Invariant System
∗
GE Quanbo
1,2
, XU Tingliang
2
, FENG Xiaoliang
3
and WEN Chenglin
2
(1.State Key Laboratory of Industrial Control Technology, Zhejiang University, Hangzhou 310027, China)
(2.Institute of Information and Control, Hangzhou Dianzi University, Hangzhou 310018, China)
(3.College of Computer and Information, Hohai University, Nanjing 310098, China)
Abstract — This paper considers the design of univer-
sal delayed Kalman filter for the networked tracking sys-
tem with arbitrary random delay. Firstly, an equivalent
Weighted summation form of the conventional Kalman fil-
ter (WSFKF) is given to provide a novel frame to more
effectively solve the delayed filtering or Out-of-sequence
measurements (OOSMs) estimate. In nature, this form
makes perfectly use of the properties of offline parameters
computation for Kalman filter and weighted summation of
initial state estimate and the ordered measurements, which
are respectively from Linear time invariant (LTI) system
and Linear minimum mean square error (LMMSE) estima-
tor. Secondly, by combing a replacement with global mea-
surement prediction and a compensation operation based
on the innovation of delayed measurement and adaptive on-
line weighted coefficient matrix, a novel universal delayed
Kalman filter which is applicable to the arbitrary random
delay is designed under the WSFKF frame. Compared
with the current delayed filters or OOSMs update meth-
ods, the proposed delayed estimator has not only more con-
cise algorithm structure and better estimate accuracy but
also stronger application range. The example is demon-
strated to validate the proposed delayed estimator in this
paper.
Key words — Kalman filter, Linear time invariant sys-
tem, Linear minimum mean square error (LMMSE), Ran-
dom delay, Measurement summation.
I. Introduction
Networked Kalman filtering is one of popular topics in
state estimation and data fusion in recent years
[1,2]
. It is
well known that data delay is common and unavoidable when
the estimate center collects data from local sensors in wireless
sensor networks
[3]
. Thereby, many researchers are interested
in focusing on the design of Kalman filter with random de-
lay. Random delay induced by random network transmission
proto col generally makes the out-of-sequence phenomenon ap-
p ear, which means that the sampling order of sensor measure-
ments can be broken and is called as Out-of-sequence measure-
ments (OOSM) estimate
[4−7]
. For the OOSM estimate, some
interesting results have been given for different delay cases or
OOSM scenes
[8−15]
.
Most of the current delayed Kalman estimators all adopt
the conventional Kalman filtering with prediction and update
scheme in Ref.[16]. As a result, the algorithm structures of
the designed delayed Kalman filters are complex and lots of
manmade noise correlations will appear. In nature, the main
reason to result in the complex algorithm design is that predic-
tion error covariance matrix, gain matrix, and estimate error
covariance matrix must be computed online according to the
arrival case of the delayed measurements. It is inevitable to
make the design process of the delayed estimator suffer many
difficulties and the reduced application ability together with
the random character of delay. Thereby, it is necessary choice
to modify the conventional Kalman filtering formulas with re-
cursive prediction and measurement update scheme. In terms
of the further analysis, we find that Minimum mean square
error (MMSE) estimator for Linear time invariant (LTI) sys-
tem has two go od properties. Firstly, the three parameter
matrixes mentioned-above in the MMSE estimator can be
computed offline for LTI system, namely Offline computation
prop erty (OCP). Secondly, the Linear MMSE (LMMSE) esti-
mator is substantially linear weighted summation with initial
state estimate and the ordered measurements which is called
as Weighted summation property (WSP)
[17]
. Thus, both prop-
erties existing in LMMSE estimator can be used to overcome
the high complexity and difficult analytic expression of the
current delayed estimators under the conventional Kalman fil-
tering formulas.
On the basis of above-mentioned analysis, this paper re-
searches the networked Kalman filtering with arbitrary ran-
dom delays. Firstly, the conventional Kalman filter with pre-
diction and update scheme is rewritten as a weighted summa-
tion form of the initial state estimate and the ordered mea-
surements with offline weighted coefficient matrixes, which is
∗
Manuscript Received Dec. 2009; Accepted Mar. 2010. This work is supported by the National Natural Science Foundation of
China (No.60934009, No.60804064, No.60572051), China Postdoctoral Science Foundation (Grant.20100471727), Project of Science and
Technology Department of Zhejiang Province (No.2009C34016), and Zhejiang Graduate Innovation Project (No.YK.2008061).