IJSTE - International Journal of Science Technology & Engineering | Volume 3 | Issue 11 | May 2017

ISSN (online): 2349-784X

A Review on Extended Kalman Filtering

Approach to Nonlinear Time Delay Systems
Aruj Kumar Verma Rakesh Mandal
Department of Electronics & Telecommunication Engineering Department of Electronics & Telecommunication Engineering
Shri Shankaracharya Technical Campus, Bhilai Shri Shankaracharya Technical Campus, Bhilai

Khemlal Sinha
Department of Electronics & Telecommunication Engineering
Shri Shankaracharya Technical Campus, Bhilai

Abstract
The Kalman Filter are used to estimates system states from the sequential noise measurements of the outputs. Another hand, real
time system is often modeled with uncertainties and time-delay. Developing Kalman filter (KF) algorithms for this type of
systems are an important problem to obtain optimal state estimated by the utilizing the information on uncertainty and time-
delays. First designing of KF for nominal discrete-time systems are studied. Considering the co-variance error in the estimation
of the KF algorithms are derived which is further tested on a numerical example. Next, development of KF for nonlinear time
delay systems are considered. Similar to the previous design, considering covariance of estimation of the error, an Extended
Kalman Filter (EKF) are developed based on the structure and normal information of the plant and The performance of the
designed EKF are also compared with that of the KF designed for the same type of nominal system. It’s observed that the
performance of the RKF are better than the KF.
Keywords: Kalman Filter (KF), Extended Kalman filter (EKF), nonlinear time-delay systems, Real-time systems
________________________________________________________________________________________________________

I. PROPERTIES/ CHARACTERISTICS

Kalman Filter (KF) is a numerical method used to track a time-varying signal in the presence of noise. It is the problems
estimating the instantaneous states of linear system from measurement of the outputs. That is linear combination of states but
corrupted with white Gaussian noise. The resulting estimator are statically optimal with respect to the quadratic function of
estimation errors. From the mathematical point of view, the KF is a set of equations that provides an efficient recursive
computational solution of linear estimation problems. The filters are very powerful in several aspects. It can be suitably used to
either smoothening, estimating or predicting respectively the past, present and the future states. This may be achieved even when
precise characteristics of the modeled system are unknown, i.e., the case of uncertain time delay system. The KF is the extremely
effective and versatile procedure for combining noisy sensor output to estimate the states of the system with uncertain quantity.
When we applied to a physical system, the observer or filter will be under the influence of two noisy sources:
1) Process noise
2) Measurement noise
The estimate of the states specified by its conditional probability density function. The purpose of filter are computes the state
estimates while a optimal filter minimizes spread of estimation error probability density. A filters are propagating in the
conditional probability function from the one sampling instant to the next keep into the view of systems dynamic function and its
inputs, and its measurements and measurement error of statistics are the estimated. There are a generation of the mean value and
covariance value in finite time can be expressed by the following:
1) State estimate Propagation,
2) Covariance estimation Propagation,
3) Filter gain computation,
4) State estimate update,
5) Covariance estimate update.
We use KF to estimate the states x k   of discrete time controlled system. The systems are described by a linear stochastic
n

difference equation: -
x k 1  Ax k  Bw k (i)
y k  Cx k  v k (ii)
where, x k   is a system state, y k   is measured output, w k   is the process noise, v k  
n m q p
is the measurement
noise. In following v k and w k will be regarded as a zero mean, uncorrelated white noise sequence with covariance R k and Q k .

All rights reserved by www.ijste.org 147

 A Review on Extended Kalman Filtering Approach to Nonlinear Time Delay Systems
(IJSTE/ Volume 3 / Issue 11 / 030)

v k  N  0, Rk  (iii)

wk  N  0, Q k  . (iv)
n n
The matrix A   the difference Equation(i) are the dynamics matrix which relate to the state at time step k to the state at
n 1 mm
time step k  1 . The matrix B   is called noise matrix. The matrix C   in the measurement Equation (ii) relates to
the state measurement y k .When measurement error covariance R k approaches zero, the weighting of K f lets the actual

measurement y k be trusted more, while the predicted measurements C k xˆ k are trusted less. On other hand, the actual

measurement y k is trusted less as the a priori estimate error covariance Pk approaches to zero, and the predicted measurement

C k xˆ k is trusted more.
The KF algorithm can be seen as a form of feedback estimation. The set of KF equations can be separated into two group:
1) Time update equations
2) Measurement update equation
The time update equation projects the current state and the error covariance estimates forward to time to obtain priori
estimates for the next time step. The measurement update equations handle the feedback. In other words, it incorporates a new
measurement into the a priori estimate to obtain a corrected posterior estimate. Therefore, time updates equations are predicator
equations, and the measurement update equation is corrector equations. That is the KF is a predictor-corrector algorithm to
provide a recursive solution of the discrete time linear system shown in Fig. 1.1.

Fig. 1.1: Discrete Kalman Filter Cycle

The time and measurement updates are presented below:

Time update equations:

xˆ k 1  Axˆ k  Bu k (v)

Pk 1  APk Ak  Q k
T
(vi)
Measurements update equations:
1
K f

 Pk C kC P C
T
k k
 Y
k
 Rk  (vii)

 xˆ  K  y  C xˆ k 

xˆ k k f k k
(viii)

 I  K C P

Pk f k k
(ix)
The time update equation project to state estimate and covariance from time step k to k  1 . To compute the Kalman gain
(KG) K f are the first job in the measurement update equations. Then y k are obtained by actual measurement of system.
Incorporating actual measurement and estimated one in equation (vii), we generate a posterior estimate and last steps are to
compute a posterior error covariance. This are the recursive operation of the KF. A complete picture of the operation of KF is

All rights reserved by www.ijste.org 148

 A Review on Extended Kalman Filtering Approach to Nonlinear Time Delay Systems
(IJSTE/ Volume 3 / Issue 11 / 030)

shown in Fig. 1.2, after that each time and measurement update pair, the recursive algorithms are repeated with the previous, a
posterior estimate to predict the new priori estimates. This recursive nature is the biggest advantage of the KF.
This are gives practical implementation of Kalman filter much easier and relevant. In contrast the KF only uses the
immediately previous data to predict the current states. The standard KF algorithm is shown in Fig. 1.2.

Fig. 1.2: Kalman Filter Algorithm

II. KALMAN FILTER DESIGN

A state space model is a mathematical model of a process, where process state x is representing numerical vector. A state space
model actually consists of two separate model:
the process model:
which describes how the state propagates in time based on external influence, such as input and noise, and measurement y are
taken from the process, typically simulating noisy and inaccurate measurements. A linear system given equation (i) and (ii) and
our objectives are to design the KF in the form of equation number (x) and determine the gain of the matrix which minimizes the
mean square of the error.
xˆ k 1  A f xˆ k  k f y k (x)
Some assumptions of discrete-time KF are:
1) The state dynamics are linear and time-invariant
2) The measurement equations are linear and time-invariant
3) The noise statistics are stationary.
j j P 1
The matrix A f   in the Equation(x), The matrix K f   in the Equation (x) are the KG matrix which relates to
estimated state xˆ k . Note that the matrix A f & K f both are time varying matrices to be determined in order that the estimation
error ek  x k  xˆ k are guaranteed to smaller than a certain bound for all uncertainty matrices.

III. CONCLUSION

KF is a very good tool to estimate the states of a system under the noisy output and its measurements. In this chapter, a
formulation has been presented for the design of KF for linear systems without considering the time-delay and some basic ideas
on KF. For the estimation, prediction and reduction technique of the error the KF performances are satisfactory. The KF is the

All rights reserved by www.ijste.org 149

 A Review on Extended Kalman Filtering Approach to Nonlinear Time Delay Systems
(IJSTE/ Volume 3 / Issue 11 / 030)

most widely used filter and its gives better performance. It’s a very simple and intuitive concept of the good computational
efficiency. But when uncertainty and time delay is included then the performance of KF may degraded, so EKF is considered
which is robust against large parameter uncertainty and time delay. In coming chapter, a performance comparison between EKF
and KF is obtained for nonlinear time delay system

REFERENCE
[1] M. S. Mahmoud, L. Xie, and Y. C. Soh, “Robust Kalman Filtering for Discrete State-Delay Systems”, IEE Proceeding on Control Theory Appllation,
Volume 147, pp-613–618, (2000).
[2] X. Zhu, Y. C. Soh, L. Xie, “Robust Kalman Filtering for Discrete Time-Delay Systems”, Circuits Systems Signal Processing, Volume 21, pp- 319-335,
(2000).
[3] X. Zhu, Y. C. Soh, L. Xie, “Design and Analysis of Discrete-Time Robust Kalman filter”, Automatica, Volume 38, pp-1069-1077, (2002).
[4] X. Zhu, X.Yin, “Robust Kalman filtering for Uncertain discrete markovin jump system”, Second IEEE Confrrence on Industrial Electronics and
Apllications, pp- 2489-2493 (2007).
[5] Z.Wang, J.Lam and X.Liu, “Robust Kalman filtering for Discrete-Timr Markovian Jump Delay Sustem”, IEEE Signal Processing Letters, Volume 11, pp-
659-662 (2004).
[6] X.Lu, H.Zhang and W.Wang, “Kalman Filtering for Multiple Time Delay System”, Automatica, Volume 41, pp- 1455-1461 (2005).
[7] C.V.Hollot, V.Mishra, D.Towsley, and W.Gong “Analysis and Design of Controllers for AQM Routers Supporting TCP Flows”, IEEE Transactions on
Automatic Control, Volume 47, pp- 954-959, (2002).
[8] W.M.Wang, S.M.Guo, and L.S.Shieh, “Discretizatuin of cascaded continuous- time controllers for state and input delayed systems”, International Journal
of System Science, Volume 31, pp- 287-296, (2000).
[9] V.Mishra, W.Gong, and D.Towsley, “Fluid based analysis of a network of AQM routers supporting TCP flows with an application to RED”, Proceeding
ACM/SIGCOMM, (2000).
[10] I. R. Petersen, A. V. Savkin, “Robust Kalman Filtering for Signaling and System with Large Uuncertanities”, Birkhauser Boston, New York, (1999).
[11] P.S.Maybeck, “Stochastic models Estimation and Contorl”, Academic Press, New York, (1979).
[12] H. Gorecki, “Analysis and Synthesis of time-delay system”, Willey, New York, (1989).
[13] D.E.Catlin, “Estimation Control and the Discrete Kalman Filter”, Springer, New York, (1989).
[14] D.J.Simon, “Optimal State Estimation”, Willy, New Jersey, (2006).

All rights reserved by www.ijste.org 150