Aerospace Science and Technology 16 (2012) 61–69
Contents lists available at ScienceDirect
Aerospace Science and Technology
www.elsevier.com/locate/aescte
Linearization error’s measure and its influence on the accuracy of MEKF based
attitude determination method
Yuanyuan Jiao
∗
, Haiyin Zhou, Jiongqi Wang, Jisheng Li
Department of Mathematics and System Science, National University of Defense Technology, Changsha, China
article info abstract
Article history:
Received 3 December 2010
Received in revised form 29 March 2011
Accepted 18 May 2011
Available online 26 May 2011
Keywords:
Attitude determination
Multiplicative Extended Kalman Filter
Linearization error
Measure
Multiplicative Extended Kalman Filter (MEKF) is one of the most widely used satellite attitude estimation
methods. However, the linearization error’s influence is an inherent limitation of this method. In this
paper, we aim to analyze this linearization error in the typical satellite attitude determination system
with star sensors and gyros. The formulation of linearization error is first derived and the curvature
metric is then employed to measure the linearization error. Additionally, we show the reason why
linearization error has influence on the performance of MEKF. Based on these analyses, we point out that
star sensors’ sampling frequency, initial estimated error and accuracy of gyro’s measurement model are
the factors that could enlarge the system model’s linearization error. They all affect the linearization error
and attitude determination accuracy by decreasing the predicted accuracy. More concretely, the influence
of star sensor’s sampling frequency is large, while initial estimated error and gyro’s measurement error
within a certain range have little influence on MEKF. Finally, combined with plenty of experiments,
validity of the above analyses is verified.
© 2011 Elsevier Masson SAS. All rights reserved.
1. Introduction
Satellite attitude determination plays an important role in sup-
porting the ability of precise pointing for satellite control and other
performance. Satellite attitude determination system with star sen-
sors and gyros has become the typical high accuracy attitude de-
termination system. Multiplicative Extended Kalman Filter (MEKF)
based attitude determination method has widely been used in
this system [7]. However, this approach needs to linearize system
model and the linearization error is not Gaussian. This would heav-
ily affect the performance of MEKF when this error is large [3].
A lot of researches have emerged to solve this problem of MEKF.
For example, Julier et al. have proposed Unscented Kalman Filter
(UKF) [2,4,9]. It employs unscented transform with determinate
sampling and does not need to linearize the system model. Be-
sides, some researchers have also proposed a new method, i.e.,
Robust Filter [10,12]. Assumed that the upper bounds of some
model errors, such as linearization error, are known, Robust Filter
aims to minimize the upper bound of the estimated error’s covari-
ance. Crassidis et al. have also presented a new approach, which
is named as Nonlinear Predictive Filter (NPF) [1]. It expands the
measurement equation up to its second order derivatives. Based on
the minimum model error principle, NPF estimates both the state
variable and the error in state equation simultaneously. For the
*
Corresponding author.
E-mail address: jyynudt@gmail.com (Y. Jiao).
inherent linearization problem of MEKF, Junkins et al. have investi-
gated satellite attitude state equation’s nonlinearity with different
attitude parameters [6], and given the conclusion that the state
equation, expressed by Modified Rodrigues Parameters (MRPs) and
quaternion, has lower nonlinearity.
Based on these previous works, considering the advantages that
quaternion representation is free from singularities and its deriva-
tive expression is convenient to use, we aim to analyze the lin-
earization error in widely used MEKF based attitude determination
system expressed by quaternion, especially for the linearization er-
ror in measurement equation. Since MEKF based attitude determi-
nation approach has been regarded as a standard technique in real
applications, our researches about linearization error’s formulation
and its influence will provide useful guidance for the investigation
about other EKF-like methods (such as Additive Extended Kalman
Filter (AEKF), etc.) with other attitude parameters (such as Euler
angles, Rodrigues parameters, etc.). More concretely, in this paper,
we first formulate and measure the system model’s linearization
error, and then analyze the influence of this linearization error.
Furthermore, we search the factors that could enlarge this influ-
ence, explore the relationship between measurement equation’s
linearization error and attitude determination accuracy. Addition-
ally, we try to determine under what condition the linearization
error does not heavily affect the performance of MEKF based atti-
tude determination approach.
The paper is structured as follows. Section 2 presents the pre-
liminary sketch of MEKF based attitude determination method. In
Section 3, we first formulate the system model’s linearization error
1270-9638/$ – see front matter © 2011 Elsevier Masson SAS. All rights reserved.
doi:10.1016/j.ast.2011.05.004