Improving Self-Alignment of Strapdown INS using
Measurement Augmentation
Arunasish Acharya
School of Education Technology
Jadavpur University
Kolkata, India.
arunasisha@yahoo.com
Smita Sadhu
Electrical Engineering Department
Jadavpur University
Kolkata, India.
ssadhuju@gmail.com
T.K.Ghoshal
Electrical Engineering Department
Jadavpur University
Kolkata, India.
tkghoshal@gmail.com
Abstract - An alternative solution to the velocity based
self-alignment of strapdown inertial navigation system is
proposed. It is well known that a simple Kalman filter
based solution to the above problem fails to provide
accurate azimuth alignment due to the inherent lack of
observability of the model in the presence of instrument
bias. Earlier researchers use external digital filters to
obtain improved estimation of selected states and substitute
these into the filter. The current paper demonstrates that a
simple augmentation of the output vector with the inertial
measurement unit signals and an extended Kalman filter
would yield similar or better alignment performance
compared to such ad hoc additional digital filters. The
proposed method improves the convergence rate of azimuth
attitude error even in the presence of gyro bias and makes
it relatively independent of the gyro noise. Results of
comparative performance of the two filters using Monte
Carlo simulation have been provided.
Keywords: Self-Alignment, Strapdown Inertial Navigation
System, Filtering, Extended Kalman filter.
1 Introduction
This paper addresses self-alignment (that is without external
aids) of a strapdown inertial navigation system (INS) [1,2]
in near stationary condition, where the knowledge of the
local gravity (vertical), latitude and earth’s spin rate are
utilized and additional instrumentation is not necessary.
Such initial self-alignment in stationary platform is often
used for tactical missiles and autonomous ground vehicles.
Use of “raw” IMU (inertial measurement unit, comprising
of rate gyro and accelerometer) outputs alone can provide
only a coarse alignment due to poor signal to noise ratio of
IMU’s on stationary platforms [3]. For “fine” alignment, it
is customary to use the velocity outputs from the INS (after
the due integrations are performed by the navigation
software) and from a previously aligned reference INS in a
so-named “transfer alignment” [1] configuration, employing
Kalman filter (KF) based estimators. For “self-alignment”
(that is without the aid of a reference IMU) in stationary
base and in presence of instrument biases, velocity based
scheme produces unacceptable results, especially in
correcting azimuth orientation errors due to observability
problem [4, 5, 6]. As to be demonstrated subsequently, the
poor observability manifests as inaccurate alignment (the
azimuth misalignment error being large and roughly
proportional to gyro bias) coupled with slow convergence in
azimuth. To overcome the observability problem, both [5]
and [6] use ad hoc first order digital filters on the gyro
measurement signal or on the estimated tilt axes alignment
errors respectively. Such filtered output is algebraically
combined and substituted in the KF update equations to
obtain a more accurate estimate of the azimuth
misalignment and a faster convergence.
In [6], in particular, the improved estimate of the azimuth
misalignment is obtained through approximate algebraic
relations involving north axis error and east axis error rate.
As the true value of these quantities are unknown, noisy
estimates of the azimuth error were filtered and then a KF
was applied.
The technique used in [5], on the other hand, utilizes filtered
gyro measurement to reconstruct estimates about the gyro
biases. These values are then substituted to the KF update
equations.
The present work reports an alternative possibility of using
the gyro outputs from the IMU, along with the velocity
outputs in a nonlinear state estimation framework, without
taking recourse to ad hoc digital filtering and demonstrates
that improvement in both convergence rate and residual
alignment are possible compared to a (linear) KF
implementation.
Performance of the proposed scheme has been compared
with that obtained in [5]. For a better and more objective
comparison, root mean square error (RMSE), its sensitivity
to change in instrument bias and measurement error
covariances and its convergence rates have been computed
for the proposed and earlier scheme [5].
12th International Conference on Information Fusion
Seattle, WA, USA, July 6-9, 2009
978-0-9824438-0-4 ©2009 ISIF 1783