Aerospace Science and Technology 15 (2011) 125–128

Contents lists available at ScienceDirect

Aerospace Science and Technology

www.elsevier.com/locate/aescte

Improved self-alignment scheme for SINS using augmented measurement

Arunasish Acharya, Smita Sadhu ∗ , T.K. Ghoshal
Jadavpur University, Kolkata-700032, India

a r t i c l e i n f o a b s t r a c t

Article history: An improved self-alignment scheme for strapdown Inertial Navigation System (SINS) based on the
Received 20 October 2009 extended Kalman filter (EKF) and augmented measurements is proposed. Monte Carlo simulations show
Received in revised form 28 April 2010 that the method provides good azimuth alignment, improved rate of convergence of azimuth attitude
Accepted 17 June 2010
error and reduction in sensitivity to gyro biases.
Available online 30 June 2010
© 2010 Elsevier Masson SAS. All rights reserved.
Keywords:
SINS
Self-alignment
Measurement augmentation
Nonlinear filtering
EKF

1. Introduction been proposed in [1]. As the linear acceleration with a stationary

platform can be rather noisy, the present contribution uses linear
This paper addresses self-alignment of a strapdown Inertial velocity and angular rate measurement as is also done in [4,3]. To
Navigation System (INS) [9,7] in near stationary condition. For a overcome the observability problem, [4,3] applied an ad-hoc first
strapdown INS, alignment requires accurate estimation of the INS order digital filter to the gyro measurement signal for filtering the
axis with respect to the navigation frame (assumed local level estimated tilt axes alignment errors. Such filtered output is alge-
NED). Any uncorrected initial misalignment would generate posi- braically combined with the Kalman filter (KF) update equations
tion error. to obtain a more accurate estimate of the azimuth misalignment
A standard method of self-alignment utilizes the horizontal and a faster convergence.
velocity outputs of the INS as measurements and then applies
However, the method used by [4,3] suffers from some disadvan-
state estimation techniques to obtain alignment information [5,9].
tages. Firstly, the method used in these papers relies on the use of
The method works well when the instrument biases are negligi-
an ad-hoc digital filter. Secondly, even with the filter, the estima-
bly small. With non-negligible instrument bias, the standard self-
tion in the azimuth axis is rather sensitive to gyro bias and gyro
alignment method in stationary base for strapdown INS (SINS) fails
noise. Further, [4] and [3] provide results indicating the errors for
because of the so-named observability problem [6,4,2]. The observ-
ability problem implies that from the observation of the horizontal single runs. While the results for single runs could provide an in-
velocities, the instrument biases cannot be re-constructed, which dicative value, a better measurement of filter performance would
results in error in the self-alignment. For small misalignments, of- be obtained by plotting the root mean square (RMS) errors across
ten linearized models are used and standard tests for observability several Monte Carlo (MC) runs. In the present work, the perfor-
can be applied to verify the lack of observability. The poor ob- mance of the ad-hoc digital filters proposed in [4] and [3] has been
servability primarily affects the azimuth alignment (see Fig. 10.4 characterized with MC studies using the data given in [4].
in [9]). Unlike several publications, e.g. [5,7], which ignore the The present work proposes an alternative possibility of using
case of instrument bias and thereby avoid the observability prob- the east gyro outputs from the Inertial Measurement Unit (IMU),
lem, this contribution focuses on this particularly difficult case. along with the velocity outputs in a nonlinear state estimation
To deal with the observability problem, additional measurements framework using extended Kalman filter (EKF). It is demonstrated
have to be used. Linear acceleration and angular rate measure- that improvement in both convergence rate and residual alignment
ment with linear approximations in an adaptive framework has is possible using the proposed method compared to the ad-hoc
digital filter implementation in [4] and [3].
The rest of this paper is organized as follows. The next section
* Corresponding author.
presents the dynamic model. This is followed by the formulation
E-mail addresses: [email protected] (A. Acharya), [email protected]
(S. Sadhu), [email protected] (T.K. Ghoshal). of the proposed extended Kalman filter (EKF) and augmented mea-

1270-9638/$ – see front matter © 2010 Elsevier Masson SAS. All rights reserved.
doi:10.1016/j.ast.2010.06.007
126 A. Acharya et al. / Aerospace Science and Technology 15 (2011) 125–128

Table 1
Parameters.

Local latitude L 45◦

Earth radius R e 6 378 100 m
g 9.81 m/s2
Initial values of velocities 0
Initial attitude misalignment angles of three gyros 1◦ = 17.452 mrad.
Constant bias of three gyros 0.02 ◦ /h
Random bias of three gyros 0.01 ◦ /h
Accelerometer constant bias 100 μg
Accelerometer random bias 50 μg
Sampling time T 1s
Process noise and measurement noise assumed to be Gaussian white
Process noise covariance matrix Q(k) diag((50 μg/Hz)2 (50 μg/Hz)2 (0.01 ◦ /h)2 (/Hz)2 (0.01 ◦ /h)2 (/Hz)2
(0.01 ◦ /h)2 (/Hz)2 0 0 0 0 0)
Measurement noise covariance matrix (considering velocity measurements only) R(k) diag((0.1 m/s)2 (0.1 m/s)2 )
Initial error covariance matrix P(0) diag((0.1 m/s)2 (0.1 m/s)2 (1 ◦ )2 (1 ◦ )2 (1 ◦ )2 (100 μg)2 (100 μg)2
(0.02 ◦ /h)2 (0.02 ◦ /h)2 (0.02 ◦ /h)2 )

surement based scheme. Finally, the performance of the proposed where ε Eb is constant bias (drift rate) of east gyro in body frame.
filter is analyzed and compared with earlier schemes [4,3]. Here the gyro output is represented without considering the
measurement noise. The NED coordinate system quantities like bi-
2. Dynamic model and simulation parameters ases can be transformed into body coordinate system using ap-
propriate transformation matrix as in [4]. It may be noted that
2.1. The dynamic model defining bias in the navigation (NED) axis makes the state equation
linear but nonlinearity appears in the gyro measurement equa-
The model for the strapdown INS is in the north–east–down tions.
(NED) coordinate system, which is similar to that used in [4,5].
The state equations can be written in the form 2.3. Parameters used for simulation
ẋ = Ax + w (1)
The parameters used for simulation are summarized in Table 1.
where the state vector As the objective of this paper is to provide results of comparison
x = [ VN VE φN φE φD ∇N ∇E εN ε E ε D ]T , primarily with the scheme proposed in [4], an attempt has been
made to maintain compatibility with [4] to the maximum extent
with V N , V E as the north and east velocity errors, Φ N , Φ E , Φ D are possible. However, the units of the process noise covariance ma-
misalignment angles along north, east, down gyros respectively. trix Q(k) provided in [4] are wrong. Hence the corrected units of
∇N , ∇ E are north and east accelerometer biases, and ε N , ε E , ε D Q(k) have been used while keeping the values the same as those
are constant gyro bias (drift rate) of north, east, down gyros re- in [4].
spectively, both sets being in NED coordinate system. When the outputs from gyros are also considered as measure-
A discrete time equivalent of the truth model is given by ments, the level of gyro noise of 0.01 ◦ /h is assigned.
It may be noted that the values of instrument biases and mea-
x(k + 1) = Φ x(k) + w (k), (2)
surement noises quoted above are considered nominal values. Dur-
y (k) = C x(k) + v (k). (3) ing the simulation experiments and sensitivity studies, effect of
higher (up to a decade) values of biases and measurement noises
The constant state matrix is computed numerically using the
have also been explored.
relation Φ = e AT , at the operating point [8], where T is sampling
time.
The measurements from the INS (as well as from most IMU’s) 3. Alignment using EKF and measurement augmentation
are generally available at discrete sample time and the associated
noise would be assumed to be Gaussian sequences with known The method proposed in this paper involves using the attitude
covariance. The respective covariances of the process noise and misalignment angle along the east gyro as an additional measure-
measurement noise are denoted as Q(k) and R(k) respectively. ment. The augmentation in the measurement results in a nonlinear
equation and hence an EKF has been used as the alignment fil-
2.2. East gyro output equation ter.
As the east gyro output is measured in the body frame, the drift
The east gyro measurement equation is being mentioned in rate needs to be transformed into the body frame. So, in the state
this subsection as it would be used as an additional measurement equations εx , ε y , εz are to be considered as constant drift errors
for the method proposed in this paper. Though the state equation in body frame. Since attitude misalignment angles are very small,
quantities are in (true) navigation frame, if gyro outputs (which are it may be assumed that the sine and cosine of these angles are
in the body axis) are to be used as additional measurements, as is equal to the attitude misalignment angles and zero respectively.
done in the proposed scheme, these quantities must be expressed Then it can be proved that constant bias measured in the body
in the NED navigation frame. This is done using the appropriate frame remains almost the same as that in the NED frame. Hence,
transformation matrix as follows: the state equation can be assumed to remain unchanged and is
In the body frame, the output of the east gyro is given by (2).
  The modified measurement equation can be written as
gyro_E = Ω cos( L ) sin(φ N ) sin(φ E ) cos(φ D ) − cos(φ N ) sin(φ D )
 
− Ω sin( L ) sin(φ N ) cos(φ E ) + ε Eb (4) z = h x(k) + v (k) (5)
 A. Acharya et al. / Aerospace Science and Technology 15 (2011) 125–128 127

Table 2
Comparison of filter performances.

Condition Error (arc min) at 60 s Final error (arc min) at 200 s

DF1 DF2 EKF DF1 DF2 EKF
Nominal condition 5.55 59.35 5.3 6 27.09 4.73
Nominal gyro measurement, noise 18.2 59.38 14.6 18.96 27.09 12.285
and gyro bias 2.5 times higher
Nominal bias and gyro measurement, 22.2 59.36 6.38 9.65 29.6 5
noise covariance 100 times higher

where h(x(k)) = [h1 h2 h3 ]T .

   
h1 1 0 0 0 0 0 0
= C x(k), where C =
h2 0 1 0 0 0 0 0
and h3 = gyro_E.
The Jacobian of h can be written as
   
H1
H1
H= H2 , where = C,
H2
H3
and
⎡ 0
⎤T
⎢ 0 ⎥
⎢ Ω sin(φ ) sin(φ ) + Ω cos(φ ) sin(φ ) cos(φ ) + Ω cos(φ ) cos(φ ) ⎥
⎢ N N D N N E D D N E ⎥
⎢ ⎥
H3 = ⎢ Ω N sin(φ N ) cos(φ E ) cos(φ D ) − Ω D sin(φ N ) sin(φ E ) ⎥
⎢ ⎥
⎢ Ω N (− cos(φ N ) cos(φ D ) − sin(φ N ) sin(φ E ) sin(φ D )) ⎥
⎣ ⎦
0
0
Fig. 1. Comparison of RMSE of DF1 (mentioned as DF in this figure legend) and EKF
(6) for nominal condition.

where Ω N = Ω cos( L ) and Ω D = −Ω sin( L ).

The filter is initialized with all states set to zero.

4. Comparison of alignment performance

An independent study was carried out to validate the results

in [4] and [3] and to provide a basis for comparison of the pro-
posed method. The Kalman filter runs were carried out to also
investigate the effect of lack of observability. It is interesting to
note that though the system is unobservable (though detectable),
steady state solution, though with large RMS errors exists. It
is the residual alignment error and not the slow convergence
which will prevent the adoption of a simple KF scheme for self-
alignment.
As the estimates would vary depending on the realization of
the process and measurement noise, it is essential that RMS values
of the errors in estimating the misalignment are also determined.
The RMS error (RMSE) and the approximate time to reach the
steady state estimate are the two metrics to define the quality of Fig. 2. Azimuth RMSE for different gyro biases using gyro output with DF1.
alignment. The estimates of the unknown instrument biases have
diagnostic value but are not considered as a part of the primary From Table 2, it may be seen that the performance of the filter
alignment objective. given in [3] (with the data given in Table 1) is unsatisfactory both
The comparative filter performances (RMS error in arc minutes w.r.t. the convergence rate and the final error. Hence, further com-
at both 60 s and 200 s) after 200 Monte Carlo runs have been parisons are carried out only between the proposed method and
summarized in Table 2. The column headings DF1, DF2, EKF in- that given in [4].
dicate the results using the methods given in [4,3], and the result For the proposed method and DF1 (proposed in [4]), the RMS
obtained from the proposed method respectively. The nominal con- errors for the estimated angular misalignment for the north and
dition has been considered as a gyro bias of 0.02 ◦ /h and a gyro east axes were found to converge within 2 to 3 samples and is
noise covariance of (0.01 ◦ /h)2 . As shown in the table, at nominal consistent with the behavior presented in the earlier work [4].
condition the performance of the proposed method is very simi- In contrast, the RMSE of azimuth misalignment angle (about the
lar to that of DF1 whereas the performance of DF2 with the given down axis) settles at a much higher value and hence, only the az-
data is unsatisfactory. It may also be noted from the same table imuth error is compared.
that the proposed method is less sensitive to gyro bias and gyro The RMSE of azimuth misalignment angle for the method in [4]
measurement noise covariance. (hereafter referred to as DF1) and the proposed method (here-
128 A. Acharya et al. / Aerospace Science and Technology 15 (2011) 125–128

after referred to as EKF) is shown in Fig. 1. Though from Fig. 1,

it might appear that there is only marginal improvement in filter
performance in the nominal case, a sensitivity study shows that
the method in [4] is much more sensitive to gyro bias and noise
covariance than the proposed method.
The alignment performance for DF1 in azimuth (RMSE with 200
MC runs) at different values of gyro bias (but nominal noise covari-
ance) is shown in Fig. 2 and that for different gyro (measurement)
noise covariance is shown in Fig. 3.
The alignment performance for the proposed filter in azimuth
(RMSE with 200 MC runs) at different values of gyro bias (but
nominal noise covariance) is shown in Fig. 4. It may be noted that
though the final error is affected by the gyro bias, the values of fi-
nal error are much lower than DF1 [4] and that the convergence is
reasonably fast, establishing the superiority of this method.
Fig. 3. Azimuth RMSE for different gyro noise covariance using gyro output with The alignment performance in azimuth for different gyro (mea-
DF1. surement) noise covariance is shown in Fig. 5.

5. Conclusion

A new method has been proposed for self-alignment of strap-

down INS using augmented measurement from the east gyro and
an EKF. The performance of the proposed scheme is better com-
pared to existing schemes [4,3]. The performance has been charac-
terized in terms of convergence time and final RMSE across several
MC runs. Sensitivity studies show that the proposed method has
lesser convergence time and lower final RMSE even in the pres-
ence of gyro biases and noises.

References

[1] Hyo-Sung Ahn, Chang-Hee Won, Fast alignment using rotation vector and adap-
Fig. 4. Azimuth RMS error of down gyro for different gyro biases for EKF.
tive Kalman filter, IEEE Transactions on Aerospace and Electronic Systems 42 (1)
(Jan. 2006) 70–83.
[2] A. Bose, S. Puri, P. Banerjee, Modern Inertial Sensors and Systems, Prentice Hall
of India, New Delhi, 2008.
[3] Z. Chuanbin, T. Weifeng, J. Zhihua, A novel method improving the alignment
accuracy of a strapdown inertial navigation system on a stationary base, Mea-
surement Science and Technology 15 (Apr. 2004) 765–769.
[4] J. Fang, D. Wan, A fast initial alignment method for strapdown inertial naviga-
tion system on stationary base, IEEE Transactions on Aerospace and Electronic
Systems 32 (4) (Oct. 1996) 1501–1504.
[5] Yan-Ling Hao, Ming-Hui Chen, Liang-Jun Li, Bo Xu, Comparison of robust H ∞
filter and Kalman filter for initial, alignment of inertial navigation system, Jour-
nal of Marine Science and Application 7 (2008) 116–121.
[6] Y. Jiang, Y. Lin, Error estimation of INS ground alignment through observabil-
ity analysis, IEEE Transactions on Aerospace and Electronic Systems 28 (1) (Jan.
1992) 92–96.
[7] P.G. Savage, Introduction to Strapdown Inertial Navigation Systems, Strapdown
Associates, Maple Plain, MN, 1996.
[8] D. Simon, Optimal State Estimation, Kalman, H ∞ and Nonlinear Approaches,
Wiley Interscience, John Wiley & Sons, Inc., New Jersey, 2006.
[9] D.H. Titterton, J.L. Weston, Strapdown Inertial Navigation Technology, 2nd edi-
Fig. 5. Azimuth RMS error of down gyro for different gyro noise covariances for EKF. tion, The Institution of Electrical Engineers, 2004.