Ver13

Transfer Alignment Technique for Shipboard Missile

Strapdown Inertial Navigation System Using an
Adaptive Kalman Filter
BOMMANA SIVA KUMAR1
1
SENIOR ENGINEER, HYDERABAD, India
1
[email protected]

ABSTRACT

Missile Guidance system needs accurate estimates from Inertial Navigation System (INS) for
guiding the vehicle towards the target. In this paper a target point, specified before launch, in a
battlefield scenario is considered for a landmark using missile Strap Down Inertial Navigation
System (SDINS) aided by Master INS (MINS) placed on a moving platform. Azimuth information
of the missile is one of the most critical navigation states for estimation on the moving platform
before launching the missile for precise impact.

An Adaptive Kalman Filter (AKF) based on the error state model is formulated. The 7-state AKF
with 4-measurement forms the core, where the filter gain of the innovation sequence
(measurements) is evaluated. This approach of adaptively computing the gain is tested in a
laboratory, on a van and in a ship trial, culminating in a successful guided missile launch. Mean
and the covariance of the measurement residuals were used in a unique way to compute adaptive
gain after the accumulation of initial samples.

A Master INS (with advanced Gyros) whose accuracy is much higher than the accuracy of the
missile’s SDINS is used for velocity matching algorithm before the launch with execution of an S-
maneuver for generation of accelerations towards observing the states more appropriately.
Estimated error states were used in a feedback mode to get near the true orientation of the
Missile’s slave INS. Error quaternions are used for this purpose in the feedback and the gains
were selected using offline matrix Riccati equation solution in a discrete domain as used in the
modern control system. The results were very encouraging with less than 5 arc minutes of error in
azimuth.

Keywords - Missile Guidance System, Inertial Navigation System, strap-down navigation

algorithm, Adaptive Kalman Filter, moving launch platforms, Velocity matching algorithm, S-
maneuver, Implicit Guidance Scheme, and Critical Error Probability.

NOMENCLATURE ~
Zk Observation vector at instant ' k '
~ Hk Observation ¿ instant ' k '
X (t ) state estimate at time t
F (t) state transition ¿ time t Vk white zero mean Gaussian
B Forcing function observation noise
U (t ) input at time t α ,β ,γ Euler angle rates
V N ,V W North , West velocities
W (t ) w hite zero mean guassian
state noise λ, Λ Latitude , Longitude
Φ (t) state transition ¿ time t ∅ , θ ,Ψ Euler angles

1
 Ver13

γk innovation sequence A computer-based recursive Adaptive Kalman Filter

Sγ error covariance ¿ (AKF) scheme is used for finding the missile
2
variance of IMU SDINS azimuth in an elegant way in this paper.
σ MIMU
2 However, due to the poor observability of azimuth
σX state covariance and the spread of noise in missile SDINS (with
^
R Residue higher drift), an S-maneuver of the ship is required
∆ qi element of quaternion at a low frequency. The maneuver requirement is
within the capability of the ship (less than
1. INTRODUCTION 2
± 0.15 m/s at 0.005 Hz and maybe still less when
Modern missile guidance system needs accurate the missile SDINS is better. SDINS without GPS
navigation for precise impact. Strap Down Inertial using an optical-based alignment scheme is tried for
Navigation System (SDINS) is used for the typical surface-launched missiles [22].
fire and forget missile launched from a moving
platform towards a land target. The principal aim is
An error propagation system involving the
to reduce the azimuth error as well as other
misalignment between Master and Slave INS is
Navigation errors for this Strapdown Inertial
Guidance System (SDIGS) where an implicit formulated for the velocity-matching algorithm
guidance Scheme based on perturbation technique is using AKF based Transfer Alignment (TA). This
used [11]. Finding the azimuth angle on a moving results in lesser azimuth error leading to improved
platform is a major technical challenge. Alignment Navigation accuracy for the ease of precision
of SDINS is a pre-requisite for self-contained Guidance. The result of the scheme led to lesser
navigation and guidance for a mission-critical Circular Error Probability (CEP) for precise impact
application. A master INS having highly accurate [4].
and stabilized gyros (drift <0.0010 /hr ),
accelerometers (bias<50 μg ) aided by gyrocompass In this paper, the System configuration and Kalman
is used for finding the azimuth. Filter formulation for the TA scheme is presented
and explained.
The problem is to find out the relative azimuth of
the missile’s SDINS [10] with respect to the Master
Typical simulation plans and results of trials are
INS on a moving platform, which will be used to
find out the near true orientation of the missile's highlighted subsequently with appropriate
SDINS. In addition, in a moving platform (like references in section 5 of this paper.
Ship) the pitch and roll rates are high and need a
stabilization system to reduce undue oscillation of 2. LITERATURE SURVEY
the launch platform.
Kalman filter as a navigation workhorse for
Transferring the missile azimuth available on the estimation of the dynamic state vector is well
ground to the SDINS of a missile is normally documented in the literature. Small-angle dynamics
performed by optical methods using theodolites and mechanization for Kalman filter implementation is
available survey information of the launch point. given by Bar Itzhack [12, 13, 14]. Transfer
alignment for moving vehicles was conducted by
Presently similar survey information in Sea is Chaudhuri et. al. in their papers [1][2][8]. PK Nandi
available by the help of superior class Master INS, et. al. conducted detailed simulation studies for
but the finding relative azimuth can be attempted by transfer alignment on oscillating moving base vide
various methods. Optical method of finding relative their papers [3][6].
azimuth being one of them.

2
 Ver13

However, both Chaudhuri and Nandi have not According to this criterion, transfer alignment
shown Kalman filter mechanization with LQG loop algorithms can be classified as follows [22]:
for variance estimation of the observation noise in
i. Acceleration / Rotation Rate Matching
adaptive Kalman filter mechanization.
ii. Velocity Matching
This paper discusses the complete mechanization of
iii. Integrated Velocity Matching
Kalman filter structures for estimation of velocity-
based attitude misalignment information including iv. Attitude Matching
the necessity of S-manoeuvre in the lateral axis for
v. Velocity and Attitude Matching
the estimation of azimuth angle and down gyro bias.
The velocity information is supplied by master and vi. Position Matching
slave together and fused with the adaptive Kalman
In acceleration/rotation rate matching the samples
filter.
were available at a high reception rate and were
very noisy as the computation of the samples does
Paper [18] describes the basic principles of not involve the integration process.
navigation. It is explained why attitude information
is critical to the start of inertial navigation. Paper Hence the position/velocity matching algorithm is
[19] explores robust strategies for in-motion inertial tried which gave a good sample with minimum
navigation explaining various statistically robust noise in the measurements as they involve
methods. integration process while calculating the samples.
Also, the velocity matching algorithm does not
assume rigid body assumption and can take care of
DARPA, US is relying on MEMS technology for lever arm and flexure moments, whilst the
positioning of small vehicles, for which attitude rotation/acceleration methods assume perfect rigid
information will be necessary [20]. The need for body for their working. The difference in velocity is
ubiquitous inertial navigation is given in Stovall used as input to the AKF for estimating the angular
[21]. misalignments between the Master and Slave INS.
The errors of the slave system are not directly The samples were available at a moderate
coupled with navigation outputs. For example, it is timestamp which was suited very much for the
not possible to calculate gyroscope drift errors by computational requirements of the AKF.
just using the velocity differences between the two The other methods like Integrated Velocity
systems. This is because drift errors generate Matching Attitude Matching, Velocity, and Attitude
attitude errors and these attitude errors generate Matching, Position Matching involve the further
velocity errors. In order to estimate those indirect integration in the calculation of samples, which in
errors, an estimation algorithm that can process the turn introduces lag in the computations of AKF, due
observed differences for a sufficiently long time is to which the filter became unstable and sometimes
necessary. The most significant factor that affects divergent [22].
the design of such an estimation algorithm is the
selection of vector types as given below that are
compared. The choice of vector type specifies the 3. SYSTEM CONFIGURATION FOR TA
overall structure of any estimation algorithm. SCHEME
Therefore, transfer alignment algorithms are In case of a launch from a moving base like a ship,
classified according to the type of vectors compared it is essential to align the INS in the missile (Slave
to generating an estimate. unit) with that mounted on the Ship (Master unit).

3
 Ver13

The Master INS is much more accurate, stable, and

precisely calibrated over a period of time.
Appropriate software is in-built which is aided by
external instruments (Velocity log equipment for
relative water velocity measurement and GPS
systems). TA is the process of aligning Slave INS to
the Master INS through velocity matching using
natural or deliberately induced excitation (i.e., S-
manoeuvre of a Ship) in the vertical direction before
the vertical launching of the missile. The pitch and
roll stabilization system provided in the Ship
reduces the missile yaw and pitch rates before
launch as shown in fig. 1.

The S-manoeuvre of the Ship excites the sensors of

both Master and Slave INS for the generation of
data to be used in aligning the slave INS of SDINS.

There is a launcher stabilizer on the ship which

isolates ship flexure and reduces the effect of the
ship’s roll-on missile’s pitch and ship’s pitch on
the missile’s yaw from 10o/s to 1.2o/s. The
missile’s horizontal axes are nearly perpendicular
to the ship’s horizontal axes.

4
 Ver13

The co-located master INS (on stabilizer) and slave and west velocity in the horizontal plane is
INS (inside missile) on the stiff stabilization system considered with sufficient observability under
(≈ 1 Hz) allow very little lever arm motion during maneuver for the azimuth channel and the bias
the S-manoeuvre performed by the ship in the states [23].
azimuth. The physical motion excites the system

states as shown in fig. 2. The master INS and slave The entire duration of the maneuver is reduced to
INS as depicted in fig. 2 give velocity and position less than 10 minutes with minimum effort of
~~
outputs (i.e., V n , λ ). manoeuvrings. The earth rotation rate is integrated
into the maneuver by the gyro output during the
The method of TA involves error modeling of both maneuver. In the specified 10 mins of maneuver, the
the systems and comparing two systems' outputs earth rotation rate is 2.5degrees. In order to estimate
misalignment, the differential output of slave INS
with the help of AKF. A conventional state
and master INS is used as an input to AKF based
estimator in the form of KF requiring exact process
TA algorithm.
noise and measurement statistics is not suitable for
this type of real-time system due to the requirement
The estimator should be capable of identifying and
of a higher-order model and longer convergence
learning the noise statistics online from the
time.
innovation sequence. An explicit adaptive optimal
estimator (in the form of AKF) is used [1], which
An error propagation system involving the makes direct use of online identification of noise
misalignment between Master and Slave INS is covariance. The estimated error states are used in a
formulated involving the 3-misalignment angles, 2- feedback mode to get near the true orientation of the
velocity errors, and 2-position errors. Error in north Slave INS.

5
 Ver13

The standard Matlab based routines were used without disturbing the existing missile launch
before selecting the final gains [24]. It was also configuration.
proved that online computations of gains are not
This total engineering was done not to disturb the
essential for this scheme [3]. The major part of the
existing validated S/W and H/W configuration.
algorithm was introduced in an additional PC
Minimum additional S/W was introduced in the
existing On Board embedded S/W.

Fig.2 System diagram with Kalman Filter & Control

4. at a much faster rate via On-Board Computer
FORMULATION OF KALMAN FILTER (OBC).
ALGORITHM The vector differential equation giving process
Error dynamics may be described as below in a
The KF used for TA has 7 states and uses 4 stochastic form:
differential measurement outputs as shown in fig
2. The dynamic system associated with the
Reference outputs are expected to be aided by
Navigation process is fundamentally in
external instruments like LOG and GPS inputs
continuous state space. Sensor signal drives the
from the ship. The detail states output, input along
navigation equations, which are non-linear
[16,17]. In order to apply linear filtering feedback are shown in fig.2. The process and
techniques linearization of these equations about a measurement noises are assumed to be stationary
reference trajectory (S-manoeuvre) have been and independent of each other, which are white
applied. Gaussian noise with a normal probability
distribution
~ ~ ~ ~
The states of the linear error model are North- X ( t )=F ( t )∗X ( t ) +B∗U ( t )+ G W (t) --- (1)
West-Vertical (NWV) have been already
described. For very small misalignment angles The sensor signals in slave IMU consist of
vertical axis can be decoupled from the error
substantial random error and may exceed the
model. However, height is updated from SDINS

6
 Ver13

deterministic error covariance used in the ~ ~ ~ ~

X k +1=Φ k∗X k + B k∗U k +G W k --- (2)
formulation. A discrete form stochastic difference
equation with a zero mean White Gaussian ~ ~ ~
Z k =H k∗X k + V k --- (3)
sequence W (k) is used,

~ ~ ~ ~ Where,
X k +1=Φ k∗X k + B k∗U ( k ) +G W k --- (2) ~
X k is the state vector

[]
where Φ k is the state transition matrix [1, 2] based α
on NWV frame. β
γ
~
X k= Δ v N
The structure of the Φ matrix need not be exactly
known in the AKF formulation. Errors in Δ vW
modeling of the Φ is taken care of by the adaptive ∆
modeling of the Q matrix, as given in equation 4 ΔɅ
below. ~
W k is the Plant noise from SINS gyros
~
~ Z k is the measurement vector

[ ][ ][ ]
W k comprises of the variances in accelerometer
~ ∆ vN Ʌ
and gyroscope noises. W k is given as vN vN
~ ∆ vW Ʌ

[ ]
2
σ gyro (3 , 1) Z k = v W − vW =
, with σ 2gyro ∧σ 2acc being the random ∆❑ Ʌ
2
σ acc (3 , 1) Ʌ Ʌ
∆ ɅɅ
noise variances of gyroscope and accelerometer
respectively. ~
V k is the Measurement noise from SINS
accelerometers
G is the coupling matrix between process noise
Bk is the control input selection Matrix
and the state of the system. The G matrix is given ~
by U k is the control input vector

[ ] [][ ]
I α ∆ϕ
n
C (1:2 , 1:2) β = Δϴ
b
zeros (2 , 3) γ ΔΨ
The observation model consists of a linear
equation of the error state as, The state transition matrix Φ k of the system[22]
~ ~ ~ comprises of seven error states, i.e. 3-angular
Z k =H k∗X k + V k --- (3)

[ ]
errors, 2-velocity errors, 2-position errors in the
1 00 00 00 form of a column vector
where H= 0 10 00 00
[ Δ ϕ , Δ θ , Δ ϕ , ∆V N , ∆ V W , ∆ λ , ∆ Λ ]
T
. The
0 01 00 00
0 00 10 00 measurements are the errors in velocities and
position between master INS and slave INS.
There are 4-measurements consisting of 2-
In the figure 2, the Kalman Filter is used with
velocity errors, 2-position errors given by a
discrete form stochastic difference equation with a
~ column matrix as [∆ V N , ∆ V W , ∆ λ , ∆ Λ]
T
. The
zero mean White Gaussian sequence W k
state transition matrix propagating the state of the
system between instants of time. Since this is only

7
 Ver13

an estimator without state vector augmentation, the accuracy will be dictated by master performance
Bk matrix is considered null matrix.w k is white only.
gaussian zero mean noise considered to be arising
due to modelling error with Slave INS (missile IMU) will be different from
true value due to the presence of slave INS Gyro
E [ w k wTk ]=Q , the process noise covariance .
and Accelerometer noises and biases. Further, they
The process noise covariance matrix has the same
get aligned to the proper NWV frame from the
dimensions of the state vector. z k is the misaligned N W V frame based on error
measurement of difference between the velocities quaternion [ ∆ q], which is fed back from the KF
of Δ vn , Δ vw estimator (ref. fig. 2). Quaternions are extracted
being the north∧west velocity errors Δ λ , ΔΛ from small angles due to ease of propagation of
being the latitude and longitude errors. the quaternion with available rates. The feedback
controller trims the feedback and the feedback
~ ~
The stochastic disturbance vectors W k and V k are applied is not instantaneous as is the case when a
treated as independent non-stationary Gaussian feedback controller is unavailable. The availability
noise sequences with following properties: of feedback controller reduces the transient
fluctuations of the filter. Propagation of these
E [~
W i ] =qi , E [ ( W i −qi ) ( W j−q j ) T ] =Qi δ ij- (4) states based on the slave INS different from the
true value is based on the state transition matrix
E~[ i] i [(
V =r , E V −r V −r T =R δ ----- (5)
i i )( j j ) ] i ij (Φ)[22] which gets updated based on the master
~ ~
INS ( V N , λ) and slave INS( ~
a NWV ) .
where δ ij =0 if i≠ j
¿ 1 if i= j The estimated states i.e. angular error, velocity
q i, r i are true means, Q i, Ri are true moments error, position errors
T
about the mean of the states and the observation [α , β , γ , ∆ V N , ∆ V W , ∆ λ , ∆ Λ] are fed back
sequences from the KF at the end of each Measurement
Update (MUP) and is used to turn the misaligned
frame of the slave to true NWV frame INS during
The state and output equations used for the INS TA as shown in the fig. 2.
error model in the TA scheme are given in the
above two equations (2) and (3). The DCM The recursive discrete KF, which runs indefinitely,
computation, velocity update, and position updates leads to a divergence problem. The basic sources
are carried out in the On Board Computer (OBC) may be modelling error, round-off error, and
~ ~
for getting SDINS system normal outputs ( V n , λ). Observability problems. Round-off error and
It is advisable not to use SDINS accelerometer Observability problems can be taken care of by
noise in the process noise term for writing the state appropriate states, measurement, and
propagation equation [i.e., ∆ V n and ∆ V w]. computational states. Since no exact mathematical
~ model is possible in the case of moving base INS
Further, handling measurement noises V k in KF
formation may include Slave IMU (SIMU) alignment problem, therefore adaptive KF gives
accelerometer latitude, longitude noises. better performance by estimating the measurement
noise under maneuver which is capable of tuning
These will help in considering the MINS as a itself to an ambient real-life situation, thus giving a
reference and the complete AKF based TA’s better accuracy in euler angle estimation.

8
 Ver13

In AKF exact knowledge of (Φ , H , Q , R) The expected value of this quantity can be shown
matrices are not essential. In the real-world as:
N
application of KF, one of the quantities available 1
E [ ^S γ ]= ∑ ¿ ¿ ¿ ¿ .... .......... (9)
in judging filter performance are the residuals and N k=1
their predicted statistics. Normal adaptive filter
models provide feedback from residuals. The expected value of covariance, as given in Bar
Shalom [24] is
4.1 ADAPTATION OF INNOVATION
(~
Z−H k ~
^
X k )¿ σ 2X −σ 2MIMU
SEQUENCE
N
Mehra [5] considered several techniques to ~
^ 1
R=E [ S^ γ ]− ∑ ¿ ¿ ¿ ¿
estimate error covariance matrices. This was N k=1
followed further in the next decade by several
researchers in this field [6]. Meyers and Tapley’s 1
N

[7] approach for Adaptive Limited Memory Filter

^
R=S^ γ −
N
∑ ¿¿¿¿
k=1
(ALMF) for implementation of innovation
correlation is followed in this case [6]. A variant
of this algorithm is implemented for the present N N
1 1
AKF, where the innovation sequence R= ∑ [(γ k −^γ k )(γ k −^γ k )¿ ¿T ]− ∑ ¿ ¿ ¿ ¿ ¿
^
N k=1 N k=1
(measurement vector) is adapted to minimize the
observability problem and faster convergence of …. (10)
the filter.
These estimates are based on N samples of γ k ,
Empirical estimators are derived in a 'batch' form k =1 … . N , which is assumed to be statistically
under the assumption of constant values for independent. The Adaptive Kalman Filter (AKF)
^ i r^ i R
q^ i Q ^ i over N successive stages of noise uses 30 samples in a moving window and uses
Joseph form for Kalman Filter gain computation
samples. The measurement residuals can be
[6]. Initially, for the first 30 samples collection,
expressed as:
non-adaptive KF is used. After getting the 30
γ k =~
Z k −H k ~
^
X k−¿= H k ¿¿
--(6)
samples, AKF is used with 30 samples moving
window for adaption.
An Unbiased estimator for γ , first-order moment
of measurement noise is taken as the sample mean: The state transition matrix is also updated during
the time update cycle. The 30 samples moving
N
1 window was chosen after studying 10 and 50
γ^ k = ∑ γ k …………………… (7)
samples moving window. With 10 samples the
N k=1
filter was diverging because of fewer samples and
An unbiased estimator for R is obtained by, first with 50 samples the convergence was taking a lot
constructing an estimator for Sγ , the covariance of of time for gathering samples and processing the
same. It was found from using the chi-square
measurement residuals as:
N distribution model, that an ensemble of more
^Sγ = 1 ∑ [(γ k − γ^ k )( γ k −^γ k ) ¿ ¿ T ]¿ ..... innovation sequences is advantageous. However,
N k =1 more time will be required to switch over from
.(8) non–adaptive to adaptive KF during S–maneuver.
This is not a desirable situation since it is shown

9
 Ver13

that non–adaptive KF gives rise to larger errors [1, for a measurement update at an interval of 1.8s
6]. It was suggested [6] to modify the (much faster than ship roll & pitch period of
measurement update of the state below where γ^ k is oscillation) and a time propagation step size of
the additional term 0.36s [2].
~
^
X +¿=
~^
X
−¿+ K k [γ k −^
k
γ k ]¿
¿
……… (11)
k
5. SIMULATION SETUP AND RESULTS
−1
where K k =P H ( HP H + R ) .
T T
Refer Bar- Various new testbeds were designed in the lab,
Shalom [24]. van, and ship (using a similar fixture, dummy
missile as well as real missile) before the final
The above modification will remove the undue launch. A test fixture was used to mount both
bias and the same was tried in the 7-states AKF for master INS and slave INS. Several ship trials were
TA trials. After deployment of adaptive estimator performed under various Sea states and the spread
the error feedback gains are selected using Linear of results for azimuth error was found to be less
Quadratic Gaussian Regulator and offline Matrix than 5 arc minutes when proper S-manoeuvre was
Riccati equation solution in discrete domain [2, 3]. performed by the ship [9] and slave IMUs. Master
IMU was mounted with its corresponding original
Final gains are selected using standard Matlab plate. The fixture was used in Laboratory as well
routines, which were computed offline [2, 3]. as Van trials. The same fixture was mounted
during the Ship trial directly on the stabilizer for
The control vector which is the output of the LQG carrying out TA before putting Slave IMU in the
~
regulator, ~
u k =−K X k as described in fig.2. The actual missile.
This result of TA was also validated optically by
control vector is fedback to the slave INS after the
measuring the relative azimuth angle between both
LQG regulator for attenuating the transient
IMUs using an alignment tool in place of Master
behaviour of the Kalman filter estimator of euler
IMU and slave IMU with its poroprism for lab,
angles.
van, and ship trials. Finally, during ship trial the
same optical validation method is used where the
The navigation computer computes and propagates
master IMU is placed on the stabilizer with its
the attitude in quaternion format. Hence the error
corresponding original reference plate and slave
angles obtained as estimates from the AKF are to
IMU in the missile.
be converted into quaternion format and feedback
to the navigation solution. The alignment
The test configurations for TA are as follows:
corrections ( ~u) are thus converted to the error
quaternion (ref. fig. 2) by the following
 Fixture with both IMUs in static condition
approximation:
T  Fixture with both IMUs in Van
u 1 u2 u3
[∆ q 0 ∆ q1 ∆ q 2 ∆ q3 ] =[1
T
] ----  Fixture with both IMUs on Ship stabilizer
2 2 2  Slave IMU in dummy missile with Master
(12)
 Slave IMU in real missile and Master IMU
on the stabilizer
The AKF based TA algorithm with the
formulation mentioned in this section along with
All the above tests were performed
the alignment correction of the error quaternions
chronologically to have minimum problems with
was implemented. After detailed design and
the last configuration which will be used during a
number of simulation runs, it was decided to go

10
 Ver13

real launch. Further, the data from the ship trial where the convergence was good and the
was brought into Hardware In Loop Simulation navigation results were satisfactory.
(HILS) set up to find any inherent bugs in the
software or in the concept. The Slave IMU in loop The figure-4 shows the on-line estimated diagonal
6 DOF simulation run was performed with initial elements (DelVn and DelVw) of measurement
quaternions from Ship TA results and oscillations noise variance, where dotted line is the true
of the simulated Ship using Missile Motion variance. Each line represents the time evolution
Simulator (MMS). In addition, AKF with Φ−¿ of the diagonal elements. Fig.4 depicts the
matrix simulating Master IMU in PC testbed was variance in the states of error velocities in North
and West directions. They do not necessarily
tried independently. It may be highlighted that
represent the error variances of the measurement
linear acceleration cannot be given to MMS for
and process noises (q i , r i ¿ .
generating S-curves for testing TA software.

6. CONCLUSION
Therefore, Ship trials for validating TA in realistic
conditions became a necessity, where S-curve is The efficacy of an implementable data fusion
given to tackle the unobservable azimuth problem. scheme using AKF for TA is validated, which
A typical TA trial test result is given in fig. 3, shows its usefulness for the missile guidance
which shows the proximity of converged system. The reduced-order AKF devoid of vertical
misalignment between Master and Slave IMUs in velocity error state for TA has shown that an
azimuth. explicit optimal estimator using online
identification of noise covariance from the
The value was validated by optical measurements innovation sequence based on stable mean is
performed near the shore in the calm Sea. The capable of giving less than 5 arc minutes azimuth
convergence was achieved within 250 sec. The in moving platform.
velocity matching in North and West directions is
near to 0.01 m/s and AKF corrections are observed Further, the poor observability of azimuth and
in γ (i.e. error in azimuth). Due to randomness in spread of noise in missile SDINS demands S-
nature and real time data under sea wave manoeuvre of the Ship at a low frequency. The
simulation conditions, the data is oscillatory and idea may be tried with Fuzzy logic for future
mean is shifted from '0' mean to 0.025m/s which is developments. AKF with the reduced number of
well within the acceptable band of +/- 0.05m/s, states may be attempted in the future upgrade.

11
 Ver13

REFERENCES 15. Savage P. G., Strap down System Algorithms, Advances

in Strap down Inertial Systems, NATO AGARD Lecture
Series, 1984, No: 133, ISBN: 92-935-035 1 -1.
1. S.K. Chaudhuri, P.K. Nandi, “Transfer Alignment for 16. Savage, P. G., Strap down Analytics Part I, Strap down
space vehicles from Moving base”. Def. Sci. J., 2005, Vol. Associates, 2000, ISBN: 0971778604 9780971778603.
55, No.3. 1 value
DOI: https://doi.org/10.14429/dsj.55.1986 17. Savage, P. G., Strap down Analytics Part II, Strap down
Associates, 2000.
2. S.K.Chaudhuri, G. V. Chalam, A. Rambabu et.al.,
“Adaptive Kalman Filter based Transfer Alignment for 18. Basic principles of Inertial Navigation. Seminar on
Moving Base”. Technical Report, Research Centre 1 value Inertial Navigation Systems" Tampere University of
Imarat, 2001. Technology, 2018, Page5, DOI:
https://fdocuments.in/document/basic-principles-of-
3. P.K. Nandi, P. P. Adhikari et. al., “Upgradation and inertial-navigation.html.
Validation of Adaptive Kalman Filter based Transfer
Alignment and Navigation”. Technical Report, Bengal 1 value 19. Wan Mohd Yakoob Wan Bejuri, Mohd Murtadha
Engineering College, 2002. Mohamad, Farhana Syed Omar, and Nurfarah Ain Limin,
"Robust special strategies resampling for Mobile Inertial
Navigation Systems". International Journal of Innovative
4. George R. Pitman, John Wiley & Sons Inc, “Inertial
Technology and exploring engineering. Vol 9(2) PP3196
Guidance”, 1962. ISBN: 0471690694. 1 value – 3024, DOI: 10.31224/osf.io/gak4x.

5. R.K. Mehra, “On the identification of variances an 20. Darpa.mil, "Micro Technology for positioning,
Adaptive Kalman Filtering”. IEEE trans. on Automatic navigation, and Timing”, 2017,
Control, 1970, DOI: 10.1109/TAC.1970.1099422 darpa.mil/program./micro-technology-for-positioning.
1 value
6. P.K. Nandi et. al, “Moving base INS Alignment based on 21. Stovall Sherryl. H, “Basic Inertial Navigation” global
Adaptive Kalman filter “. Technical Report, Bengal security Org, 1997, Report No. NAWCWPNS TM 8128
Engineering College, 2000. Sept 1997.
1 value
22. Titterton, D. H. and Weston, J. L., Strapdown Inertial
7. K.A. Mayers, B.D.Tapley, “Adaptive sequential estimation
Navigation Technology, 2nd edition, AIAA Progress in
with unknown noise statistics “. IEEE trans. on Automatic
Astronautics and Aeronautics 2004, 10.4 In-flight
Control, 1976, DOI: 10.1109/TAC.1976.1101260.
alignment, and 10.5 Alignment at Sea, pp. 289-307

1 value
8. S. K. Chaudhuri, G. Venkatachalam et. al, "Integrated
23. Rogers, R. M., Applied Mathematics in Integrated
On-Board Software ".Technical Report, Research Centre
Navigation Systems, 3r ed., AIAA Education Series, Ch.
Imarat, 1994.
11. INS Ground Alignment, Ch. 13. In-Motion Alignment
Fig.3 TA Result – the convergence of Azimuth misalignment between Master and Slave IMUs and Velocities
9. S. K. Chaudhuri, G. Venkatachalam et. al, “Analysis of 24. Bar-Shalom, Y., Rong Li, X., and Kirubarajan,
TA results using AKF“. Technical Report, Research T., Estimation with Applications to Tracking and
Centre Imarat, 2001. Navigation, John Wiley 2001, Chapter 11 Adaptive
Estimation and Maneuvering Targets
10. Bar-Itzhack I. Y., Porat B., Azimuth Observability
Enhancement during Inertial Navigation System In-Flight
Velocity Alignment, Journal of (Guidance and Control, 1980, Vol.
variance
2 23, No. 24, DOI: 10.2514/3.55999.
σ ¿ m /sec .
11. Bar-Itzhack I. Y., Vitek Y., The Enigma of False Bias
Detection in a Strapdown System during Transfer
Alignment, Journal of Guidance and Control, 1985,
Vol.8, No. 2, DOI: 10.2514/3.19956.

12. Bar-Itzhack, Berman N., Control Theoretic Approach to

Inertial Navigation Systems, Journal of Guidance and
Control, 1988 Vol.11, No. 3, DOI: 10.2514/3.20299.

13. Goshen-Meskin D., Bar-Itzhack I. Y., Observability

Analysis of Piecewise Constant Systems – Part I: Theory,
IEEE Transactions on Aerospace and Electronic Systems,
Fig.4: The adaptive measurement
1992, Vol 28, No: 4, DOI: 10.1109/7.165367. noise variance of diagonal elements
14. Goshen-Meskin D., Bar-Itzhack I. Y., Observability
Analysis of Piecewise Constant Systems – Part II:
Application to Inertial Navigation In-Flight Alignment,
IEEE Transactions on Aerospace and Electronic Systems,
1992, Vol 28, No: 4, DOI: 10.1109/7.165368.

12