Rapid transfer alignment for large and time-varying

attitude misalignment angles

Laetitia Tosoni, Minh Tu Pham, Paolo Massioni, Elliot Broussard

To cite this version:

Laetitia Tosoni, Minh Tu Pham, Paolo Massioni, Elliot Broussard. Rapid transfer alignment for
large and time-varying attitude misalignment angles. IEEE Control Systems Letters, 2023, pp.1-6.
�10.1109/LCSYS.2023.3283565�. �hal-04122701�

HAL Id: hal-04122701

[Link]
Submitted on 8 Jun 2023

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 Rapid transfer alignment for large and time-varying attitude
misaligment angles*
Laetitia Tosoni1,2 , Minh Tu Pham2 , Paolo Massioni2 and Elliot Broussard1

Abstract— Transfer alignment is an effective method to The principle of transfer alignment is as follows. Two
estimate the attitude difference between two inertial platforms. inertial measurement units, a master INS and a slave IMU,
Traditional transfer alignment methods are either meant for exist in different reference frames as part of a single vehicle.
small misalignments or use nonlinear filtering. Linear methods
for large misalignments use quaternion attitude representation, They both measure the same variable, each in their own
but their performance degrades when the misalignment is not reference frame. These two measurements are compared
constant or slowly varying. This work proposes to apply rapid in order to deduce the orientation difference between their
transfer alignment to the inertial stabilization of a Satcom On reference frames.
The Move antenna. In this application case, the two inertial Several methods of transfer alignment exist, and they
platforms are connected by a kinematic chain, meaning the
attitude difference between the inertial platforms is both large are usually classified based on the type of measurement
and time-varying. This work proposes a method to estimate this the estimator uses as inputs. According to [1], rotation
attitude difference. A Kalman filter is used for the estimation rate, attitude, velocity or acceleration are the most common
of the unknown angles. To that effect, a new propagation model methods. Combinations of several measurements are also
is developed to take into account the misalignment variations used: the technique known as rapid transfer alignment, first
using master and slave rotation rates. Simulation results show
the performance of the proposed solution in a Satcom antenna described by [5], uses a combination of attitude and velocity
inertial stabilization case. measurements. It allows a very fast and very precise estima-
tion (estimation error of less than 1 mrad within 10 seconds),
I. I NTRODUCTION and according to [6], it is regarded as the best transfer
Inertial navigation algorithms provide an effective means alignment technique. However, it estimates a small, constant
to compute the attitude and position of vehicles (aircraft, misalignment angle, and it uses small angle approximation
ships, etc.) [1]. In some applications, a subpart of the vehicle to linearize the model. As a consequence, the larger the mis-
can be equipped with its own navigation system: the most alignment, the more the algorithm’s performances degrade.
common example is weapons carried by an aircraft. Furthermore, potential variations in the misalignment angle
aren’t taken into account by the model.
In a projectile application [1], the final objective is for
Following the advent of rapid transfer alignment, efforts
the projectile to follow its intended trajectory. The aircraft’s
were made to reduce the computational load of the algorithm,
aiming systems rely on the main inertial navigation sys-
as done by [7], and to compensate the carrier flexure, as done
tem (INS) information to calculate said trajectory. As a
by [8]. Carrier flexure is the rapid change of misalignment
consequence, the projectile’s own inertial measurement unit
angle due to vibrations and deformations between the master
(IMU) must be aligned with the main INS to follow that
INS and the slave IMU, and it impacts the accuracy of the
trajectory after launch. Transfer alignment algorithms have
estimation when not taken in consideration. These methods
been originally developed to ensure that the projectile’s
are still applied to a small and constant misalignment angle,
navigation system worked in the same reference frame as
and suffer from the drawbacks previously mentioned.
the aiming system. To do so, transfer alignment algorithms
Several methods exist to estimate large misalignment
estimate the misalignment between a high-grade INS, located
angles. Some use non-linear filtering, such as the unscented
on the carrier vehicle, and a low-cost IMU, located on the
Kalman filter used by [9] or the extended Kalman filter used
missile.
by [10]. However, works such as [11] and [12] managed
In addition to the original projectile applications, other
to estimate a large misalignment angle with a linear Kalman
applications exist, such as aircraft carried by a ship as in [2],
filter, by using a quaternion attitude representation instead of
remote sensors carried by an aircraft as in [3], or spacecraft
Euler angles. In all these cases, however, the misalignment
launched from a mobile base as studied in [4].
estimated remained constant, and the model wasn’t built to
*This work is supported by the company Thales and Association nationale take into account potential variations.
de la recherche et de la technologie (ANRT). In the present work, transfer alignment is applied to the
1 Laetitia Tosoni and Elliot Broussard are with Thales SIX,
inertial stabilization of a satellite communication antenna
Gennevilliers, France. [Link]@[Link],
[Link]@[Link] (called Satcom On-The-Move, or SOTM). As presented by
2 Laetitia Tosoni, Minh Tu Pham and Paolo Massioni are [13], in order to obtain high data rates, the antenna must
with Univ Lyon, INSA Lyon, Université Claude Bernard Lyon be directional, which means that the received and emitted
1, Ecole Centrale de Lyon, CNRS, Ampère, UMR5005, 69621
Villeurbanne, France. [Link]@[Link], electromagnetic waves are oriented in a specific direction
[Link]@[Link] of space, called the line of sight or boresight. The line of
 II. P RELIMINARIES
A. Nomenclature
S (s) frame
(n) frame
Elevation axis • qkj : quaternion expressing the rotation from frame k to
E
(b)
frame j
Master INS frame A
• Ckj : rotation matrix from frame k to frame j
Misalignment
M
• q̄ the quaternion conjugate of q
Azimuth
(m) frame
axis • For a quaternion q, we write
 q0 its scalar part and ~q its
Vehicle T
vector part such as q = q0 ~qT . The vector part
 T
is made of three terms such as ~q = q1 q2 q3 .
• C(q): rotation matrix representing the same orientation
Fig. 1. Structure of the antenna positionning system
as the attitude quaternion q
l
• ω
~ j/k : rotation rate of frame j relative to frame k
sight must be steered in order to be aligned with the satellite expressed in coordinate frame l
to ensure a sufficient data rate transfer. The accuracy with • ~alj/k : acceleration of frame j relative to frame k ex-
which the antenna must be pointed depends on the antenna pressed in coordinate frame l
l
size and the radiofrequency band used to communicate. • ~vj/k : velocity of frame j relative to frame k expressed
In order to keep the antenna pointed towards the satellite, in coordinate frame l
it is mounted on an antenna positioning system (APS). The • ~g n : gravity acceleration in coordinate frame n
APS is made of several axes actuated by servomotors. In our • f~s : acceleration sensed by the slave IMU such as f~s =
application case, the APS has two axes: an azimuth axis, and ~ass/n + ~g s
an elevation axis. In order to control the APS, the line of • θAz : rotation angle around the azimuth axis.
sight’s current direction must be known; as a consequence, a • θEl : rotation angle around the elevation axis.
slave IMU is aligned with the antenna’s line-of sight. Placing • S(~u): 3-by-3 skew-symmetric matrix associated with 3-
the IMU in the same reference frame as the line of sight by-1 vector ~u
minimizes the risks of assembly misalignment errors. The • ẋ: derivative of variable x with respect to time
master INS is located on the vehicle. As seen from Fig. • x̂: estimation of variable x
1, the APS can create a large and time varying orientation
B. Quaternion attitude representation
difference between the master and the slave inertial sensors.
In the currently existing transfer alignment algorithms, this Unit quaternions can be used to represent attitude [14].
variation is an important perturbation that isn’t accounted for They are more compact than rotation matrices, and unlike
by the models. Euler angles, they avoid gimbal lock issues.
The contribution of this method relies on the inclusion of Five properties are key for attitude quaternions. The first
the propagation model for the movements of the secondary one is their derivative [14]:
IMU, namely the one that is connected to the antenna in our 1 j
application. This makes the method directly applicable to all q̇kj = k
q × ωk/j (1)
2 k
cases where large relative, time-varying angles are featured
between the two platforms; this case is usually neglected in where × is the Hamilton product between two quaternions,
other transfer alignment methods, due to the fact that the two qkj is the attitude quaternion from frame k to frame j, ω l
~ j/k is
platforms are connected to the same rigid body. the rotation rate of j relative to k expressed in l, and for any
k k
Specifically, this does not mean that the former algorithms rotation rate ω ~ k/j , ωk/j is the associated quaternion such as
h iT
cannot be used for tracking antenna movements; it means k
ωk/j = 0 ω k T .
~ k/j
though that they will need to be tuned in a way that the The second one is the use of attitude quaternion to change
propagation dynamics is considered as an unknown distur- another quaternion’s coordinate frame. For example, with
bance, which is inefficient. Adding the propagation model rotation rates:
makes it unnecessary to do this step, making an efficient
o
and performant tuning of the filter quite easy and direct. ωj/k = qlo × ωj/k
l
× qol (2)
This article is organized as follows. Section II specifies the
The third one is the conjugate of an attitude quaternion,
quaternion properties and different reference frames that will
with q̄ the conjugate of q :
be used in the model. Section III presents the measurement
¯
model, propagation model and lever arm compensation of the qkj = qjk (3)
transfer alignment algorithm. The lever arm compensation
is meant to take into consideration the physical distance Therefore,
T
between the master and slave systems, and its impact on qkj × qjk = 1 0 0 0

(4)
velocity measurements. Section IV presents the simulation  T
conditions and results that confirm the effectiveness of the with 1 0 0 0 being the multiplicative identity for
proposed method for our purposes. the quaternion algebra.
 The fourth one is the matrix form expression of quaternion A. Algorithm structure
multiplication. For any two quaternions q and p: A linear system can be defined with a measurement matrix
H and a propagation matrix A. Using these matrices and a
−~qT
 
q0
γ(q) = (5) measurement vector Y , it is possible to estimate the system’s
~q q0 I + S(~q)
state vector X with linear equations:
pT
 
p0 −~ (
χ(p) = (6) Ẋ = AX
p~ p0 I − S(~ p) (8)
Y = HX
q × p = γ(q)p = χ(p)q (7) The state vector to be estimated is:
The last property is the conversion from attitude quater- T
qsn00 qsm

X= (9)
nion to other attitude representations. A quaternion can be
converted to a rotation matrix or to Euler angles for better with qsn00 being the attitude of frame (s) relative to frame
readability. In this paper, we will display results in Euler (n) at t = 0, and qsm being the attitude of (s) relative to (m)
angles using the ZYX convention. at the current time.
The inputs of the algorithm are as follows:
n
C. Reference frames • ~
vm/e : the master velocity provided by the master INS
m
• ω
~ m/n : the master rotation rate provided by the master
The reference frames used in this paper are the same as INS
in [12]. They are illustrated in Fig. 1 and defined as follows: n
• qm : the master attitude provided by the master INS
NED frame (n): Local North-East-Down frame. ~s
• f : the slave accelerometers output provided by the
NED frame at start time (n0 ): Local North-East-Down slave IMU
frame at the start of transfer alignment. This frame is non- s
• ω
~ s/s0 : the slave rotation rate provided by the slave IMU
rotating with regard to inertial frame. • θAz and θEl : azimuth and elevation angles provided by
Master INS body frame (m): Orthogonal reference frame the APS encoders, which can yield θ̇Az and θ̇El through
aligned with master INS axes. derivation
n0 n n
Bearing body frame (b): Reference frame originating at • Cn , ω ~ i/e and ω~ e/n can be calculated from the vehicle
the base of the azimuth joint, stemming from the rotation of position in earth frame given by a Global Navigation
(m) around the azimuth axis. Satellite System (GNSS).
Slave IMU body frame (s): Orthogonal reference frame Fig. 2 gives an overview of the algorithm structure. Each
aligned with slave IMU axes. step described on this figure are further detailed in the
Slave IMU body frame at the start time (s0 ): It coincides following sections.
with the slave INS body frame at the start of transfer
alignment. This frame is non-rotating relative to inertial Simulink simulation environment
frame. 𝑛0
𝑞𝑚 𝐶𝑛𝑛0 , 𝑛
𝐶𝑛𝑛0 , 𝑣𝑚/𝑒 , 𝑓𝑠 𝑠
𝜔𝑠/𝑠0 𝑚
𝜔𝑚/𝑛
𝑛
Earth frame (e): Geocentric frame. 𝜔𝑒/𝑖 , 𝜃𝐴𝑧 , 𝜃𝐸𝑙 ,
𝑛 ሶ , 𝜃ሶ𝐸𝑙 Quaternion
𝜔𝑛/𝑒 𝜃𝐴𝑧
Inertial frame (i): Frame considered non-rotating and Lever arm
integration
Conversion (17)
unmoving for the entirety of the transfer alignment. compensation
to matrix
(Section 3.2)
𝑛0 𝐶𝑠𝑠0 𝑞𝑠𝑠0
𝑣𝑚/𝑒|𝑐𝑜𝑚𝑝
III. R APID TRANSFER ALIGNMENT ALGORITHM FOR Calculation of α (10)
Calculation
of β (11)
SOTM Calculation of A
α(t) β(t)
(24)
In traditional transfer alignment, the master INS and slave Transformation into quaternion
𝐴
IMU measure the same variable in their respective reference α(t), β(t)
Calculation of H (13)
frame. In this work’s application case, the two inertial
𝐻
platforms are separated by a kinematic chain. This kinematic
Kalman filter
chain’s motion will induce a specific velocity difference
𝑛0
𝑞ො𝑠0 𝑞ො𝑠𝑚
between the slave and the master, and will therefore need
to be compensated. It can also cause large and time-varying
attitude differences between the master frame and slave Fig. 2. Block diagram of the transfer alignment algorithm
frame.
In this section, the algorithm’s structure is described,
then the measurement and propagation equations are given. B. Measurement equation
Lever arm compensation and integration method for the The method of this work is based on the pseudo-
measurement equation are taken into consideration. measurement equation explained in [12]. This Section III.B
 m
 T
briefly recaps the elements from this former reference that only on one axis, we have ωb/m = 0 0 θ̇Az and
are needed in order to understand the rest of the paper. b
 T
ωs/b = 0 θ̇El 0 , yielding:
The measurement model is a pseudo-measurement equa-
tion as all the terms of the measurement vector Y are zeros.
~ and β~
In order to get to this, first we define two vectors α n
vs/m =
such as: n
Cm n
(S(ωm/n )(M~A + Rz (θAz )(AE
~ + Ry (θEl )ES)))
~
n0 n0
α
~ (t) = ~vn/e|comp (t) − ~vn/e|comp (0) n
+ Cm m
Rz (θAz )(S(ωb/m ~ + Ry (θEl )ES))
)(AE ~
Z t
+ Cnn0 ((2~ n
ωe/i +ω n
~ n/e n
) × ~vn/e − ~g n )dt (10)
n
+ Cm b
Rz (θAz )Ry (θEl )S(ωs/b ~
)ES
0
(15)
Z t
~ =
β(t) Css0 f~s dt (11) Using (15), the lever arm effect can be compensated
0
knowing the system’s dimensions, the vehicle position, and
n0
~vn/e|comp is the master INS velocity compensated for the angular positions and angular rates on the azimuth and
lever arm effect. Its calculation is detailed in Section III- elevation axes.
C. The details of the real-time calculation of (11) are given
in Section III-D. D. Integration
α(t) and β(t) are the quaternions associated with vectors In modern transfer alignment, the computed slave velocity
α ~
~ (t) and β(t). We can then write the pseudo-measurement and attitude are accessible measurements, although they
equation: contain sensor errors [15], [16], [17]. This is not the case in
our application case: the slave IMU only provides rotation
rates Rand acceleration. In order to compute qss0 in (13)
 
γ(α(t)) − χ(β(t)) 04×4
(08×1 ) = X (12) t
and 0 Css0 f~s dt in (11). these measurements have to be
χ(qss0 ) n0
−γ(qm )
integrated in real time.
As a conclusion, Using (1), we write:
 
γ(α(t)) − χ(β(t)) 04×4 1 s0
H= (13) q̇ss0 = q × ωs/s0s
(16)
χ(qss0 ) n0
−γ(qm ) 2 s
The details of the calculation of qss0 are given in Section Since the frame (s0) is immobile relatively to the inertial
s
III-D. frame, ω ~ s/s0 is the output measurement of the slave gyrom-
eters. Equation (16) can therefore be integrated using qss0 at
C. Lever arm compensation the previous step to yield qss0 at the current step.
n0 All that remains is the issue of initialization. Traditional
In (10), vn/e|comp refers to the velocity of the master
INS in reference frame n0 compensated for level arm effect. rapid transfer alignment methods such as the one described
n0 n0 in [5] calculate the slave attitude, making it necessary to
vn/e|comp is an estimation of vs/e , the velocity of the slave
IMU in reference frame n0 . approximate the initial slave attitude with the initial master
By decomposing vs/e n0
, we get the following: attitude. Instead, we calculate the slave orientation between
its initial attitude and its current attitude. At t = 0, the
n0
vs/e = Cnn0 (vs/m
n n
+ vm/e ) (14) reference frames  (s) and (s0) are aligned; therefore, by
T
n
definition, qss0 = 1 0 0 0 .
vm/e is the master velocity in n frame and is given by the Css0 can be deduced from qss0 . This allows the integration
master INS. Cnn0 can be calculated knowing the position of of the slave accelerometer output as done in (11). It is
n
the vehicle and the curvature of the Earth. However, vs/m is ~
initialized by β(0) = 0.
unknown and must be calculated. This calculation uses the Notice that the errors on the measurement of ω s
~ s/s0 and
measurements from the encoders located on the azimuth and
f~ due to slave IMU imperfections will accumulate over
s
elevation axes and the dimensions of the different parts of
time, and eventually they may cause these calculations of
the APS. The APS configuration is illustrated in Fig. 1. ~
qss0 and β(t) to drift from their real values. On the other
Let us define the following matrices:
hands, transfer alignment algorithms are commonly used for
• Rz (θ) is the matrix expressing the rotation of angle θ short period of times, which limits the extent of the drift due
about the z axis. to error integration; in this work as well, we have verified
• Ry (θ) is the matrix expressing the rotation of angle θ that within the simulation time, no significant drift has been
about the y axis. observed.
In our application, the dimensions of the vectors M~A, AE ~
and ES ~ are in the order of one meter. E. An innovative propagation equation
n
vs/m can then be computed through kinematic consider- [12] made the hypothesis that any variation of X is
ations. Because the azimuth and elevation rotation operate due solely to process noise. While this is true of qsn00 by
definition, it is not necessarily the case for qsm . In this work’s The simulation model includes vehicle, APS and antenna
application case, vehicle motion causes the APS to change simulation, and was developed by the company Thales.
the antenna’s direction relative to (m) frame, so that the The vehicle motion model is based on the work of [18].
antenna’s line of sight remains pointed towards the satellite. Two motion profiles have been simulated: relaxed and intense
As a consequence, the (s) frame, aligned with the antenna’s off-road conditions. An INS model in (m) frame, a two-axes
line of sight, changes its attitude relatively to (m) frame. APS and an IMU in (s) frame then use this motion infor-
Vehicle flexure also impacts the attitude of (s) relative to mation to calculate their respective outputs: master velocity,
(m). rotation rates and attitude, slave acceleration and rotation
We will now develop a propagation model so that the rate, and APS axis rotation angles. During the simulated
observer can anticipate variations of qsm . According to (1), scenario, the APS rotates the antenna until it points towards
the derivative of qsm is as follows: the satellite, then it keeps it inertially stabilized, causing a
misalignement between (m) frame and (s) frame: it is this
1 m
q̇sm = s
q × ωs/m (17) time-varying misalignement that the present work aims to
2 s estimate.
s
The rotation rate ω
~ s/m is unknown, so we decompose it: The slave IMU’s specifications and noises are given in
s s s
Table I. The slave IMU runs at 2000 Hz. The master INS is
ωs/m = ωs/n − ωm/n (18) considered ideal and runs at 100 Hz.
s
ω
~ s/n is the rotation rate provided by the slave IMU. TABLE I
s
However, ω~ m/n is unknown. Using (2) in (18) yields: A PPROXIMATE SENSOR AND NOISE SPECIFICATIONS OF THE SLAVE IMU
s s s m
ωs/m = ωs/n − qm × ωm/n × qsm (19) Error Order of magnitude
Accelerometer range 102 m/s2
m
ω
~ m/n is the rotation rate provided by the master INS. Accelerometer sensitivity 10− 8 (m/s2 )/LSB
Rewriting (17) using (19) yields: Accelerometer bias 10− 2 m/s2
Accelerometer axes misalignment 10−3 rad√
1 m 10− 4 (m/s2 )/ Hz
q̇sm =
q × (ωs/ns s
− qm m
× ωm/n × qsm ) (20) Velocity random walk
2 s Accelerometer bias instability 10− 3 m/s2
We can then develop this equation. By using (4), we get: Gyroscope range 10 rad/s
Gyroscope sensitivity 10−8 (rad/s)/LSB
1 m Gyroscope bias 10−5 rad/s
q̇sm = s
(q × ωs/n m
− ωm/n × qsm ) (21)
2 s Gyroscope axes misalignment 10− 3 rad
− 2
G-sensitivity 10 5 (rad/s)/(m/s
√ )
Using (5) and (6), we can put this equation in matrix form: −
Angular random walk 10 4 (rad/s)/ Hz
1 Gyroscope bias instability 10− 4 rad/s
q̇sm = s
(χ(ωs/n )qsm − γ(ωm/n
m
)qsm ) (22)
2
By factorizing this, we can write the propagation model The Kalman Filter noise covariance matrices were de-
of this rapid transfer alignment algorithm applied to large termined through empirical testing. The structure of the
variable angles: algorithm, which includes calculation of the model matrices
and Kalman filter, is given in Fig. 2.
The integration described in (16) and (11) are executed
 
04×4 04×4
Ẋ = 1 X (23)
04×4 s
2 χ(ωs/n ) − 12 γ(ωm/n
m
) with a three-steps Adams-Bashforth method, more accurate
than a forward Euler integration.
As a conclusion,
  B. Simulation results
04×4 04×4
A= 1 s 1 m (24) The initial convergence happens within 2.5 s in relaxed
04×4 2 χ(ωs/n − 2 γ(ωm/n )
)
conditions and 15s in intense conditions (Fig. 3). This
Equation (24) is an innovative propagation model express- is accomplished without any specific manoeuver from the
ing the misalignment quaternion’s variations in time, thus vehicle.
extending the model developed by [12] that only applied Two variables are of interest to examine the performance
to constant misalignment angles. If the angle is constant, of the proposed solution. One is the estimation error (Fig.
( 12 χ(ωs/n
s
) − 12 γ(ωm/n
m
))qsm = 0, yielding the trivial propa- 3): the angle between the estimated antenna direction and the
gation model available in the literature. actual antenna direction. It is compared by the preexisting
method developed in [12]. Simulation results show that
IV. S IMULATION
the estimation error stays beneath 0.22◦ in relaxed off-road
A. Simulation conditions conditions, and beneath 1◦ in intense off-road conditions.
A simulation in a MATLAB Simulink environment is In both cases, for the same tuning, the proposed method
used to evaluate the performance of the proposed method. performs better than the preexisting method.
 [5] J. Kain and J. Cloutier, “Rapid transfer alignment for tactical weapon
applications,” in Guidance, Navigation and Control Conference, 1989,
p. 3581.
[6] Z. Xiong, F. Sun, and Q. Nie, “Observability analysis of INS rapid
transfer alignment,” in 2006 6th World Congress on Intelligent Control
and Automation, vol. 1, June 2006, pp. 1664–1668.
[7] G. Xu, Y. Huang, Z. Gao, and Y. Zhang, “A computationally efficient
variational adaptive kalman filter for transfer alignment,” IEEE Sensors
Journal, vol. 20, no. 22, pp. 13 682–13 693, Nov 2020.
[8] X. Chen, Z. Ma, and P. Yang, “Integrated modeling of motion
decoupling and flexure deformation of carrier in transfer alignment,”
Fig. 3. Amplitude of the angle between estimated antenna direction and Mechanical Systems and Signal Processing, vol. 159, p. 107690, 2021.
actual antenna direction, top: driving in relaxed off-road conditions, bottom: [9] J. Cheng, T. Wang, D. Guan, and M. Li, “Polar transfer alignment
driving in intense off-road conditions of shipborne SINS with a large misalignment angle,” Measurement
Science and Technology, vol. 27, no. 3, p. 035101, jan 2016.
[10] X. Cui, C. Mei, Y. Qin, G. Yan, and Z. Liu, “A unified model for
Pointing error (°)

1
X 95 transfer alignment at random misalignment angles based on second-
0.5
X 95 Y 1.02 order EKF,” Measurement Science and Technology, vol. 28, no. 4, p.
Y 0.24
045106, feb 2017.
0 [11] S. Zhou, H. Xu, H. Dai, and G. Wu, “Novel rapid transfer alignment
0 10 20 30 40 50 60 70 80 90 100
Time (s) algorithm for strapdown inertial navigation system,” in 2011 Interna-
tional Conference on Electrical and Control Engineering, Sep. 2011,
Fig. 4. Cumulative distribution of the pointing error using the proposed pp. 6012–6016.
solution, in percent of the signal between t=0.5s and t=200s [12] Y. Chen and Y. Zhao, “New rapid transfer alignment method for SINS
of airborne weapon systems,” Journal of Systems Engineering and
Electronics, vol. 25, no. 2, pp. 281–287, April 2014.
[13] J. DeBruin, “Control systems for mobile Satcom antennas,” IEEE
The other variable of interest is the pointing error: in a Control Systems Magazine, vol. 28, no. 1, pp. 86–101, Feb 2008.
[14] H. Parwana and M. Kothari, “Quaternions and attitude representation,”
SOTM simulation similar to the one described in [19], the 08 2017.
algorithm results are used to control an APS and to point [15] J. Cheng, J. Cai, Z. Wang, and J. Liu, “A novel polar rapid transfer
an antenna towards a satellite. The pointing error is the alignment for shipborne SINS under arbitrary misalignments,” IEEE
Access, vol. 8, pp. 197 567–197 580, 2020.
angle between the target satellite direction and the actual [16] J. Cai, J. Cheng, J. Liu, Z. Wang, and Y. Xu, “A polar rapid transfer
pointed direction, and its cumulative distribution over 200s alignment assisted by the improved polarized-light navigation,” IEEE
is given in Fig. 4. Results show that 95% of the time, pointing Sensors Journal, vol. 22, no. 3, pp. 2508–2517, Feb 2022.
[17] W. Lyu, X. Cheng, and J. Wang, “Adaptive UT-H∞ filter for sins’
error remains under 0.25◦ in relaxed off-road conditions and transfer alignment under uncertain disturbances,” IEEE Access, vol. 8,
1.02◦ in intense off-road conditions. pp. 69 774–69 787, 2020.
From the MIL-STD-188-164 standard [20] and the for- [18] M. Alazab Elkhouly, “Standardized testing conditions for satellite
communications on-the-move (SOTM) terminals,” Ph.D. dissertation,
mula given in [21], we can conclude that this algorithm can Technische Universität Ilmenau, 2018.
be used to point antennas up to 82 cm in diameter. [19] E. Broussard, M. T. Pham, M. Ladevez, X. Brun, and B. Vion, “Hybrid
satellite position estimator using self-induced dither for SATCOM on
V. C ONCLUSION the Move,” in 20th IFAC World Congress, Toulouse, France, Jul 2017,
pp. 7208–7212.
This paper addresses the problem of estimating a large [20] NPFC, MIL-STD-188-164: Interoperability of SHF satellite commu-
varying misalignment between two inertial platforms. The nications Earth terminals, Naval Publications and Form Center Std.,
November 2018.
developed algorithm is an evolution of pre-existing algo- [21] J. Laine, “2D model-based step-track procedure,” IEEE Transactions
rithms using a linear model and quaternion representation, on Aerospace and Electronic Systems, vol. 36, no. 4, pp. 1386–1391,
supplemented with a new propagation model for large time- Oct 2000.
varying misalignment variations. Simulation results show
that the accuracy reached is sufficient for applications such
as open-loop Satcom On-The-Move pointing and inertial
stabilization.
Future research will look into experimental validation
and adaptation of the algorithm to the specific Satcom
application.

R EFERENCES
[1] Y. Yüksel, “Design and analysis of transfer aligment algorithms,”
Master’s thesis, Middle East Technical University, 2005.
[2] Y. Lu and X. Cheng, “Random misalignment and lever arm vector on-
line estimation in shipborne aircraft transfer alignment,” Measurement,
vol. 47, pp. 756–764, 2014.
[3] X. Gong and L. Chen, “A conditional cubature Kalman filter and its
application to transfer alignment of distributed position and orientation
system,” Aerospace Science and Technology, vol. 95, p. 105405, 2019.
[4] S. Chaudhuri and P. Nandi, “Transfer alignment for space vehicles
launched from a moving base,” Defence Science Journal, vol. 55, no. 3,
p. 245, 2005.