This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2021.3085742, IEEE Sensors
Journal
IEEE SENSORS JOURNAL, VOL. XX, NO. XX, XXXX 2017 1

An Improved Initial Alignment Method for

SINS/GPS Integration with Vectors Subtraction
Yiqing Yao, Member, IEEE, Xiang Xu, Tao Zhang, and Gaoge Hu

Abstract— SINS/GPS integrated navigation system has been used widely in many
application situations. In the SINS/GPS integrated navigation system, the initial
alignment process is carried out to determine the initial attitude of the vehicle. Based
on the previous works, the position loci method is a robust initial alignment method.
The outliers, which are contained in the GPS measurements, can be suppressed by
the position loci. However, the initial velocity is used to construct the observation
vectors in the previous works. When the errors are contained in the initial velocity
measurements, the initial alignment results are bad. To address this problem, the
vector subtraction operation is adopted in this paper. Then, the initial velocity is
offset by this operation. Thus, the interferences, which are caused by the initial
velocity errors, are suppressed. The simulation and field tests are designed to verify
the performance of the proposed method. It can be found that the alignment results
are similar with the methods, which contains small or no initial velocity error.
Index Terms— Vectors subtraction, in-motion coarse alignment method, observation vector, position Loci, GPS/SINS
integration.

I. I NTRODUCTION fine alignment stage [13], [14]. At the fine alignment stage, the
Kalman filtering methods are used to estimate the misalign-
T HE strapdown inertial navigation system is a self-
contained navigation system [1]. It has been applied
to many fields, such as the attitude determination for ships
ment angles, which are usually small angles [15], [16]. To get
the small misalignment angles, the coarse alignment methods
and satellites, positioning system for vehicles [2], [3]. In are investigated. In [13], the static coarse alignment method
these applications, SINS/GPS integration is the most popular was implemented with the measurements of the gyroscopes
application. The SINS/GPS integrated navigation system can and accelerometers. The Earth rate and the gravity vectors are
suppress the cumulative errors of the SINS and improve the used to determine the initial attitude of the SINS. However,
positioning accuracy of the GPS [4]. Thus, many researchers this method need the gyroscopes to measure the Earth rate.
are devoted to improve the performance of SINS/GPS inte- Thus, it can only be finished in the static situation. To extend
grated navigation system [5]– [7]. In these research hotspots, the application of the coarse alignment method, a new method
the initial alignment technique is the one of the important with the chain rule of the Direction Cosine Matrix (DCM) was
research directions [8]– [12]. investigated in [8], [9]. By the chain rule of the DCM, the
In general, the initial alignment can be divided into two varying attitude is isolated, and the initial alignment process
stages: one is the coarse alignment stage, the other one is the can be transformed as an attitude determination problem. Thus,
the optimized-based algorithm (OBA) was devised to solve this
This work was supported in part by the National Natural Sci- problem [10]. These methods implement the initial alignment
ence Foundation of China under Grants 62003085, 61803278, and
42004021, in part by the National Equipment Pre-Research Foundation process in the swaying situations.
of China under Grant 61405170207, in part by the Natural Science To further improve the performance of the initial alignment
Foundation of Jiangsu Province, China under Grant BK20190344, in methods, a velocity integration formula for the in-flight coarse
part by the Zhishan Youth Scholar Program of Southeast University
under Grant 2242021R41185, in part by the Natural Science Basic alignment was proposed in [17]. The GPS velocity is used to
Research Plan in Shanxi Province of China under Grant 2020JQ- construct the observation vectors. Then, the attitude determi-
234.(Corresponding author: Xiang Xu.) nation method with OBA was designed. The alignment results
Yiqing Yao is with the Key Laboratory of Micro-Inertial Instrument
and Advanced Navigation Technology, Ministry of Education, Southeast are shown that the proposed method can be applicable to the
University, Nanjing 210096, China (e-mail: [email protected]). in-flight situations. Based on the velocity integration formula,
Xiang Xu is with the School of Electronic and Information many in-motion alignment methods were investigated in other
Engineering, Soochow University, Suzhou 215006, China (e-mail:
[email protected]). applications, such as SINS/OD integration and SINS/DVL
Tao Zhang is with the Key Laboratory of Micro-Inertial Instrument integration [18]– [20]. Although these methods can be carried
and Advanced Navigation Technology, Ministry of Education, Southeast out in the in-motion situations, it is hard to suppress the
University, Nanjing 210096, China (e-mail: [email protected]).
Gaoge Hu is with School of Automation, Northwestern Polytechnical outliers of the GPS outputs. Thus, a position loci method
University, Xi’an 710129, China (e-mail: [email protected]). was proposed in [21], [22]. Taking the advantage of the

1530-437X (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: California State University Fresno. Downloaded on July 01,2021 at 04:22:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2021.3085742, IEEE Sensors
Journal
2 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, XXXX 2017

position loci, the outliers of the GPS outputs are suppressed, b0 denotes the boy frame at initial alignment start up, and it
the in-motion alignment method is robustness. To improve the is also a non-rotating frame during the whole initial alignment
robustness of the in-motion alignment method, the magnitude process.
n
matching method is investigated in [23]. Almost all outliers of In (1), the DCMs Cn0 and Cbb0 can be calculated by:
the GPS outputs can be detected by the magnitude matching  n0
method. Thus, the robustness of the in-motion alignment Ċn = Cnn0 [ωin n
×]
b0 b0 b (2)
method with SINS/GPS integration is improved. However, the Ċb = Cb [ωib ×]
magnitude matching method is need to calculate the modulus where, ωin n
can be calculated by the position and velocity of
of the observed and reference vectors. Moreover, the robust the vehicle. And, ωib b
can be instead with the outputs of the
attitude determination method needs to suppress the outliers gyroscopes.
of the GPS outputs. By contrast, although the accuracy of Based on the aforementioned analysis, if the DCM Cb0 n0
can
outliers’ detection of the position loci method is weak, but it be determined, the initial alignment process can be finished. In
is simpler to implement the robust alignment process. Thus, [21], a vector observation, which is constructed by the position
the position loci method is a low-cost and effective method. loci, is used to calculate the DCM Cb0n0
. The vector observation
However, due to the initial velocity in the observation vectors is given by:
of the position loci method, the initial velocity error is not
focused on. Moreover, the initial velocity error can not be n0
βp = Cb0 αp (3)
suppressed by the position loci. Thus, if there existing the
initial velocity errors, these errors will degrade the alignment where,
accuracy during the whole alignment process.  R t R τ b0 b
αp = 0 0 Cb(σ) f dσdτ
To address the defects of the position loci method, a vector



 Z t
subtraction method is devised in this paper. By constructing n0 n n
) ṙ dτ − tv (0)



 βp = Cn(τ
the new vector with the vector subtraction operation, the initial 0


Z tZ τ
velocity is eliminated, so the initial velocity errors will not de- n0
(4)
 − Cn(σ) g n dσdτ
grade the alignment performance with SINS/GPS integration.



 0 0
The rests of the paper are organized as follows: in section

 Z tZ τ
n0 n
×] ṙ n dσdτ

+ Cn(σ) [ωie

II, the previous related works and the problem statements

0 0
are introduced, the interferences of the initial velocity errors
are analyzed in detail. Then, the new observation vectors are Based on the vectors observation, some attitude determina-
n0
constructed with the vector subtraction operation in section tion method can be used to extract the DCM Cb0 .
III. Moreover, the relationship of the new observation and
reference vectors are given. In section IV, the simulation B. Problem Statement
and field tests are designed to verify the performance of the
In (4), it can be found that the reference vector αp can
proposed method. At last, a conclusion is given in section V.
be calculated by the outputs of the accelerometers and gyro-
scopes. And the observation vector βp can be calculated by the
II. R ELATED W ORKS AND P ROBLEMS S TATEMENTS outputs of the GPS. Using the position loci, the interferences
The SINS/GPS integration is one of the most research of the outliers, which are contained in the GPS outputs, can be
topics for the integrated navigation system [4]– [7]. Many key suppressed. This advantage had been verified in [21]. However,
techniques of the SINS/GPS integration have been focused on, it can be found that the initial velocity v n (0) is contained
such as the initial alignment, high precision navigation, and in the observation vector. And its weight in the observation
Kalman filtering. In these techniques, the initial alignment is vector is proportional to the alignment time. In the alignment
one of the most key techniques. Thus, many researchers are process, the initial velocity is using the outputs of the GPS. If
devoted to investigate the new initial alignment method. In the GPS outputs contain the outliers at the alignment start up,
[21], [22], a robust initial alignment method is proposed by the error of the initial velocity is existing. And this error can
the position loci. In this section, brief review of this method not be suppressed by the position loci. The initial alignment
is given, and the existing problems of the previous method is accuracy will degrade.
investigated. Based on the aforementioned analysis, an improved method
using the vector subtraction is proposed in this paper. The new
A. Related Works observation vectors are constructed where the initial velocity
are offset. Thus, the errors of the initial velocity can not
In [21], the DCM Cbn can be divided as:
degrade the alignment accuracy.
Cbn = Cn0
n n0 b0
Cb0 Cb (1)
where, n denotes the navigational frame, it is the East-North- III. T HE N EW O BSERVATION V ECTORS WITH V ECTORS
Up (ENU) frame in this paper; b denotes the body frame, S UBTRACTION
it is the Right-Forward-Up (RFU) frame; n0 denotes the In this section, the improved method is constructed, the new
navigational frame at initial alignment start up, and it is a observation vector based on the vector subtraction operation
non-rotating frame during the whole initial alignment process. is introduced, then the initial velocity error is compensated.

1530-437X (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: California State University Fresno. Downloaded on July 01,2021 at 04:22:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2021.3085742, IEEE Sensors
Journal
XU et al.: AN IMPROVED INITIAL ALIGNMENT METHOD FOR SINS/GPS INTEGRATION WITH VECTORS SUBTRACTION 3

Based on (4), the vector observation with the discrete forms where,
can be given by: M
!
2 X
α̂p (M ) = αp (M ) − αp (i) (12)
M + 1 i=1
M −1 k−1 Z tm+1
b(t )
X X
b0
αp (M ) = ∆tG Cb(tm)
Cb(σ)m f b dσ Based on the aforementioned analysis, the new vector
tm
k=0 m=0
(5) observation with the vector subtraction is constructed as (11).
M −1 Z tk+1 Z τ Since the initial velocity errors, which are contained in the
b(t )
X
b0
+ Cb(tk)
Cb(σ)k f b dσdτ GPS outputs, are offset. The new observation vectors will not
k=0 tk tk
be corrupted with the initial velocity errors. Thus, the initial
n0
DCM Cb0 can be extracted more accuracy than the previous
M
X −1 method.
n0
βp (M ) ≈Cn(tM)
r n − tv n (0) + ∆tG
k=0 IV. S IMULATION AND F IELD T ESTS
k−1
∆t2G n
 
×
X
n0
Cn(t ∆t I + [ω ×] gn (6) In this section, the simulation and field tests are designed
m) G in
m=0
2 to verify the performance of the proposed improved method.
M −1  2 Five methods are designed for comparison, they are:
∆t3G n

X
n0 ∆tG Scheme 1: The previous method in [17] with the initial
+ Cn(tk ) I+ [ωin ×] g n
2 6 velocity error of 5m/s.
k=0
Scheme 2: The previous method in [21] with the initial
where, t = M ∆tG . velocity error of 5m/s;
Assuming the initial velocity error is contained in the GPS Scheme 3: The previous method in [21] with the initial
outputs, it has: velocity error of 1m/s;
ṽ n (0) = v n (0) + δv0n (7) Scheme 4: The previous method in [21] with the initial
velocity error of 0.1m/s;
where, δv0n denotes the initial velocity errors of the GPS
Scheme 5: The previous method in [21] with no initial
outputs.
velocity error;
Then, (6) can be rewritten as:
Scheme 6: The proposed improved method with the initial
β̃p (M ) = βp (M ) − M ∆tG δv0n (8) velocity error of 5m/s.
The alignment results are shown in the next subsections.
where, β̃p (M ) is calculated with the GPS outputs.
Define a new observation vectors: A. Simulation Test
M In this subsection, a simulation test is designed for verifying.
2 X
β p (M ) = β̃p (i) (9) From Fig.1, a Zigzag in-motion trajectory is designed. The in-
M +1 i=1 motion status of the vehicle is shown in Fig.2. The start point
Subtracting (9) from (8), it has: of the vehicle is set as latitude 32.057313◦ N and longitude
118.786365◦ E. The bias drifts and angular random walk of the
β̂p (M ) = β̃p (M ) − β p (M ) gyroscopes√are set as [0.01 0.01 0.01]T◦ /h and [0.005 0.005
= βp (M ) − M ∆tG δv0n − 0.005]T◦ / h. The bias drifts and the velocity random √ walk
M are set as [100 100 100]T µg and [50 50 50]T µg/ Hz. The
2 X outputs of the GPS, which are adding Gaussian White noises
(βp (i) − i∆tG δv0n )
M + 1 i=1 and outliers, are using as the external auxiliary information.
M The sampling rate of the GPS is set as 1Hz. The position errors
2 X
= βp (M ) − βp (i)− (10) are set as δpn v N (0, (5m)2 I3 ).
M + 1 i=1

N 0, (5 m)2 I3
w.p. 0.97

n
δp = (13)
M
!
2 X N 0, (50 m)2 I3 w.p. 0.03
M− i ∆tG δv0n
M + 1 i=1
The alignment results are shown in Figs. 3 to 5. In Fig. 3, the
M pitch errors are shown. From Scheme 4, it can be found that the
2 X
= βp (M ) − βp (i) outliers of the GPS outputs can not degrade the performance
M +1 i=1 of the previous alignment method, which is proposed in [21].
Comparing (10) with (8), it can be found that the initial This is becasue the interferences of the outliers are suppressed
velocity errors δv0n have been offset. Thus, the reconstructed by the position loci. The same conclusions can be found in
observation vectors β̂p (M ) will not be corrupted by the initial [21]. However, the performance of the previous alignment
velocity errors. methods is degrade by the initial velocity error, which are
From (3), with the constant DCM Cb0 n0
, it has: shown as Scheme 1 to 3. It can be found that when the initial
velocity error is 5m/s, the alignment pitch errors of Scheme
n0
β̂p (M ) = Cb0 α̂p (M ) (11) 1 and Scheme 2 are around −0.05◦ and −0.2◦ , respectively.

1530-437X (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: California State University Fresno. Downloaded on July 01,2021 at 04:22:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2021.3085742, IEEE Sensors
Journal
4 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, XXXX 2017

32.09 50
GPS outputs 0.2 0.1 Scheme 1
32.08 Terminal Point The ture trajectory Scheme 2
40 Scheme 3
32.07 0.05 Scheme 4
0 Scheme 5

Pitch Errors (deg)

32.06 30
Lattitude (deg)

Scheme 6
0
32.05
20 -0.2
32.04 -0.05

32.03 10
-0.4 -0.1
32.02 0 500 0 500
Start Point 0
32.01

32 -10
117.96 117.97 117.98 117.99 118 118.01 0 100 200 300 400 500 600
Longitude (deg) Time (s)

Fig. 1. The in-motion trajectory. Fig. 3. The pitch errors.

5 40 10
Pitch East
Tilts (deg)

Roll 30 North
0
0 Up Scheme 1
0.2 0.04 Scheme 2
20
Roll Errors (deg) -10 Scheme 3
Velocity (m/s)

Scheme 4
-5 0.1 0.02
0 200 400 600 10 Scheme 5
-20 Scheme 6
0 0 0
100
-30
Yaw (deg)

-10 -0.1 -0.02

50
-40
-20 -0.04
-0.2
0 500 0 500
0 -30 -50
0 200 400 600 0 200 400 600 0 100 200 300 400 500 600
Time (s) Time (s) Time (s)

Fig. 2. The in-motion status of the vehicle Fig. 4. The roll errors.

When the initial velocity error is 1m/s, the alignment error of close to Schemes 5, which is no initial velocity error. The
pitch of Scheme 3 is around −0.05◦ . Comparing with Scheme proposed method, which is shown as Schemes 6, has a large
2 and Scheme 3, Scheme 4 has obtain the more accuracy results errors between 0s to 200s. This is because the new observation
because the initial velocity error is 0.1m/s, this can be taken vectors are constructed by the vector subtraction, this operation
as the normal outputs of the GPS. It is noted that the proposed weakens the effects of the position loci. It is noted that when
method named Scheme 6 , which is set the initial velocity error the initial alignment process lasts to 250s, the alignment errors
of 5m/s, gets the similar results with Scheme 4 and Scheme 5. of Schemes 6 are close to Schemes 5, which is not corrupted
These reslts show that the initial velocity error can not degrade by the initial velocity error.
the performance of the proposed method.
In Fig.4, the roll errors of the six schemes are shown. It B. Field Test
can be found that the alignment errors of Scheme 1 and 2 are In the previous subsection, the simulation test is designed
around −0.1◦ . The alignment errors of Scheme 3 are around to validate the performance of the proposed method. The test
−0.03◦ . The alignment errors of Schemes 4 to 6 have the results show that the proposed method can suppress the initial
similar value, which are less than 0.01◦ . velocity errors effectively. In this subsection, the field test
Fig.5 shows the yaw errors of the six schemes. It can be is designed for further verification. The experimental vehicle
found that the yaw errors of Scheme 1 and Scheme 2 are larger and test equipment are shown in Fig. 6. It can be found that
than 5◦ when the initial alignment process lasts for 600s. the test equipment contains the navigational computer, IMU
Moreover, due to the initial velocity error, the convergence and PHINS, GPS and monitoring computer. The self-made
rate of Scheme 1 and 2 is slow. In Scheme 3, the alignment navigational computer is used to implement the algorithm. The
errors of yaw are less than Scheme 1 and 2, but they are PHINS, which is made by the iXblue corporation, is used as
still larger than Schemes 4 to 6. Since the initial velocity of a reference system. It can outputs the attitude, velocity and
Schemes 4 is 0.1m/s, the alignment errors of Schemes 4 are position of the vehicle. The pitch and roll errors of PHINS

1530-437X (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: California State University Fresno. Downloaded on July 01,2021 at 04:22:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2021.3085742, IEEE Sensors
Journal
XU et al.: AN IMPROVED INITIAL ALIGNMENT METHOD FOR SINS/GPS INTEGRATION WITH VECTORS SUBTRACTION 5

60
Scheme 1
Scheme 2
40 Scheme 3
Scheme 4
Scheme 5
Yaw Errors (deg)

20 Scheme 6

1
-20

0
-40
-1
200 300 400 500 600
-60
0 100 200 300 400 500 600
Time (s)

Fig. 5. The yaw errors.

Fig. 7. The in-motion trajectory of the vehicle.

are less than 0.01◦ , the yaw error of PHINS is less than
0.1◦ sec L(L is latitude of the vehicle) . The GPS, which is 5 40
made by NovAtel, can outputs the position and velocity of Pitch East

Tilts (deg)
Roll 30 North
the vehicle. The IMU is equipped with a triaxial fiber optical Up
0
gyroscopes and triaxial quartz accelerometers. The bias drift
20
of the gyroscope is about 0.02◦ /h,√the angular random walk

Velocity (m/s)
of the gyroscope is about 0.005◦ / h. The bias drift of the -5 10
0 200 400 600
accelerometer is about 500µg, and the √ velocity random walk
of the accelerometer is about 50µg/ Hz. The sampling rate 50 0
of the IMU is set as 200Hz. The GPS sampling rate is 1Hz.
Yaw (deg)

The in-motion trajectory of the vehicle is shown in Fig. 7. 0 -10

The in-motion status of the vehicle is shown in Fig. 8. The
-50 -20
whole alignment time lasts 600s. The average velocity of the
vehicle is about 20m/s(72km/h). The outliers of the GPS can -100 -30
be found in Fig. 8. The initial alignment process is start at any 0 200 400 600 0 200 400 600
Time (s) Time (s)
time during the vehicle driving. It is noted that the outliers will
generate the initial velocity errors. To verify the performance Fig. 8. The in-motion status of the vehicle
of the proposed method, the initial velocity errors is added by
man made. The alignment errors are shown in Figs. 9 to 11.
30
1 0.2 Scheme 1
25 Scheme 2
0.5 0.1 Scheme 3
20 Scheme 4
Scheme 5
Pitch Errors (deg)

0 0
Scheme 6
15
-0.5 -0.1
10
-1 -0.2
5 0 500 0 500

-5

-10
0 100 200 300 400 500 600
Fig. 6. The experimental vehicle and test equipment. Time (s)

In Fig. 9, the pitch errors are shown. It can be found Fig. 9. The pitch errors.
that the errors of Schemes 1 and 2 are less than −0.1◦ and
−0.5◦ , respectively. And the errors of Schemes 3 is about 0.1◦
when the alignment process finished. Moreover, the alignment of Schemes 4 to 6 are close to each other. In Schemes 4, the
results are fluctunt, this is because the initial velocity errors of initial velocity error is 0.1m/s, it is small enough. And there is
Schemes 2 and 3 are 5m/s and 1m/s, respectively. The errors no initial velocity error in Schemes 5. In Schemes 6, the initial

1530-437X (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: California State University Fresno. Downloaded on July 01,2021 at 04:22:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2021.3085742, IEEE Sensors
Journal
6 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, XXXX 2017

velocity error is set as 5m/s. However, the initial velocity errors velocity errors decrease, the yaw errors are decrease, which
does not corrupt the initial alignment results. This conclusion is shown as Schemes 3. The initial velocity errors are 1m/s
shows that the proposed method, which is Schemes 6, can in Schemes 3, and its yaw errors are less than 5◦ . However,
suppress the initial velocity error. It is consistent with the it is still a large error. As the initial velocity errors decrease
simulation test. further, the yaw errors continue to decrease. It is shown that
the yaw errors of Schemes 4, which is contained the initial
10
velocity errors with 0.1m/s, is close to Schemes 5, which is
no initial velocity errors. It is noted that the yaw errors of
0
Schemes 6, which is contained the initial velocity errors with
Scheme 1 5m/s, is also close to Schemes 5. This conclusion verifies that
Scheme 2 the proposed method can supress the initial velocity errors.
Roll Errors (deg)

-10 Scheme 3
1 0.2
Scheme 4

-20 0.5 0.1

Scheme 5 V. C ONCLUSION
Scheme 6
In this paper, a initial alignment method for SINS/GPS
0 0 integration with the vector subtraction is proposed. Firstly,
-30
the related work of the previous method is reviewed, and the
-0.5 -0.1
-40 existing problem of the previous work is investigated. Then,
-1 -0.2 to address this problem, which is the initial velocity errors,
0 500 0 500 the new observation vectors with the vectors subtraction is
-50
0 100 200 300 400 500 600
constructed. By the vectors subtraction, the initial velocity
Time (s)
errors are offset. Thus, the interferences, which are caused by
Fig. 10. The roll errors. the initial velocity errors, are eliminated. At last, the simulation
and field tests are designed to verify the performance of the
Fig.10 is the roll errors of six schemes. It can be found proposed method. The tests results show that the proposed
that the alignment errors of Schemes 1 and 2 are fluctunt, method can suppress the initial velocity errors effectively.
their errors are about −0.1◦ when the initial alignment lasts Moreover, the proposed method can be used to improve the
for 600s. The errors of Schemes 3 is close to the other three other initial alignment methods, which are based on the vector
methods, which are Schemes 4 to 6, when the alignment observations.
process lasts for 300s. However, the convergence rate of
Schemes 3 is slow. Its error is less than 0.1◦ when the ACKNOWLEDGMENT
initial alignment process lasts for 250s, while the other three The authours would like to thank Xiaosu Xu from School
methods, which are Schemes 4 to 6, only take 200s. Schemes 6 of Instrument Science and Engineering, Southeast University,
has the similar error to Schemes 4 and 5. However, the initial China, for providing the field tests.
velocity errors of Schemes 6 is 5m/s, while the initial velocity
errors of Schemes 4 and 5 are 0.1m/s and 0m/s, respectively. R EFERENCES
[1] Y. Qin, Inertial Navigation. The Science Publishing Company, 2nd ed.,
2014.
50 [2] P. D. Groves, Principles of GNSS, inertial, and multisensor integrated
navigation systems. Artech house, 2013.
[3] K. R. Britting, Inertial navigation systems analysis. John Wiley and
0
Sons, 1971.
Scheme 1 [4] L. Chang, K. Li and B. Hu, “Huber’s M-estimation-based process
Yaw Errors (deg)

Scheme 2 uncertainty robust filter for integrated INS/GPS,” IEEE Sensors Journal.,
Scheme 3 vol. 15, pp. 3367-3374, 2015.
-50 Scheme 4 [5] Y. Zhang, C. Shen, J. Tang, and J. Liu, “Hybrid algorithm based on
Scheme 5 MDF-CKF and RF for GPS/INS system during GPS outages,” IEEE
Scheme 6 Access., vol. 6, pp. 35343-35354, 2018.
1
-100 [6] S. Sheikhpour, M. M. Atia and S. Waslander, “A Flexible Simulation
and Design Environment for IMU/GNSS Sensors Integration,” in 2018
0 IEEE 61st International Midwest Symposium on Circuits and Systems
(MWSCAS), 2018, pp. 472-475.
-150
-1 [7] Z. Wu, M. Yao, H. Ma, and W. Jia, “Improving accuracy of the
0 100 200 300 400 500 600 vehicle attitude estimation for low-cost INS/GPS integration aided by
0 100 200 300 400 500 600 the GPS-measured course angle,” IEEE Transactions on Intelligent
Time (s) Transportation Systems, vol. 14, pp. 553-564, 2013.
[8] Y. Qin, G. Yan, D. Gu, J. Zheng, “A Clever Way of SINS Coarse
Fig. 11. The yaw errors. Alignment despite Rocking Ship,” Journal of Northwestern Polytech-
nical University, vol. 23, pp. 681-684, 2005.
[9] X. Xu, X. Xu, T. Zhang, Y. Li, F, Zhou, “Improved Kalman filter for
Fig. 10 shows the alignment errors of the yaw. It can be SINS coarse alignment based on parameter identification,” Journal of
found that the errors of Schemes 1 and 2 are larger than 10◦ Chinese Inertial Technology, vol. 24, pp. 320-324,329, 2016.
[10] M. Wu, Y. Wu, X. Hu, and D. Hu, “Optimization-based alignment for
during the whole alignment process. This is because the initial inertial navigation systems: Theory and algorithm,” Aerospace Science
velocity errors of Schemes 1 and 2 are 5m/s. As the initial and Technology, vol. 15, pp. 1-17, 2011.

1530-437X (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: California State University Fresno. Downloaded on July 01,2021 at 04:22:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2021.3085742, IEEE Sensors
Journal
XU et al.: AN IMPROVED INITIAL ALIGNMENT METHOD FOR SINS/GPS INTEGRATION WITH VECTORS SUBTRACTION 7

[11] J. Fang, S. Yang, “Study on innovation adaptive EKF for in-flight Tao Zhang received the Ph.D. degree in nav-
alignment of airborne POS,” IEEE Transactions on Instrumentation and igation, guidance, and control from Southeast
Measurement, vol. 60, pp. 1378-1388, 2011. University, Nanjing, China, in 2008. He is cur-
[12] J. Li, J. Fang and M. Du, “Error Analysis and Gyro-Bias Calibration of rently an Associate Professor with the School of
Analytic Coarse Alignment for Airborne POS,” IEEE Transactions on Instrument Science and Engineering, Southeast
Instrumentation and Measurement, vol. 61, pp. 3058-3064, 2012. University.
[13] Y. F. Jiang and Y. P. Lin, “Error estimation of INS ground alignment His current research interests include the
through observability analysis,” IEEE Transactions on Aerospace and ship’s acoustic navigation system, long base
Electronic systems, vol. 28, pp. 92-97, 1992. line, ultra-short base line, inertial navigation, in-
[14] I. Y. Bar-Itzhack and N. Berman, “Control theoretic approach to inertial tegrated navigation, and MEMS inertial technol-
navigation systems,” Journal of Guidance, Control, and Dynamics, vol. ogy.
11, pp. 237-245, 1988.
[15] X. Xu, J. Lu and T. Zhang, “A Fast-Initial Alignment Method With
Angular Rate Aiding Based on Robust Kalman Filter,” IEEE Access,
vol. 7, pp. 51369-51378, 2019-01-01 2019.
[16] G. Chang, “Fast two-position initial alignment for SINS using velocity
plus angular rate measurements,” Advances in Space Research, vol. 56,
pp. 1331-1342, 2015.
[17] Y. Wu and X. Pan, “Velocity/Position integration formula part I: Appli-
cation to in-flight coarse alignment,” IEEE Transactions on Aerospace
and Electronic Systems, vol. 49, pp. 1006-1023, 2013.
[18] X. Xu, Z. Guo, Y. Yao, and T. Zhang, “Robust Initial Alignment for
SINS/DVL Based on Reconstructed Observation Vectors,” IEEE/ASME
Transactions on Mechatronics, vol. 25, pp. 1659-1667, 2020-01-01 2020.
[19] Y. Yao, X. Xu, Y. Zhu, and X. Xu, “In-motion coarse alignment method
for SINS/DVL with the attitude dynamics,” ISA Transactions, vol. 105,
pp. 377-386, 2020-01-01 2020.
[20] Y. Huang, Y. Zhang and X. Wang, “Kalman-Filtering-Based In-Motion
Coarse Alignment for Odometer-Aided SINS,” IEEE Transactions on
Instrumentation and Measurement, vol. 66, pp. 3364-3377, 2017.
[21] X. Xu, D. Xu, T. Zhang, and H. Zhao, “In-Motion Coarse Alignment
Method for SINS/GPS Using Position Loci,” IEEE Sensors Journal, vol.
19, pp. 3930-3938, 2019-01-01 2019.
[22] L. Luo, Y. Huang, Z. Zhang, and Y. Zhang, “A Position Loci-Based In-
Motion Initial Alignment Method for Low-Cost Attitude and Heading
Reference System,” IEEE Transactions on Instrumentation and Mea-
surement, vol. 70, pp. 1-18, 2020.
[23] X. Xu, Y. Sun, Y. Yao, and T. Zhang, “A Robust In-Motion Optimization-
Based Alignment for SINS/GPS Integration,” IEEE Transactions on
Intelligent Transportation Systems, 2021.
Gaoge Hu received the B.S. degree in math-
ematics and applied mathematics from North-
west Polytechnical University, Xi’an, China, in
2010, and the Ph.D. degree in control theory and
control engineering from Northwest Polytechni-
cal University in 2016. From 2016 to 2017, he
Yiqing Yao received the Ph.D degree in Nav- worked as a system engineer with the Huawei
igation, Guidance, and Control from Southeast Technologies Co Ltd. At present, he is an Asso-
University, Nanjing, China, in 2018. ciate Professor with the School of Automation,
She is currently working in the School of In- Northwest Polytechnical University.
strument Science and Engineering, Southeast His current research interests include inte-
University. Her research interests include vehicle grated navigation, Kalman filtering, nonlinear state estimation and multi-
positioning, inertial navigation and its integration sensors data fusion.
with other navigation systems.

Xiang Xu was born in Nantong, Jiangsu, in

1988. He received the M.S. degree in Naviga-
tion, Guidance and Control from Harbin Engi-
neering University, Harbin, China, in 2014, and
the Ph.D. degree in Instrument Science and
Technology from Southeast University, Nanjing,
China, in 2018.
He is currently a lecturer with School of Elec-
tronic and Information Engineering, Soochow
University, Suzhou, China. His research inter-
ests include initial alignment for inertial naviga-
tion, integrated navigation for an autonomous underwater vehicle, atti-
tude estimation for low-cost inertial measurement unit and information
fusion.

1530-437X (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: California State University Fresno. Downloaded on July 01,2021 at 04:22:45 UTC from IEEE Xplore. Restrictions apply.