A DOUBLE-FREQUENCY COMBINED POSITIONING ALGORITHM OF

BEIDOU NAVIGATION SATELLITE SYSTEM

YUWEI ZHOU1, XIAOLING CHEN1, XUCHU MAO1
1
Department of Instrument Science and Engineering, Shanghai Jiao Tong University, Shanghai 200240, China
E-MAIL: [email protected], [email protected], [email protected]

Abstract: positioning, including clock error, ionospheric error,

With the second civilian signal B2I announced, Beidou tropospheric error and measurement random error.
Navigation Satellite System has been into the era of multiple Parameters of the first three errors can be got from the
frequency application. In order to improve the accuracy of BDS navigation message and the pseudorange will be
positioning, an estimation algorithm of BDS positioning based compensated through the calculation of these parameters.
on Particle Swarm Optimization and Square-Root Unscented
Besides, a corresponding filtering algorithm is needed to
Kalman Filter is proposed, which could estimate the noise
covariance matrix accurately and compensate defects of eliminate the measurement random error. The commonly
Extended Kalman Filter and Unscented Kalman Filter. Based used filtering algorithms are Extended Kalman Filter (EKF)
on above, a new algorithm of BDS double-frequency combined [2] and Unscented Kalman Filter (UKF) [3]. However, the
positioning is proposed. Experimental results show that the defects of the EKF and UKF algorithms are obvious [4-5].
proposed method can effectively improve the accuracy and EKF is a result of local linearization of nonlinear function,
stability of Beidou System positioning. which makes large errors in estimation results. And UKF
might cause the divergence of the Kalman filter due to the
Keywords: loss of positive definiteness of the state parameters
Square-Root Unscented Kalman Filter; Particle Swarm covariance matrix in practice. In this paper, estimation
Optimization; Double-frequency combined positioning; Beidou algorithm named PSO_SRUKF, based on Particle Swarm
System
Optimization (PSO) [6] and Square-Root Unscented Kalman
Filter (SRUKF) [7] is applied to BDS positioning. The noise
1. Introduction covariance matrix is estimated through PSO algorithm. And
in the process of filter iteration, the positive definiteness is
Beidou Navigation Satellite System (BDS) is the third ensured by using square root of covariance matrix instead of
mature global navigation satellite system (GNSS) after GPS covariance matrix, which will improve the accuracy and
of the United States and GLONASS of Russia, established stability of BDS positioning.
and managed by China independently. The capable The biggest highlight of “Beidou System Space Signal
navigation service area has covered the Asia-Pacific region, Interface Control File (version II)” released in December
and will cover the global by 2020. According to the results 2013 is to publish the signal system of the second civilian
of the detection, the accuracy of BDS positioning has signal B2I. At this point, BDS has been into the era of
reached 10 meters in horizontal, 15 meters in elevation, 0.2 multiple frequency application. Users can simultaneously
m/s in speed and the timing accuracy is 10 ns. By now, there receive B1I and B2I signals to position, which means BDS
are 20 satellites working on orbit, including six geostationary has become the first GNSS that has two available civilian
orbit satellites (GEO), six middle earth orbiting satellites signals in use. In this paper, the PSO_SRUKF algorithm is
(MEO) and eight inclined geo-synchronization orbit used to solve the position with the combined signals of B1I
satellites (IGSO) [1]. BDS plays an important role in civil and B2I. The experimental results show that the double-
and military affairs and related products have been used in frequency combined positioning (DFCP) algorithm can
marine fishery, power system, transportation and many other further improve the accuracy of the BDS positioning.
fields, bringing enormous social and economic benefits. Section 2 introduces a new estimation positioning
Because of the vital function of BDS, how to further enhance algorithm (PSO_SRUKF) based on PSO and SRUKF.
the advantages of BDS and improve the accuracy of BDS Section 3 introduces DFCP algorithm based on the B1I/B2I
positioning has become the research hotspot. double-frequency combined pseudorange eliminating
As one of the GNSSs, there are multiple errors in BDS

978-1-5386-1010-7/17/$31.00 ©2017 IEEE 9

ionospheric error equation. Section 4 describes the means Cholesky first-order factor update, Q is the
positioning model of BDS including process model and measurement noise covariance in the measurement model.
measurement model. Section 5 analyses the single-frequency z Measurement update process

^ `
positioning experiment and the DFCP experiment. Finally,
section 6 concludes the paper and introduces the future work. S kZ qr ª W1c Z1:2 n , k |k 1  Z
ˆ º R
¬ k
¼

2. PSO_SRUKF Algorithm S kZ ^ ˆ  , Wc
cholupdate S kZ , W0c Z 0, k |k 1  Z k 0 `
2n T
2.1. SRUKF Algorithm Pxz ¦W
i 0
i
c
¬
ˆ  º ªZ
ª χ Ji , k |k 1  X ˆº
k ¼ ¬ i , k | k 1  Z k ¼

The general process of SRUKF is as follow: T

(5)
Kk Pxz / S kZ / S kZ
z Initialization.
xˆ 0 E x 0 ˆ
X ˆ  K z Z
X ˆ
k k k k k
(1)
chol E ª x0  xˆ 0 º
T
S0 x0  xˆ 0 U K k S kZ
¬ ¼
where x̂ 0 is the expectation of initial state, chol ^ ` means Sk cholupdate ^S k , U, 1`
Cholesky factorization. where R is the process noise covariance in the state model.
z Calculate 2n  1 Sigma points χ kD1 and their weights It can be seen from the above algorithm that the square
root of state parameters covariance matrix St|t instead of
Wi m , Wi c .
covariance matrix Pt|t is used in the process of filter iteration,
γ nO
i.e. Pt|t St|t St|tT . From the matrix theory, the product of any
λ D 2 n W  n
nonzero matrix S and its transposed matrix ST is
χ Dk 1 ª¬ xˆ Dk 1 xˆ Dk 1  J S k 1 xˆ Dk 1  J S k 1 º¼
(2) nonnegative definite matrix. If only the square roots of Pt|t
W0m O / nO and Pt |t -1 are calculated in the process of filtering, then the
W c
0 O / n  O  1 D  E 2
positive definiteness of Pt|t and Pt |t -1 is ensured. Therefore,
Wim 1/ ^2 n  O `
Wic the divergence of the Kalman filter due to the loss of positive
definiteness of the state parameters covariance matrix can be
where λ is the scaling factor, n is the dimension of the state solved effectively in SRUKF algorithm, which can further
vector, D determines the dispersion of the Sigma points, improve the accuracy and stability of the BDS positioning
usually takes a small positive value, W usually takes 0 , E
is used to describe the distribution of x , and the optimal 2.2. PSO Algorithm
value of E is 2 in Gaussian condition.
z Time update process PSO was proposed by Kennedy and Eberhart in 1995.
The idea of particle swarm is derived from the study and
χ Jk |k 1 f χ i , k 1 i 0,1,}, 2n
behavior simulation of the simplified social model of bird
2n
flock [6]. In PSO, it is assumed that each particle is flying at
ˆ
X k ¦W i 0
i
m
χ Ji , k |k 1
a certain speed in the n-dimensional search space,
(3)
^ `
Xi ( xi1 , xi 2 ,..., xin ) is the current position of particle i ,
S 
qr ª W1c χ1:2
J ˆ º
n , k | k 1  X k Q
k
¬ ¼ Vi (vi1 , vi 2 ,..., vin ) is the current velocity of particle i , and
S k ^ ˆ  , Wc
cholupdate S k , W0c χ J0, k |k 1  X k 0 ` pbesti ( pbesti1 , pbesti 2 ,..., pbestin ) is the optimal position
of particle i , which is called individual optimal position.
Z k |k 1 h χ Jk |k 1 Set f ( X) as the minimization of the objective function,
2n (4) so the individual optimal position of particle i is determined
ˆ
Z k ¦W
i 0
i
m
Zi , k |k 1 by:
­ pbesti (t ), f ( X i (t  1)) t f ( pbesti (t ))
where qr ^ ` means QR factorization, cholupdate ^ ` pbesti (t  1) ® (6)
¯ X i (t  1), f ( X i (t  1))  f ( pbesti (t ))

10
 Set the number of the particle is N , and gbest (t ) is Begin
the individual optimal position of all particles, which is Update the speed of
called global optimal position, then: particle i by (7)
gbest (t ) min{ f ( pbest1 (t )), f ( pbest2 (t )), Initialize
(7) Update the position of
parameters
..., f ( pbest N (t ))} particle i by (8)
The velocity and position of particles can be adjusted Calculate the fitness of
dynamically according to the individual flight experience Initialize NP particle particle i by SRUKF
and the group flight experience, so the update equation of the swarms in [R,Q] form
velocity and position is: Update pbest i by (5)
vij (t  1) wvij (t )  c1r1 ( pbestij (t )  xij (t ))  Calculate the fitness of
the particle by SRUKF
c2 r2 ( gbest j (t )  xij (t )) (8) i=i+1

xij (t  1) xij (t )  vij (t  1)

Calculate pbest and gbest Y
i İ NP
where w is the inertial weight to balance the performance of
N
the global search and local search, c1 , c2 are the acceleration
Update gbest by (6)
factor or learning factor, and r1 , r2 is a random number i=1
between 0 and 1.
Meet the
N
termination
2.3. SRUKF_PSO Algorithm condition
Y
Considering that the noise covariance of BDS
End
positioning model in the application of SRUKF need to be
set in advance, if the assumed noise covariance matrix cannot Fig.1 General process of PSO_SRUKF
reflect the real model, the reliability of SRUKF positioning
estimation will be reduced. The diversity of the environment 3. Double-Frequency Combined Positioning
makes it difficult to estimate the noise covariance matrix Algorithm
accurately, which affects the accuracy and stability of
SRUKF. To calculate the noise covariance accurately, PSO In the DFCP algorithm, the compensation of
algorithm is introduced in this paper and an estimation ionospheric error is different from that of single frequency
algorithm of BDS positioning (PSO_SRUKF) based on PSO positioning.
and SRUKF is proposed. The general process is shown in B1I and B2I signals are composed of ranging code and
Figure 1. In this algorithm, the particle expression is > R, Q @ . navigation message quadrature modulated on the carrier
R is the process noise covariance in the state model and Q from I branch and Q branch respectively. Eight ionospheric
delay correction model parameters are got from navigation
is the measurement noise covariance in the measurement
system. The ionospheric vertical delay correction of the B1I
model. Based on the process noise covariance information
and the measurement noise covariance information of each signal I Z` (t ) is calculated in seconds by eight parameters and
particle, the merits of particles could be evaluated by the root Klobuchar model. Expressions are as follows:
mean square (RMS) error of the result solved by SRUKF. ­ 9 2S (t  50400) A4
° 5 u 10  A2 cos[ ], t  50400 
The smaller the RMS error is, the higher the fitness of the ° A4 4
I Z` (t ) ® (9)
particle is, the more the noise covariance matrix represented ° A
by particles can reflect the real situation. °̄ 5 u 109 , t  50400 t 4
4
where t is the time of the intersection point (penetrate point
M) of ionosphere and the connection between the receiver
and the satellite, in the range of 0 to 86400, in seconds; A2 is
the amplitude of daytime ionospheric delay cosine curve, and
factor D n is needed to calculate A2 

11
 ­ 3 4. Positioning Model of BDS
°¦ α n IM , A2 t 0
n

A2 = ® n =0 (10)
° 4.1. Process Model
¯ 0, A2 <0
A4 is the period of the cosine curve, in seconds, and factor In the state model, it is assumed that the position
E n is needed to calculate A4 : coordinates of the user receiver is X tˈYtˈZt , the speed
­ 172800, A4 t 172800 of the receiver is X tˈYtˈZt , the clock skewing is cG t ,
° 3
°
A4 ®¦ E n IM ,172800 ! A4 t 72000 the rate of change of the clock skewing is cG t and the
n
(11)
°n 0 deviation between the rate of change of pseudorange and
°
¯ 72000, A4  72000 Doppler frequency shift is G R . In addition, the channel
The IM in the above two equations is the geographical error H n of receiver is also included in the state model. There
latitude of Penetrate Point M, in S . are errors in the signal receiving channel of each satellite,
I B1I (t ) is the ionospheric delay on the propagation path which is independent to each other. The state vector is as
follow:
of B1I signal, in seconds:
xt ª¬ X t X t Yt Yt Zt Zt cG tt cG tt G R H1t H nt º¼ (14)
1
I B1I (t ) x I Z` (t ) (12) The channel noise is non-white noise modelled by first-
2
§ R · order Gauss-Markov Process, as follows:
1 ¨ x cos E ¸
©Rh ¹ ­H1 t 1 DH1t  E wc1t
For B1I and B2I double-frequency users, B1I/B2I °
double-frequency combined pseudorange eliminating °H 2 t 1 DH 2t  E wc 2t
® (15)
ionospheric error equation can be used to correct the group °
delay caused by ionospheric effect. The equations are as °H
follows: ¯ n t 1 DH nt  E wcnt
U -k (f )UB1I c(TGD2 -k ( f )TGD1 ) 2W  1 2W
U = B2I - where D E 
1-k (f ) 1-k ( f ) 2W  1 2W  1
2
(13) Based on the general kinematics model of the position
f1 1561.098 2
k( f ) ( ) coordinates of the receiver and the relation of the system
f 22 1207.140 state, the process model is as follow:
where U is the pseudorange compensated by the ­ X t 1 X t  TXX t  T 2G X t 2
ionospheric error; U B1I is the observation pseudorange of the °
° X t 1 X t  T G X t
B1I signal compensated by the satellite time error, but hasn’t °Yt 1 Yt  TY Yt  T 2G Yt 2
compensated by TGD1 ; U B2I is the observation pseudorange °
°Yt 1 Yt  T G Yt
of the B2I signal compensated by the satellite time error, but °
hasn’t compensated by TGD2 ; TGD1 is the onboard delay Z t  T 2G Z t 2
® Z t 1 Z t  TZ (16)
°Z Zt  T G Zt
difference of B1I signal; TGD2 is the onboard delay difference ° t 1
of B2I signal; c is the speed of light, the value is 299792458 °cδtt 1 cG tt  TcG tt  cw11tt
°
m/s ; f1 is the nominal carrier frequency of B1I signal; f 2 is °cδtt 1 cG tt  cw2t
the nominal carrier frequency of B2I signal. °¯δRt 1 G Rt  wG R
In this paper, DFCP algorithm is proposed based on Equation (15) and (16) constitute the process equation
PSO_SRUKF. The main process is to get the double- of BDS positioning model, where T is the sampling period,
frequency combined pseudorange firstly by (12), B1I/B2I G X tǃG YtǃG Zt is the system noise including acceleration
double-frequency combined pseudorange eliminating
ionospheric error equation, and then, get the combined and other system interference, w1t and w2t are the driving
Doppler averaged by the Doppler of B1I and B2I signals. noise of the clock error model, wG R is the noise of G R , wci
Finally, the combined data are solved by PSO_SRUKF is the driving noise of the receiver channel i and W 100 .
algorithm to get the position. It is assumed that all the noised are non-correlated random

12
white noise with zero mean.

4.2. Measurement Model

The measurement model is constituted by pseudorange

and Doppler frequency shift.
The nonlinear pseudorange equation is as follow:
Uit Rit  cG tt  H it Q it
2 2 2
(17)
Rit X rt  X it  Yrt  Yit  Z rt  Zit
where Rit is the real distance between the receiver and the i
-th satellite, X rtˈYrtˈZ rt is the position coordinates of the
receiver at t time, X itˈYitˈZ it is the position coordinates
of the i -th satellite at t time i 1, 2 n , H it is the non-
white noise error of receiver channel i and vit is the
measurement white noise of receiver channel i .
Doppler frequency shift equation is as follow: Fig.2 Error comparison of B1I and B2I
X rt - X it X rtt - X it  Yrtt - Yit Yrtt - Yit  Z rtt - Z iit Z rtt - Z it
Dit Table 1 Error comparison of B1I and B2I
Rit (18) RMSE (m)
Signal
cG t  G Rt X Y Z 3D
where G Rt is the range difference caused by clock skewing. B1I 1.9800 5.0752 1.5272 5.6577
Assuming n satellites observed, the measurement B2I 6.3892 3.4537 2.1959 7.5876
equation combining pseudorange equation and Doppler It can be concluded form Figure 2 that the positioning
frequency shift equation is as follow: accuracy of B1I or B2I is high but slightly different in the
> U1t , U2t ,}, Unt , D1t , D2t ,}, Dnt @
T
zt (19) different direction. From Figure 2 and TABLE I, the accuracy
and stability of B1I is better than that of B2I.After several
5. Analysis of Experimental Results experiments, it is found that B1I or B2I has its own
advantages in positioning respectively.
5.1. Single-Frequency Positioning Experiment
5.2. Double-Frequency Combined Positioning
In this paper, the original data of B1I and B2I were Experiment
collected by OEM-628 from NovAtel Company. The
experiment was carried out under a good observation It is concluded above that B1I or B2I has its own
environment on the top of the SEIEE Building in Shanghai advantages in positioning respectively. The solved results of
Jiao Tong University. And the sampling frequency was 1Hz. DFCP algorithm and the output result of receiver are shown
The average number of received satellites is seven and 1400 in Figure 3. TABLE II shows RMSE of results from receiver
points were taken sample with the sampling interval of 1 and DFCP.
second. PSO_SRUKF algorithm was used to calculate the
data of B1I and B2I respectively. The pseudorange and
Doppler data used in the algorithm were the original data in
single-frequency condition. The reference point is the average
data of 24 hours’ positioning data at a same position received
by OEM-628.
Fig.2 shows the error compared to the reference point of
B1I and B2I by PSO_SRUKF algorithm. TABLE I shows
root-mean-square error (RMSE) of 3D position of B1I and
B2I by PSO_SRUKF algorithm.

13
 condition.

Acknowledgements

This work was supported by National Natural Science

Foundation of China (61471237).

References

[1] China Satellite Navigation System Management Office,

"Beidou System Space Signal Interface Control File
(version 2.1)", Beijing, pp.8-9, November 2016.
[2] F. Dai and X. Mao, "Dual GNSS systems positioning
based on nonlinear filter and satellites selection in bad
conditions",2014 IEEE/SICE International Symposium
on System Integration, Tokyo, pp. 334-339, 2014.
[3] F. Dai and X. Mao, "BDS/GPS dual systems
positioning based on Kalman filter in urban canyon
environments", IEEE 17th International Conference on
Fig.3 Error comparison of DFCP and receiver
Intelligent Transportation Systems, Qingdao, pp. 1882-
Table 2 Error comparison of DFCP and receiver 1883, 2014.
RMSE (m) [4] S. M. Siddiqui, "Integrated navigation and self-
Method
X Y Z 3D alignment using Square Root Unscented Kalman
DFCP 1.4891 1.5766 0.3817 2.2020 filtering", 10th International Bhurban Conference on
Applied Sciences & Technology, Islamabad, pp. 73-76,
Receiver 1.5034 2.8320 1.1853 3.4184 2013.
It can be concluded from Fig.3 that the positioning error [5] S. Yousefi, X. W. Chang and B. Champagne, "Mobile
solved by DFCP algorithm is small and the convergence Localization in Non-Line-of-Sight Using Constrained
speed is fast. The error is near to zero after transitional period. Square-Root Unscented Kalman Filter", IEEE
It is clear that the accuracy of DFCP algorithm is higher than Transactions on Vehicular Technology, Vol. 64, No. 5,
that of receiver. It is worth noting that there are jumps in the pp. 2071-2083, May. 2015.
output result of the receiver while no obvious jump in that of [6] H. Azami, M. D. Soltani and I. Tavakkolnia, "An
DFCP algorithm, which proves the efficiency of this intelligent approach for classification of GPS satellites
algorithm. Comparing TABLE I with TABLE II, it can be based on neural network, genetic algorithm and particle
seen that the RMSE of the result of DFCP algorithm is smaller swarm optimization", 2016 International Conference
than that of any single frequency, which proves that the DFCP for Students on Applied Engineering, Newcastle upon
is more stable than single frequency positioning. The Tyne, pp. 70-74, 2016.
experiment proves that the DFCP algorithm can effectively [7] X. Wang et al., "Iterated square root unscented Kalman
improve the accuracy and stability of BDS positioning. filter and its application in deep sea vehicle navigation",
27th Chinese Control and Decision Conference,
6. Conclusion Qingdao, pp. 4880-4885, 2015.

In this paper, DFCP based on the PSO_SRUFK

algorithm is proposed, which could estimate the noise
covariance matrix accurately and compensate defects of EKF
and UKF. The experimental results show that the DFCP
algorithm can effectively improve the positioning accuracy
which verifies its effectiveness and superiority in BDS
positioning.
Future works include testing the efficiency of this
algorithm in the dynamic environment and bad positioning

14