An improved fading Kalman filter in the application of BDS dynamic positioning

International Journal of Research in Engineering and Science (IJRES)
ISSN (Online): 2320-9364, ISSN (Print): 2320-9356
www.ijres.org Volume 4 Issue 2 ǁ February. 2016 ǁ PP.01-08
www.ijres.org 1 | Page
An improved fading Kalman filter in the application of BDS
dynamic positioning
Yi Qingming, Xie Jinhua, Shi Min
(College of Information Science and Technology, Jinan University, China)
ABSTRACT: Aiming at the poor dynamic performance and low navigation precision of traditional fading
Kalman filter in BDS dynamic positioning, an improved fading Kalman filter based on fading factor vector is
proposed. The fading factor is extended to a fading factor vector, and each element of the vector corresponds to
each state component. Based on the difference between the actual observed quantity and the predicted one, the
value of the vector is changed automatically. The memory length of different channel is changed in real time
according to the dynamic property of the corresponding state component. The actual observation data of BDS is
used to test the algorithm. The experimental results show that compared with the traditional fading Kalman filter
and the method of the third references, the positioning precision of the algorithm is improved by 46.3% and
23.6% respectively.
Keywords: BDS positioning, fading factor vector, Kalman filter, memory length;
I. Introduction
The Beidou system(BDS) goes on retaining the active positioning, two-way timing and short message
communication service of the Beidou satellite navigation testing system. Since December 27, 2012, The BDS
provides continuous passive location, navigation, timing and other services to the Asia Pacific, and is expected
to finish building the BDS which is global coverage in about 2020[1]
. As the BDS infrastructure getting better
gradually, BDS receiver terminal will be widely used in surveying and mapping, telecom, exploration,
transportation and other fields. Therefore, it is necessary to improve the calculating precision of positioning
algorithm.
In satellite navigation and positioning data processing, Kalman filter technology is a kind of filtering
algorithm which is used frequently. When the observation information, model and statistical information are
reliable, Kalman filter’s computing performance is perfect. But when there is large model error or states
mutation, the error between the state estimate value of Kalman filter and the actual system is huge, and can not
reflect the real system, and even cause filtering divergence in under certain conditions. To solve the problem,
the fading Klaman filter is proposed by some scholars[3,4,5]
. Using the fading factor to limit the memory length,
the fading Kalman filter could make full use of the current observations, and eliminate “outdated” observation
data gradually. The algorithm can effectively solve the problem of filtering divergence, and improves the
stability of the calculation process. However, the filtering result turns into suboptimal, and its result accuracy is
not guaranteed[6,7,8]
.
Before the BDS data is widely used in positioning analysis, some simulation experiments have been used to
study the positioning performance of Beidou system. With the improvement of the Beidou data system and the
wide application of Beidou data, more and more scholars begin to study the real metrical performance of BDS[9]
.
In order to solve the problem of low dynamic performance and low positioning accuracy of traditional fading

An improved fading Kalman filter in the application of BDS dynamic positioning
Kalman filter algorithm, an improved Kalman filter algorithm based on fading factor vector is propose. It is used
in BDS dynamic navigation and positioning, and has been tested by the actual BDS observation data. The
experimental results show that the algorithm effectively improves the dynamic performance and positioning
result’s precision of BDS.
II. Fading Kalman filter positioning model
In order to determine the position and speed of the carrier, the state vector X is defined as below.
 T
zyx bvzvyvxX (1)
In formula (1), x , xv , y , yv , z and zv are the position and velocity components of the carrier on a three
axis coordinate system, and b is the clock error. Thus, the state transition equation is defined as below.
kkkkk WXΦX   11, (2)
kX is the state vector of k moment, and kW is the system noise vector, while 1, kkΦ is state transition
matrix.























1000000
T100000
0T10000
00T1000
000T100
0000T10
00000T1
1,kkΦ (3)
 T
bzyxk  000W (4)
x , y , z and b are white Gaussian noise, whose means are zero, and variance are
2
x 、
2
y 、
2
z and
2
b respectively. T is the sampling interval. In a moment, when the observation data of N satellites are
available, the pseudorange observation equation is defined as below.
iiii
ncbzzyyxx sssi   222
)()()(~ (5)
i~ is the number i satellite’s pseudorange observation. i
n is the noise. ( isx ， isy , isz )is the 3d coordinate
for the satellite, while Ni 1 . c is the speed of light. After formula (5) is linearized by Taylor series
expansion, we can obtain system measurement equation as show below.
kkkk VXHZ  (6)
 T
Nk  ~~~
21 Z (7)

kZ is the observation vector for k moment, which is a n-dimensional column vector. N is the number of
satellites whose signals are received by the receiver. kH is the observation matrix. kV is the observation noise.
The recursive equation of fading Kalman filter is showed as below.

















1,.
1
1,1,
11
1,1,11,1,
1111
)(
)(
)(+=
=
kkkkkk
k
T
kkkk
T
kkkk
k,k-kkkk,k-k,k
k
T
kkkkkkkkk
,k-k-k,k-k,k-
PHKIP
RHPHHPK
XHZKXX
QΦPΦP
XΦX

(8)
kQ is the system process noise covariance matrix. kK is the gain matrix. I is the unit matrix. kR is the
observation noise covariance matrix. The fading Kalman filter used fading factor k to calculate the 1, kkP ,
and this differs from the standard Kalman filter. The memory length of Kalman filter is limited by the fading
factor to make full use of the current observations. The filter divergence has been avoided by aggravated the role
of current observation data in state estimation.
III. The improved fading Kalman filter
According to formula (8), When the traditional fading Kalman filter estimating the error variance matrix, all
the elements in the matrix is increased by )1( kk  times. Such a kind of indiscriminate multiplication
increases the redundancy. Although the problem of filtering divergence has been solved effectively, the
calculation result contains larger error, so that the accuracy is lower.
In order to solve the problems above, the traditional fading Kalman filter algorithm has been improved as
below.
In order to solve the problems above, the traditional fading Kalman filter algorithm has been improved as
below.
The state vector is defined as  T
nk xxx  21X . Based on the different elements in state
vector, a new fading factor vector is built, which is defined as  T
nk   21ω . And each element
of vector kω corresponds to each element of the state vector, while n is the dimension of the state vector. Thus,
the forecast state covariance is defined as below.
T
k
T
kkkkkkk diagdiag )()(ˆ
1,1,11, ωΦPΦωP  (9)
kkk QPP 
ˆ
1, (10)
The predictive value of state vector is defined as below.
11.1,
ˆ
  kkkkk XΦX (11)

In which 1,
ˆ kkX is showed as below.
 nkk xxx ˆˆˆˆ
211, X (12)
i
kXˆ is defined as below.
 T
ni
i
k xxxx ...ˆ...ˆ 21X , ,...,n,i 21 (13)
Thus, the predictive value of the observation vector is showed as below.
i
kk
i
k XHZ ˆˆ  (14)
The pseudorange residual error is defined as below.
i
kk
i
kk ZZV ˆ
1,  (15)
In which, ))(( 1,1,1,
Ti
kk
i
kk
i
kk E   VVV is the covariance matrix of the pseudorange residual error, and it
is defined as below.
 
 
M
i
Ti
ikik
i
ikik
i
kk
M 1
1,1,1, ))((
1
VVV (16)
The value of M is determined by experience. Many experiments proved that when M is 10, the covariance
matrix can be estimated effectively without large computational complexity.
Each element of the fading factor vector can be calculated by the following formulas.
]}[/][,1max{ k
i
ki trtr TN (17)
k
T
kkk
i
kk
i
k RHQHVN   1, (18)
T
k
T
kkkkkkkk ΦPΦ HHT 1,1,11,  (19)
When 1i , the fading Kalman filter degenerate into basic Kalman filter.
When ji   , nji ,...,2,1,  , ji  , the filtering process is equivalent to the traditional fading Kalman filter,
and all of the elements of the estimation error variance matrix will be increased at the same time. In this case,
when the system state change, the estimation error will increase, which leads to
i
kN increases. According to the
calculation formulas of the fading factor vector elements, the value of i and 1, kkP will increase. The
algorithm puts the new current observation data in an important position by increasing the filtering gain. And as
a result, the ill-effect of old data is decreased and the tracking ability of Kalman filter is improved.
When ji   , nji ,...,2,1,  , ji  , the elements of fading factor vector are not equal to each other. Carrier
is moving in 3d space. Assume that one of the component of the state changes,
i
kk 1, V and i will increase. In

this way, the improved fading Kalman filter can solve the problem of filtering divergence more effectively, and
the accuracy of the results will be further improved.
In the process of filtering, before estimating the state error covariance matrix, the fading factor vector is
calculated online according to the historical data, the current observation data and the given window length in
advance. The memory length of each channel is changed adaptive. For the channel with high quality observation
data, the length of memory is shortened to track the state of the system and improve the accuracy of results. For
the channel with bad observation data, the weight of the new observation data is reduced to weaken its influence
on the result of the operation.
IV. The experimental results and analysis
The algorithm above is verified by the actual collection of BDS satellite observation data. The receiver that
is used to collect satellite data is UB280 BDS/GPS double system receiver from UNICORE company, which
can receive the B1/B2 signal of BDS and L1/L2 signal of GPS. The receiver is installed in a car, and the antenna
is installed on the roof. The car is moving around the Olympic center of Guangzhou, and its path is showed as
figure 1. The data sampling interval T is 50 ms. The testing time is 10:00 am on July 25, 2015. The car traveled
along the path shown in figure 1, from the starting point, through the section A1A2, section B1B2 and section
C1C2, and to the destination point finally. The observation data of BDS and GPS received during the travel is
collected and stored for the subsequent processing.
Figure 1 the path of the car
The positioning module named TRACK of GPS data processing software named GAMIT can get the car’s
3d coordinate and speed at each moment, whose plane precision is within 2 cm. The result of TRACK module is
selected as the true value of positioning results. On this basis, the results of the three algorithms are compared
with each other with the BDS observation data in MATLAB platform, which are the algorithm in the paper , the
improved fading Kalman filter in reference [3] and the traditional Kalman filter.
The related parameters are selected as shown below. The initial position variance of x and y direction are
222
m100 yx  . The variance of speed are
2222
/sm25 yx vv  . The variance of clock is
2122
s10
b . The variance of pseudorange measurement noise is
22
m1D , and the variance of

pseudorange rate measurement noise is
222
/sm01.0D , The corresponding noise covariance matrix use
the model in reference literature[6, 10]. The initial value of state is  T
00000000 X , and
the initial value of the error variance matrix is )10251002510025100( 3
0

 diagP .
Error is the difference between the true value and estimated one. The curve of X coordinate estimation error
is shown in figure 2 and the curve of X speed estimation error is shown in figure 3.
Figure 2 the curve of X coordinate estimation error
Figure 3 the curve of X speed estimation error
The accuracy and operation time of the three algorithms are shown in table 1.

Table 1 the accuracy and operation time of the three algorithms
State component Traditional fading
Kalman filter
The improve fading Kalman
filter in reference [3]
The algorithm in
this paper
X(m) 11.56 8.04 6.19
xV (m/s)
2.52 1.75 1.24
Operation time (ms) 0.216 0.259 0.283
The data in table 1 shows that compared with the traditional Kalman filter, the precision of location of the
algorithm in this paper is improved by 46.3%, and the precision of speed is improved by 50.8%, while the
operation time only increase by 23.7%. Compared with the improved fading Kalman filter in reference
literature[3], the precision of location of the algorithm is improved by 23.6%, and the precision of speed is
improved by 29.1%, while operation time only increased by 9.3%.
V. Summary
This paper proposed an improved fading Kalman filter algorithm based on fading factor vector. On the
basis of traditional Kalman filter, a fading factor vector has been constructed and the method to calculate the
element of vector has been introduced. The algorithm is applied to BDS dynamic positioning. The simulation
results show that the algorithm not only improves the accuracy of BDS dynamic positioning, but also has good
dynamic performance and stable filtering process. Moreover, this paper provides the related experimental data
for the debugging of BD-2 system, and it is very useful for application study of high precision dynamic
positioning of BDS.
References
[1] An Xiangdong.The Comparative Analysis of Performance about GPS and BDS in Single Point
Positioning[J].GNSS world of china, 2014,39(3):8-14.
[2] Hu Guorong, Ou Jikun.The Improved Method of Adaptive Kalman Flitering for GPS High Kinematic
Positioning[J].Acta Geodaetica et Cartographica Sinica, 1999, 28(4):290-294.
[3] Wang Hu, Wang Jiexian, Bai Guixia, et al. An Improved Fading Kalman Fliter and its Application to
GPS Kinematic Positioning [J]. Journal of Tongji University(Natural Science), 2011, 39(1):124-128.
[4] Yang Liuqing, Xiao Qiangui, Niu Yan, et al.Design of Localization System Based on Reducing
Kalman Fliter[J].Journal of Nanjing University of Aeronautics & Astronautics, 2012, 44(1):134-138.
[5] Tian Chong, Wang Xingliang, Lu Yane. Combination Algorithm of Two Kinds of Adaptive Filter in
GPS Positioning [J]. Modern Deffence Technology, 2012, 40(3):72-77.
[6] XIA Qijun, SUN Youxian, ZHOU Chunhui. An optimal adaptive algorithm for fading kalman filter
and its application [J]. Acta Automatic Sinica, 1990, 16(3):210-216.
[7] Li Yongjun, Zuo Juan. Research on adaptive Kalman Filtering algorithm in GPS kinematic positioning
[J]. Engineering of Surveying and Mapping, 2012, 21(4):29-32.
[8] Mohamed A H, Schwarz K P. Adaptive Kalman filter for INS/GPS[J]. Journal of
Geodesy,1999,73(4):193-203.
[9] Wu T T, Zhang Y, Liu Y M, et al. Bei Dou/GPS combination positioning methodology[J].Journal of
Remote Sensing, 2014, 18(5):1087-1097.
[10] FANG Jiancheng.The study and application of the optional estimation theories and mothod in
integrated navigation system [D]. Nanjing: Southeast University, 1998.

[11] King R W, Bock Y. Documentation for the GAMIT GPS analysis software[M]. Cambridge:
Massachusetts Institute of Technology Press,2000.
[12] Yang Yuanxi, Gao Weiguang.Comparison of Two Fading Filters and Adaptively Robust
Filter[J].Geomatics and Information Science of Wuhan University, 2006, 31(11):980-982.
[13] Gong Zhenchun, Chen Anning, Li Ping, et al.Study of the Application of Adaptive Kalman Filtering
for GPS Kinematic Positioning[J].Bulletin of Surveying and Mapping, 2006, 1(7):9-12.
[14] Lu Chenxi, Tan Yunhua, Zhu Bochen, et al.Adaptive Kalman Filter Based Navigation Algorithm for
Single-Frequency Precise Point Positioning[J].Acta Scientiarum Universitatis Pekinensis, 2011,
47(4):587-592.
[15] Dong Xurong, Tao Daxin.An Efficient Kalman Filtering Algorithm and its Application in Kinematic
GPS Data Processing[J].Acta Geodaetica et Cartograhpica Sinica, 1997, 26(3):221-227.

An improved fading Kalman filter in the application of BDS dynamic positioning

More Related Content

What's hot (19)

Similar to An improved fading Kalman filter in the application of BDS dynamic positioning (20)

More from IJRES Journal (20)

Recently uploaded (20)

An improved fading Kalman filter in the application of BDS dynamic positioning