Proceedings of 2020 IEEE

International Conference on Mechatronics and Automation

October 13 - 16, Beijing, China

A UWB 3D Localization Algorithm Based on Residual

Weighting
Yonggang Zhang Qi Liu
Yunnan Electric Power Technology Co., Ltd College of Automation
China Southern Power Grid Company Limited Harbin Engineering University
Kunming, Yunnan Province, China Harbin, Heilongjiang Province, China
[email protected] [email protected]
Feng Shen Wenqiang Li
School of Instrumentation Science and Engineering School of Instrumentation Science and Engineering
Harbin Institute of Technology Harbin Institute of Technology
Harbin, Heilongjiang Province, China Harbin, Heilongjiang Province, China
[email protected] [email protected]

Abstract - This paper proposed an improved algorithm based difference of signal arrival (TDOA), and received signal
on Chan algorithm to improve the accuracy of three-dimensional strength (Received Signal Strength, RSSI), based on the angle
(3D) positioning. For the 3D indoor positioning based on ultra of arrival of the signal (Angle of Arrival, AOA). Because of
wide band (UWB), the positioning accuracy is reduced due to the NLOS effect (Non Line Of Sight), multipath effect, and
non-line-of-sight environment (NLOS). Analyzing the TDOA
multiple access interference of radio waves in an indoor
wireless positioning technology based on UWB and deducing the
Chan algorithm which can be applied to 3D positioning. The environment, TOA/TDOA technology is the preferred
UWB locates the base station group, uses the three-dimensional positioning method for UWB positioning systems, and the
Chan algorithm to estimate the tag coordinates for each group of implementation of TOA positioning technology requires a
base stations, and use the error by using the residual value strict overall system Clock synchronization, which is difficult
weighting method. The simulation results show that the to achieve in hardware and practical scenarios.
positioning accuracy of the estimated position of the tag is TODA technology is widely used because it does not
significantly improved by the residual weighting method. When require strict clock synchronization. To apply TDOA
the noise variance is 0.5, the average root mean square error is
technology, we need to obtain the measurement time
0.53m, which can effectively suppress the influence of NLOS.
difference, establish a non-linear equation set, and use the
algorithm to solve the position to achieve positioning. Güvenç
Index Terms - UWB; NLOS; Chan Algorithm; 3D Positioning
[1], Zhang G [2], etc. used the method of least squares to study
I. INTRODUCTION the TDOA positioning method, and only the approximate
value of the position estimate was obtained by solving the
Indoor positioning has become a research hotspot for equation, but not the optimal solution. Literature [4] proposed
domestic and foreign commercial companies and scholars with the Fang algorithm, which will use TDOA information to
the development of economy and the expansion of commercial establish a hyperbolic equation and linearize it, and then solve
activities towards indoor positioning, as well as the high the target position. However, the Fang algorithm cannot
requirements for automation and intelligence in industrial improve the estimation accuracy by measuring redundant data,
fields such as smart factories and smart storage. can only apply three base stations, and can’t be applied in the
Before 2013, most research institutions used the field of UWB three-dimensional positioning. T.Y Chan
conventional electrical signal positioning methods such as proposed the Chan algorithm. Studies in literature [5] [6] show
WIFI, RFID, ZigBee, Bluetooth, and WSN for indoor that the Chan algorithm has high accuracy under the LOS
positioning research methods, and supplemented with non- condition (Line of Sight) and does not require initial values.
electricity based on geomagnetic fields, IMU, infrared, or The zero-mean Gaussian noise used will reduce accuracy
vision signal positioning technology. under the NLOS factor.
But in 2013, Deca Wave of Ireland introduced a low-cost In this paper, we have made a model derivation from
high-precision positioning chip DW1000 based on UWB literature [7] using three-dimensional space. The derived Chan
technology. This chip’s theoretical accuracy reaches extremely algorithm is used to obtain the initial position estimate of the
high centimeter accuracy. Since then, research and tag through the TDOA positioning method of the UWB
commercialization in the field of indoor positioning using system, and then the residual value weighting processing is
UWB technology has increased. The indoor positioning performed to smooth the NLOS error interference to achieve
technology based on UWB can be summarized into four the treatment of the estimated points precise positioning [8] [9]
categories: based on time of signal arrival (TOA), time [10] [11].

978-1-7281-6416-8/20/$31.00 ©2020 IEEE 309

Authorized licensed use limited to: University of Gothenburg. Downloaded on December 19,2020 at 05:36:15 UTC from IEEE Xplore. Restrictions apply.
 II. TDOA POSITIONING AND CHAN ALGORITHM and building a linear equation with the unknown za as
A. TDOA positioning equation (4).
Based on the characteristics of the UWB positioning
system, in practice, due to different application scenarios, it is ri ,1 = ( x − X i )2 + ( y − Yi ) 2 + ( z − Z i ) 2
difficult to arrange base stations. If the communication (2)
between base stations does not use wired communication, it is − ( x − X 1 ) 2 + ( y − Y1 ) 2 + ( z − Z1 ) 2
difficult to ensure strict clock synchronization. The TDOA
method is used to measure the time difference between the ri 2,1 + 2ri r1 = ( K i − K1 ) − 2( X i ,1 x + Yi ,1 y + Z i ,1 z ) (3)
UWB signal sent from the tag and each base station, which can
reduce the difficulty in deploying the base station. The space
position of the base station is used as the focus, and the h = Ga Z a (4)
distance difference converted by the time difference of signal In formula
arrival is used as the long axis of the hyperbola. The focus
position of the hyperbola is the position of the tag. TDOA  X 2,1 Y2,1 Z 2,1 r2,1 
three-dimensional positioning schematic diagram shown in X Y2,1 Z 2,1 r2,1 
Ga = − 
3,1
Fig.1.
 ... ... ... ... 
 
 M ,1 M ,1
X Y Z M ,1 rM ,1 
 r2,1
2
+ K1 − K 2  x
 2   y
1 r + K1 − K 3 
h =  3,1 za =  
2 ...  z
 2   
 rM ,1 + K1 − K M   r1 
0
Let za be the true value of za , that is the true value of
the tag. Assume the signal delay random error generated by the
system is n , and the mean value is zero, the variance is σ ,
the covariance matrix is Q = diag {σ 2,1 ,..., σ M0 ,1} , and the
Fig.1 TDOA positioning method schematic 0
, σ 3,1
0

B. Chan algorithm TDOA measurement value is ri ,1 = ri 0,1 + nri ,1 , ri ,0 = ri 0,0 + nri ,0 is

In the UWB indoor positioning process, the number of the true value of the ri , so the error vector equation and
base stations involved in positioning to achieve three-
analytical formula are equation (5) (6).
dimensional positioning is at least four, total of M base
stations. Let the coordinates of the tag be ( x, y, z ) and the
ψ = h − Ga za0 (5)
coordinates of the base station be ( X i , Yi , Z i ) .The position of
1
the base station is known. Set one of the base stations as the ψ = cBn + c 2 n : n (6)
positioning reference station. c is the speed at which 2
electromagnetic waves travel through the air. ti is the time Among them, knowing B = diag {r20 , r30 ,..., rM0 } , use
from the tag to the i-th base station for the UWB signal, and equation (6) to find the error covariance matrix
ti ,1 is the time difference between the UWB signal from the tag ψ = E (ψψ T ) = c 2 BQB . Assuming that the elements of za are
to the base station and the first base station, ti ,1 = ti − t1 . ri is independent of each other, a preliminary weighted least
squares estimate of za can be obtained.
the true spatial distance from the tag to the i-th base station,
ri = cti , can also be expressed as
za = arg min {(h − Ga za )T Ψ −1 (h − Ga za )}
(7)
ri = ( x − X i ) 2 + ( y − Yi ) 2 + ( z − Z i ) 2 (1) = (GaT Ψ −1Ga ) −1 GaT Ψ −1h

ri ,1 is the difference between the distance from the tag to When the tag is far away from all base stations, the
distance between the tag and each base station is
the -th base station and the first base station, ri ,1 = ri ,1 − ri ,1 , approximately equal to the distance between the tag and the i-
expressed as equation (2). th base station, then B ≈ rrange I  rrange represents the distance,
According to ri 2,1 = (ri − r1 ) 2 , we can get ri 2 = (ri 2,1 + r1 ) 2 , and I is the identity matrix of order M − 1 , then the error
take-in formula (1), eliminate higher-order terms available, we covariance matrix ψ = c 2 rrange
2
Q , at this time Equation (7) is
can get equation (3). In K i = X i2 + Yi 2 + Z i2 , shifting the term approximately to be as equation (8).

310

Authorized licensed use limited to: University of Gothenburg. Downloaded on December 19,2020 at 05:36:15 UTC from IEEE Xplore. Restrictions apply.
 Using the za value obtained from the preliminary In the formula, the ψ ' error covariance matrix is
estimation to re-estimate the B and ψ values, and then bring approximated by ignoring the higher-order terms in the
Eq.(7), the position estimates can be obtained in the case calculation of ψ ' when ei is small:
where the za elements are independent of each other.
ψ ' = E ψ 'ψ 'T  = 4 B ' cov( za ) B ' (16)
−1 −1 −1
za = (G Q Ga ) G Q h
T T
B = diag { x − X 1 , y − Y1 , z − Z1 , r }
a a (8) ' 0 0 0 0
1 (17)
But, in reality, the element r1 in za is associated with
cov( za ) in equation (16) contains the unknown truth
( x, y, z ) . It is assumed that there is a sufficiently small noise
value G a0 . We can use Ga in equation (4) instead for
so that za becomes a random vector whose average value is
approximate calculations, the true coordinates of unknown
the true value. At this time, Ga , h and za are expressed as
tags contained in B ' are weighted least squares estimates
equation (9). using equation (7), use the obtained result za to calculate
approximately, and choose according to the prior information.
Ga = Ga0 + ΔGa
Finally, the tag coordinate zt can be estimated as equation
h = h 0 + Δh (9) (18).
z a = z + Δz a
0
a
 X1 
Putting formula (9) into formula (5), we can get (10), (11), zt = ± za' +  Y1  (18)
(12).  Z1 

ψ = Δh − ΔGa za0 (10) III. RESIDUAL WEIGHTING ALGORITHM

Δza = c ( Ga0Tψ −1G 0 −1
a ) Ga0Tψ −1 Bn (11) The UWB positioning system uses the TDOA positioning
method to process and eliminate NLOS errors, smooth the
(
cov( z a ) = E Δza ( Δza )
T
) = (G aψ −1G
0T
a)
0 −1
(12) residual value between the distance measurement and the
estimated value. Using the Chan algorithm to obtain a three-
dimensional estimate of the tag position. The Chan-RW
Let ei and i = 1, 2,3, 4 be noise and the za element be
algorithm (Chan-Residual Weighting) proposed in this paper
expressed as equation (13). using the squared residual value weighting method to smooth
the NLOS error of the estimated value of the Chan algorithm
T
za =  x 0 + e1 , y 0 + e2 , z 0 + e3 , r10 + e4  (13) to obtain the optimal estimation of the three-dimensional
spatial position of the tag.
The new linear equation can be established as equation
(14). p( zt , k , Sk ) =  p (z
i∈Sk
i t,k , Sk )
(19)
  r − (|| z − Ai || − || zt , k − A1 ||)  , ∀k
2
ψ ' = h ' − Ga' za' (14) = i ,1 t ,k
i∈S k

Among them
( xk − X i ) + ( xk − X i ) + ( xk − X i )
2 2 2
(( x 0 + e1 ) − X 1 )2  || zt , k − Ai ||= (20)
 0 
(( y + e2 ) − Y1 ) 2 
h' =  0 ( xk − X 1 ) + ( xk − X 1 ) + ( xk − X 1 )
2 2 2
|| zt , k − A1 ||= (21)
 (( z + e3 ) − Z1 )2 
 
 (r1 + e4 ) 
0 2
When the UWB system performs three-dimensional
1 0 0  positioning, one of the base stations is selected as a reference
0 1 0 ( x − X 1 ) 2  station, and the remaining M − 1 base stations are grouped.
Ga' =   z ' =  ( y − Y )2 
0 0 1  a  
1 Due to the three-dimensional positioning, the participation of
   ( z − Z1) 
2  the base stations is at least 4, so the grouping has a total of

1 1 1  M −1

Weighted least squares estimation of z a ' .

N= C
i =1
i
M −1 combinations, and the k-th combination Sk is

za' = ( Ga'Tψ '−1Ga' ) Ga'Tψ '−1h'

−1
(15) the set of base stations corresponding to this combination.
Under this combination, the tag sitting tag estimated by the

311

Authorized licensed use limited to: University of Gothenburg. Downloaded on December 19,2020 at 05:36:15 UTC from IEEE Xplore. Restrictions apply.
Chan algorithm is zt , k = [ xk , yk , zk ] , k = 1, 2,..., N , the seating point positioning performance of the Chan-RW algorithm and
the Chan algorithm. Firstly, set up 7 base stations to be
tag of the i-th base station is Ai = [ X i , Yi , Zi ] , and the distributed in a three-dimensional space with coordinates (0, 0,
corresponding squared residual function is equation (19), (20), 0) ǃ ( 50 3 , 50, 1.5 3 ) ǃ ( 25 3 , -80, 50 3 ) ǃ ( -25 3 ,
(21).Set each combination to use the chan algorithm's
50 2 , -50 2 )ǃ( -50 3 , 0, 65)ǃ( -25 3 , -50 2 , -70)ǃ
estimated position to give weights wk , wk = 1 / p ( zt , k , S k ) ,
( 25 3 , -75, 50 3 ), Fig.2 and Fig.3 shows the root mean
then the tag position is estimated as equation (22).
square error curves of the two algorithms under different noise
In order to further reduce the influence of errors caused
variances.
by non-line-of-sight factors, a signed residual weighted model
For the traditional Chan algorithm and Chan-RW
is used instead of a square residual weighted model. During
algorithm, 200 independent and repeated simulation
the positioning of the UWB system, when J ( zt , k , S k ) is a
experiments were performed with noise variance values of 0.5,
negative residual, the corresponding Chan algorithm's 1.0, and 1.5, respectively.
intermediate estimated value zt , k is more likely to be affected The simulation results show that the localization
by non-line-of-sight errors. At this time, a smaller weight performance of the two algorithms is improved when the noise
should be given to this value. Assigns weight wk to a linearly variance is smaller, which proves that the positioning
performance of the algorithms has convergence.
weighted weight based on the interval length.
The average RMS error of the Chan algorithm and Chan-
N N RW algorithm when the noise variance value is 0.5 m 2 For
zrw =  ( zt , k wk ) /  wk (22) 0.68, 0.53. The positioning accuracy is increased by 22% on
k =1 k =1 average.

J ( zt , k , S k ) =  J (z
i∈S k
i t ,k , Sk )
(23)
=   r − (|| z
i∈S k
i ,1 t ,k − Ai || − || zt , k − A1 ||) , ∀k

J ( zt , k , Sk ) − min  J ( zt , k , S k ) 
wk = (24)
max  J ( zt , k , S k )  − min  J ( zt , k , Sk ) 

In the formula, max[ J ( zt , k , S k )] and min[ J ( zt , k , S k )]

represent the maximum and minimum values of N J ( zt , k , S k ) , Fig.2 The Chan algorithms for 3D positioning errors
respectively. We can using equation (22) to calculate the 3D
estimated position of the tag.
IV. SIMULATION ANALYSIS
In order to analyze the optimization performance of the
algorithm of this paper on the traditional Chan algorithm. Root
Mean Squared Error (RMSE) is used to measure the accuracy
of tag position estimation. RMSE is defined as equation (25).

1 Q ˆ
RMSE =  || θq − θ 0 ||2
Q q =1
(25)

Fig.3 The Chan-RW algorithms for 3D positioning errors

In the formula, θˆq represents the estimated coordinate
value ( xˆ , yˆ , zˆ ) of the q-th experiment, θ 0 represents the true To further analysis the distribution of positioning errors,
0 0 0
coordinate ( x , y , z ) of the tag, and Q is the total number of Fig.3 shows the cumulative root mean square error
distributions of the two algorithms. Chan and Chan-RW
experiments. In reality, during the operation of the UWB
algorithms have 42.5% and 51% probabilities of positioning
system, when there is an NLOS factor, measurement errors
errors less than 0.5m, respectively. The maximum root mean
will occur. Therefore, in the simulation analysis, Gaussian
square error of the Chan algorithm is 1.54m, and the maximum
noise with different variances of the simulated zero mean is
root mean square error of the Chan-RW algorithm is 1.38m,
added, and the variance value is 0.5 m 2 . Verifies the single-
indicating that the overall positioning performance of the

312

Authorized licensed use limited to: University of Gothenburg. Downloaded on December 19,2020 at 05:36:15 UTC from IEEE Xplore. Restrictions apply.
Chan-RW algorithm is improved compared to the Chan in non-line-of-sight environments and suppress the effect of
algorithm. Fig.5 shows the simulation results of the 3D non-line-of-sight to a certain extent.
positioning of the Chan-RW algorithm. When the noise ϕ
ACKNOWLEDGMENT
variance value is 0.5 m 2 and the independent positioning is
repeated 200 times, the Chan-RW algorithm estimates the tags Thank you for the selfless risk of the team and the strong
on the xy , xz and yz planes. From the projection of each point of the leader.
plane, we can see that the algorithm has good convergence. REFERENCES
The position estimates are all around the real point. Simulation [1] A. Alarifi, A. Al-Salman, M. Alsaleh, A. Alnafessah, S. Alhadhrami, M.
experiments show that the Chan-RW algorithm can better meet Al-Ammar, and H. Al-Khalifa, “Ultra Wideband Indoor Positioning
the requirements of UWB positioning system for positioning Technologies: Analysis and Recent Advances,” Sensors, vol. 16, no. 1,
errors. pp. 1-36, 2016.
[2] E. Bonizzoni, A. Puiatti, S. Sapienza, P. M. Ros, D. Demarchi, and P.
Bonato, "UWB Tracking for Home Care Systems with Off-the-Shelf
Components," in 2018 IEEE International Symposium on Circuits and
Systems, 2018, pp. 1-5.
[3] Y. T. Chan, and K. C. Ho, “A simple and efficient estimator for
hyperbolic location,” IEEE Transactions on Signal Processing, vol. 42,
no. 8, pp. 1905-1915, 1994.
[4] B. T. Fang, “Simple solutions for hyperbolic and related position fixes,”
IEEE Transactions on Aerospace and Electronic Systems, vol. 26, no. 5,
pp. 748-753, 1990.
[5] E. García, P. Poudereux, H. Á, J. Ureña, and D. Gualda, "A robust UWB
indoor positioning system for highly complex environments," in 2015
IEEE International Conference on Industrial Technology, 2015, pp.
3386-3391.
[6] . Güvenç, C.-C. Chong, F. Watanabe, and H. Inamura, “NLOS
Fig.4 CDF of location for two algorithms Identification and Weighted Least-Squares Localization for UWB
Systems Using Multipath Channel Statistics,” EURASIP Journal on
Advances in Signal Processing, vol. 2008, no. 1, pp. 271-287, 2007.
[7] C. Qian, W. Xia, W. Cui, Z. Lan, F. Yan, and L. Shen, "A 3-D Indoor
Positioning System Using Asymmetry Double-Sided Two-Way Ranging
And Chan Assisted Extended Kalman Filter." in 2018 10th International
Conference on Wireless Communications and Signal Processing, 2018,
pp. 1-6.
[8] K. Yu, K. Wen, Y. Li, S. Zhang, and K. Zhang, “A Novel NLOS
Mitigation Algorithm for UWB Localization in Harsh Indoor
Environments,” IEEE Transactions on Vehicular Technology, vol. 68,
no. 1, pp. 686-699, 2019.
[9] G. Zhang, Z. Deng, L. Wen, L. Ge, H. Ke, and J. Jiao, An UWB location
algorithm for indoor NLOS environment, 2018.
[10] X. Zhang, and W. Lin, "Hyperbolic-weighted centroid Indoor Location
Algorithm based on Kalman Filter." in 2018 IEEE 3rd Advanced
Information Technology, Electronic and Automation Control
Fig.5 The Chan-RW algorithm 3D positioning renderings Conference, 2018, pp. 2520-2523.
[11] Y. Zhong, T. Wang, Y. Liu, and X. Luo, "Indoor UWB Location Based
on Residual Weighted Chan Algorithm." in 2018 7th International
IV. CONCLUSION Conference on Digital Home, 2018, pp. 274-279.
The improved algorithm proposed in this paper performs
three-dimensional spatial extension of the traditional Chan
algorithm, and uses residual weighting to smooth the error
using the Chan algorithm's estimated value. We proposed a
signed residual value weighting method to further improve the
residual value weighting method. The linear weighting based
on the interval length is used as the weight value of the
residual value weighting method to obtain the three-
dimensional position estimate of the tag.The two algorithms
are analysis by simulation experiments.
The experiments show that the Chan-RW algorithm
proposed in this paper has an average root-mean-square error
of 0.53m when the simulated noise variance is 0.5 m 2 . The
algorithm can improve the accuracy of 3D spatial positioning

313

Authorized licensed use limited to: University of Gothenburg. Downloaded on December 19,2020 at 05:36:15 UTC from IEEE Xplore. Restrictions apply.