Zhu 2021
Zhu 2021
Zhu 2021
Key Laboratory of Universal Wireless Communication. Key Laboratory of Universal Wireless Communication.
Ministry of Education Ministry of Education
Beijing University of Posts and Telecommunications Beijing University of Posts and Telecommunications
Beijing, China Beijing, China
zhuqing@bupt.edu.cn niukai@bupt.edu.cn
Abstract—Recently, the indoor positioning system based on locations, and the neighboring base stations can be seen as the
the angle of arrival (AOA) has been widely concerned. The positioning anchors (AN). Specifically, the wireless position-
positioning system often contains two steps. The first step is the ing technology utilizes one or more parameters measured in
AOA estimation algorithm and the second step is the positioning
algorithm using the estimated AOAs. However, due to the indoor several radio links between the equipment of a special user
signal multipath propagation, the AOA estimation errors can and different ANs for positioning. These parameters mainly
lower the positioning performance. To solve this problem, this include the strength of the received signal (RSS) [1], the
paper proposes a novel angle of arrival (AOA) positioning time of arrival (TOA) [2], [3], or the time difference of
algorithm which selects multiple candidate AOAs and selects arrival (TDOA) [4]. Also, with the development of multi-
the most reliable one from them. When multiple candidate
AOAs are selected and passed into the positioning algorithm, antenna technology, the positioning algorithm based on the
multiple estimated locations can be obtained. To select the most angle of arrival (AOA) can be well implemented, in which
reliable AOA, the location reliability prior information related synchronization requirement between adjacent base stations is
to these estimated locations is introduced. The AOA with the relaxed.
maximum location reliability prior information is the final AOA A least square (LS) solution for location estimation based on
estimation result and its corresponding estimated location is
the final positioning result. The simulation results show that AOA measurements is proposed in [5] where the location esti-
compared with the positioning algorithm which directly uses one mation is obtained as the intersection point of all angular lines
AOA for positioning, the proposed positioning algorithm provides from the ANs. When the LS algorithm is used for positioning,
better positioning performance. firstly, we need to estimate the AOA information. In [6] a
Index Terms—multiple candidate AOAs, location reliability joint angle and delay estimation multiple signal classification
prior information, positioning
(JADE MUSIC) algorithm is introduced to estimate the AOA
information. In [6] the spatial-temporal spectrum is computed
I. I NTRODUCTION
which is a function of the signal propagation delay and angle.
Positioning refers to the problem of determining the loca- The AOAs are the angles that correspond to the peaks on
tion of a target. It can be applied to many fields including the spectrum. In [5] the location estimation is assumed to be
navigation, sensing, radar and sonar. The most widely used po- independent of the AOA estimation. When the JADE MUSIC
sitioning technology is the Global Positioning System (GPS). algorithm is used for AOA estimation in [5], only the angle
GPS can provide reliable real-time positioning services for corresponding to the first largest peak is used for positioning.
user equipments in outdoor scenarios. However, due to the Because the first largest peak is assumed to correspond to the
occlusion of satellite signals in buildings, the performance line of sight (LOS) path which is the direct path between the
of GPS is often unsatisfactory in indoor scenarios. In this UE and the AN. Its propagation delay is the shortest.
case, the wireless positioning technology based on the base However, the signal multipath propagation can cause many
stations in the mobile network is used to estimate the users’ false peaks to appear on the JADE MUSIC spectrum. In this
978-1-7281-9505-6/21/$31.00 ©2021
Authorized licensed use limited to: IEEE
University of Prince Edward Island. Downloaded on June 03,2021 at 22:20:46 UTC from IEEE Xplore. Restrictions apply.
,(((:LUHOHVV&RPPXQLFDWLRQVDQG1HWZRUNLQJ&RQIHUHQFH :&1&
This paper is organized as follows. Section II introduces where i is the antenna index, ω (i) (t) is the Gaussian noise. In
the positioning system model which contains the JADE MU- this paper, the multipath channel model is the channel model
SIC algorithm and the LS positioning algorithm. Section in 5G NR [11]. The AOA information is contained in the
III describes the proposed positioning algorithm and gives complex channel gain and the relationship is as follows
the flowchart of the proposed positioning algorithm. The
(i)
simulation results are provided in section IV. In section IV αl (t) = αl (t)e(−jκd(i−1) sin(θl (t))) (3)
we evaluate and compare the positioning performance of
the proposed positioning algorithm and the LS positioning where κ = (2π/λ), d = λ/2, λ = c/fc . c is the speed of
algorithm in [5]. Finally, conclusions are drawn in Section light, fc is the frequency of the carrier, and θl (t) is the AOA
V. from the UE to the AN. It is worth noting that the relationship
between the AOA information and the complex channel gain
II. P OSITIONING S YSTEM M ODEL is based on the uniform linear antenna arrays [12]. Thus the
This section contains two parts. The first part is the AOA AOA information can be acquired from the complex channel
estimation process, where the JADE MUSIC algorithm is gains contained in the received signal.
described. The second part is the location estimation process, The JADE MUSIC algorithm estimates the AOA based on
where the LS positioning algorithm is depicted. the following three steps. Firstly, it computes the covariance
matrix of the received signal by the following equation
A. AOA Estimation Based on JADE MUSIC
Nt −1
Active positioning is assumed in this paper where the UE 1
R= Y(n)YH (n) (4)
transmits a positioning signal to the ANs. The uniform linear Nt n=0
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on June 03,2021 at 22:20:46 UTC from IEEE Xplore. Restrictions apply.
,(((:LUHOHVV&RPPXQLFDWLRQVDQG1HWZRUNLQJ&RQIHUHQFH :&1&
angle values and delay values in the equation (5), we can get
a two-dimensional spatial-temporal JADE MUSIC spectrum.
The candidate AOAs are the angles corresponding to spectrum
peaks.
Fig.2 depicts the JADE MUSIC spectrum where the x-axis
is the angle, y-axis is the dealy, and Ts is the symbol time.
In the left image of Fig.2, many false peaks appear on the
JADE MUSIC spectrum, and the first largest peak deviates
from the true AOA. In the right image of Fig.2, the first largest
peak accurately corresponds to the true AOA. From Fig. 2,
we can see that the first largest peak is sometimes reliable
sometimes unreliable. Fig. 3 depicts the AOA measurement
error variances of the first largest peak in a 12×12 rectangular
Fig. 2. Normalized JADE MUSIC spectrum room. The x axis and the y axis are the user’s location
coordinates, the black triangle icons are the positions of the
AN1 error variance AN2 error variance ANs and the four figures in Fig. 3 respectively correspond to
1.5 1.4
10 10 1.2 the AOA measurement error variances of the UE at the four
y(meter)
y(meter)
1
1
0.8
ANs. From Fig. 3, we can see that the measurement error
5 5
0.5
0.6 variances at four ANs is related to the UE position. When
0.4
0 0 0.2 the UE is far away from the AN, the probability of NLOS
0 5 10 0 5 10
x(meter) x(meter)
propagation between the UE and the AN increases, and the
AN4 error variance
1.4
AN3 error variance corresponding error variance becomes large. We can also see
10 1.2 10 1.2
1
that when the signal arrival direction is far from the normal
y(meter)
y(meter)
1
0.8 0.8 of the antenna array, the AOA measurement error variance
5 5 0.6
0.6
0.4 0.4
also becomes large. So the AOA measurement error variances
0 0.2 0
0.2 are strongly dependent on the UE positions and are known a
0 5 10 0 5 10
x(meter) x(meter) priori. These prior measurement error variances can be used
to calculate the location reliability prior information in section
Fig. 3. AOA measurement error variance of the first largest peak III.
The LS algorithm only selects the first largest peak, but from
Fig. 2 and Fig. 3, we can see that the first largest peak can
where the H denotes the Hermitian operator and Nt represents deviate from the real AOA at some UE positions. In this case,
the frames of received signal. To get the Y, the received it’s hard to find the real AOA on the JADE MUSIC spectrum,
signal y (i) (t) at the ith array element is sampled and N so multiple candidate AOAs are selected in this paper and the
represents the time sampling points. Then the Discrete Fourier most reliable one will be used for positioning. Note that when
Transformation of the sampled signal is taken and the Discrete the number of the selected peaks is M then there are M NA
Fourier transformation at all array elements are arranged in a UE location estimation results, where NA is the number of the
row. Thus the Y ∈ R(N ×K)×1 where K is the number of array ANs. To reduce the algorithm complexity only the first largest
elements. Secondly, it takes the separation of eigenvectors of peak and the second largest peak are picked in this paper.
the above covariance matrix. The former L larger eigenvalues’
eigenvectors span the signal subspace and the last N K − L
eigenvectors span the noise subspace E. Thirdly, it computes B. LS Positioning Method
the spital-temporal spectrum by the following equation
1 After selecting multiple candidate AOAs from the JADE
P = (5) MUSIC spectrum, these AOAs are translated into the po-
U EEH U H
sitioning Cartesian Coordinate System at the ANs. That is
where U is the spatial-temporal direction eigenvector it can
the θˆi is translated to ϕ̂i . Then pass the ϕ̂i into the LS
be acquired by the following equations
positioning algorithm to get the corresponding UE locations.
U(τl , θl ) = d(θl ) ⊗ d(τl ) (6) The theoretical performance of the LS positioning algorithm
is analysed in [7]. The positioning diagram is depicted in Fig.
d(τl ) = [1, e−j2πΔf τl , ..., e−j2πΔf τl (N −1) ]T (7) 4. The angled lines at each AN intersect at one point that is
−1 −1 the location estimation. ϕ̂i is the angle between the ith anchor
d(θl ) = [1, e−j2πλ d sin θl
, ..., e−j2πλ d(K−1) sin θl T
] (8)
and the UE. [xi , yi ] is the location of the ith anchor which is
where d(τl ) is the temporal direction eigenvector and d(θl ) is known. [xo , yo ] is the location of the UE which is needed to
the spatial direction eigenvector. ⊗ means Kronecker product be estimated and di is the distance between the ith anchor and
and T means transpose operator. Thus by putting different the UE.
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on June 03,2021 at 22:20:46 UTC from IEEE Xplore. Restrictions apply.
,(((:LUHOHVV&RPPXQLFDWLRQVDQG1HWZRUNLQJ&RQIHUHQFH :&1&
where tan−1 is the 4-quadrant arctangent function and ei
x y x y is an independent zero-mean Gaussian noise ei ∼ N (0, σi2 )
M M Thus the joint probability density function of the angle and
xo yo the location is a multivariate Gaussian probability density
function, given by
x y M 1 −JG
f (Φ̂/p) = e 2 (19)
(2π)NA /2 |Q|1/2
Fig. 4. Positioning diagram where Φ̂ is the estimated AOA vector and JG is as follows
We can multiply the above equations by sin ϕi and cos ϕi where e = [e1 , ..., eNA ] is the AOA measurement error vector.
respectively. Then we get the following equations The measurement noise variance [σ12 ...σN 2
] are assumed to
xi sin ϕi + di cos ϕi sin ϕi = xo sin ϕi (11) be the same in [13] leading to the geometric dilution of
precision (GDOP). That is the geometric relation between
yi cos ϕi + di sin ϕi cos ϕi = yo cos ϕi (12) the UE position and the ANs position magnifies the error
of AOA measurements. Due to the GDOP, the UEs around
Thus we can get the following equation the ANs have worse positioning performance than the UEs
at other locations. When the noise variances differ from each
xi sin ϕi − yi cos ϕi = xo sin ϕi − yo cos ϕi (13)
other largely, the GDOP is strongly dependent on them. Thus,
Combine the equations of all ANs and rewrite them as follows: when the noise variances are assumed to be the same, the
⎡ ⎤ selected optimum UE position estimation also suffers from
x1 sin ϕ1 − y1 cos ϕ1 the GDOP. To overcome this problem the noise variance can
⎢ .. ⎥
b=⎣ . ⎦ (14) be measured in the AOA estimation process [9]. In this paper,
xNA sin ϕNA − yNA cos ϕNA the AOA measurement noise variance of the JADE MUSIC
⎡ ⎤ algorithm is measured. The measurement results of the first
sin ϕ1 − cos ϕ1
⎢ .. .. ⎥ largest peak are provided in Fig. 3. From Fig. 3 we can see
Φ=⎣ . . ⎦ (15) that the noise variances can vary from different ANs and are
sin ϕNA − cos ϕNA dependent on the UE position. When the angle corresponding
b = Φp (16) to the first largest peak is used to estimate the UE position,
its noise variance is brought into the joint probability density
where p = [xo , yo ]T is the location of the UE to be estimated. function. Thus when the angle corresponding to the second
Then the equation set can be solved by the LS method as largest peak is used for positioning, its AOA measurement
follows: noise variances are also measured and brought into the joint
p̂ = Φ† b (17) probability density function. Bringing the candidate AOAs,
their corresponding positioning results, and their measurement
where Φ† is the pseudoinverse of Φ. noise variances into equation (19), their location reliability
In the LS positioning algorithm [5], only one UE position prior information can be acquired. The candidate AOA with
is acquired. However, combining the two peaks at NA ANs, the maximum location reliability prior information is the most
2NA estimated UE locations are acquired in this paper. reliable AOA estimation, and its corresponding positioning
results is the optimum location estimation.
III. T HE P ROPOSED P OSITIONING A LGORITHM
The measurement noise is assumed to be zero-mean Gaus-
From the above positioning algorithm, we get 2NA UE sian in the above joint probability density function. When
location estimation results. Then the location reliability prior calculating the above function, the matrix operations are
information related to them can be calculated. The location used. To reduce the computational complexity, the noise is
reliability prior information is defined by the joint probability assumed to be a zero-mean Laplacian in this paper. Then the
density function of the angle and the location. From the Fig. joint probability density function is a multivariate Laplacian
4 we can see that the measured ϕ̂i at location p is probability density function and can be written as follows
x o − xi
ϕ̂i = ϕi + ei , ϕi = tan−1 ( ) (18) f (Φ̂/p) = e−JL NA 1
(22)
yo − y i i=1 2λi
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on June 03,2021 at 22:20:46 UTC from IEEE Xplore. Restrictions apply.
,(((:LUHOHVV&RPPXQLFDWLRQVDQG1HWZRUNLQJ&RQIHUHQFH :&1&
where 2λ2i = σi2 and σi2 is the measured AOA noise variances. 30
The JL is as follows AN
UE
25
NA
|ϕ̂i − ϕi |
JL = (23) 20
i=1
λi
15
y(m)
Thus the estimated candidate AOAs, the corresponding UE
10
positions and the measured noise variances can also be brought
into the above joint probability density function. The AOA 5
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on June 03,2021 at 22:20:46 UTC from IEEE Xplore. Restrictions apply.
,(((:LUHOHVV&RPPXQLFDWLRQVDQG1HWZRUNLQJ&RQIHUHQFH :&1&
0.9
ACKNOWLEDGMENT
0.8 This study was supported by the Key Program of National
0.7 Natural Science Foundation of China (No. 92067202), the
0.6 General Program of National Natural Science Foundation of
Probability
Authorized licensed use limited to: University of Prince Edward Island. Downloaded on June 03,2021 at 22:20:46 UTC from IEEE Xplore. Restrictions apply.