Research Article
Open Access
NonSynchronous Sensor Network Positioning
Method Based On RSS and TOA Hybrid
Measurement
Lu Kangli^{1*}, Wu Xiaoping^{2}, LvYangmin^{3} and Hua Yuting^{4}
^{1}School of Information Engineering, Zhejiang Agriculture and Forestry University, Lin’an 311300, China
^{2}Zhejiang Key Laboratory of Forestry Intelligent Monitoring and Information Technology, Lin’an 311300, China
^{3}School of Information Engineering, Zhejiang Agriculture and Forestry University, Lin’an 311300, China
^{4}School of Information Engineering, Zhejiang Agriculture and Forestry University, Lin’an 311300, China
^{2}Zhejiang Key Laboratory of Forestry Intelligent Monitoring and Information Technology, Lin’an 311300, China
^{3}School of Information Engineering, Zhejiang Agriculture and Forestry University, Lin’an 311300, China
^{4}School of Information Engineering, Zhejiang Agriculture and Forestry University, Lin’an 311300, China
*Corresponding author: Lu Kangli, School of Information Engineering, Zhejiang Agriculture and Forestry University, Lin’an 311300, China, Email:
@
Received: August 21, 2018; Accepted: September 07, 2018; Published: September 21, 2018
Citation: Kangli Lu, Xiaoping Wu, Yangmin Lv, Hua Y (2018) NonSynchronous Sensor Network Positioning Method Based On RSS and TOA Hybrid Measurement. J Comp Sci Appl Inform Technol. 3(3): 19. DOI: 10.15226/24749257/3/3/00135
Abstract
The method of single rangebased measurement will lead to
an increase in the uncertainty of positioning results and a lower
positioning accuracy. To deal with the positioning problem in
three dimensional nonsynchronous sensor networks, an accurate
positioning method is proposed by using the hybrid Received
Signal Strength (RSS) and TimeofArrival (TOA) measurements.
The Unconstrained Linear Least Squares (ULLS) and Constrained
Linear Squares (CLLS) methods are put forward to obtain the node
position coordinates by converting the nonlinear optimization model
established by hybrid measurements into linear equations. The
accuracy of the designed algorithm was tested by simulation, and the
influence of different measurement noise on the estimation error of
the hybrid measurement method was analyzed. The results show that
the hybrid measurement positioning method has less error than the
single measurement method. The constrained CLLS is more accurate
than the unconstrained ULLS. The CLLS and ULLS linear estimation
methods have high stability and positioning accuracy under low noise
conditions.
Keywords: Positioning; Time synchronization; Received signal strength; Hybrid measurement;
Keywords: Positioning; Time synchronization; Received signal strength; Hybrid measurement;
Introduction
The Wireless Sensor Networks (WSN) use a large number
of static or mobile wireless transmission nodes for sensing and
collection. They have broad application prospects in the fields
of environmental monitoring, target tracking, military and
intelligent transportation systems, and have been concerned
and explored by many researchers [1,2]. The implementation of
functions such as target tracking, geographical location routing,
and fault alarm in sensor network applications is based on node
location [3,4]. Node location technology is one of the basic
support technologies of wireless sensor networks [5,6]. Sensor
network positioning technology has become a key factor in its
application promotion, and the node positioning technology with
superior development performance has become an important
research hotspot of sensor network positioning content. In
a real scenario, the sensor network area deployed by nodes
is usually in a 3D environment. However, most of the existing
localization algorithms are researched and optimized based on
2D space, which is different from the real environment, resulting
in the inability to locate specific locations with high precision.
Therefore, the research of 3D positioning algorithm is one of the
key contents to solve the problem of wireless sensor networks
node positioning in the future [7].
The use of known location beacon nodes to estimate the unknown node position coordinates is a common sensor network positioning method, which relies on some ranging methods between nodes, such as Time of Arrival (TOA) and Arrival Time Difference (TDOA), Angle of Arrival (AOA) and Signal Reception Strength (RSS) [811]. Since each node uses a separate clock module for timing, the clock module timing will drift with time and environmental parameters, so the timing of each clock module is not synchronized, which is called the asynchronous sensor network [12]. The singletime measurement method can solve the node position estimation of the asynchronous sensor network, but the single measurement method leads to the increase of uncertainty of the positioning result, the unreliable positioning result, and the low positioning accuracy, it is more difficult to apply to the positioning problem of the 3D sensor network.
The time measurement implementation principle is relatively simple. For example, the literature [13] uses multiple antenna receivers, and proposes a joint estimation method of node clock deviation and target position, which realizes the joint calculation of time synchronization and positioning. And the literature [14] introduced a hybrid measurement method combining TOA and AOA. The RSS measurement does not require additional hardware facilities, and the measurement cost is low, there is also a hybrid measurement method of RSS and TDOA proposed in literature [15]. Among the existing threedimensional positioning methods, some are derived from threedimensional positioning methods in threedimensional space, and some propose new positioning algorithms. For example, the literature [16] solving the problem of 3D location based on RSS and AOA measurements. Therefore, this paper introduces a hybrid RSS and TOA measurement method to solve the positioning problem in the 3D scene in the asynchronous sensor network.
With the increasing demand for location service applications, the positioning accuracy requirements are getting higher and higher. How to reduce the positioning error of positioning algorithm design has become a research hotspot in this field. By establishing a distanceconstrained optimization model between nodes, researchers have continuously proposed positioning algorithms for sensor networks, such as Maximum Likelihood (ML) estimation, linear algebra method and convex optimization, and so on [1719]. The numerical calculation method of ML estimation method relies on the selection of the initial solution, and has better precision in weak noise environment, but it may fall into local optimum. For this reason, linear algebra method and convex optimization method are proposed. The convex optimization method (including semidefinite programming, quadratic cone planning, etc.) relaxes the optimization model into a convex optimization problem, which is a popular method in the current sensor network positioning method, but the convex optimization function has more variables and equality constraint, and the computational complexity is high. Because of the relaxation, the estimation result is not optimal. In order to reduce the computational complexity, the linear algebra method directly represents the calculation result as an algebraic solution, which is faster in calculation and avoids the trouble caused by the initial value selection.
The TOA ranging method is used to establish the distance relationship between nodes by measuring the arrival time of the signal. The accuracy of the ranging depends on the accurate timing ability of the node clock. However, due to the change of environmental factors, may lead to the asynchrony of the timing clock of the node. In this paper, based on the chronograph clock model of signal transmission between nodes, the relationship between observation time and actual time is derived, and the joint estimation equation of time synchronization and positioning is established. Based on the node clock drift and deviation model, a joint linear estimation method of time synchronization and node localization is proposed to realize simultaneous estimation of node clock drift rate, deviation and position coordinates. The nodes communicate with each other in the form of electromagnetic waves, and the Received Signal Strength (RSS) is attenuated with the extension of the transmission path. The RSS method does not need additional hardware, is simple to implement, it has the characteristics of low power consumption, low cost, and the like, and is widely used. The sensor network node positioning is realized by the RSS and TOA hybrid measurement between the nodes, which can make the positioning result more reliable and the positioning accuracy higher. Since the unknown parameters of the 3D space are more than the 2D plane, the difficulty is also increased. The positioning method using the hybrid measurement is more reliable and accurate than the single ranging method, and is more suitable for the positioning problem of the asynchronous sensor network. Based on the RSS and TOA hybrid measurement technique between nodes, this paper proposes a precise positioning method for the wireless signal strength (RSS) and Time of Arrival (TOA) hybrid measurement of the 3D asynchronous sensor network. The nonlinear optimization model established by the hybrid measurement is transformed into a linear equation, and the Unconstrained Linear Least Squares (ULLS) and Constrained Linear Singularity (CLLS) methods for node position coordinate estimation are proposed respectively, and compared with the model of the CarmerRao Lower Bound( CRLB) values were compared. The design of hybrid measurement method for 3D spatial positioning depends on less number of beacon nodes and high positioning accuracy.
The first part of this paper begins with introduced the problem description of hybrid positioning of RSS and TOA; the second part deduces the calculation method of ULLS and CLLS; the third part deduces the lower bound of carmerRao (CRLB) of model; the fourth part is simulation and analysis; the last part is the conclusion.
The use of known location beacon nodes to estimate the unknown node position coordinates is a common sensor network positioning method, which relies on some ranging methods between nodes, such as Time of Arrival (TOA) and Arrival Time Difference (TDOA), Angle of Arrival (AOA) and Signal Reception Strength (RSS) [811]. Since each node uses a separate clock module for timing, the clock module timing will drift with time and environmental parameters, so the timing of each clock module is not synchronized, which is called the asynchronous sensor network [12]. The singletime measurement method can solve the node position estimation of the asynchronous sensor network, but the single measurement method leads to the increase of uncertainty of the positioning result, the unreliable positioning result, and the low positioning accuracy, it is more difficult to apply to the positioning problem of the 3D sensor network.
The time measurement implementation principle is relatively simple. For example, the literature [13] uses multiple antenna receivers, and proposes a joint estimation method of node clock deviation and target position, which realizes the joint calculation of time synchronization and positioning. And the literature [14] introduced a hybrid measurement method combining TOA and AOA. The RSS measurement does not require additional hardware facilities, and the measurement cost is low, there is also a hybrid measurement method of RSS and TDOA proposed in literature [15]. Among the existing threedimensional positioning methods, some are derived from threedimensional positioning methods in threedimensional space, and some propose new positioning algorithms. For example, the literature [16] solving the problem of 3D location based on RSS and AOA measurements. Therefore, this paper introduces a hybrid RSS and TOA measurement method to solve the positioning problem in the 3D scene in the asynchronous sensor network.
With the increasing demand for location service applications, the positioning accuracy requirements are getting higher and higher. How to reduce the positioning error of positioning algorithm design has become a research hotspot in this field. By establishing a distanceconstrained optimization model between nodes, researchers have continuously proposed positioning algorithms for sensor networks, such as Maximum Likelihood (ML) estimation, linear algebra method and convex optimization, and so on [1719]. The numerical calculation method of ML estimation method relies on the selection of the initial solution, and has better precision in weak noise environment, but it may fall into local optimum. For this reason, linear algebra method and convex optimization method are proposed. The convex optimization method (including semidefinite programming, quadratic cone planning, etc.) relaxes the optimization model into a convex optimization problem, which is a popular method in the current sensor network positioning method, but the convex optimization function has more variables and equality constraint, and the computational complexity is high. Because of the relaxation, the estimation result is not optimal. In order to reduce the computational complexity, the linear algebra method directly represents the calculation result as an algebraic solution, which is faster in calculation and avoids the trouble caused by the initial value selection.
The TOA ranging method is used to establish the distance relationship between nodes by measuring the arrival time of the signal. The accuracy of the ranging depends on the accurate timing ability of the node clock. However, due to the change of environmental factors, may lead to the asynchrony of the timing clock of the node. In this paper, based on the chronograph clock model of signal transmission between nodes, the relationship between observation time and actual time is derived, and the joint estimation equation of time synchronization and positioning is established. Based on the node clock drift and deviation model, a joint linear estimation method of time synchronization and node localization is proposed to realize simultaneous estimation of node clock drift rate, deviation and position coordinates. The nodes communicate with each other in the form of electromagnetic waves, and the Received Signal Strength (RSS) is attenuated with the extension of the transmission path. The RSS method does not need additional hardware, is simple to implement, it has the characteristics of low power consumption, low cost, and the like, and is widely used. The sensor network node positioning is realized by the RSS and TOA hybrid measurement between the nodes, which can make the positioning result more reliable and the positioning accuracy higher. Since the unknown parameters of the 3D space are more than the 2D plane, the difficulty is also increased. The positioning method using the hybrid measurement is more reliable and accurate than the single ranging method, and is more suitable for the positioning problem of the asynchronous sensor network. Based on the RSS and TOA hybrid measurement technique between nodes, this paper proposes a precise positioning method for the wireless signal strength (RSS) and Time of Arrival (TOA) hybrid measurement of the 3D asynchronous sensor network. The nonlinear optimization model established by the hybrid measurement is transformed into a linear equation, and the Unconstrained Linear Least Squares (ULLS) and Constrained Linear Singularity (CLLS) methods for node position coordinate estimation are proposed respectively, and compared with the model of the CarmerRao Lower Bound( CRLB) values were compared. The design of hybrid measurement method for 3D spatial positioning depends on less number of beacon nodes and high positioning accuracy.
The first part of this paper begins with introduced the problem description of hybrid positioning of RSS and TOA; the second part deduces the calculation method of ULLS and CLLS; the third part deduces the lower bound of carmerRao (CRLB) of model; the fourth part is simulation and analysis; the last part is the conclusion.
Problem Description
The coordinates of unknown nodes are calculated in twodimensional
plane. Suppose that there are N beacon nodes with
known position coordinates, and their coordinate positions are
respectively ${x}_{i}={[\begin{array}{cc}{x}_{i}& {y}_{i}\end{array}]}^{T}\text{}(i=1,2,\cdots ,N)$.
At the same time, there
is an unknown node to be located in the area, and the position
coordinate is assumed to be $x\text{=[}\begin{array}{cc}x& y\end{array}{\text{]}}^{T}$. Using the ranging method
to obtain the coordinates of the unknown node, it is obvious
that this ignores the gradient difference between the unknown
node and the beacon node, and it is unrealistic. The coordinate
parameters of the spatial position of the 3D sensor network
space are increased compared with the 2D plane, including three
directions of x, y, and z. Increasing the gradient of Z axis can make
more accurate calculation. Assuming that the coordinate positions
of the beacon a node of the N known position coordinates are
respectively ${x}_{i}={[\begin{array}{ccc}{x}_{i}& {y}_{i}& {z}_{i}\end{array}]}^{T}(i=1,2,\cdots ,N)$. At the same time,
the location coordinates of the unknown node to be located in the
area is assumed to be $x\text{=[}\begin{array}{ccc}x& y& z\end{array}{\text{]}}^{T}$. (Figure 1)
Figure 1: Schematic diagram of unknown node and beacon node model
To estimate the location of an unknown node, the Time of
Arrival (TOA) of the signal is measured to represent the spatial
distance between the unknown node and the beacon nodes.
However, in the nonsynchronous sensor network, the node
timing clock module is not synchronized due to environmental
impact and other reasons, resulting in the node observation time
being inconsistent with the actual real time. In order to improve
the accuracy of the measurement results, this paper uses the drift
and deviation clock timing model to represent the relationship
between observation time and actual real time.
$\omega $ and $\theta $ respectively represent the drift rate and the deviation amount
of the clock, and represent the time change rate and the time
difference between the observation time and the actual real time
on the unknown node. The observation time T and the actual real
time t of the unknown node are expressed as a relation:
$$T=\omega t+\theta \text{(1)}$$
The clock drift rate and the deviation amount of the beacon
nodes i are assumed to be ${\omega}_{i},\text{}{\theta}_{i}$. According to the same principle,
the observation time and the real time of the beacon nodes i
also have the same relationship as equation (1). In this model,
it is assumed that the clock parameters of the beacon nodes,
including the drift rate ${\omega}_{i}$ and the deviation amount
${\theta}_{i}$ are
calibrated, and are known parameters, while the unknown node
clock parameters $\omega $ and $\theta $ are unknown parameters.
The unknown node sends a signal to the beacon nodes i at ${T}_{1,i}$ time (the observation time of the unknown node clock at ${T}_{1,i}$) and the beacon nodes i receive the signal of the unknown node at ${R}_{1,i}$ time (the observation time of the beacon nodes i clock at ${R}_{1,i}$), According to equation (1) and the signal transmission process has the following relationship:
$$\frac{{R}_{1,i}}{{\omega}_{i}}\frac{{T}_{1,i}}{\omega}+\frac{\theta}{\omega}\frac{{\theta}_{i}}{{\omega}_{i}}={t}_{i}+{m}_{i}\text{(2)}$$
The unknown node sends a signal to the beacon nodes i at ${T}_{1,i}$ time (the observation time of the unknown node clock at ${T}_{1,i}$) and the beacon nodes i receive the signal of the unknown node at ${R}_{1,i}$ time (the observation time of the beacon nodes i clock at ${R}_{1,i}$), According to equation (1) and the signal transmission process has the following relationship:
In equation (2) ${t}_{i}$, is the Signal arrival time between nodes
${t}_{i}={d}_{i}/c,\text{}{d}_{i}$ is the spatial distance between the unknown node
and the beacon nodes i, c is the propagation speed of the signal,
$c=3\times {10}^{8}$ m/s ${m}_{i}$ represents the time measurement noise
between the unknown node and the beacon nodes i. It can be
assumed that ${m}_{i}$ obeys the Gaussian distribution with a mean of
zero, and a variance of ${\delta}_{i,m}^{2}$, which is denoted as ${m}_{i}\in N(0,{\delta}_{i,m}^{2})$.
When the beacon nodes i receive the signal of the unknown node, the beacon nodes i send a signal to the unknown node at ${T}_{2,i}$ time (${T}_{2,i}$ is the observation time of the beacon nodes i clock), and the unknown node at ${R}_{2,i}$ time (${R}_{2,i}$ is the observation time of the unknown node clock) receives the signal of the beacon nodes, a similar derivation process of the equation (2), there is:
$$\frac{{R}_{2,i}}{\omega}\frac{{T}_{2,i}}{{\omega}_{i}}+\frac{{\theta}_{i}}{{\omega}_{i}}\frac{\theta}{\omega}={t}_{i}+{n}_{i}\text{(3)}$$
When the beacon nodes i receive the signal of the unknown node, the beacon nodes i send a signal to the unknown node at ${T}_{2,i}$ time (${T}_{2,i}$ is the observation time of the beacon nodes i clock), and the unknown node at ${R}_{2,i}$ time (${R}_{2,i}$ is the observation time of the unknown node clock) receives the signal of the beacon nodes, a similar derivation process of the equation (2), there is:
In the formula, the definition of
${t}_{i}$ is the same as equation (2),
and ${n}_{i}$ is also the noise part. It is assumed that
${n}_{i}$ obeys the mean
of zero, and the Gaussian distribution of variance
${\delta}_{i,n}^{2}$ is , recorded as
${n}_{i}\in N(0,{\delta}_{i,n}^{2})$.
It can be seen from equations (2) and (3) that the clock parameters and the position parameters are intertwined, and the clock parameter estimation and positioning have commonality in time characteristics. For this reason, the clock parameter estimation and the position parameters can be jointly estimated, and carried out to realize both the clock parameters estimation and and the position estimation. Since the positional parameter in the 3D sensor network space has more zdirection than the 2D plane, the method of estimating the position coordinates of the unknown node only by measuring the arrival time of the signal between the unknown node and the beacon nodes, and the measurement result is not accurate. So the (RSS) positioning method is added to measure the signal strength of unknown node and beacon nodes. Express the RSS measurement value between the unknown node and the beacon node i as. ${p}_{i}$ According to the logarithmic attenuation model of RSS ranging, there is the following RSS relation expression:
$${p}_{i}={p}_{0}10\beta \mathrm{lg}{d}_{i}+{\epsilon}_{i}\text{(4)}$$
It can be seen from equations (2) and (3) that the clock parameters and the position parameters are intertwined, and the clock parameter estimation and positioning have commonality in time characteristics. For this reason, the clock parameter estimation and the position parameters can be jointly estimated, and carried out to realize both the clock parameters estimation and and the position estimation. Since the positional parameter in the 3D sensor network space has more zdirection than the 2D plane, the method of estimating the position coordinates of the unknown node only by measuring the arrival time of the signal between the unknown node and the beacon nodes, and the measurement result is not accurate. So the (RSS) positioning method is added to measure the signal strength of unknown node and beacon nodes. Express the RSS measurement value between the unknown node and the beacon node i as. ${p}_{i}$ According to the logarithmic attenuation model of RSS ranging, there is the following RSS relation expression:
In equation (4),
$\beta $ is the path attenuation index, The
$\beta $ value
can be obtained through experimental analysis in advance, and
the value varies with the change of the propagation medium,
generally between 2 and 5.
${p}_{o}$ is the signal emission intensity of
the unknown node, which is related to the antenna gain of the
node and the battery supply. ${p}_{i}$ is the signal reception strength
of the beacon nodes i.
${d}_{i}$ is the spatial linear distance between the
unknown node to be located and the beacon nodes i,
${d}_{i}=\Vert x{x}_{i}\Vert ,\text{}{\epsilon}_{i}$ represents the measurement noise between the unknown
node and the beacon nodes i, and it can be assumed that
${\varphi}_{i}$ obeys
the Gaussian distribution with a mean of zero, and variance is
${\delta}_{i,\epsilon}^{2}$, which is recorded as
${\epsilon}_{i}\in N(0,{\delta}_{i,\epsilon}^{2})$.
To estimate the unknown node position coordinate $x$ with TOA and RSS hybrid measurements as known values, and the maximum likelihood (ML) estimates by optimizing the following expressions:
$$\underset{x}{\mathrm{arg}}\mathrm{min}{\displaystyle \sum _{i=1}^{N}(\frac{1}{{\delta}_{i,m}^{2}}}{r}_{i,m}^{2}+\frac{1}{{\delta}_{i,n}^{2}}{r}_{i,n}^{2}+\frac{1}{{\delta}_{i,\epsilon}^{2}}{r}_{i,\epsilon}^{2})\text{(5)}$$
To estimate the unknown node position coordinate $x$ with TOA and RSS hybrid measurements as known values, and the maximum likelihood (ML) estimates by optimizing the following expressions:
In equation (5),
${r}_{i,m},\text{}{r}_{i,n},\text{}{r}_{i,\epsilon}$ respectively represent
the error items of the two arrival time measurements
and the wireless received signal strength measurement.
${r}_{i,m}=\frac{{R}_{1,i}}{{\omega}_{i}}\frac{{T}_{1,i}}{\omega}+\frac{\theta}{\omega}\frac{{\theta}_{i}}{{\omega}_{i}}{t}_{i},\text{}{r}_{i,n}=\frac{{R}_{2,i}}{\omega}\frac{{T}_{2,i}}{{\omega}_{i}}+\frac{{\theta}_{i}}{{\omega}_{i}}\frac{\theta}{\omega}{t}_{i},$
${r}_{i,\epsilon}={p}_{i}{p}_{o}+10\beta \mathrm{lg}{d}_{i}$
The optimization function established
by equation (5) is a nonlinear equation, and its numerical
calculation method depends on the selection of the initial value.
If the initial value selection is not suitable, it may fall into local
optimum, which causes serious deviation of the positioning
result. Therefore, this paper converts the nonlinear optimization
equation is transformed into a linear equation, and the twostep
calculation method is used to accurately calculate the position
coordinates and clock parameters of the unknown node.
Linear Least Squares Estimation Method
Considering the approximate linearization of the
measurement equation under small noise conditions, the
nonlinear optimization model described in equation (5) is
transformed into a linear equation. The designed calculation
method is divided into two steps: the Unconstrained Linear
Least Squares (ULLS) method and the Constrained Linear Least
Squares (CLLS) method.
Unconstrained Least Squares Method
Adding equations (2) and (3) to eliminate common items:
$$\frac{{R}_{2,i}{T}_{1,i}}{\omega}+\frac{{R}_{1,i}{T}_{2,i}}{{\omega}_{i}}=2{t}_{i}+{m}_{i}+{n}_{i}\text{(6)}$$
Since $\omega $ is very close to 1, it can be assumed that
$\omega =\frac{1}{1+\delta},\text{}\delta $
is a variable close to zero, so equation (6) can be rewritten as:
$${\mu}_{i}\delta +{\lambda}_{i}=2{t}_{i}+{m}_{i}+{n}_{i}\text{(7)}$$
In equation (7),
${\lambda}_{i}=\frac{{R}_{1,i}{T}_{2,i}}{{\omega}_{i}}+{R}_{2,i}{T}_{1,i},\text{}{\mu}_{i}={R}_{2,i}{T}_{1,i}$ it can be
seeing that ${\lambda}_{i}$ and ${\mu}_{i}$ are known parameters, since
${t}_{i}={d}_{i}/c$ and ${d}_{i}=\left\rightx{x}_{i}\left\right$ , square sides of the equation (7), ignoring the
second highorder term, and sorting the expression:
$$8{x}_{i}x4{x}^{T}x+2{c}^{2}{\mu}_{i}{\lambda}_{i}\delta =4{x}_{i}^{T}{x}_{i}{c}^{2}{\lambda}_{i}^{2}+{c}^{2}{d}_{i}({m}_{i}+{n}_{i})\text{(8)}$$
Equation (8) shows the transformed TOA measurement
equation, $i=1,2,\cdots ,N$. Let $z={\left[\begin{array}{cc}\begin{array}{cc}x& {x}^{T}x\end{array}& \delta \end{array}\right]}^{T}$, equation (8) can be
written as a linear matrix:
$${A}_{1}z={b}_{1}+{\gamma}_{1}\text{(9)}$$
In equation (9), the row vectors of matrix
${A}_{1}$ of dimension
$N\times 5$ is $[\begin{array}{ccc}8{x}_{i}& 4& 2{c}^{2}{\mu}_{i}{\lambda}_{i}\end{array}]$, the column vector
${b}_{1}$ and the row
element values of ${\gamma}_{1}$
of N row are
$[4{x}_{i}^{T}{x}_{i}{c}^{2}{\lambda}_{i}^{2}]$ and
$[{c}^{2}{d}_{i}({m}_{i}+{n}_{i})]$ respectively. Equivalent transformation of the RSS measurement
equation, that is, equivalent shift transformation of equation (4),
rewriting it to:
$${d}_{i}={10}^{\frac{{p}_{0}{p}_{i}+{\epsilon}_{i}}{10\beta}}\text{(10)}$$
Considering that in the smaller noise range, the Taylor series
expansion is applied to the right side of equation (10), ignore the
high order term, and equation (10) is transformed into:
$${d}_{i}={\tau}_{i}+\frac{{\tau}_{i}\mathrm{ln}10}{10\beta}{\epsilon}_{i}\text{(11)}$$
In equation (11), $i=1,2,\cdots ,N$,
${\tau}_{i}={10}^{\frac{{p}_{0}{p}_{i}}{10\beta}}$ Square the two
sides of the equation (11), omitting the highorder terms, it can
get:
$${d}_{i}^{2}={\tau}_{i}^{2}+\frac{{\tau}_{i}^{2}\mathrm{ln}10}{5\beta}{\epsilon}_{i}\text{(12)}$$
Since ${d}_{i}=\left\rightx{x}_{i}\left\right$, there is an expression for expansion of
equation (12):
$$2{x}_{i}{}^{T}x+{x}^{T}x={x}_{i}{}^{T}{x}_{i}+{\tau}_{i}{}^{2}+\frac{{\tau}_{i}{}^{2}\mathrm{ln}10}{5\beta}{\epsilon}_{i}\text{(13)}$$
Equation (13) shows the transformed RSS measurement
equation, $i=1,2,\cdots ,N$, and the same equation (13) can be written
as a linear matrix form:
$${A}_{2}z={b}_{2}+{\gamma}_{2}\text{(14)}$$
In equation (14), the row vector of matrix ${A}_{2}$ of dimension $N\times 5$ is $[\begin{array}{cc}2{x}_{i}{}^{T}& \begin{array}{cc}1& 0\end{array}\end{array}]$. The column vector
${b}_{2}$ and the row element
values of ${\gamma}_{2}$
of N row are
$[{x}_{i}^{T}{x}_{i}+{\tau}_{i}^{2}],\text{}[\frac{{\tau}_{i}^{2}\mathrm{ln}10}{5\beta}{\epsilon}_{i}]$ Establish a unified
matrix form by combining equations (9) and (14):
$$Az=b+\gamma \text{(15)}$$
In equation (15),
$A={[\begin{array}{cc}{A}_{1}^{T}& {A}_{2}^{T}\end{array}]}^{T},\text{}b={[\begin{array}{cc}{b}_{1}^{T}& {b}_{2}^{T}\end{array}]}^{T},\text{}\gamma ={[\begin{array}{cc}{\gamma}_{1}^{T}& {\gamma}_{2}^{T}\end{array}]}^{T}$ and
$A\in {R}^{2N\times 5},\text{}b\in {R}^{2N},\text{}v\in {R}^{2N}$.
According to the principle of least squares, the estimated value of parameters $z$ is:
$$z={({A}^{T}{\Sigma}_{\gamma}^{1}A)}^{1}{A}^{T}{\Sigma}_{\gamma}^{1}b\text{(16)}$$
According to the principle of least squares, the estimated value of parameters $z$ is:
In equation (16), the matrix
${\sum}_{\gamma}$ with a dimension of
$2N\times 2N$ is a weight matrix, whose value is
${\sum}_{\gamma}=\Sigma \left({\gamma}^{T}\gamma \right)$. Representing the
variance of ${\gamma}^{T}\gamma $ and its value is further expressed as:
$${\sum}_{\gamma}=\left[\begin{array}{cc}\Sigma \left({\gamma}_{1}^{T}{\gamma}_{1}\right)& \Sigma \left({\gamma}_{1}^{T}{\gamma}_{2}\right)\\ \Sigma \left({\gamma}_{2}^{T}{\gamma}_{1}\right)& \Sigma \left({\gamma}_{2}^{T}{\gamma}_{2}\right)\end{array}\right]\text{(17)}$$
Because
${\gamma}_{1}$ and ${\gamma}_{2}$ are not related, there are
$\Sigma \left({\gamma}_{1}^{T}{\gamma}_{2}\right),\text{}\Sigma \left({\gamma}_{2}^{T}{\gamma}_{1}\right)$ equal to zero, will ${\gamma}_{1},\text{}{\gamma}_{2}$ are substituted into each equation, the
other values of are as follows ${\sum}_{\gamma}$
$$\{\begin{array}{c}\Sigma \left({\gamma}_{1}^{T}{\gamma}_{1}\right)=\text{diag}\left\{\frac{{\tau}_{i}^{4}}{4.7{\beta}^{2}}{\delta}_{i,\epsilon}^{2}\right\}\\ \Sigma \left({\gamma}_{2}^{T}{\gamma}_{2}\right)F\left({\nu}_{2}^{T}{\nu}_{2}\right)=\text{diag}\left\{{c}^{4}{d}_{i}^{2}{\left({\delta}_{i,m}+{\delta}_{i,n}\right)}^{2}\right\}\end{array}\text{(18)}$$
Assume the estimated error $\Delta z$ of the parameter $z$, the value
is:
$$\Delta z={({A}^{T}{\Sigma}_{\gamma}{}^{1}A)}^{1}{A}^{T}{\Sigma}_{\gamma}{}^{1}\gamma \text{(19)}$$
The variance of the estimation error $\Delta z$ is expressed as:
$$\mathrm{cov}(\Delta z)={({A}^{T}{\sum}_{\gamma}^{1}A)}^{1}\text{(20)}$$
Extracting $z(1:3)$ slave from the parameter $z$ is the position
coordinates of the unknown node being located. The above
solution process is not considered the mutually constrained
relationship between the fourth element and the first three
element values in $z={\left[\begin{array}{cc}\begin{array}{cc}x& {x}^{T}x\end{array}& \delta \end{array}\right]}^{T}$, so the calculation method is
called the unconstrained least squares (ULLS) method of the RSS
and TOA hybrid positioning problem. Equation (16) obtains an
approximate estimate of the coordinates of the unknown node
being located, and the exact value of the unknown coordinates of
the unknown node can be calculated using the mutual constraint
relation between the vector $z={\left[\begin{array}{cc}\begin{array}{cc}x& {x}^{T}x\end{array}& \delta \end{array}\right]}^{T}$ elements. Since the
fifth element and the first four elements are not constrained, this
article only selects the constraint relationship of the first four
elements. The clock drift rate $\omega $ can be further solved by $\omega =\frac{1}{1+\delta}$, the equation (2) subtract the equation (3) can further calculate
the clock offset $\theta $.
Constrained Least Squares
The results obtained by the ULLS method are further
optimized by using the constraint relationship of each element
value in the vector $z$. Assume that the true value of the coordinate
position of the unknown node is ${x}^{o}={[\begin{array}{ccc}{x}^{o}& {y}^{o}& {z}^{o}\end{array}]}^{T}$, there are
the following relationships:
$$\{\begin{array}{c}{x}^{o2}={[z(1)+\Delta z(1)]}^{2}\approx z{(1)}^{2}+2z(1)\Delta z(1)\\ {y}^{o2}={[z(2)+\Delta z(2)]}^{2}\approx z{(2)}^{2}+2z(2)\Delta z(2)\\ {z}^{o2}={[z(3)+\Delta z(3)]}^{2}\approx z{(3)}^{2}+2z(3)\Delta z(3)\\ {x}^{o2}+{y}^{o2}+{z}^{o2}=z(4)+\Delta z(4)\end{array}\text{(21)}$$
In equation (21), $z(k),\text{}\Delta z(k)$ represent the kth element of
the vector $z,\text{}\Delta z,\text{}k=1,2,3,4$. Express equation (21) as a linear
matrix form:
$$G{\mu}^{o}=h+\eta \text{(22)}$$
In equation
$h={[\begin{array}{cccc}z{(1)}^{2}& z{(2)}^{2}& z{(3)}^{2}& z{(4)}^{2}\end{array}]}^{T},\text{}{\mu}^{o}={[\begin{array}{ccc}{x}^{o2}& {y}^{o2}& {z}^{o2}\end{array}]}^{T},\text{}\eta =L\Delta z$,
$L=\text{diag}\left\{\begin{array}{cccc}2z(1)& 2z(2)& 2z(3)& 1\end{array}\right\}$
$$G={\left[\begin{array}{cccc}1& 0& 0& 1\\ 0& 1& 0& 1\\ 0& 0& 1& 1\end{array}\right]}^{T}$$
$L=\text{diag}\left\{\begin{array}{cccc}2z(1)& 2z(2)& 2z(3)& 1\end{array}\right\}$
According to the principle of linear least squares, the unbiased
estimate of the vector ${\mu}^{o}$ is:
$${\mu}^{o}={({G}^{1}{\Sigma}_{\eta}^{1}G)}^{1}G{\Sigma}_{\eta}^{1}h\text{(23)}$$
The weight matrix ${\Sigma}_{\eta}$ value of the dimension $4\times 4$ in equation
(23) is:
$${\Sigma}_{\eta}=\Sigma ({\eta}^{T}\eta )={L}^{T}\mathrm{cov}(\Delta z)L={L}^{T}{({A}^{T}{\sum}_{\gamma}^{1}A)}^{1}L\text{(24)}$$
Due to then the exact estimated value of the coordinates of
the location node being located is
$$x=\text{sign}\left\{\text{diag}\left\{z\left(1:3\right)\right\}\right\}\sqrt{{\mu}^{o}}\text{(25)}$$
The calculation process represented by the equation (25)
takes into account the mutual constraint relationship between
the elements in the parameter z, and obtains accurate positioning
results. This calculation method is called the constrained linear
least squares (CLLS) method of the asynchronous sensor network
positioning problem of the RSS and TOA hybrid measurement
methods.
Model of the CramerRao Lower Bound (CRLB)
The CRLB value provides the lower bound of the error variance
for the unbiased estimation of the model to be estimated, and the
unknown parameter of the model is assumed to be $\rho =\left[\begin{array}{ccc}x& \omega & \theta \end{array}\right]$. Assuming that the estimated error variance of unknown
parameters $\rho $ is $\mathrm{cov}(\rho )$. Then according to CramerRao lower
bound theory, there is a relationship
$\mathrm{cov}(\rho )\ge {F}^{1}$, among them $F$ is the representation of the FIM (Fisher Information Matrix) of
parameter to be evaluated $\rho $, the matrix $F$ expressed as:
$$F=\frac{{\partial}^{2}\mathrm{ln}P(\text{p}\rho )}{\partial {\rho}^{T}\rho}\text{(26)}$$
The model is solved by RSS, TOA measures parameters,
and maximizing the matrix $F$. The probability density function
$\mathrm{ln}P(\text{p}\rho )$ can be expressed as:
$$P(\text{p}\rho )={\displaystyle \prod _{i=1}^{N}\frac{1}{\sqrt{2\pi}{\delta}_{i,m}}}\mathrm{exp}(\frac{{r}_{i,m}^{2}}{2{\delta}_{i,m}^{2}})\frac{1}{\sqrt{2\pi}{\delta}_{i,n}}\mathrm{exp}(\frac{{r}_{i,n}^{2}}{2{\delta}_{i,n}^{2}})\frac{1}{\sqrt{2\pi}{\delta}_{i,\epsilon}}\mathrm{exp}(\frac{{r}_{i,\epsilon}^{2}}{2{\delta}_{i,\epsilon}^{2}})\text{(27)}$$
Take the logarithm of both sides of equation (27) and
define the vector ${r}_{m},\text{}{r}_{n}$,
${r}_{\epsilon},\text{\hspace{0.17em}}{r}_{m}=[\begin{array}{cccc}{r}_{1,m}& {r}_{2,m}& \cdots & {r}_{N,m}\end{array}]$,
${r}_{n}=[\begin{array}{cccc}{r}_{1,n}& {r}_{2,n}& \cdots & {r}_{N,m}\end{array}],\text{\hspace{0.17em}}{r}_{\epsilon}=[\begin{array}{cccc}{r}_{1,\epsilon}& {r}_{2,\epsilon}& \cdots & {r}_{N,\epsilon}\end{array}]$ with the
following expression:
$$\mathrm{ln}P\left(p\rho \right)=e({r}_{m}^{T}{\Sigma}_{m}^{1}{r}_{m}+{r}_{n}^{T}{\Sigma}_{n}^{1}{r}_{n}+{r}_{\epsilon}^{T}{\Sigma}_{\epsilon}^{1}{r}_{\epsilon})\text{(28)}$$
In equation (28),
${\Sigma}_{m}=\text{diag}\left\{{\delta}_{i,m}^{2}\right\},\text{}{\Sigma}_{n}=\text{diag}\left\{{\delta}_{i,n}^{2}\right\},$
${\Sigma}_{\epsilon}=\text{diag}\left\{{\delta}_{i,\epsilon}^{2}\right\}\text{}i=1,2,\cdots ,N$ e is the constant part. Substituting equation (28) into equation (26), and the matrix
$F$ can be expressed as:
$$F={(\frac{\partial {r}_{m}}{\partial \rho})}^{T}{\displaystyle {\sum}_{m}^{1}\frac{\partial {r}_{m}}{\partial \rho}}+{(\frac{\partial {r}_{n}}{\partial \rho})}^{T}{\displaystyle {\sum}_{n}^{1}\frac{\partial {r}_{n}}{\partial \rho}}+{(\frac{\partial {r}_{\epsilon}}{\partial \rho})}^{T}{\displaystyle {\sum}_{\epsilon}^{1}\frac{\partial {r}_{\epsilon}}{\partial \rho}}\text{(29)}$$
In equation (29), $\frac{\partial {r}_{m}}{\partial \rho},\text{}\frac{\partial {r}_{n}}{\partial \rho},\text{}\frac{\partial {r}_{\epsilon}}{\partial \rho}$ are respectively the differential
of each vector to the unknown parameter $\rho $, which can be written
as follows:
$$\{\begin{array}{c}\frac{\partial {r}_{m}}{\partial \rho}={\left[\begin{array}{cccc}{r}_{1,m}^{\rho}& {r}_{2,m}^{\rho}& \cdots & {r}_{N,m}^{\rho}\end{array}\right]}^{T}\\ \frac{\partial {r}_{n}}{\partial \rho}={\left[\begin{array}{cccc}{r}_{1,n}^{\rho}& {r}_{2,n}^{\rho}& \cdots & {r}_{N,n}^{\rho}\end{array}\right]}^{T}\\ \frac{\partial {r}_{\epsilon}}{\partial \rho}={\left[\begin{array}{cccc}{r}_{1,\epsilon}^{\rho}& {r}_{2,\epsilon}^{\rho}& \cdots & {r}_{N,\epsilon}^{\rho}\end{array}\right]}^{T}\end{array}\text{(30)}$$
In equation (30),
${r}_{i,m}^{\rho},\text{}{r}_{i,n}^{\rho},\text{}{r}_{i,\epsilon}^{\rho}$ further expansion of the
solution is expressed :
$$\{\begin{array}{c}{r}_{i,m}^{\rho}=\frac{\partial {r}_{i,m}}{\partial \rho}=\left[\begin{array}{ccc}\frac{{x}_{i}x}{c{d}_{i}}& \frac{{T}_{1,i}\theta}{{\omega}^{2}}& \frac{1}{\omega}\end{array}\right]\\ {r}_{i,n}^{\rho}=\frac{\partial {r}_{i,n}}{\partial \rho}=\left[\begin{array}{ccc}\frac{{x}_{i}x}{c{d}_{i}}& \frac{\theta {R}_{2,i}}{{\omega}^{2}}& \frac{1}{\omega}\end{array}\right]\\ {r}_{i,\epsilon}^{\rho}=\frac{\partial {r}_{i,\epsilon}}{\partial \rho}=\left[\begin{array}{ccc}\frac{4.34\beta {(x{x}_{i})}^{T}}{{d}_{i}^{2}}& 0& 0\end{array}\right]\end{array}\text{(31)}$$
Then according to the CRLB unbiased estimate lower bound
theory has
$$\text{CRLB}({[\rho ]}_{p})={[{F}^{1}]}_{p,p}\text{(32)}$$
In the equation (32) $p=1,2,\cdots ,5,\text{}{[{F}^{1}]}_{p,p}$ represents the
$p$ th row and the $p$ th column element value of the inverse matrix
of F; $\text{CRLB}({[\rho ]}_{p})$ indicates the lower bound of the CRLB unbiased
to estimate of the $p$ row element of unknown parameters $\rho $.
Simulation Analysis
In this section, numerical results will be provided to verify
and compare the feasibility of the proposed nonsynchronous
sensor network positioning method based on RSS and TOA
hybrid measurements. According to the calculation method of
the above design, the hybrid positioning algorithm is simulated by
MATLAB software, and the simulation results are more intuitively
displayed and analyzed. Assume that the clock of the set beacon
nodes are synchronized, that are ${\omega}_{i}=1,\text{}{\theta}_{i}=0,\text{}i=1,2,\cdots ,N$, while
the unknown node has clock drift and deviation. The time
measurement noise variance ${\delta}_{i,m}$ with ${\delta}_{i,n}$, RSS noise variance
${\delta}_{i,\epsilon}$, signal emission intensity
${p}_{0}=45dB$, path attenuation index
$\beta =4$ are set in advance between the unknown nodes and beacon
nodes i. The accuracy of the positioning error of the designed
algorithm is based on the Root Mean Square Error (RMSE)
analysis. The simulation runs 1000 times, and the average value
of the results is used for error analysis.
The influence of the number of beacon nodes on the
positioning error
Assuming that the simulation environment is in a 3D region
of 100 m × 100 m × 100 m, the coordinates of the unknown node
is set to (50, 50, 50) in advance. Because the quantity of beacon
nodes is too few lead to the positioning accuracy is too low, so the
number of beacon nodes is set as
$N=4,5,\cdots ,8$. Table 1 compares
the Unconstrained Linear Least Squares (ULLS) and constrained
least squares (CLLS) positioning accuracy of the three different
methods of RSS, TOA, RSS/TOA, and the CRLB value is compared
as a system benchmark.
Table 1: Comparison of the influence of the number of beacon nodes on the positioning error
4 
5 
6 
7 
8 

ULLS 
RSS 
0.873 
0.628 
0.531 
0.456 
0.377 
TOA 
0.718 
0.573 
0.359 
0.272 
0.185 

RSS/TOA 
0.141 
0.113 
0.109 
0.089 
0.075 

CLLS 
RSS 
0.591 
0.504 
0.438 
0.356 
0.271 
TOA 
0.405 
0.311 
0.237 
0.188 
0.139 

RSS/TOA 
0.063 
0.039 
0.033 
0.028 
0.025 

CRLB 
RSS 
0.512 
0.425 
0.351 
0.288 
0.204 
TOA 
0.323 
0.255 
0.189 
0.132 
0.093 

RSS/TOA 
0.047 
0.024 
0.021 
0.019 
0.018 
Table 1 lists 1000 simulations under the premise that
the time measurement noise variance
${\delta}_{i,m},\text{}{\delta}_{i,n}$ and RSS noise
variance ${\delta}_{i,\epsilon}$ are all 0.2, and compares the positioning errors
of the ULLS and CLLS calculation methods mentioned above in
the paper and compared them with CRLB. It can be seen from
Table 1 that with the increase of the number of beacon nodes, the
measurement connection relationship between nodes increases,
and both algorithms and CRLB values decrease accordingly. For
example, when the number of beacon nodes is 5, the positioning
error of the single RSS in the ULLS method is 0.628 m, and that
in the CLLS method is 0.504 m; the positioning error of the single
TOA in the ULLS method is 0.573 m, and that in the CLLS method
is 0.311 m; The positioning error of the hybrid measured RSS and
TOA in the ULLS method is 0.425 m, and that in the CLLS method
is 0.255 m. This shows that in the case of the same number of
beacon nodes, whether in the unconstrained linear least squares
or constrained least squares method, the error of the hybrid
measurement method is smaller than the single measurement
method (RSS or TOA), which indicates the hybrid measurement
method is more accurate. It can also be seen from Table 1 that the
proposed hybrid method yield results very close to CRLB, which
confirms the feasibility and accuracy of the hybrid measurement
method, proposed in this paper.
Influence of noise on positioning error
In the linear estimation method proposed in this paper, each
noise of time measurement and signal strength measurement has
independence, which leads to a decrease in the accuracy of the
positioning method. In order to test the estimation error of the
ULLS and CLLS calculation methods proposed in this paper, the
positioning accuracy of the algorithm is compared with CRLB.
And the results are compared with the SDP method introduced
in literature [16]. In order to make the simulation results more
representative, the number of beacon nodes is selected to be 5, and
keeping the time measurement noise variance
${\delta}_{i,m}$ with
${\delta}_{i,n}$ both
are 0.2 ns, while adjusting RSS measurement noise
${\delta}_{i,\epsilon}$
varies from
0.2 dB to 2 dB. Figure 2 depicts the relation of RMSE positioning
error with RSS measurement noise under different algorithms.
As shown in figure 2, the value of the RSS measurement noise has
a great influence on the positioning accuracy of the algorithm.
With the increases of the RSS measurement noise, the positioning
error also increases. When the RSS measurement noise is equal to
0.2 dB, the positioning error RMSE of the designed ULLS method
is 0.131 m, and the positioning error of the CLLS method is only
0.039 m. The constrained CLLS method is more accurate than
the ULLS method, and its positioning result is closer to the CRLB
value. Compared with the SDP method, the positioning error of
the ULLS method is larger, and the positioning error of the CLLS
method is smaller than that of the SDP method. When the RSS
measurement noise is equal to 0.2 dB, the positioning error of
SDP method is 0.075 m, which is between the positioning error of
ULLS and CLLS method.
Figure 2: The effect of RSS measurement noise on positioning error
Similarly, keep adjusting RSS measurement noise
${\delta}_{i,n}$ to 0.2
dB, the time measurement noise variance
${\delta}_{i,m}$ with ${\delta}_{i,n}$
both vary
from 0.2 ns to 2 ns at the same time. It can be seen from figure
3 that the RMSE positioning error under different algorithms
increases as the TOA measurement noise increases. Compared
with the influence of RSS measurement noise on positioning
error, TOA measurement noise has less influence on positioning
error. When the TOA measurement noise is equal to 0.2 ns,
the positioning error RMSE of the designed ULLS method is
0.107 m, the positioning error of the SDP method is 0.058, and
the positioning error of the CLLS method is only 0.031 m. The
constrained CLLS method is closer to the CRLB than the ULLS
method. The same analysis results as in Fig. 2 show that the
positioning error using the constrained CLLS method is smaller
than that of the ULLS and SDP methods.
Figure 3: The effect of TOA measurement noise on positioning error
Influence of path attenuation index on positioning
error
In order to make the simulation results more representative,
this paper compares the influence of path attenuation index on
positioning error and sets the RSS measurement noise
${\delta}_{i,\epsilon}$ to
0.2 dB, the time measurement noise variance
${\delta}_{i,m}$ and
${\delta}_{i,n}$
are
set to 0.2 ns, the path attenuation index
$\beta $ value is obtained by
experimental analysis in advance, and its value varies with the
change of the propagation medium, generally between the typical
value of 2 and 5, and adjusts
$\beta $ value changes within the interval.
Figure 4: Influence of path attenuation index on positioning error
Figure 4 plots the variation of the RMSE positioning error
with the path attenuation index under different algorithms. It
can be seen from the figure that the positioning error gradually
decreases with the path attenuation index
$\beta $ value increases.
When the path attenuation index
$\beta $
value is equal to 2, the
positioning error RMSE of the designed ULLS method is 0.151
m, the positioning error of the SDP method is 0.114m, the
positioning error of the constrained CLLS method is 0.091 m, and
the CRLB value is 0.079 m; when the path attenuation index
$\beta $ value is equal to 5, the positioning error RMSE of the designed
ULLS method is 0.082 m, the positioning error of the SDP method
is 0.061m, the positioning error of the constrained CLLS method
is 0.039 m, and the CRLB value is 0.032 m.
Time synchronization parameter estimation error
The model designed in this paper assumes that the clock
parameters of the beacon nodes, including the drift rate,
deviation amount are calibrated, while the unknown node clock
synchronization parameter is unknown parameter. The time
synchronization parameters of the unknown node including the
drift rate, the deviation amount, and the position coordinates of
the unknown node are simultaneously estimated. Because the
noise error change of RSS in the hybrid algorithm has little effect
on the clock drift rate and deviation amount, the fixed RSS noise
error value is selected to be 0.2 dB, which makes the TOA noise
error change from 0.2 to 2 ns. The clock drift rate and deviation
amount of the unknown node are selected to analyze the
estimation error. The results are shown in Figure 5 and Figure 6.
Figure 5: Clock drift rate estimation error
Figure 6: Clock deviation amount t estimation error
Observing the change curve of the clock drift rate and
deviation amount of Figure 5 and Figure 6 with noise
${\delta}_{i,m}$
and ${\delta}_{i,n}$
,it is not difficult to find that the variation law of the algorithm is
similar to the result reflected in Figure 3. The general trend is that
the clock drift rate increases with the increase of noise, which is
an upward trend. As when
${\delta}_{i,m}$ and ${\delta}_{i,n}$
are 0.2 ns, the clock drift
rate RMSE of the ULLS is 0.008, the clock drift rate RMSE of the
SDP is 0.007, while the clock drift rate RMSE of the CLLS is 0.006;
when
${\delta}_{i,m}$ and
${\delta}_{i,n}$
are 2 ns, the clock drift rate RMSE of the ULLS
increases to 0.172, the clock drift rate RMSE of the SDP increases
to 0.131, while the CLLS clock drift rate RMSE increases to 0.125.
Similarly, the clock deviation quantity estimation error is similar to the abovementioned result, and the overall trend is also increasing with the increase of noise. As when ${\delta}_{i,m}$ and ${\delta}_{i,n}$ are 0.2 ns, the clock deviation quantity RMSE of the ULLS is 0.36, the clock deviation quantity RMSE of the SDP is 0.29, while the clock deviation quantity RMSE of the CLLS is 0.28 ; and when ${\delta}_{i,m}$ and ${\delta}_{i,n}$ are 2 ns, the clock deviation quantity RMSE of ULLS increases to 4.18 , the clock deviation quantity RMSE of SDP increases to 3.71, while the CLLS clock deviation quantity RMSE increases to 3.17.
Similarly, the clock deviation quantity estimation error is similar to the abovementioned result, and the overall trend is also increasing with the increase of noise. As when ${\delta}_{i,m}$ and ${\delta}_{i,n}$ are 0.2 ns, the clock deviation quantity RMSE of the ULLS is 0.36, the clock deviation quantity RMSE of the SDP is 0.29, while the clock deviation quantity RMSE of the CLLS is 0.28 ; and when ${\delta}_{i,m}$ and ${\delta}_{i,n}$ are 2 ns, the clock deviation quantity RMSE of ULLS increases to 4.18 , the clock deviation quantity RMSE of SDP increases to 3.71, while the CLLS clock deviation quantity RMSE increases to 3.17.
Conclusion
This paper proposes a 3D asynchronous sensor network
positioning method based on hybrid measurement of RSS and
TOA. Using the clock drift and deviation TOA ranging model,
the time synchronization parameters and the node position
coordinates are simultaneously estimated, which realizes both
time synchronization and node position coordinate estimation.
Converting the nonlinear optimization model of RSS and
TOA hybrid measurement into linear equation, obtains the
Unconstrained Linear Least Squares linear method (ULLS) of
the unknown node position coordinates, and optimizes the
positioning result of the ULLS. And convert it to a more precise
Constrained Least Squares (CLLS) result. Compared with the
SDP method introduced in the literature, the proposed ULLS
and CLLS linear estimation method calculation does not depend
on the choice of the initial solution, with low complexity and
fast operation speed. The simulation results show that the
performance of the proposed hybrid positioning algorithm is
superior to the traditional single RSS or TOA method. Compared
with the SDP method, the error of the constrained CLLS method
is smaller, which is closer to the CRLB lower bound value of the
positioning result. However, the ULLS and CLLS obtained by the
hybrid positioning method proposed in this paper have certain
limitations, and the estimation error has not reached the CRLB,
which needs further optimization.
ReferencesTop
 Qian Z, Sun D, LEUNG V. Overview of wireless network positioning. Chinese Journal of Computers. 2016;39(6):12371256.
 Xinming Z, Fan Y, Chao L, Qing D. Coverage EfficiencyBased Broadcast Protocol for Asynchronous Wireless Sensor Networks. IEEE Wireless Communications Letters. 2016;5(1):7679. doi: 10.1109/LWC.2015.2498173
 Guo H, Low KS, Nguyen HA. Optimizing the localization of a wireless sensor network in real time based on a lowcost microcontroller. IEEE Transactions on Industrial Electronics. 2011;58(3):741749. doi: 10.1109/TIE.2009.2022073
 Shi Q, He C, Chen H, Jiang L. Distributed wireless sensor network localization via sequential greedy optimization algorithm. IEEE Transactions on Signal Processing. 2010;58(6):33283340. doi: 10.1109/TSP.2010.2045416
 Yunhao L, Zheng Y, Xiaoping W. Location, Localization, and Localizability. Journal of Computer Science and Technology. 2010;25(2): 274297.
 Ye Miao, Wang Y. A new wireless sensor network security location method for tolerating malicious node attacks. Chinese Journal of Computers. 2013;36(3):532544.
 Sun X. Research on Kover covering mechanism of 3D wireless sensor network [D]. Nanjing University of Information Science and Technology. 2016.
 Yuan W, Shaodan M, Philip CCL. TOABased Passive Localization in QuasiSynchronous Networks. IEEE Communications Letters. 2014;18(4):592595. doi: 10.1109/LCOMM.2014.021214.132662
 TrungKien L, Nobutaka O. ClosedForm and Near ClosedForm Solutions for TDOABased Joint Source and Sensor Localization. IEEE Transactions on Signal Processing. 2017;65(5):12071221. doi: 10.1109/TSP.2016.2633784
 HuaJie S, XiaoPing Z, Zhi W. Efficient ClosedForm Algorithms for AOA Based SelfLocalization of Sensor Nodes Using Auxiliary Variables. IEEE Transactions on Signal Processing. 2014;62(10):25802594. doi: 10.1109/TSP.2014.2314064
 Yaming X, Jianguo Z, Peng Z. RSSBased Source Localization When PathLoss Model Parameters are Unknown. IEEE Communications Letters. 2014;18(6):10551058. doi: 10.1109/LCOMM.2014.2318031
 Huang J. Research on joint time synchronization and localization algorithm of sensor network [D]. Jiangnan University. 2014.
 Siamak Y, XiaoWen C, Benoit C. A Joint Localization and Synchronization Technique Using Time of Arrival at Multiple Antenna Receivers. 2013 Asilomar Conference on Signals, Systems and Computers. 2013:20172021. doi: 10.1109/ACSSC.2013.6810660
 Kalpana R, Madhusudhanan B. TAR: TOAAOA Based Random Transmission Directed Localization. Wireless Personal Communications. 2016;90(2):889902.
 Rajika K, Khair A, Nghi HT, Vijay D. An Improved Hybrid RSS/TDOA Wireless Sensors Localization Technique Utilizing WiFi Networks. Mobile Networks and Applications. 2016;21(2):286295.
 Slavisa T, Marko B, Rui D. 3D Target Localization in Wireless Sensor Network Using RSS and AoA measurements. IEEE Transactions on Vehicular Technology. 2017;66(4):31973210. doi: 10.1109/TVT.2016.2589923
 Xiaohong S, YuHen H. Maximum Likelihood MultipleSource Localization Using Acoustic Energy Measurements with Wireless Sensor Networks. IEEE Transactions on Signal Processing. 2005:53(1):4453. doi: 10.1109/TSP.2004.838930
 So HC, Lanxin L. Linear Least Squares Approach for Accurate Received Signal Strength Based Source Localization. IEEE Transactions on Signal Processing. 2011;59(8):40354040. doi: 10.1109/TSP.2011.2152400
 Wu X, Tan S. Performance Analysis of Wireless Sensor Network Location Algorithm Based on Semi definite Programming. Journal of Transduction Technology. 2012;25(12):17311736.