A method for tracking user positions in millimeter wave MIMO systems

Abstract

The present invention discloses a method for tracking user positions in a millimeter wave MIMO system, comprising: S1: establishing a two-dimensional spatial coordinate system, determining the physical coordinates of the millimeter wave MIMO base station, and each base station transmits the received user signal back to the data center for joint processing; S2: discretizing the channel beam domain by angle, sparsely representing the millimeter wave channel, and listing the sparse Bayesian learning problems of channel estimation and arrival angle tracking; S3: estimating the channel gain vector by generalized approximate message passing, and tracking the arrival angle by utilizing the time-varying characteristics of the user position; S4: utilizing the density-based clustering algorithm DBSCAN to screen out base stations with a direct line of sight path for positioning; S5: estimating the user position by utilizing the weighted minimum mean square error (WLS) estimator. The present invention realizes sub-meter-level high-precision user position tracking in a millimeter wave MIMO system by tracking the angle of moving users. At the same time, the algorithm has low computational complexity, meeting the demand for rapid positioning.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a method for tracking user positions in a millimeter wave MIMO system. Background Art

[0002] Millimeter-wave MIMO systems are considered a crucial technology for fifth- and sixth-generation cellular communication networks. Thanks to their high spectral efficiency, cellular wireless communications can achieve greater system capacity and improved communication reliability. Furthermore, due to the high angular resolution of millimeter-wave arrays, millimeter-wave MIMO systems can achieve sub-meter positioning, enabling the efficient implementation of applications such as the Internet of Things, autonomous driving, and social networking.

[0003] The fundamental characteristic of millimeter-wave MIMO system positioning is to extract location-related information such as delay, angle of arrival, and signal strength from user signals received by the base station, and then accurately recover the user's position based on this information. Therefore, the development of efficient millimeter-wave MIMO system positioning methods is essential.

[0004] Current multi-base station MIMO positioning algorithms are generally divided into two categories: two-step positioning methods and direct positioning methods. In the two-step positioning method, the base station first estimates the user's signal delay, arrival angle, reception strength, and other information based on the received signal. It then determines the user's location through triangulation or triangulation. In the direct positioning method, the base station directly recovers the user's position based on the received signal, eliminating the need for intermediate parameter estimation, enabling high-precision positioning of multi-base station systems.

[0005] However, in most existing research, both positioning methods consider locating stationary users. However, in real-world cellular network communication scenarios, some users are often in high-speed motion, such as in applications like autonomous driving, connected vehicles, and drones. In these situations, the traditional Global Positioning System (GPS) cannot meet the sub-meter positioning accuracy requirements. Therefore, developing a user location tracking algorithm for high-speed motion scenarios is urgently needed. Summary of the Invention

[0006] Technical Problem: Existing millimeter-wave MIMO systems only consider the positioning of stationary users, which cannot meet the positioning requirements of future high-speed mobile scenarios. This paper proposes a method for tracking user positions in millimeter-wave MIMO systems. This method can achieve high-precision position tracking of rapidly moving users in multi-base station millimeter-wave MIMO systems of any array configuration. The user channel estimation in this paper uses an approximate generalized message passing algorithm, which greatly reduces the complexity of user position tracking and meets the requirements of rapid millimeter-wave positioning.

[0007] Technical solution: To achieve the purpose of the present invention, a method for tracking user positions in a millimeter wave MIMO system of the present invention specifically includes the following steps:

[0008] S1: Establish a two-dimensional spatial coordinate system to determine the physical coordinates of the millimeter wave MIMO base station. Each base station transmits the received user signals back to the data center for joint processing.

[0009] S2: Discretize the channel beam domain by angle, sparsely represent the millimeter wave channel, and list the sparse Bayesian learning problem for channel estimation and arrival angle tracking;

[0010] S3: Use generalized approximate message passing to estimate the channel gain vector and use the time-varying characteristics of user position to track the arrival angle;

[0011] S4: Use the density-based clustering algorithm DBSCAN to filter out base stations with direct line-of-sight paths for positioning;

[0012] S5: Use the weighted minimum mean square error (WLS) estimator to estimate the user location.

[0013] in,

[0014] The step S1 is specifically as follows:

[0015] S1.1: Define the two-dimensional target location area By region Establish a two-dimensional coordinate system with the center as the origin, and set the position of the m-th millimeter wave MIMO base station to At the tth moment, the position of the user to be measured is

[0016] S1.2: At time t, define the narrowband signal vector transmitted by the user as The signal received by each base station is

[0017]

[0018] Where: P is the energy of the transmitted signal, h t,m is the channel vector between the user and the mth base station at time t, N t,m is the receiving noise matrix at time t, N t,m The elements in the equation are uncorrelated in time and space, and obey zero-mean Gaussian distribution. The noise variance is σ 2 ;

[0019] S1.3: Based on the sparse characteristics of the millimeter wave channel, determine the channel vector expression form, specifically:

[0020]

[0021] Where: P t,m represents the number of distinguishable paths at time t, a m (θ) is the array response vector, α t,m,1 and θ t,m,1 denote the gain and arrival angle of the direct-view path, and denote the gain and angle of arrival of the non-line-of-sight path, respectively;

[0022] S1.4: At time t, the spatial geometric relationship between the user position and the direct-view path arrival angle can be expressed as:

[0023]

[0024] make It is expressed as the position of the nth antenna of the mth base station relative to the center of gravity of the array. Assuming that the antennas are isotropic and uncoupled, for the incident angle θ, the array steering vector can be expressed as:

[0025]

[0026] Where j represents a unit imaginary number, and λ represents the carrier frequency.

[0027] The step S2 is specifically as follows:

[0028] S2.1: Since the energy of the millimeter wave channel is only concentrated in certain angle areas, the millimeter wave channel h is discretized by angle in the beam domain channel. m Perform sparse representation; for the mth base station, define S uniform angle discrete points, namely:

[0029]

[0030] in: is an angle discrete point; consider uniform discrete angle The number of discrete angles is much larger than the number of resolvable paths in the millimeter wave channel, that is, S>>P t,m ;

[0031] S2.2: In practice, the arrival angle θ t,m,p Usually does not exactly match discrete angle sets In particular, when the number of discrete points S is not large enough, in order to avoid this model mismatch, the off-grid angle information is introduced into the angle discretization. In this case, the arrival angle can be expressed as:

[0032]

[0033] in, represents the distance to arrival angle θ t,m,p The nearest discrete point, Indicates the off-grid angle;

[0034] According to the above parameter definitions, the millimeter wave channel h t,m Expressed as:

[0035] h t,m =A t,m (δ t,m )u t,m ,

[0036] Among them, δ t,m =[δ t,m,1 ,δ t,m,2 ,...,δ t,m,S ] T Represents the off-grid angle vector, if i=r t,m,p ,So Otherwise δ t,m,i =0; In addition, the channel response matrix Sparse channel gain vector u t,m =[u t,m,1 ,u t,m,2 ,...,u t,m,S ] T ; in u t,m Among them, the first t,m,p The elements correspond to the channel path gain, i.e. It can be seen that the off-grid angle vector δ t,m and the path gain vector u t,m have the same support set; further, in the channel corresponding matrix A t,m (δ t,m ), a first-order Taylor expansion approximation is performed at each discrete angle point, and the channel corresponding matrix can be expressed as:

[0037]

[0038] in, Δ t,m =diag(δ t,m,1 ,δ t,m,2 ...,δ t,m,S ), express At the point The derivative at ;

[0039] S2.3: Based on the angle discretization information given in S2.2, the received signal is expressed as:

[0040]

[0041] in, A t (δ t )=blkdiag(At,1 (δ t,1 ),A t,2 (δ t,2 ),...,A t,M (δ t,M )),

[0042] S2.4: Due to the change in user movement position, the arrival angle As a time-varying parameter, a first-order Markov process is used, and the user arrival angle between adjacent moments can be expressed as:

[0043] θ t,m =θ t-1,m +v m (t-1),t=2,3,...,T,

[0044] Among them, v m (t-1) represents the driving noise, which obeys Gaussian distribution and has a mean value of E[v m (t)]=0, the covariance matrix E[v m (t)(v m (t)) T ]=Q m , in the above formula, the covariance matrix Q m It is a diagonal matrix whose diagonal elements are unknown and need to be estimated in subsequent steps;

[0045] S2.5: To facilitate calculation, the received signal is vectorized, specifically:

[0046]

[0047] in,

[0048] In the sparse Bayesian learning framework, the path gain vector u t The sparsity of can be represented by a Gaussian mixture model, specifically:

[0049]

[0050] Among them, Γ t =diag(γ t ), is the control gain vector u t The sparsity hyperparameter vector, S1 = MS, Pr(γ t ) represents γ t The prior probability of

[0051] According to the above Gaussian mixture model, the joint probability density function of channel estimation and angle tracking can be expressed as:

[0052]

[0053] Where Y=[Y1,Y2,...,Y T ], Further:

[0054]

[0055] as well as

[0056]

[0057] S2.6: In the sparse Bayesian learning framework, an uninformative prior is usually used, that is, the prior probability Pr(γ t ) can be set to a unit quantity and will not affect the final channel sparsity; Based on step S2.5 and the above analysis, channel estimation and angle tracking can be expressed as the following optimization problem, namely:

[0058]

[0059] By solving the above formula, the user's channel gain and arrival angle information are obtained.

[0060] The step S3 is specifically as follows:

[0061] S3.1: Since the objective function of the optimization problem in step S2.6 is non-convex and there is no closed-form solution, solving this problem is divided into two steps. First, the generalized approximate message passing algorithm is used to estimate the path gain vector to obtain the estimated value Then the time-varying arrival angle is tracked to obtain an estimated value

[0062] S3.2: Based on the estimated value of the path gain vector in S3.1 The arrival angle information at each moment can be roughly obtained. In order to further obtain the accurate arrival angle information at each moment, the update formula for time-varying arrival angle tracking is further derived. According to S2.4, the off-grid angle δ t Its previous state δ t-1 It is time-dependent, and the joint probability density function of the off-grid angle is given in S2.5. Therefore, the off-grid angle at each moment is estimated by the maximum likelihood method, as follows:

[0063]

[0064] Wherein, the symbols appearing in the above formula can be seen in the steps described in claims 2 and 3,

[0065] S3.3: During the channel estimation and angle tracking process in steps S3.1 and S3.2 above, it is necessary to obtain estimates of the hyperparameter γ and the covariance matrix Q. Based on the maximum likelihood strategy, the estimated values ​​of the hyperparameter γ can be obtained by maximizing the joint probability density function of the hyperparameter γ. Specifically,

[0066]

[0067] Wherein, the symbols appearing in the above formula can be seen in the steps described in claims 2 and 3;

[0068] Similarly, the covariance matrix Q = blkdiag(Q1,Q2,...,Q M ) can be obtained by maximizing its joint probability density function, specifically:

[0069]

[0070] Wherein, the symbols appearing in the above formula can be seen in the steps described in claims 2 and 3;

[0071] In particular, in steps S3.1 to S3.3, the estimated channel gain, arrival angle, hyperparameters, and covariance matrix are coupled to each other. In order to obtain consistent estimates of the above parameters, it is necessary to execute S3.1 to S3.3 repeatedly until convergence.

[0072] In step S4, a density-based clustering algorithm is used to screen out base stations with direct-line paths for positioning. Specifically, the channel gain of the direct-line path in the millimeter-wave sparse channel is much greater than the channel gain of other non-direct-line paths. Based on this characteristic, the possible direct-line paths of each base station are preliminarily screened out from the channel information obtained in step S3 for positioning. Due to obstruction by objects, the direct-line path between the base station and the user may not exist. Therefore, a density-based clustering algorithm is used to screen out base stations with direct-line paths for final positioning.

[0073] In step S5, the user position is estimated using a weighted minimum mean square error (WLS) estimator. Specifically, based on the arrival angle information of the direct-line path obtained in step S4, the geometric relationship between the position angles described in step S1.4 is used to construct a triangulated positioning equation, which is solved using a weighted minimum mean square error (WLS) estimator to obtain the user position information at each moment.

[0074] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial technical effects:

[0075] 1) The present invention can achieve high-precision position tracking of fast-moving users in a multi-base station millimeter-wave MIMO system with any array form.

[0076] 2) The user channel estimation in the present invention adopts an approximate generalized message passing algorithm, which greatly reduces the complexity of user position tracking and realizes the requirement of millimeter wave rapid positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 This is a diagram of an application scenario of user location tracking in a millimeter wave MIMO system according to an embodiment of the present invention;

[0078] Figure 2 It is a flow chart of the algorithm of the present invention;

[0079] Figure 3 This is a comparison chart of the mean square error performance of the present invention and the existing base station positioning algorithm;

[0080] Figure 4 This is a comparison diagram of the cumulative probability density of positioning errors of the existing base station positioning algorithm of the present invention. DETAILED DESCRIPTION

[0081] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. The described embodiments are only some of the embodiments of the present invention, not all of them. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but rather merely represents selected embodiments of the present invention.

[0082] Example 1

[0083] The rapid positioning method provided by this application can be applied to Figure 1 In the application environment shown, a user terminal communicates with a base station. For example, the base station receives a target signal transmitted by a terminal at a preset location and obtains a real received signal. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices, and the base station is implemented as a base station cluster consisting of multiple base stations.

[0084] refer to Figure 2 This embodiment provides a method for tracking and positioning a user in a millimeter wave MIMO system, which specifically includes the following steps:

[0085] Step S1: Establish a two-dimensional spatial coordinate system to determine the physical coordinates of the millimeter wave MIMO base stations. Each base station transmits the received user signals back to the data center for joint processing, as follows:

[0086] Step S1.1: Define the two-dimensional target positioning area By region Establish a two-dimensional coordinate system with the center as the origin, and set the position of the m-th millimeter wave MIMO base station to At the tth moment, the position of the user to be measured is

[0087] Step S1.2: At time t, define the narrowband signal vector transmitted by the user as The signal received by each base station is

[0088]

[0089] Where: P is the energy of the transmitted signal, h t,m is the channel vector between the user and the mth base station at time t, n m is the received noise vector, n m The elements in the equation are uncorrelated in time and space, and obey zero-mean Gaussian distribution. The noise variance is σ 2 ;

[0090] Step S1.3: Determine the channel vector expression form based on the sparse characteristics of the millimeter wave channel, specifically:

[0091]

[0092] Where: P t,m represents the number of distinguishable paths at time t, a m (θ) is the array response vector, α t,m,1 and θ t,m,1 denote the gain and arrival angle of the direct-view path, and denote the gain and angle of arrival of the non-line-of-sight path, respectively;

[0093] Step S1.4: At time t, the spatial geometric relationship between the user position and the direct-view path arrival angle can be expressed as:

[0094]

[0095] make It is expressed as the position of the nth antenna of the mth base station relative to the center of gravity of the array. Assuming that the antennas are isotropic and uncoupled, the array steering vector can be expressed as:

[0096]

[0097] Where j represents a unit imaginary number, and λ represents the carrier frequency.

[0098] Step S2: Discretize the channel beam domain by angle, sparsely represent the millimeter wave channel, and list the sparse Bayesian learning problem for channel estimation and arrival angle tracking as follows:

[0099] Step S2.1: Since the millimeter wave channel energy is only concentrated in certain angle areas, the millimeter wave channel h can be obtained by discretizing the beam domain channel angles. m For the mth base station, define S uniform angle discrete points, namely:

[0100]

[0101] Among them: Consider uniform discrete angle The number of discrete angles is much larger than the number of resolvable paths in the millimeter wave channel, that is, S>>P t,m .

[0102] Step S2.2: In practice, the arrival angle θ t,m,p Usually does not exactly match a set of discrete angles In order to avoid this model mismatch, the off-grid angle information is introduced into the angle discretization. In this case, the arrival angle can be expressed as

[0103]

[0104] in, represents the distance to arrival angle θ t,m,p The nearest discrete point, Indicates the off-grid angle.

[0105] According to the above parameter definitions, the millimeter wave channel h t,m It can be expressed as:

[0106] h t,m =A t,m (δ t,m )u t,m ,

[0107] Among them, δ t,m =[δ t,m,1 ,δ t,m,2 ,...,δ t,m,S ] T Represents the off-grid angle vector, if i=r t,m,p ,So Otherwise δ t,m,i = 0. In addition, the channel response matrix Sparse channel gain vector u t,m =[u t,m,1 ,u t,m,2 ,...,u t,m,S ] T . in u t,m Among them, the first t,m,p The elements correspond to the channel path gain, i.e. It can be seen that the off-grid angle vector δt,m and the path gain vector u t,m have the same support set. Furthermore, in the channel corresponding matrix A t,m (δ t,m ), a first-order Taylor expansion approximation is performed at each discrete angle point, and the channel corresponding matrix can be expressed as:

[0108]

[0109] in, Δ t,m =diag(δ t,m,1 ,δ t,m,2 ...,δ t,m,S ), express At the point The derivative at .

[0110] Step S2.3: Based on the angle discretization information given in S2.2, the received signal can be expressed as:

[0111]

[0112] in, A t (δ t )=blkdiag(A t,1 (δ t,1 ),A t,2 (δ t,2 ),...,A t,M (δ t,M )),

[0113] Step S2.4: Due to the change in the user's movement position, the arrival angle is a time-varying parameter. Using a first-order Markov process, the user arrival angle between adjacent moments can be expressed as:

[0114] θ t,m =θ t-1,m +v m (t-1),t=2,3,...,T,

[0115] Among them, v m (t-1) represents the driving noise, which obeys Gaussian distribution and has a mean value of E[v m (t)]=0, the covariance matrix E[v m (t)(v m (t)) T ]=Q m In the above formula, the covariance matrix Q m It is a diagonal matrix whose diagonal elements are unknown and need to be estimated in subsequent steps.

[0116] Step S2.5: To facilitate calculation, the received signal is vectorized, specifically:

[0117]

[0118] in,

[0119] In the sparse Bayesian learning framework, the path gain vector u t The sparsity of can be represented by a Gaussian mixture model, specifically:

[0120]

[0121] Among them, Γ t =diag(γ t ), is the control gain vector u t The sparsity hyperparameter vector, S1 = MS, Pr(γ t ) represents γ t The prior probability of .

[0122] According to the above Gaussian mixture model, the joint probability density function of channel estimation and angle tracking can be expressed as:

[0123]

[0124] Where Y=[Y1,Y2,...,Y T ], Furthermore, we have

[0125]

[0126] as well as

[0127]

[0128] Step S2.6: In the sparse Bayesian learning framework, an uninformative prior is usually used, that is, the prior probability Pr(γ t ) can be set to a unit quantity and will not affect the final channel sparsity. Based on step S2.5 and the above analysis, channel estimation and angle tracking can be expressed as the following optimization problem, namely:

[0129]

[0130] By solving the above formula, the user's channel gain and arrival angle information can be obtained.

[0131] Step S3: Use generalized approximate message passing to estimate the channel gain vector and use the time-varying characteristics of the user position to track the arrival angle, specifically:

[0132] Step S3.1: Since the objective function of the optimization problem in step S2.6 is non-convex, there is no closed-form solution. Solving this problem is divided into two steps. First, the generalized approximate message passing (GAMP) algorithm is used to estimate the path gain vector u, and then the time-varying arrival angle is tracked. In the generalized approximate message passing algorithm, the input function g is defined as s (p,υ p ) and the output function g u (r,υ r ) are:

[0133]

[0134]

[0135] Among them, p,υ p , r, υ r is an intermediate parameter.

[0136] Furthermore, according to the linear model of the received signal in S2.3, the input function of the above formula can be specifically expressed as:

[0137]

[0138] Its partial derivative with respect to vector p is g' s (p,υ p )=σ -2 . / (σ -2 +υ p ). Here, AB and A. / B represent element-wise multiplication and element-wise division, respectively.

[0139] According to the Gaussian mixture model of the sparse vector u in S2.5, the output function can be specifically expressed as:

[0140] g u (r,υ r )=1. / (1+γ t .υ r ).r,

[0141] Its partial derivative with respect to vector r is g′ u (r,υ r )=1. / (1+γ t .υ r The specific steps of the channel estimation algorithm based on GAMP are given as follows:

[0142]

[0143]

[0144] Step S3.2: Based on the estimated value of the path gain vector in S3.1 The arrival angle information at each moment can be roughly obtained. In order to further obtain the accurate arrival angle information at each moment, the update formula for time-varying arrival angle tracking can be further derived. According to S2.4, the off-grid angle δ t Its previous state δ t-1 It is time-dependent, and the joint probability density function of the off-grid angle is given in S2.5. Therefore, the off-grid angle at each moment can be estimated by the maximum likelihood method, as follows:

[0145]

[0146] in, Δ t =diag(δ t ), In particular, if i = r t,m,p , otherwise In the above formula, the received signal vector And their covariance matrices are:

[0147]

[0148]

[0149] Among them, the input function [g s (p,υ p )] i 、p、υ p This is given by the generalized approximate message passing algorithm in S3.1.

[0150] In order to obtain the off-grid angle δ t The estimated value of g(δ t ) and h(δ t ) are expressed as t The functions are as follows:

[0151]

[0152]

[0153] in, b1 and b2 are two t Irrelevant constant. Rearranging the above formula, we can get δ t The estimated value of is:

[0154]

[0155] in,

[0156] Furthermore, the objective function in the above formula is about δ t Taking the derivative and setting it to zero, we can get δ t The estimated value of is:

[0157]

[0158] in,

[0159] Step S3.3: During the channel estimation and angle tracking process in steps S3.1 and S3.2 above, it is necessary to obtain estimates of the hyperparameter γ and the covariance matrix Q. Based on the maximum likelihood strategy, the estimated value of the hyperparameter γ can be obtained by maximizing the joint probability density function of the hyperparameter γ, specifically:

[0160]

[0161] in, and Obtained through the generalized approximate message passing algorithm in step S3.1.

[0162] Similarly, the covariance matrix Q = blkdiag(Q1,Q2,...,Q M ) can be obtained by maximizing its joint probability density function, specifically:

[0163]

[0164] in,

[0165] In particular, in steps S3.1 to S3.3, the estimated parameters, channel gain, angle of arrival, hyperparameters, and covariance matrix, are coupled to each other. To obtain consistent estimates of these parameters, it is necessary to loop through S3.1 to S3.3 until convergence. The specific algorithm pseudo code is as follows:

[0166]

[0167]

[0168] Step S4: Use the density-based clustering algorithm (DBSCAN) to filter out base stations with direct line-of-sight paths for positioning. Specifically, the channel gain of the direct line-of-sight path in the millimeter wave sparse channel is much greater than the channel gain of other non-direct line-of-sight paths. Based on this characteristic, the possible direct line of sight paths of each base station can be preliminarily filtered out from the channel information obtained in step S3, and then the intersection of each direct line of sight path is calculated to determine the user position. However, due to obstruction by objects, the direct line of sight path between the base station and the user may not exist, such as Figure 1 Therefore, the density-based clustering (DBSCAN) algorithm is used to filter out base stations with direct line of sight for final positioning. Specifically, the pseudo code of the DBSCAN-based direct line of sight screening algorithm is as follows:

[0169]

[0170]

[0171] Step S5: Use the weighted minimum mean square error (WLS) estimator to estimate the user position. Specifically, using the geometric relationship between the position angles described in step S1.4, we can get:

[0172]

[0173] in, Angle of arrival By rotating the coordinate axis, we can further obtain:

[0174]

[0175] According to the base station information with direct line of sight obtained in step S4, the following triangulation positioning equation can be constructed to restore the user position: Specifically:

[0176] q t =T t p t +ω t ,t=1,2,...,T,

[0177] in, represents the error vector, Definition ω t The covariance matrix is ​​Ω t In addition, in the above formula,

[0178]

[0179] Using the weighted minimum mean square error estimator, the user location is calculated as:

[0180]

[0181] in, In the actual positioning process, first set Ω t Roughly estimate the user position for the unit matrix, and then calculate Ω t The exact values ​​of the diagonal elements are fed into the weighted minimum mean square error estimator to obtain an accurate estimate of the user's position.

[0182] refer to Figure 3 , Figure 3 This is a comparison chart of the mean square error performance of the location tracking method of this embodiment and the existing positioning method. In the simulation parameter setting, the number of base stations M = 4, located at the coordinate points [-50m, -50m], [-50m, 50m], [50m, 50m], [50m, -50m], each base station uses a uniform circular array, and the number of antennas is N m =50, m=1,2,3,4, the carrier frequency is 30GHz millimeter wave band. The channel generation model adopts the Urban Marco scenario defined by the 3GPP standard. The randomly generated users are distributed in the target positioning area. The user's position is tracked for T = 7 moments, with a speed of 20 m / s and a random direction. It can be found that:

[0183] Compared with other base station positioning algorithms, the user location tracking algorithm of this embodiment can achieve lower positioning mean square error.

[0184] refer to Figure 4 , where the simulation parameters are set with Figure 3 The simulation parameters in are set the same as in . It can be found that:

[0185] Under the same positioning error requirement, the user location tracking algorithm of this embodiment has a higher cumulative probability density than other base station positioning algorithms. At the same time, in this embodiment, the user location tracking algorithm can achieve decimeter-level positioning accuracy.

[0186] The above is a schematic description of the present invention and its embodiments, which is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure and method are not limited thereto. Therefore, if a person skilled in the art is inspired by the above and, without departing from the purpose of the present invention, designs structures and embodiments similar to the technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A method for tracking user positions in a millimeter wave MIMO system, characterized in that: The user location tracking method specifically includes the following steps: S1: Establish a two-dimensional spatial coordinate system to determine the physical coordinates of the millimeter wave MIMO base station. Each base station transmits the received user signals back to the data center for joint processing. S2: Discretize the channel beam domain by angle, sparsely represent the millimeter wave channel, and list the sparse Bayesian learning problem for channel estimation and arrival angle tracking; S3: Use generalized approximate message passing to estimate the channel gain vector and use the time-varying characteristics of user position to track the arrival angle; S4: Use the density-based clustering algorithm DBSCAN to filter out base stations with direct line-of-sight paths for positioning; S5: Use the weighted minimum mean square error (WLS) estimator to estimate the user location.

2. The method for tracking user positions in a millimeter wave MIMO system according to claim 1, wherein: The step S1 is specifically as follows: S1.1: Define the two-dimensional target location area By region Establish a two-dimensional coordinate system with the center as the origin, and set the position of the m-th millimeter wave MIMO base station to At the tth moment, the position of the user to be measured is S1.2: At time t, define the narrowband signal vector transmitted by the user as The signal received by each base station is Where: P is the energy of the transmitted signal, h t,m is the channel vector between the user and the mth base station at time t, N t,m is the receiving noise matrix at time t, N t,m The elements in the equation are uncorrelated in time and space, and obey zero-mean Gaussian distribution. The noise variance is σ 2 ; S1.3: Based on the sparse characteristics of the millimeter wave channel, determine the channel vector expression form, specifically: Where: P t,m represents the number of distinguishable paths at time t, a m (θ) is the array response vector, α t,m,1 and θ t,m,1 denote the gain and arrival angle of the direct-view path, and denote the gain and angle of arrival of the non-line-of-sight path, respectively; S1.4: At time t, the spatial geometric relationship between the user position and the direct-view path arrival angle can be expressed as: make It is expressed as the position of the nth antenna of the mth base station relative to the center of gravity of the array. Assuming that the antennas are isotropic and uncoupled, for the incident angle θ, the array steering vector can be expressed as: Where j represents a unit imaginary number, and λ represents the carrier frequency.

3. A method for tracking user positions in a millimeter wave MIMO system according to claim 1 or 2, characterized in that: The step S2 is specifically as follows: S2.1: Since the energy of the millimeter wave channel is only concentrated in certain angle areas, the millimeter wave channel h is discretized by angle in the beam domain channel. m Perform sparse representation; for the mth base station, define S uniform angle discrete points, namely: in: is an angle discrete point; consider uniform discrete angle The number of discrete angles is much larger than the number of resolvable paths in the millimeter wave channel, that is, S>>P t,m ; S2.2: In practice, the arrival angle θ t,m,p Usually does not exactly match a set of discrete angles In particular, when the number of discrete points S is not large enough, in order to avoid this model mismatch, the off-grid angle information is introduced into the angle discretization. In this case, the arrival angle can be expressed as: in, represents the distance to arrival angle θ t,m,p The nearest discrete point, Indicates the off-grid angle; According to the above parameter definitions, the millimeter wave channel h t,m Expressed as: h t,m =A t,m (δ t,m )u t,m , Among them, δ t,m =[δ t,m,1 ,δ t,m,2 ,...,δ t,m,S ] T Represents the off-grid angle vector, if i=r t,m,p ,So Otherwise δ t,m,i =0; In addition, the channel response matrix Sparse channel gain vector u t,m =[u t,m,1 ,u t,m,2 ,...,u t,m,S ] T ; in u t,m Among them, the first t,m,p The elements correspond to the channel path gain, i.e. It can be seen that the off-grid angle vector δ t,m and the path gain vector u t,m have the same support set; further, in the channel corresponding matrix A t,m (δ t,m ), a first-order Taylor expansion approximation is performed at each discrete angle point, and the channel corresponding matrix can be expressed as: in, Δ t,m =diag(δ t,m,1 ,δ t,m,2 ...,δ t,m,S ), express At the point The derivative at ; S2.3: Based on the angle discretization information given in S2.2, the received signal is expressed as: Among them, A t (d t )=blkdiag(A t,1 (d t,1 ),A t,2 (d t,2 ),...,A t,M (d t,M )), S2.4: Due to the change in user movement position, the arrival angle As a time-varying parameter, a first-order Markov process is used, and the user arrival angle between adjacent moments can be expressed as: θ t,m =θ t-1,m +v m (t-1),t=2,3,...,T, Among them, v m (t-1) represents the driving noise, which obeys Gaussian distribution and has a mean value of E[v m (t)]=0, the covariance matrix E[v m (t)(v m (t)) T ]=Q m , in the above formula, the covariance matrix Q m It is a diagonal matrix whose diagonal elements are unknown and need to be estimated in subsequent steps; S2.5: To facilitate calculation, the received signal is vectorized, specifically: in, In the sparse Bayesian learning framework, the path gain vector u t The sparsity of can be represented by a Gaussian mixture model, specifically: Among them, Γ t =diag(γ t ), is the control gain vector u t The sparsity hyperparameter vector, S1 = MS, Pr(γ t ) represents γ t The prior probability of According to the above Gaussian mixture model, the joint probability density function of channel estimation and angle tracking can be expressed as: Where Y=[Y1,Y2,...,Y T ], Further: as well as S2.6: In the sparse Bayesian learning framework, an uninformative prior is usually used, that is, the prior probability Pr(γ t ) can be set to a unit quantity and will not affect the final channel sparsity; Based on step S2.5 and the above analysis, channel estimation and angle tracking can be expressed as the following optimization problem, namely: By solving the above formula, the user's channel gain and arrival angle information are obtained.

4. A method for tracking user positions in a millimeter wave MIMO system according to claim 1 or 3, characterized in that: The step S3 is specifically as follows: S3.1: Since the objective function of the optimization problem in step S2.6 is non-convex and there is no closed-form solution, solving this problem is divided into two steps. First, the generalized approximate message passing algorithm is used to estimate the path gain vector to obtain the estimated value Then the time-varying arrival angle is tracked to obtain an estimated value S3.2: Based on the estimated value of the path gain vector in S3.1 The arrival angle information at each moment can be roughly obtained. In order to further obtain the accurate arrival angle information at each moment, the update formula for time-varying arrival angle tracking is further derived. According to S2.4, the off-grid angle δ t Its previous state δ t-1 It is time-dependent, and the joint probability density function of the off-grid angle is given in S2.

5. Therefore, the off-grid angle at each moment is estimated by the maximum likelihood method, as follows: Wherein, the symbols appearing in the above formula can be seen in the steps described in claims 2 and 3, S3.3: During the channel estimation and angle tracking process in steps S3.1 and S3.2 above, it is necessary to obtain estimates of the hyperparameters and covariance matrix. Based on the maximum likelihood strategy, the estimated values ​​of the hyperparameters can be obtained by maximizing the joint probability density function of the hyperparameters. Specifically, Wherein, the symbols appearing in the above formula can be seen in the steps described in claims 2 and 3; Similarly, the covariance matrix Q = blkdiag(Q1,Q2,...,Q M ) can be obtained by maximizing its joint probability density function, specifically: Wherein, the symbols appearing in the above formula can be seen in the steps described in claims 2 and 3; In particular, in steps S3.1 to S3.3, the estimated channel gain, arrival angle, hyperparameters, and covariance matrix are coupled to each other. In order to obtain consistent estimates of the above parameters, it is necessary to execute S3.1 to S3.3 repeatedly until convergence.

5. A method for tracking user position in a millimeter wave MIMO system according to claim 1 or 4, characterized in that: In step S4, a density-based clustering algorithm is used to screen out base stations with direct-line paths for positioning. Specifically, the channel gain of the direct-line path in the millimeter-wave sparse channel is much greater than the channel gain of other non-direct-line paths. Based on this characteristic, the possible direct-line paths of each base station are preliminarily screened out from the channel information obtained in step S3 for positioning. Due to obstruction by objects, the direct-line path between the base station and the user may not exist. Therefore, a density-based clustering algorithm is used to screen out base stations with direct-line paths for final positioning.

6. A method for tracking user position in a millimeter wave MIMO system according to claim 1 or 5, characterized in that: In step S5, the user position is estimated using a weighted minimum mean square error (WLS) estimator. Specifically, based on the arrival angle information of the direct-line path obtained in step S4, the geometric relationship between the position angles described in step S1.4 is used to construct a triangulated positioning equation, which is solved using a weighted minimum mean square error (WLS) estimator to obtain the user position information at each moment.

Images