Hybrid field source positioning method based on HRIS and hierarchical sparse reconstruction

Abstract

The invention relates to a hybrid field source positioning method based on HRIS and hierarchical sparse reconstruction, and belongs to the technical field of wireless communication and signal processing. The method comprises the following steps: firstly, receiving a pilot signal sent by a base station BS by using an active RIS reflection unit, estimating a BS-RIS channel and decoupling a cascade channel; and secondly, the active RIS reflection units process pilot signals sent by user equipment (UE), and select the active RIS reflection units with specific symmetric structures to construct three groups of four-order cumulant matrixes, and sequentially obtain elevation angle, azimuth angle and distance information in three stages. Then, through constructing an over-complete steering vector dictionary and introducing a tuning factor and a step size attenuation strategy, three-stage hierarchical estimation is carried out on the information of the elevation angle, the azimuth angle and the distance; and finally, effective discrimination of a near-field source and a far-field source is realized by combining nonlinear phase response characteristics related to the distance in the array structure, so that high-precision estimation of the source position in the three-dimensional space is completed.

Description

Technical Field

[0001] The present invention relates to a hybrid field source positioning method based on a hybrid reconfigurable intelligent reflective metasurface (Hybrid RIS, HRIS) and hierarchical sparse reconstruction, which is used for three-dimensional positioning of far-field and near-field user signal sources in non-line-of-sight scenarios and belongs to the field of wireless communication and signal processing technology. Background Art

[0002] Positioning technology is a core means of acquiring spatial location information and achieving location awareness in communication scenarios such as intelligent transportation, emergency rescue, and the Industrial Internet of Things. In multi-user positioning scenarios, the high mobility of users causes the channel state between the base station and users to change rapidly, resulting in the coexistence of far-field and near-field sources in the communication network, forming a heterogeneous communication network. Far-field sources typically exhibit plane wave characteristics, while near-field sources exhibit spherical wave characteristics. Traditional positioning systems rely on a single propagation model (such as the far-field plane wave model). When users are within the near-field communication range, the curvature effect of the spherical wave gradually increases, making the far-field model inaccurate for all signals. To improve channel coverage and positioning accuracy, the introduction of large reconfigurable smart reflective metasurface (RIS) arrays in communication systems has significantly increased the operating frequency, further expanding the coverage of near-field communications. In near-field communications, the wavefront generated by the signal source exhibits spherical wave characteristics. In far-field communications, the wavefront generated by the signal source exhibits plane wave characteristics. In mixed-field source positioning, if the far-field plane wave model is continued to be used for channel modeling, ignoring the differences in signal characteristics between far-field and near-field signals, it will lead to model mismatch problems, and even cause positioning errors to accumulate and performance to degrade. In addition to the model mismatch problem, in actual positioning scenarios, building obstructions and complex environments often lead to frequent non-line-of-sight (NLOS) links between base stations and users. The NLOS effect not only affects the signal arrival time and phase characteristics, reducing the signal power, but also introduces multipath interference, causing the signal propagation path to distort, further exacerbating the decline in positioning accuracy. In addition, the far field relies only on angle parameters, while the near field requires joint angle and distance estimation, which will cause coupling between the position parameters. Traditional algorithms cannot decouple the mixed parameters and require a multi-dimensional joint search.

[0003] RIS, a new communication technology enhancement, reconstructs and enhances signal paths by establishing virtual links between base stations and users. Current positioning communication systems mostly use passive RIS, which cannot directly receive and process signals and must indirectly achieve positioning performance by back-transmitting signals to the base station. However, due to the coupling of the BS-RIS-UE cascade channel, the channel errors between the two hops can affect each other due to the coupling relationship. This cascaded channel exponentially increases the complexity of the positioning algorithm. In complex multi-user positioning scenarios, this cumulative effect of channel errors makes RIS-assisted positioning even more difficult. Traditional algorithms use a joint search for Direction of Arrival (DOA) in three-dimensional positioning, jointly estimating elevation and azimuth, known as a two-dimensional spectral search. This results in extremely high computational complexity. Therefore, it is necessary to decouple the elevation and azimuth in DOA estimation and simplify the two-dimensional spectral search into two one-dimensional spectral searches. This allows for a three-stage estimation of elevation, azimuth, and range parameters, significantly reducing the complexity of position parameter estimation. Summary of the Invention

[0004] The technical problem to be solved by the present invention is: in a non-line-of-sight environment, the positioning problem of the hybrid field source assisted by HRIS is solved by decoupling the user position parameter information and gradually estimating it in three stages, combining the tuning factor attenuation and step size reduction strategy to gradually reduce the search range, and integrating The norm and weight matrix improve positioning accuracy and solve the problems of multi-hop channel error interference, parameter coupling and computational complexity.

[0005] The technical solution of the present invention is: a hybrid source localization method based on HRIS and hierarchical sparse reconstruction, the specific steps are:

[0006] Step 1: The base station BS sends a pilot signal to the active RIS reflection unit to estimate and decouple the BS-RIS channel;

[0007] Step 2: Based on the decoupled BS-RIS channel, the user equipment UE sends a user pilot signal to the active RIS reflection unit;

[0008] Step 3: The active RIS reflector unit constructs three sets of fourth-order cumulant matrices based on the received user pilot signal and gradually completes the position parameter estimation in three stages based on the dynamic tuning factor and step size layering. The position parameters include elevation angle, azimuth angle and range.

[0009] The first stage is to decouple the elevation and azimuth angles based on the first set of fourth-order cumulant matrices, construct signal subspace projections, and estimate the elevation angle through hierarchical sparse reconstruction to eliminate azimuth and range interference and retain only the linear phase term of the elevation angle.

[0010] The second stage is: based on the elevation angle estimation result, the azimuth angle is estimated by retaining the elevation angle and azimuth angle information through the second set of fourth-order cumulant matrices;

[0011] Among them, the third stage is: based on the estimation results of the elevation angle and azimuth angle, the nonlinear phase difference between the near-field and far-field signals is combined through the third set of fourth-order cumulant matrices to separate the far-field and near-field signal sources and estimate the distance in layers.

[0012] The first set of fourth-order cumulant matrices is specifically constructed as follows:

[0013] The first set of fourth-order cumulant matrices The construction is based on four groups of symmetric reflection units, whose indices are: , , , Get the first set of fourth-order cumulant matrices Expressed as:

[0014]

[0015] Where, and Represents index, subscript Represents the row index of HRIS, subscript Represents the column index of HRIS; The representative index is The active RIS reflection unit receives the user signal, The representative index is The active RIS reflection unit receives the user signal, The representative index is The active RIS reflection unit receives the user signal, represents conjugation, The representative index is The active RIS reflection unit receives the user signal; Represents user source signal The fourth-order cumulant of is the elevation linear phase term, For users The elevation angle, For distance, is the wavelength; Represents the total number of user sources, is the number of near-field sources, is the number of far-field signal sources, and satisfies ; then build Projection to the signal subspace , used to solve the sparse coefficients in hierarchical sparse reconstruction ,in, for The signal subspace of

[0016] Design of layered elevation steering vector dictionary , gradually shrinking the dynamic tuning factor With step length To achieve the gradual reduction of the search space, combined with Norm minimization and weight matrices Solving for sparse coefficients , and finally output the elevation angle estimation result .

[0017] The method of estimating the azimuth angle by retaining the elevation angle and azimuth angle information based on the elevation angle estimation result by the second set of fourth-order cumulant matrices is specifically as follows:

[0018] Based on the elevation angle estimation results , calculate the second set of fourth-order cumulant matrices , which preserves the elevation angle and azimuth and the linear phase term and To prevent the elevation estimation error from being passed to the azimuth estimation stage, then construct Projection to the signal subspace , used to solve the sparse coefficients in hierarchical sparse reconstruction ,in, for The signal subspace of

[0019] Designing a hierarchical azimuth-steering vector dictionary , gradually shrinking the dynamic tuning factor With step length To gradually reduce the search space, calculate the weight matrix , and combined with The azimuth angle estimation result is obtained by minimizing the norm.

[0020] The specific construction method of the second group of fourth-order cumulant matrices is:

[0021] The second set of fourth-order cumulant matrices The construction of is based on four sets of symmetric unit pairs, whose indices are: , , , , and obtain the second set of fourth-order cumulant matrices Expressed as:

[0022]

[0023] Where, The representative index is The active RIS reflection unit receives the user signal; The representative index is The active RIS reflection unit receives the user signal; is the linear phase term, is the quadratic nonlinear phase term related to distance, For users Elevation angle estimated in the first stage.

[0024] The method of separating the far-field and near-field signal sources and estimating the distance in layers is specifically as follows:

[0025] Based on the estimation results of elevation and azimuth angles, the third set of fourth-order cumulant matrices Utilize the existence of quadratic nonlinear phase terms in near-field and far-field signals The characteristics of the far-field and near-field signal sources are separated and then constructed Projection to the signal subspace ,in, is the signal subspace;

[0026] Using calculated projections , and design a distance dictionary and the weight matrix Weighted Norm minimization to solve sparse coefficients , the final output estimated distance .

[0027] The third group of fourth-order cumulant matrices is specifically constructed as follows:

[0028] The third set of fourth-order cumulant matrices The construction of is based on three sets of unit pairs, whose indices are: , , , and get the third set of fourth-order cumulant matrices Expressed as:

[0029]

[0030] Where, The representative index is The active RIS reflects the user signal received by the unit.

[0031] The beneficial effects of the present invention are:

[0032] 1. This invention achieves parameter decoupling and computational complexity optimization. Through a three-stage hierarchical estimation strategy (elevation → azimuth → distance), the traditional two-dimensional angle joint search is decoupled into a one-dimensional search. Combined with hierarchical sparse reconstruction, the search space is dynamically contracted, significantly reducing computational complexity.

[0033] 2. The present invention utilizes the HRIS active sensing mode to receive pilot signals, constructs three sets of fourth-order cumulant matrices through symmetrical units, realizes analysis and processing of user signals, and obtains position parameter information.

[0034] 3. The present invention integrates a hierarchical search mechanism, designs tuning factors and trust intervals to gradually shrink the spectrum search space, and implements a progressive search from coarse to fine. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a system model diagram of the present invention;

[0036] Figure 2 It is a diagram of the hierarchical search mechanism of the present invention;

[0037] Figure 3 is a flow chart of the present invention;

[0038] Figure 4 is a flowchart of elevation angle estimation of the present invention;

[0039] Figure 5 is a flow chart of azimuth estimation of the present invention;

[0040] Figure 6 is a flow chart of distance estimation of the present invention;

[0041] Figure 7 This is a performance difference diagram of the RMSE between the present invention and the typical comparison algorithm at different SNR levels;

[0042] Figure 8 This is a performance difference diagram of the RMSE between the present invention and the typical comparison algorithm under different snapshot numbers;

[0043] Figure 9 This is a diagram comparing the computational overhead between the hierarchical search method and the global search method of the present invention. DETAILED DESCRIPTION

[0044] The present invention will be further described below in conjunction with the accompanying drawings and specific implementation methods based on the description.

[0045] Example 1: A hybrid source localization method based on HRIS and hierarchical sparse reconstruction, the specific steps are as follows:

[0046] Step 1: The base station BS sends a pilot signal to the active RIS reflection unit to estimate and decouple the BS-RIS channel.

[0047] Step 1.1: BS-RIS channel estimation phase: The base station BS digitally modulates the transmitted signal and sends it to the active RIS via free space;

[0048] Step 1.2: The active RIS reflection unit captures the BS pilot signal. The active RIS reflection unit is in sensing mode to receive the pilot signal. The active RIS reflection unit estimates the BS-RIS channel. , and then feeds the received signal back to the BS through the backhaul link. The BS receives the signal sent by the active RIS as a BS-RIS signal transmission as shown in formula (1):

[0049]

[0050] Where, The active RIS reflection unit receives the signal from the BS; is the transmit power; A pilot signal sent by a base station; , is the BS-RIS channel matrix, is the steering vector of the RIS reflection unit; is the steering vector of the BS antenna array; is the fading coefficient, is the noise of the BS-RIS channel.

[0051] Specifically, the BS-RIS channel estimation in Step 1.1-Step 1.2 is used as the initialization stage, and its channel estimation parameters are input into the positioning system as known conditions. Step 1.1-Step 1.2 realizes channel decoupling, solves the error interference between cascade channels, and prepares for subsequent user signal processing.

[0052] Step 2: Based on the decoupled BS-RIS channel, the user equipment UE sends a user pilot signal to the active RIS reflection unit.

[0053] Specifically, when the BS-RIS channel estimation is completed, it enters the positioning phase. The user sends a pilot signal to the active RIS reflection unit. The uplink pilot signal transmitted by the user source is received at the active RIS. As shown in formula (2):

[0054]

[0055] Where, For users Transmitted signal power; For users a transmitted pilot signal; , containing the steering vectors of near-field and far-field sources; is the near-field steering vector from the user source to the active RIS reflector unit; is the far-field steering vector from the user source to the active RIS reflector unit, is the distance from the user to the RIS center reflection unit, is the elevation angle from the user to the RIS central reflection unit, is the azimuth from the user to the RIS central reflection unit; For users The fading coefficient of is zero in mean and has a variance of Additive white Gaussian noise.

[0056] Users in the near field The distance to any active RIS reflector unit is expanded into:

[0057]

[0058] Where, For users The distance to any active RIS reflector unit; For users Distance to the RIS central reflector unit; are the coordinates of the RIS reflection unit, is the reflection unit index, which contains Axis Index , is the number of rows of reflection units, Axis Index , is the number of columns of reflection units; The present invention assumes that the RIS is deployed in the YOZ plane, so the x-axis coordinate of the RIS reflector unit is ,and then , is the wave vector;

[0059] When the distance between the RIS unit and the target changes very little, using the binomial expansion, equation (4) can be expanded as follows:

[0060]

[0061] In the far field, since the source point is far away from the RIS, the wavefront of the electromagnetic wave can be approximately regarded as a plane wave, so the far-field source wave arrival distance Expressed as:

[0062]

[0063] Therefore, users in the near field The array manifold is as follows:

[0064]

[0065] Where, For active RIS reflector unit 1 to user distance; Active RIS reflection unit To the user distance; is the number of active reflection units.

[0066] Far-field users The array manifold is as follows:

[0067]

[0068] is the coordinate of active RIS reflection unit 1; Active RIS reflection unit 's coordinates.

[0069] According to the above distance derivation, the user Near-field delay to any reflector for:

[0070]

[0071] Far-field delay for:

[0072]

[0073] Step 3: The active RIS reflector unit constructs three sets of fourth-order cumulant matrices based on the received user pilot signal, and gradually completes the position parameter estimation in three stages based on the dynamic tuning factor and step size layering. The position parameters include elevation angle, azimuth angle and range.

[0074] Step 3.1: The first stage is to decouple the elevation and azimuth angles based on the first set of fourth-order cumulant matrices, construct the signal subspace projection, and estimate the elevation angle through hierarchical sparse reconstruction to eliminate the azimuth and range interference and retain only the linear phase term of the elevation angle.

[0075] Specifically, four groups of symmetric reflection units are selected, and their indexes are: , , , . Cell Index , using the above-mentioned specific symmetrical reflection unit to eliminate and , construct the first set of fourth-order cumulant matrices :

[0076]

[0077] Where, and Represents index, subscript Represents the row index of HRIS, subscript Represents the column index of HRIS; The representative index is The active RIS reflection unit receives the user signal, The representative index is The active RIS reflection unit receives the user signal, The representative index is The active RIS reflection unit receives the user signal, represents conjugation, The representative index is The active RIS reflection unit receives the user signal; Represents user source signal The fourth-order cumulant of is the elevation linear phase term, For users The elevation angle, For distance, is the wavelength; Represents the total number of user sources, is the number of near-field sources, is the number of far-field signal sources, and satisfies .

[0078] In order to avoid the index problem of negative and zero, the reflection unit index needs to be updated, so , .matrix becomes:

[0079]

[0080] matrix It can be written as , is the fourth-order cumulant matrix containing all user source signals, Represents user source signal The fourth-order cumulant of is the elevation steering vector that includes all far-field and near-field signal sources, and the array manifold , For users elevation angle.

[0081] Since the matrix calculated above The elevation angle information is included in , and the eigenvalue decomposition (SVD) is performed on it:

[0082]

[0083] Where, is the signal subspace, is a diagonal matrix containing the eigenvalues ​​of the signal subspace; is the noise subspace; is a diagonal matrix containing the eigenvalues ​​of the noise subspace; stands for conjugate transpose.

[0084] In order to concentrate the signal energy in the signal subspace, construct Projection to the signal subspace , so that the noise energy is truncated, and the projection is used to solve the sparse coefficients in the following hierarchical sparse reconstruction .

[0085] use and Norm solution for sparse coefficients To achieve elevation angle estimation, in order to avoid global search, the estimation process integrates layered steering vectors, shrinking the spectral search space layer by layer and improving the angle spectrum search.

[0086] Specifically, the hierarchical search estimates the number of search layers to be ,in is the maximum number of layers. Layered elevation steering vector dictionary for:

[0087]

[0088] Where, elevation angle The steering vector of the active reflection unit is , is the number of active units; For the The elevation search range of the layer, For the Minimum search elevation angle of the layer, For the Maximum search elevation angle of the layer; For the The elevation search step of the layer; in order to detect the specific range of elevation angles to be estimated in the initial layer, the initial step size needs to be set , step size reduction factor , achieving progressively more refined searches in subsequent layers.

[0089] Furthermore, the reconstruction error is performed, that is, the signal subspace projection is constructed Dictionary of Design and sparse coefficients The difference between the reconstructions. As shown in formula (14):

[0090]

[0091] Where, is the elevation tuning factor, which controls the upper limit of the reconstruction error.

[0092] Furthermore, the reconstruction error is mainly used to find the sparse coefficient In order to further suppress the noise and false elevation angles on the sparse coefficients Solve the impact and design the weight matrix :

[0093]

[0094] Where, , and are the steering vectors of the real elevation angle and the false elevation angle respectively; is the product of the true elevation steering vector and the noise subspace matrix; is the product of the false elevation steering vector and the noise subspace matrix. and There is orthogonality between ,and .

[0095] Furthermore, we introduce weighted Norm minimization, using the weight matrix Enhanced Matrix The sparsity in , as shown in formula (16):

[0096]

[0097] That is, the weight matrix and sparse coefficients Multiplying, the redundant components corresponding to the non-real elevation angle and noise interference are The effective component corresponding to the true elevation angle is small, so the false elevation angle response and noise interference are compressed to zero during the optimization process.

[0098] Furthermore, after weighted Sparse coefficients with minimized norm The non-zero rows of correspond to the positions of the real signal sources in the parameter space. Therefore, after each layer of sparse reconstruction, the sparse coefficients The energy accumulation and summation of the row vectors are obtained As shown in formula (17):

[0099]

[0100] in, For the Layer dictionary The energy accumulation value of a codeword on all signal sources, is the codeword index; For the The code word corresponds to The above formula measures the sparse coefficient of the first signal source. The matching degree between the codeword and the received signal. Then, the energy is sorted in ascending order, as shown in formula (18):

[0101]

[0102] Where, Represents the cumulative sum of the elevation codeword energy value Sort by is the set of elevation codeword energy values ​​returned after sorting; The original index of the elevation codeword energy value returned after sorting.

[0103] Then, according to the sorted elevation codeword energy values, the front codeword with the largest energy is cut out. indexes, forming an index set :

[0104]

[0105] Further, according to the filtered codeword index After completing the adaptation angle estimation of the current layer, a half-width is constructed with each optimal steering vector as the center. The confidence interval of , For the Layer elevation angle search step size, update step size at the same time , is the step size reduction factor; tuning factor , is the regularization attenuation rate. When the number of layers does not exceed the maximum number of layers , the search step size is greater than the minimum step size , enter the next level of search, otherwise output the estimated elevation angle .

[0106] Step 3.2: The second stage is to estimate the azimuth angle based on the elevation angle estimation result by retaining the elevation angle and azimuth angle information through the second set of fourth-order cumulant matrices.

[0107] Specifically, four pairs of symmetric elements are selected, whose indexes are: , , , . Cell Index , The above-mentioned specific symmetric unit pairs are used to suppress the quadratic term related to distance. , only the linear phase related to the angle is retained and ,reserve Used to match the elevation angle to prevent the estimation error from being transferred to the azimuth estimation stage. The second set of fourth-order cumulant matrices as follows:

[0108]

[0109] Where, The representative index is The active RIS reflection unit receives the user signal; The representative index is The active RIS reflection unit receives the user signal; is the linear phase term, is the quadratic nonlinear phase term related to distance, For users Elevation angle estimated in the first stage.

[0110] Further, It only contains the angle information of the user source. In order to obtain the signal subspace, Perform eigendecomposition (SVD):

[0111]

[0112] Where, is the signal subspace, is a diagonal matrix containing the eigenvalues ​​of the signal subspace; is the noise subspace; is a diagonal matrix containing the eigenvalues ​​of the noise subspace.

[0113] Furthermore, using the obtained ,calculate Projection to the signal subspace , realizing the concentration of signal energy in the signal subspace and cutting off the noise energy.

[0114] Further, using And the elevation angle results estimated in the first stage, solve the sparse coefficient Implement azimuth estimation. Design layered steering vectors , dynamically shrink the spectrum search space, calculate the weight matrix implement Norm minimization suppresses false azimuths and noise parameters and improves angle spectrum search.

[0115] Specifically, based on the elevation angle results estimated in the first stage, the azimuth angle dictionary is constructed:

[0116]

[0117]

[0118] Where, For the The elevation angle estimation range of the layer, For the Minimum search elevation angle of the layer, For the Maximum search elevation angle of the layer; For the The search step of the layer.

[0119] Specifically, in order to further suppress noise and false elevation angles on sparse coefficients Solve the impact and construct the azimuth weight matrix :

[0120]

[0121]

[0122] Where, is the steering vector of the true azimuth; is the steering vector of the false azimuth, and Orthogonal, so , and Not orthogonal, , so that in the subsequent weighted In the process of norm minimization, The redundant components corresponding to the false azimuth and noise are compressed to zero.

[0123] Specifically, using projection ,dictionary and the weight matrix Weighted Minimize the norm to obtain sparse coefficients :

[0124]

[0125] Where, For the Tuning factor for layer azimuth.

[0126] Specifically, based on the sparse coefficient Calculate the codeword energy of the row elements and return the previous The index of the matching codeword with the largest energy:

[0127]

[0128]

[0129] Where, Representative The sparse coefficients of all user signal sources corresponding to the codewords; is the set of azimuth codeword energy values ​​returned after sorting; The original index of the azimuth codeword energy value returned after sorting.

[0130] Then, according to the sorted azimuth codeword energy values, the front with the largest energy is cut out. indexes, forming an index set :

[0131]

[0132] Specifically, while building Trust interval of the layer , For the Layer azimuth search step, update step , tuning factor , increase the number of layers When the number of layers does not exceed the maximum number of layers , the search step is greater than the minimum step , enter the next level of search, otherwise output parameters .

[0133] Step 3.3: The third stage is to separate the far-field and near-field signal sources and estimate the distance in layers based on the estimation results of the elevation and azimuth angles by combining the nonlinear phase difference between the near-field and far-field signals through the third set of fourth-order cumulant matrices.

[0134] Specifically, three groups of unit pairs are selected, whose indexes are: , , By constructing a fourth-order cumulant matrix containing far-field angles, near-field angles and distance parameters , the near-field and far-field components can be effectively separated from the mixed source:

[0135]

[0136] Furthermore, the matrix It only contains angle and distance information. In order to obtain the signal subspace, it is subjected to eigendecomposition (SVD):

[0137]

[0138] Where, is the signal subspace; is a diagonal matrix containing the eigenvalues ​​of the signal subspace; is the noise subspace; is a diagonal matrix containing the eigenvalues ​​of the noise subspace. Construct Projection to the signal subspace , the signal energy is concentrated in the signal subspace and the noise energy is truncated.

[0139] Further, using Sum norm to solve sparse coefficients To achieve elevation angle estimation, in order to avoid global search, the estimation process integrates layered steering vectors, shrinking the spectral search space layer by layer and improving the angle spectrum search.

[0140] First, design a distance dictionary . Define the hierarchical distance set , For distance step, press attenuation , is the distance step reduction factor. Initial distance range , is the minimum measurable distance ( ), is the far-field boundary of the Fresnel zone, is the array aperture, and the subsequent layers ( ) The distance search range is centered around the optimal distance estimate of the previous layer, so the array manifold for:

[0141]

[0142]

[0143]

[0144]

[0145] Wherein, formula (30) is the steering vector dictionary of far-field source and near-field source, is the far-field source dictionary, is the near-field source dictionary; the far-field source dictionary is written in detail as shown in formula (31), is the number of far-field signal sources, whose position information is only related to the direction of the incident signal and does not involve the distance parameter; the near-field source dictionary is detailed as shown in formula (32), is the number of near-field users, whose positions include angle and distance parameters; Formula (33) describes the manifold of the near-field mid-range layered array in detail, and are the users estimated in the first two stages respectively Elevation and azimuth.

[0146] Weight Matrix for:

[0147]

[0148]

[0149] Where, is the noise subspace; is the steering vector containing the true and false positions in the far and near fields, is the true distance, False distance. is the weight matrix of the real position in the far field; is the weight matrix of the false position in the near field; is the weight matrix of the real position in the near field; is the weight matrix of the false position in the near field; using the orthogonality of the steering vector and the noise subspace, we can get ,and Therefore, the spurious position information in the far and near fields is suppressed during the regularization process.

[0150] Using distance dictionary ,projection and the weight matrix Weighted Norm minimization to solve sparse coefficients :

[0151]

[0152] Where, For the Layer distance tuning factor.

[0153] Based on the sparse coefficients Perform energy accumulation and summing and sorting, and update the search range, number of layers, step size, and tuning factor:

[0154]

[0155] For the The cumulative sum of the sparse coefficients of all user signal sources corresponding to the range codewords, is the set of distance codeword energy values ​​returned after sorting; The original index of the distance codeword energy value returned after sorting.

[0156] According to the energy value of the sorted distance codewords, the top one with the largest energy is selected. indexes, forming an index set :

[0157]

[0158] Using distance index collection Build the Layer distance search range ,in, For the The optimal distance of the layer, and update the number of layers, step size, and tuning factor to start the search for the next level. The final output parameter is the distance At this point, the position parameters of the near-field source and the far-field source are estimated respectively.

[0159] The present invention will be further described below in conjunction with the accompanying drawings based on the above embodiments.

[0160] Figure 1This is a system model diagram of the hybrid source location method based on HRIS and hierarchical sparse reconstruction. MIMO array base station with 100 antennas, hybrid active and passive RIS, single-antenna users, the number of near-field sources is , the number of far-field sources is , and satisfies , and is a positive integer.

[0161] Figure 2 A schematic diagram of the hierarchical search mechanism is shown. In the first layer, the entire steering vector space is divided into several large angular ranges, and coarse-grained dictionary matching is performed within these ranges. When a codeword is determined to be the best match, a trust interval with a step length of half the width is constructed around this codeword, which serves as the search range for the next layer. A new dictionary is then generated within this subrange, and the above search process is repeated, refining the search layer by layer until the preset finest search layer is reached, achieving high-precision estimation.

[0162] Figure 3 This is a flow chart of the hybrid source location method based on HRIS and hierarchical sparse reconstruction. First, the HRIS is used to initialize the system, i.e., estimate the BS-RIS channel. The base station sends a pilot signal to the HRIS, which receives the base station signal and estimates the BS-RIS channel. This signal is then transmitted back to the base station via a backhaul link to complete the BS-RIS channel estimation. The initialization operation prevents the estimation errors between the two hop channels from interfering with each other. The user then sends a signal to the HRIS, which analyzes and processes the signal. This is to construct three sets of fourth-order cumulant matrices using the received user signals. , , They are used to estimate the elevation, azimuth and distance parameters respectively. Construct a hierarchical dictionary of three-dimensional position parameters , and set the step size And the search range is initialized for hierarchical sparse reconstruction. Then the initialization parameters are combined with weighted Norm minimization and error reconstruction to solve sparse coefficients , where the weight matrix Suppress redundant components such as non-real position parameters and noise interference. When the number of layers is greater than the maximum number of layers or the step size is less than the minimum search step size, the three-dimensional position parameters are output. .

[0163] Figure 4 The flowchart of the layered elevation angle estimation method is shown in Figure 2. In the initialization phase, based on the HRIS estimating the BS-RIS channel, the user sends a pilot signal to the active RIS. The active RIS uses the received user signal to construct a fourth-order cumulant matrix , then eigenvalue decomposition distinguishes the signal subspace from the noise subspace. In order to suppress the noise energy, construct Projection to the signal subspace Then enter the layered sparse reconstruction stage, first set the initialization parameters: the number of layers , initial search step length for elevation angle , initial tuning factor , Design Elevation Dictionary , and calculate the weight matrix , using the weight matrix calculated previously Solving weighted Norm minimization problem, suppressing sparse coefficients The redundant components corresponding to the non-real elevation angle and noise interference are then calculated. The codeword energy accumulation operation is then performed. , find The codeword energy that best matches the actual elevation angle is obtained, and the index of the best matching codeword energy is calculated to update the search range of the next level. Before entering the next level, update the elevation search range , tuning factor , search step , increase the current search level. Greater than or equal to the maximum number of layers or Greater than the minimum elevation step When , the estimated elevation angle is output , otherwise perform the angle estimation of the next level again.

[0164] Figure 5 This is the flowchart of the hierarchical azimuth estimation method. First, the fourth-order cumulant matrix , perform eigenvalue decomposition and extract the signal subspace corresponding to the user signal , and calculate the matrix Projection to the signal subspace . Hierarchical parameter initialization module, set the iteration parameters: initial search step , tuning factor ,use Figure 4 Estimated elevation angle Designing an overcomplete dictionary . Calculate the weight matrix , using the elevation angle estimated in the previous stage to design the steering vector in the weight matrix. Using the norm to solve the sparse coefficient matrix . The row elements of the accumulative summation are performed, and the steering vector index that best matches the signal is solved to determine the angle range for the next level search. Then the dynamic parameter update is performed: the azimuth search range of the next level , tuning factor , search step , increase the current level. When searching the level Greater than or equal to the maximum number of layers or Greater than the minimum azimuth search step When , the estimated azimuth is output, otherwise the next level of angle estimation is performed again until it is satisfied.

[0165] Figure 6 The active RIS reflection unit calculates the third set of fourth-order cumulants. , At the same time, the angle and distance information are retained. Since the far-field signal is only related to the angle parameter, while the near-field signal contains both angle and distance information. Therefore, the distance estimation selects three groups of sensors to construct a subarray to separate the position parameters of the far-field source and the near-field source, thereby estimating the position information of the far-field source and the near-field source respectively. SVD decomposition is performed to obtain the signal subspace , and ask Projection to the signal subspace . Set the number of iteration parameters , distance search step , tuning factor , using the angle information estimated in the first two stages to match the codeword of the real angle in the steering vector to prevent the estimation error from being passed to the distance estimation. Constructing the distance dictionary . Using the norm to solve the sparse coefficients Calculate codeword energy Find the index of the best distance guidance vector. Based on this, determine the distance range of the next level and dynamically update the parameters. The angle search range of the layer is set to , the tuning factor is updated to , the search step size is updated to , increase the current level. When searching the level Greater than or equal to the maximum number of layers or Greater than the minimum search step distance When , the distance parameter is output; otherwise, the angle estimation algorithm of the next level is executed until the stopping condition is met.

[0166] Figure 7 The RMSE performance difference between the proposed method and the classical method under different SNRs is shown in Figure 2. Figure 7 The performance of the proposed hybrid field source localization algorithm under different signal-to-noise ratio (SNR) conditions was verified. Figure 7In the figure, Figure (a) and Figure (b) respectively show the curves of the root mean square error (RMSE) of the positioning of far-field sources and near-field sources as the SNR changes. The experimental setting signal-to-noise ratio range is 0 dB to 20 dB, the step size is 5 dB, and the number of snapshots is set to 100. It can be seen from the results in the figure that the present invention is superior to the traditional two-stage multiple signal classification (TSMUSIC) method and the hybrid orthogonal matching (OMP) method in terms of far-field and near-field positioning accuracy. In far-field source positioning, when the SNR is high, the performance of the present invention is similar to that of TSMUSIC; under low SNR conditions, the method of the present invention shows significant advantages, reflecting stronger noise robustness and estimation accuracy. In the near-field source positioning task, since the angle and distance need to be estimated simultaneously, and the distance information depends on the signal model containing the nonlinear phase term, it is more susceptible to angle estimation errors and noise. However, the present invention introduces a weight matrix in the third stage, which effectively suppresses the interference caused by false angles and distances, thereby improving the estimation accuracy. From Figure 7 It can be seen that in the entire SNR range, the present invention always maintains the lowest RMSE, verifying its superior performance and stability in the mixed far-field and near-field scenarios.

[0167] Figure 8 This paper demonstrates the difference in RMSE performance between the proposed method and a typical comparison method under different snapshot numbers. In the experiment, the signal-to-noise ratio was fixed at 10 dB, and the snapshot number was increased from 50 to 300. 500 Monte Carlo simulations were performed for each snapshot number setting. Figure 8 Figures (a) and (b) show the RMSE trends of each algorithm for angle estimation and distance estimation, respectively, under different snapshot numbers. Simulation results show that the RMSE of each algorithm decreases with increasing snapshot number, indicating that more observations improve parameter estimation accuracy. In comparison, the proposed method outperforms the TSMUSIC and hybrid OMP methods in both angle and distance estimation, demonstrating superior overall performance. This performance advantage is primarily attributed to the sparse reconstruction mechanism introduced in the algorithm, which effectively suppresses the interference of non-real position parameters through regularization, maintaining high estimation accuracy, especially at low snapshot numbers. Specifically, Figure (a) shows that the RMSE reduction of the proposed method for angle estimation stabilizes when the number of snapshots reaches approximately 100, further verifying its stability and robustness under low snapshot conditions. In Figure (b), the convergence rate of distance estimation is significantly slower than that of angle estimation, indicating that the distance parameter is more sensitive to the amount of observation data. The main reason for this difference is that distance estimation needs to retain nonlinear phase information and decouple angle-distance parameters in the algorithm, which places higher requirements on the number of snapshots.

[0168] Figure 9The comparison results of the computational overhead between the hierarchical search strategy proposed in this invention and the traditional global uniform step-size search method are shown. Figure 9 (a) Comparison of computation time under different angle step sizes (e.g., from larger step size to smaller step size). Figure 9 (b) shows the computational overhead of the two methods for different distance step sizes (0.05 m to 1 m). To achieve efficient search, the present invention introduces a hierarchical multi-resolution mechanism: by gradually narrowing the search interval and dynamically adjusting the step size and tuning factor, it achieves layer-by-layer convergence from coarse to fine. Specifically, in the angle estimation stage, traditional global methods use a two-dimensional joint spectral search, which has high computational complexity. In contrast, the present invention adopts a phased optimization strategy of "elevation first, azimuth later," effectively reducing the search dimension and computational burden. In the distance estimation stage, the hierarchical mechanism constructs a trust interval based on the optimal solution of the previous layer and refines the search within this local range, thus avoiding the redundant computation caused by the global traversal search and significantly improving efficiency. It is worth noting that while the hierarchical strategy improves overall search efficiency, using an excessively large step size in the initial layer may cause the true angle or distance to fall into the grid gap, resulting in missed detections, causing initial estimation errors, and further affecting the convergence quality of subsequent layers. Therefore, to ensure the robustness and accuracy of the algorithm, the initial step size should be set to strike a balance between efficiency and resolution, avoiding excessive settings.

[0169] The specific embodiments of the present invention described above in conjunction with the accompanying drawings are only used to illustrate the technical principles and implementation methods of the present invention and do not constitute a limitation on the scope of protection of the present invention. For those skilled in the art, various equivalent transformations or substitutions can be made to the embodiments of the present invention based on the concept of the present invention without departing from the spirit and essence of the present invention, and these transformations or substitutions should all be included in the scope of protection of the present invention.

Claims

1. A hybrid source localization method based on HRIS and hierarchical sparse reconstruction, characterized by: Step 1: The base station BS sends a pilot signal to the active RIS reflection unit to estimate and decouple the BS-RIS channel; Step 2: Based on the decoupled BS-RIS channel, the user equipment UE sends a user pilot signal to the active RIS reflection unit; Step 3: The active RIS reflector unit constructs three sets of fourth-order cumulant matrices based on the received user pilot signal and gradually completes the position parameter estimation in three stages based on the dynamic tuning factor and step size layering. The position parameters include elevation angle, azimuth angle and range. The first stage is to decouple the elevation and azimuth angles based on the first set of fourth-order cumulant matrices, construct signal subspace projections, and estimate the elevation angle through hierarchical sparse reconstruction to eliminate azimuth and range interference and retain only the linear phase term of the elevation angle. The second stage is: based on the elevation angle estimation result, the azimuth angle is estimated by retaining the elevation angle and azimuth angle information through the second set of fourth-order cumulant matrices; Among them, the third stage is: based on the estimation results of the elevation angle and azimuth angle, the nonlinear phase difference between the near-field and far-field signals is combined through the third set of fourth-order cumulant matrices to separate the far-field and near-field signal sources and estimate the distance in layers.

2. The method according to claim 1, characterized in that The first set of fourth-order cumulant matrices is specifically constructed as follows: The first set of fourth-order cumulant matrices The construction is based on four groups of symmetric reflection units, whose indices are: , , , Get the first set of fourth-order cumulant matrices Expressed as: ； Where, and Represents index, subscript Represents the row index of HRIS, subscript Represents the column index of HRIS; The representative index is The active RIS reflection unit receives the user signal, The representative index is The active RIS reflection unit receives the user signal, The representative index is The active RIS reflection unit receives the user signal, represents conjugation, The representative index is The active RIS reflection unit receives the user signal; Represents user source signal The fourth-order cumulant of is the elevation linear phase term, For users The elevation angle, For distance, is the wavelength; Represents the total number of user sources, is the number of near-field sources, is the number of far-field signal sources, and satisfies ; then build Projection to the signal subspace , used to solve the sparse coefficients in hierarchical sparse reconstruction ,in, for The signal subspace of Design of layered elevation steering vector dictionary , gradually shrinking the dynamic tuning factor With step length To achieve the gradual reduction of the search space, combined with Norm minimization and weight matrices Solving for sparse coefficients , and finally output the elevation angle estimation result .

3. The method according to claim 1, characterized in that The method of estimating the azimuth angle by retaining the elevation angle and azimuth angle information based on the elevation angle estimation result by the second set of fourth-order cumulant matrices is specifically as follows: Based on the elevation angle estimation results , calculate the second set of fourth-order cumulant matrices , which preserves the elevation angle and azimuth and the linear phase term and To prevent the elevation estimation error from being passed to the azimuth estimation stage, then construct Projection to the signal subspace , used to solve the sparse coefficients in hierarchical sparse reconstruction ,in, for The signal subspace of Designing a hierarchical azimuth-steering vector dictionary , gradually shrinking the dynamic tuning factor With step length To gradually reduce the search space, calculate the weight matrix , and combined with The azimuth angle estimation result is obtained by minimizing the norm.

4. The method according to claim 3, characterized in that The specific construction method of the second group of fourth-order cumulant matrices is: The second set of fourth-order cumulant matrices The construction of is based on four sets of symmetric unit pairs, whose indices are: , , , , and obtain the second set of fourth-order cumulant matrices Expressed as: ； Where, The representative index is The active RIS reflection unit receives the user signal; The representative index is The active RIS reflection unit receives the user signal; is the linear phase term, is the quadratic nonlinear phase term related to distance, For users Elevation angle estimated in the first stage.

5. The method according to claim 1, wherein The method of separating the far-field and near-field signal sources and estimating the distance in layers is specifically as follows: Based on the estimation results of elevation and azimuth angles, the third set of fourth-order cumulant matrices Utilize the existence of quadratic nonlinear phase terms in near-field and far-field signals The characteristics of the far-field and near-field signal sources are separated and then constructed Projection to the signal subspace ,in, is the signal subspace; Using calculated projections , and design a distance dictionary and the weight matrix Weighted Norm minimization to solve sparse coefficients , the final output estimated distance .

6. The method according to claim 5, characterized in that The third group of fourth-order cumulant matrices is specifically constructed as follows: The third set of fourth-order cumulant matrices The construction of is based on three sets of unit pairs, whose indices are: , , , and get the third set of fourth-order cumulant matrices Expressed as: ； Where, The representative index is The active RIS reflects the user signal received by the unit.

Images