A phased cooperative estimation method for XL-MIMO channel oriented to mixed far and near field propagation
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- YICHUN UNIVERSITY
- Filing Date
- 2026-04-10
- Publication Date
- 2026-06-19
Smart Images
Figure CN122247803A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wireless communication technology, specifically to a phased cooperative estimation method for XL-MIMO channels oriented towards hybrid near-field and far-field propagation. Background Technology
[0002] With the development of sixth-generation (6G) mobile communication systems, extremely large-scale MIMO (XL-MIMO) has become an important technical means to achieve ultra-high spectral efficiency and high-precision spatial sensing. When the base station antenna scale reaches hundreds or even thousands of array elements, the array aperture increases significantly, and the far-field propagation assumption in traditional XL-MIMO systems gradually becomes invalid. In actual propagation environments, the propagation path between user equipment and base station may simultaneously contain far-field plane wave components and near-field spherical wave components, thus forming hybrid far-field propagation characteristics.
[0003] In mixed-field propagation environments, the far-field path is primarily characterized by the incident angle parameter, while the near-field path depends not only on the angle but also on the propagation distance, exhibiting a distinct spherical wave structure in its array response. Therefore, the far-field and near-field paths differ significantly in parameter space dimensions, array response models, and sparse structures. If traditional far-field channel models or unified parameter dictionaries are still used for channel estimation, it will be difficult to accurately characterize the mixed-field propagation characteristics, leading to a significant increase in channel estimation errors.
[0004] To address the aforementioned issues, researchers have proposed a series of channel estimation methods based on compressed sensing theory in recent years. These methods construct dictionaries in the angle or polar coordinate domains and leverage channel sparsity to recover path parameters. However, in XL-MIMO mixed-field scenarios, existing methods still have the following shortcomings: First, many methods employ a unified dictionary to jointly model the far-field and near-field paths, making it difficult to simultaneously account for the propagation characteristics of plane waves and spherical waves, leading to severe model mismatch problems. Second, when both far-field and near-field paths exist simultaneously, unified sparse recovery or simple serial recovery is prone to residual contamination and support misselection, causing interference between different propagation components and affecting estimation accuracy. Third, discrete grid dictionaries introduce grid mismatch problems, especially in XL-MIMO systems, where even small deviations in angle and distance parameters can lead to significant array response errors. Finally, due to the significant increase in array size and parameter dimensions, traditional sparse recovery algorithms also face challenges in terms of complexity and stability.
[0005] Therefore, how to achieve high-precision, low-mismatch XL-MIMO channel estimation in a mixed far-field and near-field propagation environment by using reasonable parameter modeling and estimation strategies while controlling computational complexity remains an important technical problem that urgently needs to be solved. Summary of the Invention
[0006] To address the shortcomings and deficiencies of existing technologies, this invention provides a staged collaborative estimation method for XL-MIMO channels in mixed far- and near-field propagation. This method achieves high-precision estimation of mixed-field channels by constructing a far- and near-field domain dictionary and combining it with a staged support recovery and continuous parameter refinement mechanism.
[0007] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a staged cooperative estimation method for XL-MIMO channels oriented towards hybrid near-field and far-field propagation, comprising the following steps: S1. Construct a multi-subcarrier joint signal observation model, and perform subarray partitioning and parameter initialization; S2. Hybrid field channel modeling and independent domain modeling of far-field and near-field dictionaries: The hybrid field channel model includes a far-field plane wave propagation model and a near-field spherical wave propagation model. A far-field angle domain dictionary is established based on the far-field plane wave propagation model, and a near-field polar coordinate domain dictionary with joint representation of angle and distance is established based on the near-field spherical wave propagation model, so as to realize independent domain modeling of far-field and near-field components. S3, Phased Collaborative Estimation: The first stage involves performing joint sparse support recovery on the far-field angle domain dictionary to obtain the initial parameter estimates of the far-field path, and then extracting the reconstructed signal from the observation data to obtain the updated residual signal. The second stage involves using the updated residuals to perform sparse recovery on the near-field dictionary to obtain the initial angle and distance parameters of the near-field path. S4. Continuous parameter refinement: Gradient descent algorithm is used to refine the far-field angle parameters, near-field angle parameters and distance parameters in the continuous domain to reduce the grid mismatch error caused by the discrete dictionary, and the residual coupling between far-field and near-field paths is further suppressed by the alternating update mechanism. S5. Merge the far-field and near-field refinement results, and output the hybrid field channel estimation matrix. .
[0008] Preferably, in step S1, constructing a multi-subcarrier joint signal observation model and performing subarray partitioning and parameter initialization specifically includes: Base station configuration for uplink time-division duplex XL-MIMO systems The antennas form a uniform linear array, with the antenna spacing equal to the carrier wavelength λ; the number of antennas for the user equipment is... ; Base station The antenna is uniformly divided into several subarrays, and signal processing is performed on a subarray basis. A single-time-slot signal observation model and a single-carrier received signal model are established in sequence, and all subcarrier signals are stacked together to form an equivalent matrix observation model of the joint received signal matrix. The joint received signal matrix is composed of the superposition of the pilot matrix, the multi-subcarrier channel matrix and the noise matrix.
[0009] Preferably, in step S2, the hybrid field channel modeling specifically includes: The wireless propagation channel is decomposed into far-field path components and near-field path components. The far-field path consists of plane wave signals with a propagation distance greater than the array aperture, and the array response is determined only by the incident angle and is independent of the propagation distance. The near-field path consists of spherical wave signals with a propagation distance less than the Rayleigh distance, and the array response is determined by both the incident angle and the propagation distance. Based on the above component decomposition, the complex channel response received by the base station is composed of the superposition of far-field components and near-field components. Specifically, the far-field component is obtained by superimposing the path gain and angle response of multiple far-field paths, and the near-field component is obtained by superimposing the path gain, range phase term and angle-range joint response of multiple near-field paths. The resulting hybrid channel is the superposition of the far-field channel matrix and the near-field channel matrix.
[0010] Preferably, in step S2, the independent domain modeling of the far-field and near-field dictionaries specifically includes: The far-field angle domain dictionary is generated based on the far-field plane wave propagation model. The far-field manifold is determined by the incident angle, antenna index, antenna spacing, signal wavelength, and total number of antennas. The channel response in each direction is calculated using a discrete angle sequence to form the far-field dictionary. ; The near-field polar coordinate domain dictionary is generated based on the near-field spherical wave propagation model. The near-field manifold is determined by the propagation distance, incident angle, subcarrier frequency, and speed of light. The channel response for each path is calculated using two-dimensional sequences of discrete distance and discrete angle to construct the near-field dictionary. .
[0011] Preferably, in step S3, during the phased collaborative estimation: The first stage of far-field channel estimation is as follows: taking the subarray received signal as input, a coarse far-field channel estimation is performed using the synchronous orthogonal matched pursuit algorithm on the far-field angle domain dictionary. Support set selection and path gain estimation are performed through the far-field dictionary to obtain the initial parameters of the far-field path. The far-field subarray signal is reconstructed based on the initial far-field parameters and the far-field dictionary. The reconstructed far-field subarray signal is then subtracted from the subarray received signal to obtain the updated residual signal.
[0012] Preferably, the second-stage near-field sparse recovery specifically includes: Using the updated residual signal obtained in the first stage as input, a coarse estimation of the near-field channel is performed using the synchronous orthogonal matching pursuit algorithm on the near-field polar coordinate domain dictionary. The support set selection and path gain estimation are completed through the near-field dictionary, and the initial incident angle parameters and initial propagation distance parameters corresponding to the near-field path are directly obtained.
[0013] Preferably, in step S4, the gradient descent algorithm is used to perform alternating far-field and near-field optimization and refinement, specifically including: First, using the residual signal obtained from the near-field channel estimation as a constraint, gradient descent optimization is performed on the incident angle parameter of the far-field channel to complete the far-field parameter refinement; then, using the refined far-field residual signal as a constraint, gradient descent optimization is performed on the incident angle parameter and propagation distance parameter of the near-field channel to complete the near-field parameter refinement. The alternating update mechanism continuously reduces the parameter grid mismatch error introduced by the discrete dictionary grid partitioning through cyclic iteration of far-field refinement and near-field refinement, and further suppresses the residual coupling between the far-field channel path and the near-field channel path, thereby improving the estimation accuracy of angle parameters and distance parameters.
[0014] Preferably, in step S5, the far-field and near-field refinement results are combined to output a hybrid field channel estimation matrix. Specifically, it includes: The far-field channel estimation results refined by continuous domain gradient descent are superimposed and merged with the near-field channel estimation results to obtain the hybrid field channel estimation results corresponding to a single antenna subarray level. The hybrid field channel estimation results of all subarrays are then stacked according to the order of antenna subarray division to finally generate and output a complete global hybrid field channel estimation matrix. .
[0015] The staged cooperative estimation method for XL-MIMO channels oriented towards hybrid near-field and far-field propagation proposed in this invention has the following beneficial effects: The present invention provides a phased collaborative estimation method for XL-MIMO channels in a hybrid near-field propagation environment. By constructing a near-field domain dictionary and combining it with a phased support recovery and continuous parameter refinement mechanism, the method achieves high-precision, low-mismatch, and high-robustness XL-MIMO channel estimation in a hybrid near-field environment while controlling the computational complexity of large-scale antenna arrays.
[0016] The method of this invention adopts independent modeling of far-field angle domain dictionary and near-field polar coordinate domain dictionary, which can accurately match the propagation characteristics of plane wave and spherical wave, fundamentally avoiding the model mismatch problem caused by unified dictionary; it adopts a phased collaborative strategy of far-field priority recovery and residual-driven near-field estimation, first stripping the far-field reconstruction signal and then carrying out near-field sparse recovery, which greatly reduces the mutual interference between far-field and near-field paths and improves the accuracy of support set selection.
[0017] Finally, SOMP signal recovery and gradient descent optimization are used to refine the parameters in the continuous domain, which can significantly reduce the estimation bias caused by the mismatch of the discrete dictionary grid, and further suppress residual coupling through the alternating update mechanism. Finally, the subarray-level estimation results are merged and stacked to form a global channel matrix, which greatly improves the channel estimation accuracy and system robustness while ensuring low computational complexity, and is more suitable for the actual communication scenarios of 6G ultra-large-scale antenna arrays. Attached Figure Description
[0018] Figure 1 This is a flowchart of the phased cooperative estimation method for XL-MIMO channels for hybrid near-field and far-field propagation according to the present invention. Figure 2 The graph shows a comparison of the normalized mean square error (NMSE) performance of the method of this invention and existing algorithms under different pilot length conditions. Figure 3 This is a comparison curve of the NMSE performance of the method of the present invention and existing algorithms under different SNR conditions; Figure 4 This is a comparison chart of the achievable speed performance of the method of this invention and existing algorithms. Detailed Implementation
[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Example 1
[0020] like Figure 1 As shown, the XL-MIMO channel staged cooperative estimation method for hybrid near-field and far-field propagation of the present invention includes the following steps: S1. Construct a multi-subcarrier joint signal observation model, and perform subarray division and parameter initialization.
[0021] Among them, the construction of a multi-subcarrier joint signal observation model stacks the received signals on multiple subcarriers in the matrix dimension to form a unified equivalent observation structure, thereby improving the frequency domain diversity gain and enhancing the stability of sparse recovery; subarray partitioning divides the base station's full array antenna into several non-overlapping subarray units at fixed intervals, with each subarray serving as an independent signal processing unit to reduce the computational complexity and storage overhead caused by large-scale antennas; parameter initialization can set the initial support set, path gain vector, initial angle / distance values, and iterative convergence threshold for subsequent sparse recovery and continuous refinement.
[0022] Base station configuration for uplink time-division duplex XL-MIMO systems The antennas form a uniform linear array, with the antenna spacing equal to the carrier wavelength λ; the number of antennas for the user equipment is... ; base station The antenna is uniformly divided into several subarrays, and signal processing is performed on a subarray basis. A single-time-slot signal observation model and a single-carrier received signal model are established in sequence, and all subcarrier signals are stacked together to form an equivalent matrix observation model of the joint received signal matrix. The joint received signal matrix is composed of the superposition of the pilot matrix, the multi-subcarrier channel matrix and the noise matrix.
[0023] The signal observation model can first be represented as: in, For the receiving matrix, Let m be the channel path vector of the m-th subcarrier and p-th time slot. Pilot signal, It is additive white Gaussian noise.
[0024] Considering the received signals of all subcarriers, the signal matrix received by the base station can be represented as:
[0025] in, Let m be the received signal matrix of the m-th subcarrier. For pilot matrix, For channel path vectors, This is the noise vector.
[0026] To further model the signal, we consider jointly stacking the received signals of all subcarriers to obtain an equivalent observation model in matrix form:
[0027] in, Represents the joint received signal matrix. For a multi-subcarrier channel matrix, This is the noise matrix.
[0028] S2. Hybrid field channel modeling and independent domain modeling of far-field and near-field dictionaries: The hybrid field channel model includes a far-field plane wave propagation model and a near-field spherical wave propagation model. A far-field angle domain dictionary is established based on the far-field plane wave propagation model, and a near-field polar coordinate domain dictionary with joint representation of angle and distance is established based on the near-field spherical wave propagation model, so as to realize independent domain modeling of far-field and near-field components.
[0029] Hybrid-field channel modeling specifically involves decomposing the wireless propagation channel into far-field path components and near-field path components. The far-field path consists of plane wave signals with a propagation distance greater than the array aperture, and the array response is determined solely by the incident angle and is independent of the propagation distance. The near-field path consists of spherical wave signals with a propagation distance less than the Rayleigh distance, and the array response is determined by both the incident angle and the propagation distance.
[0030] Based on the above component decomposition, the complex channel response received by the base station is composed of the superposition of far-field and near-field components. Specifically, the far-field component is obtained by superimposing the path gains and angle responses of multiple far-field paths, while the near-field component is obtained by superimposing the path gains, range-phase terms, and angle-range joint responses of multiple near-field paths. The resulting hybrid channel is the superposition of the far-field channel matrix and the near-field channel matrix, specifically expressed as follows:
[0031] in, and These represent the far-field and near-field channel matrices, respectively.
[0032] Independent domain modeling of far-field and near-field dictionaries specifically includes: The far-field angle domain dictionary is generated based on the far-field plane wave propagation model. The far-field manifold is determined by the incident angle, antenna index, antenna spacing, signal wavelength, and total number of antennas. The calculation formula is as follows:
[0033] in, Let n be the incident angle, n be the antenna index, and d be the spacing between antennas. λ is the signal wavelength, and N is the total number of antennas.
[0034] The far-field dictionary is generated by discrete angles. The process involves using the linspace function to generate multiple angles and calculating the channel response in different directions based on these angles. Each dictionary element corresponds to different angle information and the corresponding channel response, ultimately forming the far-field dictionary. .
[0035] The near-field polar coordinate domain dictionary is generated based on the near-field spherical wave propagation model. The near-field manifold is determined by the propagation distance, incident angle, subcarrier frequency, and speed of light. The calculation of the near-field manifold is based on the following formula:
[0036] Where r is the near-field distance, f is the angle, f is the signal frequency, and c is the speed of light.
[0037] The near-field dictionary first determines the distance range of the path *r*, then generates the corresponding manifold based on the spherical wave model. The dictionary matrix, by calling the near-field manifold function and combining different distance and angle information, generates the corresponding channel response, ultimately forming the near-field dictionary. .
[0038] Using the far-field dictionary Near-field dictionary The independent construction enables decoupled modeling of different physical propagation mechanisms in hybrid field channels: the far-field dictionary focuses on one-dimensional angle space, efficiently covering the directional characteristics of plane waves at low dimensional cost; the near-field dictionary extends to two-dimensional polar coordinate space, fully characterizing the distance-angle coupling structure of spherical waves; the two are strictly distinguished in terms of parameter dimension, manifold generation basis and atomic physical meaning, which not only avoids model mismatch caused by a unified dictionary, but also provides a structurally adapted and semantically clear dictionary input basis for subsequent staged sparse recovery, thereby supporting the subsequent process of far-field priority stripping and residual-driven near-field recognition.
[0039] S3, Phased Collaborative Estimation: The first stage involves performing joint sparse support recovery on the far-field angle domain dictionary to obtain initial parameter estimates for the far-field path, and then extracting the reconstructed signal from the observation data to obtain the updated residual signal. Specifically: Using the subarray received signal as input, the synchronous orthogonal matching pursuit (SOMP) algorithm is used to perform coarse estimation of the far-field channel on the far-field angle domain dictionary. Support set selection and path gain estimation are performed through the far-field dictionary to obtain the initial parameters of the far-field path. The far-field subarray signal is reconstructed based on the far-field initial parameters and the far-field dictionary. The reconstructed far-field subarray signal is then subtracted from the subarray received signal to obtain the updated residual signal.
[0040] The first-stage coarse estimation achieves efficient joint sparse support recovery of multi-subcarrier jointly received signals by introducing a synchronous orthogonal matched pursuit algorithm in the subarray dimension. The consistency of physical modeling in the far-field angle domain dictionary ensures that support identification and gain estimation conform to the far-field plane wave propagation law. Furthermore, a closed-loop signal stripping mechanism is constructed through parameter-driven signal reconstruction and algebraic subtraction operations to accurately separate far-field components. Finally, the updated residual signal output satisfies the input constraints of the second-stage near-field sparse recovery in terms of dimension, signal-to-noise ratio, and component purity. This provides an implementable, verifiable, and reproducible execution step for the phased collaborative estimation strategy's far-to-near and residual-driven technical logic.
[0041] The second stage involves using the updated residuals to perform sparse recovery on the near-field dictionary to obtain the initial angle and distance parameters of the near-field path. The updated residual signal obtained in the first stage is used as input, and the synchronous orthogonal matching pursuit algorithm is used to perform coarse estimation of the near-field channel on the near-field polar coordinate domain dictionary. The support set selection and path gain estimation are completed through the near-field dictionary, and the initial incident angle parameters and initial propagation distance parameters corresponding to the near-field path are directly obtained.
[0042] The second stage uses the updated residual signal as input and leverages the two-dimensional structural characteristics of the near-field polar coordinate domain dictionary and the multi-channel joint sparse recovery capability of the synchronous orthogonal matching pursuit algorithm to ensure that the identified support set naturally corresponds to the joint index position of the near-field path in both the angle and distance dimensions. The path gain estimation based on this is directly related to this two-dimensional parameter combination, thus avoiding the error accumulation and support misalignment risk caused by modeling and estimating the angle and distance parameters separately and then performing pairing matching. This collaborative mechanism ensures that the initial parameters output by the second stage have clear physical interpretability and structural integrity.
[0043] S4. Continuous parameter refinement: The gradient descent algorithm is used to refine the far-field angle parameters, near-field angle parameters and distance parameters in the continuous domain, reducing the grid mismatch error caused by the discrete dictionary, and further suppressing the residual coupling between the far-field and near-field paths through the alternating update mechanism.
[0044] Specifically, the residual signal obtained from the near-field channel estimation is first used as a constraint to optimize the incident angle parameter of the far-field channel by gradient descent. Based on the continuous adjustability of the incident angle parameter, the far-field angle is fine-tuned during the gradient descent process to make the reconstructed signal more accurately match the far-field component remaining in the observation residual, thereby reducing the modeling mismatch caused by angle discretization and completing the fine-tuning of the far-field parameters.
[0045] Then, using the refined residual signal in the far field as a constraint, gradient descent optimization is performed on the incident angle parameter and propagation distance parameter of the near field channel. Based on the continuous adjustability of this parameter pair, the angle and distance are optimized simultaneously during the gradient descent process, so that the reconstructed signal can more accurately match the near field component remaining in the refined residual in the far field, thereby reducing the joint modeling mismatch caused by the dual discretization of angle and distance, and completing the refinement of near field parameters.
[0046] The alternating update mechanism refreshes the residual constraints of the far-field and near-field channels in each iteration through cyclical refinement of the far-field and near-field channels. This allows the two channels to alternately approximate the real physical path in the continuous parameter space, continuously reducing the parameter grid mismatch error introduced by the discrete dictionary grid partitioning. Furthermore, it suppresses the residual coupling between the far-field and near-field channel paths, improves the estimation accuracy of angle and distance parameters, and significantly reduces the residual coupling effect caused by the initial coarse estimation deviation and grid mismatch.
[0047] The alternating update mechanism refers to establishing a closed-loop feedback relationship between far-field parameter refinement and near-field parameter refinement, so that each far-field refinement is constrained by the latest near-field residual, and each near-field refinement is constrained by the latest far-field residual, forming a bidirectional iterative optimization process for parameter estimation.
[0048] By continuously optimizing the far-field incident angle parameter, jointly optimizing the near-field incident angle parameter and the propagation distance parameter, and iteratively updating them alternately under residual constraints, effective compensation for the inherent grid mismatch error of the discrete dictionary is achieved. On this basis, by leveraging the constraint effect of the far-field refined residual on the near-field optimization, and the feedback effect of the near-field refined residual on the far-field re-optimization, the residual coupling phenomenon caused by the incomplete orthogonality of the model between the far-field path and the near-field path during the sparse recovery process is suppressed. Thus, in the mixed far-field and near-field propagation environment, the overall estimation accuracy and robustness of the channel angle parameter and distance parameter in the XL-MIMO system are significantly improved.
[0049] S5. Merge the far-field and near-field refinement results, and output the hybrid field channel estimation matrix. .
[0050] The far-field channel estimation results refined by continuous domain gradient descent are superimposed and merged with the near-field channel estimation results to obtain the hybrid field channel estimation results corresponding to a single antenna subarray.
[0051] The hybrid field channel estimation results of all subarrays are then stacked according to the order of antenna subarray division to finally generate and output a complete global hybrid field channel estimation matrix. .
[0052]
[0053] By superimposing the far-field and near-field channel components refined by gradient descent in the complex domain at the subarray level, the independence of their respective propagation mechanism modeling and the accuracy of parameter optimization are preserved. On this basis, by leveraging the subarray sequential stacking mechanism of the antenna physical layout, the local estimation results are losslessly mapped to the global channel matrix space, realizing the closed-loop construction from parameter estimation to the system's available channel output.
[0054] This invention presents a staged collaborative estimation method for mixed-field XL-MIMO channels. Based on a channel path recovery and refinement mechanism, it combines SOMP signal recovery and gradient descent optimization (GD). By alternately optimizing far-field and near-field channel paths, it further improves the accuracy of channel estimation. After each SOMP recovery, gradient descent is used to finely optimize the angle and distance parameters of the path, effectively solving the problems of grid mismatch error and insufficient path identification accuracy in traditional methods. Example 2
[0055] This invention applies the hybrid near-field and far-field XL-MIMO channel phased cooperative estimation method in Example 1 to a typical 6G XL-MIMO hybrid field communication scenario. It adopts the channel estimation process by setting system parameters and completes channel estimation and performance testing under specific simulation parameters.
[0056] The specific steps of channel estimation are as follows: S1. Construct a multi-subcarrier joint signal observation model, perform subarray partitioning and parameter initialization; System parameter settings: Number of base station receiving antennas: N=256; Number of users: K=3; Number of far-field paths: Similarly, the number of near-field paths ; Radio frequency chain number Carrier frequency: ,wavelength meters, antenna spacing Meters. The signal-to-noise ratio is fixed at [value missing]. The maximum number of iterations is set to .
[0057] Based on the uplink TDD XL-MIMO system, a multi-subcarrier joint signal observation model is constructed. 256 antennas are arranged into a uniform linear array and divided into several subarrays. The received signal and pilot matrix are extracted on a subarray basis. A single-slot, single-carrier reception model is established and multi-subcarrier signal stacking is completed to form an equivalent matrix observation model, which provides a standard input structure for subsequent staged estimation.
[0058] The received signal is modeled as follows: , Let m be the received signal matrix of the m-th subcarrier. For pilot matrix, For channel path vectors, This is the noise vector.
[0059] S2. The channel is decomposed into far-field plane wave components and near-field spherical wave components. The mixed-field channel is represented as the superposition of far-field and near-field channel matrices. A far-field angle domain dictionary, which is only related to angle, is constructed based on the plane wave model; a near-field polar coordinate domain dictionary, which is related to angle and distance, is constructed based on the spherical wave model, thus achieving decoupled characterization of far-field and near-field physical propagation characteristics.
[0060]
[0061] in, This indicates the multiple channel response received by the base station. For the far-field channel response, This represents the near-field channel response.
[0062] Far-field partial channel response : By the It consists of several far-field paths, and the gain of each path is determined by... This indicates that the response of the path is... It means that, among them, The incident angle of the far-field path. It is the gain coefficient of the far-field signal, representing the propagation gain of the far-field path at a given angle.
[0063] Near-field partial channel response : By the It consists of several near-field paths, and the gain of each path is determined by... This indicates that the response of the path is... and Joint description. f is the propagation distance of the first near-field path, f is the subcarrier frequency, and c is the speed of light. It is the gain coefficient of the near-field path, representing the propagation gain of the near-field path at a given distance, angle, and frequency.
[0064] S3, Phased Cooperative Sparse Estimation: Initialization parameters: Initialize the channel matrix Set the size of the subarray: , Initialize far-field and near-field refinement results: , .
[0065] Phase 1: Using the subarray received signal as input, Synchronous Orthogonal Matched Pursuit (SOMP) is employed to perform joint sparse support recovery on the far-field dictionary, obtaining the initial far-field path angle and gain.
[0066] in, Store the estimated gain matrix for the far-field path. Complex gain of a far-field path; Store the angle support set for the far-field path estimation. The incident angle of the far-field path; The subarray receives the signal matrix, which is the input observation data for the algorithm; This is the pilot matrix of the subarray, used for projection from the channel domain to the pilot domain; The far-field dictionary matrix of the subarray is composed of far-field angular domain guiding vectors; is the number of far-field paths, i.e., far-field sparsity, and is the sparsity constraint parameter of the algorithm.
[0067] The path and coefficients of the far-field signal are recovered using SOMP. The reconstructed far-field signal is then separated from the observations to obtain the updated residuals.
[0068] Second stage: using the residual signal after far-field stripping As input, the near-field signal is estimated:
[0069] in, Store the estimated gain matrix for the near-field path. Complex gain of a near-field path; Store the estimated angle-distance support set for the near-field path. The propagation distance and incident angle of each near-field path; The residual signal matrix after far-field stripping of the subarray is the input observation data for the algorithm; This is the pilot matrix of the subarray, used for projection from the channel domain to the pilot domain; The near-field dictionary matrix of the subarray is composed of two-dimensional guide vectors of near-field distance and angle. Here, represents the number of near-field paths, i.e., near-field sparsity, and represents the sparsity constraint parameter of the algorithm.
[0070] The SOMP algorithm is used to perform sparse recovery on the near-field polar coordinate dictionary, and the initial angle and distance parameters of the near-field path are directly output.
[0071] S4. Gradient descent (GD) is used to perform continuous domain optimization of the far-field angle, near-field angle, and distance: Far-field refinement: The far-field is refined using the near-field estimation residuals. Gradient descent is used to optimize the angle parameters of the far-field channel.
[0072]
[0073] in, This is the result of far-field channel refinement, including the corrected far-field angle and gain; For fine-tuning the far-field residuals after removing the near-field components; This is the far-field observation matrix for the subarray; The result of the far-field channel rough estimation is used as the initial iteration point for gradient descent.
[0074] Near-field refinement: The near-field refinement residual is used to refine the near-field, and gradient descent is also used to optimize the angle and distance parameters of the near-field channel.
[0075]
[0076] in, This is the result of near-field channel refinement, including the corrected near-field range, angle, and gain; For near-field refinement residuals after stripping far-field components; This is the near-field observation matrix for the subarray; The result of the coarse estimation of the near-field channel is used as the initial iteration point for gradient descent.
[0077] By alternating iterative updates, near-field and far-field residual coupling is suppressed, and discrete dictionary grid mismatch error is eliminated.
[0078] S5. Superimpose the subarray-level far-field and near-field refinement results, then stack all subarray estimation results in the order of antenna subarrays to finally output the global hybrid field channel estimation matrix. .
[0079]
[0080] The channel estimation results in this embodiment are compared with the performance of other traditional estimation algorithms. The test results are as follows: Figures 2-4 As shown.
[0081] Depend on Figure 2 The comparison curves of the normalized mean square error (NMSE) performance of the method of this invention and traditional algorithms under different pilot lengths show that, as the number of pilots gradually increases from 4 to 64, the NMSE of the method of this invention is consistently lower than that of the least squares method, the mixed-field SGP algorithm, the control far-field component OMP algorithm, and the traditional enhanced SOMP algorithm. With the increase of the number of pilots, the estimation error of each algorithm decreases, but the method of this invention decreases faster and achieves higher accuracy. This indicates that the method of this invention can achieve better channel estimation results under different pilot overheads, especially with more significant advantages at medium to high pilot lengths.
[0082] Depend on Figure 3 The NMSE performance comparison curves of the proposed method and traditional algorithms under different signal-to-noise ratio (SNR) conditions show that, within the SNR range of 0 dB to 15 dB, the proposed method maintains the best estimation performance under low, medium, and high SNR conditions, with its NMSE curve significantly lower than other comparative algorithms. Even in low SNR environments, this method can effectively suppress noise interference, maintain stable channel recovery capability, and exhibit stronger robustness and environmental adaptability.
[0083] Achievable rate performance refers to the maximum data transmission rate that a communication system can reach under certain conditions. Figure 4 As shown in the performance comparison chart of the achievable rate of the method of this invention and traditional algorithms, the achievable rate of the method of this invention is significantly higher than that of traditional least squares, mixed-field SGP, controlled far-field component OMP, and enhanced SOMP methods, and it is closer to the ideal performance curve of perfect channel state information (CSI). The achievable rate directly reflects the maximum data transmission capacity of the communication system. This result proves that the channel estimation results obtained by this method are more accurate and have less distortion, which can significantly improve the transmission efficiency and spectrum utilization of XL-MIMO systems.
[0084] The hybrid near-field and far-field XL-MIMO channel staged collaborative estimation method of this invention was verified and compared through standardized simulation experiments in a typical 6G hybrid XL-MIMO simulation scenario. The comparison results show that the overall architecture of this invention, which includes far-field and near-field domain dictionary modeling, far-field priority sparse recovery, residual-driven near-field estimation, and gradient descent continuous domain alternating refinement, can effectively solve the problems of model mismatch, residual coupling, parameter grid mismatch, and excessive computational complexity of large-scale arrays in traditional methods.
[0085] Furthermore, under different pilot lengths, signal-to-noise ratios, and multipath hybrid propagation conditions, this method achieves lower estimation errors and higher system achievable rates, comprehensively outperforming existing mainstream channel estimation algorithms in terms of estimation accuracy, anti-interference capability, and algorithm stability. Simultaneously, the introduction of subarray partitioning and vectorized gradient calculation enables the algorithm to effectively control computational complexity and storage overhead while maintaining high accuracy, making it suitable for ultra-large-scale array deployments with hundreds of antennas.
[0086] In summary, the XL-MIMO channel phased cooperative estimation method proposed in this invention for hybrid near-field and far-field propagation has comprehensive advantages of high accuracy, low mismatch, high robustness, and low complexity in hybrid field environments. It can effectively support high-precision channel sensing and efficient data transmission in 6G ultra-large-scale antenna systems and has good application prospects.
[0087] The above are preferred embodiments of the present invention. It should be noted that those skilled in the art can make various equivalent modifications or substitutions without departing from the principle of the present invention. These possible changes and substitutions also fall within the scope of protection covered by the present invention.
Claims
1. A staged cooperative estimation method for XL-MIMO channels in hybrid near-field and far-field propagation, characterized in that, The steps include the following: S1. Construct a multi-subcarrier joint signal observation model, and perform subarray partitioning and parameter initialization; S2. Hybrid field channel modeling and independent domain modeling of far-field and near-field dictionaries: The hybrid field channel model includes a far-field plane wave propagation model and a near-field spherical wave propagation model. A far-field angle domain dictionary is established based on the far-field plane wave propagation model, and a near-field polar coordinate domain dictionary with joint representation of angle and distance is established based on the near-field spherical wave propagation model, so as to realize independent domain modeling of far-field and near-field components. S3, Phased Collaborative Estimation: The first stage involves performing joint sparse support recovery on the far-field angle domain dictionary to obtain the initial parameter estimates of the far-field path, and then extracting the reconstructed signal from the observation data to obtain the updated residual signal. The second stage involves using the updated residuals to perform sparse recovery on the near-field dictionary to obtain the initial angle and distance parameters of the near-field path. S4. Continuous parameter refinement: Gradient descent algorithm is used to refine the far-field angle parameters, near-field angle parameters and distance parameters in the continuous domain to reduce the grid mismatch error caused by the discrete dictionary, and the residual coupling between far-field and near-field paths is further suppressed by the alternating update mechanism. S5. Merge the far-field and near-field refinement results, and output the hybrid field channel estimation matrix. .
2. The staged cooperative estimation method for XL-MIMO channels oriented towards hybrid near-field and far-field propagation according to claim 1, characterized in that, In step S1, the construction of a multi-subcarrier joint signal observation model, including subarray partitioning and parameter initialization, specifically includes: Base station configuration for uplink time-division duplex XL-MIMO systems The antennas form a uniform linear array, with the antenna spacing equal to the carrier wavelength λ; the number of antennas for the user equipment is... ; Base station The antenna is uniformly divided into several subarrays, and signal processing is performed on a subarray basis. A single-time-slot signal observation model and a single-carrier received signal model are established in sequence, and all subcarrier signals are stacked together to form an equivalent matrix observation model of the joint received signal matrix. The joint received signal matrix is composed of the superposition of the pilot matrix, the multi-subcarrier channel matrix and the noise matrix.
3. The staged cooperative estimation method for XL-MIMO channels oriented towards hybrid near-field and far-field propagation according to claim 1, characterized in that, In step S2, the hybrid field channel modeling specifically includes: The wireless propagation channel is decomposed into far-field path components and near-field path components. The far-field path consists of plane wave signals with a propagation distance greater than the array aperture, and the array response is determined only by the incident angle and is independent of the propagation distance. The near-field path consists of spherical wave signals with a propagation distance less than the Rayleigh distance, and the array response is determined by both the incident angle and the propagation distance. Based on the above component decomposition, the complex channel response received by the base station is composed of the superposition of far-field components and near-field components. Specifically, the far-field component is obtained by superimposing the path gain and angle response of multiple far-field paths, and the near-field component is obtained by superimposing the path gain, range phase term and angle-range joint response of multiple near-field paths. The resulting hybrid channel is the superposition of the far-field channel matrix and the near-field channel matrix.
4. The staged cooperative estimation method for XL-MIMO channels oriented towards hybrid near-field and far-field propagation according to claim 3, characterized in that, In step S2, the independent domain modeling of the far-field and near-field dictionaries specifically includes: The far-field angle domain dictionary is generated based on the far-field plane wave propagation model. The far-field manifold is determined by the incident angle, antenna index, antenna spacing, signal wavelength, and total number of antennas. The channel response in each direction is calculated using a discrete angle sequence to form the far-field dictionary. ; The near-field polar coordinate domain dictionary is generated based on the near-field spherical wave propagation model. The near-field manifold is determined by the propagation distance, incident angle, subcarrier frequency, and speed of light. The channel response for each path is calculated using two-dimensional sequences of discrete distance and discrete angle to construct the near-field dictionary. .
5. The staged cooperative estimation method for XL-MIMO channels oriented towards hybrid near-field and far-field propagation according to claim 1, characterized in that, In step S3, during the phased collaborative estimation: The first stage of far-field channel estimation is as follows: taking the subarray received signal as input, a coarse far-field channel estimation is performed using the synchronous orthogonal matched pursuit algorithm on the far-field angle domain dictionary. Support set selection and path gain estimation are performed through the far-field dictionary to obtain the initial parameters of the far-field path. The far-field subarray signal is reconstructed based on the initial far-field parameters and the far-field dictionary. The reconstructed far-field subarray signal is then subtracted from the subarray received signal to obtain the updated residual signal.
6. The staged cooperative estimation method for XL-MIMO channels oriented towards hybrid near-field and far-field propagation according to claim 5, characterized in that, The second phase of near-field sparse recovery specifically involves: Using the updated residual signal obtained in the first stage as input, a coarse estimation of the near-field channel is performed using the synchronous orthogonal matching pursuit algorithm on the near-field polar coordinate domain dictionary. The support set selection and path gain estimation are completed through the near-field dictionary, and the initial incident angle parameters and initial propagation distance parameters corresponding to the near-field path are directly obtained.
7. The staged cooperative estimation method for XL-MIMO channels oriented towards hybrid near-field and far-field propagation according to claim 1, characterized in that, In step S4, the gradient descent algorithm is used to perform alternating far-field and near-field optimization and refinement, specifically including: First, using the residual signal obtained from the near-field channel estimation as a constraint, gradient descent optimization is performed on the incident angle parameter of the far-field channel to complete the far-field parameter refinement; then, using the refined far-field residual signal as a constraint, gradient descent optimization is performed on the incident angle parameter and propagation distance parameter of the near-field channel to complete the near-field parameter refinement. The alternating update mechanism continuously reduces the parameter grid mismatch error introduced by the discrete dictionary grid partitioning through cyclic iteration of far-field refinement and near-field refinement, and further suppresses the residual coupling between the far-field channel path and the near-field channel path, thereby improving the estimation accuracy of angle parameters and distance parameters.
8. The staged cooperative estimation method for XL-MIMO channels oriented towards hybrid near-field and far-field propagation according to claim 7, characterized in that, In step S5, the far-field and near-field refinement results are merged, and the mixed-field channel estimation matrix is output. Specifically, it includes: The far-field channel estimation results refined by continuous domain gradient descent are superimposed and merged with the near-field channel estimation results to obtain the hybrid field channel estimation results corresponding to a single antenna subarray level. The hybrid field channel estimation results of all subarrays are then stacked according to the order of antenna subarray division to finally generate and output a complete global hybrid field channel estimation matrix. .