Channel feedback method and system

Abstract

The invention relates to a channel feedback method and system, which is applied to a user side, and comprises the following steps: establishing a multi-cluster channel response matrix of a downlink channel, and carrying out decoupling processing on the multi-cluster channel response matrix to obtain a plurality of single-cluster responses; wherein each single cluster response comprises a plurality of path responses; for each single-cluster response, acquiring a peak position index in the single-cluster response, determining a target path response in the single-cluster response according to the peak position index, and performing phase compensation on all path responses in the single-cluster response according to the phase of the target path response to obtain a target single-cluster response corresponding to the single-cluster response; and obtaining an encoding processing result of the target single-cluster response corresponding to each single-cluster response, and sending each encoding processing result to the base station end to instruct the base station end to decode each encoding processing result to obtain a channel feedback result corresponding to the multi-cluster channel response matrix. According to the invention, the generalization of the channel feedback technology in different application scenes can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of channel feedback, in particular to a channel feedback method and system. BACKGROUND

[0002] In a Frequency Division Duplexing (FDD) system, due to the difference between uplink and downlink frequency bands, channel reciprocity does not hold, and the downlink channel state information needs to be obtained through the process of downlink pilot estimation and channel feedback. However, in a large-scale Multiple Input Multiple Output (MIMO) scenario, with the rapid increase in the number of antennas, the overhead of channel feedback is huge.

[0003] In related technologies, the channel feedback technology based on deep learning can fully exploit the correlation of the channel in different domains, thereby significantly improving the accuracy of channel feedback and reducing the overhead of channel feedback. However, the channel feedback technology based on deep learning has the disadvantage of poor generalization, because the base station position, user distribution, object layout, and electromagnetic parameters in different environments are significantly different, causing the Channel State Information Distribution (CSI) in different environments to shift significantly. Therefore, when the trained deep learning model is deployed to a new environment, it will not be able to accurately feedback a large number of out-of-distribution channel matrices, making it difficult to generalize to new environments.

[0004] Therefore, how to improve the generalization of the channel feedback technology has become a technical problem to be solved. SUMMARY

[0005] Therefore, it is necessary to provide a channel feedback method and system to solve the above technical problems.

[0006] In a first aspect, the present application provides a channel feedback method applied to a user end, comprising:

[0007] A multi-cluster channel response matrix of a downlink channel is established, and the multi-cluster channel response matrix is decoupled to obtain a plurality of single-cluster responses; wherein each single-cluster response includes a plurality of path responses;

[0008] For each single-cluster response, the peak position index in the single-cluster response is obtained, and the target path response in the single-cluster response is determined according to the peak position index, and all path responses in the single-cluster response are phase compensated according to the phase of the target path response to obtain the target single-cluster response corresponding to the single-cluster response;

[0009] The encoding processing result of each single-cluster response corresponding to the target single-cluster response is obtained, and each encoding processing result is sent to the base station end to indicate the base station end to decode each encoding processing result to obtain the channel feedback result corresponding to the multi-cluster channel response matrix.

[0010] In one of the embodiments, the multi-cluster channel response matrix is decoupled to obtain a plurality of single-cluster responses, including:

[0011] The multi-cluster channel response matrix is singular value decomposed to obtain all initial single-cluster response channels contained in the multi-cluster channel response matrix;

[0012] The number of clusters corresponding to the multi-cluster channel response matrix is determined;

[0013] Based on the size of the singular value of each initial single-cluster response channel, the singular values are arranged in descending order to obtain a descending sequence, and the singular values ranked in the front of the descending sequence are filtered based on the number of clusters to obtain a plurality of target singular values, the corresponding single-cluster response channel is determined based on the target singular value, and the corresponding single-cluster response is obtained based on the single-cluster response channel.

[0014] In one of the embodiments, the number of clusters corresponding to the multi-cluster channel response matrix is determined, including:

[0015] The first cluster number is calculated according to the multi-cluster channel response matrix and the singular value decomposition result;

[0016] The second cluster number is calculated according to the multi-cluster channel response matrix, the singular value decomposition result and the preset energy threshold;

[0017] The cluster number with the minimum value between the first cluster number and the second cluster number is taken as the cluster number corresponding to the multi-cluster channel response matrix.

[0018] In one of the embodiments, the phase of all path responses in the single-cluster response is compensated according to the phase of the target path response to obtain the target single-cluster response corresponding to the single-cluster response, including:

[0019] The phase of the target path response is uniformly quantized to obtain a phase quantization result;

[0020] The phase of all path responses in the single-cluster response is rotated based on the phase quantization result to obtain the target single-cluster response.

[0021] In one of the embodiments, the peak position index in the single-cluster response is retrieved, including:

[0022] For each single-cluster response, the angle peak position with the strongest energy in the single-cluster response is determined based on a preset angle domain oversampling discrete Fourier transform codebook.

[0023] determine a strongest energy time-delay domain peak position in the single-cluster response based on a preset time-delay domain oversampling discrete Fourier transform codebook;

[0024] determine a peak position index based on the angle peak position and the time-delay domain peak position.

[0025] In one of the embodiments, the encoding processing results of the target single-cluster responses corresponding to the single-cluster responses are obtained, and the encoding processing results are sent to the base station end to instruct the base station end to decode the encoding processing results to obtain the channel feedback result corresponding to the multi-cluster channel response matrix, including:

[0026] The single-cluster responses are respectively encoded by the encoder in the user end to obtain the encoding processing results of the target single-cluster responses, and the encoding processing results are sent to the base station end to instruct the base station end to decode the encoding processing results to obtain the channel feedback result corresponding to the multi-cluster channel matrix.

[0027] In one of the embodiments, the method further includes:

[0028] An initial signal processing network composed of an initial encoder and an initial decoder is established, and a preset training set is obtained; the training set includes a plurality of training single-cluster responses with phase compensation on path responses;

[0029] The training set is input into the initial signal processing network, the feature representation of the training set is extracted by the initial encoder, and the feature representation is reconstructed by the initial decoder;

[0030] The reconstruction loss between the reconstruction result and the training set is calculated based on a preset loss function, and the parameters of the initial encoder and the initial decoder are iteratively calculated by back propagation until the initial signal processing network converges, to obtain a trained signal processing network including a trained complete encoder and a trained complete decoder.

[0031] In a second aspect, the application further provides a channel feedback method applied to a base station end connected with a user end, the method further includes:

[0032] receive the encoding processing results transmitted from the user end, wherein the encoding processing results are obtained by encoding the target single-cluster responses corresponding to the single-cluster responses by the user end, the target single-cluster responses are obtained by phase compensating all path responses in the single-cluster responses according to the phase of the target path response, for each single-cluster response, the target path response is determined according to the peak position index in the single-cluster response, and the single-cluster response is obtained by decoupling processing the multi-cluster channel response matrix of the downlink channel, and each single-cluster response includes a plurality of path responses;

[0033] The decoding result is decoded by a preset decoder at the base station end to obtain a decoded single-cluster response corresponding to each target single-cluster response, and the feedback result corresponding to the multi-cluster channel response matrix is obtained by superposition according to the decoded single-cluster response.

[0034] In one embodiment, the decoding result is decoded by a preset decoder at the base station end to obtain a decoded single-cluster response corresponding to each target single-cluster response, including:

[0035] The decoding result is decoded by a preset decoder at the base station end to obtain a decoded single-cluster response corresponding to each target single-cluster response, and the feedback result corresponding to the multi-cluster channel response matrix is obtained by superposition according to the decoded single-cluster response.

[0036] The peak position index of each single-cluster response and the phase of the target path response transmitted by the user end are obtained, and the corresponding single-cluster response is recovered and calculated based on the peak position index and the phase of the target path response to obtain the decoded single-cluster response corresponding to each single-cluster response after recovery.

[0037] In a third aspect, the application also provides a channel feedback system, including a user end and a base station end, and the user end and the base station end are in communication connection;

[0038] The user end is used to execute the channel feedback method applied to the user end as described above.

[0039] The base station end is used to execute the channel feedback method applied to the base station end as described above.

[0040] The above-mentioned channel feedback method and system are applied to the user end, first, the multi-cluster channel response matrix of the downlink channel is established, and the multi-cluster channel response matrix is decoupled to obtain a plurality of single-cluster responses, and each single-cluster response includes a plurality of path responses; then for each single-cluster response, the peak position index in the single-cluster response is obtained, and the target path response in the single-cluster response is determined according to the peak position index, the phase compensation of all path responses in the single-cluster response is performed according to the phase of the target path response to obtain the target single-cluster response corresponding to the single-cluster response; finally, the encoding processing result of the target single-cluster response corresponding to each single-cluster response is obtained, and each encoding processing result is sent to the base station end to instruct the base station end to decode each encoding processing result to obtain the channel feedback result corresponding to the multi-cluster channel response matrix. In the embodiment of the application, the efficient alignment of the cluster response in different environments is realized through the multi-cluster decoupling and phase alignment scheme, which has strong physical interpretability, can overcome the distribution offset of the cluster response in different environments, and thus improves the generalization of the channel feedback scheme of the application in different application scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the present application or the related art. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other related drawings can also be obtained on the basis of these drawings without any creative effort.

[0042] Figure 1 Flowchart of a channel feedback method applied to a user end in an embodiment;

[0043] Figure 2 Flowchart of a single-cluster response calculation in an embodiment;

[0044] Figure 3 Flowchart of a cluster number determination in an embodiment;

[0045] Figure 4 Flowchart of a target single-cluster response calculation in an embodiment;

[0046] Figure 5 Flowchart of a peak position index retrieval in an embodiment;

[0047] Figure 6 Flowchart of a signal processing network training in an embodiment;

[0048] Figure 7 Flowchart of a channel feedback method applied to a base station end in an embodiment;

[0049] Figure 8 Flowchart of a channel feedback method in a preferred embodiment;

[0050] Figure 9 Structure block diagram of a channel feedback system in an embodiment;

[0051] Figure 10 Structure block diagram of a channel feedback device applied to a user end in an embodiment;

[0052] Figure 11 Structure block diagram of a channel feedback device applied to a base station end in an embodiment. DETAILED DESCRIPTION

[0053] In order to make the purposes, technical solutions and advantages of the present application clearer, the following will further describe the present application with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0054] This section first introduces downlink channel acquisition techniques in FDD massive MIMO systems. Since uplink and downlink channels in different frequency bands are not reciprocal in FDD systems, existing FDD massive MIMO systems employ a downlink channel estimation plus feedback process to acquire the downlink channel. Because the channel matrix dimension in a massive MIMO system is proportional to the number of antennas, the overhead of directly performing feedback quantization on each element of the downlink channel matrix estimate is proportional to the number of antennas. To reduce downlink channel feedback overhead, existing FDD massive MIMO systems typically utilize the sparsity of the Channel State Information matrix (CSI) and employ compressed sensing and CSI codebooks to reduce channel feedback overhead. In compressed sensing-based CSI feedback, the channel feedback problem is modeled as a sparse recovery problem, and feedback channel reconstruction can be achieved using the Approximate Message Passing (AMP) algorithm and the Orthogonal Matching Pursuit (OMP) algorithm. Specifically, the 3GPP standardization organization has conducted in-depth research on channel feedback technology based on 5G NR CSI codebooks, evolving through multiple codebook versions such as R15 Type-I, R15 Type-II, and R16 eType-II. In the R15 Type-I CSI codebook, the base station transmits orthogonal beams in the downlink channel for scanning, and the user terminal selects the strongest beam direction based on the received signal strength and feeds back the beam index to the base station. Since the R15 Type-I codebook only uses the optimal beam index, it has extremely low feedback overhead. However, R15 Type-I can only reflect the strongest beam direction information and cannot accurately characterize channel information under complex multipath propagation environments. Therefore, 3GPP proposed the R15 Type-II codebook to accurately characterize channel information under complex multipath propagation conditions. Specifically, the base station first transmits CSI-RS pilot signals through the downlink channel, and the user terminal uses the received signal to complete channel estimation. Subsequently, the user selects several orthogonal beams with relatively large amplitudes from the estimated channel, subband by subband, and feeds back the corresponding beam indices, quantization amplitudes, and phases to the base station, thereby more accurately characterizing the downlink CSI. However, the feedback overhead of the R15 Type-II codebook is much greater than that of the R15 Type-I codebook, resulting in lower feedback efficiency. Furthermore, the 3GPP standardization organization proposed the R16 eType-II codebook to further reduce feedback overhead while accurately characterizing CSI under complex multipath propagation conditions. Specifically, the R16 eType-II codebook further introduces frequency domain compression technology to fully exploit the sparsity of the downlink channel in the angular delay domain, thereby reducing feedback overhead.

[0055] However, existing R16 eType-II codebooks suffer from limitations in feedback accuracy and overhead. This is because current R16 eType-II codebooks still employ a port-by-port feedback method based on angle delay, failing to achieve joint feedback across ports with different angle delays. In wireless channels, a small number of propagation paths / clusters consume the majority of the channel matrix's energy. Angle delay domain transformations in eType-II codebooks typically employ 2D DFT transformations. Due to the power leakage effect caused by 2D DFT transformations, power from the same path will leak into different angle delay ports. Therefore, different angle delay ports are not independent but exhibit complex correlations. However, channel feedback based on eType-II codebooks cannot fully utilize the correlations between different angle delay ports, thus reducing channel feedback accuracy and increasing channel feedback overhead.

[0056] In recent years, deep learning-based channel feedback techniques have received widespread attention from academia and industry. Compared with traditional feedback methods based on compressed sensing and CSI codebooks, deep learning-based channel feedback techniques can fully exploit the correlation of the channel across different domains, thereby significantly improving channel feedback accuracy and reducing channel feedback overhead. Existing deep learning-based channel feedback techniques typically employ an autoencoder neural network structure, where the encoder and decoder networks are deployed at the user end and base station, respectively. Specifically, the downlink angular delay domain CSI matrix is ​​first input to the user-end encoder network to extract features from the CSI matrix and compress it into a low-dimensional vector. After quantization, the low-dimensional channel vector is fed back to the base station through the uplink channel. The feedback is then input to the base station decoder network to reconstruct the downlink CSI matrix. To fully exploit channel features, researchers have proposed different encoder and decoder network structures. Inspired by the local correlation of the angle-delay domain CSI matrix, researchers first proposed CsiNet and CsiNet+, autoencoder structures based on the Convolutional Neural Network (CNN) architecture, to mine the angle-delay domain features of the CSI matrix. Furthermore, an attention mechanism was introduced into the design of the CSI feedback network, and a TransNet feedback network based on the Transformer architecture was proposed.

[0057] However, existing deep learning-based channel feedback techniques suffer from poor generalization, thus limiting the large-scale deployment of deep learning models. Specifically, existing deep learning-based channel feedback typically relies on the same distribution assumption, meaning that the channel data in the training and test sets satisfy the same probability distribution. Due to the diversity of wireless propagation environments, model deployment may not satisfy the same distribution condition, limiting the model's performance in new environments. Based on large-scale MIMO channel models, the CSI matrix is ​​directly determined by multipath / cluster propagation in the channel. Since base station locations, user distributions, object layouts, and electromagnetic parameters differ significantly in different environments, the CSI distribution shifts significantly. Therefore, when a trained model is directly deployed to a new environment, it cannot accurately respond to a large number of out-of-distribution (OOD) channel matrices, making it difficult to generalize to new environments. To enhance the performance of deep learning-based channel feedback in new environments, researchers have proposed various model adaptation techniques, including transfer learning, meta-learning, and plug-in module designs for environment adaptation. However, all model adaptation techniques require collecting channel data in new environments, and their large-scale deployment overhead remains significant. To directly increase the generalization ability of pre-trained models and reduce deployment overhead, researchers have proposed data fusion and UniversalNet / UniversalNet+ learning frameworks. In the data fusion learning framework, the model is pre-trained directly on a diverse channel dataset that mixes multiple environments to enhance its generalization ability. However, the generalization ability of the data fusion learning framework is directly limited by the diversity of the pre-training dataset. In summary, among related technologies, deep learning-based channel feedback techniques typically suffer from poor generalization and high model deployment overhead.

[0058] In one exemplary embodiment, such as Figure 1 As shown, a channel feedback method is provided, including:

[0059] Step S110: Establish a multi-cluster channel response matrix for the downlink channel and decouple the multi-cluster channel response matrix to obtain multiple single-cluster responses; wherein each single-cluster response includes multiple path responses.

[0060] The technical solution in this application is considered for application in a typical FDD massive MIMO system, where the number of base station antennas is N. T The number of subcarriers is N c Based on the multi-cluster channel model, it is assumed that there are N clusters in the channel. cl There are N clusters, where the l-th cluster includes N p，l There is one path, and at this time, the channel h on the k-th subcarrier... k ∈ It can be represented as:

[0061]

[0062] in , , These represent the complex gain, delay, and departure angle corresponding to the i-th path in the l-th cluster, respectively. Indicates the subcarrier spacing of the system. This represents the system's guidance vector. Therefore, for a channel matrix H = [h1, h2, ..., hNc] ∈ [...], the system needs to consider multiple subcarriers. It can be represented as:

[0063]

[0064] in, This represents the response of the l-th cluster. This represents the frequency domain response vector.

[0065] In summary, the H matrix described above is the multi-cluster channel response matrix of the downlink channel as described in the embodiments of this application.

[0066] Furthermore, in this embodiment, considering the distribution offset of multi-cluster channel response matrices in different environments, this offset includes multi-cluster structure distribution offset and single-cluster response distribution offset, specifically:

[0067] The distribution offset of multi-cluster structures comprises two parts: the distribution offset of the number of clusters and the distribution offset of inter-cluster dependencies. Specifically, the number of clusters in the channel matrix is ​​related to the number of scatterers in the environment. For example, in an open scene, the number of scatterers is relatively small, resulting in a smaller number of clusters in the channel matrix. Conversely, in a scattering-rich environment, the increased number of scatterers in the channel corresponds to a larger number of clusters in the channel matrix. Furthermore, different clusters in the channel matrix are not independent but rather have complex dependencies. As the electromagnetic waves radiated by the base station propagate through the physical channel, they interact with objects in the environment, forming different clusters. Therefore, the dependencies between different clusters are influenced by the arrangement of objects in the environment.

[0068] In the angular delay domain, the distribution offset of a single cluster response mainly consists of three parts: peak position, path gain, and power leakage distribution offset. According to the propagation process of electromagnetic waves in space, the peak value of the single cluster response in the angular delay domain is primarily determined by the location of the interaction point during propagation. Simultaneously, the path gain is affected by the propagation distance and the electromagnetic properties of the interaction point material. According to the properties of the DFT transform, the power leakage of the single cluster response in the angular delay domain is determined by two parts: peak position misalignment and intra-cluster angular delay spread. Due to differences in base station location, user location distribution, and object location and material in different environments, the peak position, path gain, and power leakage distribution of the single cluster response in the angular delay domain will change under different environments, thus causing the distribution offset of the single cluster response.

[0069] Therefore, in order to overcome the channel distribution offset in different environments, this application can overcome the multi-cluster structure distribution offset and single-cluster response distribution offset of the channel in different environments through user-end multi-cluster decoupling and phase alignment. The purpose of multi-cluster decoupling is to decompose the multi-cluster channel into a single-cluster superposition form, i.e. , where C l This represents the response of the l-th cluster obtained after decoupling. This indicates the number of clusters obtained through decoupling. In this embodiment, decoupling the multi-cluster channel response matrix yields multiple single-cluster responses, and each single-cluster response contains multiple path responses.

[0070] Because the subsequent steps in this application will perform compression and decompression operations on each cluster, the multi-cluster channel response matrix in this embodiment is decoupled so that the model will not fit the dependency relationship between different clusters and the distribution of the number of clusters. Thus, the multi-cluster decoupling step can overcome the distribution offset of the multi-cluster structure in the CSI feedback mentioned above.

[0071] Step S120: For each single cluster response, obtain the peak position index in the single cluster response, determine the target path response in the single cluster response based on the peak position index, and perform phase compensation on all path responses in the single cluster response based on the phase of the target path response to obtain the target single cluster response corresponding to the single cluster response.

[0072] The peak location index of a single cluster response is the location index of the most energetic path among the multiple path responses included in the single cluster response. The location index of this most energetic path can be obtained using... To indicate, among which, This represents the peak index in the delay domain. This represents the peak index in the angle domain. The target path response described above is the path with the strongest energy in a single cluster response, based on the peak position index.

[0073] In this embodiment, for each cluster response, the target cluster response within that cluster response is determined. Specifically, this involves: first, retrieving the peak position index from the cluster response; then, determining the target path response based on the peak position index; and finally, determining the phase of the target path response. In practical applications, the phase of the target path response can be Q-factored. p -bit uniform quantization is performed to determine the phase of the quantized target path response. Finally, phase compensation is applied to all path responses in the single cluster response based on the calculated phase of the target path response to obtain the target single cluster response corresponding to the single cluster response. It is important to emphasize that phase compensation is performed independently on the path responses within each single cluster response, thus yielding the corresponding phase-compensated target single cluster response for each single cluster response.

[0074] In summary, for each single-cluster response aligned with the angular delay domain, its peak position index is fixed at (0, 0), thus overcoming the offset of the peak position index.

[0075] Step S130: Obtain the encoding processing result of the target single cluster response corresponding to each single cluster response, and send each encoding processing result to the base station to instruct the base station to decode each encoding processing result to obtain the channel feedback result corresponding to the multi-cluster channel response matrix.

[0076] In this embodiment, the target single-cluster response corresponding to each single-cluster response is encoded by a preset encoder at the user terminal to obtain the encoding result corresponding to each target single-cluster response. The encoding result is then sent to the base station. The base station decodes the encoding result based on a preset decoder, recovers and superimposes the decoded single-cluster signal to obtain the channel feedback result corresponding to the multi-cluster channel response matrix. This channel feedback result is the channel matrix reconstructed by the base station that corresponds to the multi-cluster channel response matrix mentioned above.

[0077] In this embodiment, the multi-cluster channel response matrix is ​​first decoupled to obtain multiple single-cluster responses. After multi-cluster decoupling, the encoding and decoding network compresses and decompresses each cluster separately in subsequent encoding and decoding processes. This prevents the model calculation from fitting the dependencies between different clusters and the distribution of cluster numbers. Therefore, this application directly overcomes the distribution offset of the multi-cluster structure in CSI feedback based on multi-cluster decoupling. Furthermore, for each single-cluster response, its target path response is determined, and phase compensation is performed on all path responses in that single-cluster response based on the phase of the target path response to obtain the target single-cluster response corresponding to each single-cluster response. This effectively overcomes the distribution offset of the single-cluster response. Finally, the encoding processing results of each target single-cluster response are sent to the base station to instruct the base station to decode each encoding processing result to obtain the channel feedback result corresponding to the multi-cluster channel response matrix. In summary, by overcoming the distribution shift of multi-cluster structures and the distribution shift of single-cluster responses, the influence between different environments and multi-cluster responses is compensated, thereby achieving robust channel feedback in different environments. When the deep learning network used for encoding and decoding is deployed to a new environment, no additional channel data needs to be collected for model fine-tuning, thus enhancing the generalization of the deep learning network.

[0078] In one exemplary embodiment, such as Figure 2 As shown, the multi-cluster channel response matrix is ​​decoupled to obtain multiple single-cluster responses, including:

[0079] Step S210: Perform singular value decomposition on the multi-cluster channel response matrix to obtain all the initial single-cluster response channels contained in the multi-cluster channel response matrix.

[0080] In this application, the channel properties at the cluster level are clearly defined, including the intra-cluster rank-one property and the inter-cluster biorthogonality property. Specifically, since different paths within a single cluster have similar angular delay parameters, especially when the intra-cluster angle / delay spread is less than the system resolution, the antenna domain guidance vectors corresponding to each path... and frequency domain response vector Highly linearly correlated. Mathematically, a single-cluster response can be represented as... Therefore, the response of a single cluster can be approximated as a rank-1 matrix, i.e. The aforementioned initial single-cluster response channels are the left and right singular vectors obtained after SVD decomposition.

[0081] The aforementioned inter-cluster bioorthogonality property refers to the fact that, based on actual large-scale MIMO channel measurement experiments, the direct path and single-hop reflection path in the channel account for the vast majority of the energy. According to the propagation process of electromagnetic waves in the physical channel, the angular delay parameters of each cluster are determined by the interaction points during propagation. In general, the layout of objects in the channel is asymmetric, and the cluster angles and delay parameters generated by their interactions differ significantly. Therefore, different cluster responses are approximately orthogonal to each other in both the angular and delay domains, exhibiting bioorthogonality.

[0082] In summary, considering the aforementioned intra-cluster rank-one property and inter-cluster bioorthogonality property, this application embodiment chooses to use singular value decomposition to decompose the multi-cluster channel response matrix to obtain the initial single-cluster response. This is because: based on the rank-one property and bioorthogonality property, the cluster decoupling problem can be modeled as an optimization problem, where the optimization objective is to minimize the decoupling error, and the constraints are the intra-cluster rank-one property and the inter-cluster bioorthogonality property. According to the Eckart-Young-Mirsky theorem of low-rank matrix decomposition, it can be proved that in this application, singular value decomposition is the optimal solution to this multi-cluster decoupling optimization problem.

[0083] In this embodiment of the application, singular value decomposition is performed on the multi-cluster channel response matrix, resulting in:

[0084]

[0085] The l-th initial single-cluster response channel obtained by decoupling can be expressed as: .

[0086] Step S220: Determine the number of clusters corresponding to the multi-cluster channel response matrix.

[0087] In this embodiment, the number of clusters is explicitly specified because user-end received noise will cause estimation errors in downlink channel estimation. Under orthogonal pilot and least squares channel estimation, the channel estimate H... （e） It can be represented as Where H represents the noiseless downlink channel truth value, and N (e) This represents channel estimation noise. To enhance the robustness of channel feedback under channel estimation errors, and also to reduce the overhead of channel feedback and improve the accuracy of channel feedback, the embodiments of this application estimate the number of clusters in the multi-cluster channel response to filter out single-cluster responses corrupted by noise, or to filter out single-cluster responses with weak power. The cluster number estimation method includes, but is not limited to, the minimum description length criterion, the threshold truncation criterion, etc.

[0088] Step S230: Based on the magnitude of the singular values ​​of each initial single-cluster response channel, the singular values ​​are sorted in descending order to obtain a descending sequence. The singular values ​​ranked first in the descending sequence are filtered based on the number of clusters to obtain multiple target singular values. The corresponding single-cluster response channel is determined based on the target singular values, and the corresponding single-cluster response is obtained based on the single-cluster response channel.

[0089] In the embodiments of this application, it can be understood that the multiple initial single-cluster response channels obtained based on singular value decomposition correspond to the cluster responses actually included in the physical layer multi-cluster channel response matrix. The initial single-cluster response channels provide the optimal transmission channels in the spatial domain, and can then be mapped to the path response characteristics of the physical layer by combining a preset propagation model (such as a geometric cluster model).

[0090] Therefore, in this application, the singular values ​​of the initial single-cluster response channel are sorted in descending order to obtain a descending sequence. The singular values ​​ranked first in the descending sequence are then filtered according to the pre-calculated number of clusters. That is, multiple singular values ​​with the largest values ​​are filtered based on the number of clusters to obtain multiple target singular values. The corresponding single-cluster response channel can then be determined based on the target singular values. Based on the single-cluster response channel and the preset propagation model, the corresponding single-cluster response is determined.

[0091] Compared to traditional cluster response extraction algorithms such as Space Alternating Generalized Expectation Maximization (SAGE) and DBSCAN (Density-Based Spatial Clustering of Applications with Noise), the multi-cluster decoupling algorithm based on singular value decomposition proposed in this application does not require path-level parameter estimation, thus significantly improving its computational speed. Furthermore, in practical applications, the optimality of the multi-cluster decoupling algorithm based on singular value decomposition in this application can be guaranteed by the Eckart-Young-Mirsky theorem.

[0092] In one exemplary embodiment, such as Figure 3 As shown, determining the number of clusters corresponding to the multi-cluster channel response matrix includes:

[0093] Step S310: Calculate the number of the first cluster based on the multi-cluster channel response matrix and the singular value decomposition results; calculate the number of the second cluster based on the multi-cluster channel response matrix, the singular value decomposition results, and the preset energy threshold.

[0094] In this application embodiment, to enhance the robustness of channel feedback under channel estimation errors, a noise-robust multi-cluster quantity estimation hybrid criterion is proposed. This criterion has two design objectives:

[0095] Objective 1: When the downlink channel estimation signal-to-noise ratio is low, the cluster response that is damaged by noise after decoupling needs to be directly filtered to ensure the accuracy of the channel feedback;

[0096] Objective 2: When the downlink channel estimation signal-to-noise ratio is high, the weaker cluster responses need to be directly filtered to balance feedback overhead and accuracy.

[0097] To achieve objective 1 above, the minimum description length (MDL) criterion can be used to estimate the number of clusters. Specifically, the channel estimate H mentioned above can be set as follows: （e） Perform SVD decomposition, decomposed into: ,in, This represents the i-th largest singular value. Therefore, under the above MDL criterion, the estimated number of multiple clusters is the number of the first cluster. have:

[0098]

[0099] Here, r is a temporary variable that iterates from 1 to min(N_T, N_C) to calculate the argmin(·) part in the above formula. After the calculation is complete, the r that minimizes the value of the argmin(·) part is selected as the value of the variable. The estimated value of N. And it is understandable that... C N T The values ​​of these parameters can be obtained through the multi-cluster channel response matrix.

[0100] Similarly, to achieve objective 2 above, a threshold truncation criterion can be used. First, the defined energy threshold... Then, based on the threshold truncation criterion, the number of the second cluster... It can be estimated as follows:

[0101]

[0102] Step S320: The cluster number with the smallest value among the first cluster number and the second cluster number is taken as the cluster number corresponding to the multi-cluster channel response matrix.

[0103] In this embodiment of the application, in the hybrid criteria The cluster number with the smallest value between the first and second cluster numbers is taken as the cluster number corresponding to the multi-cluster channel response matrix. This enables accurate decoupling of the multi-cluster components, thereby balancing feedback overhead and accuracy.

[0104] In one exemplary embodiment, such as Figure 4 As shown, phase compensation is performed on all path responses in a single cluster response based on the phase of the target path response to obtain the target single cluster response corresponding to the single cluster response, including:

[0105] Step S410: Perform parameter uniform quantization on the phase of the target path response to obtain the phase quantization result.

[0106] In this embodiment of the application, the peak gain in the angle delay domain is further calculated based on the obtained peak position index. ,in, The array response vector representing the peak index in the optimal angle domain. Let H represent the frequency domain response vector corresponding to the optimal delay domain peak index, H represent the Hermitian transpose, and C represent the cluster response matrix. Through fine-grained peak position scanning, the peak gain p can effectively reflect the gains of different paths within the cluster. Therefore, the phase of the peak value p of the target path response corresponding to the peak position index can be analyzed. Bit uniform quantization is obtained That is, the phase quantization result mentioned above.

[0107] Step S420: Based on the phase quantization results, rotate the phase of all path responses in the single-cluster response to obtain the target single-cluster response.

[0108] In this embodiment, the phase of all elements in C can be rotated by -β to overcome the distribution offset of the intra-cluster path gain. Based on the above processing, the target single-cluster response in the aligned angular delay domain can be obtained. ,have:

[0109]

[0110] Among them, F a Represents a normalized DFT (Discrete Fourier Transform) matrix of N_T×N_T dimensions. S represents a normalized IDFT (Inverse Discrete Fourier Transform) matrix of N_C × N_C dimensions, where N_T represents the number of antennas, N_C represents the number of subcarriers, and S is the phase adjustment matrix. And there are In summary, alignment of single-cluster responses can be achieved. In the aligned single-cluster response, the peak position index is fixed at (0,0), thus overcoming the offset of the peak position index.

[0111] In one exemplary embodiment, such as Figure 5As shown, retrieving the peak location index in a single cluster response includes:

[0112] Step S510: For each single cluster response, based on a preset angle domain oversampled discrete Fourier transform codebook, determine the position of the angle peak with the strongest energy in the single cluster response; based on a preset time delay domain oversampled discrete Fourier transform codebook, determine the position of the time delay domain peak with the strongest energy in the single cluster response.

[0113] In this embodiment of the application, in order to overcome the peak position and power leakage factors in the single-cluster response distribution offset, this embodiment of the application proposes a fine-grained peak position search algorithm based on the oversampled DFT codebook.

[0114] Based on the preset angle domain The nth codeword in the oversampled DFT codebook (i.e., the aforementioned preset angle-domain oversampled discrete Fourier transform codebook) The search yields the index of the peak position of the angle. for:

[0115]

[0116] Similarly, a delay domain can be defined. The m-th codeword in the oversampled DFT codebook (i.e., the aforementioned pre-defined time-delay domain oversampled Discrete Fourier Transform codebook) The search yields the peak position index in the time delay domain. for:

[0117]

[0118] Step S520: Determine the peak position index based on the angular peak position and the time delay domain peak position.

[0119] In summary, based on the aforementioned angular peak position and time delay domain peak position, the peak position index can be determined based on the scan. .

[0120] In an exemplary embodiment, the encoding processing result of the target single-cluster response corresponding to each single-cluster response is obtained, and each encoding processing result is sent to the base station to instruct the base station to decode each encoding processing result to obtain the channel feedback result corresponding to the multi-cluster channel response matrix, including:

[0121] The encoder at the user terminal encodes each cluster response separately to obtain the encoding result of each target cluster response, and sends the encoding result to the base station to instruct the base station to decode the encoding result and obtain the channel feedback result corresponding to the multi-cluster channel matrix.

[0122] In this embodiment, a preset encoder is provided at the user terminal. The encoder is used to encode each aligned single-cluster response separately to obtain the encoding result corresponding to each target single-cluster response, and then sends the encoding result to the base station. In practical applications, the encoder can use existing deep learning-based encoder networks, such as CNN (Convolutional Neural Networks) or transformers.

[0123] In this embodiment of the application, the base station is equipped with a decoder corresponding to the encoder. The decoder is used to decode the encoding result and restore it to the channel feedback result corresponding to the multi-cluster channel response matrix.

[0124] In one exemplary embodiment, such as Figure 6 As shown, the above method also includes:

[0125] Step S610: Establish an initial signal processing network consisting of an initial encoder and an initial decoder, and obtain a preset training set; wherein, the training set includes multiple training single-cluster responses for path responses that have undergone phase compensation.

[0126] The initial signal processing network consists of an initial encoder and an initial decoder. Furthermore, it is necessary to perform multi-cluster decoupling and phase alignment on all channel matrices H in the original training set to obtain the aligned training set. That is, the training set consists of training single-cluster responses after phase compensation and alignment of path responses in single-cluster responses.

[0127] Step S620: Input the training set into the initial signal processing network, extract the feature representation of the training set through the initial encoder, and reconstruct the feature representation through the initial decoder; calculate the reconstruction loss between the reconstruction result and the training set based on the preset loss function, and iteratively calculate the parameters of the initial encoder and the initial decoder through backpropagation until the initial signal processing network converges, and obtain the trained signal processing network, which includes a fully trained encoder and a fully trained decoder.

[0128] In this embodiment, the training set is input into an initial signal processing network. An initial encoder extracts feature representations from the training set, and an initial decoder reconstructs these feature representations. Based on a preset loss function, the reconstruction loss between the reconstructed result and the training set is calculated, where the loss function can be defined as:

[0129]

[0130] Among them, f en For encoder networks, f deHere, Q represents the decoder network, and Q denotes the quantization function.

[0131] The optimal encoder network at this point and decoder network Minimize Received. After completing the training. and The parameters in the dataset will be maintained during the inference phase. Through multi-cluster decoupling algorithms and phase alignment algorithms in single-cluster responses, the distribution shift of each single-cluster response in the training set and the new environment can be effectively suppressed. Therefore, when and When deployed to a new environment, it can still effectively compress and decompress cluster responses, thus ensuring the generalization of the model.

[0132] In summary, the trained encoder and decoder can be obtained, and the trained encoder can be set at the user end, while the trained decoder can be set at the base station end.

[0133] Based on the same inventive concept, such as Figure 7 As shown in the embodiments of this application, a channel feedback method is also provided, applied at the base station, and a communication connection is established between the base station and the user terminal. The method further includes:

[0134] Step S710: Receive the encoding processing result from the user terminal. The encoding processing result is obtained by the user terminal encoding the target single-cluster response corresponding to each single-cluster response. The target single-cluster response is obtained by performing phase compensation on all path responses in the single-cluster response according to the phase of the target path response. For each single-cluster response, the target path response is determined according to the peak position index in the single-cluster response. The single-cluster response is obtained by decoupling the multi-cluster channel response matrix of the downlink channel. Each single-cluster response includes multiple path responses.

[0135] In this embodiment of the application, the base station receives the encoding processing result transmitted from the user terminal. The encoding processing result is obtained by the user terminal encoding the target single-cluster response corresponding to each single-cluster response. The target single-cluster response is obtained by performing phase compensation on all path responses in the single-cluster response based on the phase of the target path response with the strongest energy in each single-cluster response. For each single-cluster response, the target path response with the strongest energy is determined based on the peak position index in the single-cluster response. The single-cluster response is obtained by performing multi-cluster decoupling processing based on the multi-coarse response matrix of the downlink channel. Each single-cluster response includes multiple path responses.

[0136] Step S720: The encoding result is decoded based on the decoder preset at the base station to obtain the decoded single cluster response corresponding to each target single cluster response. The decoded single cluster responses are then superimposed to obtain the feedback result corresponding to the multi-cluster channel response matrix.

[0137] In this embodiment of the application, the base station decodes the encoding result using a preset decoder and restores the decoded result using a preset restoration operator to obtain the decoded single-cluster response corresponding to each target single-cluster response. The decoded single-cluster responses are then superimposed to obtain the feedback result corresponding to the above-mentioned multi-cluster channel response matrix.

[0138] In an exemplary embodiment, the encoding result is decoded based on a decoder preset at the base station to obtain a decoded single-cluster response corresponding to each target single-cluster response, including:

[0139] The decoder decodes the encoded result to obtain the decoded result; the peak position index of each single cluster response transmitted by the user terminal and the phase of the target path response are obtained, and the corresponding single cluster response is recovered based on the peak position index and the phase of the target path response to obtain the recovered decoded single cluster response corresponding to each single cluster response.

[0140] First, the encoding result is decoded using the decoder at the base station to obtain the decoded result.

[0141] Then, the peak position index corresponding to each single cluster response and the phase quantization bit corresponding to the target path response are obtained from the user terminal. In practical applications, the peak position index corresponding to each single cluster response and the phase quantization bit corresponding to the target path response are transmitted to the base station synchronously with the above encoding results.

[0142] The base station performs recovery calculations using a pre-defined cluster feature recovery operator, which includes the decoding results corresponding to each individual cluster response, the peak position index corresponding to each individual cluster response, and the phase quantization bit input value corresponding to each individual cluster response, to obtain the recovered cluster response. This leads to the cluster response of the l-th antenna frequency domain recovered by the base station being... The reconstructed multi-cluster channel response matrix corresponds to the antenna frequency domain channel. for That is, the decoded single-cluster response corresponding to each single-cluster response after the above recovery.

[0143] This application also provides a preferred embodiment of a channel feedback method. Figure 8 This is a schematic diagram of a channel feedback method in a preferred embodiment.

[0144] First, the user end establishes a multi-cluster channel response matrix for the downlink channel and performs multi-cluster decoupling processing on the multi-cluster channel response matrix to obtain multiple single-cluster responses, such as... Figure 8As shown in the diagram, the right column of the multi-cluster decoupling algorithm represents multiple single-cluster responses. These single-cluster responses also demonstrate that the power distribution of the decoupled clusters in the angular delay domain is more concentrated. Then, for each single-cluster response, phase compensation (phase alignment) is performed independently based on a preset fine-grained cluster alignment operator. Specifically, for each single-cluster response, the peak position index is first determined, and the target path response is determined based on this index. Phase compensation is then performed on all path responses within the single-cluster response based on the phase of the target path response, yielding the target single-cluster response for each single-cluster response. These target single-cluster responses are also cluster responses aligned in the angular delay domain. Finally, each target single-cluster response is input into a preset encoder to obtain compressed and quantized cluster response bits, i.e., the encoding result, which is then sent to the corresponding base station.

[0145] After the base station obtains the encoding result, it first inputs the encoding result into the decoder. The decoder decodes each compressed target cluster response to obtain the decoding result. Then, the decoding result, the peak position index of each cluster response transmitted from the user terminal, and the phase of the quantized target path response are input into a preset feature recovery operator, i.e., the above. The responses of each cluster are recovered to obtain the decoded cluster response corresponding to each cluster response. Finally, the decoded cluster responses are superimposed to obtain the feedback result corresponding to the multi-cluster channel response matrix established by the user terminal.

[0146] In summary, the feedback overhead of the embodiments of this application is analyzed as follows:

[0147] From the above Figure 8 It can be seen that the user-side feedback overhead includes the feature bits of each cluster. (i.e., the cluster response bits after compression and quantization, i.e., the encoding result) and metadata bits (i.e., the phase quantization bits of the peak position index of each cluster and the quantized target path response) When the feature compression dimension of each cluster is M and the quantization bit depth of each element is Q... f At that time, the total feedback cost can be expressed as:

[0148]

[0149] Among them, Q p Indicates the number of phase quantization bits, O a O represents the oversampling factor in the angular domain. dThis represents the oversampling factor in the delay domain. It can be seen that the feedback overhead of a single channel instance is proportional to the number of decoupled clusters. Therefore, the cluster number estimation method mentioned above (taking the cluster number with the smallest value between the first and second cluster numbers as the estimated cluster number corresponding to the multi-cluster channel response matrix) achieves an optimal trade-off between feedback overhead and accuracy. When the model is deployed in a specific environment, the average feedback overhead can be expressed as:

[0150]

[0151] in, This represents the average number of clusters. In practical applications, this average can be the average number of clusters that have been split over a period of time.

[0152] In summary, under the same average feedback cost, the solution in this application can reduce the number of parameters in the model. Specifically, under the same average feedback cost scenario, the dimension of the compression and decompression mapping in the feedback method described above in this application can be reduced: In contrast, for codecs, especially when compression dimensions are high, the number of parameters in the compression and decompression linear modules of the codec dominates the overall network. Therefore, the scheme in this application significantly reduces the number of feedback parameters, making it suitable for practical device deployment. Furthermore, the feedback scheme proposed in this application is compatible with existing codec network structures, exhibiting plug-and-play characteristics.

[0153] In summary, the channel feedback method proposed in this application first proposes a generalizable feedback learning framework. Through the phase compensation scheme therein, the distribution offset of multi-cluster large-scale MIMO channels under different environments can be effectively overcome. Furthermore, the low-complexity multi-cluster decoupling scheme proposed in this application can support the real-time operation of the proposed generalizable feedback learning framework.

[0154] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0155] Based on the same inventive concept, this application also provides a channel feedback system for implementing the channel feedback method described above. The solution provided by this device is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more channel feedback system embodiments provided below can be found in the limitations of the channel feedback method described above, and will not be repeated here.

[0156] In one exemplary embodiment, such as Figure 9 As shown, a channel feedback system is provided, including a user terminal 91 and a base station terminal 92, and the user terminal and the base station terminal are connected in communication.

[0157] User terminal 91 is used to execute the channel feedback method applied to the user terminal as described above;

[0158] Base station 92 is used to execute the channel feedback method applied at the base station as described above.

[0159] Each module in the aforementioned channel feedback system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware within or independently of the processor in a computer device, or stored in software within the memory of the computer device, so that the processor can invoke and execute the corresponding operations of each module.

[0160] In one exemplary embodiment, such as Figure 10 As shown, a channel feedback device is provided for use at a user end. The device includes:

[0161] The first acquisition module 101 is used to establish a multi-cluster channel response matrix for the downlink channel and to decouple the multi-cluster channel response matrix to obtain multiple single-cluster responses; wherein each single-cluster response includes multiple path responses.

[0162] The first calculation module 102 is used to obtain the peak position index in the single cluster response for each single cluster response, determine the target path response in the single cluster response based on the peak position index, and perform phase compensation on all path responses in the single cluster response based on the phase of the target path response to obtain the target single cluster response corresponding to the single cluster response.

[0163] The first generation module 103 is used to obtain the encoding processing result of the target single cluster response corresponding to each single cluster response, and send each encoding processing result to the base station to instruct the base station to decode each encoding processing result to obtain the channel feedback result corresponding to the multi-cluster channel response matrix.

[0164] In one embodiment, the channel feedback device further includes:

[0165] The second decomposition module is used to perform singular value decomposition on the multi-cluster channel response matrix to obtain all the initial single-cluster response channels contained in the multi-cluster channel response matrix.

[0166] The second calculation module is used to determine the number of clusters corresponding to the multi-cluster channel response matrix;

[0167] The second filtering module is used to sort the singular values ​​in descending order based on the magnitude of the singular values ​​of each initial single-cluster response channel to obtain a descending sequence, and to filter the singular values ​​ranked first in the descending sequence based on the number of clusters to obtain multiple target singular values. Based on the target singular values, the corresponding single-cluster response channel is determined, and the corresponding single-cluster response is obtained based on the single-cluster response channel.

[0168] In one embodiment, the channel feedback device further includes:

[0169] The third calculation module is used to calculate the number of the first cluster based on the multi-cluster channel response matrix and the singular value decomposition results; and to calculate the number of the second cluster based on the multi-cluster channel response matrix, the singular value decomposition results, and the preset energy threshold.

[0170] The third determining module is used to take the smallest cluster number among the first cluster number and the second cluster number as the cluster number corresponding to the multi-cluster channel response matrix.

[0171] In one embodiment, the channel feedback device further includes:

[0172] The fourth quantization module is used to perform parameter uniform quantization on the phase of the target path response to obtain the phase quantization result;

[0173] The fourth compensation module is used to rotate the phase of all path responses in a single cluster response based on the phase quantization results to obtain the target single cluster response.

[0174] In one embodiment, the channel feedback device further includes:

[0175] The fifth calculation module is used to determine the position of the strongest angular peak in each cluster response based on a preset angle domain oversampled discrete Fourier transform codebook; and to determine the position of the strongest time-delay domain peak in each cluster response based on a preset time-delay domain oversampled discrete Fourier transform codebook.

[0176] The fifth determination module is used to determine the peak position index based on the angular peak position and the time delay domain peak position.

[0177] In one embodiment, the channel feedback device further includes:

[0178] The sixth encoding module is used to encode each cluster response based on the encoder in the user terminal, obtain the encoding result of each target cluster response, and send the encoding result to the base station to instruct the base station to decode the encoding result to obtain the channel feedback result corresponding to the multi-cluster channel matrix.

[0179] In one embodiment, the channel feedback device further includes:

[0180] The seventh module is used to establish an initial signal processing network consisting of an initial encoder and an initial decoder, and to obtain a preset training set; wherein, the training set includes multiple training single-cluster responses for path responses that have undergone phase compensation;

[0181] The seventh training module is used to input the training set into the initial signal processing network, extract the feature representation of the training set through the initial encoder, and reconstruct the feature representation through the initial decoder. Based on the preset loss function, the reconstruction loss between the reconstruction result and the training set is calculated, and the parameters of the initial encoder and the initial decoder are iteratively calculated through backpropagation until the initial signal processing network converges, thus obtaining the trained signal processing network, which includes a fully trained encoder and a fully trained decoder.

[0182] In one exemplary embodiment, such as Figure 11 As shown, a channel feedback device is provided, applied at a base station, the device comprising:

[0183] The eighth receiving module 111 is used to receive the encoding processing result transmitted from the user terminal. The encoding processing result is obtained by the user terminal encoding the target single cluster response corresponding to each single cluster response. The target single cluster response is obtained by performing phase compensation on all path responses in the single cluster response according to the phase of the target path response. For each single cluster response, the target path response is determined according to the peak position index in the single cluster response. The single cluster response is obtained by decoupling the multi-cluster channel response matrix of the downlink channel. Each single cluster response includes multiple path responses.

[0184] The eighth decoding module 112 is used to decode the encoding result based on the decoder preset at the base station, to obtain the decoded single cluster response corresponding to each target single cluster response, and to superimpose the decoded single cluster responses to obtain the feedback result corresponding to the multi-cluster channel response matrix.

[0185] In one embodiment, the channel feedback device further includes:

[0186] The ninth decoding module is used to decode the encoded result based on the decoder to obtain the decoded result;

[0187] The ninth recovery module is used to obtain the peak position index of each single cluster response transmitted by the user terminal and the phase of the target path response, and to perform recovery calculations on the corresponding single cluster response based on the peak position index and the phase of the target path response to obtain the recovered decoded single cluster response corresponding to each single cluster response.

[0188] In one exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement any of the channel feedback methods described above.

[0189] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements any of the channel feedback methods described above.

[0190] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements any of the channel feedback methods described above.

[0191] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0192] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0193] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A channel feedback method, characterized in that, When applied to a user terminal, the method includes: A multi-cluster channel response matrix for the downlink channel is established, and the multi-cluster channel response matrix is ​​decoupled to obtain multiple single-cluster responses; wherein each single-cluster response includes multiple path responses; For each cluster response, obtain the peak position index in the cluster response, determine the target path response in the cluster response based on the peak position index, and perform phase compensation on all path responses in the cluster response based on the phase of the target path response to obtain the target cluster response corresponding to the cluster response. The encoding processing result of the target single cluster response corresponding to each single cluster response is obtained, and each encoding processing result is sent to the base station to instruct the base station to decode each encoding processing result to obtain the channel feedback result corresponding to the multi-cluster channel response matrix.

2. The method according to claim 1, characterized in that, The decoupling process of the multi-cluster channel response matrix yields multiple single-cluster responses, including: Singular value decomposition is performed on the multi-cluster channel response matrix to obtain all the initial single-cluster response channels contained in the multi-cluster channel response matrix; Determine the number of clusters corresponding to the multi-cluster channel response matrix; Based on the magnitude of the singular values ​​of each initial single-cluster response channel, the singular values ​​are sorted in descending order to obtain a descending sequence. The singular values ​​ranked first in the descending sequence are then filtered based on the number of clusters to obtain multiple target singular values. The corresponding single-cluster response channel is determined based on the target singular values, and the corresponding single-cluster response is obtained based on the single-cluster response channel.

3. The method according to claim 2, characterized in that, Determining the number of clusters corresponding to the multi-cluster channel response matrix includes: The number of the first cluster is calculated based on the multi-cluster channel response matrix and the singular value decomposition results; The number of the second cluster is calculated based on the multi-cluster channel response matrix, the singular value decomposition result, and the preset energy threshold. The cluster number with the smallest value between the first cluster number and the second cluster number is taken as the cluster number corresponding to the multi-cluster channel response matrix.

4. The method according to any one of claims 1 to 3, characterized in that, The step of performing phase compensation on all path responses in the single-cluster response based on the phase of the target path response to obtain the target single-cluster response corresponding to the single-cluster response includes: The phase of the target path response is subjected to parameter uniform quantization to obtain the phase quantization result; Based on the phase quantization result, the phases of all path responses in the single-cluster response are rotated to obtain the target single-cluster response.

5. The method according to any one of claims 1 to 3, characterized in that, Obtaining the peak position index in the single cluster response includes: For each of the single-cluster responses, the position of the angle peak with the strongest energy in the single-cluster response is determined based on a preset angle domain oversampled discrete Fourier transform codebook. Based on a preset time-delay domain oversampled discrete Fourier transform codebook, the position of the time-delay domain peak with the strongest energy in the single-cluster response is determined. The peak position index is determined based on the angle peak position and the time delay domain peak position.

6. The method according to any one of claims 1 to 3, characterized in that, The step of obtaining the encoding processing result of the target single-cluster response corresponding to each of the single-cluster responses and sending each of the encoding processing results to the base station to instruct the base station to decode each of the encoding processing results to obtain the channel feedback result corresponding to the multi-cluster channel response matrix includes: The encoder in the user terminal encodes each of the single-cluster responses to obtain the encoding result of each target single-cluster response, and sends the encoding result to the base station to instruct the base station to decode the encoding result to obtain the channel feedback result corresponding to the multi-cluster channel matrix.

7. The method according to claim 6, characterized in that, The method further includes: An initial signal processing network consisting of an initial encoder and an initial decoder is established, and a preset training set is obtained; wherein, the training set includes multiple training single-cluster responses for path responses that have undergone phase compensation; The training set is input into the initial signal processing network, the initial encoder extracts the feature representation of the training set, and the initial decoder reconstructs the feature representation. Based on a preset loss function, the reconstruction loss between the reconstruction result and the training set is calculated, and the parameters of the initial encoder and the initial decoder are iteratively calculated through backpropagation until the initial signal processing network converges, thereby obtaining the trained signal processing network, which includes the fully trained encoder and the fully trained decoder.

8. A channel feedback method, characterized in that, Applied to the base station, the method further includes a connection between the base station and the user terminal, and is further described as follows: The system receives encoding processing results from the user terminal, wherein the encoding processing results are obtained by the user terminal encoding the target single-cluster response corresponding to each single-cluster response, the target single-cluster response is obtained by performing phase compensation on all path responses in the single-cluster response according to the phase of the target path response, for each single-cluster response, the target path response is determined according to the peak position index in the single-cluster response, the single-cluster response is obtained by decoupling processing according to the multi-cluster channel response matrix of the downlink channel, and each single-cluster response includes multiple path responses; The encoding result is decoded based on the decoder preset at the base station to obtain a decoded single-cluster response corresponding to each target single-cluster response. The decoded single-cluster responses are then superimposed to obtain a feedback result corresponding to the multi-cluster channel response matrix.

9. The method according to claim 8, characterized in that, The step of decoding the encoding result based on the decoder preset at the base station to obtain the decoded single-cluster response corresponding to each target single-cluster response includes: The encoded result is decoded using the decoder to obtain the decoded result; The peak position index of each cluster response transmitted by the user terminal and the phase of the target path response are obtained. Based on the peak position index and the phase of the target path response, the corresponding cluster response is recovered to obtain the recovered decoded cluster response corresponding to each cluster response.

10. A channel feedback system, characterized in that, It includes a user terminal and a base station terminal, and the user terminal and the base station terminal are connected in communication. The user terminal is used to execute the channel feedback method as described in any one of claims 1 to 7; The base station is used to execute the channel feedback method as described in any one of claims 8 to 9.

Images