Channel state information feedback compression method based on large-scale MIMO system

Abstract

The invention discloses a channel state information feedback compression method based on a large-scale MIMO system, and the method comprises the steps: building a time and space dual-domain channel change rate quantitative index through obtaining a historical CSI matrix, and predicting a future change rate through a time convolution network; establishing a block granularity mapping function based on a prediction result, dynamically adjusting the block granularity of the CSI matrix, and generating a sub-matrix; the method comprises the following steps: aiming at the dynamic characteristics of a channel, adopting an adaptive dual-mode compression strategy, using improved Kronecker sparse representation in a dynamic scene, and using an attention-enhanced auto-encoder in a static scene; optimizing a compressed sub-matrix feedback process through a joint codebook and a closed-loop feedback mechanism; through the blocking strategy and the compression mode selection mechanism, the CSI feedback overhead is reduced, meanwhile, high-quality reconstruction and transmission of the CSI are ensured, and the method is suitable for a complex and changeable channel environment.

Description

Technical Field

[0001] The present invention relates to the technical field of channel state information processing, and particularly to a method for compressing feedback of channel state information based on a massive MIMO system. Background Art

[0002] With the rapid development of the fifth-generation mobile communication technology (5G), massive multiple-input multiple-output (Massive MIMO) technology has become the core enabling technology of the new generation of wireless communication systems due to its breakthrough advantages in spectral efficiency, energy efficiency, and system capacity. By configuring a large-scale antenna array on the base station side, this technology utilizes spatial domain resources to achieve precise beamforming and efficient spatial multiplexing, thereby significantly improving system throughput and user service quality. However, the system performance strongly depends on the accuracy of channel state information (CSI), which poses a severe challenge in practical deployment. Specifically, the user equipment (UE) needs to feedback a high-dimensional CSI matrix to the base station through the uplink. The exponential growth of the antenna scale and the number of subcarriers leads to a significant jump in the dimension of the CSI matrix, resulting in a huge feedback overhead pressure and severely restricting the effective utilization of system bandwidth resources.

[0003] Regarding the key issue of CSI feedback compression, the academic community has proposed various solutions. For example, sparse reconstruction methods based on compressed sensing (CS), intelligent coding schemes relying on deep neural networks (DNN), and dimension reduction strategies based on matrix decomposition. However, existing methods still have significant limitations in engineering practice: First, traditional compressed sensing methods are based on the theoretical framework of channel sparsity assumptions. However, in the actual communication environment, factors such as multipath effects, time-varying fading, and non-ideal scattering lead to dynamic uncertainty in the channel sparse characteristics, resulting in a significant degradation in reconstruction performance. Second, existing deep learning methods are mostly static models and do not fully consider the dynamic change characteristics of the channel in the time and space dimensions, making it difficult to achieve adaptive compression for dynamic scenarios. Third, traditional matrix decomposition methods usually adopt a fixed block strategy and cannot dynamically adjust the block granularity according to the channel change situation, resulting in it being difficult to balance compression efficiency and reconstruction quality.

[0004] Therefore, there is an urgent need to propose a CSI feedback compression method that can fully consider the spatio-temporal dynamic characteristics of the channel, achieve adaptive block division and compression strategies, so as to effectively reduce the feedback overhead and improve the channel reconstruction accuracy. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract, and the title, and such simplifications or omissions shall not be used to limit the scope of the present invention.

[0006] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides a method for compressing channel state information feedback based on a large-scale MIMO system to solve the problems raised in the background art.

[0007] To solve the above technical problems, the present invention provides the following technical solution: A method for compressing channel state information feedback based on a large-scale MIMO system, including:

[0008] Obtain a historical channel state information matrix, construct a quantization index of the channel change rate in the dual domain, and design a temporal convolutional network prediction module through the quantization index of the channel change rate in the dual domain to predict the future quantization index of the channel change rate in the dual domain;

[0009] Establish a block granularity mapping function through the predicted future quantization index of the channel change rate in the dual domain, and dynamically adjust the block granularity of the channel state information matrix according to the block granularity mapping function to obtain a sub-matrix of the channel state information after partitioning;

[0010] Compress the sub-matrix, optimize the feedback process of the sub-matrix after compression, and ensure the efficient transmission and high-quality reconstruction of the channel state information.

[0011] As a preferred embodiment of the method for compressing channel state information feedback based on a large-scale MIMO system of the present invention, wherein: the obtaining of the historical channel state information matrix and the construction of the quantization index of the channel change rate in the dual domain include:

[0012] Consider the change rates of the historical channel state information in the spatial and temporal dimensions;

[0013] Obtain the change rate in the temporal dimension by comparing the differences between adjacent time slots, and obtain the change rate in the spatial dimension by calculating the channel response power variance of each antenna pair over all subcarriers.

[0014] As a preferred embodiment of the method for compressing channel state information feedback based on a large-scale MIMO system of the present invention, wherein: designing a temporal convolutional network prediction module through the quantization index of the channel change rate in the dual domain to predict the future quantization index of the channel change rate in the dual domain includes:

[0015] Extract features from the change rate in the temporal dimension and the change rate in the spatial dimension based on causal dilated convolution in the temporal convolutional network, and predict the future change rate in the temporal dimension and the change rate in the spatial dimension;

[0016] Input the predicted future time dimension change rate and future space dimension change rate into the offline reinforcement learning model to obtain the future time dimension change rate threshold, future space dimension change rate threshold, and hyperparameters.

[0017] Among them, the causal dilated convolution is a multi-layer convolution structure, each layer of the convolution structure contains different dilation factors, and each layer of dilation factors increases by 2 l for increment, where k is the number of layers.

[0018] As a preferred solution of the channel state information feedback compression method based on a large-scale MIMO system according to the present invention, wherein: a block granularity mapping function is established through the predicted future two-domain channel change rate quantization index, including:

[0019] According to the hyperparameters, map the predicted future time dimension change rate and future space dimension change rate into the block size.

[0020] As a preferred solution of the channel state information feedback compression method based on a large-scale MIMO system according to the present invention, wherein: according to the block granularity mapping function, dynamically adjust the block granularity of the channel state information matrix to obtain a sub-matrix of the channel state information after blocking, including:

[0021] Obtain the current channel state information matrix and create a blocking strategy, and the blocking strategy is divided into a spatial domain blocking strategy and a frequency domain blocking strategy;

[0022] In the spatial domain blocking strategy, divide the current channel state information matrix through the block size, and adopt an overlapping block method according to the remaining size after division;

[0023] In the frequency domain blocking strategy, discriminate the future space dimension change rate by a preset threshold. If the future space dimension change rate is greater than the preset threshold, adopt non-uniform blocking, otherwise adopt uniform blocking.

[0024] As a preferred solution of the channel state information feedback compression method based on a large-scale MIMO system according to the present invention, wherein: compress the sub-matrix, including:

[0025] The compression method is divided into a dual mode of dynamic scenario and static scenario;

[0026] For the dynamic scenario, an improved Kronecker sparse representation is adopted, and for the static scenario, an attention-enhanced autoencoder is adopted.

[0027] As a preferred solution of the channel state information feedback compression method based on a large-scale MIMO system according to the present invention, wherein: further includes:

[0028] Define dynamic scene and static scene decision functions. When the future time dimension change rate is greater than the future time dimension change rate threshold and the future space dimension change rate is greater than the future space dimension change rate threshold, the compression method adopts the dynamic scene, otherwise the static scene is adopted.

[0029] As a preferred solution of the channel state information feedback compression method based on large-scale MIMO system described in the present invention, the improved Kronecker sparse representation includes:

[0030] A joint dictionary is constructed, wherein the joint dictionary is composed of a time dictionary and a space dictionary, wherein the time dictionary is generated by Hankel matrix decomposition.

[0031] As a preferred solution of the channel state information feedback compression method based on a large-scale MIMO system described in the present invention, wherein: the attention-enhanced autoencoder comprises:

[0032] A spatial-frequency attention gate is introduced in the autoencoder.

[0033] As a preferred solution of the channel state information feedback compression method based on a large-scale MIMO system described in the present invention, the feedback process of the sub-matrix after optimization compression includes:

[0034] According to the block size, compression mode and quantization step size, a joint codebook of a triplet structure is obtained, and the compressed sub-matrix is ​​mapped to the joint codebook as an index in the codebook;

[0035] The compressed sub-matrix is ​​reconstructed by the joint codebook and index, and the reconstructed error value is calculated. If the error value is greater than 0, feedback is triggered, and the reconstruction information is fed back to the temporal convolutional network prediction module to continuously optimize and predict the future dual-domain channel change rate quantization index and the compression process of the sub-matrix until the error value is 0.

[0036] Compared with the prior art, the invention has the following beneficial effects:

[0037] 1. The present invention constructs a dual-domain channel change rate quantitative index and uses a temporal convolutional network (TCN) to predict future change trends and dynamically perceive channel changes, thus getting rid of the traditional method's reliance on sparsity assumptions, being able to adapt to complex and changing environments and ensuring the stability of compression performance;

[0038] 2. Since the traditional matrix decomposition method adopts a fixed block strategy and cannot be flexibly adjusted according to channel changes, it is difficult to balance compression efficiency and reconstruction quality. The present invention can optimize the block granularity according to channel characteristics through a dynamic block mechanism, refine the blocks in areas with drastic changes, and coarsen the blocks in stable areas, thereby ensuring compression efficiency while ensuring reconstruction accuracy;

[0039] 3. In addition, by establishing a scenario decision function to dynamically select the compression mode, we ensure that the optimal strategy can be adopted under different channel conditions. By creating a joint codebook, we can reduce the compression feedback overhead while maintaining the accuracy of CSI reconstruction, ensuring the stability and reliability of the system in long-term operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0041] Figure 1 The present invention is an overall flow chart of a channel state information feedback compression method based on a large-scale MIMO system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0043] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0044] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0045] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0046] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0047] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0048] Example 1

[0049] Reference Figure 1 , which is the first embodiment of the present invention, and provides a channel state information feedback compression method based on a large-scale MIMO system, comprising:

[0050] S1. Obtain a historical channel state information matrix, construct a dual-domain channel change rate quantification index, and design a temporal convolutional network prediction module based on the dual-domain channel change rate quantification index to predict the future dual-domain channel change rate quantification index;

[0051] Specifically, in a massive MIMO system, a pilot signal is sent by a transmitting end, and a receiving end (such as a user equipment UE) receives the pilot signal and calculates a channel estimate, and the calculated channel estimate result is stored as a channel state information matrix H k , and record the channel state information matrix in order of time slots;

[0052] It should be noted that the pilot signal is a known sequence periodically sent by the transmitting end, and the receiving end calculates the channel estimation by using these signals. The channel estimation can adopt the minimum mean square error MMSE or zero-forcing ZF method;

[0053] Specifically, H kIt is represented as the historical channel state information matrix at the kth time slot, and its dimension is expressed as:

[0054]

[0055] Where C is the complex field, which means that each element in the complex field contains a real part and an imaginary part; N tx Expressed as the number of transmitting antennas; N rx Expressed as the number of receiving antennas; N sc Expressed as the number of subcarriers;

[0056] Specifically, the historical channel state information matrix H k Normalize the amplitude to get:

[0057]

[0058] in,‖·‖ F It is expressed as the Frobenius norm to ensure the stability of the channel state information matrix during the normalization process;

[0059] Further, the change rate of the historical channel state information in the spatial and temporal dimensions is considered;

[0060] It should be explained that since the channel environment fluctuates due to changes in time and space, such as user movement, distribution of antenna arrays, etc., it is necessary to quantify these changes; specifically, the change rate indicators are defined in the time and space dimensions, namely, the time dimension change rate and the space dimension change rate. The time dimension change rate is used to measure the speed of change of the channel in a continuous time period, and the space dimension change rate is used to measure the spatial difference of the channel, where the channel responses between different antennas may differ significantly;

[0061] Furthermore, the time dimension change rate is obtained by comparing the differences between adjacent time slots, and the spatial dimension change rate is obtained by calculating the variance of the channel response power of each antenna pair on all subcarriers;

[0062] Specifically, the time dimension change rate R t It is expressed as:

[0063]

[0064] Among them, H k (i) is represented as the historical channel state information matrix of the i-th subcarrier at the k-th time slot, where K is the time window length, indicating the time range of observation;

[0065] It should be noted that if R t If R is large, it means that the channel changes quickly, for example, the user moves at high speed. tIf it is smaller, it means that the channel is relatively stable;

[0066] Specifically, the spatial dimension change rate R s It is expressed as:

[0067]

[0068] Where Var(·) represents the variance, which calculates the power variation of each antenna pair on all subcarriers; h m,n (f) is represented by the channel response of the mth transmitting antenna and the nth receiving antenna on the i-th subcarrier;

[0069] It should be noted that if R s If R s If it is smaller, it means that the spatial distribution is more uniform;

[0070] It should be noted that it is not enough to just know the change rate of the time dimension and the space dimension. It is also necessary to predict the change rate of the time dimension and the space dimension in the future in order to prepare for the dynamic block segmentation and compression method selection. To this end, a prediction module based on the temporal convolutional network (TCN) is designed to predict the change rate of the time dimension and the space dimension.

[0071] It should be explained that TCN is a neural network that specializes in processing time series. Through causal dilated convolution, it can capture long-term dependencies in historical data while avoiding the gradient vanishing problem that may be encountered in traditional recurrent neural networks (such as LSTM).

[0072] Furthermore, based on the causal dilation convolution in the time convolution network, feature extraction is performed on the time dimension change rate and the space dimension change rate, and the future time dimension change rate and the space dimension change rate are predicted;

[0073] Specifically, the causal dilation convolution is a multi-layer convolution structure, each layer of the convolution structure contains a different dilation factor, and each layer of the dilation factor is 2 l Incrementally, l is the number of layers;

[0074] It should be noted that by expanding the dilation factor, the receptive field of the temporal convolutional network can be gradually expanded, and it can see further into the past;

[0075] Specifically, the mathematical formula for predicting the future time dimension change rate and space dimension change rate can be expressed as:

[0076]

[0077] Among them, Rt-kΔt Expressed as the time rate of change of the kth time slot, R s-kΔt Expressed as the spatial rate of change of the kth time slot; It is expressed as a predicted value, including the predicted results of the change rate of future time dimension and space dimension; L is expressed as a historical time slot;

[0078] Furthermore, the predicted future time dimension change rate and future space dimension change rate are input into the offline reinforcement learning model to obtain the future time dimension change rate threshold, the future space dimension change rate threshold and hyperparameters;

[0079] It should be explained that reinforcement learning (RL) is a machine learning method for optimizing decision-making, while offline reinforcement learning refers to using historical data or simulation environments to generate strategies during the training phase and directly applying them during actual deployment without the need for real-time online training.

[0080] Specifically, the framework of reinforcement learning includes, state: the characteristics of the current channel environment, action: the specific threshold of the change rate of the future time dimension and the change rate of the future space dimension, and reward: designed according to the performance indicators of the system;

[0081] Specifically, in the training phase, a large number of "state-action-reward" samples are generated by using the historical channel state information matrix or the simulation environment, and then the reinforcement learning model is trained to learn to select the optimal future time dimension change rate threshold and future space dimension change rate threshold under different states; in practical applications, because the training is offline, the computational overhead is low during deployment;

[0082] S2, establishing a block granularity mapping function through the predicted future dual-domain channel change rate quantization index, and dynamically adjusting the block granularity of the channel state information matrix according to the block granularity mapping function to obtain a sub-matrix of the channel state information after blocking;

[0083] Further, the predicted future time dimension change rate and future space dimension change rate are mapped to the block size by using the hyperparameters, and a block granularity mapping function is established according to the process of mapping to the block size;

[0084] Specifically, the block granularity mapping function is expressed as:

[0085]

[0086] Where B represents the block size, α, β, and γ are hyperparameters used to balance the impact of temporal and spatial change rates and adapt to different scenarios; B max The maximum allowed chunk size, which limits the upper limit of chunks; represents the floor function; exp(·) represents the natural exponential function, which is used to ensure that the channel variation is smooth and nonlinear;

[0087] It should be noted that when the channel changes rapidly ( and When the channel size is larger, the block size B will be smaller to capture more details; similarly, when the channel changes slowly, the block size B will be larger to improve the compression efficiency;

[0088] Further, obtaining a current channel state information matrix;

[0089] Specifically, the current channel state information matrix is ​​obtained, whose dimension is the same as the historical channel state information matrix H k Keep the same, but the time slot k = K + 1;

[0090] It should be explained that, since the size of the current channel state information matrix obtained may not be an integer multiple of the block size B, it is necessary to consider how to process the remaining part of the block;

[0091] Further, a block strategy is created, wherein the block strategy is divided into a space domain block strategy and a frequency domain block strategy;

[0092] Furthermore, in the spatial domain (antenna dimension) block strategy, the current channel state information matrix is ​​divided by the block size, and an overlapping block method is adopted according to the remaining size after the division;

[0093] Specifically, the current channel state information matrix is ​​divided according to B×B, for example, B=4, then the matrix is ​​divided into multiple 4×4 sub-blocks, if there are remaining blocks after the division or the size is incorrect during the division process, an overlapping block method is adopted;

[0094] Specifically, the overlapping block allows adjacent sub-blocks to have partial overlap. For example, assuming that the number of transmitting antennas is 64, the dimension of receiving antennas is 4, and the block size is 4×4, if the dimension of receiving antennas 4 is less than 16, assuming it is 4×3, a complete sub-block cannot be formed, then the receiving antenna dimension is superimposed as a sub-block as a whole;

[0095] Furthermore, in the frequency domain (subcarrier dimension) blocking strategy, the future spatial dimension change rate is judged by means of a preset threshold value. If the future spatial dimension change rate is greater than the preset threshold value, non-uniform blocking is adopted, otherwise uniform blocking is adopted;

[0096] It should be noted that if the future spatial dimension change rate is greater than the preset threshold, it means that the spatial change is large and there is obvious frequency selective fading in the frequency domain. On the contrary, it means that the spatial change is relatively gentle and uniform blocking is suitable.

[0097] Specifically, the preset threshold value can be obtained by statistical analysis, simulation experiment, machine learning or expert experience;

[0098] Specifically, uniform blocking refers to dividing the channel state information matrix into sub-blocks of equal size in a certain dimension; non-uniform blocking refers to dividing the data into sub-blocks of different sizes according to the high-frequency and low-frequency changes of the channel state information matrix in the frequency domain, for example, using larger sub-blocks in the low-frequency area and smaller sub-blocks in the high-frequency area;

[0099] It should be noted that the sub-matrix refers to the channel state information matrix after block division, which is composed of multiple sub-blocks;

[0100] S3, compressing the sub-matrix, optimizing the feedback process of the compressed sub-matrix, and ensuring efficient transmission and high-quality reconstruction of channel state information;

[0101] Furthermore, the compression method of the sub-matrix is ​​divided into two modes: dynamic scene and static scene;

[0102] It should be noted that, since different channel dynamics require different compression strategies, two compression methods are designed in the solution of the present invention, and adaptive switching is performed according to the scene decision function;

[0103] Furthermore, the dynamic scene adopts an improved Kronecker sparse representation, and the static scene adopts an attention-enhanced autoencoder;

[0104] Specifically, for the improved Kronecker sparse representation, a joint dictionary is constructed;

[0105] Specifically, the joint dictionary D is expressed as:

[0106]

[0107] Among them, D s Represented as a spatial dictionary, it is used to capture the sparse features of the CSI matrix in the spatial dimension; Expressed as Kronecker product; D t Represented as a time dictionary, it is used to capture the sparse features of the CSI matrix in the time dimension and is generated by Hankel matrix decomposition:

[0108]

[0109] in, Expressed as the reconstruction error term, λ∥X∥ 1,2 represents the regularization term; argmin represents the minimization objective; H(H t) represents the conversion of the channel state information matrix Ht in the time dimension into a Hankel matrix; X is the sparse coefficient matrix; λ is the penalty coefficient, which controls the trade-off between reconstruction error and sparsity. The larger λ is, the stronger the sparsity is, and the smaller λ is, the smaller the reconstruction error is.

[0110] Specifically, for the attention-enhanced autoencoder, a space-frequency attention gate is introduced into the autoencoder;

[0111] Specifically, the space-frequency attention gate A is expressed as follows:

[0112] A=σ(W a [H s ;H f ]+b a )

[0113] Among them, σ(·) represents the Sigmoid activation function, which is used to map the attention weight matrix output to the (0,1) interval, [H s ;H f ] represents the spatial feature H s With frequency characteristics H f Splicing; W a Represented as the attention weight matrix; b a Expressed as a bias term;

[0114] It should be noted that b a By linear transformation W a [H s ;H f ]Apply an offset to ensure that the attention mechanism can more accurately capture the importance of spatial and frequency features;

[0115] Specifically, by defining dynamic scene and static scene decision functions, when the future time dimension change rate is greater than the future time dimension change rate threshold and the future space dimension change rate is greater than the future space dimension change rate threshold, the compression method adopts the dynamic scene, otherwise the static scene is adopted;

[0116] Further, a joint codebook of a triplet structure is obtained according to the block size, the compression method and the quantization step size, and the compressed sub-matrix is ​​mapped to the joint codebook as an index in the codebook;

[0117] It should be explained that in the communication system, a codebook is a predefined set of symbols used to map data into transmittable signals. In CSI feedback, the role of the codebook is to represent the compressed CSI as a series of indexes, which can greatly reduce the amount of data that needs to be transmitted.

[0118] Specifically, the joint codebook F of the triplet structure is expressed as:

[0119] F={(B,mode,Q)}

[0120] Among them, mode represents the compression method, static scene or dynamic scene; Q represents the quantization step size, which is expressed as:

[0121]

[0122] Among them, Q min is the minimum value of the quantization step, Q max is the maximum value of the quantization step size, R t,min is the minimum value of the rate of change in the time dimension, R t,max is the maximum value of the rate of change in the time dimension;

[0123] It should be noted that the quantization step size can be based on The size of is adaptively adjusted to maintain a balance between transmission overhead and reconstruction accuracy;

[0124] It should be noted that in order to ensure the quality of the CSI reconstructed at the receiving end, the reconstructed CSI needs to be evaluated;

[0125] Furthermore, the compressed sub-matrix is ​​reconstructed by the joint codebook and the index, and the reconstructed error value is calculated. If the error value is greater than 0, feedback is triggered, and the reconstruction information is fed back to the temporal convolutional network prediction module to continuously optimize and predict the future dual-domain channel change rate quantization index and the compression process of the sub-matrix until the error value is 0;

[0126] Specifically, the evaluation formula of the reconstructed error value ∈ is as follows:

[0127]

[0128] Among them, H(i) is the original CSI, For the reconstructed CSI;

[0129] Specifically, when the feedback is triggered, the receiving end will notify the transmitting end, and the transmitting end will fine-tune the parameters in the temporal convolutional network (TCN) model according to the error value reconstructed in the trigger feedback, and repeat the above process to re-predict and compress until the error value is 0.

[0130] Those skilled in the art will appreciate that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program codes. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0131] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0132] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0133] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0134] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0135] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A channel state information feedback compression method based on a massive MIMO system, characterized in that: include: Obtaining a historical channel state information matrix, constructing a dual-domain channel change rate quantification index, and designing a temporal convolutional network prediction module through the dual-domain channel change rate quantification index to predict future dual-domain channel change rate quantification indicators; A block granularity mapping function is established by using the predicted future dual-domain channel change rate quantization index, and according to the block granularity mapping function, the block granularity of the channel state information matrix is ​​dynamically adjusted to obtain a sub-matrix of the channel state information after blocking; The sub-matrix is ​​compressed, and a feedback process of the compressed sub-matrix is ​​optimized to ensure efficient transmission and high-quality reconstruction of channel state information.

2. The channel state information feedback compression method based on a large-scale MIMO system according to claim 1, characterized in that: The obtaining of the historical channel state information matrix and the construction of the dual-domain channel change rate quantitative index include: Considering the rate of change of the historical channel state information in spatial and temporal dimensions; By comparing the differences between adjacent time slots, the time dimension change rate is obtained, and by calculating the channel response power variance of each antenna pair on all subcarriers, the spatial dimension change rate is obtained.

3. The channel state information feedback compression method based on a large-scale MIMO system as claimed in claim 2, characterized in that: Based on the dual-domain channel change rate quantification index, a temporal convolutional network prediction module is designed to predict the future dual-domain channel change rate quantification index, including: Extract features of the time dimension change rate and the space dimension change rate based on the causal dilation convolution in the time convolution network, and predict the future time dimension change rate and the space dimension change rate; The predicted future time dimension change rate and future space dimension change rate are input into the offline reinforcement learning model to obtain the future time dimension change rate threshold, future space dimension change rate threshold and hyperparameters. The causal dilation convolution is a multi-layer convolution structure, each layer of the convolution structure contains a different dilation factor, and the dilation factor of each layer increases by 2l, where l is the number of layers.

4. The channel state information feedback compression method based on a large-scale MIMO system as claimed in claim 3, characterized in that: Through the predicted quantitative index of the future dual-domain channel change rate, a block granularity mapping function is established, including: According to the hyperparameters, the predicted future time dimension change rate and future space dimension change rate are mapped to the block size.

5. The channel state information feedback compression method based on a large-scale MIMO system according to claim 4, characterized in that: According to the block granularity mapping function, dynamically adjusting the block granularity of the channel state information matrix to obtain a sub-matrix of the channel state information after blocking, including: Obtaining a current channel state information matrix, and creating a block strategy, wherein the block strategy is divided into a space domain block strategy and a frequency domain block strategy; In the spatial domain blocking strategy, the current channel state information matrix is ​​divided by the block size, and an overlapping blocking method is adopted according to the remaining size after the division; In the frequency domain blocking strategy, the change rate of the future spatial dimension is judged by means of a preset threshold. If the change rate of the future spatial dimension is greater than the preset threshold, non-uniform blocking is adopted, otherwise uniform blocking is adopted.

6. The channel state information feedback compression method based on a large-scale MIMO system as claimed in claim 5, characterized in that: Compressing the sub-matrix includes: The compression method is divided into two modes: dynamic scene and static scene; The dynamic scene adopts the improved Kronecker sparse representation, and the static scene adopts the attention-enhanced autoencoder.

7. The channel state information feedback compression method based on a large-scale MIMO system according to claim 3 or 6, characterized in that: Also includes: Define dynamic scene and static scene decision functions. When the future time dimension change rate is greater than the future time dimension change rate threshold and the future space dimension change rate is greater than the future space dimension change rate threshold, the compression method adopts the dynamic scene, otherwise the static scene is adopted.

8. The channel state information feedback compression method based on a large-scale MIMO system as claimed in claim 6, characterized in that: The improved Kronecker sparse representation includes: A joint dictionary is constructed, wherein the joint dictionary is composed of a time dictionary and a space dictionary, wherein the time dictionary is generated by Hankel matrix decomposition.

9. The channel state information feedback compression method based on a large-scale MIMO system according to claim 6, characterized in that: The attention-enhanced autoencoder comprises: A spatial-frequency attention gate is introduced in the autoencoder.

10. The channel state information feedback compression method based on a large-scale MIMO system according to claim 4 or 6, characterized in that: The feedback process of the sub-matrix after optimization compression includes: According to the block size, the compression method and the quantization step size, a joint codebook of a triplet structure is obtained, and the compressed sub-matrix is ​​mapped to the joint codebook as an index in the codebook; The compressed sub-matrix is ​​reconstructed by the joint codebook and index, and the reconstructed error value is calculated. If the error value is greater than 0, feedback is triggered, and the reconstruction information is fed back to the temporal convolutional network prediction module to continuously optimize and predict the future dual-domain channel change rate quantization index and the compression process of the sub-matrix until the error value is 0.

Images