Channel state feedback method and device

Abstract

The invention provides a channel state feedback method and device. The method comprises the following steps: determining channel state information to be fed back; splicing the channel state information to be fed back with a preset channel state matrix to obtain a target channel state matrix; the preset channel state matrix is established based on multiple pieces of historical channel state information; extracting a first channel feature from the target channel state matrix; the first channel feature refers to a channel feature in a time domain dimension; extracting a second channel feature from the first channel feature; the second channel characteristics refer to channel characteristics in a space dimension and a frequency dimension; screening an index of a code word matched with the second channel characteristic from a predefined codebook, and sending the index to the base station; the predefined codebook comprises a plurality of code words, and each code word corresponds to a channel state. According to the embodiment of the invention, through joint optimization of multi-domain features, the purpose of realizing high-precision channel state feedback with relatively low feedback overhead can be achieved.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a channel state feedback method and device. Background Art

[0002] Massive Multiple Input and Multiple Output (Massive MIMO) technology is regarded as one of the core technologies of 5G and its evolution systems. It can effectively improve channel capacity and spectrum efficiency, increase data transmission rate and energy utilization.

[0003] In a large-scale MIMO system based on frequency division duplex (FDD), users need to feed back channel state information (CSI) to the base station through the uplink. However, as the size of the antenna array increases, directly feeding back the entire channel matrix will cause very serious feedback overhead, and it needs to be compressed to reduce the amount of data transmission. However, the compression-decompression process often brings data conversion errors, which reduces the accuracy of the reconstructed CSI and affects the overall performance of the system.

[0004] Therefore, how to achieve high-precision CSI feedback with lower feedback overhead has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of this, the present application proposes a channel state feedback method and device to solve the problem of high-precision CSI feedback overhead existing in the related art.

[0006] In a first aspect, an embodiment of the present application proposes a channel state feedback method, which is used for a terminal and includes:

[0007] Determining channel state information to be fed back;

[0008] The channel state information to be fed back is concatenated with a preset channel state matrix to obtain a target channel state matrix; the preset channel state matrix is ​​established based on a plurality of historical channel state information;

[0009] Extracting a first channel feature from the target channel state matrix; the first channel feature refers to a channel feature in the time domain dimension;

[0010] Extracting a second channel feature from the first channel feature; the second channel feature refers to a channel feature in a spatial dimension and a frequency dimension;

[0011] An index of a codeword matching the second channel feature is screened out from a predefined codebook, and the index is sent to a base station; the predefined codebook includes a plurality of codewords, and each codeword corresponds to a channel state.

[0012] The embodiment of the present application can fully utilize the spatial-frequency-time domain physical characteristics in the channel state information by extracting the first channel characteristics from the target channel state matrix and extracting the second channel characteristics from the first channel characteristics, and further improve the accuracy of CSI feedback and save feedback overhead through the joint optimization of multi-domain characteristics.

[0013] In the embodiment of the present application, determining the channel state information to be fed back includes:

[0014] Dividing a signal frequency band into a plurality of sub-bands according to a preset frequency interval; the signal frequency band refers to a frequency range of a signal;

[0015] Construct a channel matrix corresponding to each subband to obtain multiple channel matrices corresponding to the multiple subbands one by one; wherein each element h in each channel matrix ij Used to represent the transmission response of the signal from the jth transmitting antenna to the i-th receiving antenna on the corresponding subband;

[0016] For any channel matrix among the multiple channel matrices, calculating an eigenvector of the channel matrix;

[0017] The multiple eigenvectors of the multiple channel matrices are used as the channel state information to be fed back.

[0018] The embodiment of the present application uses the sub-band as the minimum feedback granularity and uses the eigenvector of the frequency domain channel autocorrelation matrix instead of the initial CSI matrix for feedback, which meets the standardization requirements and has more practical deployment value.

[0019] In an embodiment of the present application, the method further includes:

[0020] Setting the time window length and time interval; the time window length refers to the time span of each historical channel state information, and the time interval refers to the time interval between two consecutive channel state measurements;

[0021] A plurality of historical channel state information measured before the current time is stored in a buffer to form the preset channel state matrix.

[0022] In an embodiment of the present application, extracting a first channel feature from the target channel state matrix includes:

[0023] The target channel state matrix is ​​input into the time memory layer, and the first channel feature is output.

[0024] The embodiment of the present application can fully utilize the time correlation of the channel through the time memory layer, extract more compact physical features, and improve the computing efficiency and the speed of model convergence.

[0025] In the embodiment of the present application, extracting the second channel feature from the first channel feature includes:

[0026] Mapping the first channel feature into a plurality of channel features of a preset dimension;

[0027] Setting a plurality of attention heads, each attention head is used to calculate the attention weight of each channel feature in the plurality of channel features for other channel features;

[0028] Using the attention weight, weighted sum is performed on each channel feature to obtain a sub-channel feature;

[0029] The sub-channel features output by the multiple attention heads are concatenated to obtain the second channel features.

[0030] In the embodiment of the present application, after extracting the second channel feature from the first channel feature, the method further includes:

[0031] The second channel feature is quantized; the quantized second channel feature is used to filter out the index.

[0032] An embodiment of a second aspect of the present application provides a channel state feedback method, which is used for a base station and includes:

[0033] An index sent by a receiving terminal; the index is used to indicate the position of the target codeword in a predefined codebook;

[0034] Filtering a target codeword corresponding to the index from the predefined codebook, and determining a third channel feature corresponding to the target codeword; the third channel feature refers to a channel feature in a spatial dimension and a frequency dimension;

[0035] Converting the third channel feature into a corresponding fourth channel feature; the fourth signal feature refers to a channel feature in the time domain dimension;

[0036] Based on a preset channel state matrix, reconstructed channel state information is extracted from the fourth channel feature; the preset channel state matrix is ​​established based on multiple historical channel state information.

[0037] In the embodiment of the present application, before receiving the index sent by the user terminal, the method further includes:

[0038] A bit stream sent by a user terminal is received, and a dequantization process is performed on the bit stream to obtain the index.

[0039] An embodiment of a third aspect of the present application provides a channel state feedback device, the device comprising:

[0040] An information determination module, used to determine the channel state information to be fed back;

[0041] A target channel state matrix determination module, used for splicing the channel state information to be fed back with a preset channel state matrix to obtain a target channel state matrix; the preset channel state matrix is ​​established based on multiple historical channel state information;

[0042] A first channel feature extraction module, used to extract a first channel feature from the target channel state matrix; the first channel feature refers to a channel feature in the time domain dimension;

[0043] A second channel feature extraction module, used to extract a second channel feature from the first channel feature; the second channel feature refers to a channel feature in a spatial dimension and a frequency dimension;

[0044] An index screening module is used to screen out the index of the codeword matching the second channel feature from a predefined codebook and send the index to a base station; the predefined codebook includes multiple codewords, each codeword corresponds to a channel state.

[0045] An embodiment of a fourth aspect of the present application provides a channel state feedback device, the device comprising:

[0046] An index receiving module, used to receive an index sent by a user terminal; the index is used to indicate the position of a target codeword in a predefined codebook;

[0047] A third channel feature determination module, configured to filter out a target codeword corresponding to the index from the predefined codebook, and determine a third channel feature corresponding to the target codeword; the third channel feature refers to a channel feature in a spatial dimension and a frequency dimension;

[0048] A feature conversion module, used to convert the third channel feature into a corresponding fourth channel feature; the fourth signal feature refers to a channel feature in the time domain dimension;

[0049] The channel state information reconstruction module is used to extract the reconstructed channel state information from the fourth channel feature based on a preset channel state matrix; the preset channel state matrix is ​​established based on multiple historical channel state information.

[0050] An embodiment of the fifth aspect of the present application provides an electronic device, which includes a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the channel state feedback method described in the first aspect above by executing the computer instructions.

[0051] An embodiment of the sixth aspect of the present application provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to enable a computer to execute the channel state feedback method described in the first aspect above.

[0052] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] By reading the detailed description of the preferred embodiment below, various other advantages and benefits will become clear to those of ordinary skill in the art. The accompanying drawings are only used for the purpose of illustrating the preferred embodiment and are not considered to be limitations of the present application. In addition, the same reference symbols are used to represent the same components throughout the accompanying drawings.

[0054] In the attached picture:

[0055] Figure 1 A schematic diagram showing a flow chart of a channel state feedback method provided by an embodiment of the present application;

[0056] Figure 2 A schematic diagram of a process of extracting first channel features using a temporal memory layer provided in an embodiment of the present application is shown;

[0057] Figure 3 A schematic diagram showing a flow chart of another channel state feedback method provided by an embodiment of the present application;

[0058] Figure 4 A schematic diagram showing the structure of a channel state feedback device provided by an embodiment of the present application is shown;

[0059] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present application is shown;

[0060] Figure 6 A schematic diagram of a storage medium provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0061] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0062] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by technicians in the field to which this application belongs.

[0063] The following describes the technical scenarios involved in the embodiments of the present application.

[0064] Existing CSI feedback schemes are divided into three categories: codebook-based CSI feedback, compressed sensing-based CSI feedback, and deep learning-based CSI feedback. In the codebook-based CSI feedback defined by the 5G NR standard, users and base stations share a set of pre-designed codebooks, including Type I codebooks with lower accuracy and Type II codebooks with higher accuracy. When performing CSI feedback, the user first uses the codebook to encode its current channel state information into a string of codeword indexes, and feeds back the indexes to the base station through the uplink instead of all CSI. After receiving the index, the base station decodes it into the original CSI matrix through the shared codebook. This feedback method effectively compresses the amount of feedback information and achieves the purpose of saving overhead to a certain extent. However, the channel state information provided by this method is relatively rough. The base station needs to estimate and infer the channel state based on the pre-set codebook, which is difficult to adapt to complex channel environments and changes in application scenarios.

[0065] In view of the shortcomings of the above-mentioned schemes, compressed sensing (CS) is also used in the prior art for CSI feedback. CSI feedback based on compressed sensing uses the sparsity assumption or structured characteristics of the channel, and extracts key information from the global channel state through random sampling technology, and feeds it back to the transmitter with less feedback overhead. Through the compressed sensing reconstruction algorithm, the global channel state information is recovered, which can significantly reduce the feedback overhead. However, in actual situations, the channel is usually non-sparse, and even exhibits mathematical inexplicability, which seriously affects the applicability of these schemes. In addition, schemes based on compressed sensing usually involve complex signal processing and optimization, and most of them rely on iterative algorithms, which leads to increased computational complexity, reduced computational efficiency, and requires higher computing resources and time costs.

[0066] In recent years, artificial intelligence (AI) technology has developed rapidly, improving the performance, efficiency and intelligence level of communication systems. In order to solve the shortcomings of traditional codebook-based CSI feedback and compressed sensing-based CSI feedback solutions, attempts are made to introduce artificial intelligence into CSI feedback. Most of the existing AI-based CSI feedback solutions adopt an auto-encoder architecture, which consists of an encoder and a decoder, where the encoder is deployed on the user side and the decoder is deployed on the base station side. The encoder on the user side compresses the initial high-dimensional CSI matrix into a more compact and meaningful low-dimensional vector representation, and then feeds it back to the base station through the uplink. After receiving the low-dimensional vector, the decoder on the base station side decompresses it and restores it to the initial CSI matrix. The network models used in existing solutions include convolutional neural network (CNN), recurrent neural network (RNN) and transformer.

[0067] However, most of the above-mentioned deep learning-based CSI feedback schemes treat the initial CSI matrix as two-dimensional image data for processing, extract data features from the angle-delay domain, and realize the compression and recovery of the CSI matrix. Therefore, these schemes do not fully utilize all the physical characteristics of the channel itself from the perspective of wireless communication. In particular, most of these schemes are oriented to the channel state information at a single moment, without considering the time correlation between adjacent time slots in the time-varying channel, and their performance still has room for further improvement. In addition, the existing schemes assume that the uplink is ideal, that is, the decoder on the base station side can perfectly receive the low-dimensional vector sent by the encoder on the user side. However, in real situations, the data transmission between the user and the base station requires the introduction of a quantization layer. These schemes ignore the quantization problem and do not consider the quantization error.

[0068] In view of the above problems, there is an urgent need for a CSI feedback solution that takes into account all physical characteristics of the channel and fully considers the quantization error of data in the CSI feedback uplink.

[0069] The present invention aims to solve the problem that the existing CSI feedback scheme fails to fully utilize the implicit correlation of wireless channel characteristics, and designs a CSI feedback scheme that takes into account the multi-domain physical characteristics of the channel. The present invention adopts an autoencoder architecture, in which the encoder is deployed on the user side and the decoder is deployed on the base station side, and the two have a symmetrical structure. The present invention uses the transformer model as the backbone to extract the spatial-frequency domain characteristics of the channel. In particular, the present invention designs a time memory layer based on a convolutional long short-term memory network (Convolutional long short-term memory, ConvLSTM) to fully extract the time correlation of the channel. By jointly optimizing the spatial-frequency-time domain characteristics of the channel, the overhead of CSI feedback is further reduced and the accuracy of feedback is improved. At the same time, the present invention designs a learnable dynamic vector quantization layer to effectively reduce the quantization error in view of the problem that the existing scheme does not consider the uplink quantization error.

[0070] According to an embodiment of the present application, a channel state feedback method embodiment is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0071] In this embodiment, a channel state feedback method is provided, which can be used in a terminal (i.e., a user end); Figure 1 is a flow chart of a channel state feedback method according to an embodiment of the present application. Figure 1 As shown, the process includes the following steps:

[0072] Step S101: determining channel state information to be fed back.

[0073] In some specific embodiments, the above step S101 includes steps S1011 to S1014:

[0074] Step S1011: divide the signal frequency band into a plurality of sub-bands according to a preset frequency interval.

[0075] Specifically, the preset frequency interval can be set according to the situation and is not specifically limited here. The signal frequency band can be understood as the distribution range of the signal in the frequency domain, which describes the distribution characteristics of the signal in frequency. Subband refers to dividing a wider signal frequency band into several narrower subbands, each of which contains a part of the frequency range of the signal.

[0076] Step S1012: construct a channel matrix corresponding to each subband to obtain a plurality of channel matrices corresponding one-to-one to the plurality of subbands.

[0077] Specifically, when N sb When there are subbands, the channel matrix corresponding to each subband can be expressed as:

[0078]

[0079] Among them, N sb represents the number of subbands, and each subband corresponds to a unique channel matrix. Each element h in each channel matrix H ij It is used to represent the transmission response of the signal from the j-th transmitting antenna to the i-th receiving antenna in the corresponding subband. The transmission response includes but is not limited to path loss, phase change and multipath effect.

[0080] Step S1013: For any channel matrix among the multiple channel matrices, calculate an eigenvector of the channel matrix.

[0081] Specifically, the eigenvalue decomposition can be used to calculate the eigenvector w of each channel matrix H, thereby obtaining multiple eigenvectors corresponding to multiple channel matrices, which are expressed as:

[0082]

[0083] The eigenvalue decomposition can be performed using the following formula:

[0084]

[0085] Where, the superscript H represents the matrix transpose, the subscript k represents the kth subband, and λ k Represents the eigenvector corresponding to the channel matrix of the kth subband.

[0086] Step S1014: using the multiple eigenvectors of the multiple channel matrices as the channel state information to be fed back.

[0087] Step S102: concatenate the channel state information to be fed back with a preset channel state matrix to obtain a target channel state matrix.

[0088] Specifically, the preset channel state matrix is ​​established based on a plurality of historical channel state information. The current channel state information to be fed back w t Can be used with the historical CSI buffer (ie: w t-k 、w t-2k ... t-(n-1)k ) to perform matrix concatenation, for example Figure 2 shown.

[0089] In some specific embodiments, before the above step S102, the method further includes:

[0090] Step a1, setting the time window length and time interval.

[0091] Specifically, the time window length n refers to the time span k of each historical channel state information, and the time interval refers to the time interval between two consecutive channel state measurements.

[0092] Step a2: storing multiple historical channel state information measured before the current time into a buffer to form the preset channel state matrix.

[0093] Specifically, for example, historical channel state information of n-1 time intervals k before the current time may be stored in a buffer to form a historical CSI matrix.

[0094] Step S103: extracting a first channel feature from the target channel state matrix.

[0095] Specifically, the first channel feature refers to the channel feature in the time domain dimension. The target channel state matrix can be input into the time memory layer in the encoder to extract the channel feature based on time correlation, that is, the first channel feature mentioned above. This process can be implemented by the following formula:

[0096] w′ t =f CLe (w t ∣w t-k ,…,w t-(n-1)k )

[0097] Among them, f CLe It is the function of the time memory layer, which is used to extract the time domain characteristics of the channel, that is, the first channel characteristics mentioned above, which describes the characteristics of the signal changing over time. t represents the first channel feature of the current time slot, k represents the time interval, and n is the length of the time window (including the CSI of the current time slot and (n-1) historical CSIs); w t-k ,…,w t-(n-1)k Represents the historical channel state information, where the adjacent interval is k, and there are a total of (n-1).

[0098] More specifically, the specific structure of the above-mentioned time memory layer is as follows Figure 2 As shown, the matrices are concatenated to obtain the target channel state matrix, which is input into the convolutional long short-term memory network ConvLSTMBlockxN, and the output result is passed through the Flatten layer and the fully connected layer to obtain the above-mentioned first channel feature.

[0099] In some specific embodiments, the relevance and data dimension of the historical CSI in the buffer are adjusted according to the specific application scenarios of the model and different channel state change configurations, thereby dynamically adapting to the needs of different application scenarios.

[0100] For example, the value of the time interval k generally refers to the pilot setting of the system, which can be k=1, k=3, k=5, etc. When the user is in a low-speed motion scene, the channel correlation of adjacent time slots is strong, and the time window length can be set to n=3 to save computing resources and overhead; when the user is in a high-speed motion scene, the channel correlation of adjacent time slots is weak, and the time window length can be set to n=5 to obtain effective information from more historical data.

[0101] Step S104: extracting a second channel feature from the first channel feature.

[0102] Specifically, the second channel feature refers to a channel feature based on the space-frequency domain, which can be understood as a characteristic parameter that describes the channel characteristics in two dimensions, space and frequency.

[0103] In some specific embodiments, the second channel feature w can be extracted from the first channel feature by a deep learning model transformer. e , and use its second channel feature as the input of the learnable dynamic vector quantization layer, so that the learnable dynamic vector quantization layer quantizes the second channel feature. Among them, the model structure of the deep learning model transformer includes the Reshape layer, Embedding layer, multi-head attention module × N and full connection layer in the encoder, for example Figure 3 shown.

[0104] In some specific embodiments, the number of multi-head attention layers and attention heads is adjusted according to changes in the channel environment and the dimension of the CSI matrix to better adapt to the channel state, improve feedback performance and reduce model complexity.

[0105] For example: when the dimension of the channel matrix is ​​small or the quality is good and the correlation is strong, the number of attention layers can be set to 5 and the number of attention heads can be set to 8 to reduce the complexity of the model and save computing resources; when the dimension of the channel matrix is ​​large or the quality is poor and the correlation is weak, the number of attention layers can be set to 10 and the number of attention heads can be set to 16 to improve model performance.

[0106] In some specific embodiments, the above step S104 includes steps S1041 to S1044:

[0107] Step S1041: Map the first channel feature into multiple channel features of a preset dimension.

[0108] Specifically, for example, the preset dimension can be 512 dimensions, and the 64-dimensional channel features can be mapped to 512-dimensional channel features through the Embedding layer, so that the multi-head attention module can more fully learn the above 512-dimensional channel features.

[0109] Step S1042, setting multiple attention heads.

[0110] Specifically, each attention head is used to calculate the attention weight of each channel feature among the multiple channel features for other channel features.

[0111] Preferably, the attention weight of each channel feature for other channel features can be calculated in the following manner: by applying different weight matrices to each channel feature, three feature vectors of each channel feature, namely, a query vector, a key vector and a value vector, are obtained; then the query vector and the key vector of each channel feature are used to calculate the attention score, which can be achieved by calculating the dot product of the query vector and the key vector; finally, the attention score of each channel feature is normalized using the Softmax function to ensure that the sum of all scores is 1, thereby obtaining the attention weight of each channel feature for other channel features.

[0112] Step S1043: Use the attention weight to perform weighted summation on each channel feature to obtain a sub-channel feature.

[0113] Specifically, the normalized attention weights can be used to perform weighted summation on the value vector of each channel feature to obtain the final sub-channel feature. This can be expressed by the following formula:

[0114]

[0115] Among them, Q represents the query vector, K represents the key vector, V represents the value vector, and d k represents the dimension of the key vector, Represents the factor by which the dot product result is scaled.

[0116] Step S1044, concatenating the sub-channel features output by the multiple attention heads to obtain the second channel features.

[0117] In some specific embodiments, after the above step S104, the method further includes:

[0118] The second channel feature is quantized, for example Figure 3 On the encoder side, the compressed CSIw e Input to the quantization layer. The loss function of the quantization layer is as follows:

[0119]

[0120] Among them, sg(·) represents the stop gradient update operation, and the first term of the loss function Used to optimize the codebook space, the second It is used to optimize the output of the encoder on the user side, thereby effectively reducing the quantization error. β is the weight parameter that balances the above two items. e is the second channel feature and the output of the encoder, s k is a codeword that matches the second channel characteristic.

[0121] Step S105: Filter out the index of the codeword matching the second channel characteristic from the predefined codebook, and send the index to the base station.

[0122] Specifically, the predefined codebook includes multiple codewords, and each codeword corresponds to a channel state.

[0123] More specifically, after obtaining the index, it needs to be quantized into a bit stream, and the quantized bit stream is sent to the base station. The quantization process is as follows:

[0124]

[0125] Among them, s i is to find the output w of the encoder e The intermediate variable when the best matching codeword is found, s k is selected with w e The best matching codeword is taken as the quantization result.

[0126] The embodiment of the present application also provides a channel state feedback method, which is used in a base station and includes:

[0127] Step S201: receiving an index sent by a terminal.

[0128] Specifically, the index is used to indicate the position of the target codeword in the predefined codebook.

[0129] Before step S201, the method further includes: after receiving the bit stream, the decoder of the base station performs dequantization processing on the bit stream to obtain a corresponding index.

[0130] Step S202: Filter out a target codeword corresponding to the index from the predefined codebook, and determine a third channel feature corresponding to the target codeword.

[0131] Specifically, the third channel characteristics refer to channel characteristics in spatial dimension and frequency dimension.

[0132] More specifically, if Figure 3 As shown, at the base station side, after the bit stream is dequantized to obtain an index, the target codeword s corresponding to the index is determined through the same codebook space as the encoder at the user end. k , and obtain the channel feature w corresponding to the target codeword through the inverse quantization layer q; and input it into the transformer in the decoder, recover the CSI from the spatial-frequency domain dimension, and output the third channel feature w based on the spatial-frequency domain recovery d , and input it into the temporal memory layer.

[0133] Step S203: convert the third channel characteristic into a corresponding fourth channel characteristic.

[0134] Specifically, the fourth signal feature refers to a channel feature in the time domain dimension. Figure 3 As shown, the third channel feature can be converted into the corresponding fourth channel feature through the time memory layer in the decoder.

[0135] Step S204: extracting reconstructed channel state information from the fourth channel feature based on a preset channel state matrix; the preset channel state matrix is ​​established based on a plurality of historical channel state information.

[0136] Specifically, based on the time window length n and the time slot interval k, the base station concatenates the CSI recovered based on the space-frequency domain with the historical CSI matrix in the buffer (i.e., the preset channel state matrix mentioned above), and calculates w from the dimension of time correlation. d Further reconstruction is performed to obtain the reconstructed CSI matrix. This process can be expressed as:

[0137]

[0138] Among them, f CLd is the function of the temporal memory layer in the decoder, It is the reconstructed CSI matrix and the final output of the entire architecture.

[0139] More specifically, the square of the generalized cosine similarity between the reconstructed CSI matrix and the CSI to be fed back at the current moment is also recorded as ρ, which is used as the optimization target of the overall architecture, and the calculation formula is:

[0140]

[0141] Among them, N sa and N sb are the number of samples and the number of subbands respectively. The loss function of the overall architecture is designed as L = L r +L q , where L r =1-ρ is used to optimize the reconstruction accuracy of CSI and reduce the feedback error, and Lq is used to optimize the dynamic vector quantization layer and reduce the quantization error.

[0142] Compared with the prior art, the embodiments of the present application have the following technical effects:

[0143] 1. From the perspective of wireless communication, the spatial-frequency-time domain physical characteristics of the channel state information are fully utilized, and the accuracy of CSI feedback is further improved through joint optimization of multi-domain characteristics, saving feedback overhead.

[0144] 2. Using sub-band as the minimum feedback granularity, the eigenvector of the frequency domain channel autocorrelation matrix is ​​used instead of the initial CSI matrix for feedback, which meets the standardization requirements and has more practical deployment value.

[0145] 3. Selecting the transformer model as the backbone of the architecture can more effectively extract channel features based on the spatial-frequency domain from the channel state information and improve the performance of the model.

[0146] 4. A time memory layer based on the ConvLSTM model is designed to make full use of the temporal correlation of the channel, extract more compact physical features, and improve computational efficiency and model convergence speed.

[0147] 5. According to different channel conditions and the dimensions of the CSI matrix, the configuration can be flexibly adjusted, including: the number of feedback bits, the transformer's multi-head attention layer and the number of attention heads, the number of ConvLSTM modules in the time memory layer, the time window length n, and the time slot interval k between adjacent CSIs, etc. This can reduce the complexity of the model and the overhead while ensuring the performance in different scenarios.

[0148] 6. Taking full account of the quantization error in the uplink, a learnable dynamic vector quantization layer is introduced. The codebook space can be dynamically adjusted to adapt to data changes of different CSIs, effectively reducing the quantization error in the uplink and improving the accuracy of quantization.

[0149] Corresponding to the implementation of the above channel state feedback method, the embodiment of the present application also provides a channel state feedback device for performing the above Figures 1 to 3 The channel state feedback method described in any of the embodiments shown in the figure. Figure 4 As shown, the channel state feedback device includes:

[0150] An information determination module, used to determine the channel state information to be fed back;

[0151] A target channel state matrix determination module, used for splicing the channel state information to be fed back with a preset channel state matrix to obtain a target channel state matrix; the preset channel state matrix is ​​established based on multiple historical channel state information;

[0152] A first channel feature extraction module, used to extract a first channel feature from the target channel state matrix; the first channel feature refers to a channel feature in the time domain dimension;

[0153] A second channel feature extraction module, used to extract a second channel feature from the first channel feature; the second channel feature refers to a channel feature in a spatial dimension and a frequency dimension;

[0154] An index screening module is used to screen out the index of the codeword matching the second channel feature from a predefined codebook and send the index to a base station; the predefined codebook includes multiple codewords, each codeword corresponds to a channel state.

[0155] Optionally, the information determination module is further used to: divide the signal frequency band into a plurality of sub-bands according to a preset frequency interval; the signal frequency band refers to the frequency range of the signal; construct a channel matrix corresponding to each sub-band, and obtain a plurality of channel matrices corresponding to the plurality of sub-bands one by one; wherein each element h in each channel matrix ij Used to represent the transmission response of the signal from the jth transmitting antenna to the ith receiving antenna on the corresponding subband; for any channel matrix among the multiple channel matrices, calculate the eigenvector of the channel matrix; and use the multiple eigenvectors of the multiple channel matrices as the channel state information to be fed back.

[0156] Optionally, the device also includes: a preset channel state matrix forming module, which is used to set the time window length and time interval; the time window length refers to the time span of each historical channel state information, and the time interval refers to the time interval between two consecutive channel state measurements; multiple historical channel state information measured before the current time is stored in a buffer to form the preset channel state matrix.

[0157] Optionally, the first channel feature extraction module is further used to: input the target channel state matrix into the time memory layer, and output the first channel feature.

[0158] Optionally, the second channel feature extraction module is also used to: map the first channel feature into multiple channel features of a preset dimension; set multiple attention heads, each attention head is used to calculate the attention weight of each channel feature in the multiple channel features for other channel features; use the attention weight to perform weighted summation on each channel feature to obtain a sub-channel feature; and splice the sub-channel features output by the multiple attention heads to obtain the second channel feature.

[0159] Optionally, the device also includes a quantization module used to: perform quantization processing on the second channel characteristics; and the quantized second channel characteristics are used to filter out the index.

[0160] The channel state feedback device provided in the above-mentioned embodiment of the present application and the channel state feedback method provided in the embodiment of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0161] The present application also provides an electronic device to perform the above-mentioned channel state feedback method. Figure 5 , which shows a schematic diagram of an electronic device provided by some embodiments of the present application. Figure 5 As shown, the electronic device 5 includes: a processor 500, a memory 501, a bus 502 and a communication interface 503, wherein the processor 500, the communication interface 503 and the memory 501 are connected via the bus 502; the memory 501 stores a computer program that can be run on the processor 500, and the processor 500 executes the aforementioned Figures 1 to 3 A channel state feedback method is provided in any of the illustrated embodiments.

[0162] The memory 501 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 503 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0163] The bus 502 may be an ISA bus, a PCI bus, or an EISA bus. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory 501 is used to store programs. After receiving the execution instruction, the processor 500 executes the program. Figures 1 to 3 The channel state feedback method disclosed in any of the illustrated embodiments may be applied to the processor 500 or implemented by the processor 500 .

[0164] The processor 500 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 500. The above processor 500 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor can be combined and executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 501, and the processor 500 reads the information in the memory 501 and completes the steps of the above method in combination with its hardware.

[0165] The electronic device provided in the embodiment of the present application and the channel state feedback method provided in the embodiment of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, operated or implemented therein.

[0166] The present application also provides a computer-readable storage medium corresponding to the channel state feedback method provided in the above embodiment. Figure 6 The computer-readable storage medium shown is a CD 30 on which a computer program (ie, a program product) is stored. When the computer program is run by a processor, the channel state feedback method provided in any of the aforementioned embodiments will be executed.

[0167] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical or magnetic storage media, which are not listed here one by one.

[0168] The computer-readable storage medium provided in the above-mentioned embodiments of the present application and the channel state feedback method provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0169] It should be noted that:

[0170] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known structures and technologies are not shown in detail so as not to obscure the understanding of this description.

[0171] Similarly, it should be understood that in order to streamline the present application and help understand one or more of the various inventive aspects, in the above description of the exemplary embodiments of the present application, the various features of the present application are sometimes grouped together into a single embodiment, figure, or description thereof. However, the disclosed method should not be interpreted as reflecting the following: the claimed application requires more features than the features clearly stated in each claim. More specifically, as reflected in the claims below, the inventive aspects are less than all the features of the single embodiment disclosed above. Therefore, the claims following the specific embodiment are hereby expressly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the present application.

[0172] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present application and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

[0173] The above is only a preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A channel state feedback method, the method being used in a terminal, characterized in that: The method comprises: Determining channel state information to be fed back; The channel state information to be fed back is concatenated with a preset channel state matrix to obtain a target channel state matrix; the preset channel state matrix is ​​established based on a plurality of historical channel state information; Extracting a first channel feature from the target channel state matrix; the first channel feature refers to a channel feature in the time domain dimension; Extracting a second channel feature from the first channel feature; the second channel feature refers to a channel feature in a spatial dimension and a frequency dimension; An index of a codeword matching the second channel feature is screened out from a predefined codebook, and the index is sent to a base station; the predefined codebook includes a plurality of codewords, and each codeword corresponds to a channel state.

2. The method according to claim 1, determining the channel state information to be fed back comprises: Dividing a signal frequency band into a plurality of sub-bands according to a preset frequency interval; the signal frequency band refers to a frequency range of a signal; Construct a channel matrix corresponding to each subband to obtain multiple channel matrices corresponding to the multiple subbands one by one; wherein each element h in each channel matrix ij Used to represent the transmission response of the signal from the jth transmitting antenna to the i-th receiving antenna on the corresponding subband; For any channel matrix among the multiple channel matrices, calculating an eigenvector of the channel matrix; The multiple eigenvectors of the multiple channel matrices are used as the channel state information to be fed back.

3. The method according to claim 1 or 2, further comprising: Setting the time window length and time interval; the time window length refers to the time span of each historical channel state information, and the time interval refers to the time interval between two consecutive channel state measurements; A plurality of historical channel state information measured before the current time is stored in a buffer to form the preset channel state matrix.

4. The method according to claim 1 or 2, characterized in that: Extracting a first channel feature from the target channel state matrix includes: The target channel state matrix is ​​input into the time memory layer, and the first channel feature is output.

5. The method according to claim 1 or 2, characterized in that: Extracting a second channel feature from the first channel feature includes: Mapping the first channel feature into a plurality of channel features of a preset dimension; Setting a plurality of attention heads, each attention head is used to calculate the attention weight of each channel feature in the plurality of channel features for other channel features; Using the attention weight, weighted sum is performed on each channel feature to obtain a sub-channel feature; The sub-channel features output by the multiple attention heads are concatenated to obtain the second channel features.

6. The method according to claim 1 or 2, after extracting the second channel feature from the first channel feature, the method further comprises: performing quantization processing on the second channel characteristic; The quantized second channel feature is used to filter out the index.

7. A channel state feedback method, the method being used in a base station, characterized in that: The method comprises: An index sent by a receiving terminal; the index is used to indicate the position of the target codeword in a predefined codebook; Filtering a target codeword corresponding to the index from the predefined codebook, and determining a third channel feature corresponding to the target codeword; the third channel feature refers to a channel feature in a spatial dimension and a frequency dimension; Converting the third channel feature into a corresponding fourth channel feature; the fourth signal feature refers to a channel feature in the time domain dimension; Based on a preset channel state matrix, reconstructed channel state information is extracted from the fourth channel feature; the preset channel state matrix is ​​established based on multiple historical channel state information.

8. The method according to claim 7, before receiving the index sent by the user terminal, the method further comprises: A bit stream sent by a user terminal is received, and a dequantization process is performed on the bit stream to obtain the index.

9. A channel state feedback device, characterized in that: The device comprises: An information determination module, used to determine the channel state information to be fed back; A target channel state matrix determination module, used for splicing the channel state information to be fed back with a preset channel state matrix to obtain a target channel state matrix; the preset channel state matrix is ​​established based on a plurality of historical channel state information; A first channel feature extraction module, used to extract a first channel feature from the target channel state matrix; the first channel feature refers to a channel feature in the time domain dimension; A second channel feature extraction module, used to extract a second channel feature from the first channel feature; the second channel feature refers to a channel feature in a spatial dimension and a frequency dimension; An index screening module is used to screen out the index of the codeword matching the second channel feature from a predefined codebook and send the index to a base station; the predefined codebook includes multiple codewords, each codeword corresponds to a channel state.

10. A channel state feedback device, characterized in that: The device comprises: An index receiving module, used to receive an index sent by a user terminal; the index is used to indicate the position of a target codeword in a predefined codebook; A third channel feature determination module, configured to filter out a target codeword corresponding to the index from the predefined codebook, and determine a third channel feature corresponding to the target codeword; the third channel feature refers to a channel feature in a spatial dimension and a frequency dimension; A feature conversion module, used to convert the third channel feature into a corresponding fourth channel feature; the fourth signal feature refers to a channel feature in the time domain dimension; The channel state information reconstruction module is used to extract the reconstructed channel state information from the fourth channel feature based on a preset channel state matrix; the preset channel state matrix is ​​established based on multiple historical channel state information.

Images