Information feedback method and apparatus, user equipment, base station, system model, and storage medium

Abstract

The present disclosure provides an information feedback method and device, user equipment, base station, system model and storage medium. The method comprises: obtaining a channel state information (CSI) matrix; screening elements in the CSI matrix based on self-information of the CSI matrix to obtain a sparse CSI matrix; determining feedback information based on elements in the sparse CSI matrix, and sending the feedback information to a base station. The method provided by the present disclosure has high accuracy and low cost in the reconstruction of the matrix by the base station.

Description

Technical Field

[0001] This disclosure relates to the field of communication technology, and in particular to an information feedback method, apparatus, user equipment, base station, system model and storage medium. Background Technology

[0002] Because m-MIMO (massive multiple-input multiple-output) technology offers good stability, energy efficiency, and interference resistance, it is commonly used for wireless communication. In m-MIMO systems, the User Equipment (UE) typically needs to feed back a downlink CSI (Channel State Information) matrix to the base station so that the base station can determine the downlink channel quality. However, due to the large number of antennas at the base station in m-MIMO systems, the number of corresponding downlinks is also large, resulting in significant overhead for feeding back the CSI matrix. Therefore, a low-overhead information feedback method is urgently needed.

[0003] In related technologies, the methods for UE feedback CSI information matrix mainly include:

[0004] Method 1: The CSI information matrix is ​​quantized into a certain number of bits using the AoD (Angle-of-departure) adaptive subspace codebook, and fed back to the base station, so that the base station can reconstruct the CSI information matrix based on the certain number of bits.

[0005] Method 2: Transform the CSI information matrix into a sparse matrix in the channel space, and perform random compression sampling on the sparse matrix to obtain low-dimensional measurement values, which are then fed back to the base station so that the base station can reconstruct the CSI information matrix based on these low-dimensional measurement values.

[0006] Method 3: Feed back the CSI information matrix to the base station using a neural network model based on DL (Deep Learning).

[0007] Among these methods, Method 1 still incurs significant overhead when the number of bits required to transform the CSI information matrix is ​​large. Method 2 uses random sampling to obtain low-dimensional measurements, failing to consider the structural characteristics of the CSI information matrix. This results in the obtained low-dimensional measurements not accurately reflecting the original CSI information matrix, leading to a large difference between the CSI information matrix reconstructed by the base station and the original CSI information matrix transmitted by the UE, resulting in low base station reconstruction accuracy. Method 3 suffers from high complexity in training the neural network model and slow convergence, also resulting in low base station reconstruction accuracy. Summary of the Invention

[0008] The information feedback method, apparatus, user equipment, base station, system model, and storage medium disclosed herein aim to solve the technical problems of low base station reconstruction accuracy and high overhead in existing information feedback methods.

[0009] The information feedback method proposed in the first aspect of this disclosure, applied to a UE, includes:

[0010] Obtain the Channel State Information (CSI) matrix;

[0011] The elements in the CSI information matrix are filtered based on the self-information of the CSI information matrix to obtain a sparse CSI information matrix.

[0012] Feedback information is determined based on the elements in the sparse CSI information matrix, and the feedback information is sent to the base station.

[0013] The information feedback method proposed in the second aspect of this disclosure, applied to a base station, includes:

[0014] Obtain feedback information sent by the UE, and determine the preliminary CSI information matrix based on the feedback information;

[0015] The CSI information matrix is ​​reconstructed based on the prepared CSI information matrix.

[0016] The information feedback device proposed in the third aspect embodiment of this disclosure includes:

[0017] The first acquisition module is used to acquire the Channel State Information (CSI) matrix.

[0018] The filtering module is used to filter the elements in the CSI information matrix based on the self-information of the CSI information matrix to obtain a sparse CSI information matrix.

[0019] A feature encoder is used to determine feedback information based on elements in the sparse CSI information matrix and send the feedback information to the base station.

[0020] The information feedback device proposed in the fourth aspect embodiment of this disclosure includes:

[0021] The second acquisition module is used to acquire feedback information sent by the UE and determine a preliminary CSI information matrix based on the feedback information;

[0022] The reconstruction module is used to reconstruct the CSI information matrix based on the prepared CSI information matrix.

[0023] A user equipment according to a fifth aspect embodiment of this disclosure includes: a transceiver; a memory; and a processor connected to the transceiver and the memory respectively, configured to control the transmission and reception of wireless signals of the transceiver by executing computer-executable instructions on the memory, and capable of implementing the method proposed in the first aspect embodiment above.

[0024] A base station according to a sixth aspect embodiment of this disclosure includes: a transceiver; a memory; and a processor connected to the transceiver and the memory respectively, configured to control the wireless signal transmission and reception of the transceiver by executing computer-executable instructions on the memory, and capable of implementing the method proposed in the second aspect embodiment above.

[0025] A seventh aspect of this disclosure provides an information feedback system model, which includes at least the information feedback devices described in the third and fourth aspects above.

[0026] An eighth aspect embodiment of this disclosure provides a method for training an information feedback system model as described in the seventh aspect, comprising: training the information feedback system model by performing the methods described in the first and / or second aspects above.

[0027] In another aspect of this disclosure, a computer storage medium is provided, wherein the computer storage medium stores computer-executable instructions; the computer-executable instructions, when executed by a processor, can implement the method described above.

[0028] In summary, in the information feedback method provided in this embodiment, the UE filters the elements in the CSI information matrix to obtain a sparse CSI information matrix. Specifically, when determining the sparse CSI information matrix, the UE first divides the CSI image information matrix corresponding to the CSI information matrix into multiple sub-image information matrices. Then, it determines the distribution function corresponding to each sub-image information matrix based on each sub-image information matrix and its neighboring sub-image information matrices. Next, it determines the self-information corresponding to each sub-image information matrix based on the distribution function. Finally, it filters redundant information based on the self-information to determine the sparse CSI information matrix. Furthermore, the UE determines feedback information based on the elements in the determined sparse CSI information matrix and sends the feedback information to the base station, so that the base station can reconstruct the CSI information matrix based on the feedback information.

[0029] Therefore, in this embodiment of the present disclosure, when determining the distribution function, the structural correlation between each sub-image matrix and its neighboring sub-image matrices is taken into account. Thus, when the sparse CSI information matrix is ​​subsequently determined based on the distribution function, the structure of the original CSI information matrix will not be destroyed. As a result, when the base station reconstructs the CSI information matrix using the feedback information determined from the sparse CSI information matrix, the accuracy of the reconstructed CSI information matrix can be ensured.

[0030] Meanwhile, since the sparse CSI information matrix in this embodiment is a matrix after redundant information has been removed, the ease of subsequent operations on the sparse CSI information matrix can be ensured, and the overhead is reduced.

[0031] Furthermore, in this embodiment, the feedback information determined by the UE includes the position information of the compressed elements in the sparse CSI information matrix. As a result, the base station can accurately reconstruct the CSI information matrix based on the position information of the compressed elements in the sparse CSI information matrix, thereby further ensuring the accuracy of the CSI information matrix reconstruction.

[0032] Additional aspects and advantages of this disclosure will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this disclosure. Attached Figure Description

[0033] The above and / or additional aspects and advantages of this disclosure will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, in which:

[0034] Figure 1 This is a flowchart illustrating an information feedback method provided in an embodiment of the present disclosure.

[0035] Figure 2 This is a schematic diagram of a process for obtaining a sparse CSI information matrix according to an embodiment of the present disclosure;

[0036] Figure 3 This is a schematic diagram of a process for determining the self-information matrix corresponding to the CSI image information matrix using a second convolutional layer, as provided in an embodiment of this disclosure.

[0037] Figure 4 This is a schematic diagram of the process for determining feedback information provided in one embodiment of this disclosure;

[0038] Figure 5 This is a flowchart illustrating an information feedback method provided in an embodiment of the present disclosure.

[0039] Figure 6 This is a schematic diagram of the structure of an information feedback device provided in an embodiment of the present disclosure;

[0040] Figure 7 This is a schematic diagram of the structure of an information feedback device provided in an embodiment of the present disclosure;

[0041] Figure 8 This is a schematic diagram of the structure of an information feedback system model provided in an embodiment of the present disclosure;

[0042] Figure 9 This is a block diagram of a user equipment provided in one embodiment of the present disclosure;

[0043] Figure 10 This is a block diagram of a base station provided in one embodiment of the present disclosure. Detailed Implementation

[0044] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this disclosure as detailed in the appended claims.

[0045] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the present disclosure. The singular forms “a” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0046] It should be understood that although the terms first, second, third, etc., may be used to describe various information in embodiments of this disclosure, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of embodiments of this disclosure, and similarly, second information may also be referred to as first information. Depending on the context, the words “if” and “suppose” as used herein may be interpreted as “when”, “when”, or “in response to a determination”.

[0047] Embodiments of this disclosure are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting this disclosure.

[0048] The information feedback method, apparatus, user equipment, base station, system model, and storage medium provided in this disclosure are described in detail below with reference to the accompanying drawings.

[0049] Figure 1 This is a flowchart illustrating an information feedback method provided in an embodiment of this disclosure, applied to a UE, such as... Figure 1 As shown, this information feedback method may include the following steps:

[0050] Step 101: Obtain the CSI information matrix.

[0051] It should be noted that the indication method of this disclosure can be applied to any UE. The UE can be a device that provides voice and / or data connectivity to a user. The UE can communicate with one or more core networks via a RAN (Radio Access Network). The UE can be an IoT terminal, such as a sensor device, a mobile phone (or "cellular" phone), and a computer with an IoT terminal. For example, it can be a fixed, portable, pocket-sized, handheld, computer-embedded, or vehicle-mounted device. Examples include a station (STA), subscriber unit, subscriber station, mobile station, mobile station, remote station, access point, remote terminal, access terminal, user terminal, or user agent. Alternatively, the UE can also be a device of an unmanned aerial vehicle. Alternatively, the UE can also be a vehicle-mounted device, such as a vehicle computer with wireless communication capabilities, or a wireless terminal connected to an external vehicle computer. Alternatively, the UE can also be a roadside device, such as a street light, traffic light, or other roadside device with wireless communication capabilities.

[0052] It should be noted that, in one embodiment of this disclosure, the UE can be configured with a single antenna, while the base station can be configured with multiple antennas. In one embodiment, the antennas at the base station can be arranged in a ULA (Uniform Linear Array) configuration; for example, N antennas can be configured at half-wavelength intervals. t = 32 antennas.

[0053] Based on this, in one embodiment of this disclosure, the UE obtaining the CSI information matrix may include: the UE obtaining the CSI information matrix corresponding to each antenna channel of the base station.

[0054] In another embodiment of this disclosure, the method for the UE to obtain the CSI information matrix may include: the UE determining the original CSI information matrix based on the pilot information sent by the base station, and then transforming the obtained original CSI information matrix from the spatial frequency domain to the angle delay domain to obtain the CSI information matrix in the angle delay domain.

[0055] In one embodiment of this disclosure, the UE can utilize a two-dimensional discrete Fourier transform to transform the original CSI information matrix. Transform from the spatial frequency domain to the angle time delay domain, where the CSI information matrix in the angle time delay domain... F d and for N t ×N t The corresponding discrete Fourier transform matrix, This indicates the number of antennas configured in the base station. For example, m-MIMO systems use OFDM (Orthogonal Frequency Division Multiplexing) technology. And, N t Indicates the number of subcarriers corresponding to each antenna number, for example, N t =32, where the superscript H denotes the conjugate transpose of the matrix. Furthermore, in one embodiment of this disclosure, the size of the CSI information matrix H in the angular time delay domain can be d1×N. c ×N t Where d1 is the number of antenna channels, and N c The number of antennas configured for the base station. Also, the CSI information matrix H in the angular delay domain may include a real part matrix H. re and the imaginary part matrix H im .

[0056] It should be noted that, in one embodiment of this disclosure, the subsequent processing of the CSI information matrix H in the angle time delay domain (such as subsequent filtering, compression quantization, padding, and reconstruction steps) specifically involves processing the real part matrix H of the CSI information matrix H. re and the imaginary part matrix H im To be processed.

[0057] In some embodiments, the above-described scheme may further include:

[0058] Step 102: Filter the elements in the CSI information matrix based on the self-information of the CSI information matrix to obtain a sparse CSI information matrix.

[0059] It should be noted that step 101 can be implemented alone or together with step 102. That is, any device can execute only step 101 to obtain the CSI information matrix. Any device can also execute steps 101 and 102 to obtain the CSI information matrix and obtain a sparse CSI information matrix based on the CSI information matrix.

[0060] in, Figure 2 This is a schematic diagram of a process for obtaining a sparse CSI information matrix according to an embodiment of the present disclosure, as shown below. Figure 2 As shown, the method may include:

[0061] Step 201: Map the CSI information matrix to a CSI image information matrix.

[0062] In one embodiment of this disclosure, the CSI information matrix can be obtained by any appropriate means; that is, the CSI information matrix can be obtained in the same manner as in step 101 described above, or by any other feasible means.

[0063] In one embodiment of this disclosure, mapping a CSI information matrix to a CSI image information matrix may include: converting the real part Hre of the CSI information matrix H into a CSI image information matrix corresponding to the real part Hre, and converting the imaginary part Him of the CSI information matrix H into a CSI image information matrix corresponding to the imaginary part Him.

[0064] Step 202: Use the first convolutional layer to perform feature processing on the CSI image information matrix to obtain n first feature maps, where n is an integer and n≥1.

[0065] In one embodiment of this disclosure, the first convolutional layer can be a convolutional layer with gradient updates. Also, as an example, in one embodiment of this disclosure, the kernel size of the first convolutional layer can be n×c×s×s, and the stride can be L, where L < s, and n, c, s, and L are all positive integers. As an example, in one embodiment of this disclosure, the kernel size of the first convolutional layer can be 64×2×3×3, and the stride can be 1.

[0066] Furthermore, in one embodiment of this disclosure, the n first feature maps can be n-dimensional feature maps corresponding to the CSI image information matrix. For example, in one embodiment of this disclosure, when n equals 64, it means that the CSI image information matrix is ​​processed using a first convolutional layer to obtain a 64-dimensional first feature map corresponding to the CSI image information matrix.

[0067] Step 203: Use the second convolutional layer to determine the self-information matrix corresponding to the CSI image information matrix. The size of the self-information matrix is ​​the same as the size of the CSI information matrix.

[0068] in, Figure 3 This is a schematic diagram of a process for determining the self-information matrix corresponding to the CSI image information matrix using a second convolutional layer, as provided in one embodiment of this disclosure. Figure 3 As shown, the method may include

[0069] Step 301: Add 0 to the outer perimeter of the CSI image information matrix to expand the CSI image information matrix and divide the expanded CSI image information matrix into m sub-image matrices; where m is an integer, m≥2, and m is the same as the number of elements included in the CSI image information matrix.

[0070] In one embodiment of this disclosure, the CSI image information matrix can be obtained by any appropriate means; that is, the CSI image information matrix can be obtained in the same manner as in step 201 described above, or by any other feasible means.

[0071] In one embodiment of this disclosure, the method for dividing the extended CSI image information matrix into m sub-image matrices may include: dividing the extended CSI image information matrix using an x×s grid, where x and s are both positive integers, and x and s may be the same or different. For example, in one embodiment of this disclosure, the extended CSI image information matrix may be divided using a 7×7 grid.

[0072] Step 302: Calculate the distribution function of each sub-image matrix using the second convolutional layer based on each sub-image matrix and its neighboring sub-image matrices. Calculate the self-information of each sub-image matrix based on the distribution function, and form a self-information matrix based on the self-information of each sub-image matrix.

[0073] In one embodiment of this disclosure, the second convolutional layer may be a convolutional layer without gradient updates.

[0074] Furthermore, in one embodiment of this disclosure, the method for calculating the distribution function of each sub-image matrix using the second convolutional layer based on each sub-image matrix and the neighboring sub-image matrices of each sub-image matrix may include the following steps:

[0075] Step a: Select the i-th sub-image matrix p input to the second convolutional layer. i , where i is an integer, i≥1.

[0076] Step b, determine p iThe nearest sub-image matrix p i ', get p i The corresponding set of neighboring sub-images Z i .

[0077] In one embodiment of this disclosure, the set of neighboring sub-images Z can be determined based on the principle of identical distribution of proximity and the Manhattan radius R. i Specifically, let's assume p i Its neighboring sub-image matrix p i If they come from the same distribution, then the set of neighboring sub-images Z i This can include: p i (2R+1) inside a circle centered at Manhattan radius R. 2 The nearest sub-image matrix p i '.

[0078] Step c: Determine p using Formula 1 i The distribution function f i (p i Formula 1 can include:

[0079]

[0080] in, h represents p i With p i The bandwidth between '.

[0081] As can be seen from the above steps ac, when calculating the distribution function of each sub-image matrix in this embodiment, the calculation is based on the sub-image matrix and its neighboring sub-image matrices. This allows the structural correlation between each sub-image matrix to be taken into account, that is, the structural correlation between each element in the CSI image information matrix is ​​taken into account. Therefore, when the CSI image information matrix is ​​subsequently compressed and quantized based on the distribution function to obtain compressed codewords, the structural correlation of the CSI information matrix can be ensured not to be destroyed. Thus, when the base station accurately reconstructs the CSI information matrix based on the compressed codewords, the reconstruction accuracy is ensured.

[0082] Furthermore, in one embodiment of this disclosure, calculating the self-information of each sub-image matrix based on the distribution function may include:

[0083] Using Formula 2, the distribution function f based on the i-th sub-image matrix i (p i Calculate the self-information I of the i-th sub-image matrix. i (p i Formula 2 includes:

[0084] I i (p i) = -log(f i (p i )).

[0085] The self-information of each sub-image matrix can be determined using the above method. Then, a self-information matrix can be constructed based on the self-information of each sub-image matrix. In one embodiment of this disclosure, the method for constructing a self-information matrix based on the self-information of each sub-image matrix may include...

[0086] Step 1: Create a first empty matrix, the size of which is the same as the size of the CSI image information matrix.

[0087] Step 2: Fill the first empty matrix with the self-information corresponding to each sub-image matrix according to the position of each sub-image matrix in the expanded CSI image information matrix to form a self-information matrix.

[0088] It should be noted that, in one embodiment of this disclosure, step 301 divides the expanded CSI image information matrix into m sub-image matrices, where m is the same as the number of elements included in the CSI image information matrix. Furthermore, since each sub-image matrix corresponds to a self-information quantity, the number of self-information quantities determined in step 302 is also the same as the number of elements included in the CSI image information matrix. Based on this, assuming the size of the first empty matrix is ​​the same as the size of the CSI image information matrix, the number of self-information quantities should correspond one-to-one with the number of empty positions in the first empty matrix. Therefore, according to the position of each sub-image matrix in the expanded CSI image information matrix, the self-information quantities corresponding to each sub-image matrix can be filled into the first empty matrix one-to-one, forming a self-information quantity matrix.

[0089] For example, in one embodiment of this disclosure, the method of filling the first empty matrix with the self-information corresponding to each sub-image matrix according to the position of each sub-image matrix in the expanded CSI image information matrix may include:

[0090] If a certain sub-image matrix is ​​located in the W-th row and Q-th column of the expanded CSI image information matrix, then the self-information corresponding to the sub-image matrix can be filled into the W-th row and Q-th column of the first empty matrix.

[0091] Step 204: Replace the elements in the self-information matrix whose values ​​are less than a preset threshold with 0 to obtain a sparse self-information matrix, and determine the position information of the non-zero elements in the sparse self-information matrix.

[0092] In one embodiment of this disclosure, the preset threshold can be preset.

[0093] Furthermore, in one embodiment of this disclosure, the method for determining elements in the self-information matrix whose values ​​are less than a preset threshold may include:

[0094] Construct a decay function δ(I) i (p i This attenuation coefficient is inversely proportional to the self-information, and the elements of the self-information matrix set to 0 are determined based on this attenuation function.

[0095] In one embodiment of this disclosure, the attenuation coefficient may follow a Boltzmann distribution, and the attenuation function may be, for example:

[0096]

[0097] T can be seen as a soft threshold. When the value of T is small, the elements with very small values ​​in the self-information matrix are set to 0. When the value of T approaches infinity, the elements in the self-information matrix are all set to 0 with equal probability, that is, randomly set to 0.

[0098] In one embodiment of this disclosure, the value of T in the attenuation function should be relatively small.

[0099] Furthermore, in one embodiment of this disclosure, based on the inverse relationship between the attenuation coefficient and the self-information, when the self-information is greater than a preset threshold, the corresponding attenuation rate is small, so it is retained; when the self-information is not greater than the preset threshold, the corresponding attenuation rate is large, so it is set to 0, thereby obtaining a sparse self-information matrix. It should also be understood that, in one embodiment of this disclosure, the size of the sparse self-information matrix is ​​the same as the size of the CSI image information matrix.

[0100] Therefore, in this embodiment of the present disclosure, by replacing elements in the self-information matrix whose values ​​are less than a preset threshold with 0, redundant information in the self-information matrix can be removed. This makes it easier to compress the CSI image matrix when compressing and quantizing it based on the self-information matrix, thus ensuring compression efficiency.

[0101] Furthermore, in one embodiment of this disclosure, after determining the sparse self-information matrix, the position information of the non-zero elements in the sparse self-information matrix can also be determined, so that the UE can send the position information to the base station, enabling the base station to reconstruct the CSI information matrix based on the position information, thus ensuring the accuracy of the reconstructed CSI information matrix.

[0102] Step 205: Use the third convolutional layer to perform feature processing on the sparse self-information matrix to obtain n second feature maps.

[0103] In one embodiment of this disclosure, the third convolutional layer may be a convolutional layer without gradient updates.

[0104] Furthermore, in one embodiment of this disclosure, the n second feature maps can be n-dimensional feature maps corresponding to the self-information matrix. For example, in one embodiment of this disclosure, n can be equal to 64, that is: the self-information matrix is ​​processed using a third convolutional layer to obtain a 64-dimensional second feature map corresponding to the self-information matrix.

[0105] It should be noted that, in one embodiment of this disclosure, the structural principles of the first convolutional layer and the third convolutional layer can be the same, both used to perform feature analysis on the input matrix to obtain n feature maps corresponding to the input matrix.

[0106] Step 206: Using the fourth convolutional layer, determine the sparse CSI information matrix based on n first feature maps and n second feature maps, as well as the position information of the non-zero elements in the sparse CSI information matrix.

[0107] In one embodiment of this disclosure, the fourth convolutional layer can be a convolutional layer with gradient updates. Also, as an example, in one embodiment of this disclosure, the kernel size of the fourth convolutional layer can be c×n×s×s, and the stride can be L. As an example, in one embodiment of this disclosure, the kernel size of the fourth convolutional layer can be 2×64×3×3, and the stride can be 1.

[0108] It should be noted that, in one embodiment of this disclosure, the sparse CSI information matrix has the same size and dimension as the CSI information matrix.

[0109] As can be seen from the above, by performing steps 201-206, the elements in the CSI information matrix can be filtered based on the self-information of the CSI information matrix to obtain a sparse CSI information matrix.

[0110] In some embodiments, the above-described scheme may further include:

[0111] Step 103: Determine the feedback information based on the elements in the sparse CSI information matrix, and send the feedback information to the base station.

[0112] Figure 4 This is a schematic diagram of the process for determining feedback information provided in one embodiment of this disclosure, such as... Figure 4 As shown, the method may include:

[0113] Step 401: Compress and quantize the elements in the sparse CSI information matrix to obtain compressed codewords.

[0114] In one embodiment of this disclosure, the method for compressing and quantizing elements in a sparse CSI information matrix may include:

[0115] Step A: Sort the values ​​of each element in the sparse CSI information matrix from largest to smallest according to their self-information.

[0116] Step B: Select the first M element values ​​from the sorted element values, compress and quantize them to form compressed codewords, where one element value corresponds to one codeword.

[0117] In one embodiment of this disclosure, the method for determining M may include:

[0118] Step B1: Determine the compression ratio σ.

[0119] In one embodiment of this disclosure, the compression ratio σ can be determined based on user requirements. For example, in one embodiment of this disclosure, the compression ratio σ can be determined to be equal to 1 / 4.

[0120] Step B2: Calculate the number of codes M for the compressed codeword based on Formula 3, where Formula 3 includes:

[0121]

[0122] Where 'a' indicates the number of bits required to transmit a codeword, 'k' indicates the number of bits required to transmit the position information of the corresponding element in a codeword, and N... c N indicates the number of antennas configured in the base station. t Indicates the number of subcarriers corresponding to each antenna number.

[0123] Furthermore, it should be noted that in one embodiment of this disclosure, since the feedback information sent by the UE to the base station does not only include compressed codewords but also location information, Formula 3 differs from the formula for compression ratio σ in the prior art. Formula 3 also includes the number of bits k required to indicate the location information of transmitting a codeword.

[0124] Furthermore, in one embodiment of this disclosure, the classic Lloyd algorithm can be used in step B to compress and quantize the first M element values, and the M element values ​​can be quantized to 8 bits.

[0125] Step C: Determine the position information of each element value in the sparse CSI information matrix from the first M element values ​​as the position information of the compressed element value in the sparse CSI information matrix.

[0126] Step 402: Determine the position information of the compressed element in the sparse CSI information matrix, and use the compressed codeword and position information as feedback information.

[0127] In summary, in the information feedback method provided in this embodiment, the UE filters the elements in the CSI information matrix to obtain a sparse CSI information matrix. Specifically, when determining the sparse CSI information matrix, the UE first divides the CSI image information matrix corresponding to the CSI information matrix into multiple sub-image information matrices. Then, it determines the distribution function corresponding to each sub-image information matrix based on each sub-image information matrix and its neighboring sub-image information matrices. Next, it determines the self-information corresponding to each sub-image information matrix based on the distribution function. Finally, it filters redundant information based on the self-information to determine the sparse CSI information matrix. Furthermore, the UE determines feedback information based on the elements in the determined sparse CSI information matrix and sends the feedback information to the base station, so that the base station can reconstruct the CSI information matrix based on the feedback information.

[0128] Therefore, in this embodiment of the present disclosure, when determining the distribution function, the structural correlation between each sub-image matrix and its neighboring sub-image matrices is taken into account. Thus, when the sparse CSI information matrix is ​​subsequently determined based on the distribution function, the structure of the original CSI information matrix will not be destroyed. As a result, when the base station reconstructs the CSI information matrix using the feedback information determined from the sparse CSI information matrix, the accuracy of the reconstructed CSI information matrix can be ensured.

[0129] Meanwhile, since the sparse CSI information matrix in this embodiment is a matrix after redundant information has been removed, the ease of subsequent operations on the sparse CSI information matrix can be ensured, and the overhead is reduced.

[0130] Furthermore, in this embodiment, the feedback information determined by the UE includes the position information of the compressed elements in the sparse CSI information matrix. As a result, the base station can accurately reconstruct the CSI information matrix based on the position information of the compressed elements in the sparse CSI information matrix, thus further ensuring the accuracy of the CSI information matrix reconstruction.

[0131] Figure 5 This is a flowchart illustrating an information feedback method provided in an embodiment of this disclosure, applied to a base station, such as... Figure 5 As shown, this information feedback method may include the following steps:

[0132] Step 501: Obtain the feedback information sent by the UE, and determine the preliminary CSI information matrix based on the feedback information.

[0133] In one embodiment of this disclosure, the feedback information may include: compressed codewords obtained by compressing and quantizing the elements in the sparse CSI information matrix corresponding to the CSI information matrix, and the position information of the compressed elements in the sparse CSI information matrix.

[0134] Based on this, in one embodiment of this disclosure, determining the preliminary CSI information matrix based on feedback information may include:

[0135] Step (1): Dequantize the compressed codeword to obtain the dequantized codeword.

[0136] In one embodiment of this disclosure, the element value corresponding to each codeword can be determined by inverse quantization of the compressed codeword.

[0137] Step (2): Construct a second empty matrix, the size of which is the same as the size of the CSI information matrix.

[0138] Step (3): Fill the dequantized codewords into the second empty matrix based on the position information.

[0139] In one embodiment of this disclosure, filling the second empty matrix with the dequantized codeword based on position information may include:

[0140] If the position information corresponding to a certain dequantized codeword is: row W, column Q, then the dequantized codeword can be filled into row W, column Q of the second empty matrix.

[0141] Step (4): Calculate the mean of the dequantized codewords and fill the other positions of the second empty matrix with the mean to obtain the preliminary CSI information matrix.

[0142] Since step 401 only compresses and quantizes M codewords in the sparse CSI information matrix, the dequantized codewords should also only have M elements. Therefore, in one embodiment of this disclosure, the number of dequantized codewords may be less than the number of positions in the second empty matrix. That is, the dequantized codewords may not be able to fill the second empty matrix.

[0143] Based on this, in one embodiment of this disclosure, the mean value of the dequantized codewords can be filled into other positions of the second empty matrix to fill the second empty matrix and obtain the preliminary CSI information matrix.

[0144] Step 502: Reconstruct the CSI information matrix based on the prepared CSI information matrix.

[0145] In one embodiment of this disclosure, determining the CSI information matrix based on the prepared CSI information matrix may include: inputting the prepared CSI information matrix into a pre-trained convolutional structure to output the reconstructed CSI information matrix. It should be noted that, in one embodiment of this disclosure, the convolutional structure may include a fifth, sixth, seventh, eighth, ninth, and tenth convolutional layer connected in sequence.

[0146] Furthermore, in one embodiment of this disclosure, the output of the fifth convolutional layer is also connected to the input of the eighth convolutional layer, and the output of the eighth convolutional layer is also connected to the input of the tenth convolutional layer.

[0147] Furthermore, in one embodiment of this disclosure, the convolutional kernel of the fifth layer can be R1×c×s×s, the convolutional kernel of the sixth layer can be R2×R1×s×s, the convolutional kernel of the seventh layer can be c×R2×s×s, the convolutional kernel of the eighth layer can be R1×c×s×s, the convolutional kernel of the ninth layer can be R2×R1×s×s, the convolutional kernel of the tenth layer can be c×R2×s×s, and the stride of the fifth, sixth, seventh, eighth, ninth, and tenth convolutional layers can be the same.

[0148] For example, in one embodiment of this disclosure, the convolutional kernel of the fourth layer can be 8×2×3×3, the convolutional kernel of the fifth layer can be 16×8×3×3, the convolutional kernel of the sixth layer can be 2×16×3×3, the convolutional kernel of the seventh layer is 8×2×3×3, the convolutional kernel of the eighth layer is 16×8×3×3, the convolutional kernel of the ninth layer is 2×16×3×3, and the stride of the fifth, sixth, seventh, eighth, ninth, and tenth convolutional layers can all be 1.

[0149] Furthermore, in one embodiment of this disclosure, the convolution operation of the convolution structure can be defined as:

[0150]

[0151] Where y d,i j is the (d,i,j)th element of the output of the convolutional structure, where d is the number of output channels and i is the value of y. d,i ,j is the row number in the reconstructed CSI information matrix, j is y d,i,j The number of columns in the reconstructed CSI information matrix, W d,c, h ,w Let be the (d, c, h, w)th element in the kernel weight matrix W, where c is the number of input channels, h is the length of the kernel, w is the width of the kernel, and b is the width of the kernel. d The d-th element in the convolution kernel bias b is... Let i be the (c, i×s1+h, j×s2+w)th element of the convolution input, where i is... The row number in the prepared CSI information matrix, j is... In the prepared CSI information matrix, s1 and s2 are the convolution strides, denoted as (s1, s2), where s1 is the horizontal stride of the convolution kernel and s2 is the vertical stride of the convolution kernel. The activation function for each convolutional layer is the Leakyrelu function, defined as:

[0152]

[0153] Furthermore, in one embodiment of this disclosure, the reconstructed CSI information matrix specifically includes a reconstructed CSI information matrix corresponding to the real part of the CSI information matrix and a reconstructed CSI information matrix corresponding to the imaginary part of the CSI information matrix. Based on this, after determining the reconstructed CSI information matrix corresponding to the real part of the CSI information matrix and the reconstructed CSI information matrix corresponding to the imaginary part of the CSI information matrix, the two reconstructed CSI information matrices can be combined into a complex matrix, and then a two-dimensional discrete Fourier inverse transform can be performed on this complex matrix to obtain the spatial frequency domain CSI information matrix.

[0154] In summary, in the information feedback method provided in this embodiment, the UE filters the elements in the CSI information matrix to obtain a sparse CSI information matrix. Specifically, when determining the sparse CSI information matrix, the UE first divides the CSI image information matrix corresponding to the CSI information matrix into multiple sub-image information matrices. Then, it determines the distribution function corresponding to each sub-image information matrix based on each sub-image information matrix and its neighboring sub-image information matrices. Next, it determines the self-information corresponding to each sub-image information matrix based on the distribution function. Finally, it filters redundant information based on the self-information to determine the sparse CSI information matrix. Furthermore, the UE determines feedback information based on the elements in the determined sparse CSI information matrix and sends the feedback information to the base station, so that the base station can reconstruct the CSI information matrix based on the feedback information.

[0155] Therefore, in this embodiment of the present disclosure, when determining the distribution function, the structural correlation between each sub-image matrix and its neighboring sub-image matrices is taken into account. Thus, when the sparse CSI information matrix is ​​subsequently determined based on the distribution function, the structure of the original CSI information matrix will not be destroyed. As a result, when the base station reconstructs the CSI information matrix using the feedback information determined from the sparse CSI information matrix, the accuracy of the reconstructed CSI information matrix can be ensured.

[0156] Meanwhile, since the sparse CSI information matrix in this embodiment is a matrix after redundant information has been removed, the ease of subsequent operations on the sparse CSI information matrix can be ensured, and the overhead is reduced.

[0157] Furthermore, in this embodiment, the feedback information determined by the UE includes the position information of the compressed elements in the sparse CSI information matrix. As a result, the base station can accurately reconstruct the CSI information matrix based on the position information of the compressed elements in the sparse CSI information matrix, thus further ensuring the accuracy of the CSI information matrix reconstruction.

[0158] Figure 6 This is a schematic diagram of the structure of an information feedback device provided in an embodiment of the present disclosure, configured in a UE, such as... Figure 6 As shown, it may include:

[0159] The first acquisition module is used to acquire the Channel State Information (CSI) matrix.

[0160] The filtering module is used to filter the elements in the CSI information matrix based on the self-information of the CSI information matrix to obtain a sparse CSI information matrix.

[0161] A feature encoder is used to determine feedback information based on elements in the sparse CSI information matrix and send the feedback information to the base station.

[0162] In summary, in the information feedback device provided in this embodiment, the UE filters the elements in the CSI information matrix to obtain a sparse CSI information matrix. Specifically, when determining the sparse CSI information matrix, the UE first divides the CSI image information matrix corresponding to the CSI information matrix into multiple sub-image information matrices. Then, it determines the distribution function corresponding to each sub-image information matrix based on each sub-image information matrix and its neighboring sub-image information matrices. Next, it determines the self-information corresponding to each sub-image information matrix based on the distribution function. Finally, it filters redundant information based on the self-information to determine the sparse CSI information matrix. Furthermore, the UE determines feedback information based on the elements in the determined sparse CSI information matrix and sends the feedback information to the base station, so that the base station can reconstruct the CSI information matrix based on the feedback information.

[0163] Therefore, in this embodiment of the present disclosure, when determining the distribution function, the structural correlation between each sub-image matrix and its neighboring sub-image matrices is taken into account. Thus, when the sparse CSI information matrix is ​​subsequently determined based on the distribution function, the structure of the original CSI information matrix will not be destroyed. As a result, when the base station reconstructs the CSI information matrix using the feedback information determined from the sparse CSI information matrix, the accuracy of the reconstructed CSI information matrix can be ensured.

[0164] Meanwhile, since the sparse CSI information matrix in this embodiment is a matrix after redundant information has been removed, the ease of subsequent operations on the sparse CSI information matrix can be ensured, and the overhead is reduced.

[0165] Furthermore, in this embodiment, the feedback information determined by the UE includes the position information of the compressed elements in the sparse CSI information matrix. As a result, the base station can accurately reconstruct the CSI information matrix based on the position information of the compressed elements in the sparse CSI information matrix, thus further ensuring the accuracy of the CSI information matrix reconstruction.

[0166] Optionally, in one embodiment of this disclosure, the first acquisition module is further configured to: transform the acquired CSI information matrix from the spatial frequency domain to the angular time delay domain, wherein the CSI information matrix includes a real part and an imaginary part.

[0167] Optionally, in another embodiment of this disclosure, the filtering module is further configured to:

[0168] Map the CSI information matrix to a CSI image information matrix;

[0169] The CSI image information matrix is ​​processed using the first convolutional layer to obtain n first feature maps, where n is an integer and n≥1;

[0170] The self-information matrix corresponding to the CSI image information matrix is ​​determined using the second convolutional layer, and the size of the self-information matrix is ​​the same as the size of the CSI image information matrix.

[0171] The elements in the self-information matrix whose values ​​are less than a preset threshold are replaced with 0 to obtain a sparse self-information matrix, and the position information of the non-zero elements in the sparse self-information matrix is ​​determined.

[0172] The sparse self-information matrix is ​​processed using a third convolutional layer to obtain n second feature maps.

[0173] The fourth convolutional layer is used to determine the sparse CSI information matrix based on the n first feature maps and the n second feature maps, as well as the position information of the non-zero elements in the sparse CSI information matrix.

[0174] Optionally, in another embodiment of this disclosure, the filtering module is further configured to:

[0175] Add 0s to the outer perimeter of the CSI image information matrix to expand it into an expanded CSI image information matrix. Divide the expanded CSI image information matrix into m sub-image matrices; where m is an integer, m≥2, and m is the same as the number of elements included in the CSI image information matrix.

[0176] The distribution function of each sub-image matrix is ​​calculated using the second convolutional layer based on each sub-image matrix and its neighboring sub-image matrices. The self-information of each sub-image matrix is ​​calculated based on the distribution function, and the self-information matrix is ​​formed based on the self-information of each sub-image matrix.

[0177] Optionally, in another embodiment of this disclosure, the filtering module is further configured to:

[0178] Select the i-th sub-image matrix p input to the second convolutional layer i where i is an integer, i≥1;

[0179] Determine the p i The nearest sub-image matrix p i ', thus obtaining p i The corresponding set of neighboring sub-images Z i ; wherein, the set of neighboring sub-images Z i Includes: p i (2R+1) inside a circle centered at Manhattan radius R. 2 The nearest sub-image matrix p i ';

[0180] p is determined by Formula 1. i The distribution function f i (p i Formula 1 includes:

[0181]

[0182] in, h represents p i With p i The bandwidth between '.

[0183] Optionally, in another embodiment of this disclosure, the filtering module is further configured to:

[0184] Using Formula 2, the distribution function f based on the i-th sub-image matrix i (p i Calculate the self-information I of the i-th sub-image matrix. i (p i Formula two includes:

[0185] I i (p i ) = -log(f i (p i )).

[0186] Optionally, in another embodiment of this disclosure, the filtering module is further configured to:

[0187] Establish a first empty matrix, the size of which is the same as the size of the CSI image information matrix;

[0188] According to the position of each sub-image matrix in the expanded CSI image information matrix, the self-information corresponding to each sub-image matrix is ​​filled into the first empty matrix to form a self-information matrix.

[0189] Optionally, in another embodiment of this disclosure, both the first convolutional layer and the fourth convolutional layer are convolutional layers with gradient updates.

[0190] Optionally, in another embodiment of this disclosure, the apparatus is further configured to: train the first convolutional layer and the fourth convolutional layer.

[0191] Optionally, in another embodiment of this disclosure, the feature encoder further includes:

[0192] A compression quantization encoder is used to compress and quantize the elements in the sparse CSI information matrix to obtain compressed codewords.

[0193] The determining unit is further configured to determine the position information of the compressed element in the sparse CSI information matrix, and to determine the compressed codeword and the position information as the feedback information.

[0194] Optionally, in one embodiment of this disclosure, the apparatus further includes a first dimension mapping module connected between the filtering module and the feature encoder, used to map the dimension of the sparse CSI information matrix output by the filtering module to a dimension suitable for the feature encoder when the dimension of the sparse CSI information matrix output by the filtering module is not applicable to the dimension of the feature encoder.

[0195] Figure 7 This is a schematic diagram of the structure of an information feedback device provided in an embodiment of this disclosure, configured in a base station, such as... Figure 7 As shown, it may include:

[0196] The second acquisition module is used to acquire feedback information sent by the UE and determine a preliminary CSI information matrix based on the feedback information;

[0197] The reconstruction module is used to reconstruct the CSI information matrix based on the prepared CSI information matrix.

[0198] In summary, in the information feedback device provided in this embodiment, the UE filters the elements in the CSI information matrix to obtain a sparse CSI information matrix. Specifically, when determining the sparse CSI information matrix, the UE first divides the CSI image information matrix corresponding to the CSI information matrix into multiple sub-image information matrices. Then, it determines the distribution function corresponding to each sub-image information matrix based on each sub-image information matrix and its neighboring sub-image information matrices. Next, it determines the self-information corresponding to each sub-image information matrix based on the distribution function. Finally, it filters redundant information based on the self-information to determine the sparse CSI information matrix. Furthermore, the UE determines feedback information based on the elements in the determined sparse CSI information matrix and sends the feedback information to the base station, so that the base station can reconstruct the CSI information matrix based on the feedback information.

[0199] Therefore, in this embodiment of the present disclosure, when determining the distribution function, the structural correlation between each sub-image matrix and its neighboring sub-image matrices is taken into account. Thus, when the sparse CSI information matrix is ​​subsequently determined based on the distribution function, the structure of the original CSI information matrix will not be destroyed. As a result, when the base station reconstructs the CSI information matrix using the feedback information determined from the sparse CSI information matrix, the accuracy of the reconstructed CSI information matrix can be ensured.

[0200] Meanwhile, since the sparse CSI information matrix in this embodiment is a matrix after redundant information has been removed, the ease of subsequent operations on the sparse CSI information matrix can be ensured, and the overhead is reduced.

[0201] Furthermore, in this embodiment, the feedback information determined by the UE includes the position information of the compressed elements in the sparse CSI information matrix. As a result, the base station can accurately reconstruct the CSI information matrix based on the position information of the compressed elements in the sparse CSI information matrix, thus further ensuring the accuracy of the CSI information matrix reconstruction.

[0202] Optionally, in one embodiment of this disclosure, the feedback information includes: compressed codewords obtained by compressing and quantizing the elements in the sparse CSI information matrix corresponding to the CSI information matrix, and the position information of the compressed elements in the sparse CSI information matrix.

[0203] The second acquisition module further includes:

[0204] A dequantizer is used to dequantize the compressed codeword to obtain the dequantized codeword;

[0205] An empty matrix construction unit is used to construct a second empty matrix, the size of which is the same as the size of the CSI information matrix.

[0206] An interpolation unit is used to fill the dequantized codeword into the second empty matrix based on the position information;

[0207] The mean filling unit is used to calculate the mean of the dequantized codeword and fill the mean into other positions of the second empty matrix to obtain the preliminary CSI information matrix.

[0208] Optionally, in one embodiment of this disclosure, the reconstruction module is further configured to:

[0209] The pre-trained convolutional structure is used to obtain the preliminary CSI information matrix, and then the CSI information matrix is ​​output.

[0210] Optionally, in one embodiment of this disclosure, the convolutional structure includes a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, a ninth convolutional layer, and a tenth convolutional layer connected in sequence.

[0211] The output of the fifth convolutional layer is also connected to the input of the eighth convolutional layer, and the output of the eighth convolutional layer is also connected to the input of the tenth convolutional layer.

[0212] Optionally, in one embodiment of this disclosure, the apparatus is further configured to: train the convolutional structure.

[0213] Optionally, in one embodiment of this disclosure, the reconstruction module further includes a second dimension mapping module connected to the output of the tenth convolutional layer, used to map the dimension of the matrix output by the tenth convolutional layer to the dimension of the original CSI information matrix.

[0214] Optionally, in one embodiment of this disclosure, the reconstructed CSI information matrix includes a reconstructed CSI information matrix corresponding to the real part of the CSI information matrix and a reconstructed CSI information matrix corresponding to the imaginary part of the CSI information matrix;

[0215] The reconstruction module is further configured to combine the reconstructed CSI information matrix corresponding to the real part of the CSI information matrix and the reconstructed CSI information matrix corresponding to the imaginary part of the CSI information matrix into a complex matrix, and then perform a two-dimensional discrete Fourier inverse transform on the complex matrix to obtain the spatial frequency domain CSI information matrix.

[0216] also, Figure 8 This is a schematic diagram of the structure of an information feedback system model provided in an embodiment of the present disclosure, such as... Figure 8As shown, the system includes a first acquisition module, a filtering module, a first dimension mapping module, a feature encoder, a second acquisition module, and a reconstruction module connected in sequence. Detailed descriptions of the first acquisition module, the filtering module, the second dimension mapping module, the feature encoder, the second acquisition module, and the reconstruction module can be found in the above description, and will not be repeated here.

[0217] In summary, in the information feedback system model provided in this embodiment, the UE filters the elements in the CSI information matrix to obtain a sparse CSI information matrix. Specifically, when determining the sparse CSI information matrix, the UE first divides the CSI image information matrix corresponding to the CSI information matrix into multiple sub-image information matrices. Then, it determines the distribution function corresponding to each sub-image information matrix based on each sub-image information matrix and its neighboring sub-image information matrices. Next, it determines the self-information corresponding to each sub-image information matrix based on the distribution function. Finally, it filters redundant information based on the self-information to determine the sparse CSI information matrix. Furthermore, the UE determines feedback information based on the elements in the determined sparse CSI information matrix and sends the feedback information to the base station, enabling the base station to reconstruct the CSI information matrix based on the feedback information.

[0218] Therefore, in this embodiment of the present disclosure, when determining the distribution function, the structural correlation between each sub-image matrix and its neighboring sub-image matrices is taken into account. Thus, when the sparse CSI information matrix is ​​subsequently determined based on the distribution function, the structure of the original CSI information matrix will not be destroyed. As a result, when the base station reconstructs the CSI information matrix using the feedback information determined from the sparse CSI information matrix, the accuracy of the reconstructed CSI information matrix can be ensured.

[0219] Meanwhile, since the sparse CSI information matrix in this embodiment is a matrix after redundant information has been removed, the ease of subsequent operations on the sparse CSI information matrix can be ensured, and the overhead is reduced.

[0220] Furthermore, in this embodiment, the feedback information determined by the UE includes the position information of the compressed elements in the sparse CSI information matrix. As a result, the base station can accurately reconstruct the CSI information matrix based on the position information of the compressed elements in the sparse CSI information matrix, thus further ensuring the accuracy of the CSI information matrix reconstruction.

[0221] Furthermore, in one embodiment of this disclosure, it is possible to... Figure 8 The information feedback system model shown is trained, and the training method may include:

[0222] Step 1: First, obtain the CSI information matrix sample set, which may include training samples, validation samples, and test samples.

[0223] For example, in one embodiment of this disclosure, the COST2100[7] channel model can be used to generate 150,000 spatial frequency domain CSI information matrix samples in a 5.3 GHz indoor microcell scenario. The training samples may include 100,000, the validation samples may include 30,000, and the test samples may include 20,000. Also, in one embodiment of this disclosure, the epoch for training the information feedback system model is 1000, the optimizer can be the Adam optimizer, the learning rate is 0.001, and the batch size for training the information feedback system model is 200.

[0224] Step 2: Using the above... Figures 1 to 5 The method shown trains the information feedback system model based on the CSI information matrix sample set and calculates the loss function L.

[0225] In one embodiment of this disclosure, the loss function L can be defined as:

[0226]

[0227] Where N is the number of training samples, H is the reconstructed CSI information matrix output by the reconstruction module configured in the base station. a This is the original CSI information matrix obtained by the first acquisition module configured in the UE. ||·|| is the Euclidean norm.

[0228] Step 3: Update the parameters of the information feedback system model according to the loss function.

[0229] In one embodiment of this disclosure, the parameters of the information feedback system model may include the weights and biases of each convolutional layer.

[0230] Repeat steps one through three above until the loss function converges, indicating that training is complete.

[0231] In summary, since the embodiments of this disclosure employ... Figures 1 to 5 The method shown trains the information feedback system model, where, based on Figures 1 to 5 The method shown has less redundant information and lower overhead in the matrix, which reduces the complexity of the training method for the information feedback system model and allows for faster convergence, thus improving the accuracy and efficiency of the training method.

[0232] In addition, such as Figure 8As shown, the output of the fifth convolutional layer in the reconstruction module is also connected to the input of the eighth convolutional layer, and the output of the eighth convolutional layer is also connected to the input of the tenth convolutional layer. This can prevent the gradient vanishing phenomenon during the training of the reconstruction module and ensure training accuracy.

[0233] The computer storage medium provided in this embodiment stores an executable program; after the executable program is executed by a processor, it can achieve the following: Figures 1 to 4 or Figure 5 Any of the methods shown.

[0234] To achieve the above embodiments, this disclosure also proposes a computer program product, including a computer program, which, when executed by a processor, implements the following: Figures 1 to 4 or Figure 5 Any of the methods shown.

[0235] Furthermore, in order to implement the above embodiments, this disclosure also proposes a computer program that, when executed by a processor, performs the following: Figures 1 to 4 or Figure 5 Any of the methods shown.

[0236] Figure 9 This is a block diagram of a user equipment UE900 provided in one embodiment of this disclosure. For example, the UE900 may be a mobile phone, computer, digital broadcasting terminal equipment, messaging transceiver, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0237] Reference Figure 9 UE900 may include at least one of the following components: processing component 902, memory 904, power supply component 906, multimedia component 908, audio component 910, input / output (I / O) interface 912, sensor component 913, and communication component 916.

[0238] Processing component 902 typically controls the overall operation of UE 900, such as operations associated with display, telephone calls, data communication, camera operation, and recording. Processing component 902 may include at least one processor 920 to execute instructions to perform all or part of the steps of the methods described above. Furthermore, processing component 902 may include at least one module to facilitate interaction between processing component 902 and other components. For example, processing component 902 may include a multimedia module to facilitate interaction between multimedia component 908 and processing component 902.

[0239] Memory 904 is configured to store various types of data to support operation on UE 900. Examples of this data include instructions for any application or method operating on UE 900, contact data, phonebook data, messages, pictures, videos, etc. Memory 904 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0240] Power supply component 906 provides power to various components of UE900. Power supply component 906 may include a power management system, at least one power supply, and other components associated with generating, managing, and distributing power to UE900.

[0241] The multimedia component 908 includes a screen that provides an output interface between the UE 900 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes at least one touch sensor to sense touch, swipe, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or swipe action but also detect the wake-up time and pressure associated with the touch or swipe operation. In some embodiments, the multimedia component 908 includes a front-facing camera and / or a rear-facing camera. When the UE 900 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0242] Audio component 910 is configured to output and / or input audio signals. For example, audio component 910 includes a microphone (MIC) configured to receive external audio signals when UE 900 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 904 or transmitted via communication component 916. In some embodiments, audio component 910 also includes a speaker for outputting audio signals.

[0243] I / O interface 912 provides an interface between processing component 902 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0244] Sensor assembly 913 includes at least one sensor for providing status assessment of various aspects of UE 900. For example, sensor assembly 913 can detect the on / off state of device 900, the relative positioning of components such as the display and keypad of UE 900, changes in position of UE 900 or one of its components, the presence or absence of user contact with UE 900, orientation or acceleration / deceleration of UE 900, and temperature changes of UE 900. Sensor assembly 913 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 913 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 913 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0245] Communication component 916 is configured to facilitate wired or wireless communication between UE 900 and other devices. UE 900 can access wireless networks based on communication standards, such as WiFi, 2G, or 3G, or combinations thereof. In one exemplary embodiment, communication component 916 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 916 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0246] In an exemplary embodiment, UE900 may be implemented by at least one application-specific integrated circuit (ASIC), digital signal processor (DSP), digital signal processing device (DSPD), programmable logic device (PLD), field-programmable gate array (FPGA), controller, microcontroller, microprocessor or other electronic component to perform the above method.

[0247] Figure 10 This is a block diagram of a base station 1000 provided in an embodiment of this disclosure. For example, the base station 1000 can be provided as a base station. (Refer to...) Figure 10 The base station 1000 includes a processing component 1026, which further includes at least one processor, and memory resources represented by a memory 1032 for storing instructions executable by the processing component 1022, such as application programs. The application programs stored in the memory 1032 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1026 is configured to execute instructions to perform any of the methods described above applied to the base station, such as... Figure 1The method shown.

[0248] Base station 1000 may also include a power supply component 1026 configured to perform power management of base station 1000, a wired or wireless network interface 1050 configured to connect base station 1000 to a network, and an input / output (I / O) interface 1058. Base station 1000 may operate on an operating system stored in memory 1032, such as Windows Server™, MacOS X™, Unix™, Linux™, Free BSD™, or similar.

[0249] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0250] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. An information feedback method, characterized in that, Applied to User Equipment (UE), including: Obtain the Channel State Information (CSI) matrix; Based on the self-information of the Channel State Information (CSI) information matrix and the convolutional layers, elements in the CSI information matrix are filtered to obtain a sparse CSI information matrix. The convolutional layers include a first convolutional layer, a third convolutional layer, and a fourth convolutional layer. The first convolutional layer is used to perform feature processing on the CSI image information matrix to obtain n first feature maps, where n is an integer and n≥1. The CSI image information matrix is ​​mapped from the CSI information matrix. The fourth convolutional layer is used to determine the sparse CSI information matrix and the position information of non-zero elements in the sparse CSI information matrix based on the n first feature maps and n second feature maps. The n second feature maps are obtained by performing feature processing on the sparse self-information matrix using the third convolutional layer. Feedback information is determined based on the elements in the sparse CSI information matrix, and the feedback information is sent to the base station; The method further includes: The first convolutional layer and the fourth convolutional layer are trained; The convolutional layer further includes a second convolutional layer. The filtering of elements in the CSI information matrix based on the self-information of the CSI information matrix and the convolutional layer includes: Add 0s to the outer perimeter of the CSI image information matrix to expand it into an expanded CSI image information matrix. Divide the expanded CSI image information matrix into m sub-image matrices; where m is an integer, m≥2, and m is the same as the number of elements included in the CSI image information matrix. The distribution function of each sub-image matrix is ​​calculated using the second convolutional layer based on each sub-image matrix and its neighboring sub-image matrices. The self-information of each sub-image matrix is ​​calculated based on the distribution function, and the self-information matrix is ​​formed based on the self-information of each sub-image matrix. The size of the self-information matrix is ​​the same as the size of the CSI image information matrix. The elements in the self-information matrix whose values ​​are less than a preset threshold are replaced with 0 to obtain a sparse self-information matrix, and the position information of the non-zero elements in the sparse self-information matrix is ​​determined. The self-information I i (p i ) is the distribution function f of the i-th sub-image matrix using the formula. i (p i The formula, calculated as follows, includes: , where p i The i-th sub-image matrix.

2. The method as described in claim 1, characterized in that, The acquisition of the CSI information matrix includes: The obtained CSI information matrix is ​​transformed from the spatial frequency domain to the angular time delay domain, wherein the CSI information matrix includes a real part and an imaginary part.

3. The method as described in claim 1, characterized in that, The step of calculating the distribution function of each sub-image matrix using the second convolutional layer based on each sub-image matrix and its neighboring sub-image matrices includes: Select the i-th sub-image matrix p input to the second convolutional layer i where i is an integer, i≥1; Determine the p i The nearest sub-image matrix p i ’ , to obtain the p i The corresponding set of neighboring sub-images Z i ; wherein, the set of neighboring sub-images Z i Includes: p i (2R+1) inside a circle centered at Manhattan radius R. 2 The nearest sub-image matrix p i ’ ; p is determined by Formula 1. i The distribution function f i (p i Formula 1 includes: ； in, h represents p i With p i ’ The bandwidth between.

4. The method as described in claim 1, characterized in that, The self-information matrix, composed of the self-information of each sub-image matrix, includes: Establish a first empty matrix, the size of which is the same as the size of the CSI image information matrix; According to the position of each sub-image matrix in the expanded CSI image information matrix, the self-information corresponding to each sub-image matrix is ​​filled into the first empty matrix to form a self-information matrix.

5. The method as described in claim 1, characterized in that, Both the first convolutional layer and the fourth convolutional layer are convolutional layers with gradient updates.

6. The method as described in claim 1, characterized in that, The step of determining feedback information based on elements in the sparse CSI information matrix includes: The elements in the sparse CSI information matrix are compressed and quantized to obtain compressed codewords; The position information of the compressed element in the sparse CSI information matrix is ​​determined, and the compressed codeword and the position information are determined as the feedback information.

7. The method as described in claim 6, characterized in that, The step of compressing and quantizing the element values ​​in the sparse CSI information matrix to obtain compressed codewords includes: Sort the values ​​of each element in the sparse CSI information matrix from largest to smallest according to their self-information; Select the first M elements from the sorted values ​​to form the compressed codeword, where each element value corresponds to one codeword; and The position information of each element value in the first M element values ​​in the sparse CSI information matrix is ​​determined as the position information of the compressed element value in the sparse CSI information matrix.

8. The method as described in claim 7, characterized in that, The method for determining M includes: Determine the compression ratio ; The code number M of the compressed codeword is calculated based on Formula 3, which includes: Where 'a' indicates the number of bits required to transmit one codeword, 'k' indicates the number of bits required to transmit the position information of one codeword, and N... c N indicates the number of antennas configured in the base station. t Indicates the number of subcarriers corresponding to each antenna number.

9. An information feedback method, characterized in that, Applied to base stations, including: Obtain feedback information sent by the UE, and determine the preliminary CSI information matrix based on the feedback information; The pre-trained convolutional structure is used to obtain the preliminary CSI information matrix and output the CSI information matrix. The convolutional structure includes a fifth, sixth, seventh, eighth, ninth, and tenth convolutional layer connected in sequence. The output of the fifth convolutional layer is also connected to the input of the eighth convolutional layer, and the output of the eighth convolutional layer is also connected to the input of the tenth convolutional layer. The feedback information is obtained based on a sparse CSI information matrix, which is obtained by filtering the elements in the CSI information matrix based on the self-information of the channel state information (CSI) information matrix and the convolutional layer. The convolutional layer includes a first convolutional layer, a second convolutional layer, a third convolutional layer, and a fourth convolutional layer. The first convolutional layer is used to perform feature processing on the CSI image information matrix to obtain n first feature maps, where n is an integer and n≥1. The CSI image information matrix is ​​mapped from the CSI information matrix. The fourth convolutional layer is used to determine the sparse CSI information matrix and the position information of non-zero elements in the sparse CSI information matrix based on the n first feature maps and n second feature maps. The n second feature maps are obtained by performing feature processing on the sparse self-information matrix using the third convolutional layer. The filtering of elements in the CSI information matrix based on the self-information of the CSI information matrix and the convolutional layer includes: Add 0s to the outer perimeter of the CSI image information matrix to expand it into an expanded CSI image information matrix. Divide the expanded CSI image information matrix into m sub-image matrices; where m is an integer, m≥2, and m is the same as the number of elements included in the CSI image information matrix. The distribution function of each sub-image matrix is ​​calculated using the second convolutional layer based on each sub-image matrix and its neighboring sub-image matrices. The self-information of each sub-image matrix is ​​calculated based on the distribution function, and the self-information matrix is ​​formed based on the self-information of each sub-image matrix. The size of the self-information matrix is ​​the same as the size of the CSI image information matrix. The elements in the self-information matrix whose values ​​are less than a preset threshold are replaced with 0 to obtain a sparse self-information matrix, and the position information of the non-zero elements in the sparse self-information matrix is ​​determined. The self-information I i (p i ) is the distribution function f of the i-th sub-image matrix using the formula. i (p i The formula, calculated as follows, includes: , where p i The i-th sub-image matrix.

10. The method as described in claim 9, characterized in that, The feedback information includes: compressed codewords obtained by compressing and quantizing the elements in the sparse CSI information matrix corresponding to the CSI information matrix, and the position information of the compressed elements in the sparse CSI information matrix; The step of determining the preliminary CSI information matrix based on the feedback information includes: The compressed codeword is dequantized to obtain the dequantized codeword; Construct a second empty matrix, the size of which is the same as the size of the CSI information matrix; Based on the location information, the dequantized codewords are filled into the second empty matrix; Calculate the mean of the dequantized codewords and fill the other positions of the second empty matrix with the mean to obtain the preliminary CSI information matrix.

11. The method as described in claim 9, characterized in that, Also includes: The convolutional structure is trained.

12. An information feedback device, characterized in that, include: The first acquisition module is used to acquire the Channel State Information (CSI) matrix. A filtering module is used to filter elements in the CSI information matrix based on the self-information of the Channel State Information (CSI) information matrix and convolutional layers to obtain a sparse CSI information matrix. The convolutional layers include a first convolutional layer, a third convolutional layer, and a fourth convolutional layer. The first convolutional layer is used to perform feature processing on the CSI image information matrix to obtain n first feature maps, where n is an integer and n≥1. The CSI image information matrix is ​​mapped from the CSI information matrix. The fourth convolutional layer is used to determine the sparse CSI information matrix and the position information of non-zero elements in the sparse CSI information matrix based on the n first feature maps and n second feature maps. The n second feature maps are obtained by performing feature processing on the sparse self-information matrix using the third convolutional layer. A feature encoder is used to determine feedback information based on elements in the sparse CSI information matrix and send the feedback information to the base station; The device is also used to: train the first convolutional layer and the fourth convolutional layer; The convolutional layer further includes a second convolutional layer. The filtering of elements in the CSI information matrix based on the self-information of the CSI information matrix and the convolutional layer includes: Add 0s to the outer perimeter of the CSI image information matrix to expand it into an expanded CSI image information matrix. Divide the expanded CSI image information matrix into m sub-image matrices; where m is an integer, m≥2, and m is the same as the number of elements included in the CSI image information matrix. The distribution function of each sub-image matrix is ​​calculated using the second convolutional layer based on each sub-image matrix and its neighboring sub-image matrices. The self-information of each sub-image matrix is ​​calculated based on the distribution function, and the self-information matrix is ​​formed based on the self-information of each sub-image matrix. The size of the self-information matrix is ​​the same as the size of the CSI image information matrix. The elements in the self-information matrix whose values ​​are less than a preset threshold are replaced with 0 to obtain a sparse self-information matrix, and the position information of the non-zero elements in the sparse self-information matrix is ​​determined. The self-information I i (p i ) is the distribution function f of the i-th sub-image matrix using the formula. i (p i The formula, calculated as follows, includes: , where p i The i-th sub-image matrix.

13. An information feedback device, characterized in that, include: The second acquisition module is used to acquire feedback information sent by the UE and determine a preliminary CSI information matrix based on the feedback information. The reconstruction module is used to obtain the preliminary CSI information matrix using a pre-trained convolutional structure to output the CSI information matrix. The convolutional structure includes a fifth, sixth, seventh, eighth, ninth, and tenth convolutional layer connected in sequence. The output of the fifth convolutional layer is also connected to the input of the eighth convolutional layer, and the output of the eighth convolutional layer is also connected to the input of the tenth convolutional layer. The feedback information is obtained based on a sparse CSI information matrix, which is obtained by filtering the elements in the CSI information matrix based on the self-information of the channel state information (CSI) information matrix and the convolutional layer. The convolutional layer includes a first convolutional layer, a second convolutional layer, a third convolutional layer, and a fourth convolutional layer. The first convolutional layer is used to perform feature processing on the CSI image information matrix to obtain n first feature maps, where n is an integer and n≥1. The CSI image information matrix is ​​mapped from the CSI information matrix. The fourth convolutional layer is used to determine the sparse CSI information matrix and the position information of non-zero elements in the sparse CSI information matrix based on the n first feature maps and n second feature maps. The n second feature maps are obtained by performing feature processing on the sparse self-information matrix using the third convolutional layer. The filtering of elements in the CSI information matrix based on the self-information of the CSI information matrix and the convolutional layer includes: Add 0s to the outer perimeter of the CSI image information matrix to expand it into an expanded CSI image information matrix. Divide the expanded CSI image information matrix into m sub-image matrices; where m is an integer, m≥2, and m is the same as the number of elements included in the CSI image information matrix. The distribution function of each sub-image matrix is ​​calculated using the second convolutional layer based on each sub-image matrix and its neighboring sub-image matrices. The self-information of each sub-image matrix is ​​calculated based on the distribution function, and the self-information matrix is ​​formed based on the self-information of each sub-image matrix. The size of the self-information matrix is ​​the same as the size of the CSI image information matrix. The elements in the self-information matrix whose values ​​are less than a preset threshold are replaced with 0 to obtain a sparse self-information matrix, and the position information of the non-zero elements in the sparse self-information matrix is ​​determined. The self-information I i (p i ) is the distribution function f of the i-th sub-image matrix using the formula. i (p i The formula, calculated as follows, includes: , where p i The i-th sub-image matrix.

14. A user equipment, characterized in that, include: transceiver; Memory; A processor, connected to both the transceiver and the memory, is configured to control the wireless signal transmission and reception of the transceiver by executing computer-executable instructions on the memory, and is capable of implementing the method described in any one of claims 1 to 8.

15. A base station, characterized in that, include: transceiver; Memory; A processor, connected to both the transceiver and the memory, is configured to control the wireless signal transmission and reception of the transceiver by executing computer-executable instructions on the memory, and to implement the method described in any one of claims 9 to 11.

16. An information feedback system model, characterized in that, The system model includes at least the information feedback device as described in claims 12 and 13.

17. A method for training the information feedback system model of claim 16, characterized in that, include: The information feedback system model is trained by performing the method as described in any one of claims 1 to 8 and / or 9 to 11.

18. A computer storage medium, wherein, The computer storage medium stores computer-executable instructions; when executed by a processor, the computer-executable instructions can implement the method described in any one of claims 1 to 8 or 9 to 11.

Images