Data transmission method and communication device

Abstract

The embodiment of the invention provides a data transmission method and a communication device, and the method comprises the steps: obtaining first data, the dimensions of the first data being M; dimension information corresponding to each dimension in the P dimensions is determined, the dimension information corresponding to each dimension is used for determining the dimension size of the full-rank matrix corresponding to each dimension, and the P dimensions are P dimensions in the M dimensions; determining a full-rank matrix and second data corresponding to each dimension according to the first data and the dimension information corresponding to each dimension; and sending the full-rank matrix and the second data corresponding to each dimension. Thus, according to the dimension information corresponding to each dimension, the first device can determine the dimension size of the full-rank matrix corresponding to each dimension without performing high-order singular value decomposition on the high-dimensional data, thereby significantly reducing the operation complexity of first data compression.

Description

Technical Field

[0001] The present application relates to the field of communications, and in particular to a data transmission method and a communication device. Background Art

[0002] Sensing, imaging, artificial intelligence (AI), or machine learning (ML) are potential applications of future communication systems (such as the 6th generation communication system). Because of the wideband multi-frequency, larger multiple-input multiple-output (MIMO) antenna arrays, and the acquisition of signals from different directions, the above applications involve large-scale high-dimensional data transmission.

[0003] In order to reduce the wireless resource overhead of data transmission, high-dimensional data can be compressed. However, how to achieve high-dimensional data compression to reduce computational complexity remains to be studied. Summary of the invention

[0004] The data transmission method and communication device provided in the embodiments of the present application provide an implementation scheme for high-dimensional data compression, which can reduce the computational complexity of high-dimensional data compression.

[0005] To achieve the above objectives, the embodiments of the present application adopt the following technical solutions:

[0006] In the first aspect, a data transmission method is provided, which can be executed by a first device, and the first device can be a terminal device itself, or a processor, module, chip, or chip system that implements the method in the terminal device; or, the first device can be an access network device itself, or a processor, module, chip, or chip system that implements the method in the access network device. The following is an example of the method being executed by the first device. The method includes: obtaining first data, where the dimension of the first data is M dimensions; determining the dimension information corresponding to each dimension of P dimensions, where the dimension information corresponding to each dimension is used to determine the dimension size of the full rank matrix corresponding to each dimension, where P dimensions are P dimensions of M dimensions; determining the full rank matrix and second data corresponding to each dimension according to the first data and the dimension information corresponding to each dimension; and sending the full rank matrix and second data corresponding to each dimension.

[0007] Because in the embodiment of the present application, the first device can first determine P dimensions among the M dimensions of the first data (that is, which dimension needs to be reduced or compressed) and the dimensional size of the full rank matrix corresponding to each of the P dimensions through the dimensional information corresponding to each of the P dimensions, and thus, there is no need to perform high-order singular value HOSVD decomposition and iterative operations on the high-dimensional data, and the dimensional size of the full rank matrix corresponding to the P dimensions and each of the P dimensions can be determined. Therefore, the first device determines the second data and the full rank matrix corresponding to each dimension according to the dimensional size of the full rank matrix corresponding to each of the P dimensions and the first data, which can significantly reduce the computational complexity of compressing the first data.

[0008] It should be understood that in the embodiment of the present application, the full rank matrix corresponding to each dimension in the P dimensions can reflect (or characterize, or represent) the main features (or components, characteristics) corresponding to each dimension in the P dimensions. It can be understood that the column vectors or row vectors in the full rank matrix are linearly independent, that is, after the data is compressed (or reduced in dimension), the above-mentioned column vectors or row vectors can reflect the main features that can be retained in each dimension. In other words, the dimensional size of the full rank matrix corresponding to each dimension can reflect the dimensional size of each dimension after dimensionality reduction.

[0009] In a possible implementation, the dimension information corresponding to each dimension includes the dimension information corresponding to the i-th dimension among the P dimensions, and the dimension information corresponding to the i-th dimension is used to indicate: the dimension size of other dimensions except the i-th dimension in the full rank matrix corresponding to the i-th dimension, 1≤i≤P. It can be understood that the dimension size of other dimensions except the i-th dimension in the full rank matrix corresponding to the i-th dimension can be used to indicate the dimension size after the i-th dimension is reduced, that is, the dimension size of other dimensions except the i-th dimension in the full rank matrix corresponding to the i-th dimension can be replaced by: the dimension size after the i-th dimension is reduced, or the rank of the full rank matrix corresponding to the i-th dimension. In other words, the dimension information corresponding to each dimension can be used to indicate the dimension size after the dimension is reduced, and then the first device does not need to decompose and iterate the first data to determine the core tensor of the first data after the dimension reduction and the dimension size of each dimension after the dimension reduction, thereby significantly reducing the amount of calculation and reducing the implementation complexity of data compression.

[0010] In a possible implementation, the dimension information corresponding to the i-th dimension is used to indicate: the dimensional size of other dimensions except the i-th dimension in the full-rank matrix corresponding to the i-th dimension, including: the scaling factor corresponding to the i-th dimension, the scaling factor corresponding to the i-th dimension is the ratio between the dimensional size of the i-th dimension and the dimensional size of the i-th dimension after dimensionality reduction; or, the scaling factor corresponding to the i-th dimension is the ratio between the dimensional size of the i-th dimension and the rank of the full-rank matrix corresponding to the i-th dimension. That is to say, the first device can determine the dimensional size of each dimension after dimensionality reduction based on the scaling factor corresponding to each dimension in the P dimensions.

[0011] Optionally, the proportional factors corresponding to each dimension in the P dimensions may be the same. That is to say, the proportion of dimensionality reduction of each dimension in the P dimensions is the same, and then the dimension size of each dimension after dimensionality reduction can be uniformly adjusted to reduce the implementation complexity. For example, for the RX (receiving antenna) dimension, TX (transmitting antenna) dimension, and channel state information CSI data in the frequency domain dimension, the dimension size of the RX dimension is 32, the dimension size of the TX dimension is 1024, and the dimension size of the frequency domain dimension is 18. For the proportional factor = the dimension size after the i-th dimension is reduced ÷ the dimension size of the i-th dimension, assuming that the proportional factor is 0.5, the first device can determine that the dimension size after the RX dimension is reduced is 16, the size after the TX dimension is reduced is 512, and the dimension size after the frequency domain dimension is reduced is 9 according to the proportional factor 0.5.

[0012] Optionally, the scaling factor corresponding to each dimension in the P dimensions is different. That is to say, the ratio of dimensionality reduction of each dimension in the P dimensions can be different, and then the dimension size of each dimension after dimensionality reduction can be flexibly adjusted according to the original dimension size of each dimension in the P dimensions. For example, for the above example, since the TX dimension is larger, the corresponding redundant information corresponding to the TX dimension is more, and then the dimension size after dimensionality reduction of the TX dimension can be smaller, such as 128, so the scaling factor is 0.125. Furthermore, since the dimensional size of the frequency domain dimension is smaller, the corresponding redundant information corresponding to the frequency domain dimension may be less, and then the dimension size after dimensionality reduction of the frequency domain dimension is 9, so the scaling factor is 0.5.

[0013] In a possible implementation, the dimension information corresponding to each dimension is associated with one or more of the following parameters: a time domain resource for transmitting the first data, a frequency domain resource for transmitting the first data, a modulation and coding scheme (MCS) for transmitting the first data, and a compression rate of the first data. It can be understood that the MCS for transmitting the first data can indicate: a target code rate for transmitting the first data, a modulation method for transmitting the first data, and a modulation order for transmitting the first data. Among them, the first device can determine the amount of data carried by the resource for transmitting the first data according to parameters such as the MCS, working bandwidth, signal-to-noise ratio, rank (or stream), or number of layers for transmitting the first data. That is, when the MCS for transmitting the first data and the number of frequency domain units allocated on each time domain unit are fixed, the amount of data carried by the resource for transmitting the first data is proportional to the time domain resource for transmitting the first data. Similarly, the amount of data carried by the resource for transmitting the first data is proportional to the frequency domain resource for transmitting the first data, and the amount of data carried by the resource for transmitting the first data is proportional to the MCS for transmitting the first data.

[0014] In a possible implementation, the second data and the full rank matrix corresponding to each dimension are obtained by decomposing the first data according to the dimensional information corresponding to each dimension. That is to say, the first device can determine the dimension size of the full rank matrix corresponding to each dimension according to the dimensional information corresponding to each dimension, and then decompose the first data according to the dimension size of the full rank matrix corresponding to each dimension to obtain the second data (i.e., the core tensor) and the full rank matrix corresponding to each dimension, which can reduce the amount of calculation of the first device to decompose the first data.

[0015] In one possible implementation, the second data is determined based on the first data and the full rank matrix corresponding to each dimension, and the full rank matrix corresponding to each dimension is determined based on the first data and the dimensional information corresponding to each dimension. In other words, the first device can decompose the first data according to the dimensional information corresponding to each dimension to obtain the full rank matrix corresponding to each dimension, and obtain the second data (i.e., the core tensor) based on the full rank matrix corresponding to each dimension and the first data. Compared with performing HOSVD decomposition on the first data and obtaining the second data and the full rank matrix corresponding to each dimension at the same time, the complexity of decomposing high-dimensional data can be reduced.

[0016] For example, the first device performs a tensor multiplication of the first data with the full-rank matrix corresponding to each dimension to obtain the second data. For example, assuming that the i-th dimension in P dimensions corresponds to the j-th dimension in M ​​dimensions (1≤i≤P, 1≤j≤M,), the full-rank matrices corresponding to each dimension in P dimensions are: Q1, Q2, Q3, ...Q i ,…,Q P , Q1 is the full rank matrix corresponding to the first dimension in M ​​dimensions (i.e., the first dimension in P dimensions), Q2 is the full rank matrix corresponding to the third dimension in M ​​dimensions (i.e., the second dimension in P dimensions), Q i is the full rank matrix corresponding to the jth dimension in M ​​dimensions (i.e. the ith dimension in P dimensions), Q P is the full rank matrix corresponding to the Mth dimension among M dimensions (i.e., the Pth dimension among P dimensions). i The row dimension × column dimension of is: L i ×N s,i , L i Actually, it is the dimension size N of the jth dimension among the M dimensions. j , N s,i is the dimension size after the j-th dimension reduction. Further, the second data Σ = the first data H1×1Q1 H ×3Q2 H …× j Q i H …× M Q P H , Q i H Indicates Q i The conjugate transposed matrix of j Q i H Represents the conjugate transposed matrix of the full rank matrix corresponding to the jth dimension in M ​​dimensions, H× j Q i H Indicates the first data H and Q i The conjugate transposed matrix of the tensor is used to perform mode-j multiplication. For another example, the first device can convert the first data H into multiple matrices, and the multiple matrices are Kronecker products or Khatri–Rao products with the full rank matrix corresponding to each dimension.

[0017] In a possible implementation, the full rank matrix corresponding to each dimension is the full rank matrix of the feature matrix corresponding to each dimension, and the feature matrix corresponding to each dimension is determined based on the first data and the dimensional information corresponding to each dimension. In other words, the first device can reduce the dimension and extract features (or feature selection) on the first data according to the dimensional information corresponding to each dimension to obtain the feature matrix corresponding to each dimension, and then the first device can directly decompose the feature matrix corresponding to each dimension to obtain the full rank matrix corresponding to each feature matrix, which can reduce the implementation complexity and computational complexity of the first device in determining the full rank matrix corresponding to each dimension.

[0018] For example, the first device may use orthogonal triangular decomposition or lower-upper triangular decomposition to decompose the feature matrix corresponding to each dimension into a full rank matrix and a matrix of other forms. It can be understood that the dimension size between the feature matrix corresponding to each dimension and its full rank matrix is ​​the same. Assume that the full rank matrix Q corresponding to the i-th dimension among the P dimensions i The dimension size is N i ×N s,i , the dimension size of the feature matrix corresponding to the i-th dimension is N i ×N s,i , that is, the feature matrix corresponding to the i-th dimension is the subspace feature matrix corresponding to the i-th dimension in the first data, which retains the main features corresponding to the i-th dimension, that is, N s,i <N j .

[0019] In one possible implementation, the feature matrix corresponding to each dimension is determined based on the first data and the dimension information corresponding to each dimension, including: the feature matrix corresponding to each dimension is determined based on the first data and the dimension reduction matrix corresponding to each dimension, wherein the dimension reduction matrix corresponding to each dimension is associated with the dimension information corresponding to each dimension. In other words, the first device performs dimensionality reduction processing on the first data based on the dimension reduction matrix corresponding to each dimension, which can reduce the implementation complexity of the first device determining the feature matrix corresponding to each dimension. It should be understood that the dimension reduction matrix corresponding to each dimension is associated with the dimension information corresponding to each dimension, and the first device can determine the dimension size of the dimension reduction matrix corresponding to each dimension in the P dimensions based on the dimension information corresponding to one or more dimensions in the P dimensions.

[0020] For example, the dimensionality reduction matrix corresponding to each dimension can be a column full rank matrix (or row full rank matrix), and the dimension size of the column dimension (or row dimension) of the dimensionality reduction matrix corresponding to each dimension is equal to the dimension size of the column dimension (or row dimension) of the full rank matrix corresponding to each dimension. Then, the first device can reduce the dimension size of the column dimension (or row dimension) in the feature matrix corresponding to each dimension to the dimension size after dimensionality reduction of each dimension through the dimensionality reduction matrix corresponding to each dimension.

[0021] In a possible implementation, when M>2, the feature matrix corresponding to each dimension is determined based on the first data and the dimension reduction matrix corresponding to each dimension, including: the feature matrix corresponding to each dimension is determined based on the two-dimensional data corresponding to each dimension and the dimension reduction matrix corresponding to each dimension, the two-dimensional data corresponding to each dimension includes the two-dimensional data corresponding to the i-th dimension among P dimensions, the two-dimensional data corresponding to the i-th dimension is the two-dimensional data obtained by merging the other dimensions in the first data except the i-th dimension, the two-dimensional data corresponding to the i-th dimension contains the same elements as the first data, 1≤i≤P. That is, the first device can first process the first data, that is, process the first data into two-dimensional data corresponding to each dimension among P dimensions, and then reduce the number of dimensions of the first data to two dimensions, which is easy to realize the calculation of the dimension reduction matrix corresponding to each dimension, and reduce the computational complexity of the first device determining the feature matrix corresponding to each dimension.

[0022] It should be understood that tensor expansion may refer to retaining a certain dimension in high-dimensional data and merging (e.g., concatenating) other dimensions, so that a two-dimensional data can be obtained, and the elements between the two-dimensional data after tensor expansion are the same as those of the high-dimensional data.

[0023] In a possible implementation, when M>2, the feature matrix corresponding to each dimension is determined according to the first data and the dimension reduction matrix corresponding to each dimension, including: the feature matrix corresponding to each dimension is determined according to the feature tensor corresponding to each dimension, the feature tensor corresponding to each dimension is determined according to the first data and the dimension reduction matrix corresponding to each dimension, the feature matrix corresponding to each dimension includes the feature matrix corresponding to the i-th dimension among P dimensions, the feature matrix corresponding to the i-th dimension is a feature matrix obtained by merging the dimensions other than the i-th dimension in the feature tensor corresponding to the i-th dimension, the feature matrix corresponding to the i-th dimension and the feature tensor corresponding to the i-th dimension contain the same elements, 1≤i≤P. That is, the first device can reduce the dimension of the first data according to the dimension reduction matrix corresponding to each dimension to obtain the feature tensor corresponding to each dimension, and determine the feature matrix corresponding to each dimension according to the feature tensor corresponding to each dimension, which can increase the flexibility of the first device in determining the feature matrix to apply to different application scenarios. It can be understood that the number of dimensions of the feature tensor is greater than the number of dimensions of the feature matrix. Therefore, the feature tensor corresponding to the i-th dimension can be considered as: reducing the dimension size of one or more dimensions of the M dimensions except the i-th dimension, and then the feature tensors corresponding to each dimension may be different, or the feature tensors corresponding to at least two different dimensions may be the same, thereby increasing the flexibility of the first device in determining the feature matrix.

[0024] For example, when P is less than M, the dimension reduction matrix and feature tensor can be shared between each dimension in the P dimensions, and the dimension reduction matrix can include MP dimension reduction matrices. Assume that M=4, the dimension size of the first dimension in the M dimensions is N1, the dimension size of the second dimension is N2, the dimension size of the third dimension is N3, and the dimension size of the fourth dimension is N4. In the case where P dimensions include the first and second dimensions of the M dimensions, the dimension reduction matrix shared by the first and second dimensions may include two (MP=2) dimension reduction matrices, namely: a first dimension reduction matrix for reducing the third dimension, and a second dimension reduction matrix for reducing the fourth dimension, the dimension size of the first dimension reduction matrix is ​​N3×N3' (i.e., the dimension size of the third dimension after dimension reduction), and the dimension size of the second dimension reduction matrix is ​​N4×N4' (i.e., the dimension size of the fourth dimension after dimension reduction), so that the dimension size of the feature tensor is N1×N2×N3'×N4', the dimension size of the feature matrix corresponding to the first dimension is N1×(N2×N3'×N4'), and the dimension size of the feature matrix corresponding to the second dimension is N2×(N1×N3'×N4'). Similarly, in the case where M=4 and P=3, the dimension reduction matrix shared between P dimensions is one dimension reduction matrix.

[0025] For another example, different dimensions among the P dimensions may not share dimensionality reduction matrices, and thus the dimensionality reduction matrix corresponding to the i-th dimension among the P dimensions may include M-1 dimensionality reduction matrices, which are used to reduce the dimensionality of the other M-1 dimensions in the first data except the i-th dimension, thereby further reducing the dimensionality of the feature tensor corresponding to the i-th dimension, so as to reduce the computational complexity of subsequently determining the feature matrix corresponding to the i-th dimension.

[0026] In a possible implementation, the dimension information corresponding to each dimension includes the dimension information corresponding to the i-th dimension among P dimensions, and the dimension information corresponding to the i-th dimension is used to indicate: the dimension size after dimension reduction of each dimension in the dimension reduction dimension set corresponding to the i-th dimension, or the dimension reduction ratio corresponding to each dimension in the dimension reduction dimension set. The dimension reduction dimension set includes at least one dimension, and the at least one dimension is a dimension for reducing dimension of at least some of the dimensions except the i-th dimension among the M dimensions. In other words, the first device can determine the dimension size after dimension reduction of each dimension through the dimension information corresponding to each dimension, and can also determine the number of dimension reduction matrices corresponding to each dimension, which dimension is used for dimension reduction, and the dimension size.

[0027] In a possible implementation, the number of non-zero elements in the dimension reduction matrix is ​​less than the number of zero elements. That is, when the two-dimensional data or the first data corresponding to each dimension is multiplied with the dimension reduction matrix, the computational complexity can be reduced because most of the elements in the dimension reduction matrix are zero elements.

[0028] In a possible implementation, the dimension reduction matrix corresponding to each dimension is associated with the dimension information corresponding to each dimension, including: the parameters of the dimension reduction matrix corresponding to each dimension are associated with the dimension information corresponding to each dimension, wherein the parameters of the dimension reduction matrix corresponding to each dimension include at least one of the following: the proportion, position distribution, or value range of the non-zero elements in the dimension reduction matrix corresponding to each dimension. In other words, the first device can determine the proportion range, position distribution range, and value range of the non-zero elements in the dimension reduction matrix corresponding to each dimension based on the association between the dimension information corresponding to each dimension and the parameters of the dimension reduction matrix corresponding to each dimension, thereby enabling the first device to select the dimension reduction matrix corresponding to each dimension based on the proportion range, position distribution range, and value range of the non-zero elements, thereby improving the flexibility of the first device in determining the dimension reduction matrix.

[0029] In a possible implementation, the proportion of non-zero elements in the reduced dimension matrix corresponding to each dimension is greater than or equal to the first threshold corresponding to each dimension, and the first threshold corresponding to each dimension is determined based on the dimensional size of the first data and the dimensional information corresponding to each dimension. In other words, the proportion of non-zero elements in the reduced dimension matrix corresponding to each dimension should not be less than the first threshold, thereby ensuring that the reduced dimension matrix is ​​a column-full rank matrix or a row-full rank matrix. It can be understood that the first device can determine the rank of the reduced dimension matrix corresponding to each dimension based on the dimensional information corresponding to each dimension, that is, the minimum number of non-zero elements that the reduced dimension matrix can contain. Further, the first device can determine the dimensional size of the reduced dimension matrix corresponding to each dimension based on the dimensional size of the first data, so as to determine the minimum proportion of non-zero elements in the reduced dimension matrix corresponding to each dimension. For example, the reduced dimension matrix corresponding to the first dimension has a row dimension × column dimension of: (N2×N3×…×N M )×N s,1 , each column vector in the reduced dimension matrix corresponding to the first dimension should contain at least one non-zero element, so that the number of non-zero elements contained in the reduced dimension matrix is ​​greater than or equal to N s,1 , the minimum proportion of non-zero elements is 1 / (N2×N3×…×N M ).

[0030] Optionally, the first threshold corresponding to each dimension is greater than the minimum ratio corresponding to each dimension. That is, the first threshold is greater than the minimum ratio corresponding to each dimension, which can increase the proportion of non-zero elements in the dimension reduction matrix, thereby increasing the amount of information extracted from the first data, thereby retaining more feature information.

[0031] Optionally, the ratio of non-zero elements in the dimension reduction moment corresponding to each dimension is inversely proportional to the proportional factor corresponding to each dimension. It can be understood that for proportional factor = dimension size after dimension reduction / dimension size before dimension reduction, the larger the proportional factor, the lower the compression rate corresponding to the dimension. The feature matrix or feature tensor obtained by the first device after dimension reduction through the dimension reduction matrix has a larger column dimension (or row dimension) and a higher computational complexity. By reducing the ratio of non-zero elements in the dimension reduction matrix, the computational complexity of the feature matrix or feature tensor obtained by the first device through the dimension reduction matrix can be reduced. Similarly, the smaller the proportional factor, the higher the compression rate corresponding to the dimension. By increasing the ratio of non-zero elements in the dimension reduction matrix, the number of non-zero elements in the feature matrix or feature tensor can be increased, thereby retaining more feature information and improving the accuracy of the full rank matrix in representing the main features corresponding to each dimension.

[0032] In a possible implementation, the values ​​of the non-zero elements in the dimension reduction matrix corresponding to each dimension are ±1. That is to say, when the first device performs the multiplication operation between the two-dimensional data corresponding to each dimension or the first data and the dimension reduction matrix, thanks to the fact that most of the non-zero elements in the dimension reduction matrix have values ​​of ±1, the first device can implement the multiplication operation between the two through a small amount of addition and / or subtraction operations, which can further reduce the complexity of the operation.

[0033] In a possible implementation, the dimension reduction matrix corresponding to each dimension is determined according to the dimension information corresponding to each dimension. The first device determines the dimension reduction matrix corresponding to each dimension through the dimension information corresponding to each dimension, which can save the indication overhead of indicating the dimension reduction matrix corresponding to each dimension.

[0034] In a possible implementation, the dimension reduction matrix corresponding to each dimension is determined according to the dimension information corresponding to each dimension, including: the dimension reduction matrix corresponding to each dimension is determined from a candidate dimension reduction matrix set according to the dimension information corresponding to each dimension, and the candidate dimension reduction matrix set includes at least two dimension reduction matrices. In other words, the first device can select the dimension reduction matrix corresponding to each dimension from the candidate dimension reduction matrix set according to the dimension information corresponding to each dimension, thereby increasing the flexibility of the first device in determining the dimension reduction matrix corresponding to each dimension.

[0035] In one possible implementation, the full rank matrix corresponding to each dimension is determined based on the first data, the third data, and the dimension information corresponding to each dimension, and the third data has the same dimension size as the first data. In other words, the first device can be combined with other data (such as the third data) with the same dimension size as the first data for joint processing to obtain the full rank matrix corresponding to each dimension that can be shared by the first data and the third data, and then the full rank matrix corresponding to each dimension does not need to be calculated again for the third data, thereby improving the efficiency of the first device in processing high-dimensional data with at least two dimensions of the same size.

[0036] For example, the first device determines the full rank matrix corresponding to each dimension according to the first data, the third data, and the dimensional information corresponding to each dimension, including: the first device performs tensor expansion on the first data and the third data to obtain two-dimensional data corresponding to each dimension, and the two-dimensional data corresponding to each dimension includes the first two-dimensional data corresponding to each dimension and the second two-dimensional data corresponding to each dimension; the first device determines the feature matrix corresponding to each dimension according to the two-dimensional data corresponding to each dimension and the dimensionality reduction matrix corresponding to each dimension; the first device decomposes according to the feature matrix corresponding to each dimension to obtain the full rank matrix corresponding to each dimension. It can be understood that the first device can also combine the fifth data, the sixth data, or more data with the same dimensional size as the first data to obtain the two-dimensional data corresponding to the i-th dimension, and then obtain the full rank matrix corresponding to each dimension.

[0037] In a possible implementation, the second data is determined based on the first data, the third data, and the full rank matrix corresponding to each dimension. That is, the first device can use the full rank matrix corresponding to each dimension to compress the first data and the third data respectively, and can obtain the dimension-reduced data of the first data and the second data, thereby further improving the data compression efficiency and the data transmission efficiency.

[0038] For example, the data after the first data dimension reduction may be the first core tensor Σ1, and the data after the third data dimension reduction may be the second core tensor Σ2. Without loss of generality, the first device may jointly process K data of the same dimensional size, and then the kth data H of the K data k The core tensor Σ k =H k ×1Q1 H ×3Q2 H …× j Q i H …× M Q P H .

[0039] In a possible implementation, the method provided in the first aspect further includes: sending first indication information, the first indication information is used to indicate the dimensional size of the first data. It should be understood that the first device can send the dimensional size of the first data to the data recipient (e.g., the second device) so that the second device can determine the configuration information corresponding to each dimension of the P dimensions according to the dimensional size of the first data, or the second device can decompress according to the dimensional size of the first data. In other words, the first device can send the first indication information to the data recipient (e.g., the second device) to indicate the dimensional size of the first data, thereby improving the flexibility of the second device in obtaining the dimensional size of the first data to adapt to different application scenarios. For example, if the first data is artificial intelligence AI data, the dimensional information of AI may not be associated with the capability information of the first device, so that the second device can determine the dimensional size of the first data according to the first indication information. It can be understood that in the case where the agreed compression rate is not mentioned, the second device can also determine the dimensional size of the first data through the first indication information.

[0040] In a possible implementation, the first indication information includes index information. The index information is used to determine the dimension size of the first data from a candidate dimension set, where the candidate dimension set includes at least two dimension sizes. That is, the first device can indicate the dimension size of the first data to the second device by indicating the index of the dimension size of the first data in the candidate set, thereby saving the indication overhead of the first indication information and improving the reliability of the first indication information.

[0041] In a possible implementation, the first indication information also includes indication information of a candidate dimension set. That is, the first device can configure a candidate dimension set for the second device to indicate to the second device the candidate dimension sizes of the first data that the first device expects to send in the next period of time.

[0042] In a possible implementation, the first indication information may be carried by at least one of the following: a radio resource control RRC message (or signaling), downlink control information DCI, a media access control protocol data unit MAC PDU, uplink control information UCI, or a physical uplink control channel PUCCH. In other words, the dimension size of the first data indicated by the first indication information can be continuously effective during the RRC connection, and there is no need to indicate the dimension size of the first data each time it is scheduled. It is suitable for scenarios where data of the same dimension size is continuously sent during the RRC connection, and the dimension size of data transmission is changed through RRC signaling due to mobility, energy saving, or changes in business requirements.

[0043] In a possible implementation, the method provided in the first aspect further includes: sending second indication information, where the second indication information is used to indicate P dimensions. That is, the first device can send the second indication information so that the second device determines, according to the second indication information, that P dimensions of the M dimensions of the first data are reduced in dimension, that is, the second device can determine that the received second data is obtained by reducing the P dimensions of the first data, and determine that the full rank matrix corresponding to each received dimension is associated with each dimension of the P dimensions, so as to decompress the second data and improve the decompression efficiency of the second device.

[0044] In a possible implementation, the method provided in the first aspect further includes: receiving second indication information, the second indication information being used to indicate P dimensions. That is, the first device can determine, based on the second indication information sent by the network side (e.g., the second device), to reduce the dimensions of P of the M dimensions of the first data, thereby improving the flexibility of the first device in determining which dimensions of the M dimensions to reduce, so as to adapt to different application scenarios. For example, the second device can determine, based on data decompression requirements, which P dimensions of the M dimensions of the first data to reduce, and send the second indication information to the first device, so that the first device can determine, based on the second indication information, to reduce the dimensions of the P dimensions of the first data, thereby matching the data compression of the first data with the data decompression requirements of the second device, thereby improving the efficiency of data transmission.

[0045] In a possible implementation, the dimension information corresponding to each dimension is determined from a candidate dimension information set according to the second indication information. The candidate dimension information set includes at least two dimension information corresponding to each dimension. That is to say, the first device and the second device can determine the dimension information corresponding to each dimension of the P dimensions according to the second indication information, and then the first device can determine the full rank matrix and second data corresponding to each dimension according to the dimension information corresponding to each dimension and the first data, and the second device can determine the dimensional size of the full rank matrix corresponding to each dimension according to each corresponding dimensional information, thereby realizing the decoding of the compressed bit stream carrying the second data and the full rank matrix corresponding to each dimension.

[0046] In a possible implementation, the second indication information is also used to indicate the dimension information corresponding to each dimension. That is, the second indication information can be used to directly indicate the dimension information corresponding to each dimension, and then for the first device to receive the second indication information, the first device can obtain the second data and the full rank matrix corresponding to each dimension according to the dimension information corresponding to each dimension and the first data; for the first device to send the second indication information, the second device can determine the dimension size of the full rank matrix corresponding to each dimension according to the dimension information corresponding to each dimension, so as to decode the compressed bit stream carrying the second data and the full rank matrix corresponding to each dimension according to the dimension size.

[0047] In a possible implementation, the second indication information is also used to indicate the dimension reduction matrix corresponding to each dimension. That is, the second indication information can also indicate the dimension reduction matrix of each dimension, so that the first device can reduce the dimension of the first data or the two-dimensional data corresponding to each dimension according to the indicated dimension reduction matrix, thereby eliminating the need for the first device to optimize the dimension reduction matrix offline, thereby reducing the power consumption of the first device.

[0048] In the second aspect, a data transmission method is provided, which can be executed by a second device, and the second device can be the terminal device itself, or can refer to a processor, module, chip, or chip system in the terminal device that implements the method; or, the second device can be the access network device itself, or can refer to a processor, module, chip, or chip system in the access network device that implements the method. The following is an example of the method being executed by the second device. The method includes: receiving second data and a full rank matrix corresponding to each dimension in P dimensions, the second data is associated with the first data, the dimension of the first data is M dimensions, and the P dimensions are P dimensions in M ​​dimensions; obtaining the dimensional size of the first data; determining the fourth data according to the dimensional size of the first data, the second data, and the full rank matrix corresponding to each dimension.

[0049] In a possible implementation, the second device obtains the dimension size of the first data, including: the second device determines the dimension size of the first data according to the capability information of the first device and / or the resources used to transmit the first data. That is, the second device can determine the dimension size of the first data according to the capability information of the first device and / or the resources used to transmit the first data, and then the second device does not need the first device to indicate the dimension size of the first data, thereby saving network overhead.

[0050] In a possible implementation, the method provided in the second aspect further includes: receiving first indication information, where the first indication information is used to indicate the dimension size of the first data.

[0051] In a possible implementation, the first indication information includes index information, wherein the index information is used to determine the dimension size of the first data from a candidate dimension set, where the candidate dimension set includes at least two dimension sizes.

[0052] In a possible implementation manner, the first indication information further includes indication information of a candidate dimension set.

[0053] In one possible implementation, the first indication information may be carried by at least one of the following: a radio resource control RRC message (or signaling), downlink control information DCI, a media access control protocol data unit MAC PDU, uplink control information UCI, or a physical uplink control channel PUCCH.

[0054] In a possible implementation manner, the method provided in the second aspect further includes: receiving second indication information, where the second indication information is used to indicate P dimensions.

[0055] In a possible implementation manner, the method provided in the second aspect further includes: sending second indication information, where the second indication information is used to indicate P dimensions.

[0056] In a possible implementation, the dimension information corresponding to each dimension is determined from a candidate dimension information set according to the second indication information, wherein the candidate dimension information set includes at least two dimension information corresponding to each dimension.

[0057] In a possible implementation, the second indication information is also used to indicate dimension information corresponding to each dimension.

[0058] In one possible implementation, the dimension information corresponding to each dimension includes the dimension information corresponding to the i-th dimension among P dimensions, and the dimension information corresponding to the i-th dimension is used to indicate: the dimension sizes of other dimensions except the i-th dimension in the full rank matrix corresponding to the i-th dimension, 1≤i≤P.

[0059] In one possible implementation, the dimension information corresponding to the i-th dimension is used to indicate: the dimensional sizes of other dimensions in the full-rank matrix corresponding to the i-th dimension except the i-th dimension, including: the scaling factor corresponding to the i-th dimension, the scaling factor corresponding to the i-th dimension is the ratio between the dimensional size of the i-th dimension and the dimensional size after dimensionality reduction of the i-th dimension; or, the scaling factor corresponding to the i-th dimension is the ratio between the dimensional size of the i-th dimension and the rank of the full-rank matrix corresponding to the i-th dimension.

[0060] In one possible implementation, the dimension information corresponding to each dimension is associated with one or more of the following parameters: time domain resources used to transmit the first data, frequency domain resources used to transmit the first data, a modulation and coding scheme MCS used to transmit the first data, and a compression rate of the first data.

[0061] In a possible implementation, the second indication information is also used to indicate the dimension reduction matrix corresponding to each dimension.

[0062] In a possible implementation, the dimension reduction matrix corresponding to each dimension is determined according to the dimension information corresponding to each dimension.

[0063] In a possible implementation, the dimensionality reduction matrix corresponding to each dimension is determined based on the dimensionality information corresponding to each dimension, including: the dimensionality reduction matrix corresponding to each dimension is determined from a set of candidate dimensionality reduction matrices based on the dimensionality information corresponding to each dimension, and the set of candidate dimensionality reduction matrices includes at least two dimensionality reduction matrices.

[0064] In a possible implementation, the dimension information corresponding to each dimension includes the dimension information corresponding to the i-th dimension among the P dimensions, and the dimension information corresponding to the i-th dimension is used to indicate: the dimension size after dimension reduction of each dimension in the dimension reduction dimension set corresponding to the i-th dimension, or the dimension reduction ratio corresponding to each dimension in the dimension reduction dimension set. The dimension reduction dimension set includes at least one dimension, and the at least one dimension is a dimension for reducing the dimension of at least some of the dimensions except the i-th dimension among the M dimensions.

[0065] In one possible implementation, the parameters of the dimensionality reduction matrix corresponding to each dimension are associated with the dimensionality information corresponding to each dimension, wherein the parameters of the dimensionality reduction matrix corresponding to each dimension include at least one of the following: the proportion, position distribution, or value range of non-zero elements in the dimensionality reduction matrix corresponding to each dimension.

[0066] In a possible implementation, the proportion of non-zero elements in the dimensionality reduction matrix corresponding to each dimension is greater than or equal to a first threshold corresponding to each dimension, and the first threshold corresponding to each dimension is determined based on the dimensional size of the first data and the dimensional information corresponding to each dimension.

[0067] Optionally, the first threshold corresponding to each dimension is greater than the minimum ratio corresponding to each of the above dimensions.

[0068] Optionally, the ratio of non-zero elements in the dimension reduction moment corresponding to each dimension is in inverse proportion to the scaling factor corresponding to each dimension.

[0069] Among them, the technical effects of the second aspect and any design method of the second aspect mentioned above can refer to the first aspect and will not be repeated here.

[0070] In a third aspect, a communication device is provided for implementing the above-mentioned various methods. The communication device may be the first device in the above-mentioned first aspect or any implementation thereof, or a device including the above-mentioned first device, or a device included in the above-mentioned first device, such as a chip; or, the communication device may be the second device in the above-mentioned second aspect or any implementation thereof, or a device including the above-mentioned second device, or a device included in the above-mentioned second device, such as a chip. The communication device includes a module, unit, or means corresponding to the implementation of the above-mentioned method, and the module, unit, or means may be implemented by hardware, software, or by executing the corresponding software implementation by hardware. The hardware or software includes one or more modules or units corresponding to the above-mentioned functions.

[0071] In some possible designs, the communication device may include a processing module and a transceiver module. The transceiver module, which may also be referred to as a transceiver unit, is used to implement the sending and / or receiving functions in any of the above aspects and any possible implementations thereof. The transceiver module may be composed of a transceiver circuit, a transceiver, a transceiver or a communication interface. The processing module may be used to implement the processing functions in any of the above aspects and any possible implementations thereof.

[0072] In some possible designs, the transceiver module includes a sending module and a receiving module, which are respectively used to implement the sending and receiving functions in any of the above aspects and any possible implementation methods thereof.

[0073] In a fourth aspect, a communication device is provided, comprising: at least one processor; the processor is used to execute a computer program or instruction so that the communication device executes the method described in any one of the above aspects.

[0074] In a possible implementation, the communication device further includes the memory. Optionally, the memory is coupled to the processor, the memory may be integrated with the processor, or the memory may be independent of the processor. Optionally, the processor is used to execute a computer program or instruction stored in the memory.

[0075] In a possible implementation, the memory is independent of the communication device.

[0076] In a possible implementation, the communication device further includes a communication interface, and the communication interface is used to communicate with a module outside the communication device.

[0077] The communication device may be the first device in the above-mentioned first aspect or any implementation manner thereof, or a device including the above-mentioned first device, or a device included in the above-mentioned first device, such as a chip; or, the communication device may be the second device in the above-mentioned second aspect or any implementation manner thereof, or a device including the above-mentioned second device, or a device included in the above-mentioned second device, such as a chip.

[0078] In a fifth aspect, a computer-readable storage medium is provided, in which a computer program or instruction is stored. When the computer-readable storage medium is run on a communication device, the communication device can execute the method described in any one of the above aspects or any one of its implementation methods.

[0079] In a sixth aspect, a computer program product comprising instructions is provided, which, when executed on a communication device, enables the communication device to execute the method described in any one of the above aspects or any one of its implementations.

[0080] In a seventh aspect, a communication device (for example, the communication device may be a chip or a chip system) is provided, wherein the communication device includes a processor for implementing the functions involved in any of the above aspects or any of its implementation methods.

[0081] In some possible designs, the communication device includes a memory for storing necessary program instructions and data.

[0082] In some possible designs, when the device is a chip system, it can be composed of a chip or include a chip and other discrete devices.

[0083] It can be understood that when the communication device provided in any one of the third aspect to the seventh aspect is a chip, the above-mentioned sending action / function can be understood as output, and the above-mentioned receiving action / function can be understood as input.

[0084] Among them, the technical effects brought about by any design method in the third to seventh aspects can refer to the technical effects brought about by different design methods in the above-mentioned first aspect, and will not be repeated here.

[0085] In an eighth aspect, a communication system is provided, comprising: the first device in the above-mentioned first aspect or any implementation thereof, and the second device in the above-mentioned second aspect or any implementation thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 This is a flow chart of channel state information CSI data compression provided by an embodiment of the present application;

[0087] Figure 2 is another CSI data compression flow chart provided in an embodiment of the present application;

[0088] Figure 3 It is a structural diagram of a communication system provided in an embodiment of the present application;

[0089] Figure 4 It is a flowchart of a data transmission method provided by an embodiment of the present application;

[0090] Figure 5 It is a schematic diagram of tensor expansion provided in an embodiment of the present application;

[0091] Figure 6 is a schematic diagram of a process for determining second data provided by an embodiment of the present application;

[0092] Figure 7 is a schematic diagram of elements in a dimensionality reduction matrix provided in an embodiment of the present application;

[0093] Figure 8 This is a schematic diagram of a communication device structure provided in an embodiment of the present application. Figure 1 ;

[0094] Fig. 9 This is a schematic diagram of a communication device structure provided in an embodiment of the present application. Figure 2 . DETAILED DESCRIPTION

[0095] To facilitate understanding of the technical solutions provided by the embodiments of the present application, a brief introduction to the relevant technical terms of the present application is first given. The brief introduction is as follows:

[0096] First, the dimension size of the data:

[0097] In an embodiment of the present application, dimension may refer to the number of dimensions of data, or the number of dimensions. Dimension size (size / dimension size) refers to the dimension size of the dimension. For example, the dimension size of data refers to the product of the dimension sizes of each dimension in multiple dimensions of data. For example, for two-dimensional data (or data matrix), the dimension of the data is two dimensions: the first dimension and the second dimension, and the dimension size of the data is the product of the dimension size N1 of the first dimension and the dimension size N2 of the second dimension, that is, N1×N2. For another example, for higher-dimensional data (or data tensors), the dimension of the data is M dimensions, and the dimension size of each dimension in the M dimensions is N1, N2,..., N respectively. i ,…,N M , N i is the dimension size of the i-th dimension, and the dimension size of the data is N1×N2×…×N i ×…×N M , 1≤i≤M, M is an integer greater than 2.

[0098] It can be understood that different dimensions can correspond to different meanings. For example, for channel state information (CSI) data, it can correspond to three dimensions, namely: receiving antenna (RX) dimension, transmitting antenna (TX) dimension, and frequency domain dimension. Among them, the dimensional size of the receiving antenna dimension can represent the number of antennas at the receiving end (or the number of receiving antenna ports), the dimensional size of the transmitting antenna dimension can represent the number of transmitting antennas at the transmitting end (or the number of transmitting antenna ports), and the dimensional size of the frequency domain dimension can represent the number of frequency domain units. The frequency domain unit can be, for example, a resource block (RB), a resource element (RE), or a subcarrier, etc. The embodiments of the present application do not make specific limitations on this.

[0099] It should be understood that for a radio access network (RAN) device, the number of antennas may reach hundreds or more, such as 512 or 1024, and thus CSI data is usually high-dimensional data. Direct transmission of the CSI data results in a large transmission overhead.

[0100] To this end, the 3rd generation partnership project (3GPP) provides a compression method to compress CSI data. The data compression process is described below.

[0101] Second, data compression process:

[0102] Figure 1 This is a CSI data compression flow chart provided by an embodiment of the present application. Figure 1 As shown in the figure, the compression process mainly includes:

[0103] S101, the transmitter performs singular value decomposition (SVD) on the two-dimensional data corresponding to each frequency domain unit in the three-dimensional CSI data H to obtain two-dimensional data corresponding to each rank in K ranks. The K ranks are obtained by reducing the RX dimension through two-dimensional SVD.

[0104] S102: The transmitter performs discrete Fourier transform (DFT) codebook projection and dimensionality reduction operations on the two-dimensional data W corresponding to each rank to obtain a coefficient matrix W2. The selected codebooks are matrix W1 and matrix W f , matrix W1 corresponds to the transmit antenna dimension, matrix W fCorresponding to the frequency domain dimension. It can be understood that the two-dimensional data corresponding to each rank can be decomposed into matrix W1, matrix W f , and coefficient matrix W2. For example, W = W1W2W f H .

[0105] S103, sending the index of the basis vector in the matrix W1, the matrix W f The index of the basis vector in , and the indication information of the elements in the coefficient matrix W2. The indication information of the elements in the coefficient matrix W2 can be used to indicate the value and position of the elements in the coefficient matrix W2.

[0106] It can be understood that the sending end can quantize the information sent in the above step S103, convert it into a compressed bit stream, and send the compressed bit stream.

[0107] It should be understood that the above Figure 1 The compression process shown is to realize high-dimensional CSI data compression by splitting the three-dimensional CSI data into multiple matrices and performing DFT codebook projection and dimensionality reduction operations on each of the multiple matrices. However, when the data dimension is large, the two-dimensional SVD decomposition of each frequency domain unit is large in computational complexity and the time consumed will increase significantly. In addition, since the three dimensions are not decomposed by three-dimensional SVD, the correlation is not fully utilized, and there will still be a certain amount of redundant information after the two-dimensional SVD decomposition, that is, the compression efficiency will be lost.

[0108] in addition, Figure 2 FIG. 1 is another schematic diagram of a CSI data compression process provided by an embodiment of the present application. Figure 2 As shown in the figure, the compression process compresses high-dimensional data based on compressed sensing, and the main process includes: the transmitter first converts the three-dimensional CSI data H into multiple vectors; the transmitter multiplies each vector in the multiple vectors with the observation matrix to obtain an observation value; the transmitter quantizes or performs entropy coding on the observation value to obtain a compressed bit stream, and sends the compressed bit stream.

[0109] The measurement matrix may be a random matrix such as a Gaussian random matrix, a Bernoulli random matrix, or a fixed matrix such as a Toplitz or cyclic measurement matrix generated by a given sequence.

[0110] However, Figure 2 The compression process shown can only guarantee average performance when using random matrices, and performance may deteriorate. Furthermore, for data with medium sparsity, the compression efficiency is low, and a large number of observations are required to ensure high average performance.

[0111] Based on the above problems, an embodiment of the present application provides a data transmission method for providing an implementation scheme for high-dimensional data compression, which can reduce the computational complexity of high-dimensional data compression.

[0112] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0113] In order to facilitate understanding of the embodiments of the present application, the following points are explained before introducing the embodiments of the present application.

[0114] 1. In the embodiment of the present application, dimensionality reduction is different from the usual data dimensionality reduction. The usual data dimensionality reduction refers to converting high-dimensional data into low-dimensional data, that is, reducing the number of dimensions of the dimensional data, while the dimensionality reduction in the embodiment of the present application refers to reducing the dimensionality of a certain (or multiple, or all) dimensions in the data. For example, for CSI data dimensionality reduction, it may refer to reducing the dimensionality of the RX dimension in the CSI data, or reducing the dimensionality of the TX dimension in the CSI data, and reducing the dimensionality of the frequency domain dimension in the CSI data. It should be understood that unless it is emphasized that dimensionality reduction refers to reducing the number of dimensions of the data, the dimensionality reduction in the embodiment of the present application refers to reducing the dimensionality of one or more dimensions of the M dimensions of the data, which is uniformly explained here and will not be repeated below.

[0115] 2. In the embodiments of the present application, when a "set" or "combination" is involved, the objects included in the set or combination may be one or at least two. For example, a vector combination may include one vector or at least two vectors. For example, a dimension set may include one dimension or at least two dimensions.

[0116] 3. In the embodiments of the present application, for the convenience of description, when numbering or indexing is involved, the numbering can be started from 1 or from 0, or from any parameter.

[0117] 4. "Predefined", "pre-defined", "pre-configured (or pre-configured)", and "protocol agreement" can be used interchangeably, and pre-definition can be achieved by pre-saving corresponding codes, tables or other methods that can be used to indicate relevant information in a device (for example, the first device or the second device). The embodiments of the present application do not limit the specific implementation method. Among them, "saving" can mean saving in one or more memories.

[0118] 5. The “protocol” involved in the embodiments of the present application may refer to a standard protocol in the field of communications, such as the long term evolution (LTE) protocol, the new radio (NR) protocol, wireless fidelity (Wi-Fi), and related protocols used in future communication systems (such as the sixth generation (6G) communication system), which is not limited in the embodiments of the present application.

[0119] 6. In the embodiments of the present application, descriptions such as “when…”, “in the case of…”, “if” and “if” all mean that under certain objective circumstances, the device (such as the first device or the second device) will perform corresponding processing. It does not limit the time, nor does it require the device to have a judgment action when implementing it, nor does it mean that there are other limitations.

[0120] 7. In the embodiments of the present application, “sending information to...(first device)” can be understood as the destination of the information being the first device, and can include directly or indirectly sending information to the first device. “Receiving information from...(second device)” or “receiving information from...(second device)” can be understood as the source of the information being the second device, and can include directly or indirectly receiving information from the second device. The information may be processed as necessary between the source and destination of the information transmission, such as format changes, but the destination can understand the valid information from the source. Similar expressions in the present application can be understood similarly and will not be repeated here.

[0121] 8. In the description of the embodiments of the present application, unless otherwise specified, the "and / or" in the embodiments of the present application indicates that there may be three relationships, for example, A and / or B, which may indicate: A exists alone, A and B exist at the same time, and B exists alone, wherein A and B may be singular or plural. Moreover, "at least one of the following" or similar expressions thereof refers to any combination of these items, including any combination of single items or plural items. In addition, in order to facilitate the clear description of the technical solutions of the embodiments of the present application, in the embodiments of the present application, the words "first", "second" and the like are used to distinguish the same items or similar items with substantially the same functions and effects. Those skilled in the art may understand that the words "first", "second" and the like do not limit the quantity and execution order, and the words "first", "second" and the like do not limit the certain difference. At the same time, in the embodiments of the present application, the words "exemplary" or "for example" are used to indicate examples, illustrations or descriptions.

[0122] The embodiments of the present application may be applicable to LTE systems or NR systems (also referred to as fifth generation (5th generation, 5G) systems), systems with hybrid networking of LTE and NR, vehicle to everything (V2X) systems, device-to-device (D2D) systems, machine to machine (M2M) communication systems, Internet of Things (IoT) systems (such as narrowband Internet of Things (NB-IoT) systems), Wi-Fi systems, non-terrestrial networks (NTN) systems, 6G systems, and other next generation communication systems. Alternatively, the communication system may also be an open radio access network (O-RAN or ORAN) or a cloud radio access network (cloud RAN, CRAN), without limitation.

[0123] It can be understood that the embodiments of the present application can be applicable to a variety of different business scenarios, such as enhanced mobile broadband (eMBB), ultra-high reliability and ultra-low latency communication (URLLC), massive machine type communication (mMTC), immersive communication, massive communication, ubiquitous connections, integrated artificial intelligence and communication, or integrated sensing and communication, etc. In order to meet the further requirements of the above-mentioned different business application scenarios for latency, reliability, and coverage, more flexible resource allocation is required.

[0124] In addition, the communication architecture and business scenarios described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Ordinary technicians in this field can know that with the evolution of the communication architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0125] like Figure 3 , which is a schematic diagram of the structure of a communication system 300 provided in an embodiment of the present application. Figure 3 The communication system 300 includes at least one access network device (eg Figure 3 310a or 310b), and at least one terminal device (eg Figure 3 It is understood that the access network device can be connected to the core network (CN) by wireless or wired means, and the CN equipment and the access network device in the CN can be different physical devices, or the same physical device integrating the CN logical function and the wireless access network logical function. It is understood that Figure 3 The number of access network devices and terminal devices is only an example, and can be more or less, and the embodiments of the present application do not specifically limit this.

[0126] In a possible implementation, the access network device in the embodiment of the present application may be a device that communicates with the terminal device. The access network device may also be referred to as a RAN device, an access node, a RAN entity, or a RAN node. Figure 3 As shown, the multiple access network devices in the communication system 300 can be nodes of the same type or nodes of different types. In some scenarios, the roles of the access network device and the terminal device are relative, for example Figure 3 The network element 320i in the figure can be a helicopter or a drone, which can be configured as a mobile base station. For the terminal devices 320j that access the communication system 300 through the network element 320i, the network element 320i can be the base station 310a; but for the base station 310a, the network element 320i is a terminal device. The access network device and the terminal device are sometimes referred to as communication devices, for example Figure 3 The network elements 310a and 310b may be understood as communication devices having base station functions, and the network elements 320a-220j may be understood as communication devices having terminal functions.

[0127] In one possible scenario, the access network device may be a transmission and reception point (TRP), a base station, a remote radio unit (RRU) or a baseband unit (BBU) (also called a digital unit (DU)) of a split base station, a broadband network gateway (BNG), an aggregation switch, a non-3GPP access device, a relay station or an access point. The access network device may be a macro base station (e.g. Figure 3310a), micro base station or indoor station (such as Figure 3 The access network device may be a network element 310b in the V2X system, a relay node or a donor node, or a wireless controller in a CRAN scenario. Optionally, the access network device may also be a server, a wearable device, a vehicle or an on-board device, etc. For example, the access network device in the V2X system may be a roadside unit (RSU). In addition, the access network device in the embodiment of the present application may be an eNB or eNodeB (evolutional NodeB) in LTE, a wireless controller in a CRAN scenario, a base station in a 5G communication system (e.g., a next-generation Node B (gNodeB, gNB)), or a base station in a future evolution system (e.g., a 6G communication system), etc., which is not specifically limited here.

[0128] In one possible implementation, in some deployments, the gNB may include a centralized unit (CU), a DU, a CU-control plane (CP), a CU-user plane (UP), or a radio unit (RU). The gNB may also include an active antenna unit (AAU). The CU implements some functions of the gNB, and the DU implements some functions of the gNB, for example, the CU is responsible for processing non-real-time protocols and services, and implementing the functions of the radio resource control (RRC) and / or packet data convergence protocol (PDCP) layers. The DU is responsible for processing physical (PHY) layer protocols and real-time services, and implementing the functions of the radio link control (RLC) layer, the media access control (MAC) layer, and the PHY layer. The AAU implements some physical layer processing functions, RF processing, and related functions of active antennas. Since the information of the RRC layer will eventually become the information of the PHY layer, or be converted from the information of the PHY layer, therefore, under this architecture, high-level signaling, such as RRC layer signaling, can also be considered to be sent by DU, or, sent by DU+AAU. It can be understood that the access network device can be a device including one or more of a CU node, a DU node, and an AAU node. In addition, the CU can be divided into an access network device in the RAN, and the CU can also be divided into an access network device in the CN, and the embodiments of the present application are not limited to this.

[0129] In different systems, CU (or CU-CP and CU-UP), DU or RU may also have different names, but those skilled in the art can understand their meanings. For example, in the ORAN system, CU may also be called O-CU (open CU), DU may also be called O-DU, CU-CP may also be called O-CU-CP, CU-UP may also be called O-CU-UP, and RU may also be called O-RU. For the convenience of description, the embodiments of the present application are described by taking CU, CU-CP, CU-UP, DU and RU as examples. Any unit of CU (or CU-CP, CU-UP), DU and RU in the embodiments of the present application may be implemented by a software module, a hardware module, or a combination of a software module and a hardware module.

[0130] In a possible implementation, the terminal device in the embodiment of the present application may be a device for implementing a wireless communication function, such as a terminal or a chip that can be used in a terminal, etc. The terminal may be a user equipment (UE), an access terminal, a terminal unit, a terminal station, a mobile station, a mobile station, a remote station, a remote terminal, a mobile device, or a terminal agent, etc. in a 5G network or a future evolved public land mobile network (PLMN). The access terminal may be a cellular phone, a cordless phone, a session initiation protocol (SIP) phone, a wireless local loop (WLL) station, a personal digital assistant (PDA), a handheld device with wireless communication function, a computing device or other processing device connected to a wireless modem, an in-vehicle device, a wearable device, a VR terminal device, an AR terminal device, a wireless terminal in industrial control, a wireless terminal in self driving, a wireless terminal in remote medical, a wireless terminal in smart grid, a wireless terminal in transportation safety, a wireless terminal in smart city, a wireless terminal in smart home, etc. In a possible implementation, the terminal device may be mobile or fixed, which is not limited.

[0131] It can be understood that the above-mentioned communication system 300 can support a variety of different business application scenarios, such as enhanced mobile broadband (eMBB), ultra-high reliability and ultra-low latency communication (URLLC), massive machine type communication (mMTC), immersive communication, massive communication, ubiquitous connections, integrated artificial intelligence and communication, or integrated sensing and communication, etc. The embodiments of the present application do not make specific limitations on this.

[0132] The present application embodiment provides an information transmission method, and the execution subject of the method may be a first device. The first device may be Figure 3 The terminal device, or a module or unit of the terminal device (such as a chip, a chip system, a chip circuit, or a circuit of the terminal device), may also be an access network device, or a module or unit of the access network device (such as a chip, a chip system, a chip circuit, or a circuit of the access network device).

[0133] In a possible implementation, a first device obtains first data, the dimension of the first data is M dimensions, M is an integer greater than 1; the first device determines the dimension information corresponding to each dimension in P dimensions, the dimension information corresponding to each dimension is used to determine the dimension size of the full rank matrix corresponding to each dimension, and P dimensions are P dimensions in M ​​dimensions; the first device determines the full rank matrix and second data corresponding to each dimension according to the first data and the dimension information corresponding to each dimension; the first device sends the full rank matrix and second data corresponding to each dimension. In this way, the first device can first determine the P dimensions of the M dimensions of the first data (that is, which dimension needs to be reduced or compressed) and the dimension size of the full rank matrix corresponding to each dimension in P dimensions through the dimension information corresponding to each dimension in P dimensions, and then the dimension size of the full rank matrix corresponding to P dimensions and each dimension in P dimensions can be determined without performing HOSVD decomposition and iterative operations on high-dimensional data, so that the first device determines the second data and the full rank matrix corresponding to each dimension according to the dimension size of the full rank matrix corresponding to each dimension in P dimensions and the first data, which can significantly reduce the computational complexity of the first data compression.

[0134] The following will be combined Figures 4 to 7 , the above method provided in the embodiment of the present application is described in detail.

[0135] It should be understood that the signals between various devices or apparatuses, the names of various parameters in the signals, or the names of the information carried by the signals in the following embodiments of the present application are merely examples, and other names may also be used in specific implementations, and the embodiments of the present application do not specifically limit this.

[0136] In addition, the method provided in the embodiment of the present application can be applicable to the interaction between the first device and the second device. Figure 3 The terminal device, or a module or unit of the terminal device (such as a chip, a chip system, a chip circuit, or a circuit of the terminal device), the second device may be the aforementioned Figure 3 The first device may be an access network device in the embodiment of the present invention, or a module or unit of the access network device (such as a chip, a chip system, a chip circuit, or a circuit of the access network device). Alternatively, the first device may be the aforementioned Figure 3 The access network device, or the module or unit of the access network device, the second device may be the aforementioned Figure 3 A terminal device, or a module or unit of a terminal device.

[0137] The first device and the second device may operate in a high frequency band, such as a millimeter wave band or a terahertz band, or in a low frequency band, such as a 700 MHz, 900 MHz, 2.1 GHz, 2.6 GHz, or 3.5 GHz band, etc. It is understood that the first device and the second device may also operate in other frequency bands supported by the 6G system, and the embodiments of the present application do not specifically limit this.

[0138] It can be understood that the first device can operate in an RRC activated state, an RRC inactivated state, an RRC idle state, or other RRC states or RRC modes defined in the 6G communication system, and the embodiments of the present application do not specifically limit this.

[0139] For ease of understanding, the following takes the interaction between the first device and the second device as an example to explain in detail Figure 4 The data transmission method flow is shown.

[0140] Figure 4 is a flow chart of a data transmission method provided by an embodiment of the present application. Figure 4 As shown, the method comprises the following steps:

[0141] S401: A first device obtains first data, wherein the first data has M dimensions, where M is an integer greater than 1.

[0142] S402: The first device determines dimension information corresponding to each dimension of P dimensions. The dimension information corresponding to each dimension is used to determine the dimension size of the full rank matrix corresponding to each dimension. The P dimensions are P dimensions of the M dimensions.

[0143] S403. The first device determines the full rank matrix and the second data corresponding to each dimension based on the first data and the dimension information corresponding to each dimension.

[0144] It can be understood that in the embodiment of the present application, the second data includes data after the first data is compressed.

[0145] S404: The first device sends the full rank matrix and second data corresponding to each dimension to the second device. Correspondingly, the second device receives the full rank matrix and second data corresponding to each dimension from the first device.

[0146] S405: The second device obtains the dimension size of the first data.

[0147] S406. The second device determines fourth data according to the dimensional size of the first data, the second data, and the full rank matrix corresponding to each dimension.

[0148] It can be understood that the fourth data includes data obtained by decompressing the second data, and the second data includes data obtained by compressing the first data, that is, the fourth data may correspond to the first data.

[0149] The above steps S401 to S406 are described in detail below.

[0150] For step S401:

[0151] It can be understood that the first data can be two-dimensional data, three-dimensional data, or data with more dimensions. The dimension size of the first data is the product of the size of each dimension in the M dimensions, that is, N1×…×N i ×…×N M , 1≤i≤M, M is an integer greater than 1.

[0152] In an embodiment of the present application, the first data may be channel data (such as CSI data), perception data, imaging data, artificial intelligence AI data, or model data, etc., without specific limitation.

[0153] Exemplarily, for channel data, the dimensions of the channel data may include: RX dimension, TX dimension, frequency domain dimension, or time domain dimension, etc. Among them, the dimension size of the time domain dimension may refer to the number of time domain units, and the time domain unit may be a time domain unit of different granularities, such as a time slot or a symbol, etc. It can be understood that the dimension of the channel data may also include a Doppler dimension, which may be used to indicate the change of the channel over time, such as the Doppler frequency corresponding to the multipath component.

[0154] The perception data is similar to the channel data, except that the perception data can be pre-processed data. For example, for lidar data, the dimensions of the perception data can include the three-dimensional coordinate dimension of the lidar, the laser reflection intensity dimension, or the laser wavelength dimension. It can be understood that the perception data can be called point cloud data after processing, and the point cloud data can also be used for positioning, so the dimensions of the perception data can also include the geographic location dimension.

[0155] The imaging data is usually two-dimensional data, which may include a target echo delay dimension (corresponding to the target distance) and a frequency deviation dimension (corresponding to the target speed). Of course, the imaging data may also be three-dimensional data, which is a higher dimensional data, and the embodiments of the present application do not specifically limit this.

[0156] The dimensions of AI data or model data are related to features (or channels). For example, with respect to AI data of images, the dimensions of AI data may include: the number of images, the number of pixels in the length direction of the image, the number of pixels in the width direction of the image, and the number of channels of the image (e.g., RGB channels).

[0157] In a possible implementation, the first device may generate the first data. The first device may obtain the first data by collecting signals. For example, the first device may obtain the perception data by receiving an echo signal. It is understood that the first device may also obtain multiple image data through a sensor.

[0158] It is understandable that the first data may be processed data. For example, the first device may collect the sensing signal and perform preprocessing to generate the first data in the form of a point cloud.

[0159] In another possible implementation, the first device may obtain the first data from other devices, where the other devices may be, for example, other terminal devices, the second device, or network elements in a core network element, etc., which is not specifically limited in the embodiment of the present application.

[0160] For example, the first device may send a request message to other devices to request to obtain the first data, and then receive the first data from other devices. It is understandable that other devices may also directly send the first data to the first device, and the embodiments of the present application do not specifically limit this.

[0161] Corresponding to step S402:

[0162] It can be understood that 1≤P≤M, that is, the first device can determine the dimension information corresponding to each dimension of some dimensions in the M dimensions, and can also determine the dimension information corresponding to each dimension in the M dimensions. For example, for three-dimensional CSI data, the three dimensions are RX dimension, TX dimension, and frequency domain dimension, and the P dimensions can be the RX dimension of the three dimensions, or the RX dimension and TX dimension of the three dimensions, or the three dimensions.

[0163] For example, the i-th dimension in P dimensions corresponds to the j-th dimension in M ​​dimensions, 1≤i≤P, 1≤j≤M, i may be equal to j or may not be equal to j. If the dimension size of the i-th dimension in P dimensions is L i , the size of the jth dimension in the M dimensions is N j , then L i =N j For example, the first dimension among the P dimensions corresponds to the first dimension among the M dimensions, and the second dimension among the P dimensions corresponds to the third dimension among the M dimensions.

[0164] It should be understood that the full rank matrix corresponding to each dimension in the P dimensions can reflect (or characterize, or represent) the main features (or components, characteristics) corresponding to each dimension in the P dimensions. It can be understood that the column vectors or row vectors in the full rank matrix are linearly independent, that is, after the data is compressed (or reduced in dimension), the above-mentioned column vectors or row vectors can reflect the main features that can be retained in each dimension. In other words, the dimensional size of the full rank matrix corresponding to each dimension can reflect the dimensional size of each dimension after dimensionality reduction.

[0165] The following are two examples to illustrate full rank matrices.

[0166] Example 1:

[0167] The full rank matrix in the embodiment of the present application may be a column full rank matrix, that is, the dimension size of the full rank matrix may be L i ×N s,i , L i is the number of row vectors in the full rank matrix, L i It may refer to the dimension size of the i-th dimension among the P dimensions (that is, the dimension size of the j-th dimension corresponding to the first data). s,i is the number of column vectors in the full rank matrix, N s,i It can refer to the dimension size after dimensionality reduction of the i-th dimension, L i Can be greater than N s,i , N i and N s,iAll are positive integers. It can be understood that the column full-rank matrix corresponding to each dimension can be used for the first data to perform right multiplication calculations between tensors and matrices, so that a core tensor can be obtained, which can reflect the degree of interaction between the main features corresponding to different dimensions in the P dimensions, that is, the data after the first data is reduced in dimension (that is, the second data in step S403), that is, the first data can be decomposed into a core tensor and a full-rank matrix corresponding to each dimension.

[0168] Example 2:

[0169] A full rank matrix can be a row full rank matrix, that is, the dimension size of a full rank matrix can be N s,i ×L i , N s,i is the number of row vectors in the full rank matrix, N s,i It can refer to the dimension size after dimensionality reduction of the i-th dimension. i is the number of column vectors in the full rank matrix, L i It may refer to the dimension size of the i-th dimension in the P dimensions. It can be understood that the row full rank matrix corresponding to each dimension can be used for the first data to perform left multiplication calculation of the tensor and the matrix, and the core tensor can also be obtained.

[0170] It can be understood that in the embodiment of the present application, the full rank matrix can be a vector; or, the full rank matrix can include at least two vectors, and any two of the at least two vectors are linearly independent. In other words, the full rank matrix can also be replaced by an orthogonal matrix, a set of orthogonal vectors (such as column vectors or row vectors), a quasi-orthogonal matrix, or a set of quasi-orthogonal vectors, etc., and the embodiment of the present application does not specifically limit this.

[0171] For example, taking the TX dimension in the CSI data of the downlink channel estimation as an example, the full rank matrix corresponding to the TX dimension can reflect the subspace characteristics corresponding to the TX dimension. Further, the dimension size of the TX dimension can be 1024, and the dimension size of the column full rank matrix corresponding to the TX dimension is 1024×128, so that the rank (or the number of column vectors, or the number of row vectors, etc.) of the column full rank matrix corresponding to the TX dimension is 128, and the dimension size of the TX dimension 1024 after dimensionality reduction processing is 128.

[0172] It should be understood that Figure 1The data compression scheme shown, because the three dimensions are not decomposed by three-dimensional SVD, the correlation is not fully utilized, which will lose compression efficiency. Based on this, high-dimensional (greater than 2 dimensions) SVD decomposition can be achieved through higher-order SVD (HOSVD). HOSVD decomposition can decompose high-dimensional data into a core tensor and a full-rank matrix corresponding to each dimension in multiple dimensions. However, in the process of decomposition, it is necessary to continuously iterate (or optimize) to obtain an accurate optimal full-rank matrix, that is, to determine the full-rank matrix corresponding to each dimension. In other words, in the process of compressing data through HOSVD decomposition, it is necessary to determine through iterative operations which dimensions (for example, P dimensions) among the M dimensions correspond to which main features that need to be retained, that is, the full-rank matrix corresponding to each dimension in the P dimensions.

[0173] In an embodiment of the present application, the method for determining the size of the full-rank matrix corresponding to each dimension is different from the above-mentioned HOSVD decomposition. The difference is that: the first device can first determine P dimensions of the M dimensions of the first data (that is, which dimension needs to be reduced in dimension) and the dimensional size of the full-rank matrix corresponding to each dimension of the P dimensions through the dimensional information corresponding to each dimension of the P dimensions. Therefore, there is no need to perform HOSVD decomposition and iterative operations on the high-dimensional data to determine the dimensional size of the full-rank matrix corresponding to the P dimensions and each dimension of the P dimensions, thereby significantly reducing the amount of calculation.

[0174] In a possible implementation, the dimension information corresponding to each dimension includes the dimension information corresponding to the i-th dimension among the P dimensions, and the dimension information corresponding to the i-th dimension is used to indicate: the dimension sizes of the other dimensions except the i-th dimension in the full rank matrix corresponding to the i-th dimension, 1≤i≤P. It can be understood that the dimension sizes of the other dimensions except the i-th dimension in the full rank matrix corresponding to the i-th dimension can be used to indicate the dimension size of the i-th dimension after dimensionality reduction, that is, the dimension sizes of the other dimensions except the i-th dimension in the full rank matrix corresponding to the i-th dimension can be replaced by: the dimension size of the i-th dimension after dimensionality reduction, or the rank of the full rank matrix corresponding to the i-th dimension.

[0175] That is to say, the dimensional information corresponding to each dimension can be used to indicate the dimensional size of each dimension after dimensionality reduction, so that the first device does not need to decompose and iterate the first data to determine the core tensor of the first data after dimensionality reduction and the dimensional size of each dimension after dimensionality reduction, thereby significantly reducing the amount of calculation and reducing the implementation complexity of data compression.

[0176] For example, for the above example 1, the full rank matrix is ​​a column full rank matrix, the dimensions of the full rank matrix other than the i-th dimension are the column dimensions of the full rank matrix, and the i-th dimension is the row dimension of the full rank matrix. For another example, for the above example 2, the full rank matrix is ​​a row full rank matrix, the dimensions of the full rank matrix other than the i-th dimension are the row dimensions of the full rank matrix, and the i-th dimension is the column dimension of the full rank matrix.

[0177] It should be understood that for ease of understanding, the technical solution described in detail below is explained using a full rank matrix as a column full rank matrix as an example. The implementation method corresponding to the row full rank matrix is ​​similar to the implementation method corresponding to the column full rank matrix, and is described here in a unified manner without further elaboration.

[0178] In a possible implementation, the dimension information corresponding to the i-th dimension is used to indicate: the dimensional size of other dimensions except the i-th dimension in the full-rank matrix corresponding to the i-th dimension, including: the scaling factor corresponding to the i-th dimension, the scaling factor corresponding to the i-th dimension is the ratio between the dimensional size of the i-th dimension and the dimensional size of the i-th dimension after dimensionality reduction; or, the scaling factor corresponding to the i-th dimension is the ratio between the dimensional size of the i-th dimension and the rank of the full-rank matrix corresponding to the i-th dimension. That is to say, the first device can determine the dimensional size of each dimension after dimensionality reduction based on the scaling factor corresponding to each dimension in the P dimensions.

[0179] It can be understood that the scale factor = the dimension size of the i-th dimension after dimensionality reduction ÷ the dimension size of the i-th dimension. For the scale factor being a linear value, the value range of the scale factor can be (0,1], that is, the larger the scale factor, the larger the dimension size after dimensionality reduction, and the lower the data compression rate. Of course, for the scale factor being a logarithmic value, the value unit of the scale factor can be 0 decibel (dB), -1dB, -3dB, or -4.77dB, etc., and the embodiment of the present application does not make specific limitations on this.

[0180] It can also be understood that the scaling factor = the dimension size of the i-th dimension ÷ the dimension size of the i-th dimension after dimensionality reduction. In this way, for the scaling factor to be a linear value, the scaling factor is greater than 1, that is, the larger the scaling factor, the smaller the dimension size after dimensionality reduction, and the higher the data compression rate; for the scaling factor to be a logarithmic value, the scaling factor can be 1dB, 3dB, or 4.77dB, etc., which are greater than 0dB, and the embodiments of the present application do not make specific limitations on this.

[0181] Optionally, the scaling factor corresponding to each dimension in the P dimensions may be the same. That is to say, the ratio of dimensionality reduction of each dimension in the P dimensions is the same, and then the dimensionality size of each dimension after dimensionality reduction can be uniformly adjusted to reduce the implementation complexity. For example, for CSI data of RX dimension, TX dimension, and frequency domain dimension, the dimension size of RX dimension is 32, the dimension size of TX dimension is 1024, and the dimension size of frequency domain dimension is 18. For scaling factor = dimension size after dimensionality reduction of the i-th dimension ÷ dimension size of the i-th dimension, assuming that the scaling factor is 0.5, the first device can determine that the dimension size after dimensionality reduction of the RX dimension is 16, the dimension size after dimensionality reduction of the TX dimension is 512, and the dimension size after dimensionality reduction of the frequency domain dimension is 9 according to the scaling factor 0.5.

[0182] Alternatively, optionally, the scaling factor corresponding to each dimension in the P dimensions is different. That is to say, the ratio of dimensionality reduction of each dimension in the P dimensions may be different, and thus the dimension size of each dimension after dimensionality reduction may be flexibly adjusted according to the original dimension size of each dimension in the P dimensions. For example, for the above example, since the TX dimension is larger, the corresponding redundant information corresponding to the TX dimension is more, and thus the dimension size after dimensionality reduction of the TX dimension may be smaller, such as 128, so the scaling factor is 0.125. Furthermore, since the dimension size of the frequency domain dimension is smaller, the corresponding redundant information corresponding to the frequency domain dimension may be less, and thus the dimension size after dimensionality reduction of the frequency domain dimension is 9, and thus the scaling factor is 0.5.

[0183] It can be understood that the dimension size of each dimension after dimensionality reduction can be the rounded value of the product between the dimension size of each dimension and the scale factor. The rounded value can be rounded up or rounded down, which is not specifically limited in the embodiments of the present application.

[0184] It should be understood that the dimensional information corresponding to each dimension may be determined by the first device; or, the dimensional information corresponding to each dimension may be indicated. Among them, the dimensional information corresponding to each dimension may be indicated, which may mean that the network side may indicate the dimensional information corresponding to each dimension to the first device. The network side may be a second device or a network element in the core network (such as a mobile management network element, a positioning management network element, a session management network element, or a policy management network element, etc.), and the embodiments of the present application do not specifically limit this.

[0185] It should also be understood that the dimension information corresponding to each dimension may be indicative, and details may be referred to the relevant description of step S409 below, which will not be repeated here.

[0186] For example, the first device determines the dimensional information corresponding to each dimension (i.e., step S402), including: the first device determines the proportional factor corresponding to each dimension according to the data size of the first data before dimensionality reduction (i.e., before compression) and the data amount carried by the resources used to transmit the first data. It can be understood that the first device can determine the reduced dimension size of each dimension after dimensionality reduction based on the dimensional size of each dimension and the proportional factor corresponding to each dimension. In addition, the data amount of the first data is proportional to the dimensional size of the first data, which is uniformly explained here and will not be repeated below.

[0187] It is understood that the amount of data carried by the resource for transmitting the first data can determine the size of the code block (CB) according to the transport block size (TB size, TBS) and the code rate (code rate) of the transport block (TB), so that the first device determines the amount of data carried by the resource for transmitting the first data according to the size of the CB. It should be understood that the amount of data after dimensionality reduction (i.e., compression) of the first data is less than or equal to the amount of data carried by the resource for transmitting the first data.

[0188] The above-mentioned resources for transmitting the first data may include at least one of the following: time domain resources for transmitting the first data, frequency domain resources for transmitting the first data, or a modulation and coding scheme (MCS) for transmitting the first data.

[0189] In addition, when the first device is a terminal device, the resources used to transmit the first data may be resources pre-configured by the network for the first device, such as resources pre-configured in grant gree transmission, and grant gree transmission may, for example, include transmission based on pre-configured uplink resources (pre-configured uplink resource, PUR) and configured grant (configured grant, CG) transmission, etc.; or, the resources used to transmit the first data may be resources dynamically scheduled by the network side, such as resources dynamically configured by the second device to the first device through downlink control information (downlink control information, DCI).

[0190] When the first device is an access network device, the resources used to transmit the first data may be resources estimated by the first device based on the dimension size of the first data and the compression rate of the first data, so that the first device can determine the dimension information corresponding to each dimension based on the compression rate of the first data. The compression rate of the first data may refer to: the ratio of the amount of data after dimensionality reduction of the first data to the amount of data before dimensionality reduction of the first data, or the ratio of the dimensional size of the first data after dimensionality reduction to the dimensional size of the first data before dimensionality reduction.

[0191] It can be understood that the compression rate of the first data can be pre-configured; or, the compression rate of the first data can be negotiated in advance between the first device and the second device; or, the compression rate of the first data can be indicated by the second device, and the embodiments of the present application do not specifically limit this.

[0192] For another example, the first device determines the dimensional information corresponding to each dimension, including: the first device determines the proportional factor corresponding to each dimension according to the dimensional size of the first data and the compression rate of the first data. For example, for CSI data of RX dimension, TX dimension, and frequency domain dimension, the dimensional size of the RX dimension is 32, the dimensional size of the TX dimension is 1024, and the dimensional size of the frequency domain dimension is 18, that is, the dimensional size of the first data before dimensionality reduction is 32×1024×18. Assuming that the compression rate of the first data is 0.125, the dimensional size of the first data after dimensionality reduction is equal to 0.125×(32×1024×18). An example is that the proportional factor corresponding to each dimension is the same, and then the proportional factor is equal to the compression rate of the first data, which is 0.5. In this way, the first device can determine that the dimensional size of the RX dimension after dimensionality reduction is 16, the size of the TX dimension after dimensionality reduction is 512, and the size of the frequency domain dimension after dimensionality reduction is 9 according to the compression rate of the first data.

[0193] Another example is that the scaling factors corresponding to each dimension are different. For example, the dimension size of the frequency domain dimension is smaller, and the corresponding redundant information corresponding to the frequency domain dimension may be less. Considering the wireless environment where the CSI changes dramatically with frequency, the first device and / or the second device expects to retain at least 80% of the information corresponding to the frequency domain dimension. Then, the first device can set the scaling factor corresponding to the frequency domain dimension to 0.889, so that the dimension size after the frequency domain dimension is reduced is 16. Since the TX dimension is larger and has more redundant information, the dimension size after the TX dimension is reduced can be 256, and the scaling factor corresponding to the TX dimension is 0.250. In this way, the dimension size after the RX dimension is reduced is 18, that is, the scaling factor corresponding to the RX dimension is 0.563.

[0194] It can be understood that in the embodiment of the present application, the dimension information corresponding to each dimension is associated with the compression rate of the first data. For example, the product of the scaling factors corresponding to each dimension can be equal to the compression rate of the first data.

[0195] It should be understood that the above implementation method can also be used for other types of data besides CSI data, such as perception data or AI data, etc., and the embodiments of the present application do not specifically limit this.

[0196] It should also be understood that the above implementation method for the first device to determine the dimensional information corresponding to each dimension is only an example. The first device or the network side can also adopt other implementation methods to determine the dimensional information corresponding to each dimension. The embodiments of the present application do not make specific limitations on this.

[0197] In a possible implementation, the dimension information corresponding to each dimension is associated with one or more of the following parameters:

[0198] A time domain resource for transmitting the first data, a frequency domain resource for transmitting the first data, a modulation and coding scheme (MCS) for transmitting the first data, and a compression rate of the first data.

[0199] It can be understood that the MCS used to transmit the first data can indicate: a target code rate for transmitting the first data, a modulation method for transmitting the first data, and a modulation order for transmitting the first data. Among them, the first device can determine the amount of data carried by the resources used to transmit the first data based on parameters such as the MCS used to transmit the first data, the working bandwidth, the signal-to-noise ratio, the rank (or stream), or the number of layers. For details, please refer to the relevant description of the technical specification (technology specific, TS) 38.214 in 3GPP, which will not be repeated here.

[0200] That is to say, when the MCS used to transmit the first data and the number of frequency domain units allocated on each time domain unit are fixed, the amount of data carried by the resources used to transmit the first data is proportional to the time domain resources used to transmit the first data. Similarly, the amount of data carried by the resources used to transmit the first data is proportional to the frequency domain resources used to transmit the first data, and the amount of data carried by the resources used to transmit the first data is proportional to the MCS used to transmit the first data.

[0201] Furthermore, in some cases (for example, the amount of data after dimensionality reduction of the first data is directly proportional to the amount of data carried by the resources used to transmit the first data), the size of each dimension after dimensionality reduction is roughly directly proportional to the amount of data carried by the resources used to transmit the first data, or, the proportional factor corresponding to each dimension is roughly inversely proportional to the amount of data carried by the resources used to transmit the first data.

[0202] For step S403:

[0203] It should be understood that the second data in step S403 may be the core tensor in the aforementioned step S402, that is, the second data may reflect the degree of interaction between the main features (full rank matrices) corresponding to different dimensions among the P dimensions.

[0204] In one possible implementation, the second data and the full rank matrix corresponding to each dimension are obtained by decomposing the first data according to the dimensional information corresponding to each dimension. For example, the first device determines the full rank matrix and the second data corresponding to each dimension (i.e., step S403) according to the first data and the dimensional information corresponding to each dimension, including: the first device decomposes the first data according to the dimensional information corresponding to each dimension to obtain the full rank matrix and the second data corresponding to each dimension.

[0205] That is to say, the first device can determine the dimensional size of the full rank matrix corresponding to each dimension based on the dimensional information corresponding to each dimension, and then decompose the first data according to the dimensional size of the full rank matrix corresponding to each dimension to obtain the second data (i.e., the core tensor) and the full rank matrix corresponding to each dimension, which can reduce the amount of computation required by the first device to decompose the first data.

[0206] In another possible implementation, the second data is determined based on the first data and the full rank matrix corresponding to each dimension, and the full rank matrix corresponding to each dimension is determined based on the first data and the dimensional information corresponding to each dimension.

[0207] For example, the first device determines the full rank matrix corresponding to each dimension and the second data according to the first data and the dimension information corresponding to each dimension (i.e., step S403), including: the first device determines the full rank matrix corresponding to each dimension according to the first data and the dimension information corresponding to each dimension; the first device determines the second data according to the full rank matrix corresponding to each dimension and the first data.

[0208] That is to say, the first device can decompose the first data according to the dimensional information corresponding to each dimension to obtain the full rank matrix corresponding to each dimension, and obtain the second data (i.e., the core tensor) according to the full rank matrix corresponding to each dimension and the first data. Compared with performing HOSVD decomposition on the first data and simultaneously obtaining the second data and the full rank matrix corresponding to each dimension, the complexity of decomposing high-dimensional data can be reduced.

[0209] For example, the first device determines the second data based on the full rank matrix corresponding to each dimension and the first data, which may include: the first device performs a tensor multiplication of the first data with the full rank matrix corresponding to each dimension, thereby obtaining the second data. For example, assuming that the i-th dimension in P dimensions corresponds to the j-th dimension in M ​​dimensions, 1≤i≤P, 1≤j≤M. The full rank matrices corresponding to each dimension in P dimensions are: Q1, Q2, Q3, ...Q i ,…,Q P , Q1 is the full rank matrix corresponding to the first dimension in M ​​dimensions (i.e., the first dimension in P dimensions), Q2 is the full rank matrix corresponding to the third dimension in M ​​dimensions (i.e., the second dimension in P dimensions), Qi is the full rank matrix corresponding to the jth dimension in M ​​dimensions (i.e. the ith dimension in P dimensions), Q P is the full rank matrix corresponding to the Mth dimension among M dimensions (i.e., the Pth dimension among P dimensions). i The row dimension × column dimension of is: L i ×N s,i , L i Actually, it is the dimension size N of the jth dimension among the M dimensions. j , N s,i is the dimension size after dimensionality reduction of the jth dimension.

[0210] Furthermore, the relationship between the first data H, the full rank matrix corresponding to each dimension, and the second data Σ can be determined by formula (1).

[0211] Σ=H×1Q1 H ×3Q2 H …× j Q i H …× M Q P H Formula (1)

[0212] Formula (1) indicates that the second data Σ is the product of the first data H and the conjugate transposed matrix of the full rank matrix corresponding to each dimension. i H Indicates Q i The conjugate transposed matrix of j Q i H Represents the conjugate transposed matrix of the full rank matrix corresponding to the jth dimension in M ​​dimensions, H× j Q i H Indicates the first data H and Q i The conjugate transpose matrix of the tensor performs mode-j multiplication.

[0213] It can be understood that, for formula (1), the number of dimensions of the second data Σ is the same as the number of dimensions of the first data H (both are M), and the dimension size of the second data Σ is N s,1 ×N2×N s,3 ×…×N s,j …×N s,M , N s,1 is the dimension size of the first dimension after dimensionality reduction among the M dimensions, that is, the dimension size of the second data Σ is smaller than the dimension size of the first data H (N1×…×N j ×…×N M ).

[0214] It should be understood that formula (1) is only an example, and other methods may be used to determine the second data, such as converting the first data H into multiple matrices, performing Kronecker products or Khatri–Rao products on the multiple matrices with the full-rank matrix corresponding to each dimension, etc. The embodiments of the present application do not specifically limit this.

[0215] In a possible implementation, the full rank matrix corresponding to each dimension is a full rank matrix of a feature matrix corresponding to each dimension, and the feature matrix corresponding to each dimension is determined based on the first data and dimension information corresponding to each dimension.

[0216] For example, the first device determines the full rank matrix corresponding to each dimension based on the first data and the dimension information corresponding to each dimension, including: the first device determines the feature matrix corresponding to each dimension based on the first data and the dimension information corresponding to each dimension; the first device decomposes the feature matrix corresponding to each dimension to obtain the full rank matrix of the feature matrix corresponding to each dimension.

[0217] Among them, the first device can use orthogonal triangular decomposition (orthogonal triangular decomposition / QR decomposition) or other decomposition methods (such as lower-upper triangular decomposition (lower-upper decomposition, LU decomposition)) to decompose the feature matrix corresponding to each dimension into a full rank matrix and other forms of matrices.

[0218] In addition, the dimension size between the feature matrix corresponding to each dimension and its full rank matrix is ​​the same. For example, assuming that the i-th dimension in P dimensions corresponds to the j-th dimension in M ​​dimensions, the full rank matrix Q corresponding to the i-th dimension in P dimensions is i The dimension size is L i ×N s,i , L i is the dimension size of the i-th dimension, which is equal to the dimension size N of the j-th dimension in the first data j , N s,i is the size of the dimensions other than the i-th dimension in the full-rank matrix corresponding to the i-th dimension, that is, the full-rank matrix Q corresponding to the i-th dimension in P dimensions i The dimension size is N j ×N s,i , that is, the dimension size of the feature matrix corresponding to the i-th dimension is N j ×N s,i .

[0219] It can be understood that the dimension size of the feature matrix corresponding to the i-th dimension is N j ×Ns,i , it can be seen that the feature matrix corresponding to the i-th dimension is the subspace feature matrix corresponding to the j-th dimension in the first data, which retains the main features corresponding to the j-th dimension, that is, N s,i <N j .

[0220] That is to say, the first device can perform dimensionality reduction and feature extraction (or feature selection) on the first data according to the dimensional information corresponding to each dimension to obtain a feature matrix corresponding to each dimension. Then the first device can directly decompose the feature matrix corresponding to each dimension to obtain a full rank matrix corresponding to each feature matrix, which can reduce the implementation complexity and computational complexity of the first device in determining the full rank matrix corresponding to each dimension.

[0221] In one possible implementation, the feature matrix corresponding to each dimension is obtained by performing a feature extraction operation on the first data. The feature extraction operation may include convolution; or, the feature extraction operation may include convolution and pooling; or, the feature extraction operation may include inputting the first data and the dimension information corresponding to each dimension into the first model. The output result of the first model may include the feature matrix corresponding to each dimension. The first model may be a neural network model, an AI model, or an ML model, etc., which is not specifically limited in the embodiments of the present application.

[0222] It can be understood that the first model can be a pre-configured model; or, the first model can be configured on the network side, and the embodiments of the present application do not specifically limit this.

[0223] It can also be understood that for convolution, the convolution kernel corresponding to each dimension is associated with the dimension information corresponding to each dimension. For example, the dimension information corresponding to the i-th dimension is used to indicate: the dimension size N of the column dimension of the full rank matrix corresponding to the i-th dimension s,i , N s,i Equal to the column dimension size of the feature matrix output by the convolution kernel.

[0224] In another possible implementation, the feature matrix corresponding to each dimension is determined based on the first data and the dimension information corresponding to each dimension, including: the feature matrix corresponding to each dimension is determined based on the first data and the dimension reduction matrix corresponding to each dimension, wherein the dimension reduction matrix corresponding to each dimension is associated with the dimension information corresponding to each dimension.

[0225] It should be understood that the dimension reduction matrix corresponding to each dimension is associated with the dimension information corresponding to each dimension, which may mean that the rank of the dimension reduction matrix corresponding to each dimension is equal to the rank of the full rank matrix determined by the dimension information corresponding to each dimension. In other words, the first device can determine the dimension size of the dimension reduction matrix corresponding to each dimension in the P dimensions based on the dimension information corresponding to one or more dimensions in the P dimensions.

[0226] In a possible implementation, the dimension reduction matrix corresponding to each dimension may be a column full rank matrix (or row full rank matrix), and the dimension size of the column dimension (or row dimension) of the dimension reduction matrix corresponding to each dimension is equal to the dimension size of the column dimension (or row dimension) of the full rank matrix corresponding to each dimension. In other words, the dimension reduction matrix corresponding to each dimension can be used to reduce the dimension size of some dimensions in the first data, so that the dimension size of the column dimension (or row dimension) in the feature matrix corresponding to each dimension is equal to the dimension size after the dimension reduction of each dimension.

[0227] For example, for the above example, the dimension size of the feature matrix corresponding to the i-th dimension of the P dimensions is N j ×N s,i , the dimension reduction matrix can reduce the dimension sizes of other dimensions except j dimensions in the M dimensions of the first data, so that the dimension size of the row dimension or column dimension of the feature matrix corresponding to the i-th dimension is equal to N s,i , the N s,i The dimension information corresponding to the i-th dimension indicates the dimension size of the column dimension of the full rank matrix corresponding to the i-th dimension.

[0228] For example, the first device determines the feature matrix corresponding to each dimension based on the first data and the dimension information corresponding to each dimension, including: the first device determines the dimensionality reduction matrix corresponding to each dimension; the first device determines the feature matrix corresponding to each dimension based on the first data and the dimensionality reduction matrix corresponding to each dimension.

[0229] That is to say, the first device performs dimensionality reduction processing on the first data according to the dimensionality reduction matrix corresponding to each dimension, which can reduce the implementation complexity of the first device determining the feature matrix corresponding to each dimension.

[0230] It can be understood that in the embodiment of the present application, the name of the dimensionality reduction matrix is ​​only an example and can also be other names (such as projection matrix), and the embodiment of the present application does not make specific limitations on this.

[0231] The following describes how to determine the feature matrix and how to determine the dimension reduction matrix.

[0232] A. Determine the implementation method of the feature matrix:

[0233] It should be understood that in the embodiment of the present application, for M>2, the first device can adopt method 1 and method 2 to determine the feature matrix corresponding to each dimension. Method 1 and method 2 are introduced below respectively.

[0234] Method 1: The first device first performs tensor expansion on the first data, and then determines the feature matrix according to the dimension reduction matrix.

[0235] In a possible implementation, when M>2, the feature matrix corresponding to each dimension is determined based on the first data and the dimension reduction matrix corresponding to each dimension, including: the feature matrix corresponding to each dimension is determined based on the two-dimensional data corresponding to each dimension and the dimension reduction matrix corresponding to each dimension. The two-dimensional data corresponding to each dimension includes the two-dimensional data corresponding to the i-th dimension among the P dimensions, the two-dimensional data corresponding to the i-th dimension is the two-dimensional data obtained by merging the other dimensions in the first data except the i-th dimension, the two-dimensional data corresponding to the i-th dimension has the same elements as the first data, and 1≤i≤P.

[0236] For example, the first device determines the feature matrix corresponding to each dimension based on the first data and the dimensionality reduction matrix corresponding to each dimension, including: the first device performs tensor unfolding / flattening on the first data to obtain two-dimensional data corresponding to each dimension; the first device determines the feature matrix corresponding to each dimension based on the two-dimensional data corresponding to each dimension and the dimensionality reduction matrix corresponding to each dimension.

[0237] It should be understood that tensor expansion may refer to retaining a certain dimension in high-dimensional data and merging (e.g., concatenating) other dimensions, so that a two-dimensional data can be obtained, and the elements between the two-dimensional data after tensor expansion are the same as those of the high-dimensional data.

[0238] For example, the number of dimensions of the first data H is M=3, and the first dimension of the three dimensions is tensor expanded as an example. Figure 5 As shown, the dimension size of the first data H is N1×N2×N3, that is, the first data H can be considered to include N3 two-dimensional data W, and the dimension size of the two-dimensional data W is N1×N2. In this way, the N3 two-dimensional data W are concatenated to obtain the two-dimensional data W' corresponding to the first dimension, and its dimension size is N1×(N2×N3), that is, the dimension size of the row dimension is N1, and the dimension size of the column dimension is N2×N3, and the elements contained in the first data H and the two-dimensional data W' corresponding to the first dimension are the same.

[0239] It is understandable that the above Figure 5 In the example shown, the first data H can also be considered to include N2 two-dimensional data T, and the dimension size of the two-dimensional data T is N1×N3. In this way, the N2 two-dimensional data T are spliced ​​to obtain the two-dimensional data T' corresponding to the first dimension, and its dimension size is N1×(N3×N2). Among them, the two-dimensional data T' and the two-dimensional data W' contain the same elements, but the position arrangement is different.

[0240] It should be understood that the above implementation method of tensor expansion is only an example, and the first device can also use other methods to implement tensor expansion, and the embodiments of the present application do not make specific limitations on this.

[0241] It should also be understood that the name of the tensor expansion in the embodiments of the present application is only an example, and the tensor expansion can also be replaced by matrix expansion, tensor expansion by dimension, tensor mode-n expansion, or matricization, etc. The embodiments of the present application do not make specific limitations on this.

[0242] Without loss of generality, for the two-dimensional data corresponding to the jth dimension among the M dimensions, its row dimension × column dimension is: N j ×(ε), where ε is the product of the dimensions of the M dimensions except the jth dimension. For example, for the two-dimensional data corresponding to the first dimension, ε = N2×N3×…×N M , and the two-dimensional data corresponding to the first dimension has a row dimension × column dimension of: N1×(N2×N3×…×N M ). Similarly, for the two-dimensional data corresponding to the second dimension, its row dimension × column dimension is: N2×(N1×N3×…×N M ); For the two-dimensional data corresponding to the Mth dimension, its row dimension × column dimension is: N M ×(N1×N2×…×N M-1 ).

[0243] It should be understood that the implementation method in which the first device determines the two-dimensional data corresponding to each dimension based on the first data is only an example, and other methods can also be used for implementation. The embodiments of the present application do not specifically limit this.

[0244] That is to say, the first device can first process the first data, that is, process the first data into two-dimensional data corresponding to each dimension of P dimensions, and then reduce the number of dimensions of the first data to two dimensions, which makes it easy to calculate the reduced dimension matrix corresponding to each dimension, thereby reducing the computational complexity of the first device in determining the feature matrix corresponding to each dimension.

[0245] In a possible implementation, the feature matrix corresponding to each dimension is the product of the two-dimensional data corresponding to each dimension and the dimension reduction matrix corresponding to each dimension. It can be understood that, as in the above example of the dimension reduction matrix, the dimension reduction matrix can reduce the dimension sizes of the other dimensions except the j dimensions in the M dimensions of the first data, so that the dimension size of the row dimension or column dimension of the feature matrix corresponding to the i-th dimension in the P dimensions is N. s,i In other words, the row or column dimension of the reduced matrix has a dimension size of N s,i .

[0246] It can also be understood that, in order to realize the product between the two-dimensional data and the reduced dimension matrix, the dimension size of the row dimension (or column dimension) of the reduced dimension matrix is ​​equal to the column dimension (or row dimension) of the two-dimensional data. For example, for the two-dimensional data right-multiplied by the reduced dimension matrix, the reduced dimension matrix is ​​a column-full rank matrix, and the dimension size of the column dimension of the two-dimensional data is equal to the dimension size of the row dimension of the reduced dimension matrix (i.e., the above-mentioned ε). For another example, for the two-dimensional data left-multiplied by the reduced dimension matrix, the reduced dimension matrix is ​​a row-full rank matrix, and the dimension size of the row dimension of the two-dimensional data is equal to the dimension size of the column dimension of the reduced dimension matrix.

[0247] That is to say, for method 1, the dimension size of the dimensionality reduction matrix corresponding to each dimension is determined according to the dimension information corresponding to each dimension.

[0248] For example, as mentioned above Figure 5 In the example shown, for the two-dimensional data corresponding to the jth dimension among the M dimensions, the row dimension × column dimension is: N j ×(ε). For the two-dimensional data corresponding to the jth dimension, right-multiply the dimension reduction matrix. The row dimension × column dimension of the dimension reduction matrix is: (ε)×N s,i , and then the row dimension × column dimension of the feature matrix is: N j ×N s,i , that is, the feature matrix is ​​a column full rank matrix.

[0249] It can be understood that the two-dimensional data corresponding to the jth dimension can be transposed to obtain the row dimension × column dimension of (ε) × N j , and multiply the dimension reduction matrix on the left. The row dimension × column dimension of the dimension reduction matrix is: N s,i ×(ε), and then the row dimension × column dimension of the feature matrix is: N s,i ×N j , that is, the characteristic matrix is ​​a full row rank matrix.

[0250] Exemplarily, taking mode 1 as an example, the first device determines the full rank matrix and the second data corresponding to each dimension. Figure 6 As shown, the first device can determine according to each dimension information of the P dimensions: the P dimensions can be the first dimension, the third dimension, and the Mth dimension of the M dimensions in the first data H, that is, P=3. Among them, the dimension size of the first dimension of the M dimensions determined by the dimension information corresponding to the first dimension of the P dimensions after dimensionality reduction is N s,1 , the dimension size of the third dimension in the M dimensions determined by the dimension information corresponding to the second dimension in the P dimensions after dimensionality reduction is N s,2 , the dimension information corresponding to the third dimension in P dimensions determines the dimension size of the Mth dimension in M ​​dimensions after dimensionality reduction to N s,3 .

[0251] The first device first performs tensor expansion on the first data H according to the dimensional information corresponding to each dimension in the P dimensions, and can obtain the two-dimensional data corresponding to each dimension. The two-dimensional data corresponding to each dimension is as follows:

[0252] The two-dimensional data corresponding to the first dimension of P dimensions has a row dimension × column dimension of: N1×(N2×N3×…×N M );

[0253] The two-dimensional data corresponding to the second dimension of P dimensions has a row dimension × column dimension of: N3×(N1×N2×…×N M );

[0254] And, the two-dimensional data corresponding to the third dimension of P dimensions has a row dimension × column dimension of: N M ×(N1×N2×…×N M-1 ).

[0255] The first device may determine the dimension reduction matrix corresponding to each dimension based on the two-dimensional data corresponding to each dimension and the dimension information corresponding to each dimension. The dimension reduction matrix corresponding to each dimension is as follows:

[0256] The dimension reduction matrix corresponding to the first dimension of P dimensions has the row dimension × column dimension: (N2×N3×…×N M )×N s,1 ;

[0257] The dimension reduction matrix corresponding to the second dimension of P dimensions has the row dimension × column dimension: (N1×N2×…×N M )×N s,2 ;

[0258] And, the dimensional matrix corresponding to the third dimension among the P dimensions has a row dimension × column dimension of (N1×N2×…×N M-1 )×N s,3 .

[0259] Afterwards, the first device multiplies the two-dimensional matrix corresponding to each dimension by the dimension reduction matrix corresponding to each dimension to obtain a feature matrix corresponding to each dimension. The feature matrix corresponding to each dimension is as follows:

[0260] The feature matrix corresponding to the first dimension of P dimensions has a row dimension × column dimension of: N1×N s,1 ;

[0261] The feature matrix corresponding to the second dimension of the P dimensions has a row dimension × column dimension of: N3×N s,2 ;

[0262] And, the dimension matrix corresponding to the third dimension among the P dimensions has a row dimension × column dimension of: N M ×N s,3 .

[0263] Afterwards, the first device performs orthogonal decomposition (such as orthogonal triangular decomposition) on the feature matrix corresponding to each dimension to obtain a full rank matrix corresponding to each dimension. The full rank matrix corresponding to each dimension is as follows:

[0264] The full rank matrix Q1 corresponding to the first dimension of P dimensions has a row dimension × column dimension of: N1×N s,1 ;

[0265] The full rank matrix Q2 corresponding to the second dimension of the P dimensions has a row dimension × column dimension of: N3×N s,2 ;

[0266] And, the full rank matrix Q3 corresponding to the third dimension among the P dimensions has a row dimension × column dimension of: N M ×N s,3 .

[0267] Finally, the first device can use the above formula (1) to determine the second data Σ = H × 1Q1 H ×3Q2 H × M Q3 H , the dimension size of the second data Σ is N s,1 ×N2×N s,2 ×N4…×N s,3 , the dimension size of Σ is smaller than the dimension size of the first data H N1×N2×N3×N4×…×N M .

[0268] It should be understood that the above Figure 6 This is only an example of method 1. The first device can also use other methods to implement method 1. For example, the two-dimensional matrix corresponding to each dimension can be transposed, and the transposed matrix can be left multiplied by the dimensionality reduction matrix corresponding to each dimension. The embodiment of the present application does not make specific limitations on this.

[0269] Method 2: The first device first determines the feature tensor according to the dimension reduction matrix, and then performs tensor expansion on the feature tensor to obtain the feature matrix.

[0270] In a possible implementation, when M>2, the feature matrix corresponding to each dimension is determined based on the first data and the dimension reduction matrix corresponding to each dimension, including: the feature matrix corresponding to each dimension is determined based on the feature tensor corresponding to each dimension. Among them, the feature tensor corresponding to each dimension is determined based on the first data and the dimension reduction matrix corresponding to each dimension, the feature matrix corresponding to each dimension includes the feature matrix corresponding to the i-th dimension among P dimensions, the feature matrix corresponding to the i-th dimension is a feature matrix obtained by merging the dimensions other than the i-th dimension in the feature tensor corresponding to the i-th dimension, the feature matrix corresponding to the i-th dimension and the feature tensor corresponding to the i-th dimension contain the same elements, 1≤i≤P.

[0271] For example, the first device determines the feature matrix corresponding to each dimension based on the first data and the dimensionality reduction matrix corresponding to each dimension, including: the first device determines the feature tensor corresponding to each dimension based on the first data and the dimensionality reduction matrix corresponding to each dimension; the first device performs tensor expansion on the feature tensor corresponding to each dimension to obtain the feature matrix corresponding to each dimension.

[0272] It should be understood that the difference between method 1 and method 2 is that method 2 first reduces one or more dimensions of the first data to obtain a feature tensor corresponding to each dimension, and then expands the feature tensor to obtain a two-dimensional feature matrix.

[0273] It can be understood that the first device determines the feature tensor corresponding to each dimension based on the first data and the dimensionality reduction corresponding to each dimension. The specific implementation is similar to the product between the tensor and the matrix in the aforementioned formula (1), and will not be repeated here.

[0274] It should be understood that the first device can also use other product operations between tensors and matrices to determine the feature tensor corresponding to each dimension, and the embodiments of the present application do not specifically limit this.

[0275] It can be understood that in method 2, the tensor expansion can specifically refer to the tensor expansion in method 1, which will not be repeated here.

[0276] That is to say, the first device can reduce the dimension of the first data according to the dimension reduction matrix corresponding to each dimension, obtain the feature tensor corresponding to each dimension, and determine the feature matrix corresponding to each dimension according to the feature tensor corresponding to each dimension, which can increase the flexibility of the first device in determining the feature matrix to adapt to different application scenarios. It can be understood that the number of dimensions of the feature tensor is greater than the number of dimensions of the feature matrix, so the feature tensor corresponding to the i-th dimension can be considered as: the dimension size of one or more dimensions other than the i-th dimension in the M dimensions is reduced, and then the feature tensors corresponding to each dimension can be different, or the feature tensors corresponding to at least two different dimensions can be the same, thereby increasing the flexibility of the first device in determining the feature matrix.

[0277] Exemplarily, taking three-dimensional CSI data (i.e., RX dimension, TX dimension, and frequency domain dimension) as the first data example, the number of antennas of the terminal device (i.e., the dimension size N1 of the RX dimension) is 32, the number of antennas of the access network device (i.e., the dimension size N2 of the TX dimension) is 128, and the number of frequency domain units for sending data (i.e., the dimension size N3 of the frequency domain dimension) is 64 frequency domain units. Among them, the TX dimension and the frequency domain dimension are large in size and have more redundant information, and the TX dimension and the frequency domain dimension in the first data can be reduced in dimension.

[0278] For channels with more severe frequency selective fading, the first device hopes to retain more features in the frequency domain dimension, and then for the frequency domain dimension, the first device can reduce the dimension of one of the RX dimension and the TX dimension through the dimension reduction matrix; for the TX dimension, the first device can reduce the dimension of the RX dimension in the first data through the dimension reduction matrix, and the frequency domain dimension is not reduced. When the first device determines that the dimension reduction matrix corresponding to the frequency domain dimension is used to reduce the dimension of the RX dimension, the dimension reduction matrices corresponding to the TX dimension and the frequency domain dimension are both used to reduce the dimension of the RX dimension, and then the dimension reduction matrix can be shared, and the corresponding feature tensors between the two are the same.

[0279] For frequency-flat fading channels, since the frequency response of the channel is flat within a certain frequency range, the inter-symbol interference caused by multipath is small, and the impact on channel estimation is small, and the correlation can be fully utilized to reduce the dimension of the frequency domain to reduce redundancy and reduce the amount of calculation for subsequent determination of the feature matrix. For example, for the TX dimension, the first device can reduce the dimension of the frequency domain and RX dimensions in the first data through a dimension reduction matrix; for the frequency domain dimension, the first device can reduce the dimension of the RX dimension and TX dimension in the first data through a dimension reduction matrix, so that the corresponding dimension reduction matrix between the TX dimension and the frequency domain dimension is different, and the corresponding feature tensors between the two are also different.

[0280] It should be understood that the above description on whether the two dimensions share the dimensionality reduction matrix is ​​only an example, and can also be applied to other application scenarios besides frequency selective fading and frequency flat fading, and can also be applied to other types of data besides CSI data (such as perception data, AI data, or other model data, etc.). The embodiments of the present application do not make specific limitations on this.

[0281] The dimension reduction matrix and feature tensor in method 2 are described in detail below.

[0282] In a possible implementation, the dimension reduction matrix corresponding to each dimension includes one or more dimension reduction matrices.

[0283] It can be understood that the dimensions of the first data are reduced in different dimensions, and thus the number of reduced dimension matrices included in the reduced dimension matrix corresponding to each dimension is also different. For example, taking the three-dimensional CSI data in the above example as an example, whether the TX dimension and the frequency domain dimension share the reduced dimension matrix is ​​described in case 1 and case 2.

[0284] Case 1: The TX dimension and the frequency domain dimension share the reduced dimension matrix.

[0285] The dimensionality reduction matrix corresponding to the TX dimension and the frequency domain dimension is the same, and the dimensionality reduction matrix is ​​used to reduce the RX dimension in the first data. The dimension size of the RX dimension after dimensionality reduction is N1', and thus the dimension size of the dimensionality reduction matrix is ​​N1×N1'. Furthermore, the first device obtains a feature tensor corresponding to the TX dimension based on the dimensionality reduction matrix corresponding to the first data and the TX dimension, and the dimension size is N1'×N2×N3. The first device performs tensor expansion on the feature tensor corresponding to the TX dimension, and the dimension size of the feature matrix corresponding to the TX dimension obtained is N2×(N1'×N3), and N1'×N3 is equal to the dimension size N after dimensionality reduction of the TX dimension. s,2 According to the description of the above step S402, N s,1 It is determined by the first device according to the dimension information corresponding to the TX dimension, that is, the dimension size of the dimensionality reduction matrix corresponding to each dimension is associated with the dimension information corresponding to each dimension.

[0286] It can be understood that the feature tensors corresponding to the TX dimension and the frequency domain dimension are the same, and the dimension size of the feature matrix corresponding to the frequency domain dimension is N3×(N1'×N2), and N1'×N2 is equal to the dimension size N after the frequency domain dimension is reduced. s,3 .

[0287] Without loss of generality, for P less than M, each dimension in the P dimensions can share a dimension reduction matrix and a feature tensor, and the dimension reduction matrix can include MP dimension reduction matrices. For example, for M=4, the dimension size of the first dimension in the M dimensions is N1, the dimension size of the second dimension is N2, the dimension size of the third dimension is N3, and the dimension size of the fourth dimension is N4. In the case where P dimensions include the first and second dimensions of the M dimensions, the dimensionality reduction matrix shared by the first and second dimensions may include two (MP=2) dimensionality reduction matrices, namely: a first dimensionality reduction matrix for reducing the third dimension, and a second dimensionality reduction matrix for reducing the fourth dimension. The dimensionality size of the first dimensionality reduction matrix is ​​N3×N3' (i.e., the dimensionality size of the third dimension after dimensionality reduction), and the dimensionality size of the second dimensionality reduction matrix is ​​N4×N4' (i.e., the dimensionality size of the fourth dimension after dimensionality reduction). In this way, the dimensionality size of the feature tensor is N1×N2×N3'×N4', the dimensionality size of the feature matrix corresponding to the first dimension is N1×(N2×N3'×N4'), and the dimensionality size of the feature matrix corresponding to the second dimension is N2×(N1×N3'×N4').

[0288] Similarly, when M=4 and P=3, the dimension reduction matrix shared among P dimensions is one dimension reduction matrix, which is similar to the case 1 and will not be described again.

[0289] That is to say, the dimensionality reduction matrix and feature tensor can be shared between each of the P dimensions. The first device can determine a feature tensor based on the first data and the shared dimensionality reduction matrix, and then determine the feature matrix corresponding to each dimension based on the feature tensor. This is easy to implement and can reduce the amount of calculation.

[0290] It should be understood that the P dimensions can be divided into at least two groups, and at least two dimensions included in each group can share a dimensionality reduction matrix, and the dimensionality reduction matrices shared by different groups can be different. This can further improve the flexibility of the first device in determining the feature matrix to suit more application scenarios.

[0291] For example, for M=6, P=5, the dimension size of the first dimension of the six dimensions is N1, the dimension size of the second dimension is N2, the dimension size of the third dimension is N3, the dimension size of the fourth dimension is N4, the dimension size of the fifth dimension is N5, the dimension size of the sixth dimension is N6, and the P dimensions are the first to fifth dimensions. The P dimensions can be divided into a first group and a second group, the first group includes: the first to second dimensions, and the second group includes: the third to fifth dimensions.

[0292] Assuming that the dimension reduction matrix shared by the first group is used to reduce the 5th to 6th dimensions, the dimension reduction matrix shared by the first group includes: a first dimension reduction matrix (N5×N5') and a second dimension reduction matrix (N6×N6'), and the dimension size of the feature tensor corresponding to the first group is N1×N2×N3×N4×N5'×N6'. Further, the dimension size of the feature matrix corresponding to the first dimension is N1×(N2×N3×N4×N5'×N6'), and the dimension size of the feature matrix corresponding to the second dimension is N2×(N1×N3×N4×N5'×N6').

[0293] Assuming that the dimensionality reduction matrix shared by the second group is used to reduce the dimensionality of the first and second dimensions, the dimensionality reduction matrix shared by the second group includes: a third dimensionality reduction matrix (N1×N1') and a fourth dimensionality reduction matrix (N2×N2'), and the dimension size of the feature tensor corresponding to the second group is N1'×N2'×N3×N4×N5×N6. Further, the dimension size of the feature matrix corresponding to the third dimension is N3×(N1'×N2'×N4×N5×N6), the dimension size of the feature matrix corresponding to the fourth dimension is N4×(N1'×N2'×N3×N5×N6), and the dimension size of the feature matrix corresponding to the fifth dimension is N5×(N1'×N2'×N3×N4×N6).

[0294] It should be understood that the above description of grouping of P dimensions is only an example. Depending on the size of P, the P dimensions can also be divided into 3 groups, 4 groups, or more groups. The dimensionality reduction matrix shared by each group in the multiple groups can be 1 matrix, 2 matrices, 3 matrices, or more matrices. The embodiments of the present application do not make any specific limitations on this.

[0295] Case 2: The TX dimension and the frequency domain dimension do not share the dimensionality reduction matrix.

[0296] The dimensionality reduction matrices corresponding to the TX dimension and the frequency domain dimension are different. For example, the dimensionality reduction matrix corresponding to the TX dimension is used to reduce the RX dimension and the frequency domain dimension in the first data. The dimension size of the RX dimension after dimensionality reduction is N1', and the dimension size of the frequency domain dimension after dimensionality reduction is N3'. The dimensionality reduction matrix corresponding to the TX dimension includes a first dimensionality reduction matrix and a second dimensionality reduction matrix. The dimension size of the first dimensionality reduction matrix is ​​N1×N1', and the dimension size of the second dimensionality reduction matrix is ​​N3×N3'. Further, the first device can determine that the dimension size of the feature tensor corresponding to the TX dimension is N1'×N2×N3' based on the first data, the first dimensionality reduction matrix, and the second dimensionality reduction matrix. It can be understood that the first device performs tensor expansion on the feature tensor corresponding to the TX dimension, and the dimension size of the feature matrix corresponding to the TX dimension obtained is N2×(N1'×N3'), and N1'×N3' is equal to the dimension size N after the TX dimension is reduced. s,2 .

[0297] Similarly, the dimension reduction matrix corresponding to the frequency domain dimension is used to reduce the RX dimension and TX dimension in the first data. The dimension size of the RX dimension after dimension reduction is N1", and the dimension size of the TX dimension after dimension reduction is N2'. The dimension reduction matrix corresponding to the frequency domain dimension includes a third dimension reduction matrix and a fourth dimension reduction matrix. The dimension size of the third dimension reduction matrix is ​​N1×N1", and the dimension size of the fourth dimension reduction matrix is ​​N2×N2'. Furthermore, the dimension size of the feature tensor corresponding to the frequency domain dimension is N1"×N2'×N3, and the dimension size of the feature matrix corresponding to the frequency domain dimension is N3×(N1"×N2'), where N1"×N2' is equal to the dimension size N after dimension reduction of the frequency domain dimension. s,3 .

[0298] That is to say, the dimensionality reduction matrix corresponding to the i-th dimension among P dimensions may include M-1 dimensionality reduction matrices, and the M-1 dimensionality reduction matrices are used to reduce the dimensions of the other M-1 dimensions in the first data except the i-th dimension, thereby further reducing the dimensionality of the feature tensor corresponding to the i-th dimension, so as to reduce the computational complexity of subsequently determining the feature matrix corresponding to the i-th dimension.

[0299] It should be understood that the above-mentioned situation 1 and situation 2 can be combined. For example, the dimensionality reduction matrix corresponding to the TX dimension includes at least two dimensionality reduction matrices (for example, the first dimensionality reduction matrix (i.e., N1×N1') and the second dimensionality reduction matrix (i.e., N3×N3')), and the dimensionality reduction matrix corresponding to the frequency domain dimension can be a dimensionality reduction matrix, which is used to reduce the RX dimension in the first data, and the dimension size of the RX dimension after dimensionality reduction is N1", that is, the dimension size of the dimensionality reduction matrix is ​​N1×N1"; or, the dimensionality reduction matrix is ​​used to reduce the TX dimension in the first data, and the dimension size of the TX dimension after dimensionality reduction is N2', that is, the dimension size of the dimensionality reduction matrix is ​​N2×N2'.

[0300] It can be understood that for the above-mentioned method 2, the dimension size of the dimensionality reduction matrix corresponding to each dimension can also be determined according to the dimension information corresponding to each dimension, which is described in detail below.

[0301] In a possible implementation, the dimension information corresponding to each dimension includes the dimension information corresponding to the i-th dimension among the P dimensions, and the dimension information corresponding to the i-th dimension is used to indicate: the dimension size after dimension reduction of each dimension in the dimension reduction dimension set corresponding to the i-th dimension, or the dimension reduction ratio corresponding to each dimension in the dimension reduction dimension set. Among them, the dimension reduction dimension set includes at least one dimension, and the at least one dimension is a dimension for dimension reduction for at least part of the dimensions other than the i-th dimension among the M dimensions. It can be understood that the dimension reduction dimension set can be used by the first device to determine which dimension the dimension reduction matrix corresponding to the i-th dimension is used for dimension reduction, and the size of each dimension in the dimension reduction dimension set after dimension reduction can be used by the first device to determine the number and dimension size of the dimension reduction matrix corresponding to the i-th dimension, and then the first device can determine the dimension size after dimension reduction of the i-th dimension based on the information of the dimension reduction matrix corresponding to the i-th dimension, that is, the dimension size of the full rank matrix corresponding to the i-th dimension.

[0302] In addition, when the number of dimensions included in the reduced dimension set is less than M-1, there are some dimensions in the M dimensions other than the i-th dimension that have not been reduced, and the dimension size of the i-th dimension after dimension reduction (that is, the rank of the full rank matrix corresponding to the i-th dimension) is equal to the product of the dimension size of each dimension in the reduced dimension set and the dimension size of the dimension that has not been reduced. When the number of dimensions included in the reduced dimension set is equal to M-1, the dimension size of the i-th dimension after dimension reduction is equal to the product of the dimension sizes of each dimension in the reduced dimension set after dimension reduction.

[0303] For example, for the above situation 1, the reduced dimension set corresponding to the TX dimension includes the RX dimension, the reduced dimension set corresponding to the frequency domain dimension includes the RX dimension, and the dimension size of the RX dimension after dimension reduction is N1'. In this way, the first device can determine that the reduced dimension matrix corresponding to the TX dimension and the frequency domain dimension is the same, and the dimension size of the reduced dimension matrix is ​​N1×N1', and the dimension size of the TX dimension after dimension reduction is N1'×N3=N s,2 , the dimension size after frequency domain dimension reduction is N1'×N2=N s,3 .

[0304] For another example, for the above situation 2, the set of reduced dimensions corresponding to the TX dimension includes: RX dimension and frequency domain dimension, the dimension size of the RX dimension after dimension reduction is N1', and the dimension size of the frequency domain dimension after dimension reduction is N3', so the first device can determine that the reduced dimension matrix corresponding to the TX dimension includes two reduced dimension matrices, and the dimension sizes are: N1×N1' and N3×N3', respectively, and the dimension size of the TX dimension after dimension reduction is N1'×N3'=N s,2The set of reduced dimensions corresponding to the frequency domain dimension in the above situation 2 includes: RX dimension and TX dimension. The dimension size of the RX dimension after dimension reduction is N1”, and the dimension size of the TX dimension after dimension reduction is N2'. In this way, the first device can determine that the reduced dimension matrix corresponding to the frequency domain dimension includes two reduced dimension matrices, and the dimension sizes are respectively: N1×N1” and N2×N2'. The dimension size of the frequency domain dimension after dimension reduction is N1”×N2'=N s,3 .

[0305] That is to say, the first device can determine the smaller dimension after dimensionality reduction of each dimension through the dimensionality information corresponding to each dimension, and can also determine the number of dimensionality reduction matrices corresponding to each dimension, which dimension is used for dimensionality reduction, and the size of the dimension.

[0306] It should be understood that the above-mentioned method 1 and method 2 are aimed at the case where the number of dimensions M of the first data is greater than 2. When M is equal to 2, the first data is two-dimensional data. In this way, the first data can be directly subjected to matrix multiplication operation with the dimensionality reduction matrix corresponding to each dimension (such as the dimensionality reduction matrix in method 1) to obtain the feature matrix corresponding to each dimension.

[0307] The above methods 1 and 2 introduce the number of dimension reduction matrices included in the dimension reduction matrix corresponding to each dimension, as well as the dimension size of the dimension reduction matrix. The elements in the dimension reduction matrix are described in detail below.

[0308] In a possible implementation, the number of non-zero elements in the dimension reduction matrix is ​​less than the number of zero elements. That is, when the two-dimensional data or the first data corresponding to each dimension is multiplied with the dimension reduction matrix, the computational complexity can be reduced because most of the elements in the dimension reduction matrix are zero elements.

[0309] It can be understood that the proportion of non-zero elements in the reduced dimension matrix should be less than 50% of all elements. Considering that the reduced dimension matrix is ​​a column full rank matrix (or full rank matrix), the relevant design of the non-zero elements in the reduced dimension matrix is ​​associated with the dimensional information corresponding to each dimension.

[0310] In one possible implementation, the dimensionality reduction matrix corresponding to each dimension is associated with the dimensionality information corresponding to each dimension, including: the parameters of the dimensionality reduction matrix corresponding to each dimension are associated with the dimensionality information corresponding to each dimension, wherein the parameters of the dimensionality reduction matrix corresponding to each dimension include at least one of the following: the proportion, position distribution, or value range of non-zero elements in the dimensionality reduction matrix corresponding to each dimension.

[0311] That is to say, the first device can determine the proportion range, position distribution range, and value range of the non-zero elements in the dimension reduction matrix corresponding to each dimension based on the correlation between the dimensional information corresponding to each dimension and the parameters of the dimension reduction matrix corresponding to each dimension, so that the first device can select the dimension reduction matrix corresponding to each dimension based on the proportion range, position distribution range, and value range of the non-zero elements, thereby improving the flexibility of the first device in determining the dimension reduction matrix.

[0312] In a possible implementation, the proportion of non-zero elements in the dimension reduction matrix corresponding to each dimension is greater than or equal to a first threshold corresponding to each dimension, and the first threshold corresponding to each dimension is determined according to the dimension size of the first data and the dimension information corresponding to each dimension. In other words, the proportion of non-zero elements in the dimension reduction matrix corresponding to each dimension should not be less than the first threshold, thereby ensuring that the dimension reduction matrix is ​​a column full rank matrix or a row full rank matrix.

[0313] It can be understood that the reduced dimension matrix corresponding to each dimension is a full rank matrix, that is, each column vector (or row vector) in the reduced dimension matrix should include at least one non-zero element to ensure that the reduced dimension matrix is ​​a full rank matrix. In other words, the first device can determine the rank of the reduced dimension matrix corresponding to each dimension based on the dimensional information corresponding to each dimension, that is, the minimum number of non-zero elements that the reduced dimension matrix can contain. Furthermore, the first device can determine the dimensional size of the reduced dimension matrix corresponding to each dimension based on the dimensional size of the first data, so that the minimum proportion of non-zero elements in the reduced dimension matrix corresponding to each dimension can be determined.

[0314] For example, in the above method 1 Figure 6 For example, the dimension reduction matrix corresponding to the first dimension has a row dimension × column dimension of (N2×N3×…×N M )×N s,1 , each column vector in the reduced dimension matrix corresponding to the first dimension should contain at least one non-zero element, so that the number of non-zero elements contained in the reduced dimension matrix is ​​greater than or equal to N s,1 , the minimum proportion of non-zero elements is 1 / (N2×N3×…×N M ).

[0315] For example, N s,1 =3, N2×N3×…×N M =10 as an example, Figure 7 As shown, the dimension reduction matrix corresponding to the first dimension can only include N s,1 =3 non-zero elements e1, e2, and e3, the position distribution of the three non-zero elements satisfies the condition that the reduced dimension matrix is ​​a column full rank matrix, that is, each column contains a non-zero element, and the three non-zero elements are not in the same row, that is, the same row does not include two non-zero elements.

[0316] It should be understood that Figure 7 The example shown is only an example of a reduced dimension matrix with the lowest proportion of non-zero elements (the lowest proportion of non-zero elements is 0.1). In practical applications (N2×N3×…×N M ) is larger, N s,1 The lowest ratio of smaller, nonzero elements is usually close to zero.

[0317] For another example, taking Case 1 in the above-mentioned method 2 as an example, the dimension size of the reduced dimension matrix shared between the TX dimension and the frequency domain dimension is N1×N1', the number of non-zero elements contained in the reduced dimension matrix is ​​greater than or equal to N1', and the minimum ratio of non-zero elements is 1 / N1.

[0318] For another example, taking Case 2 in the above-mentioned method 2 as an example, the dimensionality reduction matrix corresponding to the TX dimension includes a first dimensionality reduction matrix (N1×N1') and a second dimensionality reduction matrix (N3×N3'), the minimum proportion of non-zero elements in the first dimensionality reduction matrix is ​​1 / N1, and the minimum proportion of non-zero elements in the second dimensionality reduction matrix is ​​1 / N3.

[0319] Optionally, the first threshold corresponding to each dimension is equal to the minimum ratio corresponding to each dimension. For example, for mode 1, the first threshold is the reciprocal of the product of the dimensions of the other dimensions in the first data except the i-th dimension. For another example, for mode 2, the first threshold is the reciprocal of the size of each dimension in the set of reduced dimensions corresponding to the i-th dimension.

[0320] Alternatively, optionally, the first threshold corresponding to each dimension is greater than the minimum ratio corresponding to each dimension. That is, the first threshold is greater than the minimum ratio corresponding to each dimension, which can increase the ratio of non-zero elements in the dimension reduction matrix, thereby increasing the amount of information extracted from the first data, thereby retaining more feature information.

[0321] It should be understood that the difference or ratio between the first threshold corresponding to each dimension and the minimum ratio corresponding to each dimension depends on the actual implementation of the first device. For example, the first device can be flexibly set according to the application scenario, and the embodiments of the present application do not specifically limit this.

[0322] Optionally, the ratio of non-zero elements in the dimension reduction moment corresponding to each dimension is inversely proportional to the proportional factor corresponding to each dimension. It can be understood that for proportional factor = dimension size after dimension reduction / dimension size before dimension reduction, the larger the proportional factor, the lower the compression rate corresponding to the dimension. The feature matrix or feature tensor obtained by the first device after dimension reduction through the dimension reduction matrix has a larger column dimension (or row dimension) and a higher computational complexity. By reducing the ratio of non-zero elements in the dimension reduction matrix, the computational complexity of the feature matrix or feature tensor obtained by the first device through the dimension reduction matrix can be reduced. Similarly, the smaller the proportional factor, the higher the compression rate corresponding to the dimension. By increasing the ratio of non-zero elements in the dimension reduction matrix, the number of non-zero elements in the feature matrix or feature tensor can be increased, thereby retaining more feature information and improving the accuracy of the full rank matrix in representing the main features corresponding to each dimension.

[0323] It is understandable that, as mentioned above Figure 7 In the example shown, when the condition that the dimensionality reduction matrix corresponding to each dimension is full column rank (or full row rank) is satisfied, the position distribution of non-zero elements in the dimensionality reduction matrix corresponding to each dimension can be a random distribution or a non-random distribution, and the embodiments of the present application do not make specific limitations on this.

[0324] In a possible implementation, the values ​​of the non-zero elements in the dimension reduction matrix corresponding to each dimension are ±1. That is to say, when the first device performs the multiplication operation between the two-dimensional data corresponding to each dimension or the first data and the dimension reduction matrix, thanks to the fact that most of the non-zero elements in the dimension reduction matrix have values ​​of ±1, the first device can implement the multiplication operation between the two through a small amount of addition and / or subtraction operations, which can further reduce the complexity of the operation.

[0325] It can be understood that the values ​​of the above non-zero elements can also be normalized, that is, the value range is [-1,1]. In this way, when the value of an element in the first data or the two-dimensional data corresponding to each dimension is large, an overflow of the value obtained after multiplying the element with the non-zero element in the dimensionality reduction matrix is ​​avoided.

[0326] It can be understood that the value of the above non-zero element can also be greater than 1 or less than -1, and the embodiment of the present application does not specifically limit this.

[0327] B. Determine the implementation method of the dimensionality reduction matrix:

[0328] In a possible implementation, the dimension reduction matrix corresponding to each dimension is determined according to the dimension information corresponding to each dimension. That is, the first device determines the dimension reduction matrix corresponding to each dimension through the dimension information corresponding to each dimension, which can save the indication overhead of indicating the dimension reduction matrix corresponding to each dimension.

[0329] For example, the first device determines that the proportion of non-zero elements in the dimensionality reduction matrix corresponding to each dimension is the minimum proportion, and the value of the non-zero elements is ±1. In this way, the first device can determine the dimension size of each dimension after dimensionality reduction according to the dimensionality information corresponding to each dimension, and determine the dimensionality reduction matrix corresponding to each dimension based on the minimum proportion and value of the non-zero elements.

[0330] It should be understood that the first device can optimize the proportion and value of non-zero elements in the dimensionality reduction matrix in an offline manner. The proportion of non-zero elements can be other proportions greater than the minimum proportion, and the value of non-zero elements can also be other values. The embodiments of the present application do not specifically limit this.

[0331] Optionally, the dimension reduction matrix corresponding to each dimension is determined according to the dimension information corresponding to each dimension, including: the dimension reduction matrix corresponding to each dimension is determined from a candidate dimension reduction matrix set according to the dimension information corresponding to each dimension, and the candidate dimension reduction matrix set includes at least two dimension reduction matrices. In other words, the first device can select the dimension reduction matrix corresponding to each dimension from the candidate dimension reduction matrix set according to the dimension information corresponding to each dimension, thereby increasing the flexibility of the first device in determining the dimension reduction matrix corresponding to each dimension.

[0332] For example, the first device can obtain a set of candidate dimensionality reduction matrices, and the candidate dimensionality reduction matrix set includes at least two dimensionality reduction matrices corresponding to different dimensionality sizes after dimensionality reduction (or the number of column vectors, or the number of row vectors), and / or the proportion of non-zero elements. In this way, the first device can determine the dimensionality size of each dimension after dimensionality reduction (i.e., the number of column vectors or row vectors of the dimensionality reduction matrix) according to the dimensionality information corresponding to each dimension, and determine the dimensionality reduction matrix corresponding to each dimension according to the corresponding relationship between each dimensionality reduction matrix in the candidate dimensionality reduction matrix set and the dimensionality size after dimensionality reduction, and / or the corresponding relationship between each dimensionality reduction matrix in the candidate dimensionality reduction matrix set and the proportion of non-zero elements.

[0333] It can be understood that different dimensional sizes after dimensionality reduction may also refer to the scaling factors in the aforementioned step S402. For details of the scaling factors, please refer to the relevant instructions in step S402, which will not be repeated here.

[0334] It should be understood that the set of candidate dimensionality reduction matrices may be determined by advance training of the first device; or, the set of candidate dimensionality reduction matrices may be agreed upon by a protocol; or, the set of candidate dimensionality reduction matrices may be negotiated in advance between the first device and the second device; or, the set of candidate dimensionality reduction matrices may be indicated by the network side, and the embodiments of the present application do not specifically limit this.

[0335] It should also be understood that the dimension reduction matrix corresponding to each dimension may be indicated by the network side. For details, please refer to the relevant description of step S409 below, which will not be repeated here.

[0336] For step S404:

[0337] It can be understood that, as described in the above step S402, the resources (such as time domain resources, frequency domain resources, or MCS) used to transmit the first data are actually used to transmit the second data obtained after the first data is reduced in dimension and the full rank matrix corresponding to each dimension. In other words, the first device can transmit the second data and the full rank matrix corresponding to each dimension based on the resources used to transmit the first data.

[0338] Optionally, the first device sends the full rank matrix and second data corresponding to each dimension to the second device (i.e., step S404), including: the first device can quantize the full rank matrix and second data corresponding to each dimension to obtain a compressed bit stream; the first device sends the compressed bit stream to the second device. It can be understood that the full rank matrix and the second data are quantized to facilitate the first device to send. For example, the elements in the above-mentioned full rank matrix and second data can be quantized to 6 bits (bit). It can also be understood that the first device can use other numbers of bits for quantization, and the embodiments of the present application do not specifically limit this.

[0339] It should be understood that the second device can receive a data signal or data channel (for example, a physical uplink share channel (PUSCH) or a physical downlink share channel (PDSCH), etc.) carrying the second data and the full rank matrix corresponding to each dimension on the time domain resources and / or frequency domain resources corresponding to the first data. In this way, the second device can determine that the full rank matrix and the second data corresponding to each dimension are associated with the first data, and then the second device can obtain the dimensional information corresponding to each dimension and the dimensional size of the first data, and based on the above dimensional size, determine the dimensional size of the full rank matrix corresponding to each dimension, as well as the dimensional size of the second data, and then decode the compressed bit stream carried by the data signal (or data channel) according to the dimensional size of the full rank matrix corresponding to each dimension and the dimensional size of the second data to obtain the full rank matrix and the second data corresponding to each dimension.

[0340] In addition, the second device obtains the dimensional size of the first data, which can be specifically referred to in the following step S405, and the second device obtains the dimensional information corresponding to each dimension, which can be specifically referred to in the following step S408 or step S409, which will not be repeated here.

[0341] It can be understood that in order to further compress the first data, the first device can transform the full rank matrix corresponding to each dimension and the second data. The transformation can refer to projection (for example Figure 1Alternatively, the transformation may refer to feature extraction (such as convolution and / or pooling, etc.), which is not specifically limited in the embodiments of the present application.

[0342] It should be understood that after the first device transforms the full rank matrix and the second data corresponding to each dimension, it can also perform quantization. Alternatively, after the first device transforms the full rank matrix and the second data corresponding to each dimension, it can also reduce the dimension of the transformed data and then perform quantization, which is not specifically limited in the embodiments of the present application.

[0343] For step S405:

[0344] In a possible implementation, the second device obtains the dimension size of the first data (i.e., step S405), including: the second device determines the dimension size of the first data according to the capability information of the first device and / or the resources used to transmit the first data. In other words, the second device can determine the dimension size of the first data according to the capability information of the first device and / or the resources used to transmit the first data, and then the first device may not indicate the dimension size of the first data to the second device, saving network overhead.

[0345] For example, taking the first device as a terminal device, the second device as an access network device, and the first data as CSI data (or perception data), the CSI data generally includes an RX dimension and a TX dimension. Among them, the second device can determine the number of antennas of the first device based on the capability information reported by the first device, that is, the second device can determine the dimension size of the RX dimension and the dimension size of the TX dimension. Furthermore, in the case where the CSI data also includes a frequency domain dimension, for unlicensed transmission, the frequency domain resources used to transmit the first data are configured in advance by the second device to the first device, so that the second device can determine the dimension size of the frequency domain dimension.

[0346] It can be understood that for downlink transmission, the first device is an access network device, the second device is a terminal device, and the second device can determine the dynamically scheduled resources based on the DCI sent by the first device, and then determine the dimensional size of the frequency domain dimension. Further, the second device can estimate the dimensional size of the first data based on the compression rate and the dynamically scheduled resources. Among them, the compression rate can be pre-configured, or negotiated in advance between the first device and the second device, or indicated by the first device, and the embodiments of the present application do not specifically limit this.

[0347] It should be understood that the above-mentioned second device determining the dimensional size of the first data based on the capability information of the first device and / or the resources used to transmit the first data is merely an example, and the second device may also use other methods to determine the dimensional size of the first data based on the capability information of the first device and / or the resources used to transmit the first data, and the embodiments of the present application do not specifically limit this.

[0348] In another possible implementation, Figure 4 The data transmission method shown also includes:

[0349] S407, the first device sends the first indication information to the second device. Correspondingly, the second device receives the first indication information from the first device. The first indication information is used to indicate the dimension size of the first data. In other words, the first device can send the first indication information to the second device to indicate the dimension size of the first data, thereby improving the flexibility of the second device in obtaining the dimension size of the first data to apply to different application scenarios.

[0350] For example, if the first data is AI data, the dimension information of the AI ​​data may not be associated with the capability information of the first device, so the second device can determine the dimension size of the first data according to the first indication information. It can be understood that in the case where the agreed compression rate is not mentioned, the second device (such as the terminal device) can also determine the dimension size of the first data according to the first indication information.

[0351] Optionally, the first indication information includes index information. The index information is used to determine the dimension size of the first data from a candidate dimension set, wherein the candidate dimension set includes at least two dimension sizes. That is, the first device can indicate the dimension size of the first data to the second device by indicating the index of the dimension size of the first data in the candidate set, thereby saving the indication overhead of the first indication information and improving the reliability of the first indication information.

[0352] It can be understood that for the dimension size in the candidate dimension set, the dimension size includes two pieces of information, namely, indication information of multiple dimensions and indication information of the dimension size of each dimension in the multiple dimensions. For example, as shown in Table 1, the candidate dimension set includes N dim dimensional sizes, each of which includes the dimensional size corresponding to each dimension in the M dimensions (i.e., N1, N2, ..., N i ,…,N M ), N dim is an integer greater than 1. The indication overhead of the index information is N dim The logarithm of the integer value, such as ceil(log2(N dim )), ceil() means rounding up.

[0353] Table 1

[0354]

[0355] It should be understood that Table 1 is only an example, and the index information may also be the identifier (ID) of different dimension sizes in the candidate dimension set, and the embodiments of the present application do not specifically limit this.

[0356] It can be understood that the first indication information can be carried by at least one of the following: RRC message (or signaling), DCI carried, MAC protocol data unit (PDU), uplink control information (UCI), or physical uplink control channel (PUCCH). Among them, the RRC message can be, for example, an RRC setup message, or an RRC resume message, or an RRC reconfiguration message, etc. In other words, the dimension size of the first data indicated by the first indication information can be continuously effective during the RRC connection period, and there is no need to indicate the dimension size of the first data each time scheduling. It is suitable for scenarios in which data of the same dimension size is continuously sent during the RRC connection period, and the dimension size of data transmission is changed through RRC signaling due to mobility, energy saving, or changes in business requirements. For example, taking the first device as an access network device and the second device as a terminal device as an example, when the second device initially accesses or updates its location, the second device can establish an RRC connection with the first device, and the first device can carry the first indication information through an RRC establishment message or an RRC reconfiguration message, thereby indicating the dimension size of the first data to the second device, that is, during the RRC connection, the dimension size of the first data indicated by the RRC establishment message can be used for resource allocation and data transmission. For another example, the second device may enter the RRC inactive state from the RRC connection state due to current business reasons or energy saving requirements, and then when the second device restores the RRC connection, the first device can carry the first indication information through the RRC recovery message, thereby indicating the dimension size of the first data to the second device. For another example, due to poor wireless link quality or handover, the first device and the second device can re-establish the RRC connection through the RRC reconfiguration process, and the first device can carry the first indication information through the RRC reconfiguration message, thereby indicating the dimension size of the first data to the second device.

[0357] It can be understood that for the first indication information carried by DCI or MAC PDU (for example, the MAC control element (CE) in the MAC PDU, or the MAC service data unit (SDU)), the first device can dynamically indicate the dimension size of the first data to the second device, which is suitable for scenarios where different dimension sizes of first data are dynamically transmitted during the RRC connection.

[0358] It can be understood that after the second device receives the first indication information from the first device, it can determine the resources allocated to the first device for transmitting the first data according to the compression rate agreed in advance (for example, pre-configured by the protocol, or negotiated in advance by the first device and the second device) and the dimension size of the first data. In other words, the first device does not need to send a scheduling request (SR) or a buffer state report (BSR) to the second device to request resources for transmitting the first data, which can save network overhead.

[0359] Optionally, the first indication information also includes indication information of a candidate dimension set. That is, the first device may configure a candidate dimension set to the second device to indicate to the second device the candidate dimension sizes of the first data that the first device expects to send in the next period of time.

[0360] It can be understood that the indication information of the candidate dimension set can be carried by an RRC message, and the index information can be carried by DCI, or MAC CE, or UCI, or PUCCH, and the embodiments of the present application do not specifically limit this.

[0361] It should be understood that in the embodiment of the present application, the candidate dimension set may also be pre-configured by the protocol, or negotiated in advance between the first device and the second device, and the embodiment of the present application does not specifically limit this.

[0362] It can be understood that step S407 may be before step S404; or, step S407 may be after step S404; or, step S407 and step S404 may be performed simultaneously, which is not specifically limited in the embodiments of the present application.

[0363] In one possible implementation, Figure 4 The data transmission method shown also includes:

[0364] S408: The first device sends second indication information to the second device. Correspondingly, the second device receives the indication information from the first device. The second indication information is used to indicate P dimensions (ie, the P dimensions in step S402).

[0365] That is to say, the second device can determine, based on the second indication information, that P dimensions of the M dimensions of the first data are to be reduced in dimension, that is, the second device can determine, based on the second indication information, that the received second data is obtained by reducing the P dimensions of the first data in dimension, and determine that the full rank matrix corresponding to each received dimension is associated with each dimension of the P dimensions, so as to decompress the second data and improve the decompression efficiency of the second device.

[0366] It should be understood that step S408 may be before step S404; or, step S408 may be after step S404; or, step S408 and step S404 may be performed simultaneously, which is not specifically limited in the embodiments of the present application.

[0367] In another possible implementation, Figure 4 The data transmission method shown also includes:

[0368] S409: The second device sends second indication information to the first device. Correspondingly, the first device receives indication information from the second device. The second indication information is used to indicate P dimensions.

[0369] That is to say, the first device can determine which dimensions of the M dimensions of the first data are to be reduced according to the second indication information, thereby improving the flexibility of the first device in determining which dimensions of the M dimensions are to be reduced, so as to be applicable to different application scenarios. For example, the second device can determine which P dimensions of the M dimensions of the first data are to be reduced according to the data decompression requirements, and send the second indication information to the first device, so that the first device can determine which P dimensions of the first data are to be reduced according to the second indication information, thereby matching the data compression of the first data with the data decompression requirements of the second device, and improving the efficiency of data transmission.

[0370] It can be understood that for the method 1 in the aforementioned step S403, the first device can determine which dimensions of the M dimensions are to be tensor expanded according to the second indication information. For case 1 in the method 2, the first device can determine which dimensions correspond to the feature tensors to be tensor expanded according to the second indication information.

[0371] Optionally, the second indication information includes index information corresponding to each dimension in the P dimensions. The index information corresponding to each dimension in the P dimensions may include: an index or identifier of each dimension in the M dimensions. That is, the second indication information may indicate P dimensions in the M dimensions by indicating the index or identifier of each dimension in the P dimensions in the M dimensions, thereby reducing the indication overhead and improving the reliability of the second indication information.

[0372] It can be understood that the protocol can pre-agree on the identifier of each dimension in the M dimensions, or pre-agree on the index (or serial number) corresponding to each dimension in the M dimensions, so that the second indication information can be indicated by the identifier or index corresponding to P dimensions.

[0373] For example, taking the second indication information including the index corresponding to each dimension in P dimensions as an example, as shown in Table 2, the second indication information includes the indication information of the total number of indexes, and the index corresponding to each dimension in P dimensions. Among them, the indication information of the total number of indexes is used to indicate the number of indexes contained in the second indication information, so that the first device or the second device can determine the end position of the second indication information, and avoid miscoding other bits as the index corresponding to each dimension. Further, considering that the dimension of the first data is M dimensions, the number of bits (or the number of bits) required to represent M by bits is ceil(log2(M)), that is, the indication overhead of the index corresponding to each dimension is ceil(log2(M)). The first device and the second device (or the protocol) can agree that the indication overhead of the index corresponding to each dimension is ceil(log2(M)), and the indication overhead of the total number of indexes P can also be agreed to be ceil(log2(M)), so that the indication overhead of the second indication information is (P+1) · ceil(log2(M)).

[0374] Table 2

[0375]

[0376] Alternatively, the second indication information includes a bitmap corresponding to the P dimensions in the M dimensions. That is, the second indication information can indicate the P dimensions in a bitmap manner, which can further reduce the indication overhead and further improve the reliability of the second indication information.

[0377] For example, as shown in Table 3, the first row in Table 3 represents M dimensions arranged in ascending order, and the second row is a bitmap. It should be understood that the bitmap may only include the second row in Table 3, and the protocol may pre-agree that the M elements in the bitmap are arranged in ascending order, that is, the first element corresponds to the first dimension, the second element corresponds to the second dimension, ..., the Mth element corresponds to the Mth dimension. Among them, element 0 in the second row may indicate that the corresponding dimension is not indicated or tensor expansion is not performed, and element 1 may indicate that the corresponding dimension is indicated or tensor expansion is performed.

[0378] Table 3

[0379] 1 2 3 … M 1 0 1 … 1

[0380] It should be understood that Table 3 is only an example, and the bit map may also be adopted in other ways, for example, element 0 indicates that its corresponding dimension is indicated, element 1 indicates that its corresponding dimension is not indicated, or the elements in the bit dimension may correspond to the M dimensions in descending order, and the embodiments of the present application do not make specific limitations on this.

[0381] In a possible implementation, the dimension information corresponding to each dimension is determined from a candidate dimension information set according to the second indication information, wherein the candidate dimension information set includes at least two dimension information corresponding to each dimension.

[0382] That is to say, the first device or the second device can determine the dimension information corresponding to each dimension of P dimensions based on the second indication information, and then the first device can determine the full rank matrix and second data corresponding to each dimension based on the dimension information corresponding to each dimension and the first data, and the second device can determine the dimensional size of the full rank matrix corresponding to each dimension based on each corresponding dimensional information, thereby realizing decoding of the compressed bit stream carrying the second data and the full rank matrix corresponding to each dimension.

[0383] It can be understood that the candidate dimension information set can be pre-configured by the protocol, or negotiated in advance between the first device and the second device, or indicated by the network side, and the embodiments of the present application do not specifically limit this.

[0384] In another possible implementation, the second indication information is also used to indicate the dimension information corresponding to each dimension. That is, the second indication information can be used to directly indicate the dimension information corresponding to each dimension, so that when the first device receives the second indication information, the first device can obtain the second data and the full rank matrix corresponding to each dimension based on the dimension information corresponding to each dimension indicated and the first data; when the second device receives the second indication information, the second device can determine the dimension size of the full rank matrix corresponding to each dimension based on the dimension information corresponding to each dimension, so as to decode the compressed bit stream carrying the second data and the full rank matrix corresponding to each dimension according to the dimension size.

[0385] It should be understood that, according to the relevant description of the aforementioned step S402, the dimension information corresponding to each dimension can be used to indicate the dimension size of each dimension after dimensionality reduction, or the proportional factor corresponding to each dimension; or, according to the relevant description of the aforementioned step S403, the dimension information corresponding to each dimension can be used to indicate the dimension size of each dimension in the reduced dimension set corresponding to each dimension after dimensionality reduction.

[0386] For example, the overhead of the indication information of the dimension information corresponding to each dimension is X bits, that is, the indication overhead required for the dimension information corresponding to P dimensions is P·X bits. The size of X may be pre-configured by the protocol or negotiated in advance between the first device and the second device, and this embodiment of the application does not specifically limit this.

[0387] For example, taking the case where P dimensions are the first and third dimensions of M dimensions, and the dimension information corresponding to each dimension is the dimension size after dimensionality reduction, the indication information of the P dimensions is the index #1 corresponding to the first dimension, the index #2 corresponding to the third dimension, and the dimension size after dimensionality reduction of the first dimension is N. s,1 , the dimension size of the third dimension after dimensionality reduction is N s,3 .

[0388] The second indication information may include indication information for indicating P dimensions, and indication information for indicating dimension information corresponding to each dimension, that is, the two indication information may be located in different cells. Where the two indication information are located in different cells, the correspondence between the dimensions and the dimension information between the two indication information may be sequentially corresponding, that is, the index #1 corresponding to the first dimension is in front, and the index #2 corresponding to the third dimension is in the back. Accordingly, N s,1 Also in N s,3 Before.

[0389] Alternatively, the above two indication information may be set in pairs, for example, the second indication information is: (index #1, N s,1 ), (index #3, N s,3 ).

[0390] It can be understood that the dimension information corresponding to each dimension is used to indicate the dimension size of each dimension in the dimension reduction dimension set corresponding to each dimension after dimension reduction. Taking case 1 in mode 2 as an example, the dimension information corresponding to the RX dimension may include: indication information of the TX dimension and the dimension size after dimension reduction of the TX dimension. Taking case 2 in mode 2 as an example, the dimension information corresponding to the RX dimension may include: indication information of the TX dimension, the dimension size after dimension reduction of the TX dimension, indication information of the frequency domain dimension, and the dimension size after dimension reduction of the frequency domain dimension.

[0391] It should be understood that the above description of the second indication information is only an example, and the second indication information may also indicate P dimensions and the dimensional information corresponding to each dimension in other ways, and the embodiments of the present application do not specifically limit this.

[0392] In a possible implementation, the second indication information is also used to indicate the dimension reduction matrix corresponding to each dimension. That is, the second indication information can also indicate the dimension reduction matrix of each dimension, so that the first device can reduce the dimension of the first data or the two-dimensional data corresponding to each dimension according to the indicated dimension reduction matrix, thereby eliminating the need for the first device to optimize the dimension reduction matrix offline, thereby reducing the power consumption of the first device.

[0393] For step S406:

[0394] It should be understood that when the second device obtains the dimension size of the first data and the dimension information corresponding to each dimension, the first device can decode the compressed bit stream carrying the second data and the full rank matrix corresponding to each dimension to obtain the second data and the full rank matrix corresponding to each dimension. In addition, the first device can determine the dimension size of the fourth data after the second data is decompressed according to the dimension size of the first data.

[0395] It can be understood that according to the above formula (1), the second device can determine the fourth data by using the inverse operation of formula (1), for example, performing tensor multiplication operation on the second data and the full rank matrix corresponding to each dimension, thereby obtaining the fourth data.

[0396] It can also be understood that the second device can also use other methods to determine the fourth data based on the second data and the full rank matrix corresponding to each dimension, and the embodiments of the present application do not specifically limit this.

[0397] In one possible implementation, the full rank matrix corresponding to each dimension is determined based on the first data, the third data, and the dimension information corresponding to each dimension, and the third data has the same dimension size as the first data. In other words, the first device can be combined with other data (such as the third data) with the same dimension size as the first data for joint processing to obtain the full rank matrix corresponding to each dimension that can be shared by the first data and the third data, and then the full rank matrix corresponding to each dimension does not need to be calculated again for the third data, thereby improving the efficiency of the first device in processing high-dimensional data with at least two dimensions of the same size.

[0398] It can be understood that according to the relevant description of method 1 in the aforementioned step S403, the full rank matrix corresponding to each dimension is determined according to the first data, the third data, and the dimension information corresponding to each dimension, including: the full rank matrix corresponding to each dimension is determined according to the feature matrix corresponding to each dimension, and the feature matrix corresponding to each dimension is determined according to the two-dimensional data corresponding to each dimension and the dimension reduction matrix corresponding to each dimension. Among them, the two-dimensional data corresponding to each dimension includes the two-dimensional data corresponding to the i-th dimension in the P dimensions, and the two-dimensional data corresponding to the i-th dimension is obtained by splicing the first two-dimensional data corresponding to the i-th dimension and the second two-dimensional data. The first two-dimensional data corresponding to the i-th dimension is the two-dimensional data obtained by merging the other dimensions in the first data except the i-th dimension, and the first two-dimensional data corresponding to the i-th dimension is the same as the first data. The second two-dimensional data corresponding to the i-th dimension is the two-dimensional data obtained by merging the other dimensions in the third data except the i-th dimension, and the second two-dimensional data corresponding to the i-th dimension is the same as the first data.

[0399] For example, the first device determines the full rank matrix corresponding to each dimension based on the first data, the third data, and the dimension information corresponding to each dimension, including: the first device performs tensor expansion on the first data and the third data to obtain two-dimensional data corresponding to each dimension, and the two-dimensional data corresponding to each dimension includes the first two-dimensional data corresponding to each dimension and the second two-dimensional data corresponding to each dimension; the first device determines the feature matrix corresponding to each dimension based on the two-dimensional data corresponding to each dimension and the dimensionality reduction matrix corresponding to each dimension; the first device decomposes the feature matrix corresponding to each dimension to obtain the full rank matrix corresponding to each dimension.

[0400] Specifically, Figure 6 Taking the example shown in the figure as an example, for the first data, the first two-dimensional data corresponding to the first dimension has a row dimension × column dimension of: N1×(N2×N3×…×N M ); For the third data, the second two-dimensional data corresponding to the first dimension has the same row dimension × column dimension as the first two-dimensional data. Thus, the two two-dimensional data can be concatenated to obtain the two-dimensional data corresponding to the first dimension, and its row dimension × column dimension is: N1×(N2×N3×…×N M ×2).

[0401] Without loss of generality, the first device may also combine the fifth data, the sixth data, or more data with the same dimension size as the first data to obtain two-dimensional data corresponding to the i-th dimension. Taking the joint processing of K data with the same dimension size as an example, the dimension size of the two-dimensional matrix corresponding to the i-th dimension is: N i ×(N1×N2×…×N j ×…×N M ×K), i is not included in [1,2,…,j,…,M].

[0402] In a possible implementation, the second data is determined based on the first data, the third data, and the full rank matrix corresponding to each dimension. For example, the first device determines the second data based on the full rank matrix corresponding to each dimension and the first data, which may include: the first device determines the second data based on the full rank matrix corresponding to each dimension, the first data, and the third data. The second data may include data after dimensionality reduction of the first data and data after dimensionality reduction of the third data.

[0403] That is to say, the first device can use the full rank matrix corresponding to each dimension to compress the first data and the third data respectively, and can obtain reduced-dimensional data of the first data and the second data, thereby further improving data compression efficiency and data transmission efficiency.

[0404] It can be understood that according to the relevant description of the aforementioned formula (1), the data after the first data dimension reduction can be the first core tensor Σ1, and the data after the third data dimension reduction can be the second core tensor Σ2. Without loss of generality, the first device can jointly process K data of the same dimensional size, and then the core tensor of each of the K data can be determined by formula (2).

[0405] Σ k =H k ×1Q1 H ×3Q2 H …× j Q i H …× M Q P H Formula (2)

[0406] Σ k It can represent the kth data H among K data k The corresponding core tensor, the full-rank matrix corresponding to P dimensions is Q1, ..., Q P .

[0407] It can be understood that the data sent by the first device to the second device includes: Σ1, ..., Σ k , …, Σ K , and Q1, …, Q P The second device receives the data from the first device, and can process the data according to the number of data K to be processed jointly, the dimension size of the first data, and the second data (Σ1, ..., Σ k , …, Σ K ), and the full-rank matrix corresponding to each dimension (Q1, ..., Q P ) is processed, for details, please refer to step S406, which will not be repeated here.

[0408] Optionally, the second indication information is also used to indicate whether to perform joint processing on the first data. Whether to perform joint processing on the first data may refer to performing joint processing on multiple data having the same dimension size as the first data. For example, the second indication information includes joint processing indication information on whether to perform joint processing on the first data, and the joint processing indication information may be indicated by 1 bit, such as bit 1 is used to indicate that the first data is to be jointly processed, and bit 0 is used to indicate that the first data is not to be jointly processed.

[0409] That is, the first device sends the second indication information to the second device, which can notify the second device that the second data includes data after multiple data are reduced in dimension, and then the second device should decompress the second data according to the number of data jointly processed when receiving. In addition, the second device sends the second indication information to the first device, which can notify the first device that multiple data with the same dimensional size can be jointly processed to improve compression efficiency and data transmission efficiency.

[0410] Furthermore, the second indication information can also be used to indicate the number of data to be jointly processed on the first data. That is, the first device sends the second indication information to the second device, which can notify the second device of the number of data to be jointly processed on the first data, and then when receiving, the second device can correctly decompress the second data according to the number of data to be jointly processed. In addition, the second device sends the second indication information to the first device, which can notify the first device of how many data with the same dimensional size are to be jointly processed.

[0411] Due to the embodiments of the present application, the first device can first determine P dimensions among the M dimensions of the first data (that is, which dimension needs to be reduced or compressed) and the dimensional size of the full rank matrix corresponding to each of the P dimensions through the dimensional information corresponding to each of the P dimensions. Therefore, there is no need to perform HOSVD decomposition and iterative operations on the high-dimensional data to determine the dimensional size of the full rank matrix corresponding to the P dimensions and each of the P dimensions. Therefore, the first device determines the second data and the full rank matrix corresponding to each dimension based on the dimensional size of the full rank matrix corresponding to each of the P dimensions and the first data, which can significantly reduce the computational complexity of compressing the first data.

[0412] It can be understood that in each of the above embodiments, the methods and / or steps implemented by the first device can also be implemented by components that can be used for the first device (such as a processor, chip, chip system, circuit, logic module, or software); the methods and / or steps implemented by the second device can also be implemented by components that can be used for the second device (such as a processor, chip, chip system, circuit, logic module, or software).

[0413] The above mainly introduces the scheme provided by the present application. Accordingly, the present application also provides a communication device, which is used to implement various methods in the above method embodiments. The communication device can be the first device in the above method embodiments, or a device including the first device, or a component that can be used for the first device, such as a chip or a chip system. Alternatively, the communication device can be the second device in the above method embodiments, or a device including the second device, or a component that can be used to calculate the second device, such as a chip or a chip system.

[0414] It is understandable that, in order to realize the above functions, the communication device includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0415] The embodiment of the present application can divide the functional modules of the communication device according to the above method embodiment. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiment of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation.

[0416] Take the communication device as the first device or the second device in the above method embodiment as an example, Figure 8 Schematic diagram of a communication device provided in an embodiment of the present application. Figure 8 As shown, the communication device 800 includes: a processing module 801 and a transceiver module 802. The processing module 801 is used to perform the processing function of the first device or the second device in the above method embodiment. The transceiver module 802 is used to perform the transceiver function of the first device or the second device in the above method embodiment.

[0417] Among them, all relevant contents of each step involved in the above method embodiment can be referred to the functional description of the corresponding functional module, and will not be repeated here.

[0418] Since the communication device 800 provided in this embodiment can execute the above-mentioned data transmission method, the technical effects that can be obtained can refer to the above-mentioned method embodiments and will not be repeated here.

[0419] In a possible design scheme, in the embodiment of the present application, the transceiver module 802 may include a receiving module and a sending module ( Figure 8 The transceiver module is used to implement the sending function and the receiving function of the communication device 800.

[0420] In a possible design solution, the communication device 800 may further include a storage module ( Figure 8(not shown in the figure), the storage module stores a program or instruction. When the processing module 801 executes the program or instruction, the communication device 800 can execute Figure 4 The functionality of the first device or the second device in the method shown.

[0421] It should be understood that the processing module 801 involved in the communication device 800 can be implemented by a processor or a processor-related circuit component, which can be a processor or a processing unit; the transceiver module 802 can be implemented by a transceiver or a transceiver-related circuit component, which can be a transceiver or a transceiver unit.

[0422] For example, Fig. 9 This is a schematic diagram of the structure of another communication device provided in an embodiment of the present application. The communication device may be a first device or a second device, or may be a chip (system) or other component or assembly that can be provided in the first device or the second device. Fig. 9 As shown, the communication device 900 may include a processor 901 .

[0423] In a possible design, the communication device 900 may further include a memory 902 and / or a transceiver 903. The processor 901 is coupled to the memory 902 and the transceiver 903, for example, via a communication bus.

[0424] Combine the following Fig. 9 The components of the communication device 900 are described in detail:

[0425] The processor 901 is the control center of the communication device 900, and may be a processor or a general term for multiple processing elements. For example, the processor 901 is one or more central processing units (CPUs), or may be application specific integrated circuits (ASICs), or may be configured to implement one or more integrated circuits of the embodiments of the present application, such as one or more microprocessors (digital signal processors, DSPs), or one or more field programmable gate arrays (FPGAs).

[0426] In one possible design, the processor 901 may perform various functions of the communication device 900 by running or executing a software program stored in the memory 902 and calling data stored in the memory 902 .

[0427] In a specific implementation, as an embodiment, the processor 901 may include one or more CPUs, such as Fig. 9CPU0 and CPU1 are shown in FIG.

[0428] In a specific implementation, as an embodiment, the communication device 900 may also include multiple processors, such as Fig. 9 901 and processor 904 are shown in FIG. Each of these processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). A processor herein may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0429] The memory 902 is used to store the software program for executing the solution of the present application, and the execution is controlled by the processor 901. The specific implementation method can refer to the above method embodiment, which will not be repeated here.

[0430] In a possible design, the memory 902 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 902 may be integrated with the processor 901, or may exist independently and access the processor 901 through the interface circuit ( Fig. 9 (not shown) is coupled to the processor 901, which is not specifically limited in the embodiment of the present application.

[0431] The transceiver 903 is used for communication with other communication devices. For example, if the communication device 900 is a terminal device, the transceiver 903 can be used to communicate with an NTN device, or with another terminal device. For another example, if the communication device 900 is an NTN device, the transceiver 903 can be used to communicate with a terminal device, or with another NTN device.

[0432] In one possible design, the transceiver 903 may include a receiver and a transmitter ( Fig. 9The receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0433] In a possible design, the transceiver 903 may be integrated with the processor 901, or may exist independently and communicate with the processor 901 through the interface circuit ( Fig. 9 (not shown) is coupled to the processor 901, which is not specifically limited in the embodiment of the present application.

[0434] It should be noted that Fig. 9 The structure of the communication device 900 shown in the figure does not constitute a limitation on the communication device. The actual communication device may include more or less components than those shown in the figure, or combine certain components, or arrange the components differently.

[0435] In addition, the technical effects of the communication device 900 can refer to the technical effects of the data transmission method described in the above method embodiment, and will not be repeated here.

[0436] In a possible implementation, an embodiment of the present application further provides a computer-readable storage medium, which stores a computer program or instructions, and the computer program or instructions implement the functions of the above method embodiment when executed by a computer.

[0437] In a possible implementation, the embodiment of the present application further provides a computer program product, which implements the functions of the above method embodiment when executed by a computer.

[0438] In a possible implementation, an embodiment of the present application further provides a communication system, which includes the first device described in the above method embodiment and the second device described in the above method embodiment.

[0439] In a possible implementation, an embodiment of the present application further provides a communication method, which includes the method described in any of the above method embodiments or any of its implementations.

[0440] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using a software program, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loading and executing computer program instructions on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. Computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server, data center, etc. that contains one or more servers that can be integrated with a medium. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a DVD), or a semiconductor medium (eg, a solid state disk (SSD)).

[0441] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0442] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0443] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0444] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0445] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0446] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.

[0447] Although the present application is described herein in conjunction with various embodiments, in the process of implementing the claimed application, those skilled in the art may understand and implement other changes to the disclosed embodiments by viewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "one" or "an" does not exclude multiple situations. A single processor or other unit may implement several functions listed in the claims. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0448] Although the present application has been described in conjunction with specific features and embodiments thereof, it is obvious that various modifications and combinations may be made thereto without departing from the scope of the present application. Accordingly, this specification and the drawings are merely exemplary illustrations of the present application as defined by the appended claims, and are deemed to have covered any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A data transmission method, characterized in that: The method comprises: Acquire first data, where the dimension of the first data is M dimensions, where M is an integer greater than 1; Determine dimensional information corresponding to each dimension of the P dimensions, where the dimensional information corresponding to each dimension is used to determine the dimensional size of the full rank matrix corresponding to each dimension, and the P dimensions are P dimensions of the M dimensions; Determine a full rank matrix and second data corresponding to each dimension according to the first data and the dimension information corresponding to each dimension; Send the full rank matrix corresponding to each dimension and the second data.

2. A data transmission method, characterized in that: The method comprises: Receive second data and a full-rank matrix corresponding to each dimension in P dimensions, where the second data is associated with the first data, the dimensions of the first data are M dimensions, and the P dimensions are P dimensions in the M dimensions; Obtaining the dimension size of the first data; The fourth data is determined according to the dimension size of the first data, the second data, and the full rank matrix corresponding to each dimension.

3. The method according to claim 1 or 2, characterized in that: The dimension information corresponding to each dimension includes the dimension information corresponding to the i-th dimension among the P dimensions, and the dimension information corresponding to the i-th dimension is used to indicate: the dimension sizes of other dimensions except the i-th dimension in the full rank matrix corresponding to the i-th dimension, i is greater than or equal to 1, and i is less than or equal to P.

4. The method according to any one of claims 1 to 3, characterized in that The full rank matrix corresponding to each dimension is a full rank matrix of the characteristic matrix corresponding to each dimension, and the characteristic matrix corresponding to each dimension is determined according to the first data and the dimension information corresponding to each dimension.

5. The method according to claim 4, characterized in that The feature matrix corresponding to each dimension is determined according to the first data and the dimension information corresponding to each dimension, and includes: The feature matrix corresponding to each dimension is determined based on the first data and the dimension reduction matrix corresponding to each dimension, wherein the dimension reduction matrix corresponding to each dimension is associated with the dimension information corresponding to each dimension.

6. The method according to claim 5, characterized in that In the case where M>2, the feature matrix corresponding to each dimension is determined according to the first data and the dimension reduction matrix corresponding to each dimension, including: The feature matrix corresponding to each dimension is determined based on the two-dimensional data corresponding to each dimension and the dimensionality reduction matrix corresponding to each dimension, the two-dimensional data corresponding to each dimension includes the two-dimensional data corresponding to the i-th dimension among the P dimensions, the two-dimensional data corresponding to the i-th dimension is the two-dimensional data obtained by merging the other dimensions in the first data except the i-th dimension, the two-dimensional data corresponding to the i-th dimension has the same elements as the first data, i is greater than or equal to 1, and i is less than or equal to P.

7. The method according to claim 5, characterized in that In the case of M>2, the feature matrix corresponding to each dimension is determined based on the first data and the dimensionality reduction matrix corresponding to each dimension, including: the feature matrix corresponding to each dimension is determined based on the feature tensor corresponding to each dimension, the feature tensor corresponding to each dimension is determined based on the first data and the dimensionality reduction matrix corresponding to each dimension, the feature matrix corresponding to each dimension includes the feature matrix corresponding to the i-th dimension among the P dimensions, the feature matrix corresponding to the i-th dimension is a feature matrix obtained by merging the dimensions other than the i-th dimension in the feature tensor corresponding to the i-th dimension, the feature matrix corresponding to the i-th dimension contains the same elements as the feature tensor corresponding to the i-th dimension, i is greater than or equal to 1, and i is less than or equal to P.

8. The method according to any one of claims 5 to 7, characterized in that: The dimension reduction matrix corresponding to each dimension is associated with the dimension information corresponding to each dimension, including: the parameters of the dimension reduction matrix corresponding to each dimension are associated with the dimension information corresponding to each dimension, wherein the parameters of the dimension reduction matrix corresponding to each dimension include at least one of the following: The proportion, position distribution, or value range of non-zero elements in the dimensionality reduction matrix corresponding to each dimension.

9. The method according to any one of claims 5 to 8, characterized in that: The proportion of non-zero elements in the dimensionality reduction matrix corresponding to each dimension is greater than or equal to a first threshold corresponding to each dimension, and the first threshold corresponding to each dimension is determined based on the dimensional size of the first data and the dimensional information corresponding to each dimension.

10. The method according to any one of claims 5 to 9, characterized in that: The dimension reduction matrix corresponding to each dimension is determined according to the dimension information corresponding to each dimension.

11. The method according to claim 10, characterized in that The dimensionality reduction matrix corresponding to each dimension is determined based on the dimensionality information corresponding to each dimension, including: the dimensionality reduction matrix corresponding to each dimension is determined from a set of candidate dimensionality reduction matrices based on the dimensionality information corresponding to each dimension, and the set of candidate dimensionality reduction matrices includes at least two dimensionality reduction matrices.

12. The method according to any one of claims 1 to 11, characterized in that The second data is determined based on the first data and the full rank matrix corresponding to each dimension, and the full rank matrix corresponding to each dimension is determined based on the first data and the dimension information corresponding to each dimension.

13. The method according to any one of claims 1 to 12, characterized in that The full rank matrix corresponding to each dimension is determined based on the first data, the third data, and the dimension information corresponding to each dimension, and the third data has the same dimension size as the first data.

14. The method according to claim 13, characterized in that The second data is determined based on the first data, the third data, and the full rank matrix corresponding to each dimension.

15. The method according to any one of claims 1, 3-14, characterized in that: The method further comprises: Send first indication information, where the first indication information is used to indicate the dimension size of the first data.

16. The method according to any one of claims 2 to 14, characterized in that The method further comprises: First indication information is received, where the first indication information is used to indicate a dimension size of the first data.

17. The method according to claim 15 or 16, characterized in that The first indication information includes index information, where the index information is used to determine a dimension size of the first data from a candidate dimension set, where the candidate dimension set includes at least two dimension sizes.

18. The method according to any one of claims 15 to 17, characterized in that: The first indication information also includes indication information of the candidate dimension set.

19. The method according to any one of claims 1 to 18, characterized in that The method further comprises: Send second indication information, where the second indication information is used to indicate the P dimensions.

20. The method according to any one of claims 1 to 18, characterized in that The method further comprises: Second indication information is received, where the second indication information is used to indicate the P dimensions.

21. The method according to claim 19 or 20, characterized in that The second indication information is also used to indicate dimension information corresponding to each dimension.

22. The method according to any one of claims 19 to 21, characterized in that The second indication information is also used to indicate the dimension reduction matrix corresponding to each dimension.

23. The method according to claim 20, characterized in that The dimension information corresponding to each dimension is determined from a candidate dimension information set according to the second indication information, and the candidate dimension information set includes at least two dimension information corresponding to each dimension.

24. A communication device, characterized in that: The communication device includes a module or unit for executing the method of any one of claims 1, 3-15, 17-23, or includes a module or unit for executing the method of any one of claims 2-14, 16-23.

25. A communication device, characterized in that: The communication device comprises a processor, and the processor is used to enable the communication device to execute the method according to any one of claims 1, 3-15, 17-23 through logic circuits and / or execution instructions, or enable the communication device to execute the method according to any one of claims 2-14, 16-23.

26. The communication device according to claim 24, characterized in that The communication device further comprises a memory, wherein the memory is used to store the instruction.

27. The communication device according to claim 24 or 25, characterized in that: The communication device further comprises a communication interface, which is used for inputting and / or outputting signaling and / or data.

28. A computer-readable storage medium, characterized in that: The computer-readable storage medium comprises instructions, which, when executed by a processor, enable the method according to any one of claims 1, 3-15, 17-23, or the method according to any one of claims 2-14, 16-23 to be implemented.

29. A computer program product, characterized in that The computer program product comprises instructions, which, when executed on a computer, cause the computer to perform the method according to any one of claims 1, 3-15, 17-23, or cause the computer to perform the method according to any one of claims 2-14, 16-23.

30. A communication system, characterized in that: The communication system comprises a first device and a second device, wherein the first device is used to perform the method according to any one of claims 1, 3-15, 17-23, and the second device is used to perform the method according to any one of claims 2-14, 16-23.

Images