Training method for communication system, communication method, device, system, and storage medium

Abstract

The present application provides a training method for a communication system, a communication method, an electronic device, a communication system, and a storage medium. The method comprises: processing target data by means of a power allocation matrix generation network corresponding to a target number of streams, to obtain a power allocation matrix and a pilot position selection matrix corresponding to each stream; on the basis of the power allocation matrix corresponding to each stream, superimposing, on a time-frequency resource indicated by the pilot position selection matrix corresponding to each stream, first original data and first pilot data of each stream to be sent by a transmitter, so as to obtain first superimposed data of each stream; acquiring received data of each stream at a receiver, the received data of each stream being obtained by passing the first superimposed data of each stream through a wireless channel; decoding the received data of each stream by means of the receiver to obtain first decoded data of each stream; and training parameters of the power distribution matrix generation network and the receiver on the basis of the first original data and the first decoded data of each stream. The method can improve the communication performance of communication systems.

Description

Training method of communication system, communication method, device, system and storage medium TECHNICAL FIELD

[0001] The present application relates to the technical field of communication, for example, to a training method of a communication system, a communication method, an electronic device, a communication system and a storage medium. BACKGROUND

[0002] The current communication system generally uses time division multiplexing or frequency division multiplexing pilot sequence to obtain channel state information, but the pilot sequence occupies the time slot or bandwidth that can originally transmit data, loses the information rate, and reduces the bandwidth utilization. Therefore, the superimposed pilot technology emerges as the times require, which embeds low-power pilot symbols into the transmitted signal, and the receiver estimates the channel by using the known pilot sequence and the statistical characteristics of the received signal. Since the pilot no longer occupies an independent time slot or subcarrier, the spectrum resources can be effectively saved and the spectrum efficiency can be improved.

[0003] However, the superimposed signals will interfere with each other, especially in a multi-input multi-output (MIMO) system, the superimposed signals of multiple antennas may cause large errors in channel estimation results, resulting in low communication performance. SUMMARY

[0004] The embodiments of the present application provide a training method of a communication system, a communication method, an electronic device, a communication system and a storage medium.

[0005] In a first aspect, the embodiments of the present application provide a training method of a communication system, the communication system comprising a power allocation matrix generation network, a transmitter and a receiver, and the method comprising:

[0006] processing the target data through the power allocation matrix generation network corresponding to the target stream number to obtain the power allocation matrix corresponding to each stream and the pilot position selection matrix;

[0007] superimposing the first original data and the first pilot data of each stream to be sent by the transmitter on the time-frequency resources indicated by the pilot position selection matrix corresponding to each stream according to the power allocation matrix corresponding to each stream, to obtain the first superimposed data of each stream;

[0008] obtaining the received data of each stream of the receiver, the received data of each stream being obtained after the first superimposed data of each stream passes through a wireless channel;

[0009] decoding the received data of each stream through the receiver to obtain the first decoded data of each stream;

[0010] The power allocation matrix generation network and the parameters of the receiver are trained according to the first original data and the first decoded data of each stream.

[0011] In a second aspect, an embodiment of the present application provides a communication method, comprising:

[0012] The transmitter determines, from the first preset mapping relationship, a target power allocation matrix and a target pilot position selection matrix corresponding to each stream under the current number of streams;

[0013] The transmitter superimposes, according to the target power allocation matrix corresponding to each stream, the second original data and the second pilot data of each stream on the time-frequency resources indicated by the target pilot position selection matrix corresponding to each stream, to obtain second superimposed data of each stream;

[0014] The transmitter transmits target signaling and the second superimposed data of each stream, wherein the current number of streams is included in the target signaling.

[0015] The receiver receives the target signaling and the second superimposed data of each stream.

[0016] The receiver determines, from the second preset mapping relationship, target receiver parameters corresponding to the current number of streams.

[0017] The receiver decodes the second superimposed data of each stream based on the target receiver parameters.

[0018] The first preset mapping relationship includes power allocation matrices and pilot position selection matrices corresponding to each stream under different numbers of streams, information in the first preset mapping relationship is obtained by inputting target data into a pre-trained power allocation matrix generation network corresponding to each number of streams, the second preset mapping relationship includes receiver parameters corresponding to different numbers of streams, and the power allocation matrix generation network corresponding to each number of streams and the receiver parameters are trained by using the method provided in the first aspect of the present application.

[0019] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the method provided in the first aspect of the present application when executing the computer program.

[0020] In a fourth aspect, an embodiment of the present application provides a communication system, comprising a power allocation matrix generation network, a transmitter and a receiver.

[0021] The transmitter is configured to determine, from the first preset mapping relationship, a target power allocation matrix and a target pilot position selection matrix corresponding to each stream under the current number of streams; superimpose second original data and second pilot data of each stream on time-frequency resources indicated by the target pilot position selection matrix corresponding to each stream according to the target power allocation matrix corresponding to each stream to obtain second superimposed data of each stream; and transmit target signaling and the second superimposed data of each stream, wherein the target signaling includes the current number of streams.

[0022] The receiver is configured to receive the target signaling and the second superimposed data of each stream; determine a target receiver parameter corresponding to the current number of streams from the second preset mapping relationship; and decode the second superimposed data of each stream based on the target receiver parameter.

[0023] The first preset mapping relationship includes power allocation matrices and pilot position selection matrices corresponding to each stream under different numbers of streams, and information in the first preset mapping relationship is obtained by inputting target data into a power allocation matrix generation network corresponding to each number of streams that is pre-trained, and the second preset mapping relationship includes receiver parameters corresponding to different numbers of streams, and the power allocation matrix generation network corresponding to each number of streams and the receiver parameters are trained by using the method provided in the first aspect of the present application.

[0024] In a fifth aspect, an embodiment of the present application provides a storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method provided in the first aspect or the second aspect of the present application.

[0025] The technical solution provided in the embodiments of the present application generates a power allocation matrix and a pilot position selection matrix corresponding to each stream through a power allocation matrix generation network, superimposes first original data and first pilot data of each stream by using the obtained power allocation matrix and pilot position selection matrix, obtains first superimposed data, decodes received data of each stream through a receiver to obtain first decoded data of each stream, and trains parameters of the power allocation matrix generation network and the receiver by using the first original data and the first decoded data of each stream, so that the obtained parameters are more optimal. In this way, in a communication process based on superimposed pilots, the power allocation matrix generation network can generate more optimal power allocation matrices and pilot position selection matrices, and the received signals are decoded through more optimal receiver parameters, which reduces pilot interference between superimposed signals of each stream, improves channel estimation results, and thus improves decoding effects. BRIEF DESCRIPTION OF DRAWINGS

[0026] FIG. 1 is a structural schematic diagram of a communication system provided in an embodiment of the present application;

[0027] Figure 2 is a schematic diagram of the working principle of the communication system provided in an embodiment of this application;

[0028] Figure 3 is a flowchart illustrating a training method for a communication system provided in an embodiment of this application.

[0029] Figure 4 is a schematic diagram of a power allocation matrix generation network provided in an embodiment of this application;

[0030] Figure 5 is a schematic diagram of a receiver provided in an embodiment of this application;

[0031] Figure 6 is a flowchart illustrating a communication method provided in an embodiment of this application;

[0032] Figure 7 is a schematic diagram of a training device for a communication system provided in an embodiment of this application;

[0033] Figure 8 is a schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0034] It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0035] Figure 1 is a schematic diagram of a communication system provided in an embodiment of this application. As shown in Figure 1, the communication system may include a power allocation matrix generation network, a transmitter, and a receiver, wherein the power allocation matrix generation network is used to generate power allocation matrices (power allocation matrices A1, A2, and A3 shown in Figure 1). M The transmitter uses the generated power allocation matrix and pilot position selection matrix to superimpose the raw data and pilot data to be transmitted, and then transmits the superimposed data. The receiver decodes the received data. By transmitting the superimposed pilot data and raw data, the pilot data does not occupy time-frequency resources separately, thus improving the spectrum utilization efficiency of the communication system. The power allocation matrix generation network and the receiver described above can be constructed using neural networks. Therefore, the communication system can be trained end-to-end, enabling the power allocation matrix generation network in the communication system to output a better power allocation matrix and pilot position selection matrix, reducing pilot interference between superimposed signals of different streams, and allowing the receiver to decode the received data with more optimized parameters, thereby improving the communication performance of the communication system.

[0036] The communication system provided by the embodiments of the present application mainly comprises a model training module and a model application module, as shown in FIG. 2. The model training module adopts an end-to-end architecture for joint training and optimization. After the training is completed, the power allocation matrix and the pilot position selection matrix output by the module can be stored according to the flow number as an index (for example, the first preset mapping relationship in FIG. 2), or the matrix obtained by multiplying the power allocation matrix and the pilot position selection matrix can be stored according to the flow number as an index, and the receiver parameters can also be stored according to the flow number as an index (for example, the second preset mapping relationship in FIG. 2). In this way, the model application module can select the corresponding power allocation matrix and pilot position selection matrix according to the actual flow number for pilot superposition processing, and select the corresponding receiver parameters according to the actual flow number for decoding the received data. Here, the communication system can be regarded as a model for overall training.

[0037] Next, the model training process is introduced. FIG. 3 is a flowchart of a training method of the communication system provided by the embodiments of the present application. As shown in FIG. 3, the method can comprise the following steps.

[0038] S301, processing the target data by the power allocation matrix generation network corresponding to the target flow number to obtain the power allocation matrix and the pilot position selection matrix corresponding to each flow.

[0039] The target data is the input data of the power allocation matrix generation network, which can be any set of vectors without constraints on the length and amplitude. The target flow number is related to the flow number supported by the communication system. For example, assuming that the communication system is a Single Input Single Output (SISO) system, the target flow number is 1, and assuming that the communication system is a MIMO system, the target flow number is related to the multiple flow number supported by the MIMO system.

[0040] A set of fixed vectors is selected as the target data and input into the power allocation matrix generation network corresponding to the target flow number to obtain the power allocation matrix and the pilot position selection matrix corresponding to each flow. For example, assuming that the communication system comprises L flows, the target data is input into the power allocation matrix generation network corresponding to the L flows to obtain the power allocation matrix A1…AL and the pilot position selection matrix B1…BL corresponding to the L flows respectively. L . L Among them, A1, B1 correspond to the first flow, A2, B2 correspond to the second flow, and so on, A L , B L correspond to the Lth flow.

[0041] S302, superimposing the first original data and the first pilot data of each flow to be sent by the transmitter on the time-frequency resources indicated by the pilot position selection matrix corresponding to each flow according to the power allocation matrix corresponding to each flow to obtain the first aliasing data of each flow.

[0042] The power allocation matrix is used to indicate the power allocated by each time-frequency resource of the scheduling, and the pilot position selection matrix is used to indicate the time-frequency resource of the pilot superposition, so that after obtaining the power allocation matrix and the pilot position selection matrix corresponding to each stream, the first original data and the first pilot data of each stream to be transmitted by the transmitter can be superimposed on the time-frequency resource indicated by the pilot position selection matrix corresponding to each stream according to the power allocation matrix corresponding to each stream, to obtain the first mixed data of each stream, that is, the first original data and the first pilot data are superimposed on the same time-frequency resource. The specific superposition process can be shown in the following formula 1:

[0043] Wherein, C L is the matrix after the point multiplication of the power allocation matrix and the pilot position selection matrix of the Lth stream, is the first pilot data of the Lth stream, is the first original data of the Lth stream, is the first mixed data of the Lth stream, the first pilot data and the first original data are superimposed on the same time-frequency resource indicated by the pilot position selection matrix, and the above L is a positive integer.

[0044] After obtaining the first mixed data, the first mixed data is transmitted through a wireless channel, and the receiver receives the first mixed data after the wireless channel transmission.

[0045] S303, obtaining the received data of the receiver for each stream.

[0046] Wherein, the received data of each stream is the first mixed data of each stream after the wireless channel, which can be a channel simulated by a target channel data set. The target channel data set can include a real channel data set or a channel data set generated by simulation software.

[0047] As an optional implementation, the process of S303 can be: obtaining a target channel data set, processing the mixed data of each stream through a channel simulated by the target channel data set, to obtain the received data of the receiver for each stream.

[0048] By collecting real channel data sets or channel data sets generated by simulation software, various wireless channel environments can be simulated through these channel data sets, and the first mixed data generated by the transmitter is transmitted by using the simulated wireless channel environment. The first mixed data after the transmission of the wireless channel environment is determined as the data received by the receiver.

[0049] Optionally, after obtaining the received data of each stream by the receiver, the received data of each stream can also be energy normalized to reduce abnormal data, so that the processed received data meets the input requirements of the receiver, facilitating the processing of the receiver.

[0050] S304, decoding the received data of each stream by the receiver to obtain first decoded data of each stream.

[0051] Obtaining the received data of each stream received by the receiver, taking the received data of each stream as the input data of the receiver, and decoding the received data of each stream by the receiver to obtain first decoded data of each stream.

[0052] Optionally, the above S304 can include: fusing the received data of each stream, the first pilot data, the power allocation matrix and the pilot position selection matrix to obtain first fused data of each stream; and decoding the first fused data of each stream by the receiver to obtain first decoded data of each stream.

[0053] Specifically, the power allocation matrix and the pilot position selection matrix corresponding to each stream are multiplied to obtain a target matrix corresponding to each stream, the real and imaginary parts of the received data of each stream, the real and imaginary parts of the first pilot data, and the real and imaginary parts of the target matrix are superimposed in the third dimension of the three-dimensional matrix to obtain first fused data, and the first fused data is input to the receiver for decoding processing to obtain first decoded data of each stream.

[0054] S305, training the parameters of the power allocation matrix generation network and the receiver according to the first original data and the first decoded data of each stream.

[0055] The first original data of each stream is taken as the expected output of the communication system, the first decoded data of each stream is taken as the actual output of the communication system, and the parameters of the power allocation matrix generation network and the receiver in the communication system are trained based on a preset loss function. The loss function can be binary cross entropy or other types of loss function, which is not limited by the embodiments of the application.

[0056] Optionally, the above S305 can be: determining a loss value of the preset loss function according to the first original data and the first decoded data of each stream; updating the parameters of the power allocation matrix generation network and the receiver according to the loss value, and continuing to execute the step of processing the target data by the power allocation matrix generation network corresponding to the target stream number to obtain the power allocation matrix and the pilot position selection matrix corresponding to each stream until the loss value of the preset loss function meets the preset requirement or the model convergence condition is reached.

[0057] For a communication system with different target stream numbers, the processes of S301-S305 can be performed to train, so as to obtain the power allocation matrix generation network and the receiver corresponding to each stream under different stream numbers, and the power allocation matrix and the pilot position selection matrix output by the power allocation matrix generation network are correspondingly stored with different stream numbers as indexes, and the receiver parameters are correspondingly stored with different stream numbers as indexes, thereby saving the time and computing resources of model inference in the model application stage, and improving the real-time performance of the communication system.

[0058] Optionally, as shown in FIG. 4, the power allocation matrix generation network can include an input layer, a first feature extraction layer, an activation layer, and a classification decision layer, and the input layer is used to receive target data. Correspondingly, S301 can include: performing feature extraction on the target data through the first feature extraction layer to obtain feature data; performing activation processing on the feature data through the activation layer to obtain the power allocation matrix corresponding to each stream; and performing processing on the target data through the classification decision layer to obtain the pilot position selection matrix corresponding to each stream, and the pilot position selection matrix corresponding to each stream is different.

[0059] Specifically, the first feature extraction layer and the activation layer are used to generate the power allocation matrix. Since the target data is fixed, the power allocation matrix output by the power allocation matrix generation network is composed of the parameters of the power allocation matrix generation network. By optimizing the parameters of the power allocation matrix generation network, the optimal power allocation matrix can be obtained. The activation layer can include multiple fully connected layers and an activation function, which can be a sigmoid function or other types of activation functions. The classification decision layer performs L+1 multi-classification on L streams on each scheduled time-frequency resource, which is used to determine on which stream the time-frequency resource at this time is used to superimpose the first pilot data and the first original data. For example, assuming that there are L+1 labels in the classification result, L labels correspond to L streams one by one. If the decision result is that the time-frequency resource belongs to the first label, the time-frequency resource is used to superimpose the first pilot data and the first original data of the first stream. If the decision result is that the time-frequency resource belongs to the second label, the time-frequency resource is used to superimpose the first pilot data and the first original data of the second stream. Similarly, if the decision result is that the time-frequency resource belongs to the Lth label, the time-frequency resource is used to superimpose the first pilot data and the first original data of the Lth stream. If the decision result is that the time-frequency resource belongs to the L+1th label, the time-frequency resource is only used to transmit the first original data of each stream without superimposing the first pilot data of any stream.

[0060] The first feature extraction layer and the activation layer can allocate more optimal power allocation matrices for each stream, and the classification decision layer can stagger the pilots superimposed between different streams to reduce the pilot interference between streams, thereby improving the performance of the communication system.

[0061] Optionally, as shown in FIG. 5, the receiver can include an input layer, a dimension increasing layer, a second feature extraction layer, and a decision layer. Specifically, the data input into the receiver is first processed by the dimension increasing layer, then the data after dimension increasing is extracted by the second feature extraction layer, and then the extracted feature data is input into the decision layer for processing, so as to obtain the corresponding first decoding data.

[0062] The dimension increasing layer is used to aggregate feature information from different channels, so that the information from different channels can be better fused together to generate more abundant and diverse feature representations. The second feature extraction layer can include at least one local feature extraction layer and at least one global feature extraction layer, which are arranged alternately, wherein the local feature extraction layer adopts a multi-layer convolutional neural network (CNN) architecture, the CNN automatically learns spatial hierarchical features through multi-layer convolution, and can effectively extract local features of data; the global feature extraction layer adopts a Transformer architecture, the Transformer globally models through a self-attention mechanism, and can effectively handle long-distance dependencies. By alternately arranging the local feature extraction layer and the global feature extraction layer, the overall performance of the receiver is improved. The decision layer can include multiple convolutional layers and an activation function, which can be a sigmoid function or other types of activation functions, which are not limited in the embodiment.

[0063] The training method of the communication system provided in the embodiment of the present application generates the power allocation matrix and the pilot position selection matrix corresponding to each stream through the power allocation matrix generation network, superimposes the first original data and the first pilot data of each stream by using the obtained power allocation matrix and pilot position selection matrix to obtain first superimposed data, decodes the received data of each stream through the receiver to obtain first decoding data of each stream, and trains the parameters of the power allocation matrix generation network and the receiver by using the first original data and the first decoding data of each stream, so that the obtained parameters are more optimal. In this way, in the communication process based on superimposed pilots, the power allocation matrix generation network can generate more optimal power allocation matrix and pilot position selection matrix, and the received signal is decoded through more optimal receiver parameters, which reduces the pilot interference between superimposed signals of each stream, improves the channel estimation result, and thus improves the decoding effect.

[0064] Next, the model application process is specifically introduced. FIG. 6 is a flowchart of a communication method provided in an embodiment of the present application. As shown in FIG. 6, the method includes:

[0065] S601, the transmitter determines the target power allocation matrix and the target pilot position selection matrix corresponding to each stream under the current number of streams from the first preset mapping relationship.

[0066] wherein the first preset mapping relationship includes the power allocation matrix and the pilot position selection matrix corresponding to each stream under different stream numbers, the information in the first preset mapping relationship is obtained by inputting the target data into the pre-trained power allocation matrix generation network corresponding to each stream number, and the second preset mapping relationship includes the receiver parameters corresponding to each stream number, and the power allocation matrix generation network and the receiver parameters corresponding to each stream number are trained by using the training method of the communication system provided in any one of the above embodiments.

[0067] That is, after the communication system under different stream numbers is trained, the power allocation matrix and the pilot position selection matrix output by the power allocation matrix generation network in the communication system are indexed by the stream number to construct the first preset mapping relationship, and the receiver parameters are indexed by the stream number to construct the second preset mapping relationship.

[0068] In this way, in the model application stage, the transmitter can query the first preset mapping relationship based on the current stream number, so as to obtain the target power allocation matrix and the target pilot position selection matrix corresponding to each stream under the current stream number.

[0069] S602, the transmitter superimposes the second original data and the second pilot data of each stream on the time-frequency resources indicated by the target pilot position selection matrix corresponding to each stream according to the target power allocation matrix corresponding to each stream, to obtain the second superposition data of each stream.

[0070] The target power allocation matrix is used to indicate the power allocated by each time-frequency resource of the scheduling, and the target pilot position selection matrix is used to indicate the time-frequency resources of the pilot superposition, so after the target power allocation matrix and the target pilot position selection matrix corresponding to each stream are obtained, the second original data and the second pilot data of each stream to be sent by the transmitter can be superimposed on the time-frequency resources indicated by the target pilot position selection matrix corresponding to each stream according to the target power allocation matrix corresponding to each stream, to obtain the second superposition data of each stream, that is, the second original data and the second pilot data are superimposed on the same time-frequency resource. The specific superposition process can be shown in the following formula 2:

[0071] wherein C' = C L · C L, C L is the target power allocation matrix of the Lth stream, and C L is the target pilot position selection matrix of the Lth stream. L is the matrix after the point multiplication of the target power allocation matrix and the target pilot position selection matrix of the Lth stream, is the second pilot data of the Lth stream, is the second original data of the Lth stream, is the second superposition data of the Lth stream, and the second pilot data and the second original data are superimposed on the same time-frequency resource indicated by the target pilot position selection matrix.

[0072] It should be noted that before the pilot superposition, the second original data and the second pilot data also need to be channel encoded and modulated, and the specific encoding mode and modulation mode are not limited.

[0073] S603, the transmitter sends the target signaling and the second superposition data of each stream.

[0074] The transmitter performs Orthogonal Frequency Division Multiplexing (OFDM) symbol shaping processing on the second superposition data after the pilot superposition and loads a Cyclic Prefix (CP) to obtain an OFDM signal to be transmitted, and sends the target signaling and the OFDM signal. The current stream number is included in the target signaling, so that the receiver can determine the receiver parameters used for decoding based on the current stream number.

[0075] S604, the receiver receives the target signaling and the second superposition data of each stream.

[0076] The receiver performs CP removal and FFT and other related processing on the received OFDM signal to obtain the second superposition data. Optionally, the receiver can also perform energy normalization processing on the second superposition data.

[0077] S605, the receiver determines the target receiver parameters corresponding to the current stream number from the second preset mapping relationship.

[0078] The second preset mapping relationship includes the mapping relationship between different stream numbers and receiver parameters, so that after receiving the target signaling, the receiver can query the second preset mapping relationship based on the current stream number carried in the target signaling to obtain the target receiver parameters required for decoding.

[0079] S606, the receiver decodes the second superposition data of each stream based on the target receiver parameters.

[0080] After obtaining the target receiver parameters, the receiver directly decodes the second superposition data based on the target receiver parameters. In order to obtain better decoding effect, optionally, the receiver can also fuse the second superposition data of each stream, the second pilot data, the target power allocation matrix and the target pilot position selection matrix to obtain the second fusion data of each stream; the receiver decodes the second fusion data of each stream based on the target receiver parameters.

[0081] Specifically, the receiver side stores the first preset mapping relationship in addition to the second preset mapping relationship, and the receiver can query the first preset mapping relationship based on the current stream number carried in the target signaling to obtain the target power allocation matrix and the target pilot position selection matrix corresponding to each stream under the current stream number, and multiply the target power allocation matrix and the target pilot position selection matrix corresponding to each stream to obtain an independent comprehensive power allocation matrix between the streams. The real part and the imaginary part of the second aliasing data of each stream, the real part and the imaginary part of the second pilot data, and the real part and the imaginary part of the comprehensive power allocation matrix are superimposed in the third dimension of the three-dimensional matrix to obtain second fusion data, and the second fusion data is decoded and processed based on the target receiver parameter.

[0082] The communication method provided by the embodiments of the present application can obtain a more optimal target power allocation matrix, a target pilot position selection matrix, and a target receiver parameter by querying the first preset mapping relationship and the second preset mapping relationship based on the current stream number. The target pilot position selection matrix makes the pilot positions of the superimposed streams staggered with each other, reduces the interference between the superimposed pilots of the streams, and the target receiver parameter can better decode the received data, thereby improving the communication performance in the superimposed pilot transmission process.

[0083] The training process and the application process of the communication system are further introduced below by taking specific examples as examples:

[0084] Example one: assuming that the data in an OFDM SISO communication system is single stream, the number of scheduled RBs is 24, the number of REs is 288, the number of symbols is 14, the data signal can be modulated by Quadrature Amplitude Modulation (QAM), such as 16QAM, the pilot signal can be modulated by Binary Phase Shift Keying (BPSK), and the channel data set can be obtained by field collection. Specifically, the model training process is as follows:

[0085] 1. Channel data set collection and arrangement

[0086] The frequency domain channel data is obtained by field collection to construct a data set, and the channel data is normalized. The data set is used to simulate a wireless channel.

[0087] 2. Generate the transmitting end data

[0088] In this scenario, the communication system is a single stream signal, and the bit data X bit ′ of the transmitting end is randomly generated, and the modulated data symbol X symThen, a set of randomly generated vectors with no limit on length and amplitude are selected as the input of the power allocation matrix generation network to obtain the power allocation matrix A1 and the pilot position selection matrix B1, and the power allocation matrix A1 and the pilot position selection matrix B1 are point multiplied to obtain the matrix C1. At the same time, the pilot data is modulated using BPSK to obtain the modulated pilot data X p Then, the modulated data symbol and the pilot symbol are layer mapped, and the superposition process is shown in the following formula 3:

[0089] 3. Generating and processing the receiver data

[0090] The frequency domain signal X mix is obtained after the superposition of the transmitter data and the pilot, and then the receiver frequency domain signal Y mix is generated by simulating the channel process using the frequency domain channel data set in step 1. mix The energy normalization processing is performed on the signal Y nor using the following formula 4 to obtain the processed signal Y nor ′. max (Y mix ′) Formula 4

[0091] After that, the data is integrated, and the integration method is to superimpose the real part and the imaginary part of the signal Y nor ′, the real part and the imaginary part of the pilot signal X p ′, and the real part and the imaginary part of the matrix C1 obtained by point multiplying the power allocation matrix and the pilot position selection matrix in the third dimension of the three-dimensional matrix to obtain the integrated signal Y in ′.

[0092] 4. LOSS calculation and parameter update

[0093] The integrated signal Y in ′ in step 3 is used as the input data of the receiver, and the integrated output bit value Y bit ′ is obtained after decoding by the receiver. According to the bit value Y bit ′ output by the receiver and the original bit value X bit ′, the binary cross entropy is calculated using the following formula 5 to obtain the Loss value of the model. Loss = BCE(X bit ′, Y bit ′) Formula 5

[0094] The parameters of the power allocation matrix generation network and the receiver are updated based on the calculated loss value, and steps 2 to 4 are repeated iteratively until the model converges. After the model is trained, the power allocation matrix and pilot position selection matrix (or the matrix resulting from the dot product of the power allocation matrix and pilot position selection matrix) output by the power allocation matrix generation network are stored using the stream number as the index, and the receiver parameters are also stored using the stream number as the index.

[0095] The model application process is as follows:

[0096] 1. Generate transmitter data

[0097] In this scenario, the communication system uses a single-stream signal, transmitting bit data X from the transmitting end. bit "After channel coding and 16QAM modulation, the modulated data symbol X is obtained." sym Next, the power allocation matrix A1 and pilot position selection matrix B1 are obtained by looking up the table based on the current flow number. Simultaneously, the pilot data is modulated using BPSK to obtain the modulated pilot data X. p Then, the data symbols and pilot symbols are layer-mapped, and the data symbols and pilot symbols are superimposed on the same time-frequency resources according to the power allocation matrix and the pilot position selection matrix. The superposition process is shown in Formula 6 below:

[0098] 2. For the aliased data X after pilot signal superposition mix "Perform OFDM symbol shaping and load CP to obtain the OFDM signal to be transmitted, send target signaling to indicate the current stream number, and send the above OFDM signal after sending the target signaling."

[0099] 3. The receiver performs related processing such as deCP and FFT on the received OFDM signal.

[0100] 4. Processing received data

[0101] After processing in step 3, the received frequency domain signal Y is obtained. mix "The receiver parameters, power allocation matrix A1, and pilot position selection matrix B1 are selected according to the target signaling indication of stream number 1. Then, the signal Y is processed according to the following formula 7." mix "Perform energy normalization to obtain the processed signal Y." nor ". Y nor " = Normal max (Y mix ″) Formula 7

[0102] Next, the data is integrated. The integration method is to integrate the signal Y... norThe real and imaginary parts of the pilot signal X p The real and imaginary parts of the matrix C1, obtained by multiplying the real and imaginary parts of the matrix C1 after multiplying the power allocation matrix and pilot position selection matrix selected based on the current flux number, are superimposed in the third dimension of the three-dimensional matrix to obtain the integrated signal Y. in "The integrated signal Y" in "As input data to the receiver, the receiver parameters obtained from the above table lookup are used to adjust Y." in "Decode the output bit value Y after integration." bit ″.

[0103] Example 2: Assume an OFDM MIMO communication system with two data streams, 6 scheduled Receptors (RBs), 72 scheduled Receivers (REs), and 14 symbols. The data signal uses 64QAM modulation, and the pilot signal uses Quadrature Phase Shift Keying (QPSK) modulation. The channel dataset is channel data generated by simulation software. Specifically, the model training process is as follows:

[0104] 1. Collection and organization of channel datasets

[0105] A dataset was constructed using similar simulated frequency domain channel data, and the channel data was normalized.

[0106] 2. Generate transmitter data

[0107] In this scenario, the communication system consists of two streams of signals, with randomly generated bit data X at the transmitting end. bit1 After 64QAM modulation, the modulated data symbol X is obtained. sym1 Next, a set of randomly generated vectors of unlimited length and amplitude are selected as input data for the power allocation matrix generation network to obtain power allocation matrices A1 and A2, and pilot position selection matrices B1 and B2. A1 and B1 correspond to the first flow, and A2 and B2 correspond to the second flow. Simultaneously, QPSK is used to modulate the pilot data to obtain modulated pilot data X. p1 The modulated data symbols and pilot symbols are then layer-mapped and superimposed as shown in Formula 8 below:

[0108] Among them, C L Choose the matrix resulting from the matrix dot product for the power allocation matrix and pilot position corresponding to the Lth flow, where L takes the values ​​1 and 2.

[0109] 3. Generate and process received data.

[0110] Obtain the frequency domain signal X obtained by superimposing the transmitter data and pilot signals. mix1Then, the channel process is simulated using the frequency domain channel dataset from step 1 to generate the receiver frequency domain signal Y. mix1 And use the following formula 9 to apply the signal Y mix1 Energy normalization is performed to obtain the processed signal Y. nor1 Y nor1 =Normal max (Y mix1 ) Formula 9

[0111] Next, the data is integrated. The integration method is to integrate the signal Y... nor1 Real part, imaginary part, pilot signal X p1 The matrix C obtained by multiplying the real and imaginary parts of the power allocation matrix and the pilot position selection matrix. L The real and imaginary parts of the matrix are superimposed in the third dimension to obtain the integrated signal Y. in1 .

[0112] 4. LOSS calculation and parameter update

[0113] The integrated signal Y from step 3 is used in1 The integrated output bit value Y, which is the input data of the receiver, is obtained. bit1 According to the bit value Y output by the receiver bit1 and the original bit value X bit1 The binary cross-entropy is calculated using the following formula 10 to obtain the model's loss value. Loss = BCE(X) bit1 ,Y bit1 ) Formula 10

[0114] The parameters of the power allocation matrix generation network and the receiver are updated based on the calculated Loss value, and steps 2 to 4 are repeated iteratively until the model converges. Then, the power allocation matrix and pilot position selection matrix (or the matrix resulting from the dot product of the power allocation matrix and pilot position selection matrix) output by the power allocation matrix generation network are stored using the stream number as the index, and the receiver parameters are also stored using the stream number as the index.

[0115] The model application process is as follows:

[0116] 1. Generate transmitter data

[0117] In this scenario, the communication system uses a two-stream signal, transmitting bit data X from the transmitting end. bit2 After channel coding and 64QAM modulation, the modulated data symbol X is obtained. sym2Then, according to the current stream number, the power allocation matrix A1, A2 and pilot position selection matrix B1, B2 are obtained by looking up the table. Meanwhile, the pilot data is modulated by QPSK to obtain the modulated pilot data X p2 , and the data symbol and the pilot symbol are layer mapped, and the data symbol and the pilot symbol are superimposed on the same time-frequency resource according to the power allocation matrix and the pilot position selection matrix, and the superimposing process is shown in the following formula 11:

[0118] Wherein, C L is the matrix after the power allocation matrix and the pilot position selection matrix corresponding to the Lth stream are multiplied, and the value of L is 1 or 2.

[0119] 2, the superimposed data after the pilot is superimposed is subjected to OFDM symbol forming processing and CP loading to obtain the OFDM signal to be transmitted, the target signaling for indicating the current stream number is sent, and the above OFDM signal is sent after the target signaling is sent.

[0120] 3, the receiver performs related processing such as CP removal and FFT on the received OFDM signal.

[0121] 4, processing the data at the receiving end

[0122] The receiving end frequency domain signal Y mix2 is obtained after step 3 processing, according to the stream number 2 indicated by the target signaling, the corresponding receiver parameters and the corresponding power allocation matrix A1 and A2, and the corresponding pilot position selection matrix B1 and B2 are selected. Then, the signal Y mix2 is subjected to energy normalization processing according to the following formula 12, and the maximum value of the signal energy is normalized to 1 to obtain Y nor2 . Y nor2 = Normal max (Y mix2 ) Formula 12

[0123] Then, the data is integrated, and the integration method is to superimpose the real part and the imaginary part of the signal Y nor2 , the real part and the imaginary part of the pilot signal X p2 , and the real part and the imaginary part of the matrix C L obtained by multiplying the power allocation matrix and the pilot position selection matrix based on the current stream number in the third dimension of the three-dimensional matrix to obtain the integrated signal Y in2 . The integrated signal Y in2 is used as the input data of the receiver, the receiver parameters obtained by looking up the table are used to decode Y in2 to obtain the integrated output bit value Y bit2 .

[0124] Example three: assuming that an OFDM MIMO communication system, the data is four flow, the number of scheduled RB is 12, the number of RE is 144, the number of symbols is 14. The data signal adopts QPSK modulation, the pilot signal adopts BPSK modulation, and the channel data set is obtained by real collection. Specifically, the model training process is as follows:

[0125] 1, channel data set collection and arrangement

[0126] The frequency domain channel data is obtained by real collection to construct the data set, and the channel data is normalized.

[0127] 2, generate the transmitting end data

[0128] In this scenario, the communication system is four flow signal, randomly generate the bit data X bit3 , after QPSK modulation, the modulated data symbol X sym3 is obtained. Then, a group of randomly generated and unlimited length and amplitude vectors are selected as the input data of the power allocation matrix generation network to obtain the power allocation matrix A1, A2, A3, A4, the pilot position selection matrix B1, B2, B3, B4, A1 and B1 correspond to the first flow, A2 and B2 correspond to the second flow, A3 and B3 correspond to the third flow, A4 and B4 correspond to the fourth flow. At the same time, the pilot data is modulated by BPSK to obtain the modulated pilot data X p3 , and the data symbol and the pilot symbol are layer mapped and superimposed, and the superposition process is shown in the following formula 13:

[0129] Wherein, C L is the matrix obtained by multiplying the power allocation matrix and the pilot position selection matrix corresponding to the Lth flow, and the value of L is 1, 2, 3 and 4.

[0130] 3, generate and process the receiving end data

[0131] After obtaining the transmitting end data and the pilot superimposed frequency domain signal X mix3 , the receiving end frequency domain signal Y mix3 is generated by using the frequency domain channel data set in step 1 to simulate the channel process, and the energy normalization processing is carried out on the signal Y mix3 by using the following formula 14 to obtain Y nor3 . Y nor3 = Normal max (Y mix3 ) Formula 14

[0132] Then integrate the data, and the integration method is to integrate the real part and the imaginary part of the signal Y nor3 , the pilot signal X p3The matrix C obtained by multiplying the real and imaginary parts of the power allocation matrix and the pilot position selection matrix. L The real and imaginary parts of the matrix are superimposed in the third dimension to obtain the integrated signal Y. in3 .

[0133] 4. LOSS calculation and parameter update

[0134] The integrated signal Y from step 3 is used in3 The integrated output bit value Y, which is the input data of the receiver, is obtained. bit3 According to the bit value Y output by the receiver bit3 and the original bit value X bit3 The binary cross-entropy is calculated using the following formula 15 to obtain the model's loss value. Loss = BCE(X) bit3 ,Y bit3 ) Formula 15

[0135] The parameters of the power allocation matrix generation network and the receiver are updated based on the calculated Loss value, and steps 2 to 4 are repeated iteratively until the model converges. Then, the power allocation matrix and pilot position selection matrix (or the matrix resulting from the dot product of the power allocation matrix and pilot position selection matrix) output by the power allocation matrix generation network are stored using the stream number as the index, and the receiver parameters are also stored using the stream number as the index.

[0136] The model application process is as follows:

[0137] 1. Generate transmitter data

[0138] In this scenario, the communication system uses a four-stream signal, transmitting bit data X from the transmitting end. bit4 After channel coding and QPSK modulation, the modulated data symbol X is obtained. sym4 Next, based on the current number of streams, the power allocation matrices A1, A2, A3, and A4 for each stream, and the pilot position selection matrices B1, B2, B3, and B4 are obtained from a lookup table. Simultaneously, BPSK is used to modulate the pilot data, resulting in modulated pilot data X. p4 The data symbols and pilot symbols are then layer-mapped, and based on the power allocation matrix and pilot position selection matrix, the data symbols and pilot symbols are superimposed on the same time-frequency resources. The superposition process is shown in Formula 16 below:

[0139] Among them, C L Choose the matrix after matrix dot product for the power allocation matrix and pilot position corresponding to the Lth flow, where L takes the values ​​1, 2, 3, and 4.

[0140] 2. The superimposed aliasing data is subjected to OFDM symbol shaping processing and CP loading to obtain OFDM signals to be transmitted, target signaling indicating the current stream number is sent, and the OFDM signals are sent after the target signaling.

[0141] 3. The receiver performs relevant processing such as CP removal and FFT on the received OFDM signals.

[0142] 4. Process the received data

[0143] The received frequency domain signal Y is obtained after step 3 processing mix4 According to the stream number 4 indicated by the target signaling, the corresponding receiver parameters and the corresponding power allocation matrices A1, A2, A3, A4, and the corresponding pilot position selection matrices B1, B2, B3, B4 are selected. Then the signal Y is subjected to energy normalization processing according to the following formula 17 to obtain Y mix4 . Y nor4 . Y nor4 = Normal max (Y mix4 ) Formula 17

[0144] Then the data is integrated, the integration method is to superimpose the real and imaginary parts of the signal Y nor4 , the real and imaginary parts of the pilot signal X p4 , and the real and imaginary parts of the matrix C L obtained by multiplying the power allocation matrix and the pilot position selection matrix selected based on the current stream number in the third dimension of the three-dimensional matrix to obtain the integrated signal Y in4 . The integrated signal Y in4 is used as the input data of the receiver, the receiver parameters obtained by the above table look-up are used to decode Y in4 to obtain the integrated output bit value Y bit4 .

[0145] FIG. 7 is a structural schematic diagram of a training device of a communication system provided by an embodiment of the present application. As shown in FIG. 7, the device can include a first processing module 701, an acquisition module 702, a second processing module 703, and a training module 704.

[0146] Specifically, the first processing module 701 is configured to process target data by a target stream number corresponding power allocation matrix to obtain a stream corresponding power allocation matrix and a pilot position selection matrix; superimpose first original data and first pilot data of each stream to be sent by the transmitter on time-frequency resources indicated by the pilot position selection matrix corresponding to each stream according to the stream corresponding power allocation matrix to obtain first aliasing data of each stream.

[0147] The acquisition module 702 is configured to acquire the received data of each stream of the receiver, which is obtained by processing the first mixed data of each stream through a wireless channel.

[0148] The second processing module 703 is configured to decode the received data of each stream through the receiver to obtain the first decoded data of each stream.

[0149] The training module 704 is configured to train the power allocation matrix generation network and the parameters of the receiver according to the first original data and the first decoded data of each stream.

[0150] Optionally, the power allocation matrix generation network comprises a first feature extraction layer, an activation layer, and a classification decision layer.

[0151] The first processing module 701 is specifically configured to perform feature extraction on the target data through the first feature extraction layer to obtain feature data, perform activation processing on the feature data through the activation layer to obtain the power allocation matrix corresponding to each stream, and perform processing on the target data through the classification decision layer to obtain the pilot position selection matrix corresponding to each stream, wherein the pilot position selection matrices corresponding to the streams are different.

[0152] Optionally, the receiver comprises a dimension increasing layer, a second feature extraction layer, and a decision layer.

[0153] Optionally, the second feature extraction layer comprises at least one local feature extraction layer and at least one global feature extraction layer, and the at least one local feature extraction layer and the at least one global feature extraction layer are arranged alternately.

[0154] Optionally, the local feature extraction layer adopts a multi-layer convolutional neural network (CNN) architecture, and the global feature extraction layer adopts a Transformer architecture.

[0155] Optionally, the acquisition module 702 is specifically configured to acquire a target channel data set, the target channel data set comprising a real channel data set or a channel data set generated by simulation software; and perform processing on the mixed data of each stream through a channel simulated by the target channel data set to obtain the received data of each stream of the receiver.

[0156] Optionally, the second processing module 703 is specifically configured to fuse the received data of each stream, the first pilot data, the power allocation matrix, and the pilot position selection matrix to obtain the first fused data of each stream; and decode the first fused data of each stream through the receiver to obtain the first decoded data of each stream.

[0157] Optionally, based on the above-mentioned embodiments, the training module 704 is specifically configured to determine a loss value of a preset loss function according to the first original data and the first decoded data of each stream; update parameters of the power allocation matrix generation network and the receiver according to the loss value, and continue to perform the steps of processing target data by the power allocation matrix generation network corresponding to the target number of streams to obtain the power allocation matrix and the pilot position selection matrix corresponding to each stream until the loss value meets a preset requirement.

[0158] In one embodiment, an electronic device is provided, and an internal structure diagram of the electronic device can be as shown in FIG. 8. The electronic device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the electronic device is used to store data generated in the training process of the communication system. The network interface of the electronic device is used to communicate with the terminal outside through the network connection. The computer program is executed by the processor to implement a training method of a communication system.

[0159] Those skilled in the art can understand that the structure shown in FIG. 8 is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied. Specifically, the electronic device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0160] The embodiments of the present application also provide a communication system, as shown in FIG. 1, which can include a power allocation matrix generation network, a transmitter and a receiver.

[0161] Specifically, the power allocation matrix generation network is used to generate the power allocation matrix and the pilot position selection matrix corresponding to each stream.

[0162] The transmitter is used to determine the target power allocation matrix and the target pilot position selection matrix corresponding to each stream under the current number of streams from the first preset mapping relationship; superimpose the second original data and the second pilot data of each stream on the time-frequency resources indicated by the target pilot position selection matrix corresponding to each stream according to the target power allocation matrix corresponding to each stream, to obtain the second superposition data of each stream; and send the target signaling and the second superposition data of each stream, and the target signaling includes the current number of streams.

[0163] The receiver is configured to receive the target signaling and the second mixed data of each stream, determine the target receiver parameter corresponding to the current number of streams from the second preset mapping relationship, and decode the second mixed data of each stream based on the target receiver parameter.

[0164] The first preset mapping relationship includes the power allocation matrix and the pilot position selection matrix corresponding to each stream under different numbers of streams, and the information in the first preset mapping relationship is obtained by inputting the target data into the power allocation matrix generation network corresponding to each number of streams that is pre-trained, and the second preset mapping relationship includes the receiver parameter corresponding to each number of streams, and the power allocation matrix generation network and the receiver parameter corresponding to each number of streams are trained by using the training method of the communication system provided in any one of the above embodiments.

[0165] Optionally, the receiver is further configured to fuse the second mixed data of each stream, the second pilot data, the target power allocation matrix and the target pilot position selection matrix to obtain the second fused data of each stream, and decode the second fused data of each stream based on the target receiver parameter.

[0166] Optionally, the power allocation matrix generation network includes a first feature extraction layer, an activation layer and a classification decision layer.

[0167] Optionally, the receiver includes a dimension increasing layer, a second feature extraction layer and a decision layer.

[0168] Optionally, the second feature extraction layer includes at least one local feature extraction layer and at least one global feature extraction layer, and the at least one local feature extraction layer and the at least one global feature extraction layer are arranged alternately.

[0169] Optionally, the local feature extraction layer adopts a multi-layer convolutional neural network (CNN) architecture, and the global feature extraction layer adopts a Transformer architecture.

[0170] The embodiments of the present application further provide a computer readable storage medium, and the computer readable storage medium stores a computer program. When the computer program is executed by a processor, the training method of the communication system or the communication method provided in any one of the above embodiments is implemented.

[0171] The computer storage medium of the embodiments of the present application can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination thereof. The computer readable storage medium includes, but is not limited to, a non-exhaustive list: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an electrically erasable, programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or apparatus.

[0172] The computer readable signal medium can include a data signal propagating in a baseband or as part of a carrier wave propagating through a transmission medium, and carrying computer readable program code. Such a propagated data signal can take many forms, including but not limited to electro-magnetic, optical, or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport program for use by or in connection with an instruction execution system, apparatus, or device.

[0173] The program code contained on the computer readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire line, optical fiber, radio frequency (RF), or any suitable combination thereof.

[0174] Computer program code for carrying out operations of the present disclosure can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++, Ruby, Go, or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0175] Those skilled in the art will appreciate that the term user terminal encompasses any appropriate type of wireless user equipment, such as a mobile phone, a portable data processing apparatus, a portable web browser, or a vehicle mounted mobile station.

[0176] In general, the various embodiments of the application can be implemented in hardware or special purpose circuits, software, logic or any combination thereof. For example, some aspects can be implemented in hardware, while other aspects can be implemented in

[0177] Embodiments of the application can be implemented by the data processor of a mobile device executing computer program instructions, for example in a processor entity, or by hardware, or by a combination of software and hardware. Computer program instructions can be in assemblies, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or in any combination of one or more programming languages, executed on one or more computing devices.

[0178] The block diagrams of any logical flow of the present application in the drawings can represent program steps or can represent interconnected logic circuits, modules, and functions, or can represent a combination of program steps and logic circuits, modules, and functions. The computer program can be stored on a memory. The memory can be of any type suitable to the local technical environment and can be realized using any suitable data storage technology, such as, but not limited to, read only memory (ROM), random access memory (RAM), optical storage devices, and systems, such as digital versatile disc (DVD) or CD, and the like. The computer readable medium can include non-transitory storage media. The data processor can be of any type suitable to the local technical environment, and can include one or more of general purpose computers, special purpose computers, microprocessors, digital signal processors (DSPs), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA), and processors based on multi-core processor architectures, as examples.

Claims

1. A training method of a communication system, the communication system comprising a power allocation matrix generating network, a transmitter and a receiver, the method comprising: processing target data by the power allocation matrix generating network corresponding to a target number of streams to obtain a power allocation matrix corresponding to each stream of the target number of streams and a pilot position selection matrix corresponding to each stream; superimposing first original data of each stream to be sent by the transmitter and first pilot data on time-frequency resources indicated by the pilot position selection matrix corresponding to each stream according to the power allocation matrix corresponding to each stream to obtain first mixed data of each stream; obtaining reception data of each stream by the receiver, the reception data of each stream being obtained after the first mixed data of each stream passes through a wireless channel; decoding the reception data of each stream by the receiver to obtain first decoding data of each stream; training parameters of the power allocation matrix generating network and the receiver according to the first original data of each stream and the first decoding data of each stream.

2. The method of claim 1, wherein, The power allocation matrix generating network comprises a first feature extraction layer, an activation layer and a classification decision layer. The processing of the target data by the power allocation matrix generating network corresponding to the target number of streams to obtain the power allocation matrix corresponding to each stream of the target number of streams and the pilot position selection matrix corresponding to each stream comprises: extracting features of the target data by the first feature extraction layer to obtain feature data; activating the feature data by the activation layer to obtain the power allocation matrix corresponding to each stream; processing the target data by the classification decision layer to obtain the pilot position selection matrix corresponding to each stream, the pilot position selection matrices corresponding to different streams being different.

3. The method of claim 1, wherein, The receiver comprises a dimension increasing layer, a second feature extraction layer and a decision layer.

4. The method of claim 3, wherein, The second feature extraction layer comprises at least one local feature extraction layer and at least one global feature extraction layer, and the at least one local feature extraction layer and the at least one global feature extraction layer are arranged alternately.

5. The method of claim 4, wherein, The local feature extraction layer adopts a multi-layer convolutional neural network (CNN) architecture, and the global feature extraction layer adopts a Transformer architecture.

6. The method of claim 1, wherein, The obtaining of the reception data of each stream by the receiver comprises: obtaining a target channel data set, the target channel data set comprising a real channel data set or a channel data set generated by simulation software; processing the first mixed data of each stream by a channel simulated by the target channel data set to obtain the reception data of each stream by the receiver.

7. The method of claim 1, wherein, The decoding of the reception data of each stream by the receiver to obtain the first decoding data of each stream comprises: fusing the reception data of each stream, the first pilot data, the power allocation matrix and the pilot position selection matrix to obtain first fusion data of each stream; decoding the first fusion data of each stream by the receiver to obtain the first decoding data of each stream.

8. The method of claim 1, wherein, The training of the parameters of the power allocation matrix generation network and the receiver according to the first original data and the first decoded data of each stream comprises: determining a loss value of a preset loss function according to the first original data and the first decoded data of each stream; updating the parameters of the power allocation matrix generation network and the receiver according to the loss value, and continuing to perform the processing of the target data by the power allocation matrix generation network corresponding to the target stream number to obtain the power allocation matrix and the pilot position selection matrix corresponding to each stream of the target stream number until the loss value meets a preset requirement.

9. A communication method, comprising: a transmitter determining a target power allocation matrix and a target pilot position selection matrix corresponding to each stream under a current stream number from a first preset mapping relationship; the transmitter superimposing second original data and second pilot data of each stream on time-frequency resources indicated by a target pilot position selection matrix corresponding to each stream according to a target power allocation matrix corresponding to each stream to obtain second aliasing data of each stream; the transmitter sending target signaling and the second aliasing data of each stream, the target signaling comprising the current stream number; a receiver receiving the target signaling and the second aliasing data of each stream; the receiver determining target receiver parameters corresponding to the current stream number from a second preset mapping relationship; the receiver decoding the second aliasing data of each stream based on the target receiver parameters; wherein the first preset mapping relationship comprises power allocation matrices and pilot position selection matrices corresponding to each stream under different stream numbers, and information in the first preset mapping relationship is obtained by inputting target data into a pre-trained power allocation matrix generation network corresponding to each stream number, and the second preset mapping relationship comprises receiver parameters corresponding to different stream numbers, and the power allocation matrix generation network and the receiver parameters corresponding to each stream number are trained by the method in any one of claims 1 to 8.

10. The method of claim 9, wherein, The receiver decoding the second aliasing data of each stream based on the target receiver parameters comprises: the receiver fusing the second aliasing data of each stream, the second pilot data, the target power allocation matrix and the target pilot position selection matrix to obtain second fusion data of each stream; the receiver decoding the second fusion data of each stream based on the target receiver parameters.

11. An electronic device comprising: a memory and a processor, the memory storing a computer program, and the processor implementing the method in any one of claims 1 to 8 when executing the computer program.

12. A communication system comprising: a power allocation matrix generation network, a transmitter and a receiver; the transmitter is configured to determine a target power allocation matrix and a target pilot position selection matrix corresponding to each stream under a current stream number from a first preset mapping relationship; the transmitter is configured to superimpose second original data and second pilot data of each stream on time-frequency resources indicated by a target pilot position selection matrix corresponding to each stream according to a target power allocation matrix corresponding to each stream to obtain second aliasing data of each stream; the transmitter is configured to send target signaling and the second aliasing data of each stream, the target signaling comprising the current stream number. transmit target signaling and the second mixed data of each stream, the target signaling including the current number of streams; the receiver is configured to receive the target signaling and the second mixed data of each stream; determining the target receiver parameter corresponding to the current number of streams from a second preset mapping relationship; decoding the second mixed data of each stream based on the target receiver parameter; wherein the first preset mapping relationship includes a power allocation matrix and a pilot position selection matrix corresponding to each stream under different numbers of streams, and the information in the first preset mapping relationship is obtained by inputting target data into a pre-trained power allocation matrix generation network corresponding to each number of streams, and the second preset mapping relationship includes receiver parameters corresponding to different numbers of streams, and the power allocation matrix generation network and the receiver parameter corresponding to each number of streams are trained using the method of any one of claims 1-8.

13. The communication system of claim 12, wherein, the receiver is further configured to fuse the second mixed data, the second pilot data, the target power allocation matrix, and the target pilot position selection matrix of each stream to obtain second fused data of each stream; decoding the second fused data of each stream based on the target receiver parameter.

14. A non-transitory storage medium storing a computer program, the computer program being executed by a processor to implement the method of any one of claims 1-10.

Images