Precoding model training method, precoding method and device

Abstract

The present disclosure discloses a precoding model training method, a precoding method, and an apparatus. The precoding model training method includes: a network device receives channel state information of K terminals, where K is an integer greater than zero, and trains a precoding model DNet model based on the channel state information; wherein the DNet model includes a compression and reconstruction sub-model DiSNet model and a precoding sub-model PreNet model, and the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models. By implementing the present disclosure embodiment, the precoding model DNet model can perform distributed processing on the channel state information of multiple terminals, so that the network device can perform efficient precoding design under limited terminal feedback conditions.

Description

Technical Field

[0001] The present disclosure relates to the field of communication technology, and in particular to a precoding model training method, a precoding method, and a precoding device. Background Art

[0002] m-MIMO (massive-multiple-input multiple-output) technology is regarded as a key technology in future wireless communications and a basic component of 5G wireless communication networks.

[0003] To fully utilize this technology, appropriate precoding design is required. In an FDD (frequency division duplexing) system, the terminal estimates the downlink channel and then feeds back the channel state information to the network equipment. However, due to the large number of m-MIMO antennas and the high dimensionality of the m-MIMO channel matrix, the feedback of channel state information consumes a large amount of feedback communication resources. Furthermore, the high dimensionality of the channel matrix significantly increases the algorithmic complexity when the network equipment calculates the precoding matrix. Therefore, finding a simple precoding matrix design method with limited feedback is an urgent problem that needs to be solved. Summary of the Invention

[0004] The embodiments of the present disclosure provide a precoding model training method, a precoding method, and an apparatus, wherein channel state information of K terminals is received through a network device, wherein K is an integer greater than zero, and a precoding model DNet model is trained based on the channel state information; wherein the DNet model includes a compression and reconstruction sub-model DiSNet model and a precoding sub-model PreNet model, and the DiSNet model includes K groups of paired compression sub-models and reconstruction sub-models. The precoding model DNet model can perform distributed processing on the channel state information of multiple terminals, so that the network device can perform efficient precoding design under the condition of limited feedback from the terminals.

[0005] In a first aspect, an embodiment of the present disclosure provides a precoding model training method, which is executed by a network device, and the method includes: the network device receives channel state information of K terminals; wherein K is an integer greater than zero; according to the channel state information, a precoding model DNet model is trained; wherein the DNet model includes a compression and reconstruction sub-model DiSNet model and a precoding sub-model PreNet model, and the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models.

[0006] In this technical solution, the precoding model DNet model can perform distributed processing on the channel state information of multiple terminals, so that network equipment can perform efficient precoding design under the condition of limited feedback from the terminals.

[0007] In a second aspect, an embodiment of the present disclosure provides a precoding method, which is executed by a network device, and the method includes: obtaining a precoding DNet model according to the precoding model training method as described in some of the above embodiments; wherein the DNet model includes a channel compression and reconstruction sub-model DiSNet model and a precoding design sub-model PreNet model; the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models; the K compression sub-models are sent to the K terminals in a distributed transmission manner; wherein K is an integer greater than 0; receiving compressed channel state information sent by k terminals among the K terminals; wherein k is an integer greater than zero and less than or equal to K; inputting the compressed channel state information into the DNet model to generate a corresponding hybrid precoding matrix.

[0008] In a third aspect, an embodiment of the present disclosure provides another precoding method, which is executed by a terminal and includes: receiving a compression sub-model configured by a network device; inputting channel state information into the compression sub-model, generating compressed channel state information, and sending it to the network device.

[0009] In a fourth aspect, an embodiment of the present disclosure provides a precoding model training device, comprising: an information acquisition unit for receiving channel state information of K terminals; wherein K is an integer greater than zero; a model training unit for training a precoding model DNet model based on the channel state information; wherein the DNet model includes a compression and reconstruction sub-model DiSNet model and a precoding sub-model PreNet model, and the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models.

[0010] In a fifth aspect, an embodiment of the present disclosure provides a communication device that has the function of implementing some or all of the network devices described in the methods described in the first and second aspects above. For example, the functions of the communication device may have the functions of some or all of the embodiments of the present disclosure, or may have the function of implementing any one of the embodiments of the present disclosure alone. The functions may be implemented by hardware, or by hardware executing corresponding software. The hardware or software includes one or more units or modules corresponding to the above functions.

[0011] In one implementation, the communication device may include a transceiver module and a processing module, wherein the processing module is configured to support the communication device in performing the corresponding functions of the above-mentioned method. The transceiver module is used to support communication between the communication device and other devices. The communication device may also include a storage module, which is coupled to the transceiver module and the processing module and stores computer programs and data necessary for the communication device.

[0012] A transceiver module is used to obtain a precoding DNet model according to the precoding model training method as described in some of the above embodiments; wherein the DNet model includes a channel compression and reconstruction sub-model DiSNet model and a precoding design sub-model PreNet model; the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models; the K compression sub-models are sent to the K terminals in a distributed transmission manner; wherein K is an integer greater than 0; compressed channel state information sent by k terminals among the K terminals is received; wherein k is an integer greater than zero and less than or equal to K; a processing module is used to input the compressed channel state information into the DNet model to generate a corresponding hybrid precoding matrix.

[0013] In a sixth aspect, an embodiment of the present disclosure provides another communication device, which has some or all of the functions of the terminal device in the method example described in the third aspect above. For example, the functions of the communication device may have some or all of the functions in the embodiments of the present disclosure, or may have the functions of implementing any one of the embodiments of the present disclosure separately. The functions may be implemented by hardware, or by hardware executing corresponding software. The hardware or software includes one or more units or modules corresponding to the above functions.

[0014] In one implementation, the communication device may include a transceiver module and a processing module, the processing module being configured to support the communication device in performing the corresponding functions of the above-described method. The transceiver module is configured to support communication between the communication device and other devices. The communication device may also include a storage module, coupled to the transceiver module and the processing module, which stores computer programs and data necessary for the communication device.

[0015] In one implementation, the communication device includes: a transceiver module for receiving a compressed sub-model configured by a network device; a processing module for inputting channel state information into the compressed sub-model, generating compressed channel state information, and sending it to the network device.

[0016] In a seventh aspect, an embodiment of the present disclosure provides a communication device, which includes a processor. When the processor calls a computer program in a memory, the method described in the first and second aspects above is executed.

[0017] In an eighth aspect, an embodiment of the present disclosure provides a communication device, which includes a processor. When the processor calls a computer program in a memory, it executes the method described in the third aspect above.

[0018] In the ninth aspect, an embodiment of the present disclosure provides a communication device, which includes a processor and a memory, in which a computer program is stored; the processor executes the computer program stored in the memory so that the communication device executes the methods described in the first and second aspects above.

[0019] In the tenth aspect, an embodiment of the present disclosure provides a communication device, which includes a processor and a memory, in which a computer program is stored; the processor executes the computer program stored in the memory so that the communication device executes the method described in the third aspect above.

[0020] In the eleventh aspect, an embodiment of the present disclosure provides a communication device, which includes a processor and an interface circuit, wherein the interface circuit is used to receive code instructions and transmit them to the processor, and the processor is used to run the code instructions to enable the device to execute the methods described in the first and second aspects above.

[0021] In the twelfth aspect, an embodiment of the present disclosure provides a communication device, which includes a processor and an interface circuit, wherein the interface circuit is used to receive code instructions and transmit them to the processor, and the processor is used to run the code instructions to enable the device to execute the method described in the third aspect above.

[0022] In aspect 13, an embodiment of the present disclosure provides a communication system, which includes the communication device described in aspect 5 and the communication device described in aspect 6, or the system includes the communication device described in aspect 7 and the communication device described in aspect 8, or the system includes the communication device described in aspect 9 and the communication device described in aspect 10, or the system includes the communication device described in aspect 11 and the communication device described in aspect 12.

[0023] In a fourteenth aspect, an embodiment of the present invention provides a computer-readable storage medium for storing instructions for the above-mentioned terminal, and when the instructions are executed, the terminal executes the methods described in the first and second aspects above.

[0024] In a fifteenth aspect, an embodiment of the present invention provides a readable storage medium for storing instructions used by the above-mentioned network device, and when the instructions are executed, the network device executes the method described in the third aspect.

[0025] In a sixteenth aspect, the present disclosure further provides a computer program product comprising a computer program, which, when executed on a computer, enables the computer to execute the methods described in the first and second aspects above.

[0026] In a seventeenth aspect, the present disclosure further provides a computer program product comprising a computer program, which, when executed on a computer, enables the computer to execute the method described in the third aspect above.

[0027] In an eighteenth aspect, the present disclosure provides a chip system comprising at least one processor and an interface for supporting a terminal in implementing the functions described in the first and second aspects, such as determining or processing at least one of the data and information described in the aforementioned methods. In one possible design, the chip system further comprises a memory for storing computer programs and data necessary for the terminal. The chip system may consist of a chip alone or may include a chip and other discrete components.

[0028] In a nineteenth aspect, the present disclosure provides a chip system comprising at least one processor and an interface for supporting a network device in implementing the functions described in the third aspect, such as determining or processing at least one of the data and information described in the aforementioned method. In one possible design, the chip system further comprises a memory for storing computer programs and data necessary for the network device. The chip system may consist of a single chip or may include a chip and other discrete components.

[0029] In a twentieth aspect, the present disclosure provides a computer program, which, when executed on a computer, enables the computer to execute the methods described in the first and second aspects above.

[0030] In a twenty-first aspect, the present disclosure provides a computer program which, when executed on a computer, enables the computer to execute the method described in the third aspect above. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the background technology, the drawings required for use in the embodiments of the present disclosure or the background technology will be described below.

[0032] Figure 1 is an architecture diagram of a communication system provided by an embodiment of the present disclosure;

[0033] Figure 2 is a flowchart of a precoding model training method provided by an embodiment of the present disclosure;

[0034] Figure 3 This is a flowchart of sub-step S2 in a precoding model training method provided by an embodiment of the present disclosure;

[0035] Figure 4 This is a structural diagram of a DiSNet model pre-training in a pre-coding model training method provided by an embodiment of the present disclosure;

[0036] Figure 5 is a structural diagram of a DiSNet model in a precoding model training method provided by an embodiment of the present disclosure;

[0037] Figure 6 This is a structural diagram of a PreNet model pre-training in a pre-coding model training method provided by an embodiment of the present disclosure;

[0038] Figure 7 This is a structural diagram of DNet model joint training in a precoding model training method provided by an embodiment of the present disclosure;

[0039] Figure 8 is a flowchart of a precoding method provided by an embodiment of the present disclosure;

[0040] Figure 9 is a flowchart of another precoding method provided by an embodiment of the present disclosure;

[0041] Figure 10 is a structural diagram of a precoding model training device provided by an embodiment of the present disclosure;

[0042] Figure 11 is a structural diagram of a precoding device provided by an embodiment of the present disclosure;

[0043] Figure 12 is a structural diagram of a communication device provided by an embodiment of the present disclosure;

[0044] Figure 13 It is a schematic structural diagram of a chip provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0045] In order to better understand the method for determining the side link duration disclosed in an embodiment of the present disclosure, the communication system to which the embodiment of the present disclosure is applicable is first described below.

[0046] See Figure 1 , Figure 1 This is a schematic diagram of the architecture of a communication system provided by an embodiment of the present disclosure. The communication system may include but is not limited to a network device and a terminal. Figure 1 The number and form of devices shown are for example only and do not constitute a limitation on the embodiments of the present disclosure. In actual applications, two or more network devices and two or more terminals may be included. Figure 1 The communication system shown includes a network device 10 and a terminal 20 as an example.

[0047] It should be noted that the technical solutions of the embodiments of the present disclosure can be applied to various communication systems, such as long-term evolution (LTE) systems, fifth-generation (5G) mobile communication systems, 5G new radio (NR) systems, or other future new mobile communication systems.

[0048] The network device 10 may be a base station, which may be used to communicate with one or more terminals, or may be used to communicate with one or more base stations having partial terminal functions (such as communication between a macro base station and a micro base station). The base station may be a base transceiver station (BTS) in a time division synchronous code division multiple access (TD-SCDMA) system, or an evolutionary Node B (eNB) in an LTE system, as well as a next generation NodeB (gNB) in a 5G system or NR system, a base station in other future mobile communication systems, or an access node in a wireless fidelity (WiFi) system. In addition, the network device provided in the embodiment of the present disclosure may be composed of a central unit (CU) and a distributed unit (DU), wherein the CU may also be referred to as a control unit. The CU-DU structure may be used to split the protocol layer of a network device, such as a base station, and the functions of some protocol layers are centrally controlled by the CU, while the functions of the remaining part or all of the protocol layers are distributed in the DU, and the DU is centrally controlled by the CU. Optionally, considering that the network device 10 may face greater computing pressure, a server or a server cluster may be deployed therefor to independently provide computing power for the network device 10. In this case, the server or server cluster may be considered as part of the network device 10. The embodiments of the present disclosure do not limit the specific technology and specific device form used by the network device.

[0049] The terminals 20 can be distributed throughout the wireless communication system 1 and can be stationary or mobile, and their number is usually multiple. The terminals 20 may include handheld devices with wireless communication functions (e.g., mobile phones, tablet computers, PDAs, etc.), vehicle-mounted devices (e.g., cars, bicycles, electric vehicles, airplanes, ships, etc.), wearable devices (e.g., smart watches (such as iWatch, etc.), smart bracelets, pedometers, etc.), smart home devices (e.g., refrigerators, televisions, air conditioners, electric meters, etc.), intelligent robots, workshop equipment, other processing equipment that can be connected to wireless modems, and various forms of user equipment (UE), mobile stations (MS), terminals, terminal equipment, etc. The embodiments of the present disclosure do not limit the specific technology and specific device form adopted by the terminal.

[0050] In the communication system 1 , a transmitting end (eg, a network device) is deployed with multiple transmitting antennas, and a receiving end (eg, a terminal) is deployed with multiple receiving antennas, forming a Massive MIMO system.

[0051] It can be understood that the communication system described in the embodiment of the present disclosure is for the purpose of more clearly illustrating the technical solution of the embodiment of the present disclosure, and does not constitute a limitation on the technical solution provided by the embodiment of the present disclosure. Ordinary technicians in this field can know that with the evolution of the system architecture and the emergence of new business scenarios, the technical solution provided by the embodiment of the present disclosure is also applicable to similar technical problems.

[0052] In related technologies, in order to find a low-complexity precoding design algorithm under limited feedback conditions, an m-MIMO precoding design method based on codebook search and a precoding design method based on machine learning are proposed.

[0053] Among them, the design method based on codebook search uses a given analog precoding codebook in hybrid precoding and uses exhaustive search methods to obtain the optimal analog precoding, thereby obtaining equivalent low-dimensional channel and analog precoding codebook information at the baseband. However, codebook compression methods rely on exhaustive search, resulting in high algorithmic complexity. Precoding design methods based on machine learning treat the channel state information between network devices and terminals as image information or language information. They use CNN (convolutional neural network) or RNN (recurrent neural network) networks to find the mapping relationship between channel state information and precoding matrices to generate the corresponding precoding matrices, reducing the complexity of precoding calculations. However, most existing neural network algorithms are centralized networks and fail to reduce the feedback bandwidth consumption between terminals and network devices. At the same time, the precoding performance of existing neural network algorithms also falls short of the theoretical optimal performance.

[0054] Based on this, the embodiments of the present disclosure propose a precoding model training method, a precoding method and a device to solve the problem of high complexity of existing precoding algorithms under limited feedback conditions, while improving the spectral efficiency of precoding design using neural networks.

[0055] The following is a detailed introduction to the precoding model training method, precoding method and device provided by the present disclosure in conjunction with the accompanying drawings.

[0056] See Figure 2 , Figure 2 This is a flowchart of a precoding model training method provided by an embodiment of the present disclosure.

[0057] like Figure 2 As shown, the method is performed by a network device, and the method may include but is not limited to the following steps:

[0058] S1: The network device receives channel state information of K terminals, where K is an integer greater than zero.

[0059] The channel state information may be, for example, downlink channel estimation information. Multiple terminals obtain the downlink channel estimation information through a typical channel estimation method, and then send the downlink channel estimation information to the network device through an uplink channel.

[0060] S2: According to the channel state information, the precoding model DNet model is trained; wherein the DNet model includes a compression and reconstruction sub-model DiSNet model and a precoding sub-model PreNet model, and the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models.

[0061] In the disclosed embodiment, the DNet model includes a compression and reconstruction sub-model, a DiSNet model, and a precoding sub-model, a PreNet model. The DiSNet model performs distributed channel compression and reconstruction, and the PreNet model implements efficient precoding design. The DiSNet model includes K pairs of compression sub-models and reconstruction sub-models that perform distributed channel compression and reconstruction on the channel state information of K terminals. By compressing and reconstructing the channel state information received from each terminal, followed by channel splicing processing, distributed processing of the channel state information of multiple terminals is achieved.

[0062] Through the real-time embodiment of the present disclosure, the network device receives channel state information of K terminals, where K is an integer greater than zero, and trains the precoding model DNet model based on the channel state information; wherein the DNet model includes a compression and reconstruction sub-model DiSNet model and a precoding sub-model PreNet model, and the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models. The precoding model DNet model can perform distributed processing on the channel state information of multiple terminals, so that the network device can perform efficient precoding design under the condition of limited feedback from the terminals.

[0063] In some embodiments, as Figure 3 As shown, S2: Based on the channel state information, the DNet model is trained, including:

[0064] S21: Process the channel state information to obtain a training sample set.

[0065] S22: Pre-train the DiSNet model and PreNet model based on the training sample set.

[0066] S23: Based on the training sample set, the pre-trained DiSNet model and PreNet model are jointly trained.

[0067] In the disclosed embodiment, for the DiSNet model and PreNet model in the DNet model, first, the DiSNet model and the PreNet model are pre-trained according to the training sample set, and then the DiSNet model and the PreNet model are spliced ​​into a complete DNet model for joint training, which can improve the precoding performance.

[0068] In some embodiments, S21: Processing the channel state information to obtain a training sample set includes: obtaining channel state information {h1, ..., h K}, summarized as matrix samples Wherein, K is an integer greater than 1; Represents the conjugate transpose of the channel state information of the K-th terminal; calculates the matrix samples to obtain the initial analog precoding matrix F and the initial digital precoding matrix W; obtains the training sample set {(H, F, W)} based on the matrix samples, the initial analog precoding matrix F, and the initial digital precoding matrix W.

[0069] In an exemplary embodiment, in a downlink of a MIMO-OFDM (orthogonal frequency division multiplexing) system, N 1 / 2 wavelength arrays are configured at half wavelength intervals in a ULA (uniform linear array) manner on the network device side. t = 64 antennas (8×8 antenna array), with a single antenna configured at the terminal, using N c = 1024 subcarriers, 23040 CSI (channel state information) matrix samples are generated in a 2.4GHz outdoor picocell scenario and divided into a training sample set containing 20736 samples (accounting for 90%) and a test sample set containing 2304 samples (accounting for 10%). The training sample set H in the CSI matrix samples is decomposed into virtual and real, that is, H = H re +j*H im , where j 2 =-1, Based on the zero forcing ZF (zero forcing) precoding algorithm, the initial analog precoding matrix F and the initial digital precoding matrix W are calculated to obtain the training data set {(H, F, W)} required for neural network training.

[0070] It should be noted that the calculation of the initial analog precoding matrix F and the initial digital precoding matrix W based on the ZF-based precoding algorithm according to the training sample set proposed in the above exemplary embodiment is only for illustration, and any other algorithm for calculating the initial analog precoding matrix F and the initial digital precoding matrix W can also be used for calculation. This is only for illustration and not a specific limitation to the present disclosure.

[0071] In some embodiments, as Figure 4 As shown, S23: pre-training the DiSNet model according to the training samples, including:

[0072] K compression sub-models convert {h1,…,h K} is compressed into low-dimensional channel feature information {q1,…,q K}; K reconstruction sub-models transform the low-dimensional channel feature information {q1,…,q K}Reconstructed into imperfect channel state information Imperfect channel state information Splicing to generate an imperfect matrix Self-supervised learning and adaptive moment estimation Adam optimization algorithm are used to pre-train the compression sub-model and reconstruction sub-model.

[0073] For the DisNet model, pre-training is performed using self-supervised learning and the Adam (adaptive moment estimation) optimization algorithm. The loss function expression satisfies the following relationship:

[0074]

[0075] In some embodiments, as Figure 5 As shown in Figure 1, the K pairs of compression sub-models and reconstruction sub-models have the same structure and are independent of each other; the compression sub-model includes convolutional layers and fully connected layers to extract and compress channel state information; the reconstruction sub-model includes fully connected layers and deep residual networks to reconstruct the compressed channel state information.

[0076] In an exemplary embodiment, each independent compression sub-model includes a convolutional layer and a fully connected layer, and each independent reconstruction sub-model includes a fully connected layer and a residual network. The input of the compression sub-model is the channel state information h corresponding to the K-th terminal. k , the output is the low-dimensional channel feature code q k The input of the reconstruction sub-model is the corresponding low-dimensional channel feature code q k , the output is the reconstructed imperfect channel state information

[0077] When inputting, the imaginary and real parts of the channel state information are separated, that is, At the same time and From 1×N t The vector becomes Matrix, input the corresponding compression sub-model-reconstruction sub-model in the DisNet model.

[0078] The expression of the convolutional layer satisfies the following relationship:

[0079]

[0080] Among them, y d,i,j is the (d,i,j)th element in the convolution output y, W d,c,h,w is the (d,c,h,w)th element in the convolution kernel weight matrix W, b d is the dth element in the convolution kernel bias b, The convolution layer uses Tanh as the activation function, which satisfies the following relationship:

[0081]

[0082] All convolutional layers use the same zero-padding strategy, with a convolution stride of (1, 1) and a convolution kernel of d i ×c i ×3×3, satisfying d i-1 =c i The final output size of the convolutional layer is Where n is the number of convolutional layers.

[0083] The expression of the fully connected layer satisfies the following relationship:

[0084] y i =∑ j W i,j x i +b i (4)

[0085] Among them, y i is the i-th element in the fully connected output, W i,j is the (i,j)th element in the fully connected weight matrix, b i is the i-th element in the fully connected bias, x i is the i-th element in the fully connected input. The fully connected layer uses the ReLU function as the activation function, and its expression satisfies the following relationship:

[0086]

[0087] The expression of the residual network satisfies the following relationship:

[0088] y=f n (W n f n-1 (W n-1 f n-2 (…f1(W1x+b1)…)+b n-1 )+b n +x) (6)

[0089] Among them, y is the residual network output, x is the residual network input, and W i with b i is the weight and bias of the middle layer of the residual network, f i (·) is the activation function of the middle layer, specifically:

[0090]

[0091] Define the output size of the compressed sub-model as M and the input size as N t , then the compression ratio is γ=M / N t When M <N t When , γ<1, channel limited feedback is achieved.

[0092] In the disclosed embodiment, the reconstruction sub-model adopts a deep residual network, which can avoid gradient vanishing and improve the accuracy of channel reconstruction.

[0093] In some embodiments, as Figure 6 As shown, the PreNet model includes: an analog precoding sub-model RFNet model and a digital precoding sub-model BBNet model; pre-training the PreNet model includes: converting the imperfect matrix Input to RFNet model to generate analog precoding BBNet model based on imperfect matrix Generate digital precode Self-supervised learning and Adam optimization algorithm are used to pre-train the RFNet model and BBNet model.

[0094] For the PreNet model, supervised learning and Adam optimization algorithm are used for pre-training. The expression of the RFNet loss function satisfies the following relationship:

[0095]

[0096] In some embodiments, see Figure 6 , BBNet model, including: a first initial digital sub-model SNet model and a second initial digital sub-model MNet model.

[0097] Among them, the BBNet model is based on the imperfect matrix Generate digital precode include:

[0098] According to the imperfect matrix Input to RFNet model to generate analog precoding Get the diagonally dominant matrix And the principal element matrix diag(H eq ), respectively, the diagonally dominant matrix and the principal element matrix diag(H eq ) is input to the SNet model and the MNet model, and then combined with the simulated precoding Perform normalization processing to generate digital precoding

[0099] In the embodiment of the present disclosure, the RFNet model includes a convolutional layer and a fully connected layer, with ReLU (rectified linear unit) as the activation function. The processing formulas of the convolutional layer and the fully connected layer are the same as those in the above embodiment. Output simulation precoding The corresponding phase shifter angle Θ:

[0100]

[0101] Among them, N t is the number of base station antennas. The BBNet model includes two sub-models, the SNet model and the Mnet model. The diagonal dominant matrix is ​​input respectively. and the principal element matrix diag(H eq ), merge output After the normalization module, the output digital precoding

[0102] In the embodiment of the present disclosure, the BBNet model is a special diagonally dominant matrix in addition to the traditional fully connected network, in response to the particularity of the input data. An additional MNet model is constructed to improve the diagonal dominant matrix element H eq The weights in the model improve the precoding capability.

[0103] Among them, the normalized expression satisfies the following relationship:

[0104]

[0105] In some embodiments, as Figure 7 As shown, according to the training sample set, the pre-trained DiSNet model and the PreNet model are jointly trained, including: using the Adam optimization algorithm and end-to-end learning to jointly train the pre-trained DiSNet model and the PreNet model.

[0106] The DisNet model and the PreNet model are combined into a DNet model for joint training. The DNet model uses the Adam optimization algorithm and an end-to-end learning approach to jointly train the DisNet model and the PreNet model. The loss function expression satisfies the following relationship:

[0107]

[0108]

[0109] In some embodiments, the test sample set HS is decomposed into virtual and real, that is, HS = HS re +j*HS im , where j 2 =-1, Split into and And input it into the trained DNet model to obtain the corresponding output analog precoding matrix FS and digital precoding matrix WS.

[0110] Analyzing the simulated precoding matrix FS and the digital precoding matrix WS obtained by processing the test sample set HS through the DNet model can assist in judging the model training results.

[0111] See Figure 8 , which is a flowchart of a precoding method provided in an embodiment of the present disclosure.

[0112] like Figure 8 As shown, the precoding method provided by the embodiment of the present disclosure is performed by a network device, and the method includes but is not limited to the following steps:

[0113] S10: According to the precoding model training method in the above embodiment, a precoding DNet model is obtained; wherein the DNet model includes a channel compression and reconstruction sub-model DiSNet model and a precoding design sub-model PreNet model; the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models, and the K compression sub-models are sent to K terminals in a distributed transmission manner; wherein K is an integer greater than 0.

[0114] It can be understood that in the embodiments of the present disclosure, the terminal obtains the DNet model according to the precoding model training method in some of the above embodiments. The precoding model training method can be found in the description of the above embodiments and will not be repeated here.

[0115] In the disclosed embodiment, the network device sends K compressed sub-models in the K pairs of compression sub-model-reconstruction sub-model of the DiSNet model to K terminals using a distributed transmission method; wherein K is an integer greater than zero.

[0116] It should be noted that in the precoding model training method, the network device obtains the K pairs of compression sub-models and reconstruction sub-models of the DiSNet model based on the channel state information of the K terminals received, which can be distributedly configured on the K terminals. Therefore, the DNet model can be distributedly deployed on the K terminals to build a distributed network architecture. Moreover, the DNet model is deployed on the terminals and network devices at the same time, which can realize limited feedback of the channel on the terminal and efficient precoding design on the network device.

[0117] S20: Receive compressed channel state information sent by k terminals among the K terminals; where k is an integer greater than zero and less than or equal to K.

[0118] In an embodiment of the present disclosure, a network device includes K reconstruction sub-models, and the K compression sub-models are distributedly configured on K terminals, so that the network device can receive compressed channel state information from the K terminals. In an embodiment of the present disclosure, the network device can also receive compressed channel state information sent by k of the K terminals, where k is an integer greater than zero and less than or equal to K.

[0119] S30: Input the compressed channel state information into the DNet model to generate a corresponding hybrid precoding matrix.

[0120] Among them, when k is greater than zero and less than K, the data in the K reconstruction sub-models of the network device and the reconstruction sub-models that do not correspond to the k terminals are set to zero, so that the K reconstruction sub-models of the DiSNet model and the PreNet model in the network device can process the compressed channel information sent by the k terminals and generate the corresponding hybrid precoding matrix.

[0121] By implementing the embodiments of the present disclosure, the network device obtains a precoding DNet model according to the precoding model training method as in the above embodiment; wherein the DNet model includes a channel compression and reconstruction sub-model DiSNet model and a precoding design sub-model PreNet model; the DiSNet model includes K groups of paired compression sub-models-reconstruction sub-models, and the K compression sub-models are sent to K terminals in a distributed transmission manner; wherein K is an integer greater than 0; compressed channel state information sent by k terminals among the K terminals is received; wherein k is an integer greater than zero and less than or equal to K; the compressed channel state information is input into the DNet model to generate a corresponding hybrid precoding matrix. Thus, the DNet model is simultaneously deployed on the network device and distributedly deployed on the terminal, which can achieve limited feedback of the channel on the terminal and efficient precoding design on the network device.

[0122] In some embodiments, the compressed channel state information is input into the DNet model to generate a corresponding hybrid precoding matrix, including but not limited to the following steps: the reconstruction sub-model reconstructs the compressed channel state information to generate imperfect channel state information; and then splices to generate an imperfect matrix Imperfect Matrix Input to the PreNet model to generate a simulated precoding matrix and the digital precoding matrix

[0123] In some embodiments, the PreNet model includes: an analog precoding sub-model RFNet model and a digital precoding sub-model BBNet model; Input to the PreNet model to generate a simulated precoding matrix and the digital precoding matrix Includes: Imperfect Matrix Input to the RFNet model to generate the simulated precoding matrix Imperfect Matrix Input to the BBNet model to generate the digital precoding matrix

[0124] In some embodiments, the BBNet model includes: a first digital sub-model SNet model and a second digital sub-model MNet model; Input to the BBNet model to generate the digital precoding matrix Includes: Based on the imperfect matrix Get the diagonally dominant matrix And the principal element matrix diag(H eq ); respectively, the diagonally dominant matrix and the principal element matrix diag(H eq ) is input to the SNet model and the MNet model, and then merged with the imperfect matrix Perform normalization processing to generate a digital precoding matrix

[0125] In some embodiments, the DiSNet model includes multiple pairs of compression sub-models and reconstruction sub-models with the same structure and independent of each other; the compression sub-model includes a convolutional layer and a fully connected layer to extract and compress channel state information; the reconstruction sub-model includes a fully connected layer and a deep residual network to reconstruct the compressed channel state information.

[0126] In some embodiments, the RFNet model includes a convolutional layer and a fully connected layer, with a linear rectifier function ReLU function as the activation function, according to the imperfect matrix Generate simulated precoding matrix The corresponding shift angle is then used to calculate the simulated precoding matrix

[0127] The precoding method in the embodiment of the present disclosure has been described in detail in the above-mentioned precoding model training method. Please refer to the description in the above-mentioned precoding model training method and will not be repeated here.

[0128] It should be noted that the precoding method in the embodiment of the present disclosure directly uses the trained DNet model, while the use of self-supervised learning and Adam optimization algorithm to pre-train the RFNet model and BBNet model for DNet model training, as well as the use of Adam optimization algorithm and end-to-end learning to jointly train the pre-trained DiSNet model and PreNet model, are no longer applicable. The precoding method in the embodiment of the present disclosure does not include this part.

[0129] See Figure 9 , which is a flowchart of another precoding method provided in an embodiment of the present disclosure.

[0130] like Figure 9 As shown, the precoding method provided in the embodiment of the present disclosure is performed by a terminal, and the method includes but is not limited to the following steps:

[0131] S100: Receive a compressed sub-model of a network device configuration.

[0132] S200: Input the channel state information into the compression sub-model, generate compressed channel state information, and send it to the network device.

[0133] It is understandable that only when the terminal receives the compression sub-model configured by the network device can the terminal use the compression sub-model to compress the channel state information, thereby generating compressed channel state information, and further sending it to the network device.

[0134] By implementing the disclosed embodiments, a terminal device receives a compression sub-model configured by a network device; it inputs channel state information into the compression sub-model, generates compressed channel state information, and sends it to the network device. This allows the DNet model to be deployed simultaneously on the network device and distributed on the terminal, enabling limited feedback of the channel on the terminal and efficient precoding design on the network device.

[0135] In the embodiments provided above, the methods provided in the embodiments of the present disclosure are described from the perspectives of network devices and terminals, respectively. To implement the various functions of the methods provided in the embodiments of the present disclosure, the network devices and terminals may include hardware structures and software modules, and the aforementioned functions may be implemented in the form of hardware structures, software modules, or hardware structures and software modules. Certain of the aforementioned functions may be implemented in the form of hardware structures, software modules, or hardware structures and software modules.

[0136] See Figure 10 , which is a structural diagram of a precoding model training device provided in an embodiment of the present disclosure.

[0137] like Figure 10 As shown, the precoding model training device 100 provided by the embodiment of the present disclosure includes: an information acquisition unit 101 and a model training unit 102.

[0138] The information acquisition unit 101 is configured to receive channel state information of K terminals, where K is an integer greater than zero.

[0139] The model training unit 102 is used to train the precoding model DNet model according to the channel state information; wherein the DNet model includes a compression and reconstruction sub-model DiSNet model and a precoding sub-model PreNet model, and the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models.

[0140] Regarding the precoding model training device 100 in the above-described embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method and will not be elaborated on here. The precoding model training device 100 provided in the above-described embodiment of the present disclosure achieves the same or similar beneficial effects as the precoding model training method provided in some of the above embodiments and will not be described in detail here.

[0141] See Figure 11 , which is a structural diagram of a precoding device provided in an embodiment of the present disclosure.

[0142] Figure 11 The precoding device 70 shown may include a transceiver module 701 and a processing module 702. The transceiver module 701 may include a sending module and / or a receiving module, the sending module is used to implement a sending function, and the receiving module is used to implement a receiving function. The transceiver module 701 can implement the sending function and / or the receiving function.

[0143] The precoding device 70 may be a terminal, a device in a terminal, or a device compatible with the terminal. Alternatively, the precoding device 70 may be a network device, a device in a network device, or a device compatible with the network device.

[0144] The precoding device 70 is a network device: a transceiver module 701 is used to obtain a precoding DNet model according to the precoding model training method as described in some of the above embodiments; wherein the DNet model includes a channel compression and reconstruction sub-model DiSNet model and a precoding design sub-model PreNet model; the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models; the K compression sub-models are sent to K terminals using a distributed transmission method; wherein K is an integer greater than 0; compressed channel state information sent by k terminals among the K terminals is received; wherein k is an integer greater than zero and less than or equal to K; a processing module 702 is used to input the compressed channel state information into the DNet model to generate a corresponding hybrid precoding matrix.

[0145] The precoding device 70 is a terminal: a transceiver module 701 is used to receive the compression sub-model configured by the network device; a processing module 702 is used to input the channel state information into the compression sub-model, generate compressed channel state information, and send it to the network device.

[0146] Regarding the precoding apparatus 70 in the above-described embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method and will not be further elaborated here. The precoding apparatus 70 provided in the above-described embodiment of the present disclosure achieves the same or similar beneficial effects as the precoding method provided in some of the above-described embodiments and will not be further described here.

[0147] See Figure 12 , Figure 12 It is a structural diagram of another communication device 1000 provided in an embodiment of the present disclosure.

[0148] The communication device 1000 can be a network device, a terminal, a chip, a chip system, or a processor that supports the network device to implement the above method, or a chip, a chip system, or a processor that supports the terminal to implement the above method. The device can be used to implement the method described in the above method embodiment. For details, please refer to the description of the above method embodiment.

[0149] The communication device 1000 may include one or more processors 1001. The processor 1001 may be a general-purpose processor or a dedicated processor. For example, it may be a baseband processor or a central processing unit. The baseband processor may be used to process communication protocols and communication data, and the central processing unit may be used to control the communication device (e.g., a base station, a baseband chip, a terminal, a terminal device chip, a DU or CU, etc.), execute computer programs, and process computer program data.

[0150] Optionally, the communication device 1000 may further include one or more memories 1002, on which a computer program 1004 may be stored. The memory 1002 executes the computer program 1004 to enable the communication device 1000 to perform the method described in the above method embodiment. Optionally, the memory 1002 may also store data. The communication device 1000 and the memory 1002 may be provided separately or integrated together.

[0151] Optionally, the communication device 1000 may further include a transceiver 1005 and an antenna 1006. The transceiver 1005 may be referred to as a transceiver unit, a transceiver, or a transceiver circuit, etc., and is configured to implement transceiver functions. The transceiver 1005 may include a receiver and a transmitter. The receiver may be referred to as a receiver or a receiving circuit, etc., and is configured to implement a receiving function; the transmitter may be referred to as a transmitter or a transmitting circuit, etc., and is configured to implement a transmitting function.

[0152] Optionally, the communication device 1000 may further include one or more interface circuits 1007. The interface circuit 1007 is configured to receive code instructions and transmit the code instructions to the processor 1001. The processor 1001 executes the code instructions to enable the communication device 1000 to perform the method described in the above method embodiment.

[0153] The communication device 1000 is a network device: the transceiver 1005 is used to perform Figure 8 S10 and S20 in the process. The processor 1001 is used to execute Figure 8 S30.

[0154] The communication device 1000 is a terminal: the transceiver 1005 is used to perform Figure 9 S100 in the process. Processor 1001 is used to execute Figure 9 The S200 in the.

[0155] In one implementation, processor 1001 may include a transceiver for implementing receiving and transmitting functions. For example, the transceiver may be a transceiver circuit, an interface, or an interface circuit. The transceiver circuit, interface, or interface circuit for implementing the receiving and transmitting functions may be separate or integrated. The transceiver circuit, interface, or interface circuit may be used for reading and writing code / data, or may be used for transmitting or delivering signals.

[0156] In one implementation, processor 1001 may store a computer program 1003. Computer program 1003, when executed on processor 1001, enables communication device 1000 to perform the method described in the above method embodiment. Computer program 1003 may be embedded in processor 1001, in which case processor 1001 may be implemented by hardware.

[0157] In one implementation, the communication device 1000 may include a circuit that can implement the functions of sending, receiving, or communicating in the aforementioned method embodiments. The processor and transceiver described in the present disclosure can be implemented on an integrated circuit (IC), an analog IC, a radio frequency integrated circuit RFIC, a mixed signal IC, an application specific integrated circuit (ASIC), a printed circuit board (PCB), an electronic device, etc. The processor and transceiver can also be manufactured using various IC process technologies, such as complementary metal oxide semiconductor (CMOS), N-type metal oxide semiconductor (nMetal-oxide-semiconductor, NMOS), P-type metal oxide semiconductor (positive channel metal oxide semiconductor, PMOS), bipolar junction transistor (bipolar junction transistor, BJT), bipolar CMOS (BiCMOS), silicon germanium (SiGe), gallium arsenide (GaAs), etc.

[0158] The communication device described in the above embodiments may be a terminal, but the scope of the communication device described in this disclosure is not limited thereto, and the structure of the communication device may not be limited thereto. Figure 12 The communication device may be an independent device or may be part of a larger device. For example, the communication device may be:

[0159] (1) An independent integrated circuit (IC), or chip, or chip system or subsystem;

[0160] (2) a collection of one or more ICs, optionally including a storage component for storing data and computer programs;

[0161] (3) ASIC, such as modem;

[0162] (4) Modules that can be embedded in other devices;

[0163] (5) Receivers, terminal devices, intelligent terminal devices, cellular phones, wireless devices, handheld devices, mobile units, vehicle-mounted devices, network devices, cloud devices, artificial intelligence devices, etc.;

[0164] (6)Others, etc.

[0165] For cases where the communication device may be a chip or a chip system, see Figure 6 , is a structural diagram of a chip provided in an embodiment of the present disclosure.

[0166] like Figure 13 As shown, the chip 1100 includes a processor 1101 and an interface 1103. There may be one or more processors 1101, and there may be more than one interface 1103.

[0167] For the case where the chip is used to implement the functions of the terminal in the embodiments of the present disclosure:

[0168] The interface 1103 is used to receive code instructions and transmit them to the processor.

[0169] The processor 1101 is configured to execute code instructions to perform the precoding method as described in some of the above embodiments.

[0170] For the case where the chip is used to implement the functions of the network device in the embodiments of the present disclosure:

[0171] The interface 1103 is used to receive code instructions and transmit them to the processor.

[0172] The processor 1101 is configured to execute code instructions to perform the precoding method as described in some of the above embodiments.

[0173] Optionally, the chip 1100 further includes a memory 1102 , which is used to store necessary computer programs and data.

[0174] Those skilled in the art will also appreciate that the various illustrative logical blocks and steps listed in the embodiments of the present disclosure may be implemented by electronic hardware, computer software, or a combination of both. Whether such functionality is implemented by hardware or software depends on the specific application and the design requirements of the entire system. Those skilled in the art may use various methods to implement the described functionality for each specific application, but such implementation should not be construed as exceeding the scope of protection of the embodiments of the present disclosure.

[0175] The present disclosure also provides a readable storage medium having instructions stored thereon, which implement the functions of any of the above method embodiments when executed by a computer.

[0176] The present disclosure also provides a computer program product, which implements the functions of any of the above method embodiments when executed by a computer.

[0177] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer programs. When the computer program is loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present disclosure are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer program 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, the computer program can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated therein. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a high-density digital video disc (DVD)), or a semiconductor medium (eg, a solid state disk (SSD)).

[0178] Unless the context requires otherwise, throughout the specification and claims, the term "comprise" and its other forms, such as the third person singular form "comprises" and the present participle form "comprising", are to be interpreted as open and inclusive, that is, "including, but not limited to". In the description of the specification, the terms "some embodiments", "exemplary embodiments", etc. are intended to indicate that the particular features, structures, materials or characteristics associated with the embodiment or example are included in at least one embodiment or example of the present disclosure. The schematic representation of the above terms does not necessarily refer to the same embodiment or example. In addition, the particular features, structures, materials or characteristics may be included in any one or more embodiments or examples in any appropriate manner.

[0179] Those skilled in the art will understand that the various numerical numbers such as first and second involved in the present disclosure are only for the convenience of description and are not used to limit the scope of the embodiments of the present disclosure, and also indicate the order of precedence.

[0180] The at least one in the present disclosure can also be described as one or more, and the multiple can be two, three, four or more, which is not limited in the present disclosure. In the embodiments of the present disclosure, for a technical feature, the technical features in the technical feature are distinguished by "first", "second", "third", "A", "B", "C" and "D", and there is no order of precedence or size between the technical features described by the "first", "second", "third", "A", "B", "C" and "D". "A and / or B" includes the following three combinations: only A, only B, and a combination of A and B.

[0181] Those skilled 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. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0182] Those skilled in the art will 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.

[0183] The above description is merely a specific embodiment of the present disclosure, but the scope of protection of the present disclosure is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this disclosure should be included in the scope of protection of the present disclosure. Therefore, the scope of protection of the present disclosure should be based on the scope of protection of the claims.

Claims

1. A precoding model training method, characterized in that: The method is performed by a network device, and includes: The network device receives channel state information of K terminals, where K is an integer greater than zero; According to the channel state information, a precoding model DNet model is trained; wherein the DNet model includes a compression and reconstruction sub-model DiSNet model and a precoding sub-model PreNet model, and the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models; The training of the precoding model DNet model according to the channel state information includes: processing the channel state information to obtain a training sample set; pre-training the DiSNet model and the PreNet model according to the training sample set; and jointly training the pre-trained DiSNet model and the pre-trained PreNet model according to the training sample set; The PreNet model includes: an analog precoding sub-model RFNet model and a digital precoding sub-model BBNet model; The pre-training of the PreNet model includes: converting the imperfect matrix Input to the RFNet model to generate simulated precoding The BBNet model is based on the imperfect matrix Generate digital precode Pre-training the RFNet model and the BBNet model using self-supervised learning and Adam optimization algorithm; The BBNet model includes: a first initial digital sub-model SNet model and a second initial digital sub-model MNet model; The BBNet model is based on the imperfect matrix Generate digital precode Including: according to the imperfect matrix Input to the RFNet model to generate simulated precoding Take the diagonally dominant matrix And the principal element matrix diag(H eq ); respectively, the diagonally dominant matrix and the principal element matrix diag(H eq ) is input to the SNet model and the MNet model, and then combined with the simulated precoding Perform normalization processing to generate the digital precoding 2. The method according to claim 1, characterized in that The processing of the channel state information to obtain a training sample set includes: Get the channel state information of K terminals {h1,…,h K }, summarized as matrix samples in, represents the conjugate transpose of the channel state information of the K-th terminal; Calculating the matrix samples to obtain an initial analog precoding matrix F and an initial digital precoding matrix W; The training sample set {(H, F, W)} is obtained according to the matrix samples, the initial analog precoding matrix F and the initial digital precoding matrix W.

3. The method according to claim 2, characterized in that The pre-training of the DiSNet model according to the training sample comprises: The K compression sub-models convert {h1,…,h K } is compressed into low-dimensional channel feature information {q1,…,q K }; The K reconstruction sub-models transform the low-dimensional channel feature information {q1,…,q K }Reconstructed into imperfect channel state information For the imperfect channel state information Perform splicing to generate the imperfect matrix The compression sub-model and the reconstruction sub-model are pre-trained using self-supervised learning and adaptive moment estimation Adam optimization algorithm.

4. The method according to claim 3, characterized in that The K groups of paired compression sub-models and reconstruction sub-models have the same structure and are independent of each other; The compression sub-model includes a convolutional layer and a fully connected layer to extract and compress the channel state information; The reconstruction sub-model includes a fully connected layer and a deep residual network to reconstruct the compressed channel state information.

5. The method according to any one of claims 1 to 4, characterized in that The method of jointly training the pre-trained DiSNet model and the pre-trained PreNet model according to the training sample set includes: The pre-trained DiSNet model and the pre-trained PreNet model are jointly trained using the Adam optimization algorithm and end-to-end learning.

6. A precoding method, characterized in that: The method is performed by a network device, and includes: According to the precoding model training method according to any one of claims 1 to 5, a precoding DNet model is obtained; wherein the DNet model includes a channel compression and reconstruction sub-model DiSNet model and a precoding design sub-model PreNet model; the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models; the K compression sub-models are sent to the K terminals using a distributed transmission method; wherein K is an integer greater than 0; Receiving compressed channel state information sent by k terminals among the K terminals; wherein k is an integer greater than zero and less than or equal to K; The compressed channel state information is input into the DNet model to generate a corresponding hybrid precoding matrix.

7. The method according to claim 6, characterized in that The step of inputting the compressed channel state information into the DNet model to generate a corresponding hybrid precoding matrix includes: The reconstruction sub-model reconstructs the compressed channel state information to generate imperfect channel state information; then splices it to generate an imperfect matrix The imperfect matrix Input to the PreNet model to generate the simulated precoding matrix and the digital precoding matrix 8. The method according to claim 7, characterized in that The PreNet model includes: an analog precoding sub-model RFNet model and a digital precoding sub-model BBNet model; The imperfect matrix Input to the PreNet model to generate the simulated precoding matrix and the digital precoding matrix include: The imperfect matrix Input to the RFNet model to generate the simulated precoding matrix The imperfect matrix Input to the BBNet model to generate the digital precoding matrix 9. The method according to claim 8, characterized in that The BBNet model includes: a first digital sub-model SNet model and a second digital sub-model MNet model; The imperfect matrix Input to the BBNet model to generate the digital precoding matrix include: According to the imperfect matrix Get the diagonally dominant matrix And the principal element matrix diag(H eq ); The diagonally dominant matrix and the principal element matrix diag(H eq ) is input to the SNet model and the MNet model, and then merged with the imperfect matrix Perform normalization processing to generate the digital precoding matrix 10. The method according to claim 6, characterized in that The DiSNet model includes multiple pairs of compression sub-models and reconstruction sub-models with the same structure and independent of each other; The compression sub-model includes a convolutional layer and a fully connected layer to extract and compress the channel state information; The reconstruction sub-model includes a fully connected layer and a deep residual network to reconstruct the compressed channel state information.

11. The method according to any one of claims 8 to 10, characterized in that The RFNet model includes a convolutional layer and a fully connected layer, with a linear rectifier function ReLU function as the activation function, according to the imperfect matrix Generate the simulated precoding matrix The corresponding shift angle is then calculated to obtain the simulated precoding matrix 12. A precoding method, characterized in that: The method is executed by a terminal, and includes: Receive a compression sub-model configured by a network device, the compression sub-model being determined by the network device according to a precoding DNet model obtained by the precoding model training method according to any one of claims 1 to 5; wherein the DNet model includes a channel compression and reconstruction sub-model DiSNet model and a precoding design sub-model PreNet model; the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models; wherein K is an integer greater than 0; The channel state information is input into the compression sub-model to generate compressed channel state information, and is sent to the network device.

13. A precoding model training device, characterized in that: The device comprises: An information acquisition unit, configured to receive channel state information of K terminals; wherein K is an integer greater than zero; A model training unit, configured to train a precoding model DNet model according to the channel state information; wherein the DNet model includes a compression and reconstruction sub-model DiSNet model and a precoding sub-model PreNet model, and the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models; The model training unit is specifically configured to process the channel state information to obtain a training sample set; pre-train the DiSNet model and the PreNet model based on the training sample set; and jointly train the pre-trained DiSNet model and the pre-trained PreNet model based on the training sample set; The PreNet model includes: an analog precoding sub-model RFNet model and a digital precoding sub-model BBNet model; The model training unit is specifically used to transform the imperfect matrix Input to the RFNet model to generate simulated precoding The BBNet model is based on the imperfect matrix Generate digital precode Pre-training the RFNet model and the BBNet model using self-supervised learning and Adam optimization algorithm; The BBNet model includes: a first initial digital sub-model SNet model and a second initial digital sub-model MNet model; The model training unit is specifically used to train the model according to the imperfect matrix Input to the RFNet model to generate simulated precoding Take the diagonally dominant matrix And the principal element matrix diag(H eq ); respectively, the diagonally dominant matrix and the principal element matrix diag(H eq ) is input to the SNet model and the MNet model, and then combined with the simulated precoding Perform normalization processing to generate the digital precoding 14. A precoding device, characterized in that: The device comprises: A transceiver module, configured to obtain a precoding DNet model according to the precoding model training method according to any one of claims 1 to 5; wherein the DNet model includes a channel compression and reconstruction submodel DiSNet model and a precoding design submodel PreNet model; the DiSNet model includes K pairs of compression submodels and reconstruction submodels; the K compression submodels are sent to the K terminals using a distributed transmission method; wherein K is an integer greater than 0; and receive compressed channel state information sent by k of the K terminals; wherein k is an integer greater than zero and less than or equal to K. The processing module is used to input the compressed channel state information into the DNet model to generate a corresponding hybrid precoding matrix.

15. A precoding device, characterized in that: The device comprises: A transceiver module, configured to receive a compression sub-model configured by a network device, wherein the compression sub-model is determined by the network device according to a precoding DNet model obtained by the precoding model training method according to any one of claims 1 to 5; wherein the DNet model includes a channel compression and reconstruction sub-model DiSNet model and a precoding design sub-model PreNet model; wherein the DiSNet model includes K pairs of compression sub-models and reconstruction sub-models; wherein K is an integer greater than 0; The processing module is used to input the channel state information into the compression sub-model, generate compressed channel state information, and send it to the network device.

16. A communication device, characterized in that: The device includes a processor and a memory, wherein a computer program is stored in the memory, and the processor executes the computer program stored in the memory, so that the device performs the method according to any one of claims 1 to 11.

17. A communication device, characterized in that: The device includes a processor and a memory, wherein a computer program is stored in the memory, and the processor executes the computer program stored in the memory to enable the device to perform the method according to claim 12 .

18. A communication device, characterized in that: include: processor and interface circuits; The interface circuit is used to receive code instructions and transmit them to the processor; The processor is configured to run the code instructions to perform the method according to any one of claims 1 to 11.

19. A communication device, characterized in that: include: processor and interface circuits; The interface circuit is used to receive code instructions and transmit them to the processor; The processor is configured to execute the code instructions to perform the method according to claim 12.

20. A computer-readable storage medium storing instructions, which, when executed, enable the method according to any one of claims 1 to 11 to be implemented.

21. A computer-readable storage medium storing instructions, which, when executed, enable the method according to claim 12 to be implemented.

Images