Communication method and communication apparatus

Abstract

A communication method and a communication apparatus. In the method, a network device realizes port dimension reduction on the basis of a dimension reduction model to measure an equivalent downlink channel; a terminal device feeds back to the network device a compression feedback quantity of equivalent channel information that is obtained on the basis of a compression model matching the dimension reduction model; and then the network device acquires compressed equivalent downlink channel information reconstructed on the basis of a reconstruction model matching the dimension reduction model. In the method, the dimension reduction model, the compression model and the reconstruction model are models matching each other after joint training. The method can improve the compressibility of the equivalent channel information and further reduce feedback overheads. The models involved in the present application may be an AI model, a neural network model, a machine learning model, etc.

Description

Communication method and communication device

[0001] This application claims priority to the Chinese patent application filed with the State Intellectual Property Office of China on December 29, 2023, with application number 202311863068.8 and application name “Communication Method and Communication Device”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to the field of communications, and more specifically, to a communication method and a communication device. Background Art

[0003] In a communication system, network equipment can use port dimensionality reduction to measure the equivalent downlink channels under multiple physical antennas using a small number of downlink reference signals. Terminal equipment measures the received downlink reference signals to obtain downlink channel information and compresses the obtained downlink channel information based on an artificial intelligence (AI) model. The compressed information is then fed back to the network equipment via uplink control information (UCI). Currently, the dimensionality reduction weights used in port dimensionality reduction can be calculated using an orthogonal codebook or orthogonal decomposition. However, research has found that based on this dimensionality reduction weight, the measured equivalent channel information is not highly compressible. Therefore, how to further reduce feedback overhead has become an urgent issue to be addressed. Summary of the Invention

[0004] The present application provides a communication method and a communication device, which can further reduce the feedback overhead of downlink channel information.

[0005] In a first aspect, a communication method is provided. The method may be executed by a network device, or may be executed by a component of the network device (eg, a chip or circuit), without limitation.

[0006] The method includes: digitally precoding M reference signals corresponding to M antenna ports based on dimensionality reduction weights, the dimensionality reduction weights are used to measure equivalent downlink channels of N physical antennas using the M reference signals, N is a positive integer, M is a positive integer less than N, and the dimensionality reduction weights are determined based on a dimensionality reduction model; sending M reference signals; sending first indication information, the first indication information indicating a compression model, the compression model matching the dimensionality reduction model; receiving a first sequence, the first sequence being a compressed feedback amount obtained by taking first downlink channel information as input to the compression model, the first downlink channel information being obtained based on measurement of the M reference signals; obtaining second downlink channel information, the second downlink channel information being information obtained by taking the first sequence as input to a reconstruction model, the reconstruction model matching the dimensionality reduction model, and the second downlink channel being used to determine the downlink channel information of the N physical antennas.

[0007] It can be understood that the optimization goal of the current dimensionality reduction method is only to minimize the energy loss of the equivalent channel after dimensionality reduction. Such a method does not take into account the compressibility of the channel after dimensionality reduction. The dimensionality reduction model and the compression model in this application are matching models. That is to say, the dimensionality reduction method and the compression method in this application are jointly optimized. While realizing port dimensionality reduction, the compressibility of the channel after dimensionality reduction is also considered. Therefore, compared with the current downlink channel information feedback scheme, the above technical scheme can make the equivalent downlink channel information more compressible, thereby achieving the effect of lower feedback overhead.

[0008] In certain implementations of the first aspect, the method further includes: inputting the first sequence into a reconstruction model for decompression to obtain second downlink channel information.

[0009] In a second aspect, a communication method is provided. The method can be executed by a terminal device, or can also be executed by a component of the terminal device (such as a chip or circuit), without limitation.

[0010] The method includes: receiving M reference signals corresponding to M antenna ports, where M is a positive integer; obtaining first downlink channel information based on the M reference signals; receiving first indication information, where the first indication information indicates a compression model, the compression model matches a dimensionality reduction model, the dimensionality reduction model is used to obtain dimensionality reduction weights, and the dimensionality reduction weights are used to measure equivalent downlink channels of N physical antennas using the M reference signals, where N is a positive integer and M is a positive integer less than N; and sending a first sequence, where the first sequence is a compressed feedback amount obtained by using the first downlink channel information as input to the compression model, the first sequence is used to be decompressed as input to a reconstruction model to obtain second downlink channel information, the reconstruction model matches the dimensionality reduction model, and the second downlink channel is used to determine the downlink channel information of the N physical antennas.

[0011] For the beneficial effects of the second aspect, please refer to the description of the first aspect and will not be repeated here.

[0012] In certain implementations of the first aspect or the second aspect, the method further includes: inputting the first downlink channel information into a compression model for compression to obtain a first sequence.

[0013] In certain implementations of the first aspect or the second aspect, the dimensionality reduction weight is output information obtained by using the channel prior information as input to the dimensionality reduction model.

[0014] In certain implementations of the first aspect or the second aspect, the channel prior information is determined based on at least one of the following information: currently measured uplink channel information, historical uplink channel information, historical downlink channel information, or perceived channel information.

[0015] In certain implementations of the first aspect or the second aspect, the dimensionality reduction model, the compression model, and the reconstruction model are all related to the number of antenna ports M, the number of physical antennas N, and channel prior information.

[0016] In certain implementations of the first aspect or the second aspect, the first downlink channel information indicates a response of the equivalent downlink channel, and the first sequence is a compressed feedback amount of the response.

[0017] In certain implementations of the first or second aspect, the first downlink channel information indicates a precoding matrix corresponding to a response of the equivalent downlink channel, and the first sequence is a compressed feedback amount of the precoding matrix information corresponding to the response.

[0018] In certain implementations of the first aspect or the second aspect, the downlink channel information of the N physical antennas includes downlink precoding matrices of the N physical antennas and downlink channel responses of the N physical antennas.

[0019] In certain implementations of the first aspect, a downlink precoding matrix of N physical antennas is obtained, where the downlink precoding matrix of the N physical antennas is determined based on the dimensionality reduction weights and the second downlink channel information.

[0020] In a third aspect, a communication method is provided. The method can be performed by a first device, or can also be performed by a component (e.g., a chip or circuit) of the first device, without limitation. For example, the first device can be a network device or AI network element #1, such as a third-party network element #1.

[0021] The method includes: obtaining a dimensionality reduction weight based on a dimensionality reduction model, the dimensionality reduction weight being used to digitally precode M reference signals corresponding to M antenna ports to measure equivalent downlink channels of N physical antennas, where N is a positive integer and M is a positive integer less than N; inputting a first sequence into a reconstruction model, wherein the first sequence is a compressed feedback amount obtained by taking first downlink channel information as input to a compression model, the first downlink channel information being obtained by measuring the M reference signals, and the compression model, the reconstruction model, and the dimensionality reduction model being matched; and outputting second downlink channel information, where the second downlink channel is used to determine the downlink channel information of the N physical antennas.

[0022] In the above technical solution, the dimensionality reduction model, the compression model and the reconstruction model are matched models after joint training. This method can increase the compressibility of the equivalent downlink channel information and reduce the feedback overhead.

[0023] In a fourth aspect, a communication method is provided. The method can be performed by a second device, or can also be performed by a component of the second device (such as a chip or circuit), without limitation. For example, the second device can be a terminal device or AI network element #2, such as a third-party network element #2.

[0024] The method includes: inputting first downlink channel information into a compression model, wherein the first downlink channel information is obtained by measuring M reference signals, and the M reference signals are reference signals obtained by digitally precoding reference signals corresponding to M antenna ports based on dimensionality reduction weights, and the dimensionality reduction weights are used to measure equivalent downlink channels of N physical antennas using the M reference signals, N is a positive integer, M is a positive integer less than N, and the dimensionality reduction weights are determined based on the dimensionality reduction model, and the compression model and the dimensionality reduction model match; outputting a first sequence, and the first sequence is used to be decompressed as an input of a reconstruction model to obtain second downlink channel information, the reconstruction model is associated with the dimensionality reduction model, and the second downlink channel is used to determine the downlink channel information of the N physical antennas, and the reconstruction model is associated with the dimensionality reduction model.

[0025] For the beneficial effects of the fourth aspect, please refer to the description of the third aspect and will not be repeated here.

[0026] In certain implementations of the third aspect or the fourth aspect, the dimensionality reduction weight is output information obtained by using the channel prior information as input to the dimensionality reduction model.

[0027] In certain implementations of the third aspect or the fourth aspect, the channel prior information is determined based on at least one of the following information: currently measured uplink channel information, historical uplink channel information, historical downlink channel information, or perceived channel information.

[0028] In certain implementations of the third aspect or the fourth aspect, the dimensionality reduction model, the compression model, and the reconstruction model are all related to the number of antenna ports M, the number of physical antennas N, and channel prior information.

[0029] In certain implementations of the third aspect or the fourth aspect, the first downlink channel information indicates a response of the equivalent downlink channel, and the first sequence is a compressed feedback amount of the response.

[0030] In certain implementations of the third or fourth aspect, the first downlink channel information indicates a precoding matrix corresponding to a response of the equivalent downlink channel, and the first sequence is a compressed feedback amount of the precoding matrix information corresponding to the response.

[0031] In certain implementations of the third aspect or the fourth aspect, the downlink channel information of the N physical antennas includes downlink precoding matrices of the N physical antennas and downlink channel responses of the N physical antennas.

[0032] In a fifth aspect, a model training method is provided. The method can be performed by a first device, or can also be performed by a component of the first device (such as a chip or circuit), without limitation. For example, the first device can be a network device or AI network element #1, such as a third-party network element #1.

[0033] The method includes: obtaining a training data set, the training data set includes channel prior information and first downlink channel information, the first downlink channel information is the real downlink channel information of N physical antennas, N is a positive integer; sending second downlink channel information, the second downlink channel information is determined based on the first downlink channel information and a dimensionality reduction weight, wherein the dimensionality reduction weight is output information obtained by taking the channel prior information as the input of a dimensionality reduction model, the dimensionality reduction weight is used to digitally precode M reference signals corresponding to M antenna ports to measure the equivalent downlink channels of N physical antennas, M is a positive integer less than N; receiving a first sequence, the first sequence is output information obtained by taking the second downlink channel information as the input of a compression model; sending a first gradient signal The method comprises receiving first gradient information, wherein the first gradient information is input-side gradient information of the reconstruction model, the first gradient information is used to update network parameters of the compression model, the first gradient information is determined based on the first error information and the reconstruction model, the first error information is determined based on the third downlink channel information and the first downlink channel information, the third downlink channel information is output information obtained by taking the first sequence as input of the reconstruction model or calculated based on the output information of the reconstruction model; receiving second gradient information, wherein the second gradient information is input-side gradient information of the compression model, and the second gradient information is determined based on the first gradient information and the compression model; updating model parameters of the reconstruction model based on the first error information, and updating model parameters of the dimensionality reduction model based on the second gradient information.

[0034] While current model training methods only support dual-end, two-model training, this technical solution supports dual-end, multi-model joint training. This approach enables joint training and optimization of the dimensionality reduction model, compression model, and reconstruction model, reducing channel information feedback overhead.

[0035] In certain implementations of the fifth aspect, model training is terminated when a termination condition of model training is met. For example, the termination condition of model training may be: an error indicated by the first error information is less than or equal to a first threshold, or the number of model training rounds is equal to a second threshold.

[0036] In a sixth aspect, a model training method is provided, which can be performed by a second device, or can also be performed by a component of the second device (such as a chip or circuit), without limitation. For example, the second device can be a terminal device or AI network element #2, such as a third-party network element #2.

[0037] The method includes: receiving second downlink channel information, where the second downlink channel information is determined based on first downlink channel information and a dimensionality reduction weight, where the first downlink channel information is real downlink channel information of N physical antennas, the dimensionality reduction weight is output information obtained by using channel prior information as input to a dimensionality reduction model, the dimensionality reduction weight is used to digitally precode M reference signals corresponding to M antenna ports to measure equivalent downlink channels of the N physical antennas, where N is a positive integer, M is a positive integer less than N, and the first downlink channel information and the channel prior information are a training data set for model training; sending a first sequence, where the first sequence is output information obtained by using the second downlink channel information as input to a compression model; receiving first gradient information, The first gradient information is input-side gradient information of the reconstruction model, wherein the first gradient information is used to update model parameters of the compression model. The first gradient information is determined based on the first error information and the reconstruction model. The first error information is determined based on the third downlink channel information and the first downlink channel information. The third downlink channel information is output information obtained by using the first sequence as input to the reconstruction model or calculated based on the output information of the reconstruction model. The first error information is used to update model parameters of the reconstruction model. Second gradient information is sent. The second gradient information is input-side gradient information of the compression model. The second gradient information is determined based on the first gradient information and the compression model. The model parameters of the compression model are updated based on the first gradient information.

[0038] For the beneficial effects of the sixth aspect, please refer to the description of the fifth aspect and will not be repeated here.

[0039] In certain implementations of the fifth aspect or the sixth aspect, the channel prior information is determined based on at least one of the following information: currently measured uplink channel information, historical uplink channel information, historical downlink channel information, or perceived channel information.

[0040] In certain implementations of the fifth or sixth aspect, the first downlink channel information is actual downlink channel responses of N physical antennas.

[0041] In certain implementations of the fifth or sixth aspect, the second downlink channel information indicates a downlink channel response, which is determined based on the first downlink channel information and a dimensionality reduction weight, or the second downlink channel information indicates a weight corresponding to the downlink channel response.

[0042] In certain implementations of the fifth or sixth aspects, the third downlink channel information is the reconstructed weights of N physical antennas, and the first error information indicates the error between the third downlink channel information and the actual weights of the N physical antennas, where the actual weights of the N physical antennas are determined based on the first downlink channel information.

[0043] In a seventh aspect, a model training method is provided, which can be performed by a first device, or can also be performed by a component of the first device (such as a chip or circuit), without limitation. For example, the first device can be a network device, a terminal device, or an AI network element #1, such as a third-party network element #1.

[0044] The method includes: obtaining a first training data set, the first training data set includes channel prior information and first downlink channel information, the first downlink channel information is the real downlink channel information of N physical antennas, and N is a positive integer; jointly training a dimensionality reduction model, a compression model, and a reconstruction model, wherein the input of the dimensionality reduction model is the channel prior information, and the output is a dimensionality reduction weight value, wherein the dimensionality reduction weight value is used to digitally precode M reference signals corresponding to M antenna ports to measure the equivalent downlink channels of the N physical antennas, and M is a positive integer less than N; the input of the compression model is second downlink channel information, and the output is a first sequence, wherein the second downlink channel information is determined based on the first downlink channel information and the dimensionality reduction weight value; the input of the reconstruction model is the first sequence, and the output is third downlink channel information; and updating network equipment parameters of the dimensionality reduction model, the compression model, and the reconstruction model based on the first error information, and the first error information is determined based on the third downlink channel information and the first downlink channel information.

[0045] In the above technical solution, the difference from the joint training method of the fifth and sixth aspects is that in this method, three models are jointly trained on a single side (i.e., the first device).

[0046] In certain implementations of the seventh aspect, model training is terminated when a termination condition of model training is met. For example, the termination condition of model training may be: an error indicated by the first error information is less than or equal to a first threshold, or the number of model training rounds is equal to a second threshold.

[0047] In an eighth aspect, a method for model training is provided, which can be performed by a second device, or can also be performed by a component of the second device (such as a chip or circuit), without limitation. The second device can be a device different from the first device. For example, the second device can be a network device, a terminal device, or an AI network element #2, such as a third-party network element #2.

[0048] The method includes: receiving a second training data set, the second training data set including a dimensionality reduction model, a compression model, and a reconstruction model; after completing joint training based on the first training data set, the output information of the dimensionality reduction model obtained by reusing the first training data set, and the output information of the compression model; the first training data set includes channel prior information and first downlink channel information; wherein, in a round of joint training corresponding to the dimensionality reduction model, the compression model, and the reconstruction model, the input of the dimensionality reduction model is the channel prior information, and the output is the dimensionality reduction weight; wherein the dimensionality reduction weight is used to digitally precode M reference signals corresponding to M antenna ports to measure the equivalent downlink channels of N physical antennas, M is a positive integer less than N; the input of the compression model is the second downlink channel information, and the output is the first sequence; wherein the second downlink channel information is determined based on the first downlink channel information and the dimensionality reduction weight; the input of the reconstruction model is the first sequence, and the output is the third downlink channel information; training the first model based on the second training data set to obtain the second model.

[0049] In certain implementations of the seventh or eighth aspects, the method further includes: sending a compression model corresponding to the completion of the model training; or, sending a second training data set, the second training data set including the output information of the dimensionality reduction model obtained by using the first training data set again after the model training is completed, and the output information of the compression model.

[0050] In the above technical solution, if model A among the three models will be deployed on another device side (for example, the second device side), the first device can send the trained model A or the training data set corresponding to model A to the second device, and the second device completes the model deployment or performs secondary training based on the received model A or training data set.

[0051] In certain implementations of the seventh aspect or the eighth aspect, the channel prior information is determined based on at least one of the following information: uplink channel information currently measured, historical uplink channel information, historical downlink channel information, and perceived channel information.

[0052] In certain implementations of the seventh or eighth aspect, the first downlink channel information is actual downlink channel responses of N physical antennas.

[0053] In certain implementations of the seventh or eighth aspects, the second downlink channel information indicates a downlink channel response, which is determined based on the first downlink channel information and a dimensionality reduction weight, or the second downlink channel information indicates a weight corresponding to the downlink channel response.

[0054] In certain implementations of the seventh or eighth aspects, the third downlink channel information is the reconstructed weights of N physical antennas, and the first error information indicates the error between the third downlink channel information and the actual weights of the N physical antennas, where the actual weights of the N physical antennas are determined based on the first downlink channel information.

[0055] In a ninth aspect, a communication device is provided, the device being configured to execute the method provided in the first aspect. Specifically, the device may include units and / or modules, such as a processing unit and / or a communication unit, for executing the method in any aspect of the first aspect or any possible implementation of the first aspect.

[0056] In one implementation, the apparatus is a network device. When the apparatus is a network device, the communication unit may be a transceiver circuit; and the processing unit may be a processing circuit.

[0057] For example, the transceiver circuit may be a transceiver, an input / output interface, or an input / output circuit.

[0058] By way of example, the processing circuit may be one or more processors, or may be all or part of the circuits in one or more processors.

[0059] In another implementation, the device is a chip, chip system, or circuit used in a network device. When the device is a chip, chip system, or circuit used in a terminal device, the communication unit may be a transceiver circuit on the chip, chip system, or circuit; and the processing unit may be a processing circuit.

[0060] For example, the transceiver circuit may be a transceiver, an input / output interface, an input / output circuit, a pin, or a related circuit.

[0061] By way of example, the processing circuit may be a logic circuit or a processor.

[0062] In a tenth aspect, a communication device is provided, the device being configured to execute the method provided in the second aspect. Specifically, the device may include units and / or modules, such as a processing unit and / or a communication unit, for executing the method in any aspect or any possible implementation of the second aspect.

[0063] In one implementation, the apparatus is a terminal device. When the apparatus is a terminal device, the communication unit may be a transceiver circuit; and the processing unit may be a processing circuit.

[0064] In another implementation, the apparatus is a chip, chip system, or circuit used in a terminal device. When the apparatus is a chip, chip system, or circuit used in a terminal device, the communication unit may be a transceiver circuit on the chip, chip system, or circuit; and the processing unit may be a processing circuit.

[0065] For examples of transceiver circuits and processing circuits, please refer to the description of the ninth aspect and will not be repeated here.

[0066] In an eleventh aspect, a communication device is provided, configured to execute the method provided in the third aspect, the fifth aspect, or the seventh aspect. Specifically, the device may include units and / or modules, such as a processing unit and / or a communication unit, configured to execute the method in any of the third aspect, the fifth aspect, or the seventh aspect, or any possible implementation of the third aspect, the fifth aspect, or the seventh aspect.

[0067] In one implementation, the apparatus is a first device. When the apparatus is the first device, the communication unit may be a transceiver circuit; and the processing unit may be a processing circuit.

[0068] In another implementation, the apparatus is a chip, chip system, or circuit used in the first device. When the apparatus is a chip, chip system, or circuit used in the first device, the communication unit may be a transceiver circuit on the chip, chip system, or circuit; and the processing unit may be a processing circuit.

[0069] For examples of transceiver circuits and processing circuits, please refer to the description of the ninth aspect and will not be repeated here.

[0070] In a twelfth aspect, a communication device is provided, the device being configured to execute the method provided in the fourth aspect, the sixth aspect, or the eighth aspect. Specifically, the device may include units and / or modules, such as a processing unit and / or a communication unit, for executing the method in any one of the fourth aspect, the sixth aspect, or the eighth aspect, or any possible implementation of the fourth aspect, the sixth aspect, or the eighth aspect.

[0071] In one implementation, the apparatus is a second device. When the apparatus is the second device, the communication unit may be a transceiver circuit; and the processing unit may be a processing circuit.

[0072] In another implementation, the apparatus is a chip, chip system, or circuit used in the second device. When the apparatus is a chip, chip system, or circuit used in a terminal device, the communication unit may be a transceiver circuit on the chip, chip system, or circuit; and the processing unit may be a processing circuit.

[0073] For examples of transceiver circuits and processing circuits, please refer to the description of the ninth aspect and will not be repeated here.

[0074] In the thirteenth aspect, a communication device is provided, comprising: at least one processing circuit, the at least one processing circuit being coupled to at least one memory, the at least one memory being used to store computer programs or instructions, and the at least one processing circuit being used to call and run the computer program or instructions from the at least one memory, so that the communication device executes the method in any one of the first aspect, the third aspect, the fifth aspect, or the seventh aspect, or any possible implementation of any one of the first aspect, the third aspect, the fifth aspect, or the seventh aspect.

[0075] In the fourteenth aspect, a communication device is provided, comprising: at least one processing circuit, the at least one processing circuit being coupled to at least one memory, the at least one memory being used to store computer programs or instructions, and the at least one processing circuit being used to call and run the computer program or instructions from the at least one memory, so that the communication device executes the method in any one of the second aspect, the fourth aspect, the sixth aspect, or the eighth aspect and any possible implementation manner of any one of the second aspect, the fourth aspect, the sixth aspect, or the eighth aspect.

[0076] In a fifteenth aspect, a processing circuit is provided for executing the methods provided in the above aspects.

[0077] For operations such as sending and acquiring / receiving involved in the processing circuit, unless otherwise specified, or if they do not conflict with their actual functions or internal logic in the relevant descriptions, they can be understood as operations such as output, reception, and input of the processing circuit, or as sending and receiving operations performed by the radio frequency circuit and antenna. This application does not limit this.

[0078] In the sixteenth aspect, a computer-readable storage medium is provided, which stores a program code for execution by a device, and the program code includes a method for executing any one of the above-mentioned aspects 1 to 8 or any possible implementation of the aspects 1 to 8.

[0079] In the seventeenth aspect, a computer program product comprising instructions is provided, which, when run on a computer, enables the computer to execute the method in any one of the above-mentioned aspects from the first to the eighth aspect and any possible implementation of the aspects from the first to the eighth aspect.

[0080] In the eighteenth aspect, a chip is provided, which includes a processing circuit and a communication interface. The processing circuit reads instructions stored in a memory through the communication interface to execute the method in any one of the above-mentioned aspects one to eight or any possible implementation method of aspects one to six.

[0081] Optionally, as an implementation method, the chip also includes a memory, in which a computer program or instruction is stored, and the processing circuit is used to execute the computer program or instruction stored in the memory. When the computer program or instruction is executed, the processing circuit is used to execute the method in any aspect of the first to eighth aspects above or any possible implementation method of the first to eighth aspects.

[0082] For example, the processing circuit described in the above aspects may be one or more processors, or may be all or part of the circuits in one or more processors.

[0083] In the nineteenth aspect, a communication system is provided, which includes the communication device shown in the thirteenth aspect and the fourteenth aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] FIG1 is a schematic diagram of a possible application framework in a communication system.

[0085] FIG2 is a schematic diagram of a possible application framework in a communication system.

[0086] FIG3 is a schematic diagram of a communication system applicable to an embodiment of the present application.

[0087] FIG4 is a schematic diagram of another communication system applicable to an embodiment of the present application.

[0088] FIG5 is a schematic block diagram of an autoencoder.

[0089] Figure 6 is a schematic diagram of forward propagation and back propagation of a neural network.

[0090] FIG7 is a schematic diagram of the CSI feedback process.

[0091] FIG8 is a schematic flow chart of a communication method 800 proposed in this application.

[0092] FIG9 is a schematic flowchart of a model training method 900 proposed in this application.

[0093] FIG10 is a schematic flowchart of a model training method 1000 proposed in this application.

[0094] FIG11 is a comparison chart of simulation performances corresponding to the model dimensionality reduction and compression method provided by the present application and the existing orthogonal codebook dimensionality reduction and compression method.

[0095] FIG12 is a schematic block diagram of a communication device 1200 provided in an embodiment of the present application.

[0096] FIG13 is a schematic block diagram of a communication device 1300 provided in an embodiment of the present application. DETAILED DESCRIPTION

[0097] The technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings.

[0098] Before introducing the embodiments of the present application, the following points are first explained.

[0099] 1. In the description of the embodiments of the present application, unless otherwise specified, “multiple” means two or more.

[0100] 2. In the various embodiments of the present application, unless otherwise specified or provided for by logic, the terms and / or descriptions between different embodiments are consistent and can be referenced by each other. The technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0101] 3. The various numerical numbers involved in this application are only used for the convenience of description and are not used to limit the scope of this application. The size of the serial numbers involved in this application does not mean the order of execution. The execution order of each process should be determined by its function and internal logic. For example, the terms "first", "second", "third", "fourth" and other various terminology labels (if any) in the specification and claims and drawings of this application are used to distinguish similar objects and are not used to limit the size, content, order, timing, priority or importance of multiple objects. For example, the first information and the second information do not represent the difference in the amount of information, content, priority or importance.

[0102] 4. The terms "comprise", "include", "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed but may include other steps or units not explicitly listed or inherent to such process, method, product or apparatus.

[0103] 5. In each embodiment of the present application, "network element A sends information A to network element B" can be understood as the destination end of the information A or the intermediate network element in the transmission path between the destination end and the network element B, which may include directly or indirectly sending information to network element B. "Network element B receives information A from network element A" can be understood as the source end of the information A or the intermediate network element in the transmission path between the source end and the network element A, which may include directly or indirectly receiving information from network element A. The information may be processed as necessary between the source end and the destination end of the information transmission, such as format changes, etc., but the destination end can understand the valid information from the source end. Similar expressions in this application can be understood similarly and will not be elaborated here.

[0104] In other words, sending and receiving can be performed between devices, for example, between terminal device #1 and terminal device #2, or can be performed within a device, for example, sending or receiving between components, modules, chips, software modules or hardware modules within the device through a bus, traces or interface.

[0105] 6. In the embodiments of the present application, indications include direct indications (also called explicit indications) and implicit indications. Direct indication of information A refers to including information A; implicit indication of information A refers to indicating information A through the correspondence between information A and information B and the direct indication of information B. The correspondence between information A and information B can be predefined, pre-stored, pre-burned, or pre-configured.

[0106] 7. In the embodiments of the present application, information C is used to determine information D, which includes information D being determined solely based on information C, as well as information D being determined based on information C and other information. Furthermore, information C can also be used to determine information D indirectly, for example, where information D is determined based on information E, and information E is determined based on information C.

[0107] 8. "Storage" or "saving" in the embodiments of this application may refer to storage in one or more memories. The one or more memories may be provided separately or integrated into an encoder or decoder, a processor, or a communication device. The one or more memories may also be provided in part separately and in part integrated into a decoder, a processor, or a communication device. The type of memory may be any form of storage medium and is not limited in this application.

[0108] 9. The “protocol” involved in the embodiments of the present application may refer to a standard protocol in the field of communications, for example, it may include a fourth generation (4G) network / fifth generation (5G) network protocol, a new radio (NR) protocol, and related protocols used in future communication systems. This application does not limit this.

[0109] 10. The dotted arrows or boxes in the schematic diagrams in the accompanying drawings of this application represent optional steps or optional modules.

[0110] The technical solutions provided in this application can be applied to various communication systems, such as: fifth generation (5G) or new radio (NR) systems, long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, wireless local area networks (WLAN) systems, satellite communication systems, future communication systems, such as sixth generation (6G) mobile communication systems, or a fusion system of multiple systems. The technical solutions provided in this application can also be applied to device to device (D2D) communication, vehicle-to-everything (V2X) communication, machine to machine (M2M) communication, machine type communication (MTC), and Internet of Things (IoT) communication systems or other communication systems.

[0111] The technical solutions provided in this application can be applied to various communication systems, such as: fifth generation (5G) or new radio (NR) systems, long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, wireless local area networks (WLAN) systems, satellite communication systems, future communication systems, such as sixth generation (6G) mobile communication systems, or a fusion system of multiple systems. The technical solutions provided in this application can also be applied to device to device (D2D) communication, vehicle-to-everything (V2X) communication, machine to machine (M2M) communication, machine type communication (MTC), and Internet of Things (IoT) communication systems or other communication systems.

[0112] A device in a communication system can send a signal to another device or receive a signal from another device. The signal may include information, signaling, or data. The device may also be replaced by an entity, a network entity, a network element, a communication device, a communication module, a node, a communication node, etc. The present disclosure uses the device as an example for description. For example, the communication system may include at least one terminal device and at least one network device. The network device may send a downlink signal to the terminal device, and / or the terminal device may send an uplink signal to the network device. It is understandable that the terminal device in the present application may be replaced by the first device, and the network device may be replaced by the second device, and the two may perform the corresponding communication method in the present disclosure.

[0113] In an embodiment of the present application, the terminal device may also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent or user device.

[0114] The terminal device may be a device that provides voice / data, such as a handheld device or vehicle-mounted device with a wireless connection function. At present, some examples of terminals are: mobile phones, tablet computers, laptop computers, PDAs, mobile internet devices (MIDs), wearable devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical surgery, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, cellular phones, cordless phones, session initiation protocol (SIP) phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), handheld devices with wireless communication capabilities, computing devices or other processing devices connected to wireless modems, wearable devices, terminal devices in 5G networks or future evolved public land mobile communication networks (PLMNs). The terminal equipment in the network (PLMN), etc., is not limited to this in the embodiments of the present application.

[0115] As an example and not a limitation, in the embodiment of the present application, the terminal device may also be a wearable device. Wearable devices may also be called wearable smart devices, which are a general term for wearable devices that are intelligently designed and developed using wearable technology for daily wear, such as glasses, gloves, watches, clothing, and shoes. A wearable device is a portable device that is worn directly on the body or integrated into the user's clothes or accessories. Wearable devices are not only hardware devices, but also achieve powerful functions through software support, data interaction, and cloud interaction. Broadly speaking, wearable smart devices include those that are fully functional, large in size, and can achieve complete or partial functions without relying on smartphones, such as smart watches or smart glasses, as well as those that only focus on a certain type of application function and need to be used in conjunction with other devices such as smartphones, such as various smart bracelets and smart jewelry for vital sign monitoring.

[0116] In the embodiments of the present application, the device for realizing the function of the terminal device can be a terminal device, or a device capable of supporting the terminal device to realize the function, such as a chip system, which can be installed in the terminal device or used in combination with the terminal device. In the embodiments of the present application, the chip system can be composed of a chip, or it can include a chip and other discrete devices. In the embodiments of the present application, only the terminal device is used as an example for description, and the embodiments of the present application are not limited to the solutions of the embodiments of the present application.

[0117] The network device in the embodiments of the present application may be a device for communicating with a terminal device, and may also be referred to as an access network device or a radio access network device. For example, the network device may be a base station. The network device in the embodiments of the present application may refer to a radio access network (RAN) node (or device) that connects a terminal device to a wireless network. A base station may broadly cover various names as follows, or replace the following names, such as: NodeB, evolved NodeB (eNB), next generation NodeB (gNB), relay station, access point, transmitting and receiving point (TRP), transmitting point (TP), master station, secondary station, multi-standard radio (motor slide retainer, MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed unit (DU), radio unit (RU), positioning node, etc. A base station may be a macro base station, a micro base station, a relay node, a donor node, or the like, or a combination thereof. The base station may also refer to a communication module, modem or chip that is set in the aforementioned equipment or device. The base station may also be a mobile switching center and a device that performs the base station function in D2D, V2X, and M2M communications, a network side device in a 6G network, a device that performs the base station function in future communication systems, etc. The base station can support networks with the same or different access technologies. Optionally, the RAN node may also be a server, a wearable device, a vehicle or an on-board device, etc. For example, the access network device in the vehicle to everything (V2X) technology may be a road side unit (RSU). The embodiments of the present application do not limit the specific technology and specific device form adopted by the network equipment.

[0118] In some deployments, the network devices mentioned in the embodiments of the present application may include a CU, a DU, or both a CU and a DU, or a control plane CU node (central unit-control plane (CU-CP)), a user plane CU node (central unit-user plane (CU-UP)), and a DU node. For example, the network devices may include a gNB-CU-CP, a gNB-CU-UP, and a gNB-DU.

[0119] In some deployments, multiple RAN nodes collaborate to assist terminals in achieving wireless access, with different RAN nodes implementing portions of the base station's functionality. For example, a RAN node can be a CU, DU, CU-CP, CU-UP, or RU. The CU and DU can be separate or included in the same network element, such as the BBU. The RU can be included in a radio frequency device or radio unit, such as an RRU, AAU, or RRH.

[0120] The RAN node may support one or more types of fronthaul interfaces, and different fronthaul interfaces correspond to DUs and RUs with different functions. If the fronthaul interface between the DU and the RU is a common public radio interface (CPRI), the DU is configured to implement one or more baseband functions, and the RU is configured to implement one or more radio frequency functions. If the fronthaul interface between the DU and the RU is another type of interface, relative to the CPRI, some of the downlink and / or uplink baseband functions, such as precoding, digital beamforming (BF), or one or more of inverse fast Fourier transform (IFFT) / cyclic prefix (CP) for downlink, are moved from the DU to the RU for implementation; and for uplink, one or more of digital beamforming (BF), or fast Fourier transform (FFT) / CP removal are moved from the DU to the RU for implementation. In one possible implementation, the interface may be an enhanced common public radio interface (eCPRI). In the eCPRI architecture, the division between the DU and RU is different, corresponding to different types (category, Cat) of eCPRI, such as eCPRI Cat A, B, C, D, E, and F.

[0121] Taking eCPRI Cat A as an example, for downlink transmission, based on layer mapping, the DU is configured to implement layer mapping and one or more of the preceding functions (i.e., coding, rate matching, scrambling, modulation, and layer mapping), while other functions after layer mapping (e.g., resource element (RE) mapping, BF, or one or more of IFFT / CP addition) are moved to the RU for implementation. For uplink transmission, based on RE demapping, the DU is configured to implement demapping and one or more of the preceding functions (i.e., decoding, derate matching, descrambling, demodulation, inverse discrete Fourier transform (IDFT), channel equalization, and one or more of RE demapping), while other functions after demapping (e.g., digital BF or FFT / CP removal) are moved to the RU for implementation. It is understood that for a functional description of the DU and RU corresponding to various types of eCPRI, please refer to the eCPRI protocol and will not be detailed here.

[0122] In one possible design, the processing unit for implementing baseband functions in the BBU is called a baseband high layer (BBH) unit, and the processing unit for implementing baseband functions in the RRU / AAU / RRH is called a baseband low layer (BBL) unit.

[0123] In different systems, CU (or CU-CP and CU-UP), DU or RU may also have different names, but those skilled in the art can understand their meanings. For example, in an open radio access network (O-RAN / ORAN) system, CU may also be referred to as O-CU (open CU), DU may also be referred to as O-DU, CU-CP may also be referred to as O-CU-CP, CU-UP may also be referred to as O-CU-UP, and RU may also be referred to as O-RU. Any of the CU (or CU-CP, CU-UP), DU and RU in this application may be implemented by a software module, a hardware module, or a combination of a software module and a hardware module.

[0124] In the embodiments of the present application, the device for implementing the functions of the network device can be a network device; it can also be a device that can support the network device to implement the functions, such as a chip system, a hardware circuit, a software module, or a hardware circuit and a software module. The device can be installed in the network device or used in conjunction with the network device. In the embodiments of the present application, only the device for implementing the functions of the network device is used as an example to illustrate, and does not constitute a limitation on the solutions of the embodiments of the present application.

[0125] The network device and / or terminal device can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; it can also be deployed on the water surface; it can also be deployed on aircraft, balloons and satellites in the air. The embodiments of this application do not limit the scenarios in which the network device and the terminal device are located. In addition, the terminal device and the network device can be hardware devices, or they can be software functions running on dedicated hardware, software functions running on general-purpose hardware, such as virtualization functions instantiated on a platform (e.g., a cloud platform), or entities including dedicated or general-purpose hardware devices and software functions. This application does not limit the specific forms of the terminal device and the network device.

[0126] In wireless communication networks, such as mobile communication networks, the services supported by the networks are becoming increasingly diverse, and therefore the demands that need to be met are becoming increasingly diverse. For example, the network needs to be able to support ultra-high speeds, ultra-low latency, and / or ultra-large connections. This feature makes network planning, network configuration, and / or resource scheduling increasingly complex. In addition, as network functionality becomes increasingly powerful, such as supporting higher spectrum, supporting high-order multiple input multiple output (MIMO) technology, supporting beamforming, and / or supporting new technologies such as beam management, network energy saving has become a hot research topic. These new demands, new scenarios, and new features have brought unprecedented challenges to network planning, operation and maintenance, and efficient operation. To meet this challenge, artificial intelligence technology can be introduced into wireless communication networks to achieve network intelligence.

[0127] In order to support AI technology in wireless networks, AI nodes may also be introduced into the network.

[0128] Optionally, the AI ​​node can be deployed in one or more of the following locations in the communication system: access network equipment, terminal equipment, or core network equipment. Alternatively, the AI ​​node can be deployed separately, for example, in a location other than any of the above devices, such as a host or cloud server in an over-the-top (OTT) system. The AI ​​node can communicate with other devices in the communication system, such as one or more of the following: network equipment, terminal equipment, or core network elements.

[0129] It is understood that this application does not limit the number of AI nodes. For example, when there are multiple AI nodes, the multiple AI nodes can be divided based on function, such as different AI nodes are responsible for different functions.

[0130] It is also understood that AI nodes can be independent devices, or integrated into the same device to implement different functions, or can be network elements in hardware devices, or can be software functions running on dedicated hardware, or can be virtualized functions instantiated on a platform (e.g., a cloud platform). This application does not limit the specific form of the above-mentioned AI nodes. Among them, AI nodes can be AI network elements, AI entities, or AI modules.

[0131] Figure 1 is a schematic diagram of a possible application framework in a communication system. As shown in Figure 1, network elements in the communication system are connected through interfaces (such as NG, Xn) or air interfaces. One or more AI modules are provided in one or more devices of these network element nodes, such as core network equipment, access network nodes (RAN nodes), terminals or OAM (for clarity, only one is shown in Figure 1). The access network node can be a separate RAN node, or it can include multiple RAN nodes, for example, including CU and DU. The CU and / or DU can also be provided with one or more AI modules. Optionally, the CU can also be split into CU-CP and CU-UP. One or more AI models are provided in the CU-CP and / or CU-UP.

[0132] The AI ​​module is used to implement the corresponding AI function. The AI ​​modules deployed in different network elements may be the same or different. The model of the AI ​​module can implement different functions according to different parameter configurations. The model of the AI ​​module can be configured based on one or more of the following parameters: structural parameters (such as the number of neural network layers, the width of the neural network, the connection relationship between layers, the weight of the neuron, the activation function of the neuron, or at least one of the bias in the activation function), input parameters (such as the type of input parameters and / or the dimension of the input parameters), or output parameters (such as the type of output parameters and / or the dimension of the output parameters). Among them, the bias in the activation function can also be called the bias of the neural network.

[0133] An AI module can have one or more models. A model can infer an output, which includes one or more parameters. The learning, training, or inference processes of different models can be deployed on different nodes or devices, or on the same node or device.

[0134] Figure 2 is a schematic diagram of a possible application framework in a communication system. As shown in Figure 2, the communication system includes a RAN intelligent controller (RIC). For example, the RIC can be the AI ​​modules 117 and 118 shown in Figure 1, which are used to implement AI-related functions. The RIC includes a near-real-time RIC (near-real time RIC, near-RT RIC) and a non-real-time RIC (non-real time RIC, Non-RT RIC). Among them, the non-real-time RIC mainly processes non-real-time information, such as data that is not sensitive to delay, and the delay of this data can be in the order of seconds. The real-time RIC mainly processes near-real-time information, such as data that is relatively sensitive to delay, and the delay of this data is in the order of tens of milliseconds.

[0135] The near real-time RIC is used for model training and reasoning. For example, it is used to train an AI model and use the AI ​​model for reasoning. The near real-time RIC can obtain network-side and / or terminal-side information from a RAN node (e.g., CU, CU-CP, CU-UP, DU, and / or RU) and / or a terminal. This information can be used as training data or reasoning data. Optionally, the near real-time RIC can deliver the reasoning result to the RAN node and / or the terminal. Optionally, the reasoning result can be exchanged between the CU and the DU, and / or between the DU and the RU. For example, the near real-time RIC delivers the reasoning result to the DU, and the DU sends it to the RU.

[0136] The non-real-time RIC is also used for model training and reasoning. For example, it is used to train an AI model and use the model for reasoning. The non-real-time RIC can obtain network-side and / or terminal-side information from RAN nodes (such as CU, CU-CP, CU-UP, DU and / or RU) and / or terminals. This information can be used as training data or reasoning data, and the reasoning results can be submitted to the RAN node and / or terminal. Optionally, the reasoning results can be exchanged between the CU and the DU, and / or between the DU and the RU. For example, the non-real-time RIC submits the reasoning results to the DU, and the DU sends it to the RU.

[0137] The near real-time RIC and non-real-time RIC may also be separately configured as a network element. Optionally, the near real-time RIC and non-real-time RIC may also be part of other devices. For example, the near real-time RIC is configured in a RAN node (e.g., a CU or DU), while the non-real-time RIC is configured in an OAM, a cloud server, a core network device, or other network device.

[0138] FIG3 is a schematic diagram of a communication system applicable to the communication method of an embodiment of the present application. As shown in FIG3 , the communication system 100 may include at least one network device, such as the network device 110 shown in FIG3 ; the communication system 100 may also include at least one terminal device, such as the terminal device 120 and the terminal device 130 shown in FIG3 . The network device 110 and the terminal device (such as the terminal device 120 and the terminal device 130) can communicate via a wireless link. The communication devices in the communication system, for example, the network device 110 and the terminal device 120, can communicate via multi-antenna technology.

[0139] Figure 4 is a schematic diagram of another communication system applicable to the communication method of an embodiment of the present application. Compared to the communication system 100 shown in Figure 3, the communication system 200 shown in Figure 4 also includes an AI network element 140. AI network element 140 is used to perform AI-related operations, such as constructing a training dataset or training an AI model.

[0140] In one possible implementation, the network device 110 may send data related to the training of the AI ​​model to the AI ​​network element 140, which constructs a training data set and trains the AI ​​model. For example, the data related to the training of the AI ​​model may include data reported by the terminal device. The AI ​​network element 140 may send the results of the operations related to the AI ​​model to the network device 110, and forward them to the terminal device through the network device 110. For example, the results of the operations related to the AI ​​model may include at least one of the following: an AI model that has completed training, an evaluation result or a test result of the model, etc. Exemplarily, a portion of the trained AI model may be deployed on the network device 110, and another portion may be deployed on the terminal device. Alternatively, the trained AI model may be deployed on the network device 110. Alternatively, the trained AI model may be deployed on the terminal device.

[0141] It should be understood that Figure 4 illustrates only the example of a direct connection between AI network element 140 and network device 110. In other scenarios, AI network element 140 may also be connected to a terminal device. Alternatively, AI network element 140 may be connected to both network device 110 and a terminal device simultaneously. Alternatively, AI network element 140 may be connected to network device 110 through a third-party network element. This embodiment of the present application does not limit the connection relationship between the AI ​​network element and other network elements.

[0142] The AI ​​network element 140 may also be provided as a module in a network device and / or a terminal device, for example, in the network device 110 or the terminal device shown in FIG3 .

[0143] It should be noted that Figures 3 and 4 are simplified schematic diagrams for ease of understanding. For example, the communication system may also include other devices, such as wireless relay devices and / or wireless backhaul devices, which are not shown in Figures 3 and 4. In actual applications, the communication system may include multiple network devices and multiple terminal devices. The embodiments of the present application do not limit the number of network devices and terminal devices included in the communication system.

[0144] To facilitate understanding of the solutions of the embodiments of the present application, the terms that may be involved in the embodiments of the present application are explained below.

[0145] (1) AI model:

[0146] An AI model is an algorithm or computer program that implements AI functionality. It represents the mapping between the model's inputs and outputs. AI models can be neural networks, linear regression models, decision tree models, support vector machines (SVMs), Bayesian networks, Q-learning models, or other machine learning (ML) models.

[0147] (2) Two-end model:

[0148] The two-end model can also be called a bilateral model, collaborative model, dual model, or two-side model. A two-end model is a model composed of multiple sub-models. The sub-models that make up the model must match each other. These sub-models can be deployed on different nodes.

[0149] The sub-models involved in the embodiments of the present application include an encoder (encoder) for compressing information and a decoder (decoder) for restoring compressed information. The encoder and decoder are matched and used, and it can be understood that the encoder and decoder are matching models. An encoder may include one or more AI models, and the decoder matched by the encoder also includes one or more AI models. The number of AI models included in the matching encoder and decoder is the same and one-to-one corresponding. For example, the encoder and decoder can be deployed on the terminal device and network device respectively.

[0150] In one possible design, a matched encoder and decoder can be implemented as two components of a single autoencoder (AE). An autoencoder is a type of neural network that uses unsupervised learning. Its characteristic is that it uses input data as labeled data. Therefore, an autoencoder can also be understood as a self-supervised learning neural network. An autoencoder can be used for data compression and recovery. An AE model, in which the encoder and decoder are deployed on different nodes, is a typical bilateral model.

[0151] Figure 5 is a schematic block diagram of an autoencoder. As shown in Figure 5, the encoder in the autoencoder can compress (encode) data A to obtain data B; the decoder in the autoencoder can decompress (decode) data B to recover data A. Alternatively, the decoder can be understood as the inverse operation of the encoder.

[0152] Alternatively, the AI ​​model in the embodiment of the present application may be a single-ended model, which may be deployed on a terminal device or a network device.

[0153] (3) Neural network (NN):

[0154] Neural networks are a specific implementation of AI or machine learning. According to the universal approximation theorem, neural networks can theoretically approximate any continuous function, giving them the ability to learn arbitrary mappings.

[0155] A neural network can be composed of neural units, which can be a computational unit that takes xs and an intercept 1 as input. A neural network is formed by connecting many of these single neural units, meaning that the output of one neural unit can be the input of another. The input of each neural unit can be connected to the local receptive field of the previous layer to extract features from that local receptive field, which can be an area consisting of several neural units.

[0156] Taking the AI ​​model type as a neural network as an example, the AI ​​model involved in this disclosure can be a deep neural network (DNN). Depending on the network construction method, DNN can include feedforward neural networks (FNN), convolutional neural networks (CNN), and recurrent neural networks (RNN).

[0157] (4) Channel state information (CSI):

[0158] In communication systems (e.g., LTE or NR), network equipment needs to determine the resources, modulation and coding scheme (MCS), and precoding configurations for the downlink data channel of the terminal device based on CSI. It can be understood that CSI is a type of channel information that can reflect channel characteristics and channel quality.

[0159] CSI measurement refers to the receiver solving the channel information based on the reference signal sent by the transmitter, that is, estimating the channel information using the channel estimation method. Exemplarily, the reference signal may include one or more of a channel state information reference signal (CSI-RS), a synchronization signal / physical broadcast channel block (SSB), a sounding reference signal (SRS), or a demodulation reference signal (DMRS). One or more of CSI-RS, SSB, and DMRS can be used to measure downlink CSI. SRS and / or DMRS can be used to measure uplink CSI.

[0160] Taking FDD communication scenarios as an example, in FDD communication scenarios, because uplink and downlink channels are not reciprocal or cannot be guaranteed, network equipment typically transmits a downlink reference signal to the terminal device. The terminal device performs channel and interference measurements based on the received downlink reference signal to estimate the downlink CSI. The terminal device generates a CSI report based on a protocol predefined method or a network device configuration method and feeds it back to the network device to obtain the downlink CSI.

[0161] Exemplarily, the CSI may include at least one of the following: channel quality indication (CQI), precoding matrix indicator (PMI), rank indicator (RI), CSI-RS resource indicator (CRI), layer indicator (LI), reference signal receiving power (RSRP), or signal to interference plus noise ratio (SINR). The signal to interference plus noise ratio may also be referred to as the signal to interference plus noise ratio.

[0162] The RI indicates the number of downlink transmission layers recommended by the terminal device, the CQI indicates the modulation and coding scheme supported by the current channel conditions as determined by the terminal device, and the PMI indicates the precoding recommended by the terminal device. The number of precoding layers indicated by the PMI corresponds to the RI.

[0163] It is understood that the RI, CQI, and PMI indicated in the above CSI report are only recommended values ​​for the terminal device, and the network device may perform downlink transmission according to part or all of the information indicated in the CSI report. Alternatively, the network device may not perform downlink transmission according to the information indicated in the CSI report.

[0164] (5) Weight:

[0165] The precoding matrix / vector required for precoding on the network device side when the network device performs downlink transmission. The weights in this application are divided into two weights, one weight is the outer weight (hereinafter also referred to as the dimensionality reduction weight), which is used for the network device side to perform port dimensionality reduction when measuring downlink channel information; the other weight is the inner weight, which is the equivalent channel after the terminal device measures the port dimensionality reduction, and after completing the compression feedback of the channel information, the network device restores the compression feedback amount to obtain the weight. The network device can determine the final precoding matrix for downlink transmission based on the outer weight and the inner weight. For example, the network device calculates the precoding matrix C ultimately used for downlink transmission based on a two-level codebook. Specifically, the outer weight is A, the inner weight is B, and the precoding matrix C ultimately used is determined by C=A*B, where * represents multiplication.

[0166] It can be understood that the weights and precoding matrices in this application can be replaced with each other.

[0167] (6) Port dimensionality reduction:

[0168] As the size of the base station array increases, the dimension of the CSI measured and fed back by the terminal increases accordingly, resulting in an increased burden on measurement resource overhead and feedback overhead. Therefore, when the network device performs downlink channel information measurement, the network device digitally precodes the reference signals of a few antenna ports through the outer weights, so that the network device can use a few reference channel resources (corresponding to the number of antenna ports) to measure the channels under the majority of physical antennas. The dimensionality reduction in this application refers to port dimensionality reduction.

[0169] (7) Antenna port:

[0170] An antenna port is a logical concept and does not directly correspond to a physical antenna. An antenna port is typically associated with a reference signal and can be understood as a transceiver interface on the channel through which the reference signal travels. For low frequencies, an antenna port may correspond to one or more physical antennas (a physical antenna can refer to a digital port or an antenna array element). These physical antennas jointly transmit the reference signal, and the receiver can treat them as a whole without distinguishing between the physical antennas.

[0171] (8) AI model design:

[0172] The design of an AI model primarily involves data collection (e.g., collecting training data and / or inference data), model training, and model inference. Furthermore, it can also include the application of inference results.

[0173] It is understandable that a communication system may include network elements with artificial intelligence capabilities. The above-mentioned AI model design-related steps can be performed by one or more network elements with artificial intelligence capabilities. In one possible design, AI functions (such as AI modules or AI entities) can be configured in existing network elements in the communication system to implement AI-related operations, such as AI model training and / or inference. For example, the existing network element can be a network device or a terminal device. Alternatively, in another possible design, an independent network element can be introduced into the communication system to perform AI-related operations, such as training an AI model. The independent network element can be referred to as an AI network element (i.e., an AI entity or an AI node), and the embodiments of the present application are not limited to this name. For example, the AI ​​network element can be directly connected to the network equipment in the communication system, or it can be indirectly connected to the network equipment through a third-party network element. The third-party network element can be a core network element such as an authentication management function (AMF) network element, a user plane function (UPF) network element, an operation administration and maintenance (OAM) network element, a cloud server, or other network element, without limitation. Exemplarily, the independent AI network element can be deployed on one or more of the network device side, the terminal device side, or the core network side. Optionally, it can be deployed on a server, such as a cloud server, or an over-the-top (OTT) device. Exemplarily, an AI network element 140 is introduced into the communication system shown in FIG4 . Optionally, the server or OTT device can be a third-party network element. In addition, the third-party network element can also be referred to as a third-party device, that is, a device different from the network device or the terminal device.

[0174] (9) Training data set and inference data:

[0175] The training data set is used to train the AI ​​model. The training data set may include the input of the AI ​​model, or the input and target output of the AI ​​model. Among them, the training data set includes one or more training data. The training data may include training samples input to the AI ​​model, or may include the target output of the AI ​​model. Among them, the target output may also be referred to as a label, sample label, or labeled sample. The label is the true value. In the field of machine learning, the ground truth usually refers to data that is considered to be accurate or real data.

[0176] In the communications field, training datasets can include simulated data collected through simulation platforms, experimental data collected in experimental scenarios, or measured data collected in actual communication networks. Because the geographical environments and channel conditions in which data are generated vary, such as indoor and outdoor locations, mobile speeds, frequency bands, or antenna configurations, the collected data can be categorized during acquisition. For example, data with the same channel propagation environment and antenna configuration can be grouped together.

[0177] Model training essentially involves learning certain characteristics from training data. When training an AI model (such as a neural network), the goal is to ensure that the model's output is as close as possible to the desired predicted value. This is done by comparing the network's predictions with the desired target values. The weight vectors of each layer of the AI ​​model are then updated based on the difference between the two. (Of course, before the first update, there's usually an initialization process, which pre-configures the parameters for each layer of the AI ​​model.) For example, if the network's prediction is too high, the weight vectors are adjusted to predict a lower value. This adjustment is repeated until the AI ​​model predicts the desired target value, or a value very close to it. Therefore, it's necessary to predefine how to compare the difference between the predicted and target values. This is known as the loss function, or objective function. These are important equations used to measure the difference between the predicted and target values. For example, a higher loss function indicates a greater difference. Therefore, training an AI model becomes a process of minimizing this loss, keeping the loss function below a threshold or ensuring that the loss function meets the target requirement. For example, the AI ​​model is a neural network, and adjusting the model parameters of the neural network includes adjusting at least one of the following parameters: the number of layers, width, weights of neurons, or parameters in the activation function of neurons of the neural network.

[0178] Inference data can be used as input to a trained AI model for inference. During the inference process, the inference data is input into the AI ​​model, and the corresponding output is the inference result.

[0179] (10) Forward propagation and back propagation:

[0180] Figure 6 is a schematic diagram of the forward propagation and back propagation of a neural network. During the training process of an AI model, there is forward propagation and back propagation of data flows. As shown in Figure 6, the AI ​​module refers to a neural network layer or an AI model, where forward propagation refers to the neural network passing the intermediate calculation results of the previous neural network layer or the previous AI model to the next neural network layer or the next AI model to complete the intermediate calculation of the final output result under the parameter. Back propagation refers to the method by which a neural network calculates gradients. That is, after calculating the error between the final output result and the label (that is, the true result), in order to calculate the updated gradient of the network parameters to be trained, the neural network will calculate the gradient layer by layer from back to front through back propagation. The principle is similar to the chain rule in calculus. The error of the latter layer is calculated and propagated to the previous layer. The previous layer calculates the gradient of the network layer based on the current error.

[0181] The introduction of AI technology into wireless communication networks has resulted in a CSI feedback method based on AI models. Terminal devices use AI models to compress and feedback CSI, and network equipment uses AI models to recover the compressed CSI. AI-based CSI feedback transmits a sequence (such as a bit sequence), which reduces the overhead of traditional CSI feedback.

[0182] The following describes the current AI-based CSI feedback process after port dimensionality reduction, using Figure 7. In this process, the number of physical antennas is N, the number of antenna ports after dimensionality reduction is M, M is less than N, and both M and N are positive integers. The process includes the following steps.

[0183] P100: The terminal device sends an uplink reference signal (such as SRS) to measure the uplink channel information, and the network device receives the uplink reference signal to obtain the uplink channel information at this moment.

[0184] P101: The network device receives uplink channel information and calculates a dimensionality reduction weight w0 (wherein the dimension of w0 is N×M) based on the received uplink channel information. For example, the uplink channel information, i.e., the uplink channel response, and the second-order statistic Z of the uplink channel information calculated based on the historical uplink channel response are used. The dimensionality reduction weight w0 is then calculated based on the second-order statistic Z in the following manner. After obtaining the dimensionality reduction weight w0, the network device performs port dimensionality reduction based on the dimensionality reduction weight, i.e., digitally precoding the reference signals on the M ports using the dimensionality reduction weight to achieve measurement of the equivalent downlink channel response h under N physical antennas using M downlink reference signal (e.g., CSI-RS) resources. DL =H DL *w0, where HDL is the response of the actual downlink channel under N physical antennas (where H DL The spatial dimension of is N). Two current methods for calculating dimensionality reduction weights are given below as examples.

[0185] In one method, the network device calculates the projection energy of the second-order statistic Z under the discrete Fourier transform (DFT) codebook based on the DFT codebook, and selects the first M DFT bases {l0,l1,...,l M-1}(where l i is a 1×N vector) as the dimensionality reduction weight w0.

[0186] In another method, the network device performs singular value decomposition (SVD) on the second-order statistic Z and selects the first M feature directions {l0,l1,...,l M-1}(where l i is a 1×N vector) as the dimensionality reduction weight w0.

[0187] The starting point of the above methods is to minimize the loss of the energy of the equivalent downlink channel after dimensionality reduction compared to the energy of the real downlink channel. The energy of the equivalent downlink channel is based on the response h of the equivalent downlink channel. DL The energy of the real downlink channel is determined based on the response H of the real downlink channel. DL For example, the relationship between energy E and response H is E=||H||2, where ||H||2 represents the 2-norm of the matrix H.

[0188] P102: The terminal device measures M downlink reference signals and obtains the corresponding equivalent channel information H DL After completing the measurement, the terminal device sends a downlink measurement report to the network device, which includes the CSI.

[0189] The AI-based CSI feedback method involves designing and training an autoencoder for network or terminal devices to perform space-frequency dual-domain CSI compression, replacing the codebook in the 3rd Generation Partnership Project (3GPP) Release 16 protocol. For example, the compression amount of the encoder output and decoder input in the autoencoder is equivalent to the compression amount of the precoding matrix of the equivalent downlink channel. This compression amount is used in the feedback process to replace the precoding matrix feedback, reducing feedback overhead.

[0190] P103: The network device receives the CSI in the downlink measurement report and obtains the compressed feedback of the equivalent channel precoding matrix from it. It restores the inner layer weight v (i.e., the equivalent channel precoding matrix before compression) in the following way to complete the reconstruction of the downlink precoding matrix.

[0191] Since the compressed feedback is obtained by the terminal device using the encoder in the autoencoder, the network device inputs the compressed feedback into the autoencoder's decoder, which outputs the inner layer weights v. Finally, based on the two-stage codebook structure, the network device calculates the final downlink precoding matrix V for N physical antennas using the formula V = v * w0.

[0192] Since the above-mentioned dimensionality reduction weights are calculated using an orthogonal codebook or orthogonal decomposition, only the energy loss of the equivalent channel after dimensionality reduction is minimized, which is independent of the compression scheme. As a result, the compressibility of the measured equivalent channel information is not high. Therefore, how to further reduce the feedback overhead has become an urgent problem to be solved.

[0193] In view of this, the present application proposes a communication method that can effectively solve the above technical problems. The communication method is described in detail below.

[0194] FIG8 is a schematic flow chart of a communication method 800 proposed in this application. The method includes the following steps.

[0195] S810. The network device digitally precodes M reference signals corresponding to the M antenna ports based on a dimensionality reduction weight. The dimensionality reduction weight is used to measure the equivalent downlink channels of N physical antennas using the M reference signals, where N is a positive integer and M is a positive integer less than N. The dimensionality reduction weight is determined based on a dimensionality reduction model.

[0196] Optionally, the model in this application is an AI model, a neural network model, a machine learning model, etc., and this application does not impose any restrictions.

[0197] The network device digitally precodes M reference signals corresponding to M antenna ports based on the dimensionality reduction weights. That is, the network device implements port dimensionality reduction based on the dimensionality reduction weights, and the dimension of the dimensionality reduction weights is N*M. For example, the reference signal can be CSI-RS, SSB, or DMRS, which is not limited in this application.

[0198] Optionally, the dimensionality reduction weight is output information obtained by using a priori channel information as input to the dimensionality reduction model. For example, the a priori channel information is determined based on at least one of the following: currently measured uplink channel information, historical uplink channel information, historical downlink channel information, or perceived channel information. For example, the a priori channel information can be currently measured uplink channel information, or can be calculated based on previously measured uplink channel information and historical uplink channel information.

[0199] Optionally, the dimensionality reduction model can be implemented as a module within a network device or AI network element, such as a third-party network element. For example, the AI ​​network element can be a core network element such as an AMF network element or a UPF network element, or it can be an OAM, cloud server, OTT, or other network element, without limitation.

[0200] In example 1, the dimensionality reduction model is located in a network device, and the network device can input channel prior information into the dimensionality reduction model to obtain a dimensionality reduction weight (ie, the output of the dimensionality reduction model).

[0201] In example two, the dimensionality reduction model is located in an AI network element, such as a third-party network element. The network device can send the channel prior information and the dimensionality reduction model information to the AI ​​network element. The AI ​​network element inputs the channel prior information into the dimensionality reduction model to obtain the dimensionality reduction weight. Afterwards, the AI ​​network element sends the dimensionality reduction weight to the network device.

[0202] S820: The network device sends M reference signals to the terminal device. Correspondingly, the terminal device receives the M reference signals from the network device.

[0203] S830: The terminal device obtains first downlink channel information based on M reference signals.

[0204] It is understood that the present application does not impose any specific limitation on the type of information specifically indicated by the first downlink channel information.

[0205] For example, the first downlink channel information indicates the response of the equivalent downlink channel. DL is the actual downlink channel response under N physical antennas, the spatial dimension is N, w0 is the dimensionality reduction weight, then the equivalent downlink channel response obtained by the network device measuring M reference signals is h DL =H DL *w0, the spatial dimension is M.

[0206] For example, the first downlink channel information indicates the precoding matrix (also called weight) corresponding to the response of the equivalent downlink channel. For example, if the downlink channel response obtained by the network device measuring M reference signals is the above h DL , then the equivalent channel h can be obtained by SVD decomposition DL The weight v0 under the , spatial dimension is M.

[0207] S840: The network device sends first indication information, where the first indication information indicates a compression model that matches the dimensionality reduction model. Correspondingly, the terminal device receives the first indication information.

[0208] It is understandable that after obtaining the first downlink channel information, the terminal device feeds the obtained first downlink channel information back to the network device. However, due to feedback overhead considerations, the terminal device does not directly feed back such a high-dimensional tensor to the network device. Instead, it sends the compressed feedback amount of the first downlink channel information after it has been compressed by the compression model to the network device. Because this method introduces a dimensionality reduction model, the dimensionality reduction method corresponding to the dimensionality reduction model is associated with the compression method. Therefore, the network device indicates to the terminal device the compression model that matches the dimensionality reduction model to achieve better compression performance.

[0209] Furthermore, in scenarios where dimensionality reduction and compression are not associated (for example, scenarios where port dimensionality reduction is not performed, or scenarios where dimensionality reduction weights are orthogonal), since the terminal device only needs to feed back the precoding matrix of the equivalent downlink channel to complete the reconstruction of the real channel precoding matrix, the input of the compression model is consistent with the output of the reconstruction model. At this time, the compression model is used to compress the downlink channel information, and the reconstruction model is used to decompress the output information of the compression model. The reconstruction model and the compression model can be matched; in scenarios where dimensionality reduction and compression are associated (for example, the solution proposed in this application), the dimensionality reduction weights do not restrict orthogonality, the compression method, the reconstruction method and the dimensionality reduction method are strongly correlated, and the input of the compression model is inconsistent with the output of the reconstruction model. Therefore, the operating principles of the compression model and the reconstruction model are essentially different from those in scenarios where dimensionality reduction and compression are not associated. Therefore, in the embodiment of the present application, the first indication information indicates a compression model that matches the dimensionality reduction model. The compression model is used for feature extraction and compression of the inner weights of the first downlink channel information, which can not only improve the compression performance, but also further guarantee the decompression performance of the reconstruction model.

[0210] For example, the dimensionality reduction model, compression model, and reconstruction model in the embodiments of the present application are all related to the number of antenna ports M, the number of physical antennas N, and channel prior information.

[0211] Several possible specific implementations of the first indication information are given below.

[0212] Implementation method 1: the first indication information indicates an identifier (ID) of the compression model.

[0213] Implementation method 2: The first indication information indicates the dimensionality reduction model ID and / or the reconstruction model ID.

[0214] For example, after the joint training of the compression model, dimensionality reduction model and reconstruction model is completed, the compression model ID, dimensionality reduction model ID and reconstruction model ID can be predefined and the three model IDs can be associated. Then, the network device can implicitly indicate the compression model by indicating the dimensionality reduction model ID and / or reconstruction model ID.

[0215] Implementation method three: the first indication information indicates a dimensionality reduction method or a compression method.

[0216] It can be understood that the dimensionality reduction model can be determined based on the dimensionality reduction method, and the compression model can be determined based on the dimensionality reduction model. Therefore, the network device can implicitly indicate the compression model by indicating the dimensionality reduction method.

[0217] Similarly, the compression model can be determined based on the compression method, so the network device can implicitly indicate the compression model by indicating the compression method.

[0218] It should be noted that the dimensionality reduction model, compression model and reconstruction model are named based on the main functions of each model in this application. It can be understood that this application does not specifically limit the names of the three models. The three models may also have other functions in addition to the functions corresponding to their names. For example, the compression model can also realize the quantization function, and / or the reconstruction model can also realize the dequantization function. This application does not limit it.

[0219] S850: The terminal device sends a first sequence to the network device, where the first sequence is a compression feedback amount obtained by using the first downlink channel information as an input of the compression model. Correspondingly, the network device receives the first sequence from the terminal device.

[0220] Optionally, the compression model can be located as a module in the terminal device or AI network element, such as a third-party network element.

[0221] In example 1, the compression model is located in the terminal device, and the terminal device can input the first downlink channel information into the compression model to obtain a first sequence (ie, the output of the compression model).

[0222] Example 2: The compression model is located in the AI ​​network element. The terminal device can send the first downlink channel information and the compression model information to the AI ​​network element. The AI ​​network element inputs the first downlink channel information into the indicated compression model to obtain a first sequence. After that, the AI ​​network element sends the first sequence to the terminal device.

[0223] S860: The network device obtains second downlink channel information. The second downlink channel information is information obtained by using the first sequence as input to a reconstruction model. The reconstruction model matches the dimensionality reduction model. The second downlink channel is used to determine downlink channel information of N physical antennas.

[0224] Optionally, the compression model may be an encoder, and correspondingly, the reconstruction model may be a decoder.

[0225] Optionally, the reconstruction model can be located as a module in a network device or AI network element, such as a third-party network element.

[0226] In example 1, the reconstruction model is located in the network device, and the network device can input the first sequence into the reconstruction model to obtain the second downlink channel information (ie, the output of the reconstruction model).

[0227] In example two, the reconstruction model is located in the AI ​​network element. The network device can send the first downlink channel information and the reconstruction model information to the AI ​​network element. The AI ​​network element inputs the first sequence into the reconstruction model to obtain the second downlink channel information. After that, the AI ​​network element sends the second downlink channel information to the network device.

[0228] Optionally, if the dimensionality reduction model, compression model, and reconstruction model are located in an AI network element, such as a third-party network element, the third-party network elements where the models are located may be the same or different, and this application does not impose any restrictions on this. In one possible implementation, the dimensionality reduction model and the reconstruction model may be located in the third-party network element #1, and the compression model may be located in the third-party network element #2. The inputs and outputs corresponding to each model are described above and will not be repeated here.

[0229] Optionally, the second downlink channel is used to determine downlink channel information of N physical antennas. Specifically, the second downlink channel information is used to determine a downlink precoding matrix V of the N physical antennas. Possible second downlink channel information is described below with examples.

[0230] For example, the input of the reconstruction model (i.e., the first sequence) is the response h of the equivalent downlink channel. DL The compression amount, the output (ie, the second downlink channel information) can be the channel response h DL ′.

[0231] For example 2, the input of the reconstruction model (ie, the first sequence) is the compression amount of the precoding matrix v0 corresponding to the response of the equivalent downlink channel, and the output (ie, the second downlink channel information) is the inner layer weight v.

[0232] Based on the above example 1 and example 2, if the network device determines V, then for example, in a possible implementation, the network device determines h based on example 1. DL ′ and the dimensionality reduction weight w0 to obtain V, or, based on the v and the dimensionality reduction weight w0 in Example 2, V is calculated. For example, the network device calculates V by V=v*w0 based on the existing two-level codebook structure (i.e., the structure of the outer layer weight and the inner layer weight mentioned above); In another implementation, the network device can determine V through model #1, for example, the network device inputs the dimensionality reduction weight w0 and the second downlink channel information (h′) into the model #1. DL or v), calculate V.

[0233] It can be understood that model #1 is a model that has the function of calculating V. Specifically, the specific data of the input parameters required by model #1 to calculate V is input into model #1, and the output data of model #1 is obtained, that is, the corresponding V is obtained. For example, model #1 is an AI model, a neural network model, or a machine learning model, which is not limited.

[0234] Example 3: The input of the reconstruction model (i.e. the first sequence) is h in Example 1. DL The compression amount and dimensionality reduction weight w0, or the input is the compression amount and dimensionality reduction weight w0 of v0 in Example 2, and the output of the reconstruction model (ie, the second downlink channel information) is V.

[0235] Optionally, the second downlink channel is used to determine the downlink channel information of N physical antennas, which may be specifically: the second downlink channel information is used to determine the downlink channel response H of the N physical antennas. DL The following is an example of possible second downlink channel information.

[0236] For example, the input of the reconstruction model (i.e., the first sequence) is the response h of the equivalent downlink channel. DL The compression amount, the output (ie, the second downlink channel information) can be the channel response h DL ′.

[0237] Based on the above example 1, if the network device needs to determine H DL , then for example, in a possible implementation, the network device can be based on h DL ′ and the dimensionality reduction weight w0 are used to calculate H DL , such as network devices through H DL =w0*h DL Calculate H DL In another implementation, the network device can determine H through Model #2 DL For example, the network device inputs the dimensionality reduction weights w0 and h to model #2 DL ′, calculate H DL .

[0238] It can be understood that Model #2 is a model with calculated H DL The functional model, specifically, the model #2 calculates H DL The specific data of the required input parameters are input into model #2, and the output data of model #2 is obtained, that is, the corresponding H DL For example, model #2 is an AI model, a neural network model, or a machine learning model, without limitation.

[0239] In Example 2, the input of the reconstruction model (i.e. the first sequence) is h in Example 1. DL ′ and the dimensionality reduction weight w0, the output (i.e., the second downlink channel information) is H.

[0240] It can be seen that the method shown in FIG7 achieves port dimensionality reduction by means of an orthogonal codebook or orthogonal decomposition, and only considers the minimum loss of equivalent channel energy. Such a method does not take into account the compressibility of the channel after dimensionality reduction, resulting in poor compressibility of the equivalent channel information. In the method proposed in this application, the dimensionality reduction model, compression model, and reconstruction model are matching models. That is to say, the dimensionality reduction method and the compression method in this application are jointly optimized, and can achieve port dimensionality reduction while also considering the compressibility of the channel after dimensionality reduction. Therefore, compared with the method shown in FIG7, the method proposed in this application can improve the compressibility of the equivalent channel information, thereby achieving the effect of further reducing feedback overhead.

[0241] Based on the above description, to achieve the aforementioned effect of reducing feedback overhead while ensuring decoding (also known as decompression) performance, the dimensionality reduction model, compression model, and reconstruction model are matched models. Therefore, before the corresponding models are actually applied, the three models can be jointly trained to obtain three matching models. The following describes the joint training process of the three models in detail.

[0242] Figure 9 is a schematic flow chart of a method 900 for model training proposed in the present application. It is understandable that the present method 900 can be used independently or in combination with the embodiment corresponding to the aforementioned Figure 8, and is not limited here. In this method, the first device and the second device can deploy the model locally, transmit the intermediate layer calculation results and gradients in real time, and complete the training of the bilateral model. Among them, the dimensionality reduction model and the reconstruction model are deployed on the first device side, and the compression model is deployed on the second device side. Optionally, the first device can be a network device or an AI network element #1, such as a third-party network element #1, and the second device can be a terminal device or an AI network element #2, such as a third-party network element #2. The method includes the following steps.

[0243] S910. The first device obtains a training data set, where the training data set includes channel prior information and downlink channel information #1. Downlink channel information #1 is actual downlink channel information of N physical antennas, where N is a positive integer.

[0244] Before jointly training the three models, the network device collects training data. The training data used to train the AI ​​model includes training samples and sample labels. Channel prior information can be considered as training samples, and downlink channel information #1 can be considered as sample labels (i.e., true values). For more information about channel prior information, see S810 and will not be repeated here.

[0245] For ease of description, the following steps use the channel prior information as the uplink channel response H UL , downlink channel information #1 is the actual downlink channel response H of N physical antennas DL For example, during the model training process, HUL Used as input for the dimensionality reduction model, H DL The input v0 used to calculate the compression model (equivalent downlink channel response H DL w0 after SVD operation) and the calculation of the final label V (i.e., the precoding matrix of N physical antennas).

[0246] For example, the network device can obtain the uplink channel response H by measuring UL and complete the corresponding downlink channel response H through the existing air interface process DL However, it should be noted that when collecting downlink channel responses, each downlink channel response H DL The sample should be indicated by the network device with each upstream channel response H UL The samples correspond to each other, for example, H UL With H DL Located in the same or adjacent bandwidth (frequency domain position interval is less than K RBs), and the time domain interval between the two should be less than t time slots, that is, so that H UL With H DL The reflected channel response has the reciprocity that meets the requirements, such as the uplink and downlink reciprocity in the delay angle domain.

[0247] S920: The first device sends downlink channel information #2 to the second device. Downlink channel information #2 is determined based on downlink channel information #1 and a dimensionality reduction weight. The dimensionality reduction weight is output information obtained by using prior channel information as input to a dimensionality reduction model. The dimensionality reduction weight is used to digitally precode M reference signals corresponding to M antenna ports to measure the equivalent downlink channels of N physical antennas, where M is a positive integer less than N. Correspondingly, the second device receives downlink channel information #2 from the first device.

[0248] Optionally, before S920, the method further includes: the first device responds to the uplink channel H UL The dimensionality reduction model is input and the dimensionality reduction weight w0 is output. Then, the first device determines downlink channel information #2 (ie, equivalent downlink channel information) based on the dimensionality reduction weight w0.

[0249] For example, downlink channel information #2 is the response H of the equivalent downlink channel DL w0;

[0250] For example, downlink channel information #2 is the response H of the equivalent downlink channel DL w0 is the equivalent channel weight v0 after SVD.

[0251] S930: The second device sends a first sequence to the first device, where the first sequence is output information obtained by using the downlink channel information #2 as input to the compression model. Correspondingly, the first device receives the first sequence from the second device.

[0252] Optionally, before S930, the method further includes: the second device inputs the downlink channel information #2 into the compression model and outputs the first sequence.

[0253] For example, downlink channel information #2 is the response H of the equivalent downlink channel DL w0, then the first sequence can be the response H of the equivalent downlink channel DL The compression feedback amount of w0 may also be the compression feedback amount of the equivalent channel weight v0, without limitation.

[0254] The downlink channel information #2 and the first sequence can be regarded as intermediate calculation results of the forward propagation.

[0255] S940: The first device sends first gradient information to the second device. The first gradient information is input-side gradient information of the reconstruction model. The first gradient information is used to update model parameters of the compression model. The first gradient information is determined based on the first error information and the reconstruction model. The first error information is determined based on downlink channel information #3 and downlink channel information #1. Downlink channel information #3 is output information obtained by using the first sequence as input to the reconstruction model or calculated based on the output information of the reconstruction model. The first error information is used to update model parameters of the reconstruction model. Correspondingly, the second device receives the first gradient information from the first device.

[0256] Optionally, before S940, the method further includes: the first device inputting the first sequence into the reconstruction model to obtain output information of the reconstruction model (the output information is downlink channel information #3 or is used to determine downlink channel information #3); then, the first device determining first error information based on downlink channel information #1 and downlink channel information #3. Then, the first device performs a backpropagation calculation based on the first error information and the reconstruction model to obtain gradient information on the input side of the reconstruction model (i.e., first gradient information).

[0257] Among them, the first error information is determined based on the downlink channel information #3 and the downlink channel information #1, which can be understood as: the downlink channel information #3 is the actual output information of the reconstructed model in the current training round obtained based on the channel prior information in the training data set, or the information calculated based on the actual output information. For example, the downlink channel information #3 is the channel response (or channel weight) actually obtained during the training process. Correspondingly, the first device can determine the expected channel response (or channel weight) based on the downlink information #1 in the training data set. Then the first device can determine the first error information based on the difference between the expected channel response (or channel weight) and the actual channel response (or channel weight).

[0258] The following is an example with specific parameters. For example, if the first sequence is the compressed feedback amount of the equivalent channel weight v0, the first device inputs v0 into the reconstruction model, obtains the output information of the reconstruction model, and determines the downlink weights of N physical antennas based on the output information of the reconstruction model. (i.e. downlink channel information #3), where the output information of the reconstructed AI model can be Alternatively, the output information of the reconstructed AI model can also be the inner layer weight v, and then the first device can calculate based on the inner layer weight v and the dimensionality reduction weight w0 Afterwards, the first device responds to the real downlink channel H DL (ie downlink channel information #1) performs SVD to obtain the actual weights V of the N physical antennas, and calculates V and Based on the calculated error, the first device calculates the input side gradient of the reconstructed model (i.e., the first gradient information) through back propagation and transmits it to the second device.

[0259] At S950, the second device sends second gradient information to the first device. The second gradient information is input-side gradient information of the compression model. The second gradient information is determined based on the first gradient information and the compression model. The second gradient information is used to update model parameters of the dimensionality reduction model. In response, the first device receives the second gradient information from the second device.

[0260] Before S950 , the method further includes: the second device performs back propagation calculation based on the first gradient information and the compression model to obtain gradient information on the input side of the compression model (ie, second gradient information).

[0261] S960, the first device updates the model parameters of the reconstruction model based on the first error information, and updates the model parameters of the dimensionality reduction model based on the second gradient information; the second device updates the model parameters of the compression model based on the first gradient information.

[0262] For example, the model parameters may include at least one of the following parameters: a weight of a neuron, or a bias, etc.

[0263] Optionally, the method further includes: S970, when a termination condition of the model training is met, the first device determines to terminate the model training.

[0264] For example, the termination condition of the model training may be: the error indicated by the first error information is less than or equal to a first threshold, or the number of model training rounds is equal to a second threshold.

[0265] The above S920 to S960 can be regarded as a round of model training. After each round of model training, the first device and the second device need to update the model parameters of their respective locally deployed models. After the update, they continue with the next round of model training until the model training is terminated.

[0266] For example, when the first device determines to terminate model training, the first device can send relevant instructions to the second device to inform the second device to terminate model training.

[0267] For example, the above threshold (first threshold or second threshold) can be predefined, or can be set by the first device itself, or can be indicated by other devices (such as the second device), and this application does not limit this.

[0268] It can be understood that the dimensionality reduction model, compression model, and reconstruction model described in method 800 shown in Figure 8 can be the dimensionality reduction model, compression model, and reconstruction model obtained after the model training is completed in method 900. For example, if the first device and the second device are AI network element #1 and AI network element #2, such as third-party network element #1 and third-party network element #2, and the dimensionality reduction model and reconstruction model will be deployed on the network device side, and the compression model will be deployed on the terminal device side, then the first device will send the trained (i.e., completed) dimensionality reduction model and reconstruction model, or the training data set corresponding to the trained model (dimensionality reduction model and reconstruction model) to the network device, and the second device will send the trained compression model or the training data set corresponding to the trained compression model to the terminal device. Here, the second device sends the training data of the compression model to the terminal device as an example. After the above three models are trained, the three trained models are obtained. The second device may not directly send the trained compression model, but instead send the training data set #1 corresponding to the trained compression model to the terminal device. The terminal device trains the corresponding compression model based on the received training data set #1. The training dataset #1 corresponding to the trained compression model may include information A and information B, where information A is output information obtained by using the channel prior information in the training dataset in S910 as input to the trained dimensionality reduction model, and information B is output information obtained by using information A as input to the trained compression model. That is, information A and information B in training dataset #1 are the input information and output information of the trained compression model.

[0269] The above describes in detail the model training method 900 provided in this application. While traditional model training methods only support dual-end two-model training, method 900 supports dual-end multi-model joint training. Based on method 900, the dimensionality reduction model, compression model, and reconstruction model can be jointly trained and optimized, further reducing the channel information feedback overhead.

[0270] The following is an introduction to another model training method proposed in this application. The main difference from method 900 is that in method 900, the two ends jointly train three models, while in this method, one side (for example, the first device) jointly trains three models. Afterwards, if one or more of the three models, such as model A, will be deployed on another device side (for example, the second device side), the first device can send the trained model A or the training data set corresponding to the trained model A to the second device, and the second device completes the model deployment or performs secondary training based on the received model A or training data set. The following is a detailed explanation in conjunction with Figure 10.

[0271] Figure 10 is a schematic flow chart of a model training method 1000 proposed in this application. In this method, the first device can complete the training of the unilateral model by locally deploying the model and transmitting the intermediate layer calculation results and gradients in real time. Among them, the dimensionality reduction model, compression model, and reconstruction model are all deployed on the first device side. Optionally, the first device can be a terminal device, a network device, or an AI network element #1, such as a third-party network element #1. The method includes the following steps.

[0272] S1010: The first device obtains a first training data set, where the first training data set includes channel prior information and downlink channel information #1. Downlink channel information #1 is actual downlink channel information of N physical antennas, where N is a positive integer.

[0273] For S1010, please refer to the description in S910 and will not be repeated here.

[0274] It can be understood that the training method of the unilateral joint training of method 1000 is consistent with the training method of the dual-end joint training of method 900. For more specific training process, please refer to the description of method 900.

[0275] S1020, the first device jointly trains a dimensionality reduction model, a compression model, and a reconstruction model, wherein the input of the dimensionality reduction model is the channel prior information, and the output is the dimensionality reduction weight, wherein the dimensionality reduction weight is used to digitally precode M reference signals corresponding to M antenna ports to measure the equivalent downlink channels of N physical antennas, M is a positive integer less than N, the input of the compression model is downlink channel information #2, and the output is a first sequence, wherein the downlink channel information #2 is determined based on the downlink channel information #1 and the dimensionality reduction weight, the input of the reconstruction model is the first sequence, and the output is downlink channel information #3.

[0276] For example, the channel prior information may be the uplink channel response H UL , downlink channel information #1 can be the actual downlink channel response H of N physical antennas DL , the input and output of each model and examples can be found in the description of S920 to S950, which will not be repeated here.

[0277] S1030: The first device updates network device parameters of a dimensionality reduction model, a compression model, and a reconstruction model based on first error information, where the first error information is determined based on downlink channel information #3 and downlink channel information #1.

[0278] It is understood that in this method, the first device obtains the first gradient information and the second gradient information described in S940 and S950. Further, the first device updates the model parameters of the reconstruction model based on the first error information, updates the model parameters of the dimensionality reduction model based on the second gradient information, and updates the model parameters of the compression model based on the first gradient information.

[0279] For the first error information, related examples of the first error information, the first gradient information, and the second gradient information, please refer to the description in S940 to S950 and will not be repeated here.

[0280] Optionally, the method further includes: S1040, when a termination condition of the model training is met, the first device determines to terminate the model training.

[0281] For examples of possible termination conditions for model training, please refer to the description in S970 and will not be repeated here.

[0282] The above S1020 and S1030 can be regarded as a round of model training. After each round of model training, the first device needs to update the model parameters of the three locally deployed models. After the update, continue with the next round of model training until the model training is terminated.

[0283] In one possible scenario, if the trained dimensionality reduction model and reconstruction model are deployed on the network device side, and the trained compression model is deployed on the terminal device side, and the first device is the network device, then after the network device completes the joint training of the three models, the network device can send the corresponding information of (1) or (2) to the terminal device:

[0284] (1) The network device sends the trained compression model to the terminal device. Correspondingly, the terminal device receives the trained compression model from the network device.

[0285] In example 1, the terminal device directly uses the received compression model as the compression model in method 800 .

[0286] In a second example, the terminal device may modify the received compression model and use it as the compression model in method 800 .

[0287] It can be understood that the correction in Example 2 can be to add some data sets of the terminal device's own characteristics for re-training, and / or to modify the structure of the received compression model and then train again.

[0288] (2) The network device sends a second training data set to the terminal device, where the second training data set is a training data set consisting of the input and output of the trained compression model. Correspondingly, the terminal device receives the second training data set from the network device.

[0289] It can be understood that the second training data set includes information C and information D, wherein information C is the output information obtained by using the channel prior information in the first training data set as the input of the trained dimensionality reduction model, and information D is the output information obtained by using information C as the input of the trained compression model.

[0290] In example one, the terminal device trains a compression model based on the second training data set as the compression model in method 800 .

[0291] In example two, the terminal device trains a compression model based on the second training data set and uses the revised compression model as the compression model in method 800.

[0292] In another possible scenario, if the trained dimensionality reduction model and reconstruction model are deployed on the network device side, and the trained compression model is deployed on the terminal device side, and the above-mentioned first device is an AI network element, such as a third-party network element, then the third-party network element can send the trained model or the corresponding training data set to the network device and terminal device respectively. Examples will not be given one by one here.

[0293] It should be noted that Method 900 and Method 1000 provide a detailed description of the unilateral and dual-end joint training process using three models as an example. The training method proposed in this application is also applicable to the joint training process of Y (Y>3) models.

[0294] The above describes in detail the two model training methods provided by the present application. The following is a comparison chart of the simulation performance of the model dimensionality reduction method based on the present application and the dimensionality reduction method based on the traditional orthogonal codebook, in conjunction with FIG11 .

[0295] Figure 11 is a performance comparison chart of the dimensionality reduction weights obtained by determining the dimensionality reduction weights using the model dimensionality reduction compression method (method 1) proposed in this application and the dimensionality reduction compression method (method 2) under the traditional orthogonal codebook. The dimensionality reduction weights are used to measure the equivalent downlink channels of 128 physical antennas using 32 reference signals (corresponding to 32 antenna ports), and the AI ​​model structures used for compression in the two schemes are consistent. The horizontal axis of Figure 11 represents the feedback overhead, and the vertical axis represents the square generalized cosine similarity (SGCS) (SGCS can also be simply understood as the accuracy of the reconstructed downlink precoding matrix of 128 physical antennas), where SGCS satisfies the following formula Among them, K is the rank (rank of the channel), N fis the number of frequency domain units (a frequency domain unit can be one or more subcarriers or RBs, without limitation), E{·} represents the average value of the samples in the brackets, is the reconstructed downlink precoding matrix of 128 physical antennas, V is the label, that is, V is the true value of the downlink precoding matrix of 128 physical antennas, (.) H represents the conjugate transpose of a vector, ||.|| represents the modulus of a vector, and ||.|| represents the absolute value of a vector. As shown in Figure 11, the proposed scheme reduces the feedback overhead by approximately 70% compared to the dimensionality reduction scheme using an orthogonal codebook, while maintaining the same performance (i.e., the vertical axis corresponds to the same value).

[0296] It can be understood that the scope of application of the above method is not limited to port dimensionality reduction, but can also be applied to dimensionality reduction scenarios in other dimensions, such as frequency domain dimensionality reduction and analog beam domain dimensionality reduction. This application does not limit this.

[0297] It can be understood that the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0298] It is also understood that in some of the above embodiments, devices in existing network architectures are mainly used as examples for illustrative purposes. It is understood that the embodiments of the present application do not limit the specific form of the devices. For example, devices that can achieve the same functions in the future are applicable to the embodiments of the present application.

[0299] It can be understood that in the above-mentioned method embodiments, the methods and operations implemented by devices (such as the above-mentioned network devices, terminal devices, first devices, second devices, etc.) can also be implemented by components of the devices (such as chips or circuits).

[0300] The method provided by the embodiment of the present application is described in detail above with reference to Figures 1 to 11. The above method is mainly introduced from the perspective of the interaction between the network device and the terminal device (or the first device and the second device). It is understandable that the network device and the terminal device (or the first device and the second device) include hardware structures and / or software modules corresponding to the execution of each function in order to implement the above functions.

[0301] Those skilled in the art should be aware that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is performed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0302] Below, the communication device provided by the embodiment of the present application is described in conjunction with Figures 12 and 13. It can be understood that the description of the device embodiment corresponds to the description of the method embodiment. Therefore, for the content that is not described in detail, please refer to the above method embodiment. For the sake of brevity, some content will not be repeated. The embodiment of the present application can divide the functional modules of the device according to the above method example. For example, each functional module can be divided corresponding to each function, or two or more functions can be integrated into one processing module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiment of the present application is schematic and is only a logical functional division. There may be other division methods in actual implementation. The following is an example of dividing each functional module corresponding to each function.

[0303] The method provided in this application has been described in detail above. The communication device provided in this application is described below. In one possible implementation, the device is used to implement the steps or processes corresponding to the terminal device in the above method embodiment. In another possible implementation, the device is used to implement the steps or processes corresponding to the network device in the above method embodiment. In yet another possible implementation, the device is used to implement the steps or processes corresponding to the first device in the above method embodiment. For example, the first device may be a network device or AI network element #1, such as a third-party network element #1. In yet another possible implementation, the device is used to implement the steps or processes corresponding to the second device in the above method embodiment. For example, the second device may be a terminal device or AI network element #2, such as a third-party network element #2. In yet another possible implementation, the device is used to implement the steps or processes corresponding to the network device and the AI ​​network element that deploys the dimensionality reduction model and the reconstruction model in method 800. In yet another possible implementation, the device is used to implement the steps or processes corresponding to the terminal device and the AI ​​network element that deploys the compression model in the above method embodiment. That is, the first device may be a network device and an AI network element #1, and / or the second device may be a terminal device and an AI network element #2.

[0304] Figure 12 is a schematic block diagram of a communication device 1200 provided in an embodiment of the present application. As shown in Figure 12, the device 1200 may include a communication unit 1210 and a processing unit 1220. The communication unit 1210 can communicate with the outside world, and the processing unit 1220 is used for data processing. The communication unit 1210 may also be referred to as a communication interface or a transceiver unit.

[0305] In one possible design, the device 1200 can implement steps or processes corresponding to those performed by the network device in the above method embodiment, wherein the processing unit 1220 is used to perform processing-related operations of the network device in the above method embodiment, and the communication unit 1210 is used to perform sending-related operations of the network device in the above method embodiment.

[0306] In another possible design, the device 1200 can implement steps or processes corresponding to those performed by the terminal device in the above method embodiment, wherein the communication unit 1210 is used to perform reception-related operations of the terminal device in the above method embodiment, and the processing unit 1220 is used to perform processing-related operations of the terminal device in the above method embodiment.

[0307] In another possible design, the device 1200 can implement steps or processes corresponding to those performed by the first device in the above method embodiment, wherein the communication unit 1210 is used to perform reception-related operations of the first device in the above method embodiment, and the processing unit 1220 is used to perform processing-related operations of the first device in the above method embodiment.

[0308] In another possible design, the device 1200 can implement steps or processes corresponding to those performed by the second device in the above method embodiment, wherein the communication unit 1210 is used to perform reception-related operations of the second device in the above method embodiment, and the processing unit 1220 is used to perform processing-related operations of the second device in the above method embodiment.

[0309] It will be appreciated that the apparatus 1200 herein is embodied in the form of a functional unit. The term "unit" herein may refer to an application specific integrated circuit (ASIC), an electronic circuit, a processor (e.g., a shared processor, a dedicated processor, or a group processor, etc.) and memory for executing one or more software or firmware programs, combined logic circuits, and / or other suitable components that support the described functionality.

[0310] The apparatus 1200 of each of the above-described solutions has the function of implementing the corresponding steps performed by the device in the above-described method. The functions can be implemented by hardware, or by hardware executing corresponding software implementations. The hardware or software includes one or more modules corresponding to the above-described functions; for example, the communication unit can be replaced by a transceiver (for example, the sending unit in the communication unit can be replaced by a transmitter, and the receiving unit in the communication unit can be replaced by a receiver), and other units, such as the processing unit, can be replaced by a processor to respectively perform the sending and receiving operations and related processing operations in each method embodiment.

[0311] In addition, the above-mentioned communication unit can also be a transceiver circuit (for example, it can include a receiving circuit and a transmitting circuit), and the processing unit can be a processing circuit. In an embodiment of the present application, the device in Figure 12 can be the device in the aforementioned embodiment, or it can be a chip or a chip system, such as a system on chip (SoC). Among them, the communication unit can be an input and output circuit, a communication interface; the processing unit is a processor or microprocessor or integrated circuit integrated on the chip. This is not limited here.

[0312] Figure 13 is a schematic block diagram of a communication device 1300 provided in an embodiment of the present application. Device 1300 includes a processing circuit 1310 and a transceiver circuit 1320. Processing circuit 1310 and transceiver circuit 1320 communicate with each other via an internal connection path. Processing circuit 1310 is configured to execute instructions to control transceiver circuit 1320 to send and / or receive signals.

[0313] Optionally, the apparatus 1300 may further include a memory 1330, which communicates with the processing circuit 1310 and the transceiver circuit 1320 via an internal connection path. The memory 1330 is used to store instructions, and the processing circuit 1310 may execute the instructions stored in the memory 1330.

[0314] For example, the processing circuit 1310 may be one or more processors, or may be all or part of the circuits in one or more processors.

[0315] For example, the transceiver circuit 1320 may be a transceiver, an interface circuit, or an input / output circuit.

[0316] In one possible implementation, apparatus 1300 is used to implement the various processes and steps corresponding to the network device in the above-described method embodiment. In another possible implementation, apparatus 1300 is used to implement the various processes and steps corresponding to the terminal device in the above-described method embodiment. In yet another possible implementation, apparatus 1300 is used to implement the various processes and steps corresponding to the first device in the above-described method embodiment. In yet another possible implementation, apparatus 1300 is used to implement the various processes and steps corresponding to the second device in the above-described method embodiment.

[0317] It can be understood that the device 1300 can be specifically the device in the above-mentioned embodiment, or it can be a chip or a chip system. Correspondingly, the transceiver circuit 1320 can be the transceiver circuit of the chip, which is not limited here. Specifically, the device 1300 can be used to execute the various steps and / or processes corresponding to the device in the above-mentioned method embodiment. Optionally, the memory 1330 may include a read-only memory and a random access memory, and provide instructions and data to the processing circuit. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type. The processing circuit 1310 can be used to execute instructions stored in the memory, and when the processing circuit 1310 executes instructions stored in the memory, the processing circuit 1310 is used to execute the various steps and / or processes of the above-mentioned method embodiment corresponding to the device.

[0318] During implementation, each step of the above method can be completed by an integrated logic circuit of the hardware in the processor or by instructions in the form of software. The steps of the method disclosed in conjunction with the embodiments of the present application can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in conjunction with its hardware. To avoid repetition, it will not be described in detail here.

[0319] For example, the processor in the embodiments of the present application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above-mentioned method embodiment can be completed by hardware integrated logic circuits in the processor or software instructions. The above-mentioned processor can be a general-purpose processor, digital signal processing (DSP), ASIC, field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. The processor in the embodiments of the present application can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of the present application can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above-mentioned method.

[0320] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus RAM (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0321] It should be noted that when the processor is a general-purpose processor, DSP, ASIC, FPGA or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, the memory (storage module) can be integrated into the processor.

[0322] In addition, the present application also provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed on a computer, the operations and / or processes performed by the network device or terminal device or the first device or the second device in each method embodiment of the present application are executed.

[0323] The present application also provides a computer program product, which includes computer program code or instructions. When the computer program code or instructions are run on a computer, the operations and / or processes performed by the network device or terminal device or the first device or the second device in each method embodiment of the present application are executed.

[0324] In addition, the present application further provides a chip, the chip including a processing circuit. A memory for storing a computer program is provided independently of the chip, and the processing circuit is configured to execute the computer program stored in the memory, so that the operations and / or processes performed by the network device, the terminal device, the first device, or the second device in any one of the method embodiments are performed.

[0325] Furthermore, the chip may further include a communication interface. The communication interface may be an input / output interface, or an interface circuit, etc. Furthermore, the chip may further include a memory.

[0326] In addition, the present application also provides a communication system, including the network device and terminal device in the embodiments of the present application, and / or, the first device and the second device.

[0327] It should also be noted that the memory described herein is intended to comprise, but not be limited to, these and any other suitable types of memory.

[0328] Those skilled in the art will appreciate that the various exemplary units and algorithmic steps described in conjunction with the embodiments disclosed herein can be implemented using electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented using hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application. Those skilled in the art will clearly understand that, for ease of description and brevity, the specific operating 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. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units described is merely a logical functional division. In actual implementation, other divisions may be used, such as multiple units or components being combined or integrated into another system, or some features being omitted or not implemented. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interface, or indirect coupling or communication connection between devices or units, which may be electrical, mechanical, or other forms. The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, the functional units in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0329] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

[0330] It should be understood that references to "embodiments" throughout this specification mean that a particular feature, structure, or characteristic associated with the embodiment is included in at least one embodiment of the present application. Therefore, various embodiments throughout this specification do not necessarily refer to the same embodiment. Furthermore, these particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0331] It can also be understood that in this application, "when", "if" and "if" all mean that the network element will make corresponding processing under certain objective circumstances, which is not a time limit, and does not require the network element to have a judgment action when implementing it, nor does it mean that there are other limitations.

[0332] It is also understood that in each embodiment of the present application, "A corresponds to B" means that B is associated with A, and B can be determined based on A. However, it is also understood that determining B based on A does not mean determining B based solely on A, and B can also be determined based on A and / or other information.

[0333] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application 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 application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A communication method, characterized in that, Comprising: Performing digital precoding on M reference signals corresponding to M antenna ports based on the dimension-reduced weight values, where the dimension-reduced weight values are used to measure the equivalent downlink channel of N physical antennas by using the M reference signals, N is a positive integer, M is a positive integer less than N, and the dimension-reduced weight values are determined based on a dimension reduction model; Transmitting the M reference signals; Transmitting first indication information indicating a compression model, where the compression model matches the dimension reduction model; Receiving a first sequence, where the first sequence is a compressed feedback quantity obtained by using first downlink channel information as an input to the compression model, and the first downlink channel information is obtained based on the measurement of the M reference signals; Obtaining second downlink channel information, where the second downlink channel information is information obtained by using the first sequence as an input to a reconstruction model, the reconstruction model matches the dimension reduction model, and the second downlink channel is used to determine the downlink channel information of the N physical antennas.

2. A communication method, characterized in that, Comprising: Receiving M reference signals corresponding to M antenna ports, where M is a positive integer; Obtaining first downlink channel information based on the M reference signals; Receiving first indication information indicating a compression model, where the compression model matches a dimension reduction model, the dimension reduction model is used to obtain dimension-reduced weight values, and the dimension-reduced weight values are used to measure the equivalent downlink channel of N physical antennas by using the M reference signals, N is a positive integer, and M is a positive integer less than N; Transmitting a first sequence, where the first sequence is a compressed feedback quantity obtained by using first downlink channel information as an input to the compression model, and the first sequence is used as an input to a reconstruction model to be decompressed to obtain second downlink channel information, the reconstruction model matches the dimension reduction model, and the second downlink channel is used to determine the downlink channel information of the N physical antennas.

3. The method according to claim 1 or 2, characterized in that, The dimension-reduced weight values are output information obtained by using channel prior information as an input to the dimension reduction model.

4. The method according to claim 3, characterized in that, The channel prior information is determined based on at least one of the following information: currently measured uplink channel information, historical uplink channel information, historical downlink channel information, or sensed channel information.

5. The method according to claim 3 or 4, characterized in that, The dimension reduction model, the compression model, and the reconstruction model are all related to the number of antenna ports M, the number of physical antennas N, and the channel prior information.

6. The method according to any one of claims 1 to 5, characterized in that The first downlink channel information indicates the response of the equivalent downlink channel, and the first sequence is the compressed feedback quantity of the response.

7. The method according to any one of claims 1 to 5, characterized in that The first downlink channel information indicates the precoding matrix corresponding to the response of the equivalent downlink channel, and the first sequence is the compressed feedback quantity of the precoding matrix information corresponding to the response.

8. The method according to any one of claims 1 to 7, characterized in that, The downlink channel information of the N physical antennas includes the downlink precoding matrix of the N physical antennas.

9. A model training method, characterized in that, Comprising: Obtain a training data set, where the training data set includes channel prior information and first downlink channel information, and the first downlink channel information is the true downlink channel information of N physical antennas, and N is a positive integer; Transmit second downlink channel information, where the second downlink channel information is determined based on the first downlink channel information and a dimension reduction weight value. Among them, the dimension reduction weight value is the output information obtained by taking the channel prior information as the input of a dimension reduction model. The dimension reduction weight value is used for digital precoding of M reference signals corresponding to M antenna ports to measure the equivalent downlink channel of the N physical antennas, and M is a positive integer less than N; Receive a first sequence, where the first sequence is the output information obtained by taking the second downlink channel information as the input of a compression model; Transmit first gradient information, where the first gradient information is the input-side gradient information of a reconstruction model. Among them, the first gradient information is used to update the network parameters of the compression model, and the first gradient information is determined based on first error information and the reconstruction model. The first error information is determined based on third downlink channel information and the first downlink channel information, and the third downlink channel information is the output information obtained by taking the first sequence as the input of the reconstruction model or is calculated based on the output information of the reconstruction model; Receive second gradient information, where the second gradient information is the input-side gradient information of the compression model, and the second gradient information is determined based on the first gradient information and the compression model; Update the model parameters of the reconstruction model based on the first error information, and update the model parameters of the dimension reduction model based on the second gradient information.

10. The method according to claim 9, wherein The method further includes: Terminate model training when the termination condition for model training is satisfied.

11. A model training method, characterized in that, It includes: Receive second downlink channel information, where the second downlink channel information is determined based on first downlink channel information and a dimension reduction weight value. The first downlink channel information is the true downlink channel information of N physical antennas, and the dimension reduction weight value is the output information obtained by taking channel prior information as the input of a dimension reduction model. The dimension reduction weight value is used for digital precoding of M reference signals corresponding to M antenna ports to measure the equivalent downlink channel of the N physical antennas. N is a positive integer, M is a positive integer less than N, and the first downlink channel information and the channel prior information are the training data set for model training; Transmit a first sequence, where the first sequence is the output information obtained by taking the information of the second downlink channel as the input of a compression model; Receive first gradient information, where the first gradient information is the input - side gradient information of the reconstruction model. The first gradient information is used to update the model parameters of the compression model and is determined based on the first error information and the reconstruction model. The first error information is determined based on the third downlink channel information and the first downlink channel information. The third downlink channel information is the output information obtained by using the first sequence as the input of the reconstruction model or is calculated based on the output information of the reconstruction model. The first error information is used to update the model parameters of the reconstruction model; Send second gradient information, where the second gradient information is the input - side gradient information of the compression model and is determined based on the first gradient information and the compression model; Update the model parameters of the compression model based on the first gradient information.

12. The method according to any one of claims 9 to 11, characterized in that The channel prior information is determined based on at least one of the following information: the currently measured uplink channel information, historical uplink channel information, historical downlink channel information, or sensed channel information.

13. The method according to any one of claims 9 to 12, characterized in that The first downlink channel information is the true downlink channel response of the N physical antennas.

14. The method according to claim 13, wherein The second downlink channel information indicates the downlink channel response, where the downlink channel response is determined based on the first downlink channel information and the dimension - reduction weight, or the second downlink channel information indicates the weight corresponding to the downlink channel response.

15. The method according to claim 14, wherein The third downlink channel information is the reconstructed weights of the N physical antennas. The first error information indicates the error between the third downlink channel information and the true weights of the N physical antennas, and the true weights of the N physical antennas are determined based on the first downlink channel information.

16. A model training method, characterized in that, Comprising: Obtain a first training data set, where the first training data set includes channel prior information and first downlink channel information. The first downlink channel information is the true downlink channel information of N physical antennas, and N is a positive integer; Jointly train a dimension - reduction model, a compression model, and a reconstruction model. The input of the dimension - reduction model is the channel prior information, and the output is the dimension - reduction weight. The dimension - reduction weight is used for digital precoding of M reference signals corresponding to M antenna ports to measure the equivalent downlink channel of the N physical antennas, where M is a positive integer less than N. The input of the compression model is the second downlink channel information, and the output is a first sequence. The second downlink channel information is determined based on the first downlink channel information and the dimension - reduction weight. The input of the reconstruction model is the first sequence, and the output is the third downlink channel information; Update the network device parameters of the dimension - reduction model, the compression model, and the reconstruction model based on the first error information. The first error information is determined based on the third downlink channel information and the first downlink channel information.

17. The method according to claim 16, wherein When the error indicated by the first error information is less than or equal to a first threshold, or the number of model training rounds is equal to a second threshold, terminate the model training.

18. The method according to claim 17, wherein the second downlink channel information indicates a downlink channel response, the downlink channel response is determined based on the first downlink channel information and the dimension-reduced weight, or the second downlink channel information indicates a weight corresponding to the downlink channel response.

19. The method according to claim 18, characterized in that, the third downlink channel information is the weight of the reconfigured N physical antennas, the first error information indicates an error between the third downlink channel information and the true weights of the N physical antennas, and the true weights of the N physical antennas are determined based on the first downlink channel information.

20. A model training method, characterized in that, comprising: receiving a second training data set, the second training data set including output information of the dimension-reduced model obtained by using the first training data set again after the dimension-reduced model, the compression model, and the reconstruction model complete joint training based on the first training data set, and output information of the compression model, the first training data set including channel prior information and first downlink channel information, wherein, in one round of joint training corresponding to the dimension-reduced model, the compression model, and the reconstruction model, the input of the dimension-reduced model is the channel prior information and the output is the dimension-reduced weight, wherein the dimension-reduced weight is used for digital precoding of M reference signals corresponding to M antenna ports to measure the equivalent downlink channel of N physical antennas, M being a positive integer less than N, the input of the compression model is the second downlink channel information and the output is a first sequence, wherein the second downlink channel information is determined based on the first downlink channel information and the dimension-reduced weight, and the input of the reconstruction model is the first sequence and the output is the third downlink channel information; training a first model based on the second training data set to obtain a second model.

21. The method according to claim 20, wherein The method further comprises: sending the compression model corresponding to after completing model training; or sending the second training data set, the second training data set including output information of the dimension-reduced model obtained by using the first training data set again after completing model training, and output information of the compression model.

22. The method according to any one of claims 16 to 21, characterized in that, The channel prior information is determined based on at least one of the following information: currently measured uplink channel information, historical uplink channel information, historical downlink channel information, or sensed channel information.

23. The method according to any one of claims 16 to 22, characterized in that The first downlink channel information is the true downlink channel response of the N physical antennas.

24. The method according to claim 23, wherein The second downlink channel information indicates a downlink channel response, the downlink channel response is determined based on the first downlink channel information and the dimension-reduced weight, or the second downlink channel information indicates a weight corresponding to the downlink channel response.

25. A communication device, characterized in that, Comprising a module or unit for performing the method according to any one of claims 1, 3 to 8, or comprising a module or unit for performing the method according to any one of claims 2 to 8, or comprising a module or unit for performing the method according to any one of claims 9, 10, 12 to 15, or comprising a module or unit for performing the method according to any one of claims 11 to 15, or comprising a module or unit for performing the method according to any one of claims 16 to 19 or any one of claims 22 to 24, or comprising a module or unit for performing the method according to any one of claims 20 to 24.

26. A communication device, characterized in that, Comprising: A processor for executing a computer program stored in a memory, so that the device performs the method according to any one of claims 1, 3 to 8, or so that the device performs the method according to any one of claims 2 to 8, or so that the device performs the method according to any one of claims 9, 10, 12 to 15, or so that the device performs the method according to any one of claims 11 to 15, or so that the device performs the method according to any one of claims 16 to 19 or any one of claims 22 to 24, or so that the device performs the method according to any one of claims 20 to 24.

27. A computer program product, characterized in that, The computer program product comprises instructions for performing the method according to any one of claims 1, 3 to 8, or the computer program product comprises instructions for performing the method according to any one of claims 2 to 8, or the computer program product comprises instructions for performing the method according to any one of claims 9, 10, 12 to 15, or the computer program product comprises instructions for performing the method according to any one of claims 11 to 15, or the computer program product comprises instructions for performing the method according to any one of claims 16 to 19 or any one of claims 22 to 24, or the computer program product comprises instructions for performing the method according to any one of claims 20 to 24.

28. A computer-readable storage medium, characterized in that, Comprising: The computer-readable storage medium stores a computer program; when the computer program runs on a computer, it causes the computer to perform the method according to any one of claims 1, 3 to 8, or causes the computer to perform the method according to any one of claims 2 to 8, or causes the computer to perform the method according to any one of claims 9, 10, 12 to 15, or causes the computer to perform the method according to any one of claims 11 to 15, or causes the computer to perform the method according to any one of claims 16 to 19 or any one of claims 22 to 24, or causes the computer to perform the method according to any one of claims 20 to 24.

Images