Channel State Information Feedback System and Method

Abstract

The present invention provides a channel state information feedback system and method. The system includes: a first CSI-RS feedback inference network agent deployed on the network side and a second CSI-RS feedback inference network agent deployed on the terminal side; the first CSI-RS feedback inference network agent is used to send a feedback inference network model to the second CSI-RS feedback inference network agent and, based on the federated learning mechanism, cooperate with the second CSI-RS feedback inference network agent to perform online training on the feedback inference network model; the second CSI-RS feedback inference network agent is used to, based on the federated learning mechanism, cooperate with the first CSI-RS feedback inference network agent to perform online training on the feedback inference network model. By separately deploying agents on the network side and the terminal side and implementing federated learning on a specific feedback inference network model, the present invention can improve the accuracy of channel state information feedback.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technologies, and in particular, to a channel state information feedback system and method. Background Art

[0002] Massive MIMO (Multiple-Input Multiple-Output) is one of the core technologies of wireless communication. With the rapid growth of the demand for bandwidth resources in wireless mobile communication technologies and the rapid development of multi-antenna technologies, the advantages of massive MIMO technologies are gradually emerging. Larger-scale antenna arrays can multiply the communication capacity, optimize the energy transmission efficiency of information, and shorten the air interface transmission delay.

[0003] The effectiveness of the massive MIMO transmission link, especially the improvement of capacity, the most critical determining factor is the proximity between the downlink CSI (Channel State Information) report obtained by the base station and the actual downlink CSI during actual transmission. If the CSI information can accurately reflect the link situation at the actual transmission moment, the base station scheduler can schedule wireless resources according to the CSI information and send information to the receiving terminal in a manner closer to the maximum capacity of the actual channel. The CSI information fed back by multiple terminals can also assist the base station in performing MU-MIMO (Multi-User Multiple-Input Multiple-Output) paired transmission.

[0004] For an FDD (Frequency Division Duplexing) system, since the downlink carrier and the uplink carrier are not in the same frequency band and the channel reciprocity cannot be utilized, the base station cannot optimize the transmission on the downlink carrier using the channel estimation information of the uplink signal sent by the UE (User Equipment). The general method is for the UE to measure the downlink channel and then report the measured channel state information CSI to the base station through the uplink channel. After receiving it, the base station uses this reported information to optimize and schedule the downlink signal in the next downlink transmission window, so as to achieve the purpose of improving the downlink transmission efficiency, improving the user's downlink experience, and optimizing various performances.

[0005] The existing CSI reporting configuration of 5G NR (New Radio) usually adopts a codebook-based CSI feedback method. However, with the increase in the number of antennas in the massive MIMO system, the codebook size and computational complexity will increase significantly. When a terminal reports CSI information to a base station in broadband wireless communication, related technologies compress the CSI information through deep learning technology and parse and recover the CSI information after the base station receives it. However, for terminals in different cells, the "CSI-RS (Channel State Information-Reference Signal) feedback inference network" used by the terminal and the base station ("UE-gNB pair") will vary due to different channel environments in different cells and different antenna panel / array combinations when using multi-TRP (multi-Transmission and Reception Point). There are differences in aspects such as the network model and the weights in the same model, and it is necessary for both the UE and the gNB to separately update the encoder (carried by the UE) and decoder (carried by the gNB) in the local "CSI-RS feedback inference network". However, in the existing artificial intelligence-based channel state information feedback method, due to the difficulty in corresponding updates of the "CSI-RS feedback inference network" used by the terminal and the base station, the accuracy of the CSI information parsed and recovered on the base station side is not high.

[0006] Therefore, how to improve the accuracy of artificial intelligence-based channel state information feedback has become an urgent problem to be solved in the industry. Summary of the Invention

[0007] In view of the problems existing in the prior art, the present invention provides a channel state information feedback system and method.

[0008] In a first aspect, the present invention provides a channel state information feedback system, including:

[0009] A first CSI-RS feedback inference network agent deployed on the network side and a second CSI-RS feedback inference network agent deployed on the terminal side;

[0010] The first CSI-RS feedback inference network agent is used to send a CSI-RS feedback inference network model to the second CSI-RS feedback inference network agent corresponding to the terminal side in the cell, and based on the federated learning mechanism, cooperate with the second CSI-RS feedback inference network agent to perform online training on the CSI-RS feedback inference network model;

[0011] The second CSI-RS feedback inference network agent is used to cooperate with the first CSI-RS feedback inference network agent to perform online training on the CSI-RS feedback inference network model based on a federated learning mechanism.

[0012] Optionally, a channel state information feedback system provided according to the present invention further includes: a third CSI-RS feedback inference network agent deployed on the core network side;

[0013] The third CSI-RS feedback inference network agent is used to manage and evaluate the performance of the candidate CSI-RS feedback inference network model of the entire network.

[0014] In a second aspect, the present invention further provides a channel state information feedback method based on any channel state information feedback system described in the first aspect, applied to a first CSI-RS feedback inference network agent, comprising:

[0015] In the case of determining that the network side receives a connection establishment request message sent by the target terminal side, establishing a connection with a second CSI-RS feedback inference network agent corresponding to the target terminal side;

[0016] By interacting with the second CSI-RS feedback inference network agent, determining whether the encoding codeword of the first CSI-RS feedback inference network encoder model currently used by the target terminal side is the same as the decoding codeword of the first CSI-RS feedback inference network decoder model currently used by the network side;

[0017] When it is determined that they are the same, a first CSI-RS feedback inference network model is established between the network side and the target terminal side based on the first CSI-RS feedback inference network encoder model and the first CSI-RS feedback inference network decoder model.

[0018] Optionally, according to a channel state information feedback method provided by the present invention, after establishing a first CSI-RS feedback inference network model between the network side and the target terminal side, the method further includes:

[0019] Determine whether the first CSI-RS feedback inference network encoder model is available in the current cell;

[0020] When it is determined that it is unavailable, first indication information carrying a first model update message is sent to the second CSI-RS feedback inference network agent, wherein the first indication information is used to instruct the second CSI-RS feedback inference network agent to update the first CSI-RS feedback inference network encoder model based on the first model update message.

[0021] Optionally, a method for channel state information feedback provided by the present invention further includes:

[0022] In the case of determination of non - identity, determine whether the first CSI - RS feedback inference network encoder model and the first weights corresponding to the first CSI - RS feedback inference network encoder model are available in the current cell;

[0023] In the case where it is determined that the first CSI - RS feedback inference network encoder model is available in the current cell and some or all of the first weights are unavailable, send second indication information carrying a second model update message to the second CSI - RS feedback inference network agent, where the second indication information is used to instruct the second CSI - RS feedback inference network agent to update the first CSI - RS feedback inference network encoder model based on the second model update message.

[0024] Optionally, a method for channel state information feedback provided by the present invention further includes:

[0025] In the case of determination of non - identity, send abnormal fallback information to the second CSI - RS feedback inference network agent, where the abnormal fallback information is used to instruct the second CSI - RS feedback inference network agent to perform abnormal fallback on the first CSI - RS feedback inference network encoder model.

[0026] Optionally, after sending the abnormal fallback information to the second CSI - RS feedback inference network agent in a method for channel state information feedback provided by the present invention, it further includes:

[0027] Determine the second CSI - RS feedback inference network encoder model obtained after the second CSI - RS feedback inference network agent performs abnormal fallback on the first CSI - RS feedback inference network encoder model;

[0028] Load a second CSI - RS feedback inference network decoder model that matches the second CSI - RS feedback inference network encoder model, and based on the second CSI - RS feedback inference network encoder model and the second CSI - RS feedback inference network decoder model, establish a second CSI - RS feedback inference network model between the network side and the target terminal side.

[0029] Optionally, a method for channel state information feedback provided by the present invention further includes:

[0030] Based on the federated learning mechanism, cooperate with the second CSI - RS feedback inference network agent to perform online training on the second CSI - RS feedback inference network model;

[0031] When it is determined that the performance of the second CSI-RS feedback inference network model meets the first preset threshold, the second weight corresponding to the second CSI-RS feedback inference network model is optimized based on the federated learning mechanism.

[0032] Optionally, a channel state information feedback method provided according to the present invention further includes:

[0033] When it is determined that the performance of the second CSI-RS feedback inference network model meets the second preset threshold, a candidate CSI-RS feedback inference network model provided by a third CSI-RS feedback inference network agent is determined, and based on a federated learning mechanism, federated learning is performed on the candidate CSI-RS feedback inference network model.

[0034] Optionally, a channel state information feedback method provided according to the present invention further includes:

[0035] Collect statistics on the training results of the local CSI-RS feedback inference network model;

[0036] The training result information is reported to the third CSI-RS feedback reasoning network agent, so that the third CSI-RS feedback reasoning network agent can manage the candidate CSI-RS feedback reasoning network model of the entire network based on the model training result information.

[0037] In a third aspect, the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the channel state information feedback method as described in the second aspect is implemented.

[0038] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the channel state information feedback method as described in the second aspect.

[0039] In a fifth aspect, the present invention further provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the channel state information feedback method as described in the second aspect is implemented.

[0040] The channel state information feedback system and method provided by the present invention respectively deploy CSI-RS feedback reasoning network agents on the network side and the terminal side, and implement federated learning of a specific feedback reasoning network model based on the CSI-RS feedback reasoning network agents on the network side and the terminal side to obtain a feedback reasoning network model with better performance, thereby improving the accuracy of channel state information feedback. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0042] Figure 1 It is a schematic structural diagram of a channel state information feedback system provided by the present invention;

[0043] Figure 2 It is a schematic flowchart of a channel state information feedback method provided by the present invention;

[0044] Figure 3 It is a schematic physical structure diagram of an electronic device provided by the present invention. Detailed implementation manners

[0045] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments in the present invention belong to the scope of protection of the present invention.

[0046] To facilitate a clearer understanding of the embodiments of the present invention, some relevant background knowledge will be introduced as follows.

[0047] The existing CSI reporting configuration for 5G NR includes two parts: CSI-RS related resource configuration (indicating that the UE measures at this time-frequency resource) and the configuration for the UE to perform CSI reporting (indicating that the UE sends a CSI measurement report at this time-frequency resource). The resource configuration is used to configure the reference signal for calculating CSI, while the reporting configuration is used to configure the behavior of reporting CSI. The RRC (Radio Resource Control) layer signaling CSI-ReportConfig IE indicates the resources for channel measurement and interference measurement, which also includes codebook configuration, including Type I, Type II, or enhanced Type II codebooks and codebook restricted subsets, and indicates that the time-domain CSI reporting adopts a periodic manner, a semi-persistent manner, a semi-persistent manner based on PUSCH (Physical Uplink Shared Channel), or an aperiodic manner, etc. The settings also include: the frequency-domain wideband and sub-band granularity of CQI (Channel Quality Indication) and PMI (Precoding Matrix Indicator); the restrictions on channel measurement and interference measurement; the types of information that the CSI reported by the UE needs to indicate, including CQI, PMI, CRI (CSI-RS Resource Indicator), SSBRI (SS / PBCH Block Resource Indicator), LI (Lay Indicator), RI (Rank Indicato), L1-RSRP (Layer 1 Reference Signal Received Power), or L1-SINR (Layer 1 Signal to Interference plus Noise Ratio), etc.

[0048] For the above codebook-based CSI feedback method, as the number of antennas in the massive MIMO system increases, both the codebook size and the computational complexity will increase significantly.

[0049] Using artificial intelligence methods, all or part of the channel state information measured by the UE is encoded and compressed through a neural network encoder deployed on the UE side to form a dedicated codeword, and the codeword is sent to the base station through the uplink of the air interface. On the base station side, by using a neural network that matches the UE side, based on the received codeword and the prior knowledge learned and stored in the network by the neural network on the base station side, all channel features with a high similarity to the UE measurement information are restored. To implement this function, the following problems need to be solved:

[0050] (1) For terminals in different cells, the "CSI-RS feedback inference network" used by the "UE-gNB pair" will have differences in terms of network models and weights in the same model due to different channel environments in each cell and different antenna panel / array combinations when using multi-TRP;

[0051] (2) When the UE powers on and registers or switches to a new cell, the above differences require both the UE and the gNB to update the encoders (carried by the UE) and decoders (carried by the gNB) in the local "CSI-RS feedback inference network" respectively. How the gNB can efficiently update the encoder of the UE part of the "CSI-RS feedback inference network" without occupying too much system downlink radio resources is a problem that needs to be solved;

[0052] (3) After the UE obtains the update of the encoder of the UE part of the "CSI-RS feedback inference network" from the current gNB, while using the existing network for inference, it is also necessary to optimize the performance of the local encoder in a certain way. A more common approach is to perform online learning on the local encoder while performing inference; considering the limited computing power of the UE, usually a certain way is adopted to learn and optimize the weights of the encoder on the UE side, that is, the online learning of the weights of the encoder on the UE side is carried out by using the distributed federated learning mechanism of the "UE-gNB pair". When certain conditions are met, the UE can feedback the weight combination with better effects to the gNB through the uplink, and then the gNB updates it to the suitable UE in this cell;

[0053] (4) Due to the large number of UEs in the cell, the gNB can also adopt the method of configuring different encoder models for different UEs and jointly training different decoders on the base station side to monitor the performance and perform online training on different "CSI-RS feedback inference network" models;

[0054] (5) To prevent network security risks against artificial intelligence, it is necessary to design an efficient and trustworthy privacy computing security prevention mechanism based on federated learning for the "CSI-RS feedback inference network".

[0055] To overcome the above-mentioned deficiencies, the present invention provides a channel state information feedback system and method. The following will be combined with Figures 1 - 3 to describe the channel state information feedback system and method provided by the present invention.

[0056] Figure 1 is a schematic structural diagram of the channel state information feedback system provided by the present invention. As Figure 1 shown, the system includes:

[0057] a first CSI-RS feedback inference network agent deployed on the network side and a second CSI-RS feedback inference network agent deployed on the terminal side;

[0058] The first CSI-RS feedback inference network agent is used to send a CSI-RS feedback inference network model to the second CSI-RS feedback inference network agent corresponding to the terminal side in the cell, and based on the federated learning mechanism, cooperate with the second CSI-RS feedback inference network agent to perform online training on the CSI-RS feedback inference network model;

[0059] The second CSI-RS feedback inference network agent is used to, based on the federated learning mechanism, cooperate with the first CSI-RS feedback inference network agent to perform online training on the CSI-RS feedback inference network model.

[0060] Specifically, in the embodiments of the present invention, the channel state information feedback system may include a first CSI-RS feedback inference network agent deployed on the network side and a second CSI-RS feedback inference network agent deployed on the terminal side. Among them, the first CSI-RS feedback inference network agent may be used to send a CSI-RS feedback inference network model to the second CSI-RS feedback inference network agent corresponding to the terminal side in the cell, and based on the federated learning mechanism, cooperate with the second CSI-RS feedback inference network agent to perform online training on the CSI-RS feedback inference network model. The second CSI-RS feedback inference network agent is used to, based on the federated learning mechanism, cooperate with the first CSI-RS feedback inference network agent to perform online training on the CSI-RS feedback inference network model.

[0061] Optionally, as Figure 1 shown, the first CSI-RS feedback inference network agent deployed on the network side may include the following functional modules: a CSI-RS feedback inference model update module, a CSI-RS inference module, and a CSI-RS feedback inference model federated learning management module. Among them, the CSI-RS feedback inference model update module may include various model performance evaluation sub-modules, a model distribution sub-module, and a model training sub-module.

[0062] Optionally, as Figure 1As shown in the figure, the second feedback inference network agent deployed on the terminal side may include the following functional modules: a CSI-RS feedback inference model update module, a CSI-RS inference module, and a CSI-RS feedback inference model federated learning management module.

[0063] Specifically, the working processes of the above functional modules are introduced below.

[0064] Optionally, in the embodiments of the present invention, the network side may be a base station.

[0065] Optionally, the model distribution sub-module on the network side may be responsible for managing the terminals in the cell, configuring, by means of a certain strategy, which CSI-RS feedback inference network model the terminals in the cell use for CQI feedback, and performing online federated learning at the same time.

[0066] Optionally, the model training sub-module may be a management and control unit on the base station side for online training and federated learning of the "CSI-RS feedback inference network model" used by different UEs, and is responsible for configuring various parameters and training methods of the encoder on the terminal side and the corresponding decoder on the base station side. The model training sub-module determines the effect of downlink scheduling and transmission based on this information according to the CQI codeword information reported by the terminal after compression by the encoder each time and the downlink channel information restored by the decoder on the base station side. The specific method includes: in each training cycle, by calling a preset training set, online evaluating the accuracy and inference ability of the model.

[0067] Optionally, the CSI-RS inference module is equivalent to an actuator. When the neural network model is trained, it will be loaded into the execution environment, and the neural network working in the execution environment is the module that performs the inference function. The CSI-RS inference modules on the base station side and the terminal side are an integrated model, and the two parts of the network cooperate with each other. By transmitting the codeword formed after compression by the encoder on the terminal side to the decoder on the base station side through the air interface, all information channel information is restored to reduce the air interface transmission overhead, and at the same time, a relatively accurate downlink channel estimation value can be obtained.

[0068] For example, the CSI-RS inference process can be understood as that the networks on the terminal side and the base station side have learned the laws and knowledge of the current downlink channel. Only by indicating the mode of the current downlink channel through the codeword carried by a smaller number of bits, the base station can make better downlink resource scheduling according to the indication. This principle is similar to the codebook-based CSI feedback, but the neural network can use its non-linear ability to achieve a more accurate indication of the current channel state.

[0069] It can be understood that in a more general model of an AI (Artificial Intelligence)-based CSI feedback network, a joint AI model is deployed on the terminal side and the base station side. Generally, a certain neural network is used as a CSI information encoding and compression module to compress and encode all or part of the channel characteristics obtained by the UE through measuring the CSI-RS (Channel State Information-Reference Signal) sent by the gNB (base station) and performing channel estimation, forming a codeword, which is reported to the base station through an uplink channel such as the PUCCH (Physical Uplink Control Channel) or the PUSCH. On the base station side, a matching artificial intelligence decoding network is used to send the compressed encoding of all or part of the received channel characteristics into a decoding network, and all or part of the channel characteristics are restored from the received codeword through the prior knowledge stored in the decoding network.

[0070] To implement the above functions, a relatively stable, secure, and computationally acceptable network architecture at the gNB granularity is required for support. The channel state information feedback system provided by the present invention can ensure the performance and security of the channel state information artificial intelligence compression reporting system, and realize online training, inference, update of the "UE encoder + gNB decoder" artificial intelligence network model, as well as security protection and management of distributed intelligent agents.

[0071] The channel state information feedback system provided by the embodiments of the present invention supports multiple UEs under the same base station to respectively utilize the wireless environment, service conditions, and built-in AI computing power of the UEs themselves, cooperate with the base station to perform online AI compression of CSI, and then decompress it after transmitting it to the base station.

[0072] The channel state information feedback system provided by the present invention deploys CSI-RS feedback inference network agents on the network side and the terminal side respectively, and realizes federated learning on a specific feedback inference network model based on the CSI-RS feedback inference network agents on the network side and the terminal side, so as to obtain a feedback inference network model with better performance, thereby improving the accuracy of channel state information feedback.

[0073] Optionally, the channel state information feedback system provided by the present invention further includes: a third CSI-RS feedback inference network agent deployed on the core network side;

[0074] The third CSI-RS feedback inference network agent is used to manage and evaluate the performance of the candidate CSI-RS feedback inference network models for the entire network.

[0075] Specifically, asFigure 1 As shown, in the embodiment of the present invention, the channel state information feedback system may further include a third CSI-RS feedback inference network agent deployed on the core network side, where the third CSI-RS feedback inference network agent is used to manage and evaluate the performance of the whole network candidate CSI-RS feedback inference network model.

[0076] Optionally, as Figure 1 shown, the third CSI-RS feedback inference network agent deployed on the core network side may include the following functional modules: a CSI-RS feedback inference model update module, which includes multiple model performance evaluation sub-modules and a model distribution sub-module.

[0077] Optionally, the model distribution sub-module on the core network side may control the model strategies of each gNB according to the performance evaluation of multiple CSI feedback neural network models in the whole network.

[0078] Specifically, the working process of the channel state information feedback system provided by the present invention includes the following steps:

[0079] (1) The target UE establishes a connection with the gNB. When the target UE powers on and registers or switches to a new cell, the target UE side establishes a connection with the gNB model update module to update the model in the CSI-RS feedback inference network agent on the target UE side, and then proceeds to step (2);

[0080] (2) The model distribution sub-module in the gNB establishes a connection with the model update module of the target UE. Through information interaction, the model update module on the gNB side evaluates the CSI-RS feedback inference network encoder model and weights currently used by the UE. If an update is required, it proceeds to step (4); if it is available, it proceeds to step (3);

[0081] (3) If the current UE model and weights are available, it immediately notifies the CSI-RS inference module in the gNB to load the corresponding decoder model and weights, establishes a "CSI-RS feedback inference network" between the gNB and the UE, and at the same time incorporates this network into the "model federated learning management module" of the gNB to monitor and manage this network, and then proceeds to step (5); if the feedback inference network of the current UE is abnormal, the gNB model federated learning management module will instruct the UE to perform abnormal fallback. If the abnormal fallback fails, it controls the UE to proceed to step (1); at the same time, the gNB federated learning management module reports the abnormality to the gNB model training sub-module and proceeds to step (6);

[0082] (4) The model update module on the gNB side evaluates the CSI-RS feedback encoder model currently in use by the UE and the weights, etc. If the model is available in this cell, but some or all of the weights in the encoder weights cannot be used, then the process of updating some or all of the weights in the weights corresponding to the UE encoder model is immediately executed, which specifically includes: notifying the model update module of the UE agent to execute the update of some or all of the weights corresponding to the encoder model. The UE agent will receive the updated weight information using the unicast or multicast channel and update the weights of the UE local CSI-RS inference module. When the CSI-RS feedback encoder model and weights on the UE side are available, go to step (3); if the model is not available, then go to step (5);

[0083] (5) The model update module on the gNB side evaluates the CSI-RS feedback encoder model currently in use by the UE and the weights, etc. If the model and weights are not available in this cell, then the process of updating the UE encoder model and weights is immediately executed, which specifically includes: notifying the model update module of the UE agent to execute the update of the encoder model and weights. The UE agent will receive the updated weight information using the unicast or multicast channel and update the weights of the UE local CSI-RS inference module, and then go to step (2);

[0084] (6) After the gNB agent loads the CSI-RS decoder model and weights corresponding to the current UE, it starts and monitors the "CSI-RS inference module" between the gNB and the UE to work; the "model federated learning management modules" on the gNB and UE sides use the online learning mechanism to train the encoder and decoder according to the federated learning rules. When the CSI-RS inference network model has been continuously trained for a period of time and it is found that it reaches a better performance threshold, the "model federated learning management module" on the gNB side sends a notification to the model training sub-module in the model update module of the local (gNB). The model training sub-module manages the online learning of the gNB-UEs using the same model within the local (gNB) and optimizes the weights of the CSI-RS inference network model using the federated learning method. If at a certain moment the gNB model training sub-module evaluates a set of weights with better performance or the model performance deteriorates to a certain threshold, it will notify the model performance evaluation sub-module of the gNB and go to step (7);

[0085] (7) The performance evaluation sub-module selects different UEs within the cell based on relevant rules according to the candidate "CSI-RS feedback inference network models" provided by the model performance evaluation sub-module in the core network, and performs federated learning on multiple different "CSI-RS feedback inference network models" respectively; when receiving the report from the model training sub-module, after local summarization and relevant processing, it reports the relevant information of various models and the model training environment to the model performance evaluation sub-module on the core network side, and proceeds to step (8);

[0086] Optionally, since the CSI-RS feedback inference network model is usually related to the terminal types in the network, the relevant rules may include: for terminals with strong inference capabilities and strong GPU computing resources, relatively complex neural networks can be used.

[0087] Optionally, different UEs can be selected for federated learning according to different CSI-RS feedback inference network models. For example, for UEs with low indoor mobility rates, due to low mobility and stable channel environments, networks with very high compression rates can be used for CSI feedback. For UEs with high mobility, the feedback codewords cannot be too short, and the timeliness requirements are also relatively high.

[0088] (8) The model performance evaluation sub-module on the core network side manages and analyzes multiple CSI-RS feedback inference network models and model applicable environment information, and manages the "CSI-RS feedback inference network models" of different gNBs in the whole network, and proceeds to step (9);

[0089] (9) According to the feedback from each gNB, the core network evaluation sub-module controls the core network model distribution sub-module to initiate all or part of the update of the model and the corresponding weights. The update methods include unicast, multicast or broadcast session methods, and the model or weights of the UEs grouped according to different models in the gNB are updated, and proceeds to step (10);

[0090] It can be understood that by using the multicast and broadcast channels of 5G NR and the multicast method in the same wave speed, it is possible to send the encoder weights of the UE locally to the terminals within the same group using one time-frequency resource. Such a mechanism improves the model update efficiency and saves time-frequency resources.

[0091] (10) The gNB model distribution sub-module controls the UEs within the cell to perform model updates, and all or part of the model parameters are updated, and proceeds to step (2).

[0092] It can be understood that the channel state information feedback system provided by the present invention manages the "CSI-RS feedback inference network model" used between the base station side and the terminal side through the agents distributed on the core network side, the base station side and the terminal side.

[0093] In the embodiments of the present invention, the core network agent manages the candidate "CSI-RS feedback inference network model" for the entire network; the agents distributed on the base station side select a suitable UE group for model distribution, inference application, and online learning according to the prefabricated conditions of different models; when a certain UE-gNB reaches a set threshold and achieves a better effect after a certain amount of online training on the used model, the current weight of the model and information such as external conditions related to the model are collected and reported to the agents of the base station and the core network, and the effects of the model are evaluated at the base station level and the entire network level respectively. The base station agent can expand the evaluation scope of the model and use more UEs to evaluate the model performance.

[0094] Adopt the method of federated learning to perform online training on the model using multiple UEs. The core network side, as the multi-model management and control and model source service provider, supports different gNBs to perform model building, online training, and inference for the models suitable for their own cells. When a certain UE fails to load the model, first, the current gNB collects relevant information and controls the UE to perform model fallback; on the other hand, the gNB notifies the core network to re-initialize the model for the UE according to the relevant conditions of the current UE.

[0095] During the federated learning process of the "CSI-RS feedback inference network model", the federated learning management module in the gNB agent adopts a method based on horizontal federated learning in this application, monitors the federated learning processes of multiple UEs using the same model at the same time, and processes and integrates the model training results to achieve online learning and performance optimization of a specific model.

[0096] The model performance evaluation sub-module on the core network side performs online learning and performance analysis on the candidate models imported in the background, and optimizes and evaluates the model inference ability from different levels, different scenarios and other angles.

[0097] It can be understood that the channel state information feedback system provided by the present invention deploys "CSI-RS feedback inference network" agents on the network side and the terminal side. The agent on the network side manages and maintains a structure of a "CSI-RS feedback inference network" based on federated learning, so as to realize the security protection and behavior traceability of the "CSI-RS feedback inference network" artificial intelligence system in the cell.

[0098] In the embodiments of the present invention, a federated learning mechanism is adopted, which takes the "CSI-RS feedback inference network" agent as the core and involves the joint participation of terminals within the cell, to achieve online training and performance monitoring of various "CSI-RS feedback inference network" models. Through the management and maintenance of the "CSI-RS feedback inference network" agent, functions such as model update, full and partial weight update of the existing model, online learning of the model, performance evaluation, and inference are performed on the "CSI-RS feedback inference network" encoder used by the UEs within the cell and the decoder on the network side that matches it.

[0099] It can be understood that in the embodiments of the present invention, each base station can be responsible for organizing the terminals within its own coverage area to execute the "AI-based CSI feedback" function. Since online training and a federated learning mechanism are adopted, during the use of the model, the model is continuously trained with real service data each time, and the weights of the current model are optimized according to the set federated learning mechanism. When the "model performance evaluation sub-module" on the base station side finds that the performance of the current model has been significantly improved compared to the original model after a period of federated learning, it will update the weights of the model and also report the new model to the "agent management entity" on the core network side. In addition, the "model performance evaluation sub-module" on the base station side can set multiple different models to multiple different terminals within the cell for separate learning, and evaluate the effectiveness of each model based on the obtained results.

[0100] The channel state information feedback system provided by the present invention deploys CSI-RS feedback inference network agents on the network side and the terminal side respectively, and realizes federated learning on a specific feedback inference network model based on the CSI-RS feedback inference network agents on the network side and the terminal side, to obtain a feedback inference network model with better performance, thereby improving the accuracy of channel state information feedback.

[0101] The channel state information feedback method provided by the present invention will be described below. The channel state information feedback method described below can be mutually corresponding and referred to the channel state information feedback system described above.

[0102] Figure 2 is a schematic flowchart of the channel state information feedback method provided by the present invention. As Figure 2 shown, the channel state information feedback method provided by the present invention is applied to the first CSI-RS feedback inference network agent, and may include:

[0103] Step 200, when it is determined that the network side receives a connection establishment request message sent by the target terminal side, establish a connection with the second CSI-RS feedback inference network agent corresponding to the target terminal side;

[0104] Step 210: By interacting with the second CSI-RS feedback inference network agent, determine whether the encoding codewords of the first CSI-RS feedback inference network encoder model currently used on the target terminal side are the same as the decoding codewords of the first CSI-RS feedback inference network decoder model currently used on the network side;

[0105] Step 230: When it is determined that they are the same, based on the first CSI-RS feedback inference network encoder model and the first CSI-RS feedback inference network decoder model, establish the first CSI-RS feedback inference network model between the network side and the target terminal side.

[0106] Specifically, in the embodiments of the present invention, when the first CSI-RS feedback inference network agent determines that the network side receives the connection establishment request message sent by the target terminal side, the first CSI-RS feedback inference network agent may establish a connection with the second CSI-RS feedback inference network agent corresponding to the target terminal side, and then by interacting with the second CSI-RS feedback inference network agent, determine whether the encoding codewords of the first CSI-RS feedback inference network encoder model currently used on the target terminal side are the same as the decoding codewords of the first CSI-RS feedback inference network decoder model currently used on the network side. When it is determined that they are the same, based on the first CSI-RS feedback inference network encoder model and the first CSI-RS feedback inference network decoder model, establish the first CSI-RS feedback inference network model between the network side and the target terminal side.

[0107] For example, when a UE establishes a connection with a gNB, when the UE powers on and registers or switches to a new cell, the second CSI-RS feedback inference network agent on the UE side establishes a connection with the first CSI-RS feedback inference network agent on the gNB side to update the CSI-RS feedback inference network model in the CSI-RS feedback inference network agent on the UE side.

[0108] The channel state information feedback method provided by the present invention establishes a connection between the first CSI-RS feedback inference network agent and the second CSI-RS feedback inference network agent, and then through information interaction, determines whether the encoding codewords of the first CSI-RS feedback inference network encoder model currently used on the target terminal side match the first CSI-RS feedback inference network decoder model currently used on the network side. When they match and are the same, based on the first CSI-RS feedback inference network encoder model and the first CSI-RS feedback inference network decoder model, establish the first CSI-RS feedback inference network model between the network side and the target terminal side, which can improve the accuracy of channel state information feedback.

[0109] Optionally, after establishing the first CSI-RS feedback inference network model between the network side and the target terminal side, the following steps are further included:

[0110] Determine whether the first CSI-RS feedback inference network encoder model is available in this cell;

[0111] In the case of determining that it is not available, send a first indication message carrying a first model update message to the second CSI-RS feedback inference network agent, where the first indication message is used to instruct the second CSI-RS feedback inference network agent to update the first CSI-RS feedback inference network encoder model based on the first model update message.

[0112] Specifically, after establishing the first CSI-RS feedback inference network model between the network side and the target terminal side, the first CSI-RS feedback inference network agent can determine whether the first CSI-RS feedback inference network encoder model is available in this cell. In the case of determining that it is not available, send a first indication message carrying a first model update message to the second CSI-RS feedback inference network agent, where the first indication message is used to instruct the second CSI-RS feedback inference network agent to update the first CSI-RS feedback inference network encoder model based on the first model update message.

[0113] Optionally, the first model update message may include weight information for updating the CSI-RS feedback inference network model.

[0114] Optionally, the channel state information feedback method provided by the present invention further includes:

[0115] In the case of determining that they are not the same, determine whether the first CSI-RS feedback inference network encoder model and the first weights corresponding to the first CSI-RS feedback inference network encoder model are available in this cell;

[0116] In the case of determining that the first CSI-RS feedback inference network encoder model is available in this cell and some or all of the first weights are not available, send a second indication message carrying a second model update message to the second CSI-RS feedback inference network agent, where the second indication message is used to instruct the second CSI-RS feedback inference network agent to update the first CSI-RS feedback inference network encoder model based on the second model update message.

[0117] Specifically, in the embodiments of the present invention, when the first CSI-RS feedback inference network agent determines that the encoded codeword of the first CSI-RS feedback inference network encoder model currently used on the target terminal side is different from the decoded codeword of the first CSI-RS feedback inference network decoder model currently used on the network side, it can determine whether the first CSI-RS feedback inference network encoder model and the first weight corresponding to the first CSI-RS feedback inference network encoder model are available in the current cell; when it is determined that the first CSI-RS feedback inference network encoder model is available in the current cell and some or all of the weights in the first weight are unavailable, it sends second indication information carrying a second model update message to the second CSI-RS feedback inference network agent, and the second indication information is used to instruct the second CSI-RS feedback inference network agent to update the first CSI-RS feedback inference network encoder model based on the second model update message.

[0118] Optionally, the second model update message may include weight information for updating the CSI-RS feedback inference network model.

[0119] Optionally, the second indication information may be used to instruct the second CSI-RS feedback inference network agent to update some or all of the weights in the first weight corresponding to the first CSI-RS feedback inference network encoder model based on the second model update message.

[0120] Optionally, the channel state information feedback method provided by the present invention further includes:

[0121] When it is determined that they are different, abnormal fallback information is sent to the second CSI-RS feedback inference network agent, and the abnormal fallback information is used to instruct the second CSI-RS feedback inference network agent to perform abnormal fallback on the first CSI-RS feedback inference network encoder model.

[0122] Specifically, in the embodiments of the present invention, when the first CSI-RS feedback inference network agent determines that the encoded codeword of the first CSI-RS feedback inference network encoder model currently used on the target terminal side is different from the decoded codeword of the first CSI-RS feedback inference network decoder model currently used on the network side, it can send abnormal fallback information to the second CSI-RS feedback inference network agent, and the abnormal fallback information can be used to instruct the second CSI-RS feedback inference network agent to perform abnormal fallback on the first CSI-RS feedback inference network encoder model.

[0123] It can be understood that when the CSI inference network model is unavailable, it is necessary to update the decoder model on the network side and the encoder model on the terminal side simultaneously. In the embodiments of the present invention, the CSI inference neural network model can be replaced, or the CSI inference neural network model is not replaced, but the weight parameters corresponding to the CSI inference neural network model are rolled back to the previous available version.

[0124] Optionally, after sending the abnormal rollback information to the second CSI-RS feedback inference network agent, it further includes:

[0125] Determine the second CSI-RS feedback inference network encoder model obtained after the second CSI-RS feedback inference network agent performs abnormal rollback on the first CSI-RS feedback inference network encoder model;

[0126] Load the second CSI-RS feedback inference network decoder model that matches the second CSI-RS feedback inference network encoder model, and based on the second CSI-RS feedback inference network encoder model and the second CSI-RS feedback inference network decoder model, establish the second CSI-RS feedback inference network model between the network side and the target terminal side.

[0127] Specifically, after the first CSI-RS feedback inference network agent sends the abnormal rollback information to the second CSI-RS feedback inference network agent, the second CSI-RS feedback inference network encoder model obtained after the second CSI-RS feedback inference network agent performs abnormal rollback on the first CSI-RS feedback inference network encoder model can be determined; then load the second CSI-RS feedback inference network decoder model that matches the second CSI-RS feedback inference network encoder model, and based on the second CSI-RS feedback inference network encoder model and the second CSI-RS feedback inference network decoder model, establish the second CSI-RS feedback inference network model between the network side and the target terminal side.

[0128] Optionally, the channel state information feedback method provided by the present invention further includes:

[0129] Based on the federated learning mechanism, cooperate with the second CSI-RS feedback inference network agent to perform online training on the second CSI-RS feedback inference network model;

[0130] When it is determined that the performance of the second CSI-RS feedback inference network model meets the first preset threshold, optimize the second weight corresponding to the second CSI-RS feedback inference network model based on the federated learning mechanism.

[0131] Specifically, in an embodiment of the present invention, the first CSI-RS feedback inference network agent can cooperate with the second CSI-RS feedback inference network agent to perform online training on the second CSI-RS feedback inference network model based on a federated learning mechanism; after training for a period of time, when it is determined that the performance of the second CSI-RS feedback inference network model meets the first preset threshold, the second weight corresponding to the second CSI-RS feedback inference network model is optimized based on the federated learning mechanism.

[0132] Optionally, the first preset threshold may be that the accuracy of the number of inferences of the second CSI-RS feedback inference network model reaches a certain preset value, or increases by a certain preset value.

[0133] Optionally, the channel state information feedback method provided by the present invention further includes:

[0134] When it is determined that the performance of the second CSI-RS feedback inference network model meets the second preset threshold, a candidate CSI-RS feedback inference network model provided by a third CSI-RS feedback inference network agent is determined, and based on a federated learning mechanism, federated learning is performed on the candidate CSI-RS feedback inference network model.

[0135] Specifically, in an embodiment of the present invention, when determining that the performance of the second CSI-RS feedback inference network model meets the second preset threshold, the first CSI-RS feedback inference network agent determines a candidate CSI-RS feedback inference network model provided by the third CSI-RS feedback inference network agent, and performs federated learning on the candidate CSI-RS feedback inference network model based on the federated learning mechanism.

[0136] Optionally, the channel state information feedback method provided by the present invention further includes:

[0137] Collect statistics on the training results of the local CSI-RS feedback inference network model;

[0138] The training result information is reported to the third CSI-RS feedback reasoning network agent, so that the third CSI-RS feedback reasoning network agent can manage the candidate CSI-RS feedback reasoning network model of the entire network based on the model training result information.

[0139] Specifically, in an embodiment of the present invention, while the first CSI-RS feedback inference network agent executes the federated online training process, the training result information of the local CSI-RS feedback inference network model can be counted; then the training result information is reported to the third CSI-RS feedback inference network agent, so that the third CSI-RS feedback inference network agent can manage the candidate CSI-RS feedback inference network models of the entire network based on the model training result information.

[0140] The channel state information feedback method provided by the present invention establishes a connection between a first CSI-RS feedback inference network agent and a second CSI-RS feedback inference network agent, and then determines whether the encoding codewords of the first CSI-RS feedback inference network encoder model currently used on the target terminal side match the first CSI-RS feedback inference network decoder model currently used on the network side through information interaction. When they match, a first CSI-RS feedback inference network model between the network side and the target terminal side is established based on the first CSI-RS feedback inference network encoder model and the first CSI-RS feedback inference network decoder model, which can improve the accuracy of channel state information feedback.

[0141] Figure 3 It is a schematic diagram of the physical structure of the electronic device provided by the present invention, as Figure 3 shown. The electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communication bus 340. Among them, the processor 310, the communications interface 320, and the memory 330 complete mutual communication through the communication bus 340. The processor 310 can call the logical instructions in the memory 330 to execute the channel state information feedback method provided by each of the above methods, and the method includes:

[0142] When it is determined that the network side receives a connection establishment request message sent by the target terminal side, establish a connection with the second CSI-RS feedback inference network agent corresponding to the target terminal side;

[0143] Through information interaction with the second CSI-RS feedback inference network agent, determine whether the encoding codewords of the first CSI-RS feedback inference network encoder model currently used on the target terminal side are the same as the decoding codewords of the first CSI-RS feedback inference network decoder model currently used on the network side;

[0144] When it is determined that they are the same, establish a first CSI-RS feedback inference network model between the network side and the target terminal side based on the first CSI-RS feedback inference network encoder model and the first CSI-RS feedback inference network decoder model.

[0145] In addition, when the logical instructions in the above-mentioned memory 330 can be 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 such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0146] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the channel state information feedback method provided by the above-mentioned various methods. The method includes:

[0147] When it is determined that the network side receives a connection establishment request message sent by the target terminal side, establish a connection with the second CSI-RS feedback inference network agent corresponding to the target terminal side;

[0148] Through information interaction with the second CSI-RS feedback inference network agent, determine whether the encoding codeword of the first CSI-RS feedback inference network encoder model currently used by the target terminal side is the same as the decoding codeword of the first CSI-RS feedback inference network decoder model currently used by the network side;

[0149] When it is determined that they are the same, based on the first CSI-RS feedback inference network encoder model and the first CSI-RS feedback inference network decoder model, establish a first CSI-RS feedback inference network model between the network side and the target terminal side.

[0150] On yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the channel state information feedback method provided above. The method includes:

[0151] When it is determined that the network side receives a connection establishment request message sent by the target terminal side, establish a connection with the second CSI-RS feedback inference network agent corresponding to the target terminal side;

[0152] By interacting with the information of the second CSI-RS feedback inference network agent, it is determined whether the encoding codeword of the first CSI-RS feedback inference network encoder model currently used on the target terminal side is the same as the decoding codeword of the first CSI-RS feedback inference network decoder model currently used on the network side;

[0153] When it is determined that they are the same, based on the first CSI-RS feedback inference network encoder model and the first CSI-RS feedback inference network decoder model, a first CSI-RS feedback inference network model between the network side and the target terminal side is established.

[0154] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative work.

[0155] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solutions or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A channel state information feedback system, characterized in that, include: A first CSI-RS feedback inference network agent deployed on the network side and a second CSI-RS feedback inference network agent deployed on the terminal side; The first CSI-RS feedback inference network agent is used to send a CSI-RS feedback inference network model to a second CSI-RS feedback inference network agent corresponding to a terminal side in a cell, and cooperate with the second CSI-RS feedback inference network agent to perform online training on the CSI-RS feedback inference network model based on a federated learning mechanism; The second CSI-RS feedback inference network agent is used to cooperate with the first CSI-RS feedback inference network agent to perform online training on the CSI-RS feedback inference network model based on a federated learning mechanism; The first CSI-RS feedback inference network agent is used to: In the case of determining that the network side receives a connection establishment request message sent by the target terminal side, establishing a connection with a second CSI-RS feedback inference network agent corresponding to the target terminal side; By interacting with the second CSI-RS feedback inference network agent, determining whether the encoding codeword of the first CSI-RS feedback inference network encoder model currently used by the target terminal side is the same as the decoding codeword of the first CSI-RS feedback inference network decoder model currently used by the network side; In the case of determining that they are the same, establishing a first CSI-RS feedback inference network model between the network side and the target terminal side based on the first CSI-RS feedback inference network encoder model and the first CSI-RS feedback inference network decoder model; If they are determined to be different, determining whether the first CSI-RS feedback inference network encoder model and the first weight corresponding to the first CSI-RS feedback inference network encoder model are available in the current cell; When it is determined that the first CSI-RS feedback inference network encoder model is available in the current cell and some or all of the first weights are unavailable, a second indication message carrying a second model update message is sent to the second CSI-RS feedback inference network agent, wherein the second indication information is used to instruct the second CSI-RS feedback inference network agent to update the first CSI-RS feedback inference network encoder model based on the second model update message.

2. The channel state information feedback system according to claim 1, characterized in that, Also includes: A third CSI-RS feedback inference network agent deployed on the core network side; The third CSI-RS feedback inference network agent is used to manage and evaluate the performance of the candidate CSI-RS feedback inference network model of the entire network.

3. A channel state information feedback method applied to the channel state information feedback system according to claim 1 or 2, characterized in that, Applied to the first CSI-RS feedback reasoning network agent, including: In the case of determining that the network side receives a connection establishment request message sent by the target terminal side, establishing a connection with a second CSI-RS feedback inference network agent corresponding to the target terminal side; By interacting with the information of the second CSI-RS feedback inference network agent, determine whether the encoded codeword of the first CSI-RS feedback inference network encoder model currently used on the target terminal side is the same as the decoded codeword of the first CSI-RS feedback inference network decoder model currently used on the network side; When it is determined that they are the same, based on the first CSI-RS feedback inference network encoder model and the first CSI-RS feedback inference network decoder model, establish the first CSI-RS feedback inference network model between the network side and the target terminal side; When it is determined that they are not the same, determine whether the first CSI-RS feedback inference network encoder model and the first weight corresponding to the first CSI-RS feedback inference network encoder model are available in the current cell; When it is determined that the first CSI-RS feedback inference network encoder model is available in the current cell and some or all of the first weights are not available, send second indication information carrying a second model update message to the second CSI-RS feedback inference network agent, where the second indication information is used to instruct the second CSI-RS feedback inference network agent to update the first CSI-RS feedback inference network encoder model based on the second model update message.

4. The channel state information feedback method according to claim 3, characterized in that, After establishing the first CSI-RS feedback inference network model between the network side and the target terminal side, it further includes: Determine whether the first CSI-RS feedback inference network encoder model is available in the current cell; When it is determined that it is not available, send first indication information carrying a first model update message to the second CSI-RS feedback inference network agent, where the first indication information is used to instruct the second CSI-RS feedback inference network agent to update the first CSI-RS feedback inference network encoder model based on the first model update message.

5. The channel state information feedback method according to claim 3, characterized in that, It further includes: When it is determined that they are not the same, send abnormal fallback information to the second CSI-RS feedback inference network agent, where the abnormal fallback information is used to instruct the second CSI-RS feedback inference network agent to perform abnormal fallback on the first CSI-RS feedback inference network encoder model.

6. The channel state information feedback method according to claim 5, characterized in that, After sending the abnormal fallback information to the second CSI-RS feedback inference network agent, it further includes: Determine the second CSI-RS feedback inference network encoder model obtained after the second CSI-RS feedback inference network agent performs abnormal fallback on the first CSI-RS feedback inference network encoder model; Load the second CSI-RS feedback inference network decoder model matching the second CSI-RS feedback inference network encoder model, and based on the second CSI-RS feedback inference network encoder model and the second CSI-RS feedback inference network decoder model, establish the second CSI-RS feedback inference network model between the network side and the target terminal side.

7. The channel state information feedback method according to claim 6, characterized in that, It further includes: Based on the federated learning mechanism, cooperate with the second CSI-RS feedback inference network agent to perform online training on the second CSI-RS feedback inference network model; When it is determined that the performance of the second CSI-RS feedback inference network model meets the first preset threshold, optimize the second weights corresponding to the second CSI-RS feedback inference network model based on the federated learning mechanism.

8. The channel state information feedback method according to claim 7, wherein It further includes: When it is determined that the performance of the second CSI-RS feedback inference network model meets the second preset threshold, determine the candidate CSI-RS feedback inference network model provided by the third CSI-RS feedback inference network agent, and perform federated learning on the candidate CSI-RS feedback inference network model based on the federated learning mechanism.

9. The channel state information feedback method according to claim 8, wherein It further includes: Statistically analyze the training result information of the local CSI-RS feedback inference network model; Report the training result information to the third CSI-RS feedback inference network agent for the third CSI-RS feedback inference network agent to manage the candidate CSI-RS feedback inference network models across the network based on the model training result information.

Images